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 1.98 by root, Sun Jul 24 05:53:34 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	204	#endif
92		205
93	#ifdef _D_EXACT_NAMLEN	206	#if HAVE_SYS_SYSCALL_H
94	# undef D_NAMLEN	207	# include <sys/syscall.h>
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	208	#endif
101		209
102	# ifndef EIO_STRUCT_DIRENT	210	#if HAVE_SYS_PRCTL_H
103	# define EIO_STRUCT_DIRENT struct dirent	211	# include <sys/prctl.h>
104	# endif
105
106	#endif	212	#endif
107		213
108	#if HAVE_SENDFILE	214	#if HAVE_SENDFILE
109	# if __linux	215	# if __linux
110	# include <sys/sendfile.h>	216	# include <sys/sendfile.h>
111	# elif __freebsd	217	# elif __FreeBSD__ || defined __APPLE__
112	# include <sys/socket.h>	218	# include <sys/socket.h>
113	# include <sys/uio.h>	219	# include <sys/uio.h>
114	# elif __hpux	220	# elif __hpux
115	# include <sys/socket.h>	221	# include <sys/socket.h>
116	# elif __solaris /* not yet */	222	# elif __solaris
117	# include <sys/sendfile.h>	223	# include <sys/sendfile.h>
118	# else	224	# else
119	# error sendfile support requested but not available	225	# error sendfile support requested but not available
120	# endif	226	# endif
121	#endif	227	#endif
…		…
125	#endif	231	#endif
126	#ifndef D_INO	232	#ifndef D_INO
127	# define D_INO(de) 0	233	# define D_INO(de) 0
128	#endif	234	#endif
129	#ifndef D_NAMLEN	235	#ifndef D_NAMLEN
130	# define D_NAMLEN(de) strlen ((de)->d_name)	236	# define D_NAMLEN(entp) strlen (D_NAME (entp))
131	#endif	237	#endif
132
133	/* number of seconds after which an idle threads exit */
134	#define IDLE_TIMEOUT 10
135		238
136	/* used for struct dirent, AIX doesn't provide it */	239	/* used for struct dirent, AIX doesn't provide it */
137	#ifndef NAME_MAX	240	#ifndef NAME_MAX
138	# define NAME_MAX 4096	241	# define NAME_MAX 4096
139	#endif	242	#endif
140		243
		244	/* used for readlink etc. */
		245	#ifndef PATH_MAX
		246	# define PATH_MAX 4096
		247	#endif
		248
141	/* buffer size for various temporary buffers */	249	/* buffer size for various temporary buffers */
142	#define EIO_BUFSIZE 65536	250	#define EIO_BUFSIZE 65536
143		251
144	#define dBUF \	252	#define dBUF \
145	char *eio_buf; \
146	ETP_WORKER_LOCK (self); \
147	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	253	char *eio_buf = malloc (EIO_BUFSIZE); \
148	ETP_WORKER_UNLOCK (self); \
149	errno = ENOMEM; \	254	errno = ENOMEM; \
150	if (!eio_buf) \	255	if (!eio_buf) \
151	return -1;	256	return -1
		257
		258	#define FUBd \
		259	free (eio_buf)
152		260
153	#define EIO_TICKS ((1000000 + 1023) >> 10)	261	#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		262
168	#define ETP_PRI_MIN EIO_PRI_MIN	263	#define ETP_PRI_MIN EIO_PRI_MIN
169	#define ETP_PRI_MAX EIO_PRI_MAX	264	#define ETP_PRI_MAX EIO_PRI_MAX
170		265
171	struct etp_worker;	266	struct etp_worker;
…		…
175	static int eio_finish (eio_req *req);	270	static int eio_finish (eio_req *req);
176	#define ETP_FINISH(req) eio_finish (req)	271	#define ETP_FINISH(req) eio_finish (req)
177	static void eio_execute (struct etp_worker *self, eio_req *req);	272	static void eio_execute (struct etp_worker *self, eio_req *req);
178	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	273	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
179		274
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	/*****************************************************************************/	275	/*****************************************************************************/
198		276
199	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	277	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200		278
201	/* calculcate time difference in ~1/EIO_TICKS of a second */	279	/* calculate time difference in ~1/EIO_TICKS of a second */
		280	ecb_inline int
202	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	281	tvdiff (struct timeval *tv1, struct timeval *tv2)
203	{	282	{
204	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	283	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	284	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
206	}	285	}
207		286
…		…
214	static unsigned int max_poll_reqs; /* reslock */	293	static unsigned int max_poll_reqs; /* reslock */
215		294
216	static volatile unsigned int nreqs; /* reqlock */	295	static volatile unsigned int nreqs; /* reqlock */
217	static volatile unsigned int nready; /* reqlock */	296	static volatile unsigned int nready; /* reqlock */
218	static volatile unsigned int npending; /* reqlock */	297	static volatile unsigned int npending; /* reqlock */
219	static volatile unsigned int max_idle = 4;	298	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		299	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
220		300
221	static mutex_t wrklock = X_MUTEX_INIT;	301	static xmutex_t wrklock;
222	static mutex_t reslock = X_MUTEX_INIT;	302	static xmutex_t reslock;
223	static mutex_t reqlock = X_MUTEX_INIT;	303	static xmutex_t reqlock;
224	static cond_t reqwait = X_COND_INIT;	304	static xcond_t reqwait;
225		305
226	#if !HAVE_PREADWRITE	306	#if !HAVE_PREADWRITE
227	/*	307	/*
228	* make our pread/pwrite emulation safe against themselves, but not against	308	* make our pread/pwrite emulation safe against themselves, but not against
229	* normal read/write by using a mutex. slows down execution a lot,	309	* normal read/write by using a mutex. slows down execution a lot,
230	* but that's your problem, not mine.	310	* but that's your problem, not mine.
231	*/	311	*/
232	static mutex_t preadwritelock = X_MUTEX_INIT;	312	static xmutex_t preadwritelock;
233	#endif	313	#endif
234		314
235	typedef struct etp_worker	315	typedef struct etp_worker
236	{	316	{
237	/* locked by wrklock */	317	/* locked by wrklock */
238	struct etp_worker *prev, *next;	318	struct etp_worker *prev, *next;
239		319
240	thread_t tid;	320	xthread_t tid;
241		321
242	/* locked by reslock, reqlock or wrklock */	322	/* locked by reslock, reqlock or wrklock */
243	ETP_REQ *req; /* currently processed request */	323	ETP_REQ *req; /* currently processed request */
244		324
		325	#ifdef ETP_WORKER_COMMON
245	ETP_WORKER_COMMON	326	ETP_WORKER_COMMON
		327	#endif
246	} etp_worker;	328	} etp_worker;
247		329
248	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	330	static etp_worker wrk_first; /* NOT etp */
249		331
250	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	332	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
251	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	333	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
252		334
253	/* worker threads management */	335	/* worker threads management */
254		336
		337	static void ecb_cold
255	static void etp_worker_clear (etp_worker *wrk)	338	etp_worker_clear (etp_worker *wrk)
256	{	339	{
257	ETP_WORKER_CLEAR (wrk);
258	}	340	}
259		341
		342	static void ecb_cold
260	static void etp_worker_free (etp_worker *wrk)	343	etp_worker_free (etp_worker *wrk)
261	{	344	{
262	wrk->next->prev = wrk->prev;	345	wrk->next->prev = wrk->prev;
263	wrk->prev->next = wrk->next;	346	wrk->prev->next = wrk->next;
264		347
265	free (wrk);	348	free (wrk);
266	}	349	}
267		350
268	static unsigned int etp_nreqs (void)	351	static unsigned int
		352	etp_nreqs (void)
269	{	353	{
270	int retval;	354	int retval;
271	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	355	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
272	retval = nreqs;	356	retval = nreqs;
273	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	357	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
274	return retval;	358	return retval;
275	}	359	}
276		360
277	static unsigned int etp_nready (void)	361	static unsigned int
		362	etp_nready (void)
278	{	363	{
279	unsigned int retval;	364	unsigned int retval;
280		365
281	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	366	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282	retval = nready;	367	retval = nready;
283	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	368	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284		369
285	return retval;	370	return retval;
286	}	371	}
287		372
288	static unsigned int etp_npending (void)	373	static unsigned int
		374	etp_npending (void)
289	{	375	{
290	unsigned int retval;	376	unsigned int retval;
291		377
292	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	378	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
293	retval = npending;	379	retval = npending;
294	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	380	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
295		381
296	return retval;	382	return retval;
297	}	383	}
298		384
299	static unsigned int etp_nthreads (void)	385	static unsigned int
