ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/cvsroot/libeio/eio.c

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.38 by root, Fri Jun 12 20:01:42 2009 UTC vs.
Revision 1.97 by root, Sun Jul 24 03:32:54 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	#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
		77	# define ENOTSOCK WSAENOTSOCK
		78	#endif
		79
		80	static void eio_destroy (eio_req *req);
		81
58	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
59	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	83	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60	#endif	84	#endif
61		85
62	#ifndef EIO_DESTROY	86	#ifndef EIO_DESTROY
…		…
65		89
66	#ifndef EIO_FEED	90	#ifndef EIO_FEED
67	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	91	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68	#endif	92	#endif
69		93
		94	#ifndef EIO_FD_TO_WIN32_HANDLE
		95	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		96	#endif
		97	#ifndef EIO_WIN32_HANDLE_TO_FD
		98	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		99	#endif
		100
		101	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		102
		103	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		104
70	#ifdef _WIN32	105	#ifdef _WIN32
71		106
72	/*doh*/	107	#undef PAGESIZE
		108	#define PAGESIZE 4096 /* GetSystemInfo? */
		109
		110	#ifdef EIO_STRUCT_STATI64
		111	#define stat(path,buf) _stati64 (path,buf)
		112	#define fstat(fd,buf) _fstati64 (fd,buf)
		113	#endif
		114	#define lstat(path,buf) stat (path,buf)
		115	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		116	#define mkdir(path,mode) _mkdir (path)
		117	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		118
		119	#define chmod(path,mode) _chmod (path, mode)
		120	#define dup(fd) _dup (fd)
		121	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		122
		123	#define fchmod(fd,mode) EIO_ENOSYS ()
		124	#define chown(path,uid,gid) EIO_ENOSYS ()
		125	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		126	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		127	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		128	#define mknod(path,mode,dev) EIO_ENOSYS ()
		129	#define sync() EIO_ENOSYS ()
		130	#define readlink(path,buf,s) EIO_ENOSYS ()
		131	#define statvfs(path,buf) EIO_ENOSYS ()
		132	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		133
		134	/* we could even stat and see if it exists */
		135	static int
		136	symlink (const char *old, const char *neu)
		137	{
		138	#if WINVER >= 0x0600
		139	if (CreateSymbolicLink (neu, old, 1))
		140	return 0;
		141
		142	if (CreateSymbolicLink (neu, old, 0))
		143	return 0;
		144	#endif
		145
		146	return EIO_ERRNO (ENOENT, -1);
		147	}
		148
		149	/* POSIX API only */
		150	#define CreateHardLink(neu,old,flags) 0
		151	#define CreateSymbolicLink(neu,old,flags) 0
		152
		153	struct statvfs
		154	{
		155	int dummy;
		156	};
		157
		158	#define DT_DIR EIO_DT_DIR
		159	#define DT_REG EIO_DT_REG
		160	#define D_NAME(entp) entp.cFileName
		161	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
		162
73	#else	163	#else
74		164
75	# include "config.h"
76	# include <sys/time.h>	165	#include <sys/time.h>
77	# include <sys/select.h>	166	#include <sys/select.h>
78	# include <sys/mman.h>	167	#include <sys/statvfs.h>
79	# include <unistd.h>	168	#include <unistd.h>
		169	#include <signal.h>
		170	#include <dirent.h>
		171
		172	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		173	#include <sys/mman.h>
		174	#endif
		175
		176	#define D_NAME(entp) entp->d_name
		177
		178	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		179	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		180	#define _DIRENT_HAVE_D_TYPE /* sigh */
		181	#define D_INO(de) (de)->d_fileno
		182	#define D_NAMLEN(de) (de)->d_namlen
		183	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		184	#define D_INO(de) (de)->d_ino
		185	#endif
		186
		187	#ifdef _D_EXACT_NAMLEN
		188	#undef D_NAMLEN
		189	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		190	#endif
		191
		192	#ifdef _DIRENT_HAVE_D_TYPE
		193	#define D_TYPE(de) (de)->d_type
		194	#endif
		195
		196	#ifndef EIO_STRUCT_DIRENT
		197	#define EIO_STRUCT_DIRENT struct dirent
		198	#endif
		199
		200	#endif
		201
		202	#if HAVE_UTIMES
80	# include <utime.h>	203	# include <utime.h>
81	# include <signal.h>
82	# include <dirent.h>
83
84	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85	# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
86	# define _DIRENT_HAVE_D_TYPE /* sigh */
87	# define D_INO(de) (de)->d_fileno
88	# define D_NAMLEN(de) (de)->d_namlen
89	# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
90	# define D_INO(de) (de)->d_ino
91	# endif
92
93	#ifdef _D_EXACT_NAMLEN
94	# undef D_NAMLEN
95	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96	#endif
97
98	# ifdef _DIRENT_HAVE_D_TYPE
99	# define D_TYPE(de) (de)->d_type
100	# endif
101
102	# ifndef EIO_STRUCT_DIRENT
103	# define EIO_STRUCT_DIRENT struct dirent
104	# endif
105
106	#endif	204	#endif
107		205
108	#if HAVE_SENDFILE	206	#if HAVE_SENDFILE
109	# if __linux	207	# if __linux
110	# include <sys/sendfile.h>	208	# include <sys/sendfile.h>
111	# elif __freebsd	209	# elif __FreeBSD__ || defined __APPLE__
112	# include <sys/socket.h>	210	# include <sys/socket.h>
113	# include <sys/uio.h>	211	# include <sys/uio.h>
114	# elif __hpux	212	# elif __hpux
115	# include <sys/socket.h>	213	# include <sys/socket.h>
116	# elif __solaris /* not yet */	214	# elif __solaris
117	# include <sys/sendfile.h>	215	# include <sys/sendfile.h>
118	# else	216	# else
119	# error sendfile support requested but not available	217	# error sendfile support requested but not available
120	# endif	218	# endif
121	#endif	219	#endif
122		220
		221	#if HAVE_SYS_SYSCALL_H
		222	# include <sys/syscall.h>
		223	#endif
		224
123	#ifndef D_TYPE	225	#ifndef D_TYPE
124	# define D_TYPE(de) 0	226	# define D_TYPE(de) 0
125	#endif	227	#endif
126	#ifndef D_INO	228	#ifndef D_INO
127	# define D_INO(de) 0	229	# define D_INO(de) 0
128	#endif	230	#endif
129	#ifndef D_NAMLEN	231	#ifndef D_NAMLEN
130	# define D_NAMLEN(de) strlen ((de)->d_name)	232	# define D_NAMLEN(entp) strlen (D_NAME (entp))
131	#endif	233	#endif
132
133	/* number of seconds after which an idle threads exit */
134	#define IDLE_TIMEOUT 10
135		234
136	/* used for struct dirent, AIX doesn't provide it */	235	/* used for struct dirent, AIX doesn't provide it */
137	#ifndef NAME_MAX	236	#ifndef NAME_MAX
138	# define NAME_MAX 4096	237	# define NAME_MAX 4096
139	#endif	238	#endif
140		239
		240	/* used for readlink etc. */
		241	#ifndef PATH_MAX
		242	# define PATH_MAX 4096
		243	#endif
		244
141	/* buffer size for various temporary buffers */	245	/* buffer size for various temporary buffers */
142	#define EIO_BUFSIZE 65536	246	#define EIO_BUFSIZE 65536
143		247
144	#define dBUF \	248	#define dBUF \
145	char *eio_buf; \
146	ETP_WORKER_LOCK (self); \
147	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	249	char *eio_buf = malloc (EIO_BUFSIZE); \
148	ETP_WORKER_UNLOCK (self); \
149	errno = ENOMEM; \	250	errno = ENOMEM; \
150	if (!eio_buf) \	251	if (!eio_buf) \
151	return -1;	252	return -1
		253
		254	#define FUBd \
		255	free (eio_buf)
152		256
153	#define EIO_TICKS ((1000000 + 1023) >> 10)	257	#define EIO_TICKS ((1000000 + 1023) >> 10)
154
155	/*****************************************************************************/
156
157	#if __GNUC__ >= 3
158	# define expect(expr,value) __builtin_expect ((expr),(value))
159	#else
160	# define expect(expr,value) (expr)
161	#endif
162
163	#define expect_false(expr) expect ((expr) != 0, 0)
164	#define expect_true(expr) expect ((expr) != 0, 1)
165
166	/*****************************************************************************/
167		258
168	#define ETP_PRI_MIN EIO_PRI_MIN	259	#define ETP_PRI_MIN EIO_PRI_MIN
169	#define ETP_PRI_MAX EIO_PRI_MAX	260	#define ETP_PRI_MAX EIO_PRI_MAX
170		261
171	struct etp_worker;	262	struct etp_worker;
…		…
175	static int eio_finish (eio_req *req);	266	static int eio_finish (eio_req *req);
176	#define ETP_FINISH(req) eio_finish (req)	267	#define ETP_FINISH(req) eio_finish (req)
177	static void eio_execute (struct etp_worker *self, eio_req *req);	268	static void eio_execute (struct etp_worker *self, eio_req *req);
178	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	269	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
179		270
180	#define ETP_WORKER_CLEAR(req) \
181	if (wrk->dbuf) \
182	{ \
183	free (wrk->dbuf); \
184	wrk->dbuf = 0; \
185	} \
186	\
187	if (wrk->dirp) \
188	{ \
189	closedir (wrk->dirp); \
190	wrk->dirp = 0; \
191	}
192
193	#define ETP_WORKER_COMMON \
194	void *dbuf; \
195	DIR *dirp;
196
197	/*****************************************************************************/	271	/*****************************************************************************/
198		272
199	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	273	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200		274
201	/* calculcate time difference in ~1/EIO_TICKS of a second */	275	/* calculate time difference in ~1/EIO_TICKS of a second */
		276	ecb_inline int
202	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	277	tvdiff (struct timeval *tv1, struct timeval *tv2)
203	{	278	{
204	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	279	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	280	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
206	}	281	}
207		282
…		…
214	static unsigned int max_poll_reqs; /* reslock */	289	static unsigned int max_poll_reqs; /* reslock */
215		290
216	static volatile unsigned int nreqs; /* reqlock */	291	static volatile unsigned int nreqs; /* reqlock */
217	static volatile unsigned int nready; /* reqlock */	292	static volatile unsigned int nready; /* reqlock */
218	static volatile unsigned int npending; /* reqlock */	293	static volatile unsigned int npending; /* reqlock */
219	static volatile unsigned int max_idle = 4;	294	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		295	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
220		296
221	static mutex_t wrklock = X_MUTEX_INIT;	297	static xmutex_t wrklock;
222	static mutex_t reslock = X_MUTEX_INIT;	298	static xmutex_t reslock;
223	static mutex_t reqlock = X_MUTEX_INIT;	299	static xmutex_t reqlock;
224	static cond_t reqwait = X_COND_INIT;	300	static xcond_t reqwait;
225		301
226	#if !HAVE_PREADWRITE	302	#if !HAVE_PREADWRITE
227	/*	303	/*
228	* make our pread/pwrite emulation safe against themselves, but not against	304	* make our pread/pwrite emulation safe against themselves, but not against
229	* normal read/write by using a mutex. slows down execution a lot,	305	* normal read/write by using a mutex. slows down execution a lot,
230	* but that's your problem, not mine.	306	* but that's your problem, not mine.
