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 1.94 by root, Mon Jul 18 02:59:58 2011 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libeio@schmorp.de>	4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5	* All rights reserved.	5	* All rights reserved.
6	*	6	*
7	* Redistribution and use in source and binary forms, with or without modifica-	7	* Redistribution and use in source and binary forms, with or without modifica-
8	* tion, are permitted provided that the following conditions are met:	8	* tion, are permitted provided that the following conditions are met:
9	*	9	*
…		…
35	* and other provisions required by the GPL. If you do not delete the	35	* and other provisions required by the GPL. If you do not delete the
36	* provisions above, a recipient may use your version of this file under	36	* provisions above, a recipient may use your version of this file under
37	* either the BSD or the GPL.	37	* either the BSD or the GPL.
38	*/	38	*/
39		39
		40	#ifndef _WIN32
		41	# include "config.h"
		42	#endif
		43
40	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
41		46
42	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
43	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
44	#endif	49	#endif
45	#include "xthread.h"	50	#include "xthread.h"
…		…
53	#include <sys/stat.h>	58	#include <sys/stat.h>
54	#include <limits.h>	59	#include <limits.h>
55	#include <fcntl.h>	60	#include <fcntl.h>
56	#include <assert.h>	61	#include <assert.h>
57		62
		63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
		64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
		65	#if HAVE_STDINT_H
		66	# include <stdint.h>
		67	#endif
		68
		69	#ifndef ECANCELED
		70	# define ECANCELED EDOM
		71	#endif
		72	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	static void eio_destroy (eio_req *req);
		77
58	#ifndef EIO_FINISH	78	#ifndef EIO_FINISH
59	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	79	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60	#endif	80	#endif
61		81
62	#ifndef EIO_DESTROY	82	#ifndef EIO_DESTROY
…		…
65		85
66	#ifndef EIO_FEED	86	#ifndef EIO_FEED
67	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	87	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68	#endif	88	#endif
69		89
		90	#ifndef EIO_FD_TO_WIN32_HANDLE
		91	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		92	#endif
		93	#ifndef EIO_WIN32_HANDLE_TO_FD
		94	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		95	#endif
		96
		97	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		98
		99	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		100
70	#ifdef _WIN32	101	#ifdef _WIN32
71		102
72	/*doh*/	103	#undef PAGESIZE
		104	#define PAGESIZE 4096 /* GetSystemInfo? */
		105
		106	#ifdef EIO_STRUCT_STATI64
		107	#define stat(path,buf) _stati64 (path,buf)
		108	#define fstat(fd,buf) _fstati64 (path,buf)
		109	#endif
		110	#define lstat(path,buf) stat (path,buf)
		111	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		112	#define mkdir(path,mode) _mkdir (path)
		113	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		114
		115	#define chmod(path,mode) _chmod (path, mode)
		116	#define dup(fd) _dup (fd)
		117	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		118
		119	#define fchmod(fd,mode) EIO_ENOSYS ()
		120	#define chown(path,uid,gid) EIO_ENOSYS ()
		121	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		122	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		123	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		124	#define mknod(path,mode,dev) EIO_ENOSYS ()
		125	#define sync() EIO_ENOSYS ()
		126	#define readlink(path,buf,s) EIO_ENOSYS ()
		127	#define statvfs(path,buf) EIO_ENOSYS ()
		128	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		129
		130	/* we could even stat and see if it exists */
		131	static int
		132	symlink (const char *old, const char *neu)
		133	{
		134	#if WINVER >= 0x0600
		135	if (CreateSymbolicLink (neu, old, 1))
		136	return 0;
		137
		138	if (CreateSymbolicLink (neu, old, 0))
		139	return 0;
		140	#endif
		141
		142	return EIO_ERRNO (ENOENT, -1);
		143	}
		144
		145	/* POSIX API only */
		146	#define CreateHardLink(neu,old,flags) 0
		147	#define CreateSymbolicLink(neu,old,flags) 0
		148
		149	struct statvfs
		150	{
		151	int dummy;
		152	};
		153
		154	#define DT_DIR EIO_DT_DIR
		155	#define DT_REG EIO_DT_REG
		156	#define D_NAME(entp) entp.cFileName
		157	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
		158
73	#else	159	#else
74		160
75	# include "config.h"
76	# include <sys/time.h>	161	#include <sys/time.h>
77	# include <sys/select.h>	162	#include <sys/select.h>
78	# include <sys/mman.h>	163	#include <sys/statvfs.h>
79	# include <unistd.h>	164	#include <unistd.h>
80	# include <utime.h>	165	#include <utime.h>
81	# include <signal.h>	166	#include <signal.h>
82	# include <dirent.h>	167	#include <dirent.h>
83		168
		169	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		170	#include <sys/mman.h>
		171	#endif
		172
		173	#define D_NAME(entp) entp->d_name
		174
84	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */	175	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85	# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)	176	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
86	# define _DIRENT_HAVE_D_TYPE /* sigh */	177	#define _DIRENT_HAVE_D_TYPE /* sigh */
87	# define D_INO(de) (de)->d_fileno	178	#define D_INO(de) (de)->d_fileno
88	# define D_NAMLEN(de) (de)->d_namlen	179	#define D_NAMLEN(de) (de)->d_namlen
89	# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600	180	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
90	# define D_INO(de) (de)->d_ino	181	#define D_INO(de) (de)->d_ino
91	# endif	182	#endif
92		183
93	#ifdef _D_EXACT_NAMLEN	184	#ifdef _D_EXACT_NAMLEN
94	# undef D_NAMLEN	185	#undef D_NAMLEN
95	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)	186	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96	#endif	187	#endif
97		188
98	# ifdef _DIRENT_HAVE_D_TYPE	189	#ifdef _DIRENT_HAVE_D_TYPE
99	# define D_TYPE(de) (de)->d_type	190	#define D_TYPE(de) (de)->d_type
100	# endif	191	#endif
101		192
102	# ifndef EIO_STRUCT_DIRENT	193	#ifndef EIO_STRUCT_DIRENT
103	# define EIO_STRUCT_DIRENT struct dirent	194	#define EIO_STRUCT_DIRENT struct dirent
104	# endif	195	#endif
105		196
106	#endif	197	#endif
107		198
108	#if HAVE_SENDFILE	199	#if HAVE_SENDFILE
109	# if __linux	200	# if __linux
110	# include <sys/sendfile.h>	201	# include <sys/sendfile.h>
111	# elif __freebsd	202	# elif __FreeBSD__ || defined __APPLE__
112	# include <sys/socket.h>	203	# include <sys/socket.h>
113	# include <sys/uio.h>	204	# include <sys/uio.h>
114	# elif __hpux	205	# elif __hpux
115	# include <sys/socket.h>	206	# include <sys/socket.h>
116	# elif __solaris /* not yet */	207	# elif __solaris
117	# include <sys/sendfile.h>	208	# include <sys/sendfile.h>
118	# else	209	# else
119	# error sendfile support requested but not available	210	# error sendfile support requested but not available
120	# endif	211	# endif
