ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/cvsroot/libeio/eio.c

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.33 by root, Sat Jun 6 19:44:17 2009 UTC vs.
Revision 1.94 by root, Mon Jul 18 02:59:58 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	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	static void eio_destroy (eio_req *req);
		77
58	#ifndef EIO_FINISH	78	#ifndef EIO_FINISH
59	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	79	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60	#endif	80	#endif
61		81
62	#ifndef EIO_DESTROY	82	#ifndef EIO_DESTROY
…		…
65		85
66	#ifndef EIO_FEED	86	#ifndef EIO_FEED
67	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	87	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68	#endif	88	#endif
69		89
		90	#ifndef EIO_FD_TO_WIN32_HANDLE
		91	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		92	#endif
		93	#ifndef EIO_WIN32_HANDLE_TO_FD
		94	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		95	#endif
		96
		97	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		98
		99	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		100
70	#ifdef _WIN32	101	#ifdef _WIN32
71		102
72	/*doh*/	103	#undef PAGESIZE
		104	#define PAGESIZE 4096 /* GetSystemInfo? */
		105
		106	#ifdef EIO_STRUCT_STATI64
		107	#define stat(path,buf) _stati64 (path,buf)
		108	#define fstat(fd,buf) _fstati64 (path,buf)
		109	#endif
		110	#define lstat(path,buf) stat (path,buf)
		111	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		112	#define mkdir(path,mode) _mkdir (path)
		113	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		114
		115	#define chmod(path,mode) _chmod (path, mode)
		116	#define dup(fd) _dup (fd)
		117	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		118
		119	#define fchmod(fd,mode) EIO_ENOSYS ()
		120	#define chown(path,uid,gid) EIO_ENOSYS ()
		121	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		122	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		123	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		124	#define mknod(path,mode,dev) EIO_ENOSYS ()
		125	#define sync() EIO_ENOSYS ()
		126	#define readlink(path,buf,s) EIO_ENOSYS ()
		127	#define statvfs(path,buf) EIO_ENOSYS ()
		128	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		129
		130	/* we could even stat and see if it exists */
		131	static int
		132	symlink (const char *old, const char *neu)
		133	{
		134	#if WINVER >= 0x0600
		135	if (CreateSymbolicLink (neu, old, 1))
		136	return 0;
		137
		138	if (CreateSymbolicLink (neu, old, 0))
		139	return 0;
		140	#endif
		141
		142	return EIO_ERRNO (ENOENT, -1);
		143	}
		144
		145	/* POSIX API only */
		146	#define CreateHardLink(neu,old,flags) 0
		147	#define CreateSymbolicLink(neu,old,flags) 0
		148
		149	struct statvfs
		150	{
		151	int dummy;
		152	};
		153
		154	#define DT_DIR EIO_DT_DIR
		155	#define DT_REG EIO_DT_REG
		156	#define D_NAME(entp) entp.cFileName
		157	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
		158
73	#else	159	#else
74		160
75	# include "config.h"
76	# include <sys/time.h>	161	#include <sys/time.h>
77	# include <sys/select.h>	162	#include <sys/select.h>
		163	#include <sys/statvfs.h>
78	# include <unistd.h>	164	#include <unistd.h>
79	# include <utime.h>	165	#include <utime.h>
80	# include <signal.h>	166	#include <signal.h>
81	# include <dirent.h>	167	#include <dirent.h>
82		168
		169	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		170	#include <sys/mman.h>
		171	#endif
		172
		173	#define D_NAME(entp) entp->d_name
		174
83	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */	175	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84	# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)	176	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
85	# define _DIRENT_HAVE_D_TYPE /* sigh */	177	#define _DIRENT_HAVE_D_TYPE /* sigh */
86	# define D_INO(de) (de)->d_fileno	178	#define D_INO(de) (de)->d_fileno
87	# define D_NAMLEN(de) (de)->d_namlen	179	#define D_NAMLEN(de) (de)->d_namlen
88	# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600	180	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
89	# define D_INO(de) (de)->d_ino	181	#define D_INO(de) (de)->d_ino
90	# endif	182	#endif
91		183
92	#ifdef _D_EXACT_NAMLEN	184	#ifdef _D_EXACT_NAMLEN
93	# undef D_NAMLEN	185	#undef D_NAMLEN
94	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)	186	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
95	#endif	187	#endif
96		188
97	# ifdef _DIRENT_HAVE_D_TYPE	189	#ifdef _DIRENT_HAVE_D_TYPE
98	# define D_TYPE(de) (de)->d_type	190	#define D_TYPE(de) (de)->d_type
99	# endif	191	#endif
100		192
101	# ifndef EIO_STRUCT_DIRENT	193	#ifndef EIO_STRUCT_DIRENT
102	# define EIO_STRUCT_DIRENT struct dirent	194	#define EIO_STRUCT_DIRENT struct dirent
103	# endif	195	#endif
104		196
105	#endif	197	#endif
106		198
107	#if HAVE_SENDFILE	199	#if HAVE_SENDFILE
108	# if __linux	200	# if __linux
109	# include <sys/sendfile.h>	201	# include <sys/sendfile.h>
110	# elif __freebsd	202	# elif __FreeBSD__ || defined __APPLE__
111	# include <sys/socket.h>	203	# include <sys/socket.h>
112	# include <sys/uio.h>	204	# include <sys/uio.h>
113	# elif __hpux	205	# elif __hpux
114	# include <sys/socket.h>	206	# include <sys/socket.h>
115	# elif __solaris /* not yet */	207	# elif __solaris
116	# include <sys/sendfile.h>	208	# include <sys/sendfile.h>
117	# else	209	# else
118	# error sendfile support requested but not available	210	# error sendfile support requested but not available
119	# endif	211	# endif
120	#endif	212	#endif
…		…
124	#endif	216	#endif
125	#ifndef D_INO	217	#ifndef D_INO
126	# define D_INO(de) 0	218	# define D_INO(de) 0
127	#endif	219	#endif
128	#ifndef D_NAMLEN	220	#ifndef D_NAMLEN
129	# define D_NAMLEN(de) strlen ((de)->d_name)	221	# define D_NAMLEN(entp) strlen (D_NAME (entp))
130	#endif	222	#endif
131
132	/* number of seconds after which an idle threads exit */
133	#define IDLE_TIMEOUT 10
134		223
135	/* used for struct dirent, AIX doesn't provide it */	224	/* used for struct dirent, AIX doesn't provide it */
136	#ifndef NAME_MAX	225	#ifndef NAME_MAX
137	# define NAME_MAX 4096	226	# define NAME_MAX 4096
138	#endif	227	#endif
139		228
		229	/* used for readlink etc. */
		230	#ifndef PATH_MAX
		231	# define PATH_MAX 4096
		232	#endif
		233
140	/* buffer size for various temporary buffers */	234	/* buffer size for various temporary buffers */
141	#define EIO_BUFSIZE 65536	235	#define EIO_BUFSIZE 65536
142		236
143	#define dBUF \	237	#define dBUF \
144	char *eio_buf; \
145	ETP_WORKER_LOCK (self); \
146	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	238	char *eio_buf = malloc (EIO_BUFSIZE); \
147	ETP_WORKER_UNLOCK (self); \
148	errno = ENOMEM; \	239	errno = ENOMEM; \
149	if (!eio_buf) \	240	if (!eio_buf) \
150	return -1;	241	return -1
		242
		243	#define FUBd \
		244	free (eio_buf)
151		245
152	#define EIO_TICKS ((1000000 + 1023) >> 10)	246	#define EIO_TICKS ((1000000 + 1023) >> 10)
153
154	/*****************************************************************************/
155
156	#if __GNUC__ >= 3
157	# define expect(expr,value) __builtin_expect ((expr),(value))
158	#else
159	# define expect(expr,value) (expr)
160	#endif
161
162	#define expect_false(expr) expect ((expr) != 0, 0)
163	#define expect_true(expr) expect ((expr) != 0, 1)
164
165	/*****************************************************************************/
166		247
167	#define ETP_PRI_MIN EIO_PRI_MIN	248	#define ETP_PRI_MIN EIO_PRI_MIN
168	#define ETP_PRI_MAX EIO_PRI_MAX	249	#define ETP_PRI_MAX EIO_PRI_MAX
169		250
170	struct etp_worker;	251	struct etp_worker;
…		…
174	static int eio_finish (eio_req *req);	255	static int eio_finish (eio_req *req);
175	#define ETP_FINISH(req) eio_finish (req)	256	#define ETP_FINISH(req) eio_finish (req)
176	static void eio_execute (struct etp_worker *self, eio_req *req);	257	static void eio_execute (struct etp_worker *self, eio_req *req);
177	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	258	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
178		259
179	#define ETP_WORKER_CLEAR(req) \
180	if (wrk->dbuf) \
181	{ \
182	free (wrk->dbuf); \
183	wrk->dbuf = 0; \
184	} \
185	\
186	if (wrk->dirp) \
187	{ \
188	closedir (wrk->dirp); \
189	wrk->dirp = 0; \
190	}
191
192	#define ETP_WORKER_COMMON \
193	void *dbuf; \
194	DIR *dirp;
195
196	/*****************************************************************************/	260	/*****************************************************************************/
197		261
198	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	262	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
199		263
200	/* calculcate time difference in ~1/EIO_TICKS of a second */	264	/* calculate time difference in ~1/EIO_TICKS of a second */
		265	ecb_inline int
201	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	266	tvdiff (struct timeval *tv1, struct timeval *tv2)
202	{	267	{
203	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	268	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
204	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	269	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
205	}	270	}
206		271
…		…
213	static unsigned int max_poll_reqs; /* reslock */	278	static unsigned int max_poll_reqs; /* reslock */
214		279
215	static volatile unsigned int nreqs; /* reqlock */	280	static volatile unsigned int nreqs; /* reqlock */
216	static volatile unsigned int nready; /* reqlock */	281	static volatile unsigned int nready; /* reqlock */
217	static volatile unsigned int npending; /* reqlock */	282	static volatile unsigned int npending; /* reqlock */
218	static volatile unsigned int max_idle = 4;	283	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		284	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
219		285
220	static mutex_t wrklock = X_MUTEX_INIT;	286	static xmutex_t wrklock;
221	static mutex_t reslock = X_MUTEX_INIT;	287	static xmutex_t reslock;
222	static mutex_t reqlock = X_MUTEX_INIT;	288	static xmutex_t reqlock;
223	static cond_t reqwait = X_COND_INIT;	289	static xcond_t reqwait;
224		290
225	#if !HAVE_PREADWRITE	291	#if !HAVE_PREADWRITE
226	/*	292	/*
227	* make our pread/pwrite emulation safe against themselves, but not against	293	* make our pread/pwrite emulation safe against themselves, but not against
228	* normal read/write by using a mutex. slows down execution a lot,	294	* normal read/write by using a mutex. slows down execution a lot,
229	* but that's your problem, not mine.	295	* but that's your problem, not mine.