231	*/	307	*/
232	static mutex_t preadwritelock = X_MUTEX_INIT;	308	static xmutex_t preadwritelock;
233	#endif	309	#endif
234		310
235	typedef struct etp_worker	311	typedef struct etp_worker
236	{	312	{
237	/* locked by wrklock */	313	/* locked by wrklock */
238	struct etp_worker *prev, *next;	314	struct etp_worker *prev, *next;
239		315
240	thread_t tid;	316	xthread_t tid;
241		317
242	/* locked by reslock, reqlock or wrklock */	318	/* locked by reslock, reqlock or wrklock */
243	ETP_REQ *req; /* currently processed request */	319	ETP_REQ *req; /* currently processed request */
244		320
		321	#ifdef ETP_WORKER_COMMON
245	ETP_WORKER_COMMON	322	ETP_WORKER_COMMON
		323	#endif
246	} etp_worker;	324	} etp_worker;
247		325
248	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	326	static etp_worker wrk_first; /* NOT etp */
249		327
250	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	328	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
251	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	329	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
252		330
253	/* worker threads management */	331	/* worker threads management */
254		332
		333	static void ecb_cold
255	static void etp_worker_clear (etp_worker *wrk)	334	etp_worker_clear (etp_worker *wrk)
256	{	335	{
257	ETP_WORKER_CLEAR (wrk);
258	}	336	}
259		337
		338	static void ecb_cold
260	static void etp_worker_free (etp_worker *wrk)	339	etp_worker_free (etp_worker *wrk)
261	{	340	{
262	wrk->next->prev = wrk->prev;	341	wrk->next->prev = wrk->prev;
263	wrk->prev->next = wrk->next;	342	wrk->prev->next = wrk->next;
264		343
265	free (wrk);	344	free (wrk);
266	}	345	}
267		346
268	static unsigned int etp_nreqs (void)	347	static unsigned int
		348	etp_nreqs (void)
269	{	349	{
270	int retval;	350	int retval;
271	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	351	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
272	retval = nreqs;	352	retval = nreqs;
273	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	353	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
274	return retval;	354	return retval;
275	}	355	}
276		356
277	static unsigned int etp_nready (void)	357	static unsigned int
		358	etp_nready (void)
278	{	359	{
279	unsigned int retval;	360	unsigned int retval;
280		361
281	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	362	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282	retval = nready;	363	retval = nready;
283	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	364	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284		365
285	return retval;	366	return retval;
286	}	367	}
287		368
288	static unsigned int etp_npending (void)	369	static unsigned int
		370	etp_npending (void)
289	{	371	{
290	unsigned int retval;	372	unsigned int retval;
291		373
292	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	374	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
293	retval = npending;	375	retval = npending;
294	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	376	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
295		377
296	return retval;	378	return retval;
297	}	379	}
298		380
299	static unsigned int etp_nthreads (void)	381	static unsigned int
		382	etp_nthreads (void)
300	{	383	{
301	unsigned int retval;	384	unsigned int retval;
302		385
303	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	386	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
304	retval = started;	387	retval = started;
…		…
318	} etp_reqq;	401	} etp_reqq;
319		402
320	static etp_reqq req_queue;	403	static etp_reqq req_queue;
321	static etp_reqq res_queue;	404	static etp_reqq res_queue;
322		405
		406	static void ecb_noinline ecb_cold
		407	reqq_init (etp_reqq *q)
		408	{
		409	int pri;
		410
		411	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		412	q->qs[pri] = q->qe[pri] = 0;
		413
		414	q->size = 0;
		415	}
		416
		417	static int ecb_noinline
323	static int reqq_push (etp_reqq *q, ETP_REQ *req)	418	reqq_push (etp_reqq *q, ETP_REQ *req)
324	{	419	{
325	int pri = req->pri;	420	int pri = req->pri;
326	req->next = 0;	421	req->next = 0;
327		422
328	if (q->qe[pri])	423	if (q->qe[pri])
…		…
334	q->qe[pri] = q->qs[pri] = req;	429	q->qe[pri] = q->qs[pri] = req;
335		430
336	return q->size++;	431	return q->size++;
337	}	432	}
338		433
		434	static ETP_REQ * ecb_noinline
339	static ETP_REQ *reqq_shift (etp_reqq *q)	435	reqq_shift (etp_reqq *q)
340	{	436	{
341	int pri;	437	int pri;
342		438
343	if (!q->size)	439	if (!q->size)
344	return 0;	440	return 0;
…		…
359	}	455	}
360		456
361	abort ();	457	abort ();
362	}	458	}
363		459
364	static void etp_atfork_prepare (void)	460	static int ecb_cold
		461	etp_init (void (*want_poll)(void), void (*done_poll)(void))
365	{	462	{
366	X_LOCK (wrklock);	463	X_MUTEX_CREATE (wrklock);
367	X_LOCK (reqlock);	464	X_MUTEX_CREATE (reslock);
368	X_LOCK (reslock);	465	X_MUTEX_CREATE (reqlock);
369	#if !HAVE_PREADWRITE	466	X_COND_CREATE (reqwait);
370	X_LOCK (preadwritelock);
371	#endif
372	}
373		467
374	static void etp_atfork_parent (void)	468	reqq_init (&req_queue);
375	{	469	reqq_init (&res_queue);
376	#if !HAVE_PREADWRITE
377	X_UNLOCK (preadwritelock);
378	#endif
379	X_UNLOCK (reslock);
380	X_UNLOCK (reqlock);
381	X_UNLOCK (wrklock);
382	}
383		470
384	static void etp_atfork_child (void)	471	wrk_first.next =
385	{	472	wrk_first.prev = &wrk_first;
386	ETP_REQ *prv;
387
388	while ((prv = reqq_shift (&req_queue)))
389	ETP_DESTROY (prv);
390
391	while ((prv = reqq_shift (&res_queue)))
392	ETP_DESTROY (prv);
393
394	while (wrk_first.next != &wrk_first)
395	{
396	etp_worker *wrk = wrk_first.next;
397
398	if (wrk->req)
399	ETP_DESTROY (wrk->req);
400
401	etp_worker_clear (wrk);
402	etp_worker_free (wrk);
403	}
404		473
405	started = 0;	474	started = 0;
406	idle = 0;	475	idle = 0;
407	nreqs = 0;	476	nreqs = 0;
408	nready = 0;	477	nready = 0;
409	npending = 0;	478	npending = 0;
410		479
411	etp_atfork_parent ();
412	}
413
414	static void
415	etp_once_init (void)
416	{
417	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
418	}
419
420	static int
421	etp_init (void (*want_poll)(void), void (*done_poll)(void))
422	{
423	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
424
425	pthread_once (&doinit, etp_once_init);
426
427	want_poll_cb = want_poll;	480	want_poll_cb = want_poll;
428	done_poll_cb = done_poll;	481	done_poll_cb = done_poll;
429		482
430	return 0;	483	return 0;
431	}	484	}
432		485
433	X_THREAD_PROC (etp_proc);	486	X_THREAD_PROC (etp_proc);
434		487
		488	static void ecb_cold
435	static void etp_start_thread (void)	489	etp_start_thread (void)
436	{	490	{
437	etp_worker *wrk = calloc (1, sizeof (etp_worker));	491	etp_worker *wrk = calloc (1, sizeof (etp_worker));
438		492
439	/*TODO*/	493	/*TODO*/
440	assert (("unable to allocate worker thread data", wrk));	494	assert (("unable to allocate worker thread data", wrk));
…		…
453	free (wrk);	507	free (wrk);
454		508
455	X_UNLOCK (wrklock);	509	X_UNLOCK (wrklock);
456	}	510	}
457		511
		512	static void
458	static void etp_maybe_start_thread (void)	513	etp_maybe_start_thread (void)
459	{	514	{
460	if (expect_true (etp_nthreads () >= wanted))	515	if (ecb_expect_true (etp_nthreads () >= wanted))
461	return;	516	return;
462		517
463	/* todo: maybe use idle here, but might be less exact */	518	/* todo: maybe use idle here, but might be less exact */
464	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	519	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
465	return;	520	return;
466		521
467	etp_start_thread ();	522	etp_start_thread ();
468	}	523	}
469		524
		525	static void ecb_cold
470	static void etp_end_thread (void)	526	etp_end_thread (void)
471	{	527	{