121	#endif	212	#endif
…		…
125	#endif	216	#endif
126	#ifndef D_INO	217	#ifndef D_INO
127	# define D_INO(de) 0	218	# define D_INO(de) 0
128	#endif	219	#endif
129	#ifndef D_NAMLEN	220	#ifndef D_NAMLEN
130	# define D_NAMLEN(de) strlen ((de)->d_name)	221	# define D_NAMLEN(entp) strlen (D_NAME (entp))
131	#endif	222	#endif
132
133	/* number of seconds after which an idle threads exit */
134	#define IDLE_TIMEOUT 10
135		223
136	/* used for struct dirent, AIX doesn't provide it */	224	/* used for struct dirent, AIX doesn't provide it */
137	#ifndef NAME_MAX	225	#ifndef NAME_MAX
138	# define NAME_MAX 4096	226	# define NAME_MAX 4096
139	#endif	227	#endif
140		228
		229	/* used for readlink etc. */
		230	#ifndef PATH_MAX
		231	# define PATH_MAX 4096
		232	#endif
		233
141	/* buffer size for various temporary buffers */	234	/* buffer size for various temporary buffers */
142	#define EIO_BUFSIZE 65536	235	#define EIO_BUFSIZE 65536
143		236
144	#define dBUF \	237	#define dBUF \
145	char *eio_buf; \
146	ETP_WORKER_LOCK (self); \
147	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	238	char *eio_buf = malloc (EIO_BUFSIZE); \
148	ETP_WORKER_UNLOCK (self); \
149	errno = ENOMEM; \	239	errno = ENOMEM; \
150	if (!eio_buf) \	240	if (!eio_buf) \
151	return -1;	241	return -1
		242
		243	#define FUBd \
		244	free (eio_buf)
152		245
153	#define EIO_TICKS ((1000000 + 1023) >> 10)	246	#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		247
168	#define ETP_PRI_MIN EIO_PRI_MIN	248	#define ETP_PRI_MIN EIO_PRI_MIN
169	#define ETP_PRI_MAX EIO_PRI_MAX	249	#define ETP_PRI_MAX EIO_PRI_MAX
170		250
171	struct etp_worker;	251	struct etp_worker;
…		…
175	static int eio_finish (eio_req *req);	255	static int eio_finish (eio_req *req);
176	#define ETP_FINISH(req) eio_finish (req)	256	#define ETP_FINISH(req) eio_finish (req)
177	static void eio_execute (struct etp_worker *self, eio_req *req);	257	static void eio_execute (struct etp_worker *self, eio_req *req);
178	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	258	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
179		259
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	/*****************************************************************************/	260	/*****************************************************************************/
198		261
199	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	262	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200		263
201	/* calculcate time difference in ~1/EIO_TICKS of a second */	264	/* calculate time difference in ~1/EIO_TICKS of a second */
		265	ecb_inline int
202	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	266	tvdiff (struct timeval *tv1, struct timeval *tv2)
203	{	267	{
204	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	268	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	269	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
206	}	270	}
207		271
…		…
214	static unsigned int max_poll_reqs; /* reslock */	278	static unsigned int max_poll_reqs; /* reslock */
215		279
216	static volatile unsigned int nreqs; /* reqlock */	280	static volatile unsigned int nreqs; /* reqlock */
217	static volatile unsigned int nready; /* reqlock */	281	static volatile unsigned int nready; /* reqlock */
218	static volatile unsigned int npending; /* reqlock */	282	static volatile unsigned int npending; /* reqlock */
219	static volatile unsigned int max_idle = 4;	283	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		284	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
220		285
221	static mutex_t wrklock = X_MUTEX_INIT;	286	static xmutex_t wrklock;
222	static mutex_t reslock = X_MUTEX_INIT;	287	static xmutex_t reslock;
223	static mutex_t reqlock = X_MUTEX_INIT;	288	static xmutex_t reqlock;
224	static cond_t reqwait = X_COND_INIT;	289	static xcond_t reqwait;
225		290
226	#if !HAVE_PREADWRITE	291	#if !HAVE_PREADWRITE
227	/*	292	/*
228	* make our pread/pwrite emulation safe against themselves, but not against	293	* make our pread/pwrite emulation safe against themselves, but not against
229	* normal read/write by using a mutex. slows down execution a lot,	294	* normal read/write by using a mutex. slows down execution a lot,
230	* but that's your problem, not mine.	295	* but that's your problem, not mine.
231	*/	296	*/
232	static mutex_t preadwritelock = X_MUTEX_INIT;	297	static xmutex_t preadwritelock;
233	#endif	298	#endif
234		299
235	typedef struct etp_worker	300	typedef struct etp_worker
236	{	301	{
237	/* locked by wrklock */	302	/* locked by wrklock */
238	struct etp_worker *prev, *next;	303	struct etp_worker *prev, *next;
239		304
240	thread_t tid;	305	xthread_t tid;
241		306
242	/* locked by reslock, reqlock or wrklock */	307	/* locked by reslock, reqlock or wrklock */
243	ETP_REQ *req; /* currently processed request */	308	ETP_REQ *req; /* currently processed request */
244		309
		310	#ifdef ETP_WORKER_COMMON
245	ETP_WORKER_COMMON	311	ETP_WORKER_COMMON
		312	#endif
246	} etp_worker;	313	} etp_worker;
247		314
248	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	315	static etp_worker wrk_first; /* NOT etp */
249		316
250	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	317	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
251	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	318	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
252		319
253	/* worker threads management */	320	/* worker threads management */
254		321
		322	static void ecb_cold
255	static void etp_worker_clear (etp_worker *wrk)	323	etp_worker_clear (etp_worker *wrk)
256	{	324	{
257	ETP_WORKER_CLEAR (wrk);
258	}	325	}
259		326
		327	static void ecb_cold
260	static void etp_worker_free (etp_worker *wrk)	328	etp_worker_free (etp_worker *wrk)
261	{	329	{
262	wrk->next->prev = wrk->prev;	330	wrk->next->prev = wrk->prev;
263	wrk->prev->next = wrk->next;	331	wrk->prev->next = wrk->next;
264		332
265	free (wrk);	333	free (wrk);
266	}	334	}
267		335
268	static unsigned int etp_nreqs (void)	336	static unsigned int
		337	etp_nreqs (void)
269	{	338	{
270	int retval;	339	int retval;
271	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	340	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
272	retval = nreqs;	341	retval = nreqs;
273	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	342	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
274	return retval;	343	return retval;
275	}	344	}
276		345
277	static unsigned int etp_nready (void)	346	static unsigned int
		347	etp_nready (void)
278	{	348	{
279	unsigned int retval;	349	unsigned int retval;
280		350
281	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	351	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282	retval = nready;	352	retval = nready;
283	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	353	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284		354
285	return retval;	355	return retval;
286	}	356	}
287		357
288	static unsigned int etp_npending (void)	358	static unsigned int
		359	etp_npending (void)
289	{	360	{