		386	etp_nthreads (void)
300	{	387	{
301	unsigned int retval;	388	unsigned int retval;
302		389
303	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	390	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
304	retval = started;	391	retval = started;
…		…
318	} etp_reqq;	405	} etp_reqq;
319		406
320	static etp_reqq req_queue;	407	static etp_reqq req_queue;
321	static etp_reqq res_queue;	408	static etp_reqq res_queue;
322		409
		410	static void ecb_noinline ecb_cold
		411	reqq_init (etp_reqq *q)
		412	{
		413	int pri;
		414
		415	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		416	q->qs[pri] = q->qe[pri] = 0;
		417
		418	q->size = 0;
		419	}
		420
		421	static int ecb_noinline
323	static int reqq_push (etp_reqq *q, ETP_REQ *req)	422	reqq_push (etp_reqq *q, ETP_REQ *req)
324	{	423	{
325	int pri = req->pri;	424	int pri = req->pri;
326	req->next = 0;	425	req->next = 0;
327		426
328	if (q->qe[pri])	427	if (q->qe[pri])
…		…
334	q->qe[pri] = q->qs[pri] = req;	433	q->qe[pri] = q->qs[pri] = req;
335		434
336	return q->size++;	435	return q->size++;
337	}	436	}
338		437
		438	static ETP_REQ * ecb_noinline
339	static ETP_REQ *reqq_shift (etp_reqq *q)	439	reqq_shift (etp_reqq *q)
340	{	440	{
341	int pri;	441	int pri;
342		442
343	if (!q->size)	443	if (!q->size)
344	return 0;	444	return 0;
…		…
359	}	459	}
360		460
361	abort ();	461	abort ();
362	}	462	}
363		463
364	static void etp_atfork_prepare (void)	464	static int ecb_cold
		465	etp_init (void (*want_poll)(void), void (*done_poll)(void))
365	{	466	{
366	X_LOCK (wrklock);	467	X_MUTEX_CREATE (wrklock);
367	X_LOCK (reqlock);	468	X_MUTEX_CREATE (reslock);
368	X_LOCK (reslock);	469	X_MUTEX_CREATE (reqlock);
369	#if !HAVE_PREADWRITE	470	X_COND_CREATE (reqwait);
370	X_LOCK (preadwritelock);
371	#endif
372	}
373		471
374	static void etp_atfork_parent (void)	472	reqq_init (&req_queue);
375	{	473	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		474
384	static void etp_atfork_child (void)	475	wrk_first.next =
385	{	476	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		477
405	started = 0;	478	started = 0;
406	idle = 0;	479	idle = 0;
407	nreqs = 0;	480	nreqs = 0;
408	nready = 0;	481	nready = 0;
409	npending = 0;	482	npending = 0;
410		483
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;	484	want_poll_cb = want_poll;
428	done_poll_cb = done_poll;	485	done_poll_cb = done_poll;
429		486
430	return 0;	487	return 0;
431	}	488	}
432		489
433	X_THREAD_PROC (etp_proc);	490	X_THREAD_PROC (etp_proc);
434		491
		492	static void ecb_cold
435	static void etp_start_thread (void)	493	etp_start_thread (void)
436	{	494	{
437	etp_worker *wrk = calloc (1, sizeof (etp_worker));	495	etp_worker *wrk = calloc (1, sizeof (etp_worker));
438		496
439	/*TODO*/	497	/*TODO*/
440	assert (("unable to allocate worker thread data", wrk));	498	assert (("unable to allocate worker thread data", wrk));
…		…
453	free (wrk);	511	free (wrk);
454		512
455	X_UNLOCK (wrklock);	513	X_UNLOCK (wrklock);
456	}	514	}
457		515
		516	static void
458	static void etp_maybe_start_thread (void)	517	etp_maybe_start_thread (void)
459	{	518	{
460	if (expect_true (etp_nthreads () >= wanted))	519	if (ecb_expect_true (etp_nthreads () >= wanted))
461	return;	520	return;
462		521
463	/* todo: maybe use idle here, but might be less exact */	522	/* todo: maybe use idle here, but might be less exact */
464	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	523	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
465	return;	524	return;
466		525
467	etp_start_thread ();	526	etp_start_thread ();
468	}	527	}
469		528
		529	static void ecb_cold
470	static void etp_end_thread (void)	530	etp_end_thread (void)
471	{	531	{
472	eio_req *req = calloc (1, sizeof (eio_req));	532	eio_req *req = calloc (1, sizeof (eio_req));
473		533
474	req->type = -1;	534	req->type = -1;
475	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	535	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
482	X_LOCK (wrklock);	542	X_LOCK (wrklock);
483	--started;	543	--started;
484	X_UNLOCK (wrklock);	544	X_UNLOCK (wrklock);
485	}	545	}
486		546
487	static int etp_poll (void)	547	static int
		548	etp_poll (void)
488	{	549	{
489	unsigned int maxreqs;	550	unsigned int maxreqs;
490	unsigned int maxtime;	551	unsigned int maxtime;
491	struct timeval tv_start, tv_now;	552	struct timeval tv_start, tv_now;
492		553
…		…
522		583
523	X_LOCK (reqlock);	584	X_LOCK (reqlock);
524	--nreqs;	585	--nreqs;
525	X_UNLOCK (reqlock);	586	X_UNLOCK (reqlock);
526		587
527	if (expect_false (req->type == EIO_GROUP && req->size))	588	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
528	{	589	{
529	req->int1 = 1; /* mark request as delayed */	590	req->int1 = 1; /* mark request as delayed */
530	continue;	591	continue;
531	}	592	}
532	else	593	else
533	{	594	{
534	int res = ETP_FINISH (req);	595	int res = ETP_FINISH (req);
535	if (expect_false (res))	596	if (ecb_expect_false (res))
536	return res;	597	return res;
537	}	598	}
538		599
539	if (expect_false (maxreqs && !--maxreqs))	600	if (ecb_expect_false (maxreqs && !--maxreqs))
540	break;	601	break;
541		602
542	if (maxtime)	603	if (maxtime)
543	{	604	{
544	gettimeofday (&tv_now, 0);	605	gettimeofday (&tv_now, 0);
…		…
550		611
551	errno = EAGAIN;	612	errno = EAGAIN;
552	return -1;	613	return -1;
553	}	614	}
554		615
		616	static void
555	static void etp_cancel (ETP_REQ *req)	617	etp_cancel (ETP_REQ *req)
556	{	618	{
557	X_LOCK (wrklock);	619	req->cancelled = 1;
558	req->flags |= EIO_FLAG_CANCELLED;
559	X_UNLOCK (wrklock);
560		620
561	eio_grp_cancel (req);	621	eio_grp_cancel (req);
562	}	622	}
563		623
		624	static void
564	static void etp_submit (ETP_REQ *req)	625	etp_submit (ETP_REQ *req)
565	{	626	{
566	req->pri -= ETP_PRI_MIN;	627	req->pri -= ETP_PRI_MIN;
567		628
568	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	629	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;	630	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
570		631
571	if (expect_false (req->type == EIO_GROUP))	632	if (ecb_expect_false (req->type == EIO_GROUP))
572	{	633	{
573	/* I hope this is worth it :/ */	634	/* I hope this is worth it :/ */
574	X_LOCK (reqlock);	635	X_LOCK (reqlock);
575	++nreqs;	636	++nreqs;
576	X_UNLOCK (reqlock);	637	X_UNLOCK (reqlock);
…		…
595		656
596	etp_maybe_start_thread ();	657	etp_maybe_start_thread ();
597	}	658	}
598	}	659	}
599		660
		661	static void ecb_cold
600	static void etp_set_max_poll_time (double nseconds)	662	etp_set_max_poll_time (double nseconds)
601	{	663	{
602	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	664	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603	max_poll_time = nseconds;	665	max_poll_time = nseconds * EIO_TICKS;
604	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	666	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605	}	667	}
606		668
		669	static void ecb_cold
607	static void etp_set_max_poll_reqs (unsigned int maxreqs)	670	etp_set_max_poll_reqs (unsigned int maxreqs)
608	{	671	{
609	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	672	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
610	max_poll_reqs = maxreqs;	673	max_poll_reqs = maxreqs;
611	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	674	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
612	}	675	}
613		676
		677	static void ecb_cold
614	static void etp_set_max_idle (unsigned int nthreads)	678	etp_set_max_idle (unsigned int nthreads)
615	{	679	{
616	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	680	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617	max_idle = nthreads <= 0 ? 1 : nthreads;	681	max_idle = nthreads;
618	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	682	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619	}	683	}
620		684
		685	static void ecb_cold
		686	etp_set_idle_timeout (unsigned int seconds)
		687	{
		688	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		689	idle_timeout = seconds;
		690	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		691	}
		692
		693	static void ecb_cold
621	static void etp_set_min_parallel (unsigned int nthreads)	694	etp_set_min_parallel (unsigned int nthreads)
622	{	695	{
623	if (wanted < nthreads)	696	if (wanted < nthreads)
624	wanted = nthreads;	697	wanted = nthreads;
625	}	698	}
626		699
		700	static void ecb_cold
627	static void etp_set_max_parallel (unsigned int nthreads)	701	etp_set_max_parallel (unsigned int nthreads)
628	{	702	{
629	if (wanted > nthreads)	703	if (wanted > nthreads)
630	wanted = nthreads;	704	wanted = nthreads;
631		705
632	while (started > wanted)	706	while (started > wanted)
633	etp_end_thread ();	707	etp_end_thread ();
634	}	708	}
635		709