472	eio_req *req = calloc (1, sizeof (eio_req));	528	eio_req *req = calloc (1, sizeof (eio_req));
473		529
474	req->type = -1;	530	req->type = -1;
475	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	531	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
482	X_LOCK (wrklock);	538	X_LOCK (wrklock);
483	--started;	539	--started;
484	X_UNLOCK (wrklock);	540	X_UNLOCK (wrklock);
485	}	541	}
486		542
487	static int etp_poll (void)	543	static int
		544	etp_poll (void)
488	{	545	{
489	unsigned int maxreqs;	546	unsigned int maxreqs;
490	unsigned int maxtime;	547	unsigned int maxtime;
491	struct timeval tv_start, tv_now;	548	struct timeval tv_start, tv_now;
492		549
…		…
522		579
523	X_LOCK (reqlock);	580	X_LOCK (reqlock);
524	--nreqs;	581	--nreqs;
525	X_UNLOCK (reqlock);	582	X_UNLOCK (reqlock);
526		583
527	if (expect_false (req->type == EIO_GROUP && req->size))	584	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
528	{	585	{
529	req->int1 = 1; /* mark request as delayed */	586	req->int1 = 1; /* mark request as delayed */
530	continue;	587	continue;
531	}	588	}
532	else	589	else
533	{	590	{
534	int res = ETP_FINISH (req);	591	int res = ETP_FINISH (req);
535	if (expect_false (res))	592	if (ecb_expect_false (res))
536	return res;	593	return res;
537	}	594	}
538		595
539	if (expect_false (maxreqs && !--maxreqs))	596	if (ecb_expect_false (maxreqs && !--maxreqs))
540	break;	597	break;
541		598
542	if (maxtime)	599	if (maxtime)
543	{	600	{
544	gettimeofday (&tv_now, 0);	601	gettimeofday (&tv_now, 0);
…		…
550		607
551	errno = EAGAIN;	608	errno = EAGAIN;
552	return -1;	609	return -1;
553	}	610	}
554		611
		612	static void
555	static void etp_cancel (ETP_REQ *req)	613	etp_cancel (ETP_REQ *req)
556	{	614	{
557	X_LOCK (wrklock);	615	req->cancelled = 1;
558	req->flags |= EIO_FLAG_CANCELLED;
559	X_UNLOCK (wrklock);
560		616
561	eio_grp_cancel (req);	617	eio_grp_cancel (req);
562	}	618	}
563		619
		620	static void
564	static void etp_submit (ETP_REQ *req)	621	etp_submit (ETP_REQ *req)
565	{	622	{
566	req->pri -= ETP_PRI_MIN;	623	req->pri -= ETP_PRI_MIN;
567		624
568	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	625	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
569	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	626	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
570		627
571	if (expect_false (req->type == EIO_GROUP))	628	if (ecb_expect_false (req->type == EIO_GROUP))
572	{	629	{
573	/* I hope this is worth it :/ */	630	/* I hope this is worth it :/ */
574	X_LOCK (reqlock);	631	X_LOCK (reqlock);
575	++nreqs;	632	++nreqs;
576	X_UNLOCK (reqlock);	633	X_UNLOCK (reqlock);
…		…
595		652
596	etp_maybe_start_thread ();	653	etp_maybe_start_thread ();
597	}	654	}
598	}	655	}
599		656
		657	static void ecb_cold
600	static void etp_set_max_poll_time (double nseconds)	658	etp_set_max_poll_time (double nseconds)
601	{	659	{
602	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	660	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603	max_poll_time = nseconds;	661	max_poll_time = nseconds * EIO_TICKS;
604	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	662	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605	}	663	}
606		664
		665	static void ecb_cold
607	static void etp_set_max_poll_reqs (unsigned int maxreqs)	666	etp_set_max_poll_reqs (unsigned int maxreqs)
608	{	667	{
609	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	668	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
610	max_poll_reqs = maxreqs;	669	max_poll_reqs = maxreqs;
611	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	670	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
612	}	671	}
613		672
		673	static void ecb_cold
614	static void etp_set_max_idle (unsigned int nthreads)	674	etp_set_max_idle (unsigned int nthreads)
615	{	675	{
616	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	676	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617	max_idle = nthreads <= 0 ? 1 : nthreads;	677	max_idle = nthreads;
618	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	678	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619	}	679	}
620		680
		681	static void ecb_cold
		682	etp_set_idle_timeout (unsigned int seconds)
		683	{
		684	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		685	idle_timeout = seconds;
		686	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		687	}
		688
		689	static void ecb_cold
621	static void etp_set_min_parallel (unsigned int nthreads)	690	etp_set_min_parallel (unsigned int nthreads)
622	{	691	{
623	if (wanted < nthreads)	692	if (wanted < nthreads)
624	wanted = nthreads;	693	wanted = nthreads;
625	}	694	}
626		695
		696	static void ecb_cold
627	static void etp_set_max_parallel (unsigned int nthreads)	697	etp_set_max_parallel (unsigned int nthreads)
628	{	698	{
629	if (wanted > nthreads)	699	if (wanted > nthreads)
630	wanted = nthreads;	700	wanted = nthreads;
631		701
632	while (started > wanted)	702	while (started > wanted)
633	etp_end_thread ();	703	etp_end_thread ();
634	}	704	}
635		705
636	/*****************************************************************************/	706	/*****************************************************************************/
637		707
		708	static void
638	static void grp_try_feed (eio_req *grp)	709	grp_try_feed (eio_req *grp)
639	{	710	{
640	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	711	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641	{	712	{
642	grp->flags &= ~EIO_FLAG_GROUPADD;	713	grp->flags &= ~EIO_FLAG_GROUPADD;
643		714
…		…
650	break;	721	break;
651	}	722	}
652	}	723	}
653	}	724	}
654		725
		726	static int
655	static int grp_dec (eio_req *grp)	727	grp_dec (eio_req *grp)
656	{	728	{
657	--grp->size;	729	--grp->size;
658		730
659	/* call feeder, if applicable */	731	/* call feeder, if applicable */
660	grp_try_feed (grp);	732	grp_try_feed (grp);
…		…
664	return eio_finish (grp);	736	return eio_finish (grp);
665	else	737	else
666	return 0;	738	return 0;
667	}	739	}
668		740
		741	static void
669	void eio_destroy (eio_req *req)	742	eio_destroy (eio_req *req)
670	{	743	{
671	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	744	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	745	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673		746
674	EIO_DESTROY (req);	747	EIO_DESTROY (req);
675	}	748	}
676		749
		750	static int
677	static int eio_finish (eio_req *req)	751	eio_finish (eio_req *req)
678	{	752	{
679	int res = EIO_FINISH (req);	753	int res = EIO_FINISH (req);
680		754
681	if (req->grp)	755	if (req->grp)
682	{	756	{
…		…
690	if (grp->grp_first == req)	764	if (grp->grp_first == req)
691	grp->grp_first = req->grp_next;	765	grp->grp_first = req->grp_next;
692		766
693	res2 = grp_dec (grp);	767	res2 = grp_dec (grp);
694		768
695	if (!res && res2)	769	if (!res)
696	res = res2;	770	res = res2;
697	}	771	}
698		772
699	eio_destroy (req);	773	eio_destroy (req);
700		774
701	return res;	775	return res;
702	}	776	}
703		777
		778	void
704	void eio_grp_cancel (eio_req *grp)	779	eio_grp_cancel (eio_req *grp)
705	{	780	{
706	for (grp = grp->grp_first; grp; grp = grp->grp_next)	781	for (grp = grp->grp_first; grp; grp = grp->grp_next)
707	eio_cancel (grp);	782	eio_cancel (grp);
708	}	783	}
709		784
		785	void
710	void eio_cancel (eio_req *req)	786	eio_cancel (eio_req *req)
711	{	787	{
712	etp_cancel (req);	788	etp_cancel (req);
713	}	789	}
714		790
		791	void
715	void eio_submit (eio_req *req)	792	eio_submit (eio_req *req)
716	{	793	{
717	etp_submit (req);	794	etp_submit (req);
718	}	795	}
719		796
720	unsigned int eio_nreqs (void)	797	unsigned int
		798	eio_nreqs (void)
721	{	799	{
722	return etp_nreqs ();	800	return etp_nreqs ();
723	}	801	}