290	unsigned int retval;	361	unsigned int retval;
291		362
292	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	363	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
293	retval = npending;	364	retval = npending;
294	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	365	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
295		366
296	return retval;	367	return retval;
297	}	368	}
298		369
299	static unsigned int etp_nthreads (void)	370	static unsigned int
		371	etp_nthreads (void)
300	{	372	{
301	unsigned int retval;	373	unsigned int retval;
302		374
303	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	375	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
304	retval = started;	376	retval = started;
…		…
318	} etp_reqq;	390	} etp_reqq;
319		391
320	static etp_reqq req_queue;	392	static etp_reqq req_queue;
321	static etp_reqq res_queue;	393	static etp_reqq res_queue;
322		394
		395	static void ecb_noinline ecb_cold
		396	reqq_init (etp_reqq *q)
		397	{
		398	int pri;
		399
		400	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		401	q->qs[pri] = q->qe[pri] = 0;
		402
		403	q->size = 0;
		404	}
		405
		406	static int ecb_noinline
323	static int reqq_push (etp_reqq *q, ETP_REQ *req)	407	reqq_push (etp_reqq *q, ETP_REQ *req)
324	{	408	{
325	int pri = req->pri;	409	int pri = req->pri;
326	req->next = 0;	410	req->next = 0;
327		411
328	if (q->qe[pri])	412	if (q->qe[pri])
…		…
334	q->qe[pri] = q->qs[pri] = req;	418	q->qe[pri] = q->qs[pri] = req;
335		419
336	return q->size++;	420	return q->size++;
337	}	421	}
338		422
		423	static ETP_REQ * ecb_noinline
339	static ETP_REQ *reqq_shift (etp_reqq *q)	424	reqq_shift (etp_reqq *q)
340	{	425	{
341	int pri;	426	int pri;
342		427
343	if (!q->size)	428	if (!q->size)
344	return 0;	429	return 0;
…		…
359	}	444	}
360		445
361	abort ();	446	abort ();
362	}	447	}
363		448
364	static void etp_atfork_prepare (void)	449	static int ecb_cold
		450	etp_init (void (*want_poll)(void), void (*done_poll)(void))
365	{	451	{
366	X_LOCK (wrklock);	452	X_MUTEX_CREATE (wrklock);
367	X_LOCK (reqlock);	453	X_MUTEX_CREATE (reslock);
368	X_LOCK (reslock);	454	X_MUTEX_CREATE (reqlock);
369	#if !HAVE_PREADWRITE	455	X_COND_CREATE (reqwait);
370	X_LOCK (preadwritelock);
371	#endif
372	}
373		456
374	static void etp_atfork_parent (void)	457	reqq_init (&req_queue);
375	{	458	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		459
384	static void etp_atfork_child (void)	460	wrk_first.next =
385	{	461	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		462
405	started = 0;	463	started = 0;
406	idle = 0;	464	idle = 0;
407	nreqs = 0;	465	nreqs = 0;
408	nready = 0;	466	nready = 0;
409	npending = 0;	467	npending = 0;
410		468
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;	469	want_poll_cb = want_poll;
428	done_poll_cb = done_poll;	470	done_poll_cb = done_poll;
429		471
430	return 0;	472	return 0;
431	}	473	}
432		474
433	X_THREAD_PROC (etp_proc);	475	X_THREAD_PROC (etp_proc);
434		476
		477	static void ecb_cold
435	static void etp_start_thread (void)	478	etp_start_thread (void)
436	{	479	{
437	etp_worker *wrk = calloc (1, sizeof (etp_worker));	480	etp_worker *wrk = calloc (1, sizeof (etp_worker));
438		481
439	/*TODO*/	482	/*TODO*/
440	assert (("unable to allocate worker thread data", wrk));	483	assert (("unable to allocate worker thread data", wrk));
…		…
453	free (wrk);	496	free (wrk);
454		497
455	X_UNLOCK (wrklock);	498	X_UNLOCK (wrklock);
456	}	499	}
457		500
		501	static void
458	static void etp_maybe_start_thread (void)	502	etp_maybe_start_thread (void)
459	{	503	{
460	if (expect_true (etp_nthreads () >= wanted))	504	if (ecb_expect_true (etp_nthreads () >= wanted))
461	return;	505	return;
462		506
463	/* todo: maybe use idle here, but might be less exact */	507	/* todo: maybe use idle here, but might be less exact */
464	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	508	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
465	return;	509	return;
466		510
467	etp_start_thread ();	511	etp_start_thread ();
468	}	512	}
469		513
		514	static void ecb_cold
470	static void etp_end_thread (void)	515	etp_end_thread (void)
471	{	516	{
472	eio_req *req = calloc (1, sizeof (eio_req));	517	eio_req *req = calloc (1, sizeof (eio_req));
473		518
474	req->type = -1;	519	req->type = -1;
475	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	520	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
482	X_LOCK (wrklock);	527	X_LOCK (wrklock);
483	--started;	528	--started;
484	X_UNLOCK (wrklock);	529	X_UNLOCK (wrklock);
485	}	530	}
486		531
487	static int etp_poll (void)	532	static int
		533	etp_poll (void)
488	{	534	{
489	unsigned int maxreqs;	535	unsigned int maxreqs;
490	unsigned int maxtime;	536	unsigned int maxtime;
491	struct timeval tv_start, tv_now;	537	struct timeval tv_start, tv_now;
492		538
…		…
522		568
523	X_LOCK (reqlock);	569	X_LOCK (reqlock);
524	--nreqs;	570	--nreqs;
525	X_UNLOCK (reqlock);	571	X_UNLOCK (reqlock);
526		572
527	if (expect_false (req->type == EIO_GROUP && req->size))	573	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
528	{	574	{
529	req->int1 = 1; /* mark request as delayed */	575	req->int1 = 1; /* mark request as delayed */
530	continue;	576	continue;
531	}	577	}
532	else	578	else
533	{	579	{
534	int res = ETP_FINISH (req);	580	int res = ETP_FINISH (req);
535	if (expect_false (res))	581	if (ecb_expect_false (res))
536	return res;	582	return res;
537	}	583	}
538		584
539	if (expect_false (maxreqs && !--maxreqs))	585	if (ecb_expect_false (maxreqs && !--maxreqs))
540	break;	586	break;
541		587
542	if (maxtime)	588	if (maxtime)
543	{	589	{
544	gettimeofday (&tv_now, 0);	590	gettimeofday (&tv_now, 0);
…		…
550		596
551	errno = EAGAIN;	597	errno = EAGAIN;
552	return -1;	598	return -1;
553	}	599	}
554		600
		601	static void
555	static void etp_cancel (ETP_REQ *req)	602	etp_cancel (ETP_REQ *req)
556	{	603	{
557	X_LOCK (wrklock);	604	req->cancelled = 1;
558	req->flags |= EIO_FLAG_CANCELLED;
559	X_UNLOCK (wrklock);
560		605
561	eio_grp_cancel (req);	606	eio_grp_cancel (req);
562	}	607	}
563		608
		609	static void
564	static void etp_submit (ETP_REQ *req)	610	etp_submit (ETP_REQ *req)
565	{	611	{
566	req->pri -= ETP_PRI_MIN;	612	req->pri -= ETP_PRI_MIN;
567		613
568	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	614	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
569	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	615	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
570		616
571	if (expect_false (req->type == EIO_GROUP))	617	if (ecb_expect_false (req->type == EIO_GROUP))
572	{	618	{
573	/* I hope this is worth it :/ */	619	/* I hope this is worth it :/ */
574	X_LOCK (reqlock);	620	X_LOCK (reqlock);
575	++nreqs;	621	++nreqs;
576	X_UNLOCK (reqlock);	622	X_UNLOCK (reqlock);
…		…
595		641
596	etp_maybe_start_thread ();	642	etp_maybe_start_thread ();
597	}	643	}
598	}	644	}
599		645
		646	static void ecb_cold
600	static void etp_set_max_poll_time (double nseconds)	647	etp_set_max_poll_time (double nseconds)