636	/*****************************************************************************/	710	/*****************************************************************************/
637		711
		712	static void
638	static void grp_try_feed (eio_req *grp)	713	grp_try_feed (eio_req *grp)
639	{	714	{
640	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	715	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641	{	716	{
642	grp->flags &= ~EIO_FLAG_GROUPADD;	717	grp->flags &= ~EIO_FLAG_GROUPADD;
643		718
…		…
650	break;	725	break;
651	}	726	}
652	}	727	}
653	}	728	}
654		729
		730	static int
655	static int grp_dec (eio_req *grp)	731	grp_dec (eio_req *grp)
656	{	732	{
657	--grp->size;	733	--grp->size;
658		734
659	/* call feeder, if applicable */	735	/* call feeder, if applicable */
660	grp_try_feed (grp);	736	grp_try_feed (grp);
…		…
664	return eio_finish (grp);	740	return eio_finish (grp);
665	else	741	else
666	return 0;	742	return 0;
667	}	743	}
668		744
		745	static void
669	void eio_destroy (eio_req *req)	746	eio_destroy (eio_req *req)
670	{	747	{
671	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	748	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	749	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673		750
674	EIO_DESTROY (req);	751	EIO_DESTROY (req);
675	}	752	}
676		753
		754	static int
677	static int eio_finish (eio_req *req)	755	eio_finish (eio_req *req)
678	{	756	{
679	int res = EIO_FINISH (req);	757	int res = EIO_FINISH (req);
680		758
681	if (req->grp)	759	if (req->grp)
682	{	760	{
…		…
690	if (grp->grp_first == req)	768	if (grp->grp_first == req)
691	grp->grp_first = req->grp_next;	769	grp->grp_first = req->grp_next;
692		770
693	res2 = grp_dec (grp);	771	res2 = grp_dec (grp);
694		772
695	if (!res && res2)	773	if (!res)
696	res = res2;	774	res = res2;
697	}	775	}
698		776
699	eio_destroy (req);	777	eio_destroy (req);
700		778
701	return res;	779	return res;
702	}	780	}
703		781
		782	void
704	void eio_grp_cancel (eio_req *grp)	783	eio_grp_cancel (eio_req *grp)
705	{	784	{
706	for (grp = grp->grp_first; grp; grp = grp->grp_next)	785	for (grp = grp->grp_first; grp; grp = grp->grp_next)
707	eio_cancel (grp);	786	eio_cancel (grp);
708	}	787	}
709		788
		789	void
710	void eio_cancel (eio_req *req)	790	eio_cancel (eio_req *req)
711	{	791	{
712	etp_cancel (req);	792	etp_cancel (req);
713	}	793	}
714		794
		795	void
715	void eio_submit (eio_req *req)	796	eio_submit (eio_req *req)
716	{	797	{
717	etp_submit (req);	798	etp_submit (req);
718	}	799	}
719		800
720	unsigned int eio_nreqs (void)	801	unsigned int
		802	eio_nreqs (void)
721	{	803	{
722	return etp_nreqs ();	804	return etp_nreqs ();
723	}	805	}
724		806
725	unsigned int eio_nready (void)	807	unsigned int
		808	eio_nready (void)
726	{	809	{
727	return etp_nready ();	810	return etp_nready ();
728	}	811	}
729		812
730	unsigned int eio_npending (void)	813	unsigned int
		814	eio_npending (void)
731	{	815	{
732	return etp_npending ();	816	return etp_npending ();
733	}	817	}
734		818
735	unsigned int eio_nthreads (void)	819	unsigned int ecb_cold
		820	eio_nthreads (void)
736	{	821	{
737	return etp_nthreads ();	822	return etp_nthreads ();
738	}	823	}
739		824
		825	void ecb_cold
740	void eio_set_max_poll_time (double nseconds)	826	eio_set_max_poll_time (double nseconds)
741	{	827	{
742	etp_set_max_poll_time (nseconds);	828	etp_set_max_poll_time (nseconds);
743	}	829	}
744		830
		831	void ecb_cold
745	void eio_set_max_poll_reqs (unsigned int maxreqs)	832	eio_set_max_poll_reqs (unsigned int maxreqs)
746	{	833	{
747	etp_set_max_poll_reqs (maxreqs);	834	etp_set_max_poll_reqs (maxreqs);
748	}	835	}
749		836
		837	void ecb_cold
750	void eio_set_max_idle (unsigned int nthreads)	838	eio_set_max_idle (unsigned int nthreads)
751	{	839	{
752	etp_set_max_idle (nthreads);	840	etp_set_max_idle (nthreads);
753	}	841	}
754		842
		843	void ecb_cold
		844	eio_set_idle_timeout (unsigned int seconds)
		845	{
		846	etp_set_idle_timeout (seconds);
		847	}
		848
		849	void ecb_cold
755	void eio_set_min_parallel (unsigned int nthreads)	850	eio_set_min_parallel (unsigned int nthreads)
756	{	851	{
757	etp_set_min_parallel (nthreads);	852	etp_set_min_parallel (nthreads);
758	}	853	}
759		854
		855	void ecb_cold
760	void eio_set_max_parallel (unsigned int nthreads)	856	eio_set_max_parallel (unsigned int nthreads)
761	{	857	{
762	etp_set_max_parallel (nthreads);	858	etp_set_max_parallel (nthreads);
763	}	859	}
764		860
765	int eio_poll (void)	861	int eio_poll (void)
…		…
774	# undef pread	870	# undef pread
775	# undef pwrite	871	# undef pwrite
776	# define pread eio__pread	872	# define pread eio__pread
777	# define pwrite eio__pwrite	873	# define pwrite eio__pwrite
778		874
779	static ssize_t	875	static eio_ssize_t
780	eio__pread (int fd, void *buf, size_t count, off_t offset)	876	eio__pread (int fd, void *buf, size_t count, off_t offset)
781	{	877	{
782	ssize_t res;	878	eio_ssize_t res;
783	off_t ooffset;	879	off_t ooffset;
784		880
785	X_LOCK (preadwritelock);	881	X_LOCK (preadwritelock);
786	ooffset = lseek (fd, 0, SEEK_CUR);	882	ooffset = lseek (fd, 0, SEEK_CUR);
787	lseek (fd, offset, SEEK_SET);	883	lseek (fd, offset, SEEK_SET);
…		…
790	X_UNLOCK (preadwritelock);	886	X_UNLOCK (preadwritelock);
791		887
792	return res;	888	return res;
793	}	889	}
794		890
795	static ssize_t	891	static eio_ssize_t
796	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	892	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
797	{	893	{
798	ssize_t res;	894	eio_ssize_t res;
799	off_t ooffset;	895	off_t ooffset;
800		896
801	X_LOCK (preadwritelock);	897	X_LOCK (preadwritelock);
802	ooffset = lseek (fd, 0, SEEK_CUR);	898	ooffset = lseek (fd, 0, SEEK_CUR);
803	lseek (fd, offset, SEEK_SET);	899	lseek (fd, offset, SEEK_SET);
…		…
807		903
808	return res;	904	return res;
809	}	905	}
810	#endif	906	#endif
811		907
812	#ifndef HAVE_FUTIMES	908	#ifndef HAVE_UTIMES
813		909
814	# undef utimes	910	# undef utimes
815	# undef futimes
816	# define utimes(path,times) eio__utimes (path, times)	911	# define utimes(path,times) eio__utimes (path, times)
817	# define futimes(fd,times) eio__futimes (fd, times)
818		912
819	static int	913	static int
820	eio__utimes (const char *filename, const struct timeval times[2])	914	eio__utimes (const char *filename, const struct timeval times[2])
821	{	915	{
822	if (times)	916	if (times)
…		…
830	}	924	}
831	else	925	else
832	return utime (filename, 0);	926	return utime (filename, 0);
833	}	927	}
834		928
		929	#endif
		930
		931	#ifndef HAVE_FUTIMES
		932
		933	# undef futimes
		934	# define futimes(fd,times) eio__futimes (fd, times)
		935
		936	static int
835	static int eio__futimes (int fd, const struct timeval tv[2])	937	eio__futimes (int fd, const struct timeval tv[2])
836	{	938	{
837	errno = ENOSYS;	939	errno = ENOSYS;
838	return -1;	940	return -1;
839	}	941	}
840		942
…		…
843	#if !HAVE_FDATASYNC	945	#if !HAVE_FDATASYNC
844	# undef fdatasync	946	# undef fdatasync
845	# define fdatasync(fd) fsync (fd)	947	# define fdatasync(fd) fsync (fd)
846	#endif	948	#endif
847		949
		950	static int
		951	eio__syncfs (int fd)
		952	{
		953	int res;
		954
		955	#if HAVE_SYS_SYNCFS
		956	res = (int)syscall (__NR_syncfs, (int)(fd));
		957	#else
		958	res = -1;
		959	errno = ENOSYS;
		960	#endif
		961
		962	if (res < 0 && errno == ENOSYS && fd >= 0)
		963	sync ();
		964
		965	return res;
		966	}
		967
848	/* sync_file_range always needs emulation */	968	/* sync_file_range always needs emulation */
849	int	969	static int
850	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	970	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
851	{	971	{
852	#if HAVE_SYNC_FILE_RANGE	972	#if HAVE_SYNC_FILE_RANGE
853	int res;	973	int res;
854		974
…		…
867	if (!res || errno != ENOSYS)	987	if (!res || errno != ENOSYS)
868	return res;	988	return res;
869	#endif	989	#endif
870		990
871	/* even though we could play tricks with the flags, it's better to always	991	/* 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 */	992	* call fdatasync, as that matches the expectation of its users best */
873	return fdatasync (fd);	993	return fdatasync (fd);
		994	}
		995
		996	static int
		997	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		998	{
		999	#if HAVE_FALLOCATE
		1000	return fallocate (fd, mode, offset, len);
		1001	#else
		1002	errno = ENOSYS;
		1003	return -1;
		1004	#endif
874	}	1005	}
875		1006
876	#if !HAVE_READAHEAD	1007	#if !HAVE_READAHEAD
877	# undef readahead	1008	# undef readahead
878	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1009	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
879		1010
880	static ssize_t	1011	static eio_ssize_t
881	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1012	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