724		802
725	unsigned int eio_nready (void)	803	unsigned int
		804	eio_nready (void)
726	{	805	{
727	return etp_nready ();	806	return etp_nready ();
728	}	807	}
729		808
730	unsigned int eio_npending (void)	809	unsigned int
		810	eio_npending (void)
731	{	811	{
732	return etp_npending ();	812	return etp_npending ();
733	}	813	}
734		814
735	unsigned int eio_nthreads (void)	815	unsigned int ecb_cold
		816	eio_nthreads (void)
736	{	817	{
737	return etp_nthreads ();	818	return etp_nthreads ();
738	}	819	}
739		820
		821	void ecb_cold
740	void eio_set_max_poll_time (double nseconds)	822	eio_set_max_poll_time (double nseconds)
741	{	823	{
742	etp_set_max_poll_time (nseconds);	824	etp_set_max_poll_time (nseconds);
743	}	825	}
744		826
		827	void ecb_cold
745	void eio_set_max_poll_reqs (unsigned int maxreqs)	828	eio_set_max_poll_reqs (unsigned int maxreqs)
746	{	829	{
747	etp_set_max_poll_reqs (maxreqs);	830	etp_set_max_poll_reqs (maxreqs);
748	}	831	}
749		832
		833	void ecb_cold
750	void eio_set_max_idle (unsigned int nthreads)	834	eio_set_max_idle (unsigned int nthreads)
751	{	835	{
752	etp_set_max_idle (nthreads);	836	etp_set_max_idle (nthreads);
753	}	837	}
754		838
		839	void ecb_cold
		840	eio_set_idle_timeout (unsigned int seconds)
		841	{
		842	etp_set_idle_timeout (seconds);
		843	}
		844
		845	void ecb_cold
755	void eio_set_min_parallel (unsigned int nthreads)	846	eio_set_min_parallel (unsigned int nthreads)
756	{	847	{
757	etp_set_min_parallel (nthreads);	848	etp_set_min_parallel (nthreads);
758	}	849	}
759		850
		851	void ecb_cold
760	void eio_set_max_parallel (unsigned int nthreads)	852	eio_set_max_parallel (unsigned int nthreads)
761	{	853	{
762	etp_set_max_parallel (nthreads);	854	etp_set_max_parallel (nthreads);
763	}	855	}
764		856
765	int eio_poll (void)	857	int eio_poll (void)
…		…
774	# undef pread	866	# undef pread
775	# undef pwrite	867	# undef pwrite
776	# define pread eio__pread	868	# define pread eio__pread
777	# define pwrite eio__pwrite	869	# define pwrite eio__pwrite
778		870
779	static ssize_t	871	static eio_ssize_t
780	eio__pread (int fd, void *buf, size_t count, off_t offset)	872	eio__pread (int fd, void *buf, size_t count, off_t offset)
781	{	873	{
782	ssize_t res;	874	eio_ssize_t res;
783	off_t ooffset;	875	off_t ooffset;
784		876
785	X_LOCK (preadwritelock);	877	X_LOCK (preadwritelock);
786	ooffset = lseek (fd, 0, SEEK_CUR);	878	ooffset = lseek (fd, 0, SEEK_CUR);
787	lseek (fd, offset, SEEK_SET);	879	lseek (fd, offset, SEEK_SET);
…		…
790	X_UNLOCK (preadwritelock);	882	X_UNLOCK (preadwritelock);
791		883
792	return res;	884	return res;
793	}	885	}
794		886
795	static ssize_t	887	static eio_ssize_t
796	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	888	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
797	{	889	{
798	ssize_t res;	890	eio_ssize_t res;
799	off_t ooffset;	891	off_t ooffset;
800		892
801	X_LOCK (preadwritelock);	893	X_LOCK (preadwritelock);
802	ooffset = lseek (fd, 0, SEEK_CUR);	894	ooffset = lseek (fd, 0, SEEK_CUR);
803	lseek (fd, offset, SEEK_SET);	895	lseek (fd, offset, SEEK_SET);
…		…
807		899
808	return res;	900	return res;
809	}	901	}
810	#endif	902	#endif
811		903
812	#ifndef HAVE_FUTIMES	904	#ifndef HAVE_UTIMES
813		905
814	# undef utimes	906	# undef utimes
815	# undef futimes
816	# define utimes(path,times) eio__utimes (path, times)	907	# define utimes(path,times) eio__utimes (path, times)
817	# define futimes(fd,times) eio__futimes (fd, times)
818		908
819	static int	909	static int
820	eio__utimes (const char *filename, const struct timeval times[2])	910	eio__utimes (const char *filename, const struct timeval times[2])
821	{	911	{
822	if (times)	912	if (times)
…		…
830	}	920	}
831	else	921	else
832	return utime (filename, 0);	922	return utime (filename, 0);
833	}	923	}
834		924
		925	#endif
		926
		927	#ifndef HAVE_FUTIMES
		928
		929	# undef futimes
		930	# define futimes(fd,times) eio__futimes (fd, times)
		931
		932	static int
835	static int eio__futimes (int fd, const struct timeval tv[2])	933	eio__futimes (int fd, const struct timeval tv[2])
836	{	934	{
837	errno = ENOSYS;	935	errno = ENOSYS;
838	return -1;	936	return -1;
839	}	937	}
840		938
…		…
843	#if !HAVE_FDATASYNC	941	#if !HAVE_FDATASYNC
844	# undef fdatasync	942	# undef fdatasync
845	# define fdatasync(fd) fsync (fd)	943	# define fdatasync(fd) fsync (fd)
846	#endif	944	#endif
847		945
		946	static int
		947	eio__syncfs (int fd)
		948	{
		949	int res;
		950
		951	#if HAVE_SYS_SYNCFS
		952	res = (int)syscall (__NR_syncfs, (int)(fd));
		953	#else
		954	res = -1;
		955	errno = ENOSYS;
		956	#endif
		957
		958	if (res < 0 && errno == ENOSYS && fd >= 0)
		959	sync ();
		960
		961	return res;
		962	}
		963
848	/* sync_file_range always needs emulation */	964	/* sync_file_range always needs emulation */
849	int	965	static int
850	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	966	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
851	{	967	{
852	#if HAVE_SYNC_FILE_RANGE	968	#if HAVE_SYNC_FILE_RANGE
853	int res;	969	int res;
854		970
…		…
867	if (!res || errno != ENOSYS)	983	if (!res || errno != ENOSYS)
868	return res;	984	return res;
869	#endif	985	#endif
870		986
871	/* even though we could play tricks with the flags, it's better to always	987	/* even though we could play tricks with the flags, it's better to always
872	* call fdatasync, as thta matches the expectation of it's users best */	988	* call fdatasync, as that matches the expectation of its users best */
873	return fdatasync (fd);	989	return fdatasync (fd);
		990	}
		991
		992	static int
		993	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		994	{
		995	#if HAVE_FALLOCATE
		996	return fallocate (fd, mode, offset, len);
		997	#else
		998	errno = ENOSYS;
		999	return -1;
		1000	#endif
874	}	1001	}
875		1002
876	#if !HAVE_READAHEAD	1003	#if !HAVE_READAHEAD
877	# undef readahead	1004	# undef readahead
878	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1005	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
879		1006
880	static ssize_t	1007	static eio_ssize_t
881	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1008	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
882	{	1009	{
883	size_t todo = count;	1010	size_t todo = count;
884	dBUF;	1011	dBUF;
885		1012
…		…
890	pread (fd, eio_buf, len, offset);	1017	pread (fd, eio_buf, len, offset);
891	offset += len;	1018	offset += len;
892	todo -= len;	1019	todo -= len;
893	}	1020	}
894		1021
		1022	FUBd;
		1023
895	errno = 0;	1024	errno = 0;
896	return count;	1025	return count;
897	}	1026	}
898		1027
899	#endif	1028	#endif
900		1029
901	/* sendfile always needs emulation */	1030	/* sendfile always needs emulation */
902	static ssize_t	1031	static eio_ssize_t
903	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1032	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
904	{	1033	{
		1034	eio_ssize_t written = 0;
905	ssize_t res;	1035	eio_ssize_t res;
906		1036
907	if (!count)	1037	if (!count)
908	return 0;	1038	return 0;
909		1039
		1040	for (;;)
		1041	{
		1042	#ifdef __APPLE__
		1043	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1044	#endif
910	#if HAVE_SENDFILE	1045	#if HAVE_SENDFILE
911	# if __linux	1046	# if __linux
		1047	off_t soffset = offset;
912	res = sendfile (ofd, ifd, &offset, count);	1048	res = sendfile (ofd, ifd, &soffset, count);
913		1049
914	# elif __freebsd	1050	# elif __FreeBSD__
915	/*	1051	/*
916	* Of course, the freebsd sendfile is a dire hack with no thoughts	1052	* Of course, the freebsd sendfile is a dire hack with no thoughts
917	* wasted on making it similar to other I/O functions.	1053	* wasted on making it similar to other I/O functions.