601	{	648	{
602	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	649	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603	max_poll_time = nseconds;	650	max_poll_time = nseconds * EIO_TICKS;
604	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	651	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605	}	652	}
606		653
		654	static void ecb_cold
607	static void etp_set_max_poll_reqs (unsigned int maxreqs)	655	etp_set_max_poll_reqs (unsigned int maxreqs)
608	{	656	{
609	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	657	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
610	max_poll_reqs = maxreqs;	658	max_poll_reqs = maxreqs;
611	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	659	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
612	}	660	}
613		661
		662	static void ecb_cold
614	static void etp_set_max_idle (unsigned int nthreads)	663	etp_set_max_idle (unsigned int nthreads)
615	{	664	{
616	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	665	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617	max_idle = nthreads <= 0 ? 1 : nthreads;	666	max_idle = nthreads;
618	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	667	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619	}	668	}
620		669
		670	static void ecb_cold
		671	etp_set_idle_timeout (unsigned int seconds)
		672	{
		673	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		674	idle_timeout = seconds;
		675	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		676	}
		677
		678	static void ecb_cold
621	static void etp_set_min_parallel (unsigned int nthreads)	679	etp_set_min_parallel (unsigned int nthreads)
622	{	680	{
623	if (wanted < nthreads)	681	if (wanted < nthreads)
624	wanted = nthreads;	682	wanted = nthreads;
625	}	683	}
626		684
		685	static void ecb_cold
627	static void etp_set_max_parallel (unsigned int nthreads)	686	etp_set_max_parallel (unsigned int nthreads)
628	{	687	{
629	if (wanted > nthreads)	688	if (wanted > nthreads)
630	wanted = nthreads;	689	wanted = nthreads;
631		690
632	while (started > wanted)	691	while (started > wanted)
633	etp_end_thread ();	692	etp_end_thread ();
634	}	693	}
635		694
636	/*****************************************************************************/	695	/*****************************************************************************/
637		696
		697	static void
638	static void grp_try_feed (eio_req *grp)	698	grp_try_feed (eio_req *grp)
639	{	699	{
640	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	700	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641	{	701	{
642	grp->flags &= ~EIO_FLAG_GROUPADD;	702	grp->flags &= ~EIO_FLAG_GROUPADD;
643		703
…		…
650	break;	710	break;
651	}	711	}
652	}	712	}
653	}	713	}
654		714
		715	static int
655	static int grp_dec (eio_req *grp)	716	grp_dec (eio_req *grp)
656	{	717	{
657	--grp->size;	718	--grp->size;
658		719
659	/* call feeder, if applicable */	720	/* call feeder, if applicable */
660	grp_try_feed (grp);	721	grp_try_feed (grp);
…		…
664	return eio_finish (grp);	725	return eio_finish (grp);
665	else	726	else
666	return 0;	727	return 0;
667	}	728	}
668		729
		730	static void
669	void eio_destroy (eio_req *req)	731	eio_destroy (eio_req *req)
670	{	732	{
671	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	733	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	734	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673		735
674	EIO_DESTROY (req);	736	EIO_DESTROY (req);
675	}	737	}
676		738
		739	static int
677	static int eio_finish (eio_req *req)	740	eio_finish (eio_req *req)
678	{	741	{
679	int res = EIO_FINISH (req);	742	int res = EIO_FINISH (req);
680		743
681	if (req->grp)	744	if (req->grp)
682	{	745	{
…		…
690	if (grp->grp_first == req)	753	if (grp->grp_first == req)
691	grp->grp_first = req->grp_next;	754	grp->grp_first = req->grp_next;
692		755
693	res2 = grp_dec (grp);	756	res2 = grp_dec (grp);
694		757
695	if (!res && res2)	758	if (!res)
696	res = res2;	759	res = res2;
697	}	760	}
698		761
699	eio_destroy (req);	762	eio_destroy (req);
700		763
701	return res;	764	return res;
702	}	765	}
703		766
		767	void
704	void eio_grp_cancel (eio_req *grp)	768	eio_grp_cancel (eio_req *grp)
705	{	769	{
706	for (grp = grp->grp_first; grp; grp = grp->grp_next)	770	for (grp = grp->grp_first; grp; grp = grp->grp_next)
707	eio_cancel (grp);	771	eio_cancel (grp);
708	}	772	}
709		773
		774	void
710	void eio_cancel (eio_req *req)	775	eio_cancel (eio_req *req)
711	{	776	{
712	etp_cancel (req);	777	etp_cancel (req);
713	}	778	}
714		779
		780	void
715	void eio_submit (eio_req *req)	781	eio_submit (eio_req *req)
716	{	782	{
717	etp_submit (req);	783	etp_submit (req);
718	}	784	}
719		785
720	unsigned int eio_nreqs (void)	786	unsigned int
		787	eio_nreqs (void)
721	{	788	{
722	return etp_nreqs ();	789	return etp_nreqs ();
723	}	790	}
724		791
725	unsigned int eio_nready (void)	792	unsigned int
		793	eio_nready (void)
726	{	794	{
727	return etp_nready ();	795	return etp_nready ();
728	}	796	}
729		797
730	unsigned int eio_npending (void)	798	unsigned int
		799	eio_npending (void)
731	{	800	{
732	return etp_npending ();	801	return etp_npending ();
733	}	802	}
734		803
735	unsigned int eio_nthreads (void)	804	unsigned int ecb_cold
		805	eio_nthreads (void)
736	{	806	{
737	return etp_nthreads ();	807	return etp_nthreads ();
738	}	808	}
739		809
		810	void ecb_cold
740	void eio_set_max_poll_time (double nseconds)	811	eio_set_max_poll_time (double nseconds)
741	{	812	{
742	etp_set_max_poll_time (nseconds);	813	etp_set_max_poll_time (nseconds);
743	}	814	}
744		815
		816	void ecb_cold
745	void eio_set_max_poll_reqs (unsigned int maxreqs)	817	eio_set_max_poll_reqs (unsigned int maxreqs)
746	{	818	{
747	etp_set_max_poll_reqs (maxreqs);	819	etp_set_max_poll_reqs (maxreqs);
748	}	820	}
749		821
		822	void ecb_cold
750	void eio_set_max_idle (unsigned int nthreads)	823	eio_set_max_idle (unsigned int nthreads)
751	{	824	{
752	etp_set_max_idle (nthreads);	825	etp_set_max_idle (nthreads);
753	}	826	}
754		827
		828	void ecb_cold
		829	eio_set_idle_timeout (unsigned int seconds)
		830	{
		831	etp_set_idle_timeout (seconds);
		832	}
		833
		834	void ecb_cold
755	void eio_set_min_parallel (unsigned int nthreads)	835	eio_set_min_parallel (unsigned int nthreads)
756	{	836	{
757	etp_set_min_parallel (nthreads);	837	etp_set_min_parallel (nthreads);
758	}	838	}
759		839
		840	void ecb_cold
760	void eio_set_max_parallel (unsigned int nthreads)	841	eio_set_max_parallel (unsigned int nthreads)
761	{	842	{
762	etp_set_max_parallel (nthreads);	843	etp_set_max_parallel (nthreads);
763	}	844	}
764		845
765	int eio_poll (void)	846	int eio_poll (void)
…		…
774	# undef pread	855	# undef pread
775	# undef pwrite	856	# undef pwrite
776	# define pread eio__pread	857	# define pread eio__pread
777	# define pwrite eio__pwrite	858	# define pwrite eio__pwrite
778		859
779	static ssize_t	860	static eio_ssize_t
780	eio__pread (int fd, void *buf, size_t count, off_t offset)	861	eio__pread (int fd, void *buf, size_t count, off_t offset)