230	*/	296	*/
231	static mutex_t preadwritelock = X_MUTEX_INIT;	297	static xmutex_t preadwritelock;
232	#endif	298	#endif
233		299
234	typedef struct etp_worker	300	typedef struct etp_worker
235	{	301	{
236	/* locked by wrklock */	302	/* locked by wrklock */
237	struct etp_worker *prev, *next;	303	struct etp_worker *prev, *next;
238		304
239	thread_t tid;	305	xthread_t tid;
240		306
241	/* locked by reslock, reqlock or wrklock */	307	/* locked by reslock, reqlock or wrklock */
242	ETP_REQ *req; /* currently processed request */	308	ETP_REQ *req; /* currently processed request */
243		309
		310	#ifdef ETP_WORKER_COMMON
244	ETP_WORKER_COMMON	311	ETP_WORKER_COMMON
		312	#endif
245	} etp_worker;	313	} etp_worker;
246		314
247	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	315	static etp_worker wrk_first; /* NOT etp */
248		316
249	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	317	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
250	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	318	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
251		319
252	/* worker threads management */	320	/* worker threads management */
253		321
		322	static void ecb_cold
254	static void etp_worker_clear (etp_worker *wrk)	323	etp_worker_clear (etp_worker *wrk)
255	{	324	{
256	ETP_WORKER_CLEAR (wrk);
257	}	325	}
258		326
		327	static void ecb_cold
259	static void etp_worker_free (etp_worker *wrk)	328	etp_worker_free (etp_worker *wrk)
260	{	329	{
261	wrk->next->prev = wrk->prev;	330	wrk->next->prev = wrk->prev;
262	wrk->prev->next = wrk->next;	331	wrk->prev->next = wrk->next;
263		332
264	free (wrk);	333	free (wrk);
265	}	334	}
266		335
267	static unsigned int etp_nreqs (void)	336	static unsigned int
		337	etp_nreqs (void)
268	{	338	{
269	int retval;	339	int retval;
270	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	340	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271	retval = nreqs;	341	retval = nreqs;
272	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	342	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273	return retval;	343	return retval;
274	}	344	}
275		345
276	static unsigned int etp_nready (void)	346	static unsigned int
		347	etp_nready (void)
277	{	348	{
278	unsigned int retval;	349	unsigned int retval;
279		350
280	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	351	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281	retval = nready;	352	retval = nready;
282	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	353	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283		354
284	return retval;	355	return retval;
285	}	356	}
286		357
287	static unsigned int etp_npending (void)	358	static unsigned int
		359	etp_npending (void)
288	{	360	{
289	unsigned int retval;	361	unsigned int retval;
290		362
291	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	363	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
292	retval = npending;	364	retval = npending;
293	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	365	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
294		366
295	return retval;	367	return retval;
296	}	368	}
297		369
298	static unsigned int etp_nthreads (void)	370	static unsigned int
		371	etp_nthreads (void)
299	{	372	{
300	unsigned int retval;	373	unsigned int retval;
301		374
302	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	375	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303	retval = started;	376	retval = started;
…		…
317	} etp_reqq;	390	} etp_reqq;
318		391
319	static etp_reqq req_queue;	392	static etp_reqq req_queue;
320	static etp_reqq res_queue;	393	static etp_reqq res_queue;
321		394
		395	static void ecb_noinline ecb_cold
		396	reqq_init (etp_reqq *q)
		397	{
		398	int pri;
		399
		400	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		401	q->qs[pri] = q->qe[pri] = 0;
		402
		403	q->size = 0;
		404	}
		405
		406	static int ecb_noinline
322	static int reqq_push (etp_reqq *q, ETP_REQ *req)	407	reqq_push (etp_reqq *q, ETP_REQ *req)
323	{	408	{
324	int pri = req->pri;	409	int pri = req->pri;
325	req->next = 0;	410	req->next = 0;
326		411
327	if (q->qe[pri])	412	if (q->qe[pri])
…		…
333	q->qe[pri] = q->qs[pri] = req;	418	q->qe[pri] = q->qs[pri] = req;
334		419
335	return q->size++;	420	return q->size++;
336	}	421	}
337		422
		423	static ETP_REQ * ecb_noinline
338	static ETP_REQ *reqq_shift (etp_reqq *q)	424	reqq_shift (etp_reqq *q)
339	{	425	{
340	int pri;	426	int pri;
341		427
342	if (!q->size)	428	if (!q->size)
343	return 0;	429	return 0;
…		…
358	}	444	}
359		445
360	abort ();	446	abort ();
361	}	447	}
362		448
363	static void etp_atfork_prepare (void)	449	static int ecb_cold
		450	etp_init (void (*want_poll)(void), void (*done_poll)(void))
364	{	451	{
365	X_LOCK (wrklock);	452	X_MUTEX_CREATE (wrklock);
366	X_LOCK (reqlock);	453	X_MUTEX_CREATE (reslock);
367	X_LOCK (reslock);	454	X_MUTEX_CREATE (reqlock);
368	#if !HAVE_PREADWRITE	455	X_COND_CREATE (reqwait);
369	X_LOCK (preadwritelock);
370	#endif
371	}
372		456
373	static void etp_atfork_parent (void)	457	reqq_init (&req_queue);
374	{	458	reqq_init (&res_queue);
375	#if !HAVE_PREADWRITE
376	X_UNLOCK (preadwritelock);
377	#endif
378	X_UNLOCK (reslock);
379	X_UNLOCK (reqlock);
380	X_UNLOCK (wrklock);
381	}
382		459
383	static void etp_atfork_child (void)	460	wrk_first.next =
384	{	461	wrk_first.prev = &wrk_first;
385	ETP_REQ *prv;
386
387	while ((prv = reqq_shift (&req_queue)))
388	ETP_DESTROY (prv);
389
390	while ((prv = reqq_shift (&res_queue)))
391	ETP_DESTROY (prv);
392
393	while (wrk_first.next != &wrk_first)
394	{
395	etp_worker *wrk = wrk_first.next;
396
397	if (wrk->req)
398	ETP_DESTROY (wrk->req);
399
400	etp_worker_clear (wrk);
401	etp_worker_free (wrk);
402	}
403		462
404	started = 0;	463	started = 0;
405	idle = 0;	464	idle = 0;
406	nreqs = 0;	465	nreqs = 0;
407	nready = 0;	466	nready = 0;
408	npending = 0;	467	npending = 0;
409		468
410	etp_atfork_parent ();
411	}
412
413	static void
414	etp_once_init (void)
415	{
416	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
417	}
418
419	static int
420	etp_init (void (*want_poll)(void), void (*done_poll)(void))
421	{
422	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
423
424	pthread_once (&doinit, etp_once_init);
425
426	want_poll_cb = want_poll;	469	want_poll_cb = want_poll;
427	done_poll_cb = done_poll;	470	done_poll_cb = done_poll;
428		471
429	return 0;	472	return 0;
430	}	473	}
431		474
432	X_THREAD_PROC (etp_proc);	475	X_THREAD_PROC (etp_proc);
433		476
		477	static void ecb_cold
434	static void etp_start_thread (void)	478	etp_start_thread (void)
435	{	479	{
436	etp_worker *wrk = calloc (1, sizeof (etp_worker));	480	etp_worker *wrk = calloc (1, sizeof (etp_worker));
437		481
438	/*TODO*/	482	/*TODO*/
439	assert (("unable to allocate worker thread data", wrk));	483	assert (("unable to allocate worker thread data", wrk));
…		…
452	free (wrk);	496	free (wrk);
453		497
454	X_UNLOCK (wrklock);	498	X_UNLOCK (wrklock);
455	}	499	}
456		500
		501	static void
457	static void etp_maybe_start_thread (void)	502	etp_maybe_start_thread (void)
458	{	503	{
459	if (expect_true (etp_nthreads () >= wanted))	504	if (ecb_expect_true (etp_nthreads () >= wanted))
460	return;	505	return;
461		506
462	/* todo: maybe use idle here, but might be less exact */	507	/* todo: maybe use idle here, but might be less exact */
463	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	508	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
464	return;	509	return;
465		510
466	etp_start_thread ();	511	etp_start_thread ();
467	}	512	}
468		513