918	*/	1054	*/
919	{
920	off_t sbytes;	1055	off_t sbytes;
921	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1056	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922		1057
923	if (res < 0 && sbytes)	1058	#if 0 /* according to the manpage, this is correct, but broken behaviour */
924	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1059	/* freebsd' sendfile will return 0 on success */
		1060	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1061	/* not on e.g. EIO or EPIPE - sounds broken */
		1062	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
925	res = sbytes;	1063	res = sbytes;
926	}	1064	#endif
		1065
		1066	/* according to source inspection, this is correct, and useful behaviour */
		1067	if (sbytes)
		1068	res = sbytes;
		1069
		1070	# elif defined (__APPLE__)
		1071	off_t sbytes = count;
		1072	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1073
		1074	/* according to the manpage, sbytes is always valid */
		1075	if (sbytes)
		1076	res = sbytes;
927		1077
928	# elif __hpux	1078	# elif __hpux
929	res = sendfile (ofd, ifd, offset, count, 0, 0);	1079	res = sendfile (ofd, ifd, offset, count, 0, 0);
930		1080
931	# elif __solaris	1081	# elif __solaris
932	{
933	struct sendfilevec vec;	1082	struct sendfilevec vec;
934	size_t sbytes;	1083	size_t sbytes;
935		1084
936	vec.sfv_fd = ifd;	1085	vec.sfv_fd = ifd;
937	vec.sfv_flag = 0;	1086	vec.sfv_flag = 0;
938	vec.sfv_off = offset;	1087	vec.sfv_off = offset;
939	vec.sfv_len = count;	1088	vec.sfv_len = count;
940		1089
941	res = sendfilev (ofd, &vec, 1, &sbytes);	1090	res = sendfilev (ofd, &vec, 1, &sbytes);
942		1091
943	if (res < 0 && sbytes)	1092	if (res < 0 && sbytes)
944	res = sbytes;	1093	res = sbytes;
945	}
946		1094
947	# endif	1095	# endif
		1096
		1097	#elif defined (_WIN32) && 0
		1098	/* does not work, just for documentation of what would need to be done */
		1099	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1100	/* libeio guarantees that the file offset does not change, and windows */
		1101	/* has no way to get an independent handle to the same file description */
		1102	HANDLE h = TO_SOCKET (ifd);
		1103	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1104	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1105
948	#else	1106	#else
949	res = -1;	1107	res = -1;
950	errno = ENOSYS;	1108	errno = ENOSYS;
951	#endif	1109	#endif
952		1110
		1111	/* we assume sendfile can copy at least 128mb in one go */
		1112	if (res <= 128 * 1024 * 1024)
		1113	{
		1114	if (res > 0)
		1115	written += res;
		1116
		1117	if (written)
		1118	return written;
		1119
		1120	break;
		1121	}
		1122	else
		1123	{
		1124	/* if we requested more, then probably the kernel was lazy */
		1125	written += res;
		1126	offset += res;
		1127	count -= res;
		1128
		1129	if (!count)
		1130	return written;
		1131	}
		1132	}
		1133
953	if (res < 0	1134	if (res < 0
954	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1135	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1136	/* BSDs */
		1137	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1138	|| errno == ENOTSUP
		1139	#endif
		1140	#ifdef EOPNOTSUPP /* windows */
		1141	|| errno == EOPNOTSUPP /* BSDs */
		1142	#endif
955	#if __solaris	1143	#if __solaris
956	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1144	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
957	#endif	1145	#endif
958	)	1146	)
959	)	1147	)
…		…
963		1151
964	res = 0;	1152	res = 0;
965		1153
966	while (count)	1154	while (count)
967	{	1155	{
968	ssize_t cnt;	1156	eio_ssize_t cnt;
969		1157
970	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1158	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
971		1159
972	if (cnt <= 0)	1160	if (cnt <= 0)
973	{	1161	{
…		…
985		1173
986	offset += cnt;	1174	offset += cnt;
987	res += cnt;	1175	res += cnt;
988	count -= cnt;	1176	count -= cnt;
989	}	1177	}
		1178
		1179	FUBd;
990	}	1180	}
991		1181
992	return res;	1182	return res;
		1183	}
		1184
		1185	#ifdef PAGESIZE
		1186	# define eio_pagesize() PAGESIZE
		1187	#else
		1188	static intptr_t
		1189	eio_pagesize (void)
		1190	{
		1191	static intptr_t page;
		1192
		1193	if (!page)
		1194	page = sysconf (_SC_PAGESIZE);
		1195
		1196	return page;
		1197	}
		1198	#endif
		1199
		1200	static void
		1201	eio_page_align (void **addr, size_t *length)
		1202	{
		1203	intptr_t mask = eio_pagesize () - 1;
		1204
		1205	/* round down addr */
		1206	intptr_t adj = mask & (intptr_t)*addr;
		1207
		1208	*addr = (void *)((intptr_t)*addr - adj);
		1209	*length += adj;
		1210
		1211	/* round up length */
		1212	*length = (*length + mask) & ~mask;
		1213	}
		1214
		1215	#if !_POSIX_MEMLOCK
		1216	# define eio__mlockall(a) EIO_ENOSYS ()
		1217	#else
		1218
		1219	static int
		1220	eio__mlockall (int flags)
		1221	{
		1222	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1223	extern int mallopt (int, int);
		1224	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1225	#endif
		1226
		1227	if (EIO_MCL_CURRENT != MCL_CURRENT
		1228	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1229	{
		1230	flags = 0
		1231	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1232	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1233	}
		1234
		1235	return mlockall (flags);
		1236	}
		1237	#endif
		1238
		1239	#if !_POSIX_MEMLOCK_RANGE
		1240	# define eio__mlock(a,b) EIO_ENOSYS ()
		1241	#else
		1242
		1243	static int
		1244	eio__mlock (void *addr, size_t length)
		1245	{
		1246	eio_page_align (&addr, &length);
		1247
		1248	return mlock (addr, length);
		1249	}
		1250
		1251	#endif
		1252
		1253	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1254	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1255	#else
		1256
		1257	static int
		1258	eio__msync (void *mem, size_t len, int flags)
		1259	{
		1260	eio_page_align (&mem, &len);
		1261
		1262	if (EIO_MS_ASYNC != MS_SYNC
		1263	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1264	|| EIO_MS_SYNC != MS_SYNC)
		1265	{
		1266	flags = 0
		1267	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1268	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1269	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1270	}
		1271
		1272	return msync (mem, len, flags);
		1273	}
		1274
		1275	#endif
		1276
		1277	static int
		1278	eio__mtouch (eio_req *req)
		1279	{
		1280	void *mem = req->ptr2;
		1281	size_t len = req->size;
		1282	int flags = req->int1;
		1283
		1284	eio_page_align (&mem, &len);
		1285
		1286	{
		1287	intptr_t addr = (intptr_t)mem;
		1288	intptr_t end = addr + len;
		1289	intptr_t page = eio_pagesize ();
		1290
		1291	if (addr < end)
		1292	if (flags & EIO_MT_MODIFY) /* modify */
		1293	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1294	else
		1295	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1296	}
		1297
		1298	return 0;
		1299	}
		1300
		1301	/*****************************************************************************/
		1302	/* requests implemented outside eio_execute, because they are so large */
		1303
		1304	static void
		1305	eio__realpath (eio_req *req, etp_worker *self)
		1306	{
		1307	char *rel = req->ptr1;
		1308	char *res;
		1309	char *tmp1, *tmp2;
		1310	#if SYMLOOP_MAX > 32
		1311	int symlinks = SYMLOOP_MAX;
		1312	#else
		1313	int symlinks = 32;
		1314	#endif
		1315
		1316	req->result = -1;
		1317
		1318	errno = EINVAL;
		1319	if (!rel)
		1320	return;
		1321
		1322	errno = ENOENT;
		1323	if (!*rel)
		1324	return;
		1325
		1326	if (!req->ptr2)
		1327	{
		1328	X_LOCK (wrklock);
		1329	req->flags |= EIO_FLAG_PTR2_FREE;
		1330	X_UNLOCK (wrklock);
		1331	req->ptr2 = malloc (PATH_MAX * 3);
		1332
		1333	errno = ENOMEM;
		1334	if (!req->ptr2)
		1335	return;
		1336	}
		1337
		1338	res = req->ptr2;
		1339	tmp1 = res + PATH_MAX;
		1340	tmp2 = tmp1 + PATH_MAX;
		1341
		1342	#if 0 /* disabled, the musl way to do things is just too racy */
		1343	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1344	/* on linux we may be able to ask the kernel */
		1345	{
		1346	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1347
		1348	if (fd >= 0)
		1349	{
		1350	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1351	req->result = readlink (tmp1, res, PATH_MAX);
		1352	close (fd);
		1353
		1354	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1355
		1356	if (req->result > 0)
		1357	goto done;
		1358	}
		1359	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1360	return;
		1361	}
		1362	#endif
		1363	#endif
		1364
		1365	if (*rel != '/')
		1366	{
		1367	if (!getcwd (res, PATH_MAX))
		1368	return;
		1369
		1370	if (res [1]) /* only use if not / */
		1371	res += strlen (res);
		1372	}
		1373
		1374	while (*rel)
		1375	{
		1376	eio_ssize_t len, linklen;
		1377	char *beg = rel;
		1378
		1379	while (*rel && *rel != '/')
		1380	++rel;
		1381
		1382	len = rel - beg;
		1383
		1384	if (!len) /* skip slashes */
		1385	{
		1386	++rel;
		1387	continue;
		1388	}
		1389
		1390	if (beg [0] == '.')