781	{	862	{
782	ssize_t res;	863	eio_ssize_t res;
783	off_t ooffset;	864	off_t ooffset;
784		865
785	X_LOCK (preadwritelock);	866	X_LOCK (preadwritelock);
786	ooffset = lseek (fd, 0, SEEK_CUR);	867	ooffset = lseek (fd, 0, SEEK_CUR);
787	lseek (fd, offset, SEEK_SET);	868	lseek (fd, offset, SEEK_SET);
…		…
790	X_UNLOCK (preadwritelock);	871	X_UNLOCK (preadwritelock);
791		872
792	return res;	873	return res;
793	}	874	}
794		875
795	static ssize_t	876	static eio_ssize_t
796	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	877	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
797	{	878	{
798	ssize_t res;	879	eio_ssize_t res;
799	off_t ooffset;	880	off_t ooffset;
800		881
801	X_LOCK (preadwritelock);	882	X_LOCK (preadwritelock);
802	ooffset = lseek (fd, 0, SEEK_CUR);	883	ooffset = lseek (fd, 0, SEEK_CUR);
803	lseek (fd, offset, SEEK_SET);	884	lseek (fd, offset, SEEK_SET);
…		…
807		888
808	return res;	889	return res;
809	}	890	}
810	#endif	891	#endif
811		892
812	#ifndef HAVE_FUTIMES	893	#ifndef HAVE_UTIMES
813		894
814	# undef utimes	895	# undef utimes
815	# undef futimes
816	# define utimes(path,times) eio__utimes (path, times)	896	# define utimes(path,times) eio__utimes (path, times)
817	# define futimes(fd,times) eio__futimes (fd, times)
818		897
819	static int	898	static int
820	eio__utimes (const char *filename, const struct timeval times[2])	899	eio__utimes (const char *filename, const struct timeval times[2])
821	{	900	{
822	if (times)	901	if (times)
…		…
830	}	909	}
831	else	910	else
832	return utime (filename, 0);	911	return utime (filename, 0);
833	}	912	}
834		913
		914	#endif
		915
		916	#ifndef HAVE_FUTIMES
		917
		918	# undef futimes
		919	# define futimes(fd,times) eio__futimes (fd, times)
		920
		921	static int
835	static int eio__futimes (int fd, const struct timeval tv[2])	922	eio__futimes (int fd, const struct timeval tv[2])
836	{	923	{
837	errno = ENOSYS;	924	errno = ENOSYS;
838	return -1;	925	return -1;
839	}	926	}
840		927
…		…
844	# undef fdatasync	931	# undef fdatasync
845	# define fdatasync(fd) fsync (fd)	932	# define fdatasync(fd) fsync (fd)
846	#endif	933	#endif
847		934
848	/* sync_file_range always needs emulation */	935	/* sync_file_range always needs emulation */
849	int	936	static int
850	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	937	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
851	{	938	{
852	#if HAVE_SYNC_FILE_RANGE	939	#if HAVE_SYNC_FILE_RANGE
853	int res;	940	int res;
854		941
…		…
867	if (!res || errno != ENOSYS)	954	if (!res || errno != ENOSYS)
868	return res;	955	return res;
869	#endif	956	#endif
870		957
871	/* even though we could play tricks with the flags, it's better to always	958	/* even though we could play tricks with the flags, it's better to always
872	* call fdatasync, as thta matches the expectation of it's users best */	959	* call fdatasync, as that matches the expectation of its users best */
873	return fdatasync (fd);	960	return fdatasync (fd);
		961	}
		962
		963	static int
		964	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		965	{
		966	#if HAVE_FALLOCATE
		967	return fallocate (fd, mode, offset, len);
		968	#else
		969	errno = ENOSYS;
		970	return -1;
		971	#endif
874	}	972	}
875		973
876	#if !HAVE_READAHEAD	974	#if !HAVE_READAHEAD
877	# undef readahead	975	# undef readahead
878	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	976	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
879		977
880	static ssize_t	978	static eio_ssize_t
881	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	979	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
882	{	980	{
883	size_t todo = count;	981	size_t todo = count;
884	dBUF;	982	dBUF;
885		983
…		…
890	pread (fd, eio_buf, len, offset);	988	pread (fd, eio_buf, len, offset);
891	offset += len;	989	offset += len;
892	todo -= len;	990	todo -= len;
893	}	991	}
894		992
		993	FUBd;
		994
895	errno = 0;	995	errno = 0;
896	return count;	996	return count;
897	}	997	}
898		998
899	#endif	999	#endif
900		1000
901	/* sendfile always needs emulation */	1001	/* sendfile always needs emulation */
902	static ssize_t	1002	static eio_ssize_t
903	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1003	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
904	{	1004	{
		1005	eio_ssize_t written = 0;
905	ssize_t res;	1006	eio_ssize_t res;
906		1007
907	if (!count)	1008	if (!count)
908	return 0;	1009	return 0;
909		1010
		1011	for (;;)
		1012	{
		1013	#ifdef __APPLE__
		1014	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1015	#endif
910	#if HAVE_SENDFILE	1016	#if HAVE_SENDFILE
911	# if __linux	1017	# if __linux
		1018	off_t soffset = offset;
912	res = sendfile (ofd, ifd, &offset, count);	1019	res = sendfile (ofd, ifd, &soffset, count);
913		1020
914	# elif __freebsd	1021	# elif __FreeBSD__
915	/*	1022	/*
916	* Of course, the freebsd sendfile is a dire hack with no thoughts	1023	* Of course, the freebsd sendfile is a dire hack with no thoughts
917	* wasted on making it similar to other I/O functions.	1024	* wasted on making it similar to other I/O functions.
918	*/	1025	*/
919	{
920	off_t sbytes;	1026	off_t sbytes;
921	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1027	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922		1028
923	if (res < 0 && sbytes)	1029	#if 0 /* according to the manpage, this is correct, but broken behaviour */
924	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1030	/* freebsd' sendfile will return 0 on success */
		1031	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1032	/* not on e.g. EIO or EPIPE - sounds broken */
		1033	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
925	res = sbytes;	1034	res = sbytes;
926	}	1035	#endif
		1036
		1037	/* according to source inspection, this is correct, and useful behaviour */
		1038	if (sbytes)
		1039	res = sbytes;
		1040
		1041	# elif defined (__APPLE__)
		1042	off_t sbytes = count;
		1043	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1044
		1045	/* according to the manpage, sbytes is always valid */
		1046	if (sbytes)
		1047	res = sbytes;
927		1048
928	# elif __hpux	1049	# elif __hpux
929	res = sendfile (ofd, ifd, offset, count, 0, 0);	1050	res = sendfile (ofd, ifd, offset, count, 0, 0);
930		1051
931	# elif __solaris	1052	# elif __solaris
932	{
933	struct sendfilevec vec;	1053	struct sendfilevec vec;
934	size_t sbytes;	1054	size_t sbytes;
935		1055
936	vec.sfv_fd = ifd;	1056	vec.sfv_fd = ifd;
937	vec.sfv_flag = 0;	1057	vec.sfv_flag = 0;
938	vec.sfv_off = offset;	1058	vec.sfv_off = offset;
939	vec.sfv_len = count;	1059	vec.sfv_len = count;
940		1060
941	res = sendfilev (ofd, &vec, 1, &sbytes);	1061	res = sendfilev (ofd, &vec, 1, &sbytes);
942		1062
943	if (res < 0 && sbytes)	1063	if (res < 0 && sbytes)
944	res = sbytes;	1064	res = sbytes;
945	}
946		1065
947	# endif	1066	# endif
		1067
		1068	#elif defined (_WIN32) && 0
		1069	/* does not work, just for documentation of what would need to be done */