		514	static void ecb_cold
469	static void etp_end_thread (void)	515	etp_end_thread (void)
470	{	516	{
471	eio_req *req = calloc (1, sizeof (eio_req));	517	eio_req *req = calloc (1, sizeof (eio_req));
472		518
473	req->type = -1;	519	req->type = -1;
474	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	520	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
481	X_LOCK (wrklock);	527	X_LOCK (wrklock);
482	--started;	528	--started;
483	X_UNLOCK (wrklock);	529	X_UNLOCK (wrklock);
484	}	530	}
485		531
486	static int etp_poll (void)	532	static int
		533	etp_poll (void)
487	{	534	{
488	unsigned int maxreqs;	535	unsigned int maxreqs;
489	unsigned int maxtime;	536	unsigned int maxtime;
490	struct timeval tv_start, tv_now;	537	struct timeval tv_start, tv_now;
491		538
…		…
521		568
522	X_LOCK (reqlock);	569	X_LOCK (reqlock);
523	--nreqs;	570	--nreqs;
524	X_UNLOCK (reqlock);	571	X_UNLOCK (reqlock);
525		572
526	if (expect_false (req->type == EIO_GROUP && req->size))	573	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
527	{	574	{
528	req->int1 = 1; /* mark request as delayed */	575	req->int1 = 1; /* mark request as delayed */
529	continue;	576	continue;
530	}	577	}
531	else	578	else
532	{	579	{
533	int res = ETP_FINISH (req);	580	int res = ETP_FINISH (req);
534	if (expect_false (res))	581	if (ecb_expect_false (res))
535	return res;	582	return res;
536	}	583	}
537		584
538	if (expect_false (maxreqs && !--maxreqs))	585	if (ecb_expect_false (maxreqs && !--maxreqs))
539	break;	586	break;
540		587
541	if (maxtime)	588	if (maxtime)
542	{	589	{
543	gettimeofday (&tv_now, 0);	590	gettimeofday (&tv_now, 0);
…		…
549		596
550	errno = EAGAIN;	597	errno = EAGAIN;
551	return -1;	598	return -1;
552	}	599	}
553		600
		601	static void
554	static void etp_cancel (ETP_REQ *req)	602	etp_cancel (ETP_REQ *req)
555	{	603	{
556	X_LOCK (wrklock);	604	req->cancelled = 1;
557	req->flags |= EIO_FLAG_CANCELLED;
558	X_UNLOCK (wrklock);
559		605
560	eio_grp_cancel (req);	606	eio_grp_cancel (req);
561	}	607	}
562		608
		609	static void
563	static void etp_submit (ETP_REQ *req)	610	etp_submit (ETP_REQ *req)
564	{	611	{
565	req->pri -= ETP_PRI_MIN;	612	req->pri -= ETP_PRI_MIN;
566		613
567	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	614	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
568	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	615	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
569		616
570	if (expect_false (req->type == EIO_GROUP))	617	if (ecb_expect_false (req->type == EIO_GROUP))
571	{	618	{
572	/* I hope this is worth it :/ */	619	/* I hope this is worth it :/ */
573	X_LOCK (reqlock);	620	X_LOCK (reqlock);
574	++nreqs;	621	++nreqs;
575	X_UNLOCK (reqlock);	622	X_UNLOCK (reqlock);
…		…
594		641
595	etp_maybe_start_thread ();	642	etp_maybe_start_thread ();
596	}	643	}
597	}	644	}
598		645
		646	static void ecb_cold
599	static void etp_set_max_poll_time (double nseconds)	647	etp_set_max_poll_time (double nseconds)
600	{	648	{
601	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	649	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
602	max_poll_time = nseconds;	650	max_poll_time = nseconds * EIO_TICKS;
603	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	651	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
604	}	652	}
605		653
		654	static void ecb_cold
606	static void etp_set_max_poll_reqs (unsigned int maxreqs)	655	etp_set_max_poll_reqs (unsigned int maxreqs)
607	{	656	{
608	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	657	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609	max_poll_reqs = maxreqs;	658	max_poll_reqs = maxreqs;
610	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	659	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611	}	660	}
612		661
		662	static void ecb_cold
613	static void etp_set_max_idle (unsigned int nthreads)	663	etp_set_max_idle (unsigned int nthreads)
614	{	664	{
615	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	665	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
616	max_idle = nthreads <= 0 ? 1 : nthreads;	666	max_idle = nthreads;
617	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	667	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
618	}	668	}
619		669
		670	static void ecb_cold
		671	etp_set_idle_timeout (unsigned int seconds)
		672	{
		673	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		674	idle_timeout = seconds;
		675	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		676	}
		677
		678	static void ecb_cold
620	static void etp_set_min_parallel (unsigned int nthreads)	679	etp_set_min_parallel (unsigned int nthreads)
621	{	680	{
622	if (wanted < nthreads)	681	if (wanted < nthreads)
623	wanted = nthreads;	682	wanted = nthreads;
624	}	683	}
625		684
		685	static void ecb_cold
626	static void etp_set_max_parallel (unsigned int nthreads)	686	etp_set_max_parallel (unsigned int nthreads)
627	{	687	{
628	if (wanted > nthreads)	688	if (wanted > nthreads)
629	wanted = nthreads;	689	wanted = nthreads;
630		690
631	while (started > wanted)	691	while (started > wanted)
632	etp_end_thread ();	692	etp_end_thread ();
633	}	693	}
634		694
635	/*****************************************************************************/	695	/*****************************************************************************/
636		696
		697	static void
637	static void grp_try_feed (eio_req *grp)	698	grp_try_feed (eio_req *grp)
638	{	699	{
639	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	700	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
640	{	701	{
641	grp->flags &= ~EIO_FLAG_GROUPADD;	702	grp->flags &= ~EIO_FLAG_GROUPADD;
642		703
…		…
649	break;	710	break;
650	}	711	}
651	}	712	}
652	}	713	}
653		714
		715	static int
654	static int grp_dec (eio_req *grp)	716	grp_dec (eio_req *grp)
655	{	717	{
656	--grp->size;	718	--grp->size;
657		719
658	/* call feeder, if applicable */	720	/* call feeder, if applicable */
659	grp_try_feed (grp);	721	grp_try_feed (grp);
…		…
663	return eio_finish (grp);	725	return eio_finish (grp);
664	else	726	else
665	return 0;	727	return 0;
666	}	728	}
667		729
		730	static void
668	void eio_destroy (eio_req *req)	731	eio_destroy (eio_req *req)
669	{	732	{
670	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	733	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
671	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	734	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
672		735
673	EIO_DESTROY (req);	736	EIO_DESTROY (req);
674	}	737	}
675		738
		739	static int
676	static int eio_finish (eio_req *req)	740	eio_finish (eio_req *req)
677	{	741	{
678	int res = EIO_FINISH (req);	742	int res = EIO_FINISH (req);
679		743
680	if (req->grp)	744	if (req->grp)
681	{	745	{
…		…
689	if (grp->grp_first == req)	753	if (grp->grp_first == req)
690	grp->grp_first = req->grp_next;	754	grp->grp_first = req->grp_next;
691		755
692	res2 = grp_dec (grp);	756	res2 = grp_dec (grp);
693		757
694	if (!res && res2)	758	if (!res)
695	res = res2;	759	res = res2;
696	}	760	}
697		761
698	eio_destroy (req);	762	eio_destroy (req);
699		763
700	return res;	764	return res;
701	}	765	}
702		766
		767	void
703	void eio_grp_cancel (eio_req *grp)	768	eio_grp_cancel (eio_req *grp)
704	{	769	{
705	for (grp = grp->grp_first; grp; grp = grp->grp_next)	770	for (grp = grp->grp_first; grp; grp = grp->grp_next)
706	eio_cancel (grp);	771	eio_cancel (grp);
707	}	772	}
708		773
		774	void
709	void eio_cancel (eio_req *req)	775	eio_cancel (eio_req *req)
710	{	776	{
711	etp_cancel (req);	777	etp_cancel (req);
712	}	778	}
713		779
		780	void