		1391	{
		1392	if (len == 1)
		1393	continue; /* . - nop */
		1394
		1395	if (beg [1] == '.' && len == 2)
		1396	{
		1397	/* .. - back up one component, if possible */
		1398
		1399	while (res != req->ptr2)
		1400	if (*--res == '/')
		1401	break;
		1402
		1403	continue;
		1404	}
		1405	}
		1406
		1407	errno = ENAMETOOLONG;
		1408	if (res + 1 + len + 1 >= tmp1)
		1409	return;
		1410
		1411	/* copy one component */
		1412	*res = '/';
		1413	memcpy (res + 1, beg, len);
		1414
		1415	/* zero-terminate, for readlink */
		1416	res [len + 1] = 0;
		1417
		1418	/* now check if it's a symlink */
		1419	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1420
		1421	if (linklen < 0)
		1422	{
		1423	if (errno != EINVAL)
		1424	return;
		1425
		1426	/* it's a normal directory. hopefully */
		1427	res += len + 1;
		1428	}
		1429	else
		1430	{
		1431	/* yay, it was a symlink - build new path in tmp2 */
		1432	int rellen = strlen (rel);
		1433
		1434	errno = ENAMETOOLONG;
		1435	if (linklen + 1 + rellen >= PATH_MAX)
		1436	return;
		1437
		1438	errno = ELOOP;
		1439	if (!--symlinks)
		1440	return;
		1441
		1442	if (*tmp1 == '/')
		1443	res = req->ptr2; /* symlink resolves to an absolute path */
		1444
		1445	/* we need to be careful, as rel might point into tmp2 already */
		1446	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1447	tmp2 [linklen] = '/';
		1448	memcpy (tmp2, tmp1, linklen);
		1449
		1450	rel = tmp2;
		1451	}
		1452	}
		1453
		1454	/* special case for the lone root path */
		1455	if (res == req->ptr2)
		1456	*res++ = '/';
		1457
		1458	req->result = res - (char *)req->ptr2;
		1459
		1460	done:
		1461	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
993	}	1462	}
994		1463
995	static signed char	1464	static signed char
996	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1465	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997	{	1466	{
998	return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */	1467	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
999	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;	1468	: a->inode < b->inode ? -1
		1469	: a->inode > b->inode ? 1
		1470	: 0;
1000	}	1471	}
1001		1472
		1473	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1474
1002	#define EIO_QSORT_CUTOFF 20 /* quite high, but performs well on many filesystems */	1475	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1476	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1003		1477
1004	static void	1478	static void
1005	eio_dent_sort (eio_dirent *dents, int size)	1479	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1006	{	1480	{
		1481	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
		1482	unsigned char *bit = bits;
		1483
		1484	assert (CHAR_BIT == 8);
		1485	assert (sizeof (eio_dirent) * 8 < 256);
		1486	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1487	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1488
		1489	if (size <= EIO_SORT_FAST)
		1490	return;
		1491
		1492	/* first prepare an array of bits to test in our radix sort */
		1493	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
		1494	/* inode_bits must contain all inodes ORed together */
		1495	/* which is used to skip bits that are 0 everywhere, which is very common */
		1496	{
		1497	eio_ino_t endianness;
1007	int i, j;	1498	int i, j;
1008		1499
1009	if (size > EIO_QSORT_CUTOFF * 3) /* skip quicksort for small directories */	1500	/* we store the byte offset of byte n into byte n of "endianness" */
		1501	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1502	((unsigned char *)&endianness)[i] = i;
		1503
		1504	*bit++ = 0;
		1505
		1506	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1507	{
		1508	/* shifting off the byte offsets out of "endianness" */
		1509	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1510	endianness >>= 8;
		1511
		1512	for (j = 0; j < 8; ++j)
		1513	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
		1514	*bit++ = offs + j;
		1515	}
		1516
		1517	for (j = 0; j < 8; ++j)
		1518	if (score_bits & (1 << j))
		1519	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1520	}
		1521
		1522	/* now actually do the sorting (a variant of MSD radix sort) */
		1523	{
		1524	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
		1525	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
		1526	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
		1527	int stk_idx = 0;
		1528
		1529	base_stk [stk_idx] = dents;
		1530	end_stk [stk_idx] = dents + size;
		1531	bit_stk [stk_idx] = bit - 1;
		1532
		1533	do
		1534	{
		1535	base = base_stk [stk_idx];
		1536	end = end_stk [stk_idx];
		1537	bit = bit_stk [stk_idx];
		1538
		1539	for (;;)
		1540	{
		1541	unsigned char O = *bit >> 3;
		1542	unsigned char M = 1 << (*bit & 7);
		1543
		1544	eio_dirent *a = base;
		1545	eio_dirent *b = end;
		1546
		1547	if (b - a < EIO_SORT_CUTOFF)
		1548	break;
		1549
		1550	/* now bit-partition the array on the bit */
		1551	/* this ugly asymmetric loop seems to perform much better than typical */
		1552	/* partition algos found in the literature */
		1553	do
		1554	if (!(((unsigned char *)a)[O] & M))
		1555	++a;
		1556	else if (!(((unsigned char *)--b)[O] & M))
		1557	{
		1558	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1559	++a;
		1560	}
		1561	while (b > a);
		1562
		1563	/* next bit, or stop, if no bits left in this path */
		1564	if (!*--bit)
		1565	break;
		1566
		1567	base_stk [stk_idx] = a;
		1568	end_stk [stk_idx] = end;
		1569	bit_stk [stk_idx] = bit;
		1570	++stk_idx;
		1571
		1572	end = a;
		1573	}
		1574	}
		1575	while (stk_idx--);
		1576	}
		1577	}
		1578
		1579	static void
		1580	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1581	{
		1582	/* first move the smallest element to the front, to act as a sentinel */
		1583	{
		1584	int i;
		1585	eio_dirent *min = dents;
		1586
		1587	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1588	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1589	if (EIO_DENT_CMP (dents [i], <, *min))
		1590	min = &dents [i];
		1591
		1592	/* swap elements 0 and j (minimum) */
1010	{	1593	{
1011	/* first, use quicksort */	1594	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1012	/* should be good for 2**31 entries */
1013	struct rng { int l, r; } rng [32];
1014
1015	i = 0;
1016	rng[0].l = 0;
1017	rng[0].r = size;
1018
1019	while (expect_true (i >= 0))
1020	{
1021	int L = rng [i].l;
1022	int R = rng [i].r - 1;
1023
1024	if (expect_false (L + EIO_QSORT_CUTOFF < R))
1025	{
1026	eio_dirent piv = dents [L];
1027
1028	while (L < R)
1029	{
1030	while (eio_dent_cmp (&dents [R], &piv) >= 0 && L < R)
1031	--R;
1032
1033	if (L < R)
1034	dents [L++] = dents [R];
1035
1036	while (eio_dent_cmp (&dents [L], &piv) <= 0 && L < R)
1037	++L;
1038
1039	if (L < R)
1040	dents [R--] = dents [L];
1041	}
1042
1043	dents [L] = piv;
1044
1045	++i;
1046	rng [i].l = L + 1;
1047	rng [i].r = rng [i - 1].r;
1048	rng [i - 1].r = L;
1049
1050	if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l)
1051	{
1052	struct rng t;
1053
1054	t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t;
1055	}
1056	}
1057	else
1058	--i;
1059	}
1060	}	1595	}
		1596	}
1061		1597
1062	/* use a simple insertion sort at the end */	1598	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1599	{
		1600	eio_dirent *i, *j;
		1601
1063	for (i = 1; i < size; ++i)	1602	for (i = dents + 1; i < dents + size; ++i)
1064	{	1603	{
1065	eio_dirent value = dents [i];	1604	eio_dirent value = *i;
1066		1605
1067	for (j = i - 1; j >= 0 && eio_dent_cmp (&dents [j], &value) > 0; --j)	1606	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1068	dents [j + 1] = dents [j];	1607	j [1] = j [0];
1069		1608
1070	dents [j + 1] = value;	1609	j [1] = value;
1071	}	1610	}
		1611	}
		1612	}
		1613
		1614	static void
		1615	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1616	{
		1617	if (size <= 1)
		1618	return; /* our insertion sort relies on size > 0 */
		1619
		1620	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1621	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1622	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1623
		1624	/* use an insertion sort at the end, or for small arrays, */
		1625	/* as insertion sort is more efficient for small partitions */
		1626	eio_dent_insertion_sort (dents, size);
1072	}	1627	}
1073		1628
1074	/* read a full directory */	1629	/* read a full directory */
1075	static void	1630	static void
1076	eio__scandir (eio_req *req, etp_worker *self)	1631	eio__scandir (eio_req *req, etp_worker *self)
1077	{	1632	{
1078	DIR *dirp;
1079	EIO_STRUCT_DIRENT *entp;
1080	char *name, *names;	1633	char *name, *names;
1081	int namesalloc = 4096;	1634	int namesalloc = 4096 - sizeof (void *) * 4;
1082	int namesoffs = 0;	1635	int namesoffs = 0;
1083	int flags = req->int1;	1636	int flags = req->int1;
1084	eio_dirent *dents = 0;	1637	eio_dirent *dents = 0;
1085	int dentalloc = 128;	1638	int dentalloc = 128;
1086	int dentoffs = 0;	1639	int dentoffs = 0;
		1640	eio_ino_t inode_bits = 0;
		1641	#ifdef _WIN32
		1642	HANDLE dirp;
		1643	WIN32_FIND_DATA entp;
		1644	#else
		1645	DIR *dirp;
		1646	EIO_STRUCT_DIRENT *entp;
		1647	#endif
1087		1648
1088	req->result = -1;	1649	req->result = -1;
1089		1650
1090	if (!(flags & EIO_READDIR_DENTS))	1651	if (!(flags & EIO_READDIR_DENTS))
1091	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1652	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1092		1653
1093	X_LOCK (wrklock);	1654	#ifdef _WIN32
1094	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1655	{
		1656	int len = strlen ((const char *)req->ptr1);
		1657	char *path = malloc (MAX_PATH);
		1658	const char *fmt;
		1659
		1660	if (!len)
		1661	fmt = "./*";
		1662	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1663	fmt = "%s*";
		1664	else
		1665	fmt = "%s/*";
		1666
		1667	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1668	dirp = FindFirstFile (path, &entp);
		1669	free (path);
		1670
		1671	if (dirp == INVALID_HANDLE_VALUE)
		1672	{
		1673	dirp = 0;
		1674
		1675	switch (GetLastError ())
		1676	{
		1677	case ERROR_FILE_NOT_FOUND:
		1678	req->result = 0;
		1679	break;
		1680
		1681	case ERROR_INVALID_NAME:
		1682	case ERROR_PATH_NOT_FOUND:
		1683	case ERROR_NO_MORE_FILES:
		1684	errno = ENOENT;
		1685	break;
		1686
		1687	case ERROR_NOT_ENOUGH_MEMORY:
		1688	errno = ENOMEM;
		1689	break;
		1690
		1691	default:
		1692	errno = EINVAL;
		1693	break;
		1694	}
		1695	}
		1696	}
		1697	#else
1095	self->dirp = dirp = opendir (req->ptr1);	1698	dirp = opendir (req->ptr1);
		1699	#endif
		1700
		1701	if (req->flags & EIO_FLAG_PTR1_FREE)
		1702	free (req->ptr1);
		1703
1096	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1704	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1097	req->ptr1 = names = malloc (namesalloc);
1098	req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1705	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1099	X_UNLOCK (wrklock);	1706	req->ptr2 = names = malloc (namesalloc);
1100		1707
1101	if (dirp && names && (!flags || dents))	1708	if (dirp && names && (!flags || dents))
1102	for (;;)	1709	for (;;)
1103	{	1710	{
		1711	int done;
		1712
		1713	#ifdef _WIN32
		1714	done = !dirp;
		1715	#else
1104	errno = 0;	1716	errno = 0;
1105	entp = readdir (dirp);	1717	entp = readdir (dirp);
		1718	done = !entp;
		1719	#endif
1106		1720
1107	if (!entp)	1721	if (done)
1108	{	1722	{
		1723	#ifndef _WIN32
		1724	int old_errno = errno;
		1725	closedir (dirp);
		1726	errno = old_errno;
		1727
1109	if (errno)	1728	if (errno)
1110	break;	1729	break;
		1730	#endif
1111		1731
1112	/* sort etc. */	1732	/* sort etc. */
1113	req->int1 = flags;	1733	req->int1 = flags;
1114	req->result = dentoffs;	1734	req->result = dentoffs;
1115		1735
1116	if (dents)
1117	{
1118	eio_dirent *ent = dents + dentoffs;
1119
1120	while (ent > dents)
1121	(--ent)->name = names + (size_t)ent->name;
1122	}
1123
1124	if (flags & EIO_READDIR_STAT_ORDER	1736	if (flags & EIO_READDIR_STAT_ORDER)