882	{	1013	{
883	size_t todo = count;	1014	size_t todo = count;
884	dBUF;	1015	dBUF;
885		1016
…		…
890	pread (fd, eio_buf, len, offset);	1021	pread (fd, eio_buf, len, offset);
891	offset += len;	1022	offset += len;
892	todo -= len;	1023	todo -= len;
893	}	1024	}
894		1025
		1026	FUBd;
		1027
895	errno = 0;	1028	errno = 0;
896	return count;	1029	return count;
897	}	1030	}
898		1031
899	#endif	1032	#endif
900		1033
901	/* sendfile always needs emulation */	1034	/* sendfile always needs emulation */
902	static ssize_t	1035	static eio_ssize_t
903	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1036	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
904	{	1037	{
		1038	eio_ssize_t written = 0;
905	ssize_t res;	1039	eio_ssize_t res;
906		1040
907	if (!count)	1041	if (!count)
908	return 0;	1042	return 0;
909		1043
		1044	for (;;)
		1045	{
		1046	#ifdef __APPLE__
		1047	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1048	#endif
910	#if HAVE_SENDFILE	1049	#if HAVE_SENDFILE
911	# if __linux	1050	# if __linux
		1051	off_t soffset = offset;
912	res = sendfile (ofd, ifd, &offset, count);	1052	res = sendfile (ofd, ifd, &soffset, count);
913		1053
914	# elif __freebsd	1054	# elif __FreeBSD__
915	/*	1055	/*
916	* Of course, the freebsd sendfile is a dire hack with no thoughts	1056	* Of course, the freebsd sendfile is a dire hack with no thoughts
917	* wasted on making it similar to other I/O functions.	1057	* wasted on making it similar to other I/O functions.
918	*/	1058	*/
919	{
920	off_t sbytes;	1059	off_t sbytes;
921	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1060	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922		1061
923	if (res < 0 && sbytes)	1062	#if 0 /* according to the manpage, this is correct, but broken behaviour */
924	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1063	/* freebsd' sendfile will return 0 on success */
		1064	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1065	/* not on e.g. EIO or EPIPE - sounds broken */
		1066	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
925	res = sbytes;	1067	res = sbytes;
926	}	1068	#endif
		1069
		1070	/* according to source inspection, this is correct, and useful behaviour */
		1071	if (sbytes)
		1072	res = sbytes;
		1073
		1074	# elif defined (__APPLE__)
		1075	off_t sbytes = count;
		1076	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1077
		1078	/* according to the manpage, sbytes is always valid */
		1079	if (sbytes)
		1080	res = sbytes;
927		1081
928	# elif __hpux	1082	# elif __hpux
929	res = sendfile (ofd, ifd, offset, count, 0, 0);	1083	res = sendfile (ofd, ifd, offset, count, 0, 0);
930		1084
931	# elif __solaris	1085	# elif __solaris
932	{
933	struct sendfilevec vec;	1086	struct sendfilevec vec;
934	size_t sbytes;	1087	size_t sbytes;
935		1088
936	vec.sfv_fd = ifd;	1089	vec.sfv_fd = ifd;
937	vec.sfv_flag = 0;	1090	vec.sfv_flag = 0;
938	vec.sfv_off = offset;	1091	vec.sfv_off = offset;
939	vec.sfv_len = count;	1092	vec.sfv_len = count;
940		1093
941	res = sendfilev (ofd, &vec, 1, &sbytes);	1094	res = sendfilev (ofd, &vec, 1, &sbytes);
942		1095
943	if (res < 0 && sbytes)	1096	if (res < 0 && sbytes)
944	res = sbytes;	1097	res = sbytes;
945	}
946		1098
947	# endif	1099	# endif
		1100
		1101	#elif defined (_WIN32) && 0
		1102	/* does not work, just for documentation of what would need to be done */
		1103	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1104	/* libeio guarantees that the file offset does not change, and windows */
		1105	/* has no way to get an independent handle to the same file description */
		1106	HANDLE h = TO_SOCKET (ifd);
		1107	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1108	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1109
948	#else	1110	#else
949	res = -1;	1111	res = -1;
950	errno = ENOSYS;	1112	errno = ENOSYS;
951	#endif	1113	#endif
952		1114
		1115	/* we assume sendfile can copy at least 128mb in one go */
		1116	if (res <= 128 * 1024 * 1024)
		1117	{
		1118	if (res > 0)
		1119	written += res;
		1120
		1121	if (written)
		1122	return written;
		1123
		1124	break;
		1125	}
		1126	else
		1127	{
		1128	/* if we requested more, then probably the kernel was lazy */
		1129	written += res;
		1130	offset += res;
		1131	count -= res;
		1132
		1133	if (!count)
		1134	return written;
		1135	}
		1136	}
		1137
953	if (res < 0	1138	if (res < 0
954	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1139	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1140	/* BSDs */
		1141	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1142	|| errno == ENOTSUP
		1143	#endif
		1144	#ifdef EOPNOTSUPP /* windows */
		1145	|| errno == EOPNOTSUPP /* BSDs */
		1146	#endif
955	#if __solaris	1147	#if __solaris
956	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1148	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
957	#endif	1149	#endif
958	)	1150	)
959	)	1151	)
…		…
963		1155
964	res = 0;	1156	res = 0;
965		1157
966	while (count)	1158	while (count)
967	{	1159	{
968	ssize_t cnt;	1160	eio_ssize_t cnt;
969		1161
970	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1162	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
971		1163
972	if (cnt <= 0)	1164	if (cnt <= 0)
973	{	1165	{
…		…
985		1177
986	offset += cnt;	1178	offset += cnt;
987	res += cnt;	1179	res += cnt;
988	count -= cnt;	1180	count -= cnt;
989	}	1181	}
		1182
		1183	FUBd;
990	}	1184	}
991		1185
992	return res;	1186	return res;
		1187	}
		1188
		1189	#ifdef PAGESIZE
		1190	# define eio_pagesize() PAGESIZE
		1191	#else
		1192	static intptr_t
		1193	eio_pagesize (void)
		1194	{
		1195	static intptr_t page;
		1196
		1197	if (!page)
		1198	page = sysconf (_SC_PAGESIZE);
		1199
		1200	return page;
		1201	}
		1202	#endif
		1203
		1204	static void
		1205	eio_page_align (void **addr, size_t *length)
		1206	{
		1207	intptr_t mask = eio_pagesize () - 1;
		1208
		1209	/* round down addr */
		1210	intptr_t adj = mask & (intptr_t)*addr;
		1211
		1212	*addr = (void *)((intptr_t)*addr - adj);
		1213	*length += adj;
		1214
		1215	/* round up length */
		1216	*length = (*length + mask) & ~mask;
		1217	}
		1218
		1219	#if !_POSIX_MEMLOCK
		1220	# define eio__mlockall(a) EIO_ENOSYS ()
		1221	#else
		1222
		1223	static int
		1224	eio__mlockall (int flags)
		1225	{
		1226	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1227	extern int mallopt (int, int);
		1228	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1229	#endif
		1230
		1231	if (EIO_MCL_CURRENT != MCL_CURRENT
		1232	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1233	{
		1234	flags = 0
		1235	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1236	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1237	}
		1238
		1239	return mlockall (flags);
		1240	}
		1241	#endif
		1242
		1243	#if !_POSIX_MEMLOCK_RANGE
		1244	# define eio__mlock(a,b) EIO_ENOSYS ()
		1245	#else
		1246
		1247	static int
		1248	eio__mlock (void *addr, size_t length)
		1249	{
		1250	eio_page_align (&addr, &length);
		1251
		1252	return mlock (addr, length);
		1253	}
		1254
		1255	#endif
		1256
		1257	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1258	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1259	#else
		1260
		1261	static int
		1262	eio__msync (void *mem, size_t len, int flags)
		1263	{
		1264	eio_page_align (&mem, &len);
		1265
		1266	if (EIO_MS_ASYNC != MS_SYNC
		1267	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1268	|| EIO_MS_SYNC != MS_SYNC)
		1269	{
		1270	flags = 0
		1271	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1272	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1273	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1274	}
		1275
		1276	return msync (mem, len, flags);
		1277	}
		1278
		1279	#endif
		1280
		1281	static int
		1282	eio__mtouch (eio_req *req)
		1283	{
		1284	void *mem = req->ptr2;
		1285	size_t len = req->size;
		1286	int flags = req->int1;
		1287
		1288	eio_page_align (&mem, &len);
		1289
		1290	{
		1291	intptr_t addr = (intptr_t)mem;
		1292	intptr_t end = addr + len;
		1293	intptr_t page = eio_pagesize ();
		1294
		1295	if (addr < end)
		1296	if (flags & EIO_MT_MODIFY) /* modify */
		1297	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1298	else
		1299	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1300	}
		1301
		1302	return 0;
		1303	}
		1304
		1305	/*****************************************************************************/
		1306	/* requests implemented outside eio_execute, because they are so large */
		1307
		1308	static void
		1309	eio__realpath (eio_req *req, etp_worker *self)
		1310	{
		1311	char *rel = req->ptr1;
		1312	char *res;
		1313	char *tmp1, *tmp2;
		1314	#if SYMLOOP_MAX > 32