714	void eio_submit (eio_req *req)	781	eio_submit (eio_req *req)
715	{	782	{
716	etp_submit (req);	783	etp_submit (req);
717	}	784	}
718		785
719	unsigned int eio_nreqs (void)	786	unsigned int
		787	eio_nreqs (void)
720	{	788	{
721	return etp_nreqs ();	789	return etp_nreqs ();
722	}	790	}
723		791
724	unsigned int eio_nready (void)	792	unsigned int
		793	eio_nready (void)
725	{	794	{
726	return etp_nready ();	795	return etp_nready ();
727	}	796	}
728		797
729	unsigned int eio_npending (void)	798	unsigned int
		799	eio_npending (void)
730	{	800	{
731	return etp_npending ();	801	return etp_npending ();
732	}	802	}
733		803
734	unsigned int eio_nthreads (void)	804	unsigned int ecb_cold
		805	eio_nthreads (void)
735	{	806	{
736	return etp_nthreads ();	807	return etp_nthreads ();
737	}	808	}
738		809
		810	void ecb_cold
739	void eio_set_max_poll_time (double nseconds)	811	eio_set_max_poll_time (double nseconds)
740	{	812	{
741	etp_set_max_poll_time (nseconds);	813	etp_set_max_poll_time (nseconds);
742	}	814	}
743		815
		816	void ecb_cold
744	void eio_set_max_poll_reqs (unsigned int maxreqs)	817	eio_set_max_poll_reqs (unsigned int maxreqs)
745	{	818	{
746	etp_set_max_poll_reqs (maxreqs);	819	etp_set_max_poll_reqs (maxreqs);
747	}	820	}
748		821
		822	void ecb_cold
749	void eio_set_max_idle (unsigned int nthreads)	823	eio_set_max_idle (unsigned int nthreads)
750	{	824	{
751	etp_set_max_idle (nthreads);	825	etp_set_max_idle (nthreads);
752	}	826	}
753		827
		828	void ecb_cold
		829	eio_set_idle_timeout (unsigned int seconds)
		830	{
		831	etp_set_idle_timeout (seconds);
		832	}
		833
		834	void ecb_cold
754	void eio_set_min_parallel (unsigned int nthreads)	835	eio_set_min_parallel (unsigned int nthreads)
755	{	836	{
756	etp_set_min_parallel (nthreads);	837	etp_set_min_parallel (nthreads);
757	}	838	}
758		839
		840	void ecb_cold
759	void eio_set_max_parallel (unsigned int nthreads)	841	eio_set_max_parallel (unsigned int nthreads)
760	{	842	{
761	etp_set_max_parallel (nthreads);	843	etp_set_max_parallel (nthreads);
762	}	844	}
763		845
764	int eio_poll (void)	846	int eio_poll (void)
…		…
773	# undef pread	855	# undef pread
774	# undef pwrite	856	# undef pwrite
775	# define pread eio__pread	857	# define pread eio__pread
776	# define pwrite eio__pwrite	858	# define pwrite eio__pwrite
777		859
778	static ssize_t	860	static eio_ssize_t
779	eio__pread (int fd, void *buf, size_t count, off_t offset)	861	eio__pread (int fd, void *buf, size_t count, off_t offset)
780	{	862	{
781	ssize_t res;	863	eio_ssize_t res;
782	off_t ooffset;	864	off_t ooffset;
783		865
784	X_LOCK (preadwritelock);	866	X_LOCK (preadwritelock);
785	ooffset = lseek (fd, 0, SEEK_CUR);	867	ooffset = lseek (fd, 0, SEEK_CUR);
786	lseek (fd, offset, SEEK_SET);	868	lseek (fd, offset, SEEK_SET);
…		…
789	X_UNLOCK (preadwritelock);	871	X_UNLOCK (preadwritelock);
790		872
791	return res;	873	return res;
792	}	874	}
793		875
794	static ssize_t	876	static eio_ssize_t
795	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	877	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
796	{	878	{
797	ssize_t res;	879	eio_ssize_t res;
798	off_t ooffset;	880	off_t ooffset;
799		881
800	X_LOCK (preadwritelock);	882	X_LOCK (preadwritelock);
801	ooffset = lseek (fd, 0, SEEK_CUR);	883	ooffset = lseek (fd, 0, SEEK_CUR);
802	lseek (fd, offset, SEEK_SET);	884	lseek (fd, offset, SEEK_SET);
…		…
806		888
807	return res;	889	return res;
808	}	890	}
809	#endif	891	#endif
810		892
811	#ifndef HAVE_FUTIMES	893	#ifndef HAVE_UTIMES
812		894
813	# undef utimes	895	# undef utimes
814	# undef futimes
815	# define utimes(path,times) eio__utimes (path, times)	896	# define utimes(path,times) eio__utimes (path, times)
816	# define futimes(fd,times) eio__futimes (fd, times)
817		897
818	static int	898	static int
819	eio__utimes (const char *filename, const struct timeval times[2])	899	eio__utimes (const char *filename, const struct timeval times[2])
820	{	900	{
821	if (times)	901	if (times)
…		…
829	}	909	}
830	else	910	else
831	return utime (filename, 0);	911	return utime (filename, 0);
832	}	912	}
833		913
		914	#endif
		915
		916	#ifndef HAVE_FUTIMES
		917
		918	# undef futimes
		919	# define futimes(fd,times) eio__futimes (fd, times)
		920
		921	static int
834	static int eio__futimes (int fd, const struct timeval tv[2])	922	eio__futimes (int fd, const struct timeval tv[2])
835	{	923	{
836	errno = ENOSYS;	924	errno = ENOSYS;
837	return -1;	925	return -1;
838	}	926	}
839		927
…		…
843	# undef fdatasync	931	# undef fdatasync
844	# define fdatasync(fd) fsync (fd)	932	# define fdatasync(fd) fsync (fd)
845	#endif	933	#endif
846		934
847	/* sync_file_range always needs emulation */	935	/* sync_file_range always needs emulation */
848	int	936	static int
849	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	937	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
850	{	938	{
851	#if HAVE_SYNC_FILE_RANGE	939	#if HAVE_SYNC_FILE_RANGE
852	int res;	940	int res;
853		941
…		…
866	if (!res || errno != ENOSYS)	954	if (!res || errno != ENOSYS)
867	return res;	955	return res;
868	#endif	956	#endif
869		957
870	/* even though we could play tricks with the flags, it's better to always	958	/* even though we could play tricks with the flags, it's better to always
871	* call fdatasync, as thta matches the expectation of it's users best */	959	* call fdatasync, as that matches the expectation of its users best */
872	return fdatasync (fd);	960	return fdatasync (fd);
		961	}
		962
		963	static int
		964	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		965	{
		966	#if HAVE_FALLOCATE
		967	return fallocate (fd, mode, offset, len);
		968	#else
		969	errno = ENOSYS;
		970	return -1;
		971	#endif
873	}	972	}
874		973
875	#if !HAVE_READAHEAD	974	#if !HAVE_READAHEAD
876	# undef readahead	975	# undef readahead
877	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	976	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
878		977
879	static ssize_t	978	static eio_ssize_t
880	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	979	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
881	{	980	{
882	size_t todo = count;	981	size_t todo = count;
883	dBUF;	982	dBUF;
884		983
…		…
889	pread (fd, eio_buf, len, offset);	988	pread (fd, eio_buf, len, offset);
890	offset += len;	989	offset += len;
891	todo -= len;	990	todo -= len;
892	}	991	}
893		992
		993	FUBd;
		994
894	errno = 0;	995	errno = 0;
895	return count;	996	return count;
896	}	997	}
897		998
898	#endif	999	#endif
899		1000
900	/* sendfile always needs emulation */	1001	/* sendfile always needs emulation */
901	static ssize_t	1002	static eio_ssize_t
902	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1003	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
903	{	1004	{
		1005	eio_ssize_t written = 0;
904	ssize_t res;	1006	eio_ssize_t res;
905		1007
906	if (!count)	1008	if (!count)
907	return 0;	1009	return 0;
908		1010
		1011	for (;;)
		1012	{
		1013	#ifdef __APPLE__
		1014	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1015	#endif
909	#if HAVE_SENDFILE	1016	#if HAVE_SENDFILE
910	# if __linux	1017	# if __linux
		1018	off_t soffset = offset;
911	res = sendfile (ofd, ifd, &offset, count);	1019	res = sendfile (ofd, ifd, &soffset, count);
912		1020
913	# elif __freebsd	1021	# elif __FreeBSD__
914	/*	1022	/*
915	* Of course, the freebsd sendfile is a dire hack with no thoughts	1023	* Of course, the freebsd sendfile is a dire hack with no thoughts
916	* wasted on making it similar to other I/O functions.	1024	* wasted on making it similar to other I/O functions.