		1070	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1071	/* libeio guarantees that the file offset does not change, and windows */
		1072	/* has no way to get an independent handle to the same file description */
		1073	HANDLE h = TO_SOCKET (ifd);
		1074	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1075	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1076
948	#else	1077	#else
949	res = -1;	1078	res = -1;
950	errno = ENOSYS;	1079	errno = ENOSYS;
951	#endif	1080	#endif
952		1081
		1082	/* we assume sendfile can copy at least 128mb in one go */
		1083	if (res <= 128 * 1024 * 1024)
		1084	{
		1085	if (res > 0)
		1086	written += res;
		1087
		1088	if (written)
		1089	return written;
		1090
		1091	break;
		1092	}
		1093	else
		1094	{
		1095	/* if we requested more, then probably the kernel was lazy */
		1096	written += res;
		1097	offset += res;
		1098	count -= res;
		1099
		1100	if (!count)
		1101	return written;
		1102	}
		1103	}
		1104
953	if (res < 0	1105	if (res < 0
954	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1106	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1107	/* BSDs */
		1108	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1109	|| errno == ENOTSUP
		1110	#endif
		1111	#ifdef EOPNOTSUPP /* windows */
		1112	|| errno == EOPNOTSUPP /* BSDs */
		1113	#endif
955	#if __solaris	1114	#if __solaris
956	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1115	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
957	#endif	1116	#endif
958	)	1117	)
959	)	1118	)
…		…
963		1122
964	res = 0;	1123	res = 0;
965		1124
966	while (count)	1125	while (count)
967	{	1126	{
968	ssize_t cnt;	1127	eio_ssize_t cnt;
969		1128
970	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1129	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
971		1130
972	if (cnt <= 0)	1131	if (cnt <= 0)
973	{	1132	{
…		…
985		1144
986	offset += cnt;	1145	offset += cnt;
987	res += cnt;	1146	res += cnt;
988	count -= cnt;	1147	count -= cnt;
989	}	1148	}
		1149
		1150	FUBd;
990	}	1151	}
991		1152
992	return res;	1153	return res;
		1154	}
		1155
		1156	#ifdef PAGESIZE
		1157	# define eio_pagesize() PAGESIZE
		1158	#else
		1159	static intptr_t
		1160	eio_pagesize (void)
		1161	{
		1162	static intptr_t page;
		1163
		1164	if (!page)
		1165	page = sysconf (_SC_PAGESIZE);
		1166
		1167	return page;
		1168	}
		1169	#endif
		1170
		1171	static void
		1172	eio_page_align (void **addr, size_t *length)
		1173	{
		1174	intptr_t mask = eio_pagesize () - 1;
		1175
		1176	/* round down addr */
		1177	intptr_t adj = mask & (intptr_t)*addr;
		1178
		1179	*addr = (void *)((intptr_t)*addr - adj);
		1180	*length += adj;
		1181
		1182	/* round up length */
		1183	*length = (*length + mask) & ~mask;
		1184	}
		1185
		1186	#if !_POSIX_MEMLOCK
		1187	# define eio__mlockall(a) EIO_ENOSYS ()
		1188	#else
		1189
		1190	static int
		1191	eio__mlockall (int flags)
		1192	{
		1193	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1194	extern int mallopt (int, int);
		1195	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1196	#endif
		1197
		1198	if (EIO_MCL_CURRENT != MCL_CURRENT
		1199	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1200	{
		1201	flags = 0
		1202	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1203	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1204	}
		1205
		1206	return mlockall (flags);
		1207	}
		1208	#endif
		1209
		1210	#if !_POSIX_MEMLOCK_RANGE
		1211	# define eio__mlock(a,b) EIO_ENOSYS ()
		1212	#else
		1213
		1214	static int
		1215	eio__mlock (void *addr, size_t length)
		1216	{
		1217	eio_page_align (&addr, &length);
		1218
		1219	return mlock (addr, length);
		1220	}
		1221
		1222	#endif
		1223
		1224	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1225	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1226	#else
		1227
		1228	static int
		1229	eio__msync (void *mem, size_t len, int flags)
		1230	{
		1231	eio_page_align (&mem, &len);
		1232
		1233	if (EIO_MS_ASYNC != MS_SYNC
		1234	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1235	|| EIO_MS_SYNC != MS_SYNC)
		1236	{
		1237	flags = 0
		1238	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1239	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1240	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1241	}
		1242
		1243	return msync (mem, len, flags);
		1244	}
		1245
		1246	#endif
		1247
		1248	static int
		1249	eio__mtouch (eio_req *req)
		1250	{
		1251	void *mem = req->ptr2;
		1252	size_t len = req->size;
		1253	int flags = req->int1;
		1254
		1255	eio_page_align (&mem, &len);
		1256
		1257	{
		1258	intptr_t addr = (intptr_t)mem;
		1259	intptr_t end = addr + len;
		1260	intptr_t page = eio_pagesize ();
		1261
		1262	if (addr < end)
		1263	if (flags & EIO_MT_MODIFY) /* modify */
		1264	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1265	else
		1266	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1267	}
		1268
		1269	return 0;
		1270	}
		1271
		1272	/*****************************************************************************/
		1273	/* requests implemented outside eio_execute, because they are so large */
		1274
		1275	static void
		1276	eio__realpath (eio_req *req, etp_worker *self)
		1277	{
		1278	char *rel = req->ptr1;
		1279	char *res;
		1280	char *tmp1, *tmp2;
		1281	#if SYMLOOP_MAX > 32
		1282	int symlinks = SYMLOOP_MAX;
		1283	#else
		1284	int symlinks = 32;
		1285	#endif
		1286
		1287	req->result = -1;
		1288
		1289	errno = EINVAL;
		1290	if (!rel)
		1291	return;
		1292
		1293	errno = ENOENT;
		1294	if (!*rel)
		1295	return;
		1296
		1297	if (!req->ptr2)
		1298	{
		1299	X_LOCK (wrklock);
		1300	req->flags |= EIO_FLAG_PTR2_FREE;
		1301	X_UNLOCK (wrklock);
		1302	req->ptr2 = malloc (PATH_MAX * 3);
		1303
		1304	errno = ENOMEM;
		1305	if (!req->ptr2)
		1306	return;
		1307	}
		1308
		1309	res = req->ptr2;
		1310	tmp1 = res + PATH_MAX;
		1311	tmp2 = tmp1 + PATH_MAX;
		1312
		1313	#if 0 /* disabled, the musl way to do things is just too racy */
		1314	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1315	/* on linux we may be able to ask the kernel */
		1316	{
		1317	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1318
		1319	if (fd >= 0)
		1320	{
		1321	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1322	req->result = readlink (tmp1, res, PATH_MAX);
		1323	close (fd);
		1324
		1325	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1326
		1327	if (req->result > 0)
		1328	goto done;
		1329	}
		1330	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1331	return;
		1332	}
		1333	#endif
		1334	#endif
		1335
		1336	if (*rel != '/')
		1337	{
		1338	if (!getcwd (res, PATH_MAX))
		1339	return;
		1340
		1341	if (res [1]) /* only use if not / */
		1342	res += strlen (res);
		1343	}
		1344
		1345	while (*rel)
		1346	{
		1347	eio_ssize_t len, linklen;
		1348	char *beg = rel;
		1349
		1350	while (*rel && *rel != '/')
		1351	++rel;
		1352
		1353	len = rel - beg;
		1354
		1355	if (!len) /* skip slashes */
		1356	{
		1357	++rel;
		1358	continue;
		1359	}
		1360
		1361	if (beg [0] == '.')