1125	|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))	1737	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1126	eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */
1127	else if (flags & EIO_READDIR_DIRS_FIRST)	1738	else if (flags & EIO_READDIR_DIRS_FIRST)
		1739	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1740	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1741	else
1128	{	1742	{
1129	/* in this case, all is known, and we just put dirs first and sort them */	1743	/* in this case, all is known, and we just put dirs first and sort them */
1130	eio_dirent *ent = dents + dentoffs;	1744	eio_dirent *oth = dents + dentoffs;
1131	eio_dirent *dir = dents;	1745	eio_dirent *dir = dents;
1132		1746
1133	/* now move dirs to the front, and non-dirs to the back */	1747	/* now partition dirs to the front, and non-dirs to the back */
1134	/* by walking from both sides and swapping if necessary */	1748	/* by walking from both sides and swapping if necessary */
1135	while (ent > dir)	1749	while (oth > dir)
1136	{	1750	{
1137	if (dir->type == DT_DIR)	1751	if (dir->type == EIO_DT_DIR)
1138	++dir;	1752	++dir;
1139	else
1140	{
1141	--ent;
1142
1143	if (ent->type == DT_DIR)	1753	else if ((--oth)->type == EIO_DT_DIR)
1144	{	1754	{
1145	eio_dirent tmp = *dir;	1755	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1146	*dir = *ent;
1147	*ent = tmp;
1148		1756
1149	++dir;	1757	++dir;
1150	}	1758	}
1151	}	1759	}
		1760
		1761	/* now sort the dirs only (dirs all have the same score) */
		1762	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1152	}	1763	}
1153
1154	/* now sort the dirs only */
1155	eio_dent_sort (dents, dir - dents);
1156	}
1157
1158	/* only provide the names array unless DENTS is specified */
1159	if (!(flags & EIO_READDIR_DENTS))
1160	{
1161	X_LOCK (wrklock);
1162	assert (!dents);
1163	req->ptr1 = 0;
1164	req->ptr2 = names;
1165	X_UNLOCK (wrklock);
1166	}
1167		1764
1168	break;	1765	break;
1169	}	1766	}
1170		1767
1171	/* now add the entry to our list(s) */	1768	/* now add the entry to our list(s) */
1172	name = entp->d_name;	1769	name = D_NAME (entp);
1173		1770
1174	/* skip . and .. entries */	1771	/* skip . and .. entries */
1175	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1772	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1176	{	1773	{
1177	int len = D_NAMLEN (entp) + 1;	1774	int len = D_NAMLEN (entp) + 1;
1178		1775
1179	while (expect_false (namesoffs + len > namesalloc))	1776	while (ecb_expect_false (namesoffs + len > namesalloc))
1180	{	1777	{
1181	namesalloc *= 2;	1778	namesalloc *= 2;
1182	X_LOCK (wrklock);
1183	req->ptr1 = names = realloc (names, namesalloc);	1779	req->ptr2 = names = realloc (names, namesalloc);
1184	X_UNLOCK (wrklock);
1185		1780
1186	if (!names)	1781	if (!names)
1187	break;	1782	break;
1188	}	1783	}
1189		1784
…		…
1191		1786
1192	if (dents)	1787	if (dents)
1193	{	1788	{
1194	struct eio_dirent *ent;	1789	struct eio_dirent *ent;
1195		1790
1196	if (expect_false (dentoffs == dentalloc))	1791	if (ecb_expect_false (dentoffs == dentalloc))
1197	{	1792	{
1198	dentalloc *= 2;	1793	dentalloc *= 2;
1199	X_LOCK (wrklock);
1200	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1794	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1201	X_UNLOCK (wrklock);
1202		1795
1203	if (!dents)	1796	if (!dents)
1204	break;	1797	break;
1205	}	1798	}
1206		1799
1207	ent = dents + dentoffs;	1800	ent = dents + dentoffs;
1208		1801
1209	ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */	1802	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1210	ent->namelen = len - 1;	1803	ent->namelen = len - 1;
1211	ent->inode = D_INO (entp);	1804	ent->inode = D_INO (entp);
		1805
		1806	inode_bits |= ent->inode;
1212		1807
1213	switch (D_TYPE (entp))	1808	switch (D_TYPE (entp))
1214	{	1809	{
1215	default:	1810	default:
1216	ent->type = EIO_DT_UNKNOWN;	1811	ent->type = EIO_DT_UNKNOWN;
…		…
1259	#ifdef DT_WHT	1854	#ifdef DT_WHT
1260	case DT_WHT: ent->type = EIO_DT_WHT; break;	1855	case DT_WHT: ent->type = EIO_DT_WHT; break;
1261	#endif	1856	#endif
1262	}	1857	}
1263		1858
1264	ent->score = 0;	1859	ent->score = 7;
1265		1860
1266	if (flags & EIO_READDIR_DIRS_FIRST)	1861	if (flags & EIO_READDIR_DIRS_FIRST)
1267	{	1862	{
1268	if (ent->type == EIO_DT_UNKNOWN)	1863	if (ent->type == EIO_DT_UNKNOWN)
1269	{	1864	{
1270	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */	1865	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1271	ent->score = 98;	1866	ent->score = 1;
1272	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */	1867	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1273	ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */	1868	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1274	}	1869	}
1275	else if (ent->type == EIO_DT_DIR)	1870	else if (ent->type == EIO_DT_DIR)
1276	ent->score = 100;	1871	ent->score = 0;
1277	}	1872	}
1278	}	1873	}
1279		1874
1280	namesoffs += len;	1875	namesoffs += len;
1281	++dentoffs;	1876	++dentoffs;
…		…
1284	if (EIO_CANCELLED (req))	1879	if (EIO_CANCELLED (req))
1285	{	1880	{
1286	errno = ECANCELED;	1881	errno = ECANCELED;
1287	break;	1882	break;
1288	}	1883	}
		1884
		1885	#ifdef _WIN32
		1886	if (!FindNextFile (dirp, &entp))
		1887	{
		1888	FindClose (dirp);
		1889	dirp = 0;
		1890	}
		1891	#endif
1289	}	1892	}
1290	}
1291
1292	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1293	# undef msync
1294	# define msync(a,b,c) ((errno = ENOSYS), -1)
1295	#endif
1296
1297	int
1298	eio__mtouch (void *mem, size_t len, int flags)
1299	{
1300	intptr_t addr = (intptr_t)mem;
1301	intptr_t end = addr + len;
1302	#ifdef PAGESIZE
1303	const intptr_t page = PAGESIZE;
1304	#else
1305	static intptr_t page;
1306
1307	if (!page)
1308	page = sysconf (_SC_PAGESIZE);
1309	#endif
1310
1311	addr &= ~(page - 1); /* assume page size is always a power of two */
1312
1313	if (addr < end)
1314	if (flags) /* modify */
1315	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1316	else
1317	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1318
1319	return 0;
1320	}	1893	}
1321		1894
1322	/*****************************************************************************/	1895	/*****************************************************************************/
1323		1896
1324	#define ALLOC(len) \	1897	#define ALLOC(len) \
…		…
1340	{	1913	{
1341	ETP_REQ *req;	1914	ETP_REQ *req;
1342	struct timespec ts;	1915	struct timespec ts;
1343	etp_worker *self = (etp_worker *)thr_arg;	1916	etp_worker *self = (etp_worker *)thr_arg;
1344		1917
1345	/* try to distribute timeouts somewhat randomly */	1918	/* try to distribute timeouts somewhat evenly */
1346	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1919	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1347		1920
1348	for (;;)	1921	for (;;)
1349	{	1922	{
		1923	ts.tv_sec = 0;
		1924
1350	X_LOCK (reqlock);	1925	X_LOCK (reqlock);
1351		1926
1352	for (;;)	1927	for (;;)
1353	{	1928	{
1354	self->req = req = reqq_shift (&req_queue);	1929	self->req = req = reqq_shift (&req_queue);
1355		1930
1356	if (req)	1931	if (req)
1357	break;	1932	break;
1358		1933
		1934	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1935	{
		1936	X_UNLOCK (reqlock);
		1937	X_LOCK (wrklock);
		1938	--started;
		1939	X_UNLOCK (wrklock);
		1940	goto quit;
		1941	}
		1942
1359	++idle;	1943	++idle;
1360		1944
1361	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1945	if (idle <= max_idle)
1362	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1946	/* we are allowed to idle, so do so without any timeout */
		1947	X_COND_WAIT (reqwait, reqlock);
		1948	else
1363	{	1949	{
1364	if (idle > max_idle)	1950	/* initialise timeout once */
1365	{	1951	if (!ts.tv_sec)
1366	--idle;	1952	ts.tv_sec = time (0) + idle_timeout;
1367	X_UNLOCK (reqlock);
1368	X_LOCK (wrklock);
1369	--started;
1370	X_UNLOCK (wrklock);
1371	goto quit;
1372	}
1373		1953
1374	/* we are allowed to idle, so do so without any timeout */
1375	X_COND_WAIT (reqwait, reqlock);	1954	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1955	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1376	}	1956	}
1377		1957
1378	--idle;	1958	--idle;
1379	}	1959	}
1380		1960
…		…
1383	X_UNLOCK (reqlock);	1963	X_UNLOCK (reqlock);
1384		1964
1385	if (req->type < 0)	1965	if (req->type < 0)
1386	goto quit;	1966	goto quit;
1387		1967
1388	if (!EIO_CANCELLED (req))
1389	ETP_EXECUTE (self, req);	1968	ETP_EXECUTE (self, req);
1390		1969
1391	X_LOCK (reslock);	1970	X_LOCK (reslock);
1392		1971
1393	++npending;	1972	++npending;
1394		1973
…		…
1409	return 0;	1988	return 0;
1410	}	1989	}
1411		1990
1412	/*****************************************************************************/	1991	/*****************************************************************************/
1413		1992
		1993	int ecb_cold
1414	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1994	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1415	{	1995	{
		1996	#if !HAVE_PREADWRITE
		1997	X_MUTEX_CREATE (preadwritelock);
		1998	#endif
		1999
1416	return etp_init (want_poll, done_poll);	2000	return etp_init (want_poll, done_poll);
1417	}	2001	}
1418		2002
		2003	ecb_inline void
1419	static void eio_api_destroy (eio_req *req)	2004	eio_api_destroy (eio_req *req)
1420	{	2005	{
1421	free (req);	2006	free (req);
1422	}	2007	}
1423		2008
1424	#define REQ(rtype) \	2009	#define REQ(rtype) \
…		…
1443	{ \	2028	{ \
1444	eio_api_destroy (req); \	2029	eio_api_destroy (req); \
1445	return 0; \	2030	return 0; \
1446	}	2031	}
1447		2032
		2033	static void
1448	static void eio_execute (etp_worker *self, eio_req *req)	2034	eio_execute (etp_worker *self, eio_req *req)
1449	{	2035	{
1450	errno = 0;	2036	if (ecb_expect_false (EIO_CANCELLED (req)))
		2037	{
		2038	req->result = -1;
		2039	req->errorno = ECANCELED;
		2040	return;
		2041	}
1451		2042
1452	switch (req->type)	2043	switch (req->type)
1453	{	2044	{
1454	case EIO_READ: ALLOC (req->size);	2045	case EIO_READ: ALLOC (req->size);
1455	req->result = req->offs >= 0	2046	req->result = req->offs >= 0
…		…
1458	case EIO_WRITE: req->result = req->offs >= 0	2049	case EIO_WRITE: req->result = req->offs >= 0
1459	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2050	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1460	: write (req->int1, req->ptr2, req->size); break;	2051	: write (req->int1, req->ptr2, req->size); break;
1461		2052
1462	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2053	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1463	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2054	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1464		2055
1465	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2056	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1466	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2057	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1467	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2058	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1468	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2059	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1469	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2060	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1470	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2061	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2062
		2063	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2064	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2065	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2066	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1471		2067
1472	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	2068	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1473	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	2069	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1474	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	2070	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1475	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	2071	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