917	*/	1025	*/
918	{
919	off_t sbytes;	1026	off_t sbytes;
920	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1027	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
921		1028
922	if (res < 0 && sbytes)	1029	#if 0 /* according to the manpage, this is correct, but broken behaviour */
923	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1030	/* freebsd' sendfile will return 0 on success */
		1031	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1032	/* not on e.g. EIO or EPIPE - sounds broken */
		1033	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
924	res = sbytes;	1034	res = sbytes;
925	}	1035	#endif
		1036
		1037	/* according to source inspection, this is correct, and useful behaviour */
		1038	if (sbytes)
		1039	res = sbytes;
		1040
		1041	# elif defined (__APPLE__)
		1042	off_t sbytes = count;
		1043	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1044
		1045	/* according to the manpage, sbytes is always valid */
		1046	if (sbytes)
		1047	res = sbytes;
926		1048
927	# elif __hpux	1049	# elif __hpux
928	res = sendfile (ofd, ifd, offset, count, 0, 0);	1050	res = sendfile (ofd, ifd, offset, count, 0, 0);
929		1051
930	# elif __solaris	1052	# elif __solaris
931	{
932	struct sendfilevec vec;	1053	struct sendfilevec vec;
933	size_t sbytes;	1054	size_t sbytes;
934		1055
935	vec.sfv_fd = ifd;	1056	vec.sfv_fd = ifd;
936	vec.sfv_flag = 0;	1057	vec.sfv_flag = 0;
937	vec.sfv_off = offset;	1058	vec.sfv_off = offset;
938	vec.sfv_len = count;	1059	vec.sfv_len = count;
939		1060
940	res = sendfilev (ofd, &vec, 1, &sbytes);	1061	res = sendfilev (ofd, &vec, 1, &sbytes);
941		1062
942	if (res < 0 && sbytes)	1063	if (res < 0 && sbytes)
943	res = sbytes;	1064	res = sbytes;
944	}
945		1065
946	# endif	1066	# endif
		1067
		1068	#elif defined (_WIN32) && 0
		1069	/* does not work, just for documentation of what would need to be done */
		1070	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1071	/* libeio guarantees that the file offset does not change, and windows */
		1072	/* has no way to get an independent handle to the same file description */
		1073	HANDLE h = TO_SOCKET (ifd);
		1074	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1075	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1076
947	#else	1077	#else
948	res = -1;	1078	res = -1;
949	errno = ENOSYS;	1079	errno = ENOSYS;
950	#endif	1080	#endif
951		1081
		1082	/* we assume sendfile can copy at least 128mb in one go */
		1083	if (res <= 128 * 1024 * 1024)
		1084	{
		1085	if (res > 0)
		1086	written += res;
		1087
		1088	if (written)
		1089	return written;
		1090
		1091	break;
		1092	}
		1093	else
		1094	{
		1095	/* if we requested more, then probably the kernel was lazy */
		1096	written += res;
		1097	offset += res;
		1098	count -= res;
		1099
		1100	if (!count)
		1101	return written;
		1102	}
		1103	}
		1104
952	if (res < 0	1105	if (res < 0
953	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1106	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1107	/* BSDs */
		1108	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1109	|| errno == ENOTSUP
		1110	#endif
		1111	#ifdef EOPNOTSUPP /* windows */
		1112	|| errno == EOPNOTSUPP /* BSDs */
		1113	#endif
954	#if __solaris	1114	#if __solaris
955	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1115	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
956	#endif	1116	#endif
957	)	1117	)
958	)	1118	)
…		…
962		1122
963	res = 0;	1123	res = 0;
964		1124
965	while (count)	1125	while (count)
966	{	1126	{
967	ssize_t cnt;	1127	eio_ssize_t cnt;
968		1128
969	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1129	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
970		1130
971	if (cnt <= 0)	1131	if (cnt <= 0)
972	{	1132	{
…		…
984		1144
985	offset += cnt;	1145	offset += cnt;
986	res += cnt;	1146	res += cnt;
987	count -= cnt;	1147	count -= cnt;
988	}	1148	}
		1149
		1150	FUBd;
989	}	1151	}
990		1152
991	return res;	1153	return res;
992	}	1154	}
993		1155
		1156	#ifdef PAGESIZE
		1157	# define eio_pagesize() PAGESIZE
		1158	#else
		1159	static intptr_t
		1160	eio_pagesize (void)
		1161	{
		1162	static intptr_t page;
		1163
		1164	if (!page)
		1165	page = sysconf (_SC_PAGESIZE);
		1166
		1167	return page;
		1168	}
		1169	#endif
		1170
		1171	static void
		1172	eio_page_align (void **addr, size_t *length)
		1173	{
		1174	intptr_t mask = eio_pagesize () - 1;
		1175
		1176	/* round down addr */
		1177	intptr_t adj = mask & (intptr_t)*addr;
		1178
		1179	*addr = (void *)((intptr_t)*addr - adj);
		1180	*length += adj;
		1181
		1182	/* round up length */
		1183	*length = (*length + mask) & ~mask;
		1184	}
		1185
		1186	#if !_POSIX_MEMLOCK
		1187	# define eio__mlockall(a) EIO_ENOSYS ()
		1188	#else
		1189
994	static int	1190	static int
		1191	eio__mlockall (int flags)
		1192	{
		1193	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1194	extern int mallopt (int, int);
		1195	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1196	#endif
		1197
		1198	if (EIO_MCL_CURRENT != MCL_CURRENT
		1199	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1200	{
		1201	flags = 0
		1202	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1203	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1204	}
		1205
		1206	return mlockall (flags);
		1207	}
		1208	#endif
		1209
		1210	#if !_POSIX_MEMLOCK_RANGE
		1211	# define eio__mlock(a,b) EIO_ENOSYS ()
		1212	#else
		1213
		1214	static int
		1215	eio__mlock (void *addr, size_t length)
		1216	{
		1217	eio_page_align (&addr, &length);
		1218
		1219	return mlock (addr, length);
		1220	}
		1221
		1222	#endif
		1223
		1224	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1225	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1226	#else
		1227
		1228	static int
		1229	eio__msync (void *mem, size_t len, int flags)
		1230	{
		1231	eio_page_align (&mem, &len);
		1232
		1233	if (EIO_MS_ASYNC != MS_SYNC
		1234	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1235	|| EIO_MS_SYNC != MS_SYNC)
		1236	{
		1237	flags = 0
		1238	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1239	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1240	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1241	}
		1242
		1243	return msync (mem, len, flags);
		1244	}
		1245
		1246	#endif
		1247
		1248	static int
		1249	eio__mtouch (eio_req *req)
		1250	{
		1251	void *mem = req->ptr2;
		1252	size_t len = req->size;
		1253	int flags = req->int1;
		1254
		1255	eio_page_align (&mem, &len);
		1256
		1257	{
		1258	intptr_t addr = (intptr_t)mem;
		1259	intptr_t end = addr + len;
		1260	intptr_t page = eio_pagesize ();
		1261
		1262	if (addr < end)
		1263	if (flags & EIO_MT_MODIFY) /* modify */
		1264	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1265	else
		1266	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1267	}
		1268
		1269	return 0;
		1270	}
		1271
		1272	/*****************************************************************************/
		1273	/* requests implemented outside eio_execute, because they are so large */
		1274
		1275	static void
		1276	eio__realpath (eio_req *req, etp_worker *self)
		1277	{
		1278	char *rel = req->ptr1;
		1279	char *res;
		1280	char *tmp1, *tmp2;
		1281	#if SYMLOOP_MAX > 32
		1282	int symlinks = SYMLOOP_MAX;
		1283	#else
		1284	int symlinks = 32;
		1285	#endif
		1286
		1287	req->result = -1;
		1288
		1289	errno = EINVAL;
		1290	if (!rel)
		1291	return;
		1292
		1293	errno = ENOENT;
		1294	if (!*rel)
		1295	return;
		1296
		1297	if (!req->ptr2)
		1298	{
		1299	X_LOCK (wrklock);
		1300	req->flags |= EIO_FLAG_PTR2_FREE;
		1301	X_UNLOCK (wrklock);
		1302	req->ptr2 = malloc (PATH_MAX * 3);
		1303
		1304	errno = ENOMEM;
		1305	if (!req->ptr2)
		1306	return;
		1307	}
		1308
		1309	res = req->ptr2;
		1310	tmp1 = res + PATH_MAX;
		1311	tmp2 = tmp1 + PATH_MAX;
		1312
		1313	#if 0 /* disabled, the musl way to do things is just too racy */
		1314	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1315	/* on linux we may be able to ask the kernel */
		1316	{
		1317	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1318
		1319	if (fd >= 0)
		1320	{
		1321	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1322	req->result = readlink (tmp1, res, PATH_MAX);
		1323	close (fd);
		1324
		1325	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1326
		1327	if (req->result > 0)
		1328	goto done;
		1329	}
		1330	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1331	return;
		1332	}
		1333	#endif
		1334	#endif
		1335
		1336	if (*rel != '/')
		1337	{
		1338	if (!getcwd (res, PATH_MAX))
		1339	return;
		1340
		1341	if (res [1]) /* only use if not / */
		1342	res += strlen (res);
		1343	}
		1344
		1345	while (*rel)
		1346	{
		1347	eio_ssize_t len, linklen;
		1348	char *beg = rel;
		1349
		1350	while (*rel && *rel != '/')
		1351	++rel;
		1352
		1353	len = rel - beg;
		1354
		1355	if (!len) /* skip slashes */
		1356	{
		1357	++rel;
		1358	continue;
		1359	}
		1360
		1361	if (beg [0] == '.')