		1362	{
		1363	if (len == 1)
		1364	continue; /* . - nop */
		1365
		1366	if (beg [1] == '.' && len == 2)
		1367	{
		1368	/* .. - back up one component, if possible */
		1369
		1370	while (res != req->ptr2)
		1371	if (*--res == '/')
		1372	break;
		1373
		1374	continue;
		1375	}
		1376	}
		1377
		1378	errno = ENAMETOOLONG;
		1379	if (res + 1 + len + 1 >= tmp1)
		1380	return;
		1381
		1382	/* copy one component */
		1383	*res = '/';
		1384	memcpy (res + 1, beg, len);
		1385
		1386	/* zero-terminate, for readlink */
		1387	res [len + 1] = 0;
		1388
		1389	/* now check if it's a symlink */
		1390	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1391
		1392	if (linklen < 0)
		1393	{
		1394	if (errno != EINVAL)
		1395	return;
		1396
		1397	/* it's a normal directory. hopefully */
		1398	res += len + 1;
		1399	}
		1400	else
		1401	{
		1402	/* yay, it was a symlink - build new path in tmp2 */
		1403	int rellen = strlen (rel);
		1404
		1405	errno = ENAMETOOLONG;
		1406	if (linklen + 1 + rellen >= PATH_MAX)
		1407	return;
		1408
		1409	errno = ELOOP;
		1410	if (!--symlinks)
		1411	return;
		1412
		1413	if (*tmp1 == '/')
		1414	res = req->ptr2; /* symlink resolves to an absolute path */
		1415
		1416	/* we need to be careful, as rel might point into tmp2 already */
		1417	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1418	tmp2 [linklen] = '/';
		1419	memcpy (tmp2, tmp1, linklen);
		1420
		1421	rel = tmp2;
		1422	}
		1423	}
		1424
		1425	/* special case for the lone root path */
		1426	if (res == req->ptr2)
		1427	*res++ = '/';
		1428
		1429	req->result = res - (char *)req->ptr2;
		1430
		1431	done:
		1432	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
993	}	1433	}
994		1434
995	static signed char	1435	static signed char
996	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1436	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997	{	1437	{
998	return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */	1438	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;	1439	: a->inode < b->inode ? -1
		1440	: a->inode > b->inode ? 1
		1441	: 0;
1000	}	1442	}
1001		1443
		1444	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1445
1002	#define EIO_QSORT_CUTOFF 20 /* quite high, but performs well on many filesystems */	1446	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1447	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1003		1448
1004	static void	1449	static void
1005	eio_dent_sort (eio_dirent *dents, int size)	1450	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1006	{	1451	{
		1452	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
		1453	unsigned char *bit = bits;
		1454
		1455	assert (CHAR_BIT == 8);
		1456	assert (sizeof (eio_dirent) * 8 < 256);
		1457	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1458	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1459
		1460	if (size <= EIO_SORT_FAST)
		1461	return;
		1462
		1463	/* first prepare an array of bits to test in our radix sort */
		1464	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
		1465	/* inode_bits must contain all inodes ORed together */
		1466	/* which is used to skip bits that are 0 everywhere, which is very common */
		1467	{
		1468	eio_ino_t endianness;
1007	int i, j;	1469	int i, j;
1008		1470
1009	if (size > EIO_QSORT_CUTOFF * 3) /* skip quicksort for small directories */	1471	/* we store the byte offset of byte n into byte n of "endianness" */
		1472	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1473	((unsigned char *)&endianness)[i] = i;
		1474
		1475	*bit++ = 0;
		1476
		1477	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1478	{
		1479	/* shifting off the byte offsets out of "endianness" */
		1480	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1481	endianness >>= 8;
		1482
		1483	for (j = 0; j < 8; ++j)
		1484	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
		1485	*bit++ = offs + j;
		1486	}
		1487
		1488	for (j = 0; j < 8; ++j)
		1489	if (score_bits & (1 << j))
		1490	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1491	}
		1492
		1493	/* now actually do the sorting (a variant of MSD radix sort) */
		1494	{
		1495	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
		1496	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
		1497	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
		1498	int stk_idx = 0;
		1499
		1500	base_stk [stk_idx] = dents;
		1501	end_stk [stk_idx] = dents + size;
		1502	bit_stk [stk_idx] = bit - 1;
		1503
		1504	do
		1505	{
		1506	base = base_stk [stk_idx];
		1507	end = end_stk [stk_idx];
		1508	bit = bit_stk [stk_idx];
		1509
		1510	for (;;)
		1511	{
		1512	unsigned char O = *bit >> 3;
		1513	unsigned char M = 1 << (*bit & 7);
		1514
		1515	eio_dirent *a = base;
		1516	eio_dirent *b = end;
		1517
		1518	if (b - a < EIO_SORT_CUTOFF)
		1519	break;
		1520
		1521	/* now bit-partition the array on the bit */
		1522	/* this ugly asymmetric loop seems to perform much better than typical */
		1523	/* partition algos found in the literature */
		1524	do
		1525	if (!(((unsigned char *)a)[O] & M))
		1526	++a;
		1527	else if (!(((unsigned char *)--b)[O] & M))
		1528	{
		1529	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1530	++a;
		1531	}
		1532	while (b > a);
		1533
		1534	/* next bit, or stop, if no bits left in this path */
		1535	if (!*--bit)
		1536	break;
		1537
		1538	base_stk [stk_idx] = a;
		1539	end_stk [stk_idx] = end;
		1540	bit_stk [stk_idx] = bit;
		1541	++stk_idx;
		1542
		1543	end = a;
		1544	}
		1545	}
		1546	while (stk_idx--);
		1547	}
		1548	}
		1549
		1550	static void
		1551	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1552	{
		1553	/* first move the smallest element to the front, to act as a sentinel */
		1554	{
		1555	int i;
		1556	eio_dirent *min = dents;
		1557
		1558	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1559	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1560	if (EIO_DENT_CMP (dents [i], <, *min))
		1561	min = &dents [i];
		1562
		1563	/* swap elements 0 and j (minimum) */
1010	{	1564	{
1011	/* first, use quicksort */	1565	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	}	1566	}
		1567	}
1061		1568
1062	/* use a simple insertion sort at the end */	1569	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1570	{
		1571	eio_dirent *i, *j;
		1572
1063	for (i = 1; i < size; ++i)	1573	for (i = dents + 1; i < dents + size; ++i)
1064	{	1574	{
1065	eio_dirent value = dents [i];	1575	eio_dirent value = *i;
1066		1576
1067	for (j = i - 1; j >= 0 && eio_dent_cmp (&dents [j], &value) > 0; --j)	1577	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1068	dents [j + 1] = dents [j];	1578	j [1] = j [0];
1069		1579
1070	dents [j + 1] = value;	1580	j [1] = value;
1071	}	1581	}
		1582	}
		1583	}
		1584
		1585	static void
		1586	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1587	{
		1588	if (size <= 1)
		1589	return; /* our insertion sort relies on size > 0 */
		1590
		1591	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1592	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1593	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1594
		1595	/* use an insertion sort at the end, or for small arrays, */
		1596	/* as insertion sort is more efficient for small partitions */
		1597	eio_dent_insertion_sort (dents, size);
1072	}	1598	}
1073		1599
1074	/* read a full directory */	1600	/* read a full directory */
1075	static void	1601	static void
1076	eio__scandir (eio_req *req, etp_worker *self)	1602	eio__scandir (eio_req *req, etp_worker *self)
1077	{	1603	{
1078	DIR *dirp;
1079	EIO_STRUCT_DIRENT *entp;
1080	char *name, *names;	1604	char *name, *names;
1081	int namesalloc = 4096;	1605	int namesalloc = 4096 - sizeof (void *) * 4;
1082	int namesoffs = 0;	1606	int namesoffs = 0;
1083	int flags = req->int1;	1607	int flags = req->int1;
1084	eio_dirent *dents = 0;	1608	eio_dirent *dents = 0;
1085	int dentalloc = 128;	1609	int dentalloc = 128;
1086	int dentoffs = 0;	1610	int dentoffs = 0;