		1315	int symlinks = SYMLOOP_MAX;
		1316	#else
		1317	int symlinks = 32;
		1318	#endif
		1319
		1320	req->result = -1;
		1321
		1322	errno = EINVAL;
		1323	if (!rel)
		1324	return;
		1325
		1326	errno = ENOENT;
		1327	if (!*rel)
		1328	return;
		1329
		1330	if (!req->ptr2)
		1331	{
		1332	X_LOCK (wrklock);
		1333	req->flags |= EIO_FLAG_PTR2_FREE;
		1334	X_UNLOCK (wrklock);
		1335	req->ptr2 = malloc (PATH_MAX * 3);
		1336
		1337	errno = ENOMEM;
		1338	if (!req->ptr2)
		1339	return;
		1340	}
		1341
		1342	res = req->ptr2;
		1343	tmp1 = res + PATH_MAX;
		1344	tmp2 = tmp1 + PATH_MAX;
		1345
		1346	#if 0 /* disabled, the musl way to do things is just too racy */
		1347	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1348	/* on linux we may be able to ask the kernel */
		1349	{
		1350	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1351
		1352	if (fd >= 0)
		1353	{
		1354	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1355	req->result = readlink (tmp1, res, PATH_MAX);
		1356	close (fd);
		1357
		1358	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1359
		1360	if (req->result > 0)
		1361	goto done;
		1362	}
		1363	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1364	return;
		1365	}
		1366	#endif
		1367	#endif
		1368
		1369	if (*rel != '/')
		1370	{
		1371	if (!getcwd (res, PATH_MAX))
		1372	return;
		1373
		1374	if (res [1]) /* only use if not / */
		1375	res += strlen (res);
		1376	}
		1377
		1378	while (*rel)
		1379	{
		1380	eio_ssize_t len, linklen;
		1381	char *beg = rel;
		1382
		1383	while (*rel && *rel != '/')
		1384	++rel;
		1385
		1386	len = rel - beg;
		1387
		1388	if (!len) /* skip slashes */
		1389	{
		1390	++rel;
		1391	continue;
		1392	}
		1393
		1394	if (beg [0] == '.')
		1395	{
		1396	if (len == 1)
		1397	continue; /* . - nop */
		1398
		1399	if (beg [1] == '.' && len == 2)
		1400	{
		1401	/* .. - back up one component, if possible */
		1402
		1403	while (res != req->ptr2)
		1404	if (*--res == '/')
		1405	break;
		1406
		1407	continue;
		1408	}
		1409	}
		1410
		1411	errno = ENAMETOOLONG;
		1412	if (res + 1 + len + 1 >= tmp1)
		1413	return;
		1414
		1415	/* copy one component */
		1416	*res = '/';
		1417	memcpy (res + 1, beg, len);
		1418
		1419	/* zero-terminate, for readlink */
		1420	res [len + 1] = 0;
		1421
		1422	/* now check if it's a symlink */
		1423	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1424
		1425	if (linklen < 0)
		1426	{
		1427	if (errno != EINVAL)
		1428	return;
		1429
		1430	/* it's a normal directory. hopefully */
		1431	res += len + 1;
		1432	}
		1433	else
		1434	{
		1435	/* yay, it was a symlink - build new path in tmp2 */
		1436	int rellen = strlen (rel);
		1437
		1438	errno = ENAMETOOLONG;
		1439	if (linklen + 1 + rellen >= PATH_MAX)
		1440	return;
		1441
		1442	errno = ELOOP;
		1443	if (!--symlinks)
		1444	return;
		1445
		1446	if (*tmp1 == '/')
		1447	res = req->ptr2; /* symlink resolves to an absolute path */
		1448
		1449	/* we need to be careful, as rel might point into tmp2 already */
		1450	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1451	tmp2 [linklen] = '/';
		1452	memcpy (tmp2, tmp1, linklen);
		1453
		1454	rel = tmp2;
		1455	}
		1456	}
		1457
		1458	/* special case for the lone root path */
		1459	if (res == req->ptr2)
		1460	*res++ = '/';
		1461
		1462	req->result = res - (char *)req->ptr2;
		1463
		1464	done:
		1465	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
993	}	1466	}
994		1467
995	static signed char	1468	static signed char
996	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1469	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997	{	1470	{
998	return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */	1471	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;	1472	: a->inode < b->inode ? -1
		1473	: a->inode > b->inode ? 1
		1474	: 0;
1000	}	1475	}
1001		1476
		1477	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1478
1002	#define EIO_QSORT_CUTOFF 20 /* quite high, but performs well on many filesystems */	1479	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1480	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1003		1481
1004	static void	1482	static void
1005	eio_dent_sort (eio_dirent *dents, int size)	1483	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1006	{	1484	{
		1485	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
		1486	unsigned char *bit = bits;
		1487
		1488	assert (CHAR_BIT == 8);
		1489	assert (sizeof (eio_dirent) * 8 < 256);
		1490	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1491	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1492
		1493	if (size <= EIO_SORT_FAST)
		1494	return;
		1495
		1496	/* first prepare an array of bits to test in our radix sort */
		1497	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
		1498	/* inode_bits must contain all inodes ORed together */
		1499	/* which is used to skip bits that are 0 everywhere, which is very common */
		1500	{
		1501	eio_ino_t endianness;
1007	int i, j;	1502	int i, j;
1008		1503
1009	if (size > EIO_QSORT_CUTOFF * 3) /* skip quicksort for small directories */	1504	/* we store the byte offset of byte n into byte n of "endianness" */
		1505	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1506	((unsigned char *)&endianness)[i] = i;
		1507
		1508	*bit++ = 0;
		1509
		1510	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1511	{
		1512	/* shifting off the byte offsets out of "endianness" */
		1513	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1514	endianness >>= 8;
		1515
		1516	for (j = 0; j < 8; ++j)
		1517	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
		1518	*bit++ = offs + j;
		1519	}
		1520
		1521	for (j = 0; j < 8; ++j)
		1522	if (score_bits & (1 << j))
		1523	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1524	}
		1525
		1526	/* now actually do the sorting (a variant of MSD radix sort) */
		1527	{
		1528	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
		1529	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
		1530	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
		1531	int stk_idx = 0;
		1532
		1533	base_stk [stk_idx] = dents;
		1534	end_stk [stk_idx] = dents + size;
		1535	bit_stk [stk_idx] = bit - 1;
		1536
		1537	do
		1538	{
		1539	base = base_stk [stk_idx];
		1540	end = end_stk [stk_idx];
		1541	bit = bit_stk [stk_idx];
		1542
		1543	for (;;)
		1544	{
		1545	unsigned char O = *bit >> 3;
		1546	unsigned char M = 1 << (*bit & 7);
		1547
		1548	eio_dirent *a = base;
		1549	eio_dirent *b = end;
		1550
		1551	if (b - a < EIO_SORT_CUTOFF)
		1552	break;
		1553
		1554	/* now bit-partition the array on the bit */
		1555	/* this ugly asymmetric loop seems to perform much better than typical */
		1556	/* partition algos found in the literature */
		1557	do
		1558	if (!(((unsigned char *)a)[O] & M))
		1559	++a;
		1560	else if (!(((unsigned char *)--b)[O] & M))
		1561	{
		1562	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1563	++a;
		1564	}
		1565	while (b > a);
		1566
		1567	/* next bit, or stop, if no bits left in this path */
		1568	if (!*--bit)
		1569	break;
		1570
		1571	base_stk [stk_idx] = a;
		1572	end_stk [stk_idx] = end;
		1573	bit_stk [stk_idx] = bit;
		1574	++stk_idx;
		1575
		1576	end = a;
		1577	}
		1578	}
		1579	while (stk_idx--);
		1580	}
		1581	}
		1582
		1583	static void
		1584	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1585	{
		1586	/* first move the smallest element to the front, to act as a sentinel */
		1587	{
		1588	int i;
		1589	eio_dirent *min = dents;
		1590
		1591	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1592	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1593	if (EIO_DENT_CMP (dents [i], <, *min))
		1594	min = &dents [i];
		1595
		1596	/* swap elements 0 and j (minimum) */
1010	{	1597	{
1011	/* first, use quicksort */	1598	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	}	1599	}
		1600	}
1061		1601
1062	/* use a simple insertion sort at the end */	1602	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1603	{
		1604	eio_dirent *i, *j;
		1605
1063	for (i = 1; i < size; ++i)	1606	for (i = dents + 1; i < dents + size; ++i)
1064	{	1607	{
1065	eio_dirent value = dents [i];	1608	eio_dirent value = *i;
1066		1609
1067	for (j = i - 1; j >= 0 && eio_dent_cmp (&dents [j], &value) > 0; --j)	1610	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1068	dents [j + 1] = dents [j];	1611	j [1] = j [0];
1069		1612
1070	dents [j + 1] = value;	1613	j [1] = value;
1071	}	1614	}
		1615	}
		1616	}
		1617
		1618	static void
		1619	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1620	{
		1621	if (size <= 1)
		1622	return; /* our insertion sort relies on size > 0 */
		1623