		1362	{
		1363	if (len == 1)
		1364	continue; /* . - nop */
		1365
		1366	if (beg [1] == '.' && len == 2)
		1367	{
		1368	/* .. - back up one component, if possible */
		1369
		1370	while (res != req->ptr2)
		1371	if (*--res == '/')
		1372	break;
		1373
		1374	continue;
		1375	}
		1376	}
		1377
		1378	errno = ENAMETOOLONG;
		1379	if (res + 1 + len + 1 >= tmp1)
		1380	return;
		1381
		1382	/* copy one component */
		1383	*res = '/';
		1384	memcpy (res + 1, beg, len);
		1385
		1386	/* zero-terminate, for readlink */
		1387	res [len + 1] = 0;
		1388
		1389	/* now check if it's a symlink */
		1390	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1391
		1392	if (linklen < 0)
		1393	{
		1394	if (errno != EINVAL)
		1395	return;
		1396
		1397	/* it's a normal directory. hopefully */
		1398	res += len + 1;
		1399	}
		1400	else
		1401	{
		1402	/* yay, it was a symlink - build new path in tmp2 */
		1403	int rellen = strlen (rel);
		1404
		1405	errno = ENAMETOOLONG;
		1406	if (linklen + 1 + rellen >= PATH_MAX)
		1407	return;
		1408
		1409	errno = ELOOP;
		1410	if (!--symlinks)
		1411	return;
		1412
		1413	if (*tmp1 == '/')
		1414	res = req->ptr2; /* symlink resolves to an absolute path */
		1415
		1416	/* we need to be careful, as rel might point into tmp2 already */
		1417	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1418	tmp2 [linklen] = '/';
		1419	memcpy (tmp2, tmp1, linklen);
		1420
		1421	rel = tmp2;
		1422	}
		1423	}
		1424
		1425	/* special case for the lone root path */
		1426	if (res == req->ptr2)
		1427	*res++ = '/';
		1428
		1429	req->result = res - (char *)req->ptr2;
		1430
		1431	done:
		1432	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1433	}
		1434
		1435	static signed char
995	eio_dent_cmp (const void *a_, const void *b_)	1436	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
996	{	1437	{
997	const eio_dirent *a = (const eio_dirent *)a_;	1438	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
998	const eio_dirent *b = (const eio_dirent *)b_;	1439	: a->inode < b->inode ? -1
		1440	: a->inode > b->inode ? 1
		1441	: 0;
		1442	}
999		1443
1000	return (int)b->score - (int)a->score ? (int)b->score - (int)a->score	1444	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1001	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */	1445
		1446	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1447	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1448
		1449	static void
		1450	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1451	{
		1452	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
		1453	unsigned char *bit = bits;
		1454
		1455	assert (CHAR_BIT == 8);
		1456	assert (sizeof (eio_dirent) * 8 < 256);
		1457	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1458	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1459
		1460	if (size <= EIO_SORT_FAST)
		1461	return;
		1462
		1463	/* first prepare an array of bits to test in our radix sort */
		1464	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
		1465	/* inode_bits must contain all inodes ORed together */
		1466	/* which is used to skip bits that are 0 everywhere, which is very common */
		1467	{
		1468	eio_ino_t endianness;
		1469	int i, j;
		1470
		1471	/* we store the byte offset of byte n into byte n of "endianness" */
		1472	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1473	((unsigned char *)&endianness)[i] = i;
		1474
		1475	*bit++ = 0;
		1476
		1477	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1478	{
		1479	/* shifting off the byte offsets out of "endianness" */
		1480	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1481	endianness >>= 8;
		1482
		1483	for (j = 0; j < 8; ++j)
		1484	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
		1485	*bit++ = offs + j;
		1486	}
		1487
		1488	for (j = 0; j < 8; ++j)
		1489	if (score_bits & (1 << j))
		1490	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1491	}
		1492
		1493	/* now actually do the sorting (a variant of MSD radix sort) */
		1494	{
		1495	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
		1496	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
		1497	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
		1498	int stk_idx = 0;
		1499
		1500	base_stk [stk_idx] = dents;
		1501	end_stk [stk_idx] = dents + size;
		1502	bit_stk [stk_idx] = bit - 1;
		1503
		1504	do
		1505	{
		1506	base = base_stk [stk_idx];
		1507	end = end_stk [stk_idx];
		1508	bit = bit_stk [stk_idx];
		1509
		1510	for (;;)
		1511	{
		1512	unsigned char O = *bit >> 3;
		1513	unsigned char M = 1 << (*bit & 7);
		1514
		1515	eio_dirent *a = base;
		1516	eio_dirent *b = end;
		1517
		1518	if (b - a < EIO_SORT_CUTOFF)
		1519	break;
		1520
		1521	/* now bit-partition the array on the bit */
		1522	/* this ugly asymmetric loop seems to perform much better than typical */
		1523	/* partition algos found in the literature */
		1524	do
		1525	if (!(((unsigned char *)a)[O] & M))
		1526	++a;
		1527	else if (!(((unsigned char *)--b)[O] & M))
		1528	{
		1529	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1530	++a;
		1531	}
		1532	while (b > a);
		1533
		1534	/* next bit, or stop, if no bits left in this path */
		1535	if (!*--bit)
		1536	break;
		1537
		1538	base_stk [stk_idx] = a;
		1539	end_stk [stk_idx] = end;
		1540	bit_stk [stk_idx] = bit;
		1541	++stk_idx;
		1542
		1543	end = a;
		1544	}
		1545	}
		1546	while (stk_idx--);
		1547	}
		1548	}
		1549
		1550	static void
		1551	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1552	{
		1553	/* first move the smallest element to the front, to act as a sentinel */
		1554	{
		1555	int i;
		1556	eio_dirent *min = dents;
		1557
		1558	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1559	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1560	if (EIO_DENT_CMP (dents [i], <, *min))
		1561	min = &dents [i];
		1562
		1563	/* swap elements 0 and j (minimum) */
		1564	{
		1565	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1566	}
		1567	}
		1568
		1569	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1570	{
		1571	eio_dirent *i, *j;
		1572
		1573	for (i = dents + 1; i < dents + size; ++i)
		1574	{
		1575	eio_dirent value = *i;
		1576
		1577	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1578	j [1] = j [0];
		1579
		1580	j [1] = value;
		1581	}
		1582	}
		1583	}
		1584
		1585	static void
		1586	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1587	{
		1588	if (size <= 1)
		1589	return; /* our insertion sort relies on size > 0 */
		1590
		1591	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1592	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1593	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1594
		1595	/* use an insertion sort at the end, or for small arrays, */
		1596	/* as insertion sort is more efficient for small partitions */
		1597	eio_dent_insertion_sort (dents, size);
1002	}	1598	}
1003		1599
1004	/* read a full directory */	1600	/* read a full directory */
1005	static void	1601	static void
1006	eio__scandir (eio_req *req, etp_worker *self)	1602	eio__scandir (eio_req *req, etp_worker *self)
1007	{	1603	{
1008	DIR *dirp;
1009	EIO_STRUCT_DIRENT *entp;
1010	unsigned char *name, *names;	1604	char *name, *names;
1011	int namesalloc = 4096;	1605	int namesalloc = 4096 - sizeof (void *) * 4;
1012	int namesoffs = 0;	1606	int namesoffs = 0;
1013	int flags = req->int1;	1607	int flags = req->int1;
1014	eio_dirent *dents = 0;	1608	eio_dirent *dents = 0;
1015	int dentalloc = 128;	1609	int dentalloc = 128;
1016	int dentoffs = 0;	1610	int dentoffs = 0;
		1611	eio_ino_t inode_bits = 0;
		1612	#ifdef _WIN32
		1613	HANDLE dirp;
		1614	WIN32_FIND_DATA entp;
		1615	#else
		1616	DIR *dirp;
		1617	EIO_STRUCT_DIRENT *entp;
		1618	#endif
1017		1619
1018	req->result = -1;	1620	req->result = -1;
1019		1621
1020	if (!(flags & EIO_READDIR_DENTS))	1622	if (!(flags & EIO_READDIR_DENTS))
1021	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1623	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1022		1624
1023	X_LOCK (wrklock);	1625	#ifdef _WIN32
1024	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1626	{
		1627	int len = strlen ((const char *)req->ptr1);
		1628	char *path = malloc (MAX_PATH);
		1629	const char *fmt;
		1630
		1631	if (!len)
		1632	fmt = "./*";
		1633	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1634	fmt = "%s*";
		1635	else
		1636	fmt = "%s/*";
		1637
		1638	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1639	dirp = FindFirstFile (path, &entp);
		1640	free (path);
		1641
		1642	if (dirp == INVALID_HANDLE_VALUE)
		1643	{
		1644	dirp = 0;
		1645
		1646	switch (GetLastError ())
		1647	{
		1648	case ERROR_FILE_NOT_FOUND:
		1649	req->result = 0;
		1650	break;
		1651
		1652	case ERROR_INVALID_NAME:
		1653	case ERROR_PATH_NOT_FOUND:
		1654	case ERROR_NO_MORE_FILES:
		1655	errno = ENOENT;
		1656	break;
		1657
		1658	case ERROR_NOT_ENOUGH_MEMORY:
		1659	errno = ENOMEM;
		1660	break;
		1661
		1662	default:
		1663	errno = EINVAL;
		1664	break;
		1665	}
		1666	}
		1667	}
		1668	#else
1025	self->dirp = dirp = opendir (req->ptr1);	1669	dirp = opendir (req->ptr1);
		1670	#endif
		1671
		1672	if (req->flags & EIO_FLAG_PTR1_FREE)
		1673	free (req->ptr1);
		1674
1026	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1675	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1027	req->ptr1 = names = malloc (namesalloc);
1028	req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1676	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1029	X_UNLOCK (wrklock);	1677	req->ptr2 = names = malloc (namesalloc);
1030		1678
1031	if (dirp && names && (!flags || dents))	1679	if (dirp && names && (!flags || dents))
1032	for (;;)	1680	for (;;)
1033	{	1681	{
		1682	int done;
		1683
		1684	#ifdef _WIN32
		1685	done = !dirp;
		1686	#else
1034	errno = 0;	1687	errno = 0;
1035	entp = readdir (dirp);	1688	entp = readdir (dirp);
		1689	done = !entp;
		1690	#endif
1036		1691
1037	if (!entp)	1692	if (done)
1038	{	1693	{
		1694	#ifndef _WIN32
		1695	int old_errno = errno;
		1696	closedir (dirp);
		1697	errno = old_errno;
		1698
1039	if (errno)	1699	if (errno)
1040	break;	1700	break;
		1701	#endif
1041		1702
1042	/* sort etc. */	1703	/* sort etc. */
1043	req->int1 = flags;	1704	req->int1 = flags;
1044	req->result = dentoffs;	1705	req->result = dentoffs;
1045		1706
1046	if (dents)
1047	{
1048	eio_dirent *ent = dents + dentoffs;
1049
1050	while (ent > dents)
1051	(--ent)->name = names + (size_t)ent->name;
1052	}
1053
1054	if (flags & EIO_READDIR_STAT_ORDER	1707	if (flags & EIO_READDIR_STAT_ORDER)
1055	|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))	1708	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1056	{
1057	/* pray your qsort doesn't use quicksort */
1058	qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1059	}
1060	else if (flags & EIO_READDIR_DIRS_FIRST)	1709	else if (flags & EIO_READDIR_DIRS_FIRST)
		1710	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1711	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1712	else
1061	{	1713	{
1062	/* in this case, all is known, and we just put dirs first and sort them */	1714	/* in this case, all is known, and we just put dirs first and sort them */
1063	eio_dirent *ent = dents + dentoffs;	1715	eio_dirent *oth = dents + dentoffs;