…		…
1483	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	2079	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1484	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	2080	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1485	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2081	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1486	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2082	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1487	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2083	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1488	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;	2084	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1489		2085
		2086	case EIO_REALPATH: eio__realpath (req, self); break;
		2087
1490	case EIO_READLINK: ALLOC (NAME_MAX);	2088	case EIO_READLINK: ALLOC (PATH_MAX);
1491	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	2089	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1492		2090
1493	case EIO_SYNC: req->result = 0; sync (); break;	2091	case EIO_SYNC: req->result = 0; sync (); break;
1494	case EIO_FSYNC: req->result = fsync (req->int1); break;	2092	case EIO_FSYNC: req->result = fsync (req->int1); break;
1495	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2093	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1496	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;	2094	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
1497	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1498	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	2095	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2096	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2097	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		2098	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2099	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		2100	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1499		2101
1500	case EIO_READDIR: eio__scandir (req, self); break;	2102	case EIO_READDIR: eio__scandir (req, self); break;
1501		2103
1502	case EIO_BUSY:	2104	case EIO_BUSY:
1503	#ifdef _WIN32	2105	#ifdef _WIN32
1504	Sleep (req->nv1 * 1000.);	2106	Sleep (req->nv1 * 1e3);
1505	#else	2107	#else
1506	{	2108	{
1507	struct timeval tv;	2109	struct timeval tv;
1508		2110
1509	tv.tv_sec = req->nv1;	2111	tv.tv_sec = req->nv1;
1510	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	2112	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1511		2113
1512	req->result = select (0, 0, 0, 0, &tv);	2114	req->result = select (0, 0, 0, 0, &tv);
1513	}	2115	}
1514	#endif	2116	#endif
1515	break;	2117	break;
…		…
1530	times = tv;	2132	times = tv;
1531	}	2133	}
1532	else	2134	else
1533	times = 0;	2135	times = 0;
1534		2136
1535
1536	req->result = req->type == EIO_FUTIME	2137	req->result = req->type == EIO_FUTIME
1537	? futimes (req->int1, times)	2138	? futimes (req->int1, times)
1538	: utimes (req->ptr1, times);	2139	: utimes (req->ptr1, times);
1539	}	2140	}
1540	break;	2141	break;
…		…
1545	case EIO_NOP:	2146	case EIO_NOP:
1546	req->result = 0;	2147	req->result = 0;
1547	break;	2148	break;
1548		2149
1549	case EIO_CUSTOM:	2150	case EIO_CUSTOM:
1550	((void (*)(eio_req *))req->feed) (req);	2151	req->feed (req);
1551	break;	2152	break;
1552		2153
1553	default:	2154	default:
		2155	errno = ENOSYS;
1554	req->result = -1;	2156	req->result = -1;
1555	break;	2157	break;
1556	}	2158	}
1557		2159
1558	req->errorno = errno;	2160	req->errorno = errno;
…		…
1583	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2185	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1584	{	2186	{
1585	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2187	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1586	}	2188	}
1587		2189
		2190	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		2191	{
		2192	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2193	}
		2194
		2195	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2196	{
		2197	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2198	}
		2199
		2200	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2201	{
		2202	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2203	}
		2204
1588	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2205	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1589	{	2206	{
1590	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2207	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1591	}	2208	}
1592		2209
1593	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2210	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1594	{	2211	{
1595	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2212	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1596	}	2213	}
1597		2214
1598	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2215	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1599	{	2216	{
1600	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2217	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2218	}
		2219
		2220	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2221	{
		2222	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1601	}	2223	}
1602		2224
1603	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2225	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1604	{	2226	{
1605	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2227	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
1623	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2245	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1624	{	2246	{
1625	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2247	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1626	}	2248	}
1627		2249
		2250	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2251	{
		2252	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2253	}
		2254
1628	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2255	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1629	{	2256	{
1630	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2257	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1631	}	2258	}
1632		2259
…		…
1638	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2265	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1639	{	2266	{
1640	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2267	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1641	}	2268	}
1642		2269
1643	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2270	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1644	{	2271	{
1645	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2272	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1646	}	2273	}
1647		2274
1648	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2275	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1668	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2295	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1669	{	2296	{
1670	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2297	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1671	}	2298	}
1672		2299
1673	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2300	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1674	{	2301	{
1675	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2302	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1676	}	2303	}
1677		2304
1678	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2305	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1694	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2321	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1695	{	2322	{
1696	return eio__1path (EIO_READLINK, path, pri, cb, data);	2323	return eio__1path (EIO_READLINK, path, pri, cb, data);
1697	}	2324	}
1698		2325
		2326	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2327	{
		2328	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2329	}
		2330
1699	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2331	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1700	{	2332	{
1701	return eio__1path (EIO_STAT, path, pri, cb, data);	2333	return eio__1path (EIO_STAT, path, pri, cb, data);
1702	}	2334	}
1703		2335
1704	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2336	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1705	{	2337	{
1706	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2338	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1707	}	2339	}
1708		2340
		2341	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2342	{
		2343	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2344	}
		2345
1709	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2346	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1710	{	2347	{
1711	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2348	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1712	}	2349	}
1713		2350
…		…
1721	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2358	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1722	}	2359	}
1723		2360
1724	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2361	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1725	{	2362	{
1726	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2363	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1727	}	2364	}
1728		2365
1729	static eio_req *	2366	static eio_req *
1730	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2367	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1731	{	2368	{
…		…
1755	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2392	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1756	{	2393	{
1757	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2394	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1758	}	2395	}
1759		2396
1760	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2397	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1761	{	2398	{
1762	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2399	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1763	}	2400	}
1764		2401
1765	#endif	2402	#endif
1766		2403
1767	eio_req *eio_grp (eio_cb cb, void *data)	2404	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1776	#undef SEND	2413	#undef SEND
1777		2414
1778	/*****************************************************************************/	2415	/*****************************************************************************/
1779	/* grp functions */	2416	/* grp functions */
1780		2417
		2418	void
1781	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2419	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1782	{	2420	{
1783	grp->int2 = limit;	2421	grp->int2 = limit;
1784	grp->feed = feed;	2422	grp->feed = feed;
1785		2423
1786	grp_try_feed (grp);	2424	grp_try_feed (grp);
1787	}	2425	}
1788		2426
		2427	void
1789	void eio_grp_limit (eio_req *grp, int limit)	2428	eio_grp_limit (eio_req *grp, int limit)
1790	{	2429	{
1791	grp->int2 = limit;	2430	grp->int2 = limit;
1792		2431
1793	grp_try_feed (grp);	2432	grp_try_feed (grp);
1794	}	2433	}
1795		2434
		2435	void
1796	void eio_grp_add (eio_req *grp, eio_req *req)	2436	eio_grp_add (eio_req *grp, eio_req *req)
1797	{	2437	{
1798	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2438	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1799		2439
1800	grp->flags |= EIO_FLAG_GROUPADD;	2440	grp->flags |= EIO_FLAG_GROUPADD;
1801		2441
…		…
1812	}	2452	}
1813		2453
1814	/*****************************************************************************/	2454	/*****************************************************************************/
1815	/* misc garbage */	2455	/* misc garbage */
1816		2456
		2457	eio_ssize_t
1817	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2458	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1818	{	2459	{
1819	etp_worker wrk;
1820	ssize_t ret;
1821
1822	wrk.dbuf = 0;
1823
1824	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2460	return eio__sendfile (ofd, ifd, offset, count);
1825
1826	if (wrk.dbuf)
1827	free (wrk.dbuf);
1828
1829	return ret;
1830	}	2461	}
1831		2462

Diff Legend

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