		1624	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1625	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1626	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1627
		1628	/* use an insertion sort at the end, or for small arrays, */
		1629	/* as insertion sort is more efficient for small partitions */
		1630	eio_dent_insertion_sort (dents, size);
1072	}	1631	}
1073		1632
1074	/* read a full directory */	1633	/* read a full directory */
1075	static void	1634	static void
1076	eio__scandir (eio_req *req, etp_worker *self)	1635	eio__scandir (eio_req *req, etp_worker *self)
1077	{	1636	{
1078	DIR *dirp;
1079	EIO_STRUCT_DIRENT *entp;
1080	char *name, *names;	1637	char *name, *names;
1081	int namesalloc = 4096;	1638	int namesalloc = 4096 - sizeof (void *) * 4;
1082	int namesoffs = 0;	1639	int namesoffs = 0;
1083	int flags = req->int1;	1640	int flags = req->int1;
1084	eio_dirent *dents = 0;	1641	eio_dirent *dents = 0;
1085	int dentalloc = 128;	1642	int dentalloc = 128;
1086	int dentoffs = 0;	1643	int dentoffs = 0;
		1644	eio_ino_t inode_bits = 0;
		1645	#ifdef _WIN32
		1646	HANDLE dirp;
		1647	WIN32_FIND_DATA entp;
		1648	#else
		1649	DIR *dirp;
		1650	EIO_STRUCT_DIRENT *entp;
		1651	#endif
1087		1652
1088	req->result = -1;	1653	req->result = -1;
1089		1654
1090	if (!(flags & EIO_READDIR_DENTS))	1655	if (!(flags & EIO_READDIR_DENTS))
1091	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1656	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1092		1657
1093	X_LOCK (wrklock);	1658	#ifdef _WIN32
1094	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1659	{
		1660	int len = strlen ((const char *)req->ptr1);
		1661	char *path = malloc (MAX_PATH);
		1662	const char *fmt;
		1663
		1664	if (!len)
		1665	fmt = "./*";
		1666	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1667	fmt = "%s*";
		1668	else
		1669	fmt = "%s/*";
		1670
		1671	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1672	dirp = FindFirstFile (path, &entp);
		1673	free (path);
		1674
		1675	if (dirp == INVALID_HANDLE_VALUE)
		1676	{
		1677	dirp = 0;
		1678
		1679	switch (GetLastError ())
		1680	{
		1681	case ERROR_FILE_NOT_FOUND:
		1682	req->result = 0;
		1683	break;
		1684
		1685	case ERROR_INVALID_NAME:
		1686	case ERROR_PATH_NOT_FOUND:
		1687	case ERROR_NO_MORE_FILES:
		1688	errno = ENOENT;
		1689	break;
		1690
		1691	case ERROR_NOT_ENOUGH_MEMORY:
		1692	errno = ENOMEM;
		1693	break;
		1694
		1695	default:
		1696	errno = EINVAL;
		1697	break;
		1698	}
		1699	}
		1700	}
		1701	#else
1095	self->dirp = dirp = opendir (req->ptr1);	1702	dirp = opendir (req->ptr1);
		1703	#endif
		1704
		1705	if (req->flags & EIO_FLAG_PTR1_FREE)
		1706	free (req->ptr1);
		1707
1096	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1708	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;	1709	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1099	X_UNLOCK (wrklock);	1710	req->ptr2 = names = malloc (namesalloc);
1100		1711
1101	if (dirp && names && (!flags || dents))	1712	if (dirp && names && (!flags || dents))
1102	for (;;)	1713	for (;;)
1103	{	1714	{
		1715	int done;
		1716
		1717	#ifdef _WIN32
		1718	done = !dirp;
		1719	#else
1104	errno = 0;	1720	errno = 0;
1105	entp = readdir (dirp);	1721	entp = readdir (dirp);
		1722	done = !entp;
		1723	#endif
1106		1724
1107	if (!entp)	1725	if (done)
1108	{	1726	{
		1727	#ifndef _WIN32
		1728	int old_errno = errno;
		1729	closedir (dirp);
		1730	errno = old_errno;
		1731
1109	if (errno)	1732	if (errno)
1110	break;	1733	break;
		1734	#endif
1111		1735
1112	/* sort etc. */	1736	/* sort etc. */
1113	req->int1 = flags;	1737	req->int1 = flags;
1114	req->result = dentoffs;	1738	req->result = dentoffs;
1115		1739
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	1740	if (flags & EIO_READDIR_STAT_ORDER)
1125	|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))	1741	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)	1742	else if (flags & EIO_READDIR_DIRS_FIRST)
		1743	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1744	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1745	else
1128	{	1746	{
1129	/* in this case, all is known, and we just put dirs first and sort them */	1747	/* in this case, all is known, and we just put dirs first and sort them */
1130	eio_dirent *ent = dents + dentoffs;	1748	eio_dirent *oth = dents + dentoffs;
1131	eio_dirent *dir = dents;	1749	eio_dirent *dir = dents;
1132		1750
1133	/* now move dirs to the front, and non-dirs to the back */	1751	/* now partition dirs to the front, and non-dirs to the back */
1134	/* by walking from both sides and swapping if necessary */	1752	/* by walking from both sides and swapping if necessary */
1135	while (ent > dir)	1753	while (oth > dir)
1136	{	1754	{
1137	if (dir->type == DT_DIR)	1755	if (dir->type == EIO_DT_DIR)
1138	++dir;	1756	++dir;
1139	else
1140	{
1141	--ent;
1142
1143	if (ent->type == DT_DIR)	1757	else if ((--oth)->type == EIO_DT_DIR)
1144	{	1758	{
1145	eio_dirent tmp = *dir;	1759	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1146	*dir = *ent;
1147	*ent = tmp;
1148		1760
1149	++dir;	1761	++dir;
1150	}	1762	}
1151	}	1763	}
		1764
		1765	/* now sort the dirs only (dirs all have the same score) */
		1766	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1152	}	1767	}
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		1768
1168	break;	1769	break;
1169	}	1770	}
1170		1771
1171	/* now add the entry to our list(s) */	1772	/* now add the entry to our list(s) */
1172	name = entp->d_name;	1773	name = D_NAME (entp);
1173		1774
1174	/* skip . and .. entries */	1775	/* skip . and .. entries */
1175	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1776	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1176	{	1777	{
1177	int len = D_NAMLEN (entp) + 1;	1778	int len = D_NAMLEN (entp) + 1;
1178		1779
1179	while (expect_false (namesoffs + len > namesalloc))	1780	while (ecb_expect_false (namesoffs + len > namesalloc))
1180	{	1781	{
1181	namesalloc *= 2;	1782	namesalloc *= 2;
1182	X_LOCK (wrklock);
1183	req->ptr1 = names = realloc (names, namesalloc);	1783	req->ptr2 = names = realloc (names, namesalloc);
1184	X_UNLOCK (wrklock);
1185		1784
1186	if (!names)	1785	if (!names)
1187	break;	1786	break;
1188	}	1787	}
1189		1788
…		…
1191		1790
1192	if (dents)	1791	if (dents)
1193	{	1792	{
1194	struct eio_dirent *ent;	1793	struct eio_dirent *ent;
1195		1794
1196	if (expect_false (dentoffs == dentalloc))	1795	if (ecb_expect_false (dentoffs == dentalloc))
1197	{	1796	{
1198	dentalloc *= 2;	1797	dentalloc *= 2;
1199	X_LOCK (wrklock);
1200	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1798	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1201	X_UNLOCK (wrklock);
1202		1799
1203	if (!dents)	1800	if (!dents)
1204	break;	1801	break;
1205	}	1802	}
1206		1803
1207	ent = dents + dentoffs;	1804	ent = dents + dentoffs;
1208		1805
1209	ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */	1806	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1210	ent->namelen = len - 1;	1807	ent->namelen = len - 1;
1211	ent->inode = D_INO (entp);	1808	ent->inode = D_INO (entp);
		1809
		1810	inode_bits |= ent->inode;
1212		1811
1213	switch (D_TYPE (entp))	1812	switch (D_TYPE (entp))
1214	{	1813	{
1215	default:	1814	default:
1216	ent->type = EIO_DT_UNKNOWN;	1815	ent->type = EIO_DT_UNKNOWN;
…		…
1259	#ifdef DT_WHT	1858	#ifdef DT_WHT
1260	case DT_WHT: ent->type = EIO_DT_WHT; break;	1859	case DT_WHT: ent->type = EIO_DT_WHT; break;
1261	#endif	1860	#endif
1262	}	1861	}
1263		1862
1264	ent->score = 0;	1863	ent->score = 7;
1265		1864
1266	if (flags & EIO_READDIR_DIRS_FIRST)	1865	if (flags & EIO_READDIR_DIRS_FIRST)
1267	{	1866	{
1268	if (ent->type == EIO_DT_UNKNOWN)	1867	if (ent->type == EIO_DT_UNKNOWN)
1269	{	1868	{
1270	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */	1869	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1271	ent->score = 98;	1870	ent->score = 1;
1272	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */	1871	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 */	1872	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1274	}	1873	}
1275	else if (ent->type == EIO_DT_DIR)	1874	else if (ent->type == EIO_DT_DIR)
1276	ent->score = 100;	1875	ent->score = 0;
1277	}	1876	}
1278	}	1877	}
1279		1878
1280	namesoffs += len;	1879	namesoffs += len;
1281	++dentoffs;	1880	++dentoffs;
…		…
1284	if (EIO_CANCELLED (req))	1883	if (EIO_CANCELLED (req))
1285	{	1884	{
1286	errno = ECANCELED;	1885	errno = ECANCELED;
1287	break;	1886	break;
1288	}	1887	}
		1888
		1889	#ifdef _WIN32
		1890	if (!FindNextFile (dirp, &entp))
		1891	{
		1892	FindClose (dirp);
		1893	dirp = 0;
		1894	}
		1895	#endif
1289	}	1896	}
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	}	1897	}