1064	eio_dirent *dir = dents;	1716	eio_dirent *dir = dents;
1065		1717
		1718	/* now partition dirs to the front, and non-dirs to the back */
		1719	/* by walking from both sides and swapping if necessary */
1066	while (ent > dir)	1720	while (oth > dir)
1067	{	1721	{
1068	if (dir->type == DT_DIR)	1722	if (dir->type == EIO_DT_DIR)
1069	++dir;	1723	++dir;
1070	else
1071	{
1072	--ent;
1073
1074	if (ent->type == DT_DIR)	1724	else if ((--oth)->type == EIO_DT_DIR)
1075	{	1725	{
1076	eio_dirent tmp = *dir;	1726	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1077	*dir = *ent;
1078	*ent = tmp;
1079		1727
1080	++dir;	1728	++dir;
1081	}	1729	}
1082	}	1730	}
		1731
		1732	/* now sort the dirs only (dirs all have the same score) */
		1733	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1083	}	1734	}
1084
1085	/* now sort the dirs only */
1086	qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1087	}
1088
1089	/* only provide the names array unless DENTS is specified */
1090	if (!(flags & EIO_READDIR_DENTS))
1091	{
1092	X_LOCK (wrklock);
1093	assert (!dents);
1094	req->ptr1 = 0;
1095	req->ptr2 = names;
1096	X_UNLOCK (wrklock);
1097	}
1098		1735
1099	break;	1736	break;
1100	}	1737	}
1101		1738
1102	/* now add the entry to our list(s) */	1739	/* now add the entry to our list(s) */
1103	name = entp->d_name;	1740	name = D_NAME (entp);
1104		1741
1105	/* skip . and .. entries */	1742	/* skip . and .. entries */
1106	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1743	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1107	{	1744	{
1108	int len = D_NAMLEN (entp) + 1;	1745	int len = D_NAMLEN (entp) + 1;
1109		1746
1110	while (expect_false (namesoffs + len > namesalloc))	1747	while (ecb_expect_false (namesoffs + len > namesalloc))
1111	{	1748	{
1112	namesalloc *= 2;	1749	namesalloc *= 2;
1113	X_LOCK (wrklock);
1114	req->ptr1 = names = realloc (names, namesalloc);	1750	req->ptr2 = names = realloc (names, namesalloc);
1115	X_UNLOCK (wrklock);
1116		1751
1117	if (!names)	1752	if (!names)
1118	break;	1753	break;
1119	}	1754	}
1120		1755
…		…
1122		1757
1123	if (dents)	1758	if (dents)
1124	{	1759	{
1125	struct eio_dirent *ent;	1760	struct eio_dirent *ent;
1126		1761
1127	if (expect_false (dentoffs == dentalloc))	1762	if (ecb_expect_false (dentoffs == dentalloc))
1128	{	1763	{
1129	dentalloc *= 2;	1764	dentalloc *= 2;
1130	X_LOCK (wrklock);
1131	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1765	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1132	X_UNLOCK (wrklock);
1133		1766
1134	if (!dents)	1767	if (!dents)
1135	break;	1768	break;
1136	}	1769	}
1137		1770
1138	ent = dents + dentoffs;	1771	ent = dents + dentoffs;
1139		1772
1140	ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */	1773	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1141	ent->namelen = len - 1;	1774	ent->namelen = len - 1;
1142	ent->inode = D_INO (entp);	1775	ent->inode = D_INO (entp);
		1776
		1777	inode_bits |= ent->inode;
1143		1778
1144	switch (D_TYPE (entp))	1779	switch (D_TYPE (entp))
1145	{	1780	{
1146	default:	1781	default:
1147	ent->type = EIO_DT_UNKNOWN;	1782	ent->type = EIO_DT_UNKNOWN;
…		…
1190	#ifdef DT_WHT	1825	#ifdef DT_WHT
1191	case DT_WHT: ent->type = EIO_DT_WHT; break;	1826	case DT_WHT: ent->type = EIO_DT_WHT; break;
1192	#endif	1827	#endif
1193	}	1828	}
1194		1829
1195	ent->score = 0;	1830	ent->score = 7;
1196		1831
1197	if (flags & EIO_READDIR_DIRS_FIRST)	1832	if (flags & EIO_READDIR_DIRS_FIRST)
1198	{	1833	{
1199	if (ent->type == EIO_DT_UNKNOWN)	1834	if (ent->type == EIO_DT_UNKNOWN)
1200	{	1835	{
1201	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */	1836	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1202	ent->score = 98;	1837	ent->score = 1;
1203	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */	1838	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1204	ent->score = len <= 2 ? len + 60 : len <= 4 ? 50 : len <= 7 ? 40 : 10; /* shorter == more likely dir, but avoid too many classes */	1839	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1205	}	1840	}
1206	else if (ent->type == EIO_DT_DIR)	1841	else if (ent->type == EIO_DT_DIR)
1207	ent->score = 100;	1842	ent->score = 0;
1208	}	1843	}
1209	}	1844	}
1210		1845
1211	namesoffs += len;	1846	namesoffs += len;
1212	++dentoffs;	1847	++dentoffs;
1213	}	1848	}
		1849
		1850	if (EIO_CANCELLED (req))
		1851	{
		1852	errno = ECANCELED;
		1853	break;
		1854	}
		1855
		1856	#ifdef _WIN32
		1857	if (!FindNextFile (dirp, &entp))
		1858	{
		1859	FindClose (dirp);
		1860	dirp = 0;
		1861	}
		1862	#endif
1214	}	1863	}
1215	else
1216	req->result = -1;
1217	}
1218
1219	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1220	# undef msync
1221	# define msync(a,b,c) ((errno = ENOSYS), -1)
1222	#endif
1223
1224	int
1225	eio__mtouch (void *mem, size_t len, int flags)
1226	{
1227	intptr_t addr = (intptr_t)mem;
1228	intptr_t end = addr + len;
1229	#ifdef PAGESIZE
1230	const intptr_t page = PAGESIZE;
1231	#else
1232	static intptr_t page;
1233
1234	if (!page)
1235	page = sysconf (_SC_PAGESIZE);
1236	#endif
1237
1238	addr &= ~(page - 1); /* assume page size is always a power of two */
1239
1240	if (addr < end)
1241	if (flags) /* modify */
1242	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1243	else
1244	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1245
1246	return 0;
1247	}	1864	}
1248		1865
1249	/*****************************************************************************/	1866	/*****************************************************************************/
1250		1867
1251	#define ALLOC(len) \	1868	#define ALLOC(len) \
…		…
1267	{	1884	{
1268	ETP_REQ *req;	1885	ETP_REQ *req;
1269	struct timespec ts;	1886	struct timespec ts;
1270	etp_worker *self = (etp_worker *)thr_arg;	1887	etp_worker *self = (etp_worker *)thr_arg;
1271		1888
1272	/* try to distribute timeouts somewhat randomly */	1889	/* try to distribute timeouts somewhat evenly */
1273	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1890	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1274		1891
1275	for (;;)	1892	for (;;)
1276	{	1893	{
		1894	ts.tv_sec = 0;
		1895
1277	X_LOCK (reqlock);	1896	X_LOCK (reqlock);
1278		1897
1279	for (;;)	1898	for (;;)
1280	{	1899	{
1281	self->req = req = reqq_shift (&req_queue);	1900	self->req = req = reqq_shift (&req_queue);
1282		1901
1283	if (req)	1902	if (req)
1284	break;	1903	break;
1285		1904
		1905	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1906	{
		1907	X_UNLOCK (reqlock);
		1908	X_LOCK (wrklock);
		1909	--started;
		1910	X_UNLOCK (wrklock);
		1911	goto quit;
		1912	}
		1913
1286	++idle;	1914	++idle;
1287		1915
1288	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1916	if (idle <= max_idle)
1289	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1917	/* we are allowed to idle, so do so without any timeout */
		1918	X_COND_WAIT (reqwait, reqlock);
		1919	else
1290	{	1920	{
1291	if (idle > max_idle)	1921	/* initialise timeout once */
1292	{	1922	if (!ts.tv_sec)
1293	--idle;	1923	ts.tv_sec = time (0) + idle_timeout;
1294	X_UNLOCK (reqlock);
1295	X_LOCK (wrklock);
1296	--started;
1297	X_UNLOCK (wrklock);
1298	goto quit;
1299	}
1300		1924
1301	/* we are allowed to idle, so do so without any timeout */
1302	X_COND_WAIT (reqwait, reqlock);	1925	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1926	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1303	}	1927	}
1304		1928
1305	--idle;	1929	--idle;
1306	}	1930	}
1307		1931
…		…
1310	X_UNLOCK (reqlock);	1934	X_UNLOCK (reqlock);
1311		1935
1312	if (req->type < 0)	1936	if (req->type < 0)
1313	goto quit;	1937	goto quit;
1314		1938
1315	if (!EIO_CANCELLED (req))
1316	ETP_EXECUTE (self, req);	1939	ETP_EXECUTE (self, req);
1317		1940
1318	X_LOCK (reslock);	1941	X_LOCK (reslock);
1319		1942
1320	++npending;	1943	++npending;
1321		1944
…		…
1336	return 0;	1959	return 0;
1337	}	1960	}
1338		1961
1339	/*****************************************************************************/	1962	/*****************************************************************************/
1340		1963
		1964	int ecb_cold
1341	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1965	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1342	{	1966	{
		1967	#if !HAVE_PREADWRITE
		1968	X_MUTEX_CREATE (preadwritelock);
		1969	#endif
		1970
1343	return etp_init (want_poll, done_poll);	1971	return etp_init (want_poll, done_poll);
1344	}	1972	}
1345		1973
		1974	ecb_inline void
1346	static void eio_api_destroy (eio_req *req)	1975	eio_api_destroy (eio_req *req)
1347	{	1976	{
1348	free (req);	1977	free (req);
1349	}	1978	}
1350		1979
1351	#define REQ(rtype) \	1980	#define REQ(rtype) \
…		…
1370	{ \	1999	{ \
1371	eio_api_destroy (req); \	2000	eio_api_destroy (req); \
1372	return 0; \	2001	return 0; \
1373	}	2002	}
1374		2003
		2004	static void
1375	static void eio_execute (etp_worker *self, eio_req *req)	2005	eio_execute (etp_worker *self, eio_req *req)
1376	{	2006	{
1377	errno = 0;	2007	if (ecb_expect_false (EIO_CANCELLED (req)))
		2008	{
		2009	req->result = -1;
		2010	req->errorno = ECANCELED;
		2011	return;
		2012	}
1378		2013
1379	switch (req->type)	2014	switch (req->type)
1380	{	2015	{
1381	case EIO_READ: ALLOC (req->size);	2016	case EIO_READ: ALLOC (req->size);
1382	req->result = req->offs >= 0	2017	req->result = req->offs >= 0
…		…
1385	case EIO_WRITE: req->result = req->offs >= 0	2020	case EIO_WRITE: req->result = req->offs >= 0
1386	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2021	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1387	: write (req->int1, req->ptr2, req->size); break;	2022	: write (req->int1, req->ptr2, req->size); break;
1388		2023
1389	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2024	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1390	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2025	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1391		2026
1392	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2027	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1393	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2028	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1394	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2029	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1395	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2030	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1396	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2031	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1397	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2032	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2033
		2034	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2035	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2036	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2037	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1398		2038
1399	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	2039	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1400	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	2040	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1401	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	2041	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1402	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	2042	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
…		…
1410	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	2050	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1411	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	2051	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1412	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2052	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1413	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2053	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1414	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2054	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1415	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;	2055	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1416		2056
		2057	case EIO_REALPATH: eio__realpath (req, self); break;
		2058
1417	case EIO_READLINK: ALLOC (NAME_MAX);	2059	case EIO_READLINK: ALLOC (PATH_MAX);
1418	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	2060	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1419		2061
1420	case EIO_SYNC: req->result = 0; sync (); break;	2062	case EIO_SYNC: req->result = 0; sync (); break;
1421	case EIO_FSYNC: req->result = fsync (req->int1); break;	2063	case EIO_FSYNC: req->result = fsync (req->int1); break;
1422	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2064	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1423	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;	2065	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2066	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1424	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	2067	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2068	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1425	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	2069	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2070	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1426		2071
1427	case EIO_READDIR: eio__scandir (req, self); break;	2072	case EIO_READDIR: eio__scandir (req, self); break;
1428		2073
1429	case EIO_BUSY:	2074	case EIO_BUSY:
1430	#ifdef _WIN32	2075	#ifdef _WIN32
1431	Sleep (req->nv1 * 1000.);	2076	Sleep (req->nv1 * 1e3);
1432	#else	2077	#else
1433	{	2078	{
1434	struct timeval tv;	2079	struct timeval tv;
1435		2080
1436	tv.tv_sec = req->nv1;	2081	tv.tv_sec = req->nv1;
1437	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	2082	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1438		2083
1439	req->result = select (0, 0, 0, 0, &tv);	2084	req->result = select (0, 0, 0, 0, &tv);
1440	}	2085	}
1441	#endif	2086	#endif
1442	break;	2087	break;
…		…
1457	times = tv;	2102	times = tv;
1458	}	2103	}
1459	else	2104	else
1460	times = 0;	2105	times = 0;
1461		2106
1462
1463	req->result = req->type == EIO_FUTIME	2107	req->result = req->type == EIO_FUTIME
1464	? futimes (req->int1, times)	2108	? futimes (req->int1, times)
1465	: utimes (req->ptr1, times);	2109	: utimes (req->ptr1, times);
1466	}	2110	}
1467	break;	2111	break;
…		…
1472	case EIO_NOP:	2116	case EIO_NOP:
1473	req->result = 0;	2117	req->result = 0;
1474	break;	2118	break;
1475		2119
1476	case EIO_CUSTOM:	2120	case EIO_CUSTOM:
1477	((void (*)(eio_req *))req->feed) (req);	2121	req->feed (req);
1478	break;	2122	break;
1479		2123
1480	default:	2124	default:
		2125	errno = ENOSYS;
1481	req->result = -1;	2126	req->result = -1;
1482	break;	2127	break;
1483	}	2128	}
1484		2129
1485	req->errorno = errno;	2130	req->errorno = errno;
…		…
1515	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2160	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1516	{	2161	{
1517	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2162	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1518	}	2163	}
1519		2164
		2165	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2166	{
		2167	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2168	}
		2169
		2170	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2171	{
		2172	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2173	}
		2174
1520	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2175	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1521	{	2176	{
1522	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2177	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1523	}	2178	}
1524		2179
		2180	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2181	{
		2182	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2183	}
		2184
1525	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2185	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1526	{	2186	{
1527	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2187	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1528	}	2188	}
1529		2189
…		…
1550	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2210	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1551	{	2211	{
1552	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2212	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1553	}	2213	}
1554		2214
		2215	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2216	{
		2217	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2218	}
		2219
1555	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2220	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1556	{	2221	{
1557	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2222	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1558	}	2223	}
1559		2224
…		…
1565	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2230	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1566	{	2231	{
1567	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2232	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1568	}	2233	}
1569		2234
1570	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2235	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1571	{	2236	{
1572	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2237	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1573	}	2238	}
1574		2239
1575	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2240	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1595	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2260	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1596	{	2261	{
1597	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2262	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1598	}	2263	}
1599		2264
1600	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2265	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1601	{	2266	{
1602	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2267	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1603	}	2268	}
1604		2269
1605	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2270	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1621	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2286	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1622	{	2287	{
1623	return eio__1path (EIO_READLINK, path, pri, cb, data);	2288	return eio__1path (EIO_READLINK, path, pri, cb, data);
1624	}	2289	}
1625		2290
		2291	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2292	{
		2293	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2294	}
		2295
1626	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2296	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1627	{	2297	{
1628	return eio__1path (EIO_STAT, path, pri, cb, data);	2298	return eio__1path (EIO_STAT, path, pri, cb, data);
1629	}	2299	}
1630		2300
1631	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2301	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1632	{	2302	{
1633	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2303	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1634	}	2304	}
1635		2305
		2306	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2307	{
		2308	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2309	}
		2310
1636	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2311	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1637	{	2312	{
1638	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2313	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1639	}	2314	}
1640		2315
…		…
1648	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2323	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1649	}	2324	}
1650		2325
1651	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2326	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1652	{	2327	{
1653	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2328	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1654	}	2329	}
1655		2330
1656	static eio_req *	2331	static eio_req *
1657	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2332	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1658	{	2333	{
…		…
1682	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2357	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1683	{	2358	{
1684	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2359	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1685	}	2360	}
1686		2361
1687	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2362	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1688	{	2363	{
1689	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2364	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1690	}	2365	}
1691		2366
1692	#endif	2367	#endif
1693		2368
1694	eio_req *eio_grp (eio_cb cb, void *data)	2369	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1703	#undef SEND	2378	#undef SEND
1704		2379
1705	/*****************************************************************************/	2380	/*****************************************************************************/
1706	/* grp functions */	2381	/* grp functions */
1707		2382
		2383	void
1708	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2384	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1709	{	2385	{
1710	grp->int2 = limit;	2386	grp->int2 = limit;
1711	grp->feed = feed;	2387	grp->feed = feed;
1712		2388
1713	grp_try_feed (grp);	2389	grp_try_feed (grp);
1714	}	2390	}
1715		2391
		2392	void
1716	void eio_grp_limit (eio_req *grp, int limit)	2393	eio_grp_limit (eio_req *grp, int limit)
1717	{	2394	{
1718	grp->int2 = limit;	2395	grp->int2 = limit;
1719		2396
1720	grp_try_feed (grp);	2397	grp_try_feed (grp);
1721	}	2398	}
1722		2399
		2400	void
1723	void eio_grp_add (eio_req *grp, eio_req *req)	2401	eio_grp_add (eio_req *grp, eio_req *req)
1724	{	2402	{
1725	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2403	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1726		2404
1727	grp->flags |= EIO_FLAG_GROUPADD;	2405	grp->flags |= EIO_FLAG_GROUPADD;
1728		2406
…		…
1739	}	2417	}
1740		2418
1741	/*****************************************************************************/	2419	/*****************************************************************************/
1742	/* misc garbage */	2420	/* misc garbage */
1743		2421
		2422	eio_ssize_t
1744	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2423	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1745	{	2424	{
1746	etp_worker wrk;
1747
1748	wrk.dbuf = 0;
1749
1750	eio__sendfile (ofd, ifd, offset, count, &wrk);	2425	return eio__sendfile (ofd, ifd, offset, count);
1751
1752	if (wrk.dbuf)
1753	free (wrk.dbuf);
1754	}	2426	}
1755		2427

Diff Legend

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