1321		1898
1322	/*****************************************************************************/	1899	/*****************************************************************************/
1323		1900
1324	#define ALLOC(len) \	1901	#define ALLOC(len) \
…		…
1340	{	1917	{
1341	ETP_REQ *req;	1918	ETP_REQ *req;
1342	struct timespec ts;	1919	struct timespec ts;
1343	etp_worker *self = (etp_worker *)thr_arg;	1920	etp_worker *self = (etp_worker *)thr_arg;
1344		1921
		1922	#if HAVE_PRCTL_SET_NAME
		1923	prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
		1924	#endif
		1925
1345	/* try to distribute timeouts somewhat randomly */	1926	/* try to distribute timeouts somewhat evenly */
1346	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1927	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1347		1928
1348	for (;;)	1929	for (;;)
1349	{	1930	{
		1931	ts.tv_sec = 0;
		1932
1350	X_LOCK (reqlock);	1933	X_LOCK (reqlock);
1351		1934
1352	for (;;)	1935	for (;;)
1353	{	1936	{
1354	self->req = req = reqq_shift (&req_queue);	1937	self->req = req = reqq_shift (&req_queue);
1355		1938
1356	if (req)	1939	if (req)
1357	break;	1940	break;
1358		1941
		1942	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1943	{
		1944	X_UNLOCK (reqlock);
		1945	X_LOCK (wrklock);
		1946	--started;
		1947	X_UNLOCK (wrklock);
		1948	goto quit;
		1949	}
		1950
1359	++idle;	1951	++idle;
1360		1952
1361	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1953	if (idle <= max_idle)
1362	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1954	/* we are allowed to idle, so do so without any timeout */
		1955	X_COND_WAIT (reqwait, reqlock);
		1956	else
1363	{	1957	{
1364	if (idle > max_idle)	1958	/* initialise timeout once */
1365	{	1959	if (!ts.tv_sec)
1366	--idle;	1960	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		1961
1374	/* we are allowed to idle, so do so without any timeout */
1375	X_COND_WAIT (reqwait, reqlock);	1962	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1963	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1376	}	1964	}
1377		1965
1378	--idle;	1966	--idle;
1379	}	1967	}
1380		1968
…		…
1383	X_UNLOCK (reqlock);	1971	X_UNLOCK (reqlock);
1384		1972
1385	if (req->type < 0)	1973	if (req->type < 0)
1386	goto quit;	1974	goto quit;
1387		1975
1388	if (!EIO_CANCELLED (req))
1389	ETP_EXECUTE (self, req);	1976	ETP_EXECUTE (self, req);
1390		1977
1391	X_LOCK (reslock);	1978	X_LOCK (reslock);
1392		1979
1393	++npending;	1980	++npending;
1394		1981
…		…
1409	return 0;	1996	return 0;
1410	}	1997	}
1411		1998
1412	/*****************************************************************************/	1999	/*****************************************************************************/
1413		2000
		2001	int ecb_cold
1414	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	2002	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1415	{	2003	{
		2004	#if !HAVE_PREADWRITE
		2005	X_MUTEX_CREATE (preadwritelock);
		2006	#endif
		2007
1416	return etp_init (want_poll, done_poll);	2008	return etp_init (want_poll, done_poll);
1417	}	2009	}
1418		2010
		2011	ecb_inline void
1419	static void eio_api_destroy (eio_req *req)	2012	eio_api_destroy (eio_req *req)
1420	{	2013	{
1421	free (req);	2014	free (req);
1422	}	2015	}
1423		2016
1424	#define REQ(rtype) \	2017	#define REQ(rtype) \
…		…
1443	{ \	2036	{ \
1444	eio_api_destroy (req); \	2037	eio_api_destroy (req); \
1445	return 0; \	2038	return 0; \
1446	}	2039	}
1447		2040
		2041	static void
1448	static void eio_execute (etp_worker *self, eio_req *req)	2042	eio_execute (etp_worker *self, eio_req *req)
1449	{	2043	{
1450	errno = 0;	2044	if (ecb_expect_false (EIO_CANCELLED (req)))
		2045	{
		2046	req->result = -1;
		2047	req->errorno = ECANCELED;
		2048	return;
		2049	}
1451		2050
1452	switch (req->type)	2051	switch (req->type)
1453	{	2052	{
1454	case EIO_READ: ALLOC (req->size);	2053	case EIO_READ: ALLOC (req->size);
1455	req->result = req->offs >= 0	2054	req->result = req->offs >= 0
…		…
1458	case EIO_WRITE: req->result = req->offs >= 0	2057	case EIO_WRITE: req->result = req->offs >= 0
1459	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2058	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1460	: write (req->int1, req->ptr2, req->size); break;	2059	: write (req->int1, req->ptr2, req->size); break;
1461		2060
1462	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2061	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;	2062	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1464		2063
1465	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2064	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1466	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2065	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1467	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2066	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1468	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2067	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1469	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2068	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1470	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2069	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2070
		2071	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2072	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2073	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2074	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1471		2075
1472	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	2076	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;	2077	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;	2078	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;	2079	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
…		…
1483	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	2087	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1484	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	2088	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;	2089	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1486	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2090	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1487	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2091	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;	2092	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1489		2093
		2094	case EIO_REALPATH: eio__realpath (req, self); break;
		2095
1490	case EIO_READLINK: ALLOC (NAME_MAX);	2096	case EIO_READLINK: ALLOC (PATH_MAX);
1491	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	2097	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1492		2098
1493	case EIO_SYNC: req->result = 0; sync (); break;	2099	case EIO_SYNC: req->result = 0; sync (); break;
1494	case EIO_FSYNC: req->result = fsync (req->int1); break;	2100	case EIO_FSYNC: req->result = fsync (req->int1); break;
1495	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2101	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1496	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;	2102	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;	2103	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2104	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2105	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		2106	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2107	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		2108	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1499		2109
1500	case EIO_READDIR: eio__scandir (req, self); break;	2110	case EIO_READDIR: eio__scandir (req, self); break;
1501		2111
1502	case EIO_BUSY:	2112	case EIO_BUSY:
1503	#ifdef _WIN32	2113	#ifdef _WIN32
1504	Sleep (req->nv1 * 1000.);	2114	Sleep (req->nv1 * 1e3);
1505	#else	2115	#else
1506	{	2116	{
1507	struct timeval tv;	2117	struct timeval tv;
1508		2118
1509	tv.tv_sec = req->nv1;	2119	tv.tv_sec = req->nv1;
1510	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	2120	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1511		2121
1512	req->result = select (0, 0, 0, 0, &tv);	2122	req->result = select (0, 0, 0, 0, &tv);
1513	}	2123	}
1514	#endif	2124	#endif
1515	break;	2125	break;
…		…
1530	times = tv;	2140	times = tv;
1531	}	2141	}
1532	else	2142	else
1533	times = 0;	2143	times = 0;
1534		2144
1535
1536	req->result = req->type == EIO_FUTIME	2145	req->result = req->type == EIO_FUTIME
1537	? futimes (req->int1, times)	2146	? futimes (req->int1, times)
1538	: utimes (req->ptr1, times);	2147	: utimes (req->ptr1, times);
1539	}	2148	}
1540	break;	2149	break;
…		…
1545	case EIO_NOP:	2154	case EIO_NOP:
1546	req->result = 0;	2155	req->result = 0;
1547	break;	2156	break;
1548		2157
1549	case EIO_CUSTOM:	2158	case EIO_CUSTOM:
1550	((void (*)(eio_req *))req->feed) (req);	2159	req->feed (req);
1551	break;	2160	break;
1552		2161
1553	default:	2162	default:
		2163	errno = ENOSYS;
1554	req->result = -1;	2164	req->result = -1;
1555	break;	2165	break;
1556	}	2166	}
1557		2167
1558	req->errorno = errno;	2168	req->errorno = errno;
…		…
1583	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2193	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1584	{	2194	{
1585	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2195	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1586	}	2196	}
1587		2197
		2198	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		2199	{
		2200	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2201	}
		2202
		2203	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2204	{
		2205	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2206	}
		2207
		2208	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2209	{
		2210	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2211	}
		2212
1588	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2213	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1589	{	2214	{
1590	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2215	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1591	}	2216	}
1592		2217
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)	2218	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1594	{	2219	{
1595	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2220	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1596	}	2221	}
1597		2222
1598	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2223	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1599	{	2224	{
1600	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2225	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2226	}
		2227
		2228	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2229	{
		2230	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1601	}	2231	}
1602		2232
1603	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2233	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1604	{	2234	{
1605	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2235	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
1623	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2253	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1624	{	2254	{
1625	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2255	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1626	}	2256	}
1627		2257
		2258	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2259	{
		2260	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2261	}
		2262
1628	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2263	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1629	{	2264	{
1630	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2265	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1631	}	2266	}
1632		2267
…		…
1638	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2273	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1639	{	2274	{
1640	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2275	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1641	}	2276	}
1642		2277
1643	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2278	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1644	{	2279	{
1645	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2280	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1646	}	2281	}
1647		2282
1648	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2283	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)	2303	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1669	{	2304	{
1670	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2305	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1671	}	2306	}
1672		2307
1673	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2308	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1674	{	2309	{
1675	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2310	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1676	}	2311	}
1677		2312
1678	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2313	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)	2329	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1695	{	2330	{
1696	return eio__1path (EIO_READLINK, path, pri, cb, data);	2331	return eio__1path (EIO_READLINK, path, pri, cb, data);
1697	}	2332	}
1698		2333
		2334	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2335	{
		2336	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2337	}
		2338
1699	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2339	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1700	{	2340	{
1701	return eio__1path (EIO_STAT, path, pri, cb, data);	2341	return eio__1path (EIO_STAT, path, pri, cb, data);
1702	}	2342	}
1703		2343
1704	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2344	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1705	{	2345	{
1706	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2346	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1707	}	2347	}
1708		2348
		2349	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2350	{
		2351	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2352	}
		2353
1709	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2354	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1710	{	2355	{
1711	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2356	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1712	}	2357	}
1713		2358
…		…
1721	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2366	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1722	}	2367	}
1723		2368
1724	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2369	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1725	{	2370	{
1726	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2371	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1727	}	2372	}
1728		2373
1729	static eio_req *	2374	static eio_req *
1730	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2375	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1731	{	2376	{
…		…
1755	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2400	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1756	{	2401	{
1757	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2402	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1758	}	2403	}
1759		2404
1760	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2405	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1761	{	2406	{
1762	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2407	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1763	}	2408	}
1764		2409
1765	#endif	2410	#endif
1766		2411
1767	eio_req *eio_grp (eio_cb cb, void *data)	2412	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1776	#undef SEND	2421	#undef SEND
1777		2422
1778	/*****************************************************************************/	2423	/*****************************************************************************/
1779	/* grp functions */	2424	/* grp functions */
1780		2425
		2426	void
1781	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2427	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1782	{	2428	{
1783	grp->int2 = limit;	2429	grp->int2 = limit;
1784	grp->feed = feed;	2430	grp->feed = feed;
1785		2431
1786	grp_try_feed (grp);	2432	grp_try_feed (grp);
1787	}	2433	}
1788		2434
		2435	void
1789	void eio_grp_limit (eio_req *grp, int limit)	2436	eio_grp_limit (eio_req *grp, int limit)
1790	{	2437	{
1791	grp->int2 = limit;	2438	grp->int2 = limit;
1792		2439
1793	grp_try_feed (grp);	2440	grp_try_feed (grp);
1794	}	2441	}
1795		2442
		2443	void
1796	void eio_grp_add (eio_req *grp, eio_req *req)	2444	eio_grp_add (eio_req *grp, eio_req *req)
1797	{	2445	{
1798	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2446	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1799		2447
1800	grp->flags |= EIO_FLAG_GROUPADD;	2448	grp->flags |= EIO_FLAG_GROUPADD;
1801		2449
…		…
1812	}	2460	}
1813		2461
1814	/*****************************************************************************/	2462	/*****************************************************************************/
1815	/* misc garbage */	2463	/* misc garbage */
1816		2464
		2465	eio_ssize_t
1817	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2466	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1818	{	2467	{
1819	etp_worker wrk;
1820	ssize_t ret;
1821
1822	wrk.dbuf = 0;
1823
1824	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2468	return eio__sendfile (ofd, ifd, offset, count);
1825
1826	if (wrk.dbuf)
1827	free (wrk.dbuf);
1828
1829	return ret;
1830	}	2469	}
1831		2470

Diff Legend

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