		1611	eio_ino_t inode_bits = 0;
		1612	#ifdef _WIN32
		1613	HANDLE dirp;
		1614	WIN32_FIND_DATA entp;
		1615	#else
		1616	DIR *dirp;
		1617	EIO_STRUCT_DIRENT *entp;
		1618	#endif
1087		1619
1088	req->result = -1;	1620	req->result = -1;
1089		1621
1090	if (!(flags & EIO_READDIR_DENTS))	1622	if (!(flags & EIO_READDIR_DENTS))
1091	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1623	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1092		1624
1093	X_LOCK (wrklock);	1625	#ifdef _WIN32
1094	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1626	{
		1627	int len = strlen ((const char *)req->ptr1);
		1628	char *path = malloc (MAX_PATH);
		1629	const char *fmt;
		1630
		1631	if (!len)
		1632	fmt = "./*";
		1633	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1634	fmt = "%s*";
		1635	else
		1636	fmt = "%s/*";
		1637
		1638	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1639	dirp = FindFirstFile (path, &entp);
		1640	free (path);
		1641
		1642	if (dirp == INVALID_HANDLE_VALUE)
		1643	{
		1644	dirp = 0;
		1645
		1646	switch (GetLastError ())
		1647	{
		1648	case ERROR_FILE_NOT_FOUND:
		1649	req->result = 0;
		1650	break;
		1651
		1652	case ERROR_INVALID_NAME:
		1653	case ERROR_PATH_NOT_FOUND:
		1654	case ERROR_NO_MORE_FILES:
		1655	errno = ENOENT;
		1656	break;
		1657
		1658	case ERROR_NOT_ENOUGH_MEMORY:
		1659	errno = ENOMEM;
		1660	break;
		1661
		1662	default:
		1663	errno = EINVAL;
		1664	break;
		1665	}
		1666	}
		1667	}
		1668	#else
1095	self->dirp = dirp = opendir (req->ptr1);	1669	dirp = opendir (req->ptr1);
		1670	#endif
		1671
		1672	if (req->flags & EIO_FLAG_PTR1_FREE)
		1673	free (req->ptr1);
		1674
1096	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1675	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;	1676	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1099	X_UNLOCK (wrklock);	1677	req->ptr2 = names = malloc (namesalloc);
1100		1678
1101	if (dirp && names && (!flags || dents))	1679	if (dirp && names && (!flags || dents))
1102	for (;;)	1680	for (;;)
1103	{	1681	{
		1682	int done;
		1683
		1684	#ifdef _WIN32
		1685	done = !dirp;
		1686	#else
1104	errno = 0;	1687	errno = 0;
1105	entp = readdir (dirp);	1688	entp = readdir (dirp);
		1689	done = !entp;
		1690	#endif
1106		1691
1107	if (!entp)	1692	if (done)
1108	{	1693	{
		1694	#ifndef _WIN32
		1695	int old_errno = errno;
		1696	closedir (dirp);
		1697	errno = old_errno;
		1698
1109	if (errno)	1699	if (errno)
1110	break;	1700	break;
		1701	#endif
1111		1702
1112	/* sort etc. */	1703	/* sort etc. */
1113	req->int1 = flags;	1704	req->int1 = flags;
1114	req->result = dentoffs;	1705	req->result = dentoffs;
1115		1706
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	1707	if (flags & EIO_READDIR_STAT_ORDER)
1125	|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))	1708	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)	1709	else if (flags & EIO_READDIR_DIRS_FIRST)
		1710	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1711	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1712	else
1128	{	1713	{
1129	/* in this case, all is known, and we just put dirs first and sort them */	1714	/* in this case, all is known, and we just put dirs first and sort them */
1130	eio_dirent *ent = dents + dentoffs;	1715	eio_dirent *oth = dents + dentoffs;
1131	eio_dirent *dir = dents;	1716	eio_dirent *dir = dents;
1132		1717
1133	/* now move dirs to the front, and non-dirs to the back */	1718	/* now partition dirs to the front, and non-dirs to the back */
1134	/* by walking from both sides and swapping if necessary */	1719	/* by walking from both sides and swapping if necessary */
1135	while (ent > dir)	1720	while (oth > dir)
1136	{	1721	{
1137	if (dir->type == DT_DIR)	1722	if (dir->type == EIO_DT_DIR)
1138	++dir;	1723	++dir;
1139	else
1140	{
1141	--ent;
1142
1143	if (ent->type == DT_DIR)	1724	else if ((--oth)->type == EIO_DT_DIR)
1144	{	1725	{
1145	eio_dirent tmp = *dir;	1726	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1146	*dir = *ent;
1147	*ent = tmp;
1148		1727
1149	++dir;	1728	++dir;
1150	}	1729	}
1151	}	1730	}
		1731
		1732	/* now sort the dirs only (dirs all have the same score) */
		1733	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1152	}	1734	}
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		1735
1168	break;	1736	break;
1169	}	1737	}
1170		1738
1171	/* now add the entry to our list(s) */	1739	/* now add the entry to our list(s) */
1172	name = entp->d_name;	1740	name = D_NAME (entp);
1173		1741
1174	/* skip . and .. entries */	1742	/* skip . and .. entries */
1175	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1743	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1176	{	1744	{
1177	int len = D_NAMLEN (entp) + 1;	1745	int len = D_NAMLEN (entp) + 1;
1178		1746
1179	while (expect_false (namesoffs + len > namesalloc))	1747	while (ecb_expect_false (namesoffs + len > namesalloc))
1180	{	1748	{
1181	namesalloc *= 2;	1749	namesalloc *= 2;
1182	X_LOCK (wrklock);
1183	req->ptr1 = names = realloc (names, namesalloc);	1750	req->ptr2 = names = realloc (names, namesalloc);
1184	X_UNLOCK (wrklock);
1185		1751
1186	if (!names)	1752	if (!names)
1187	break;	1753	break;
1188	}	1754	}
1189		1755
…		…
1191		1757
1192	if (dents)	1758	if (dents)
1193	{	1759	{
1194	struct eio_dirent *ent;	1760	struct eio_dirent *ent;
1195		1761
1196	if (expect_false (dentoffs == dentalloc))	1762	if (ecb_expect_false (dentoffs == dentalloc))
1197	{	1763	{
1198	dentalloc *= 2;	1764	dentalloc *= 2;
1199	X_LOCK (wrklock);
1200	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1765	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1201	X_UNLOCK (wrklock);
1202		1766
1203	if (!dents)	1767	if (!dents)
1204	break;	1768	break;
1205	}	1769	}
1206		1770
1207	ent = dents + dentoffs;	1771	ent = dents + dentoffs;
1208		1772
1209	ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */	1773	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1210	ent->namelen = len - 1;	1774	ent->namelen = len - 1;
1211	ent->inode = D_INO (entp);	1775	ent->inode = D_INO (entp);
		1776
		1777	inode_bits |= ent->inode;
1212		1778
1213	switch (D_TYPE (entp))	1779	switch (D_TYPE (entp))
1214	{	1780	{
1215	default:	1781	default:
1216	ent->type = EIO_DT_UNKNOWN;	1782	ent->type = EIO_DT_UNKNOWN;
…		…
1259	#ifdef DT_WHT	1825	#ifdef DT_WHT
1260	case DT_WHT: ent->type = EIO_DT_WHT; break;	1826	case DT_WHT: ent->type = EIO_DT_WHT; break;
1261	#endif	1827	#endif
1262	}	1828	}
1263		1829
1264	ent->score = 0;	1830	ent->score = 7;
1265		1831
1266	if (flags & EIO_READDIR_DIRS_FIRST)	1832	if (flags & EIO_READDIR_DIRS_FIRST)
1267	{	1833	{
1268	if (ent->type == EIO_DT_UNKNOWN)	1834	if (ent->type == EIO_DT_UNKNOWN)
1269	{	1835	{
1270	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */	1836	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1271	ent->score = 98;	1837	ent->score = 1;
1272	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */	1838	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 */	1839	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1274	}	1840	}
1275	else if (ent->type == EIO_DT_DIR)	1841	else if (ent->type == EIO_DT_DIR)
1276	ent->score = 100;	1842	ent->score = 0;
1277	}	1843	}
1278	}	1844	}
1279		1845
1280	namesoffs += len;	1846	namesoffs += len;
1281	++dentoffs;	1847	++dentoffs;
…		…
1284	if (EIO_CANCELLED (req))	1850	if (EIO_CANCELLED (req))
1285	{	1851	{
1286	errno = ECANCELED;	1852	errno = ECANCELED;
1287	break;	1853	break;
1288	}	1854	}
		1855
		1856	#ifdef _WIN32
		1857	if (!FindNextFile (dirp, &entp))
		1858	{
		1859	FindClose (dirp);
		1860	dirp = 0;
		1861	}
		1862	#endif
1289	}	1863	}
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	}	1864	}
1321		1865
1322	/*****************************************************************************/	1866	/*****************************************************************************/
1323		1867
1324	#define ALLOC(len) \	1868	#define ALLOC(len) \
…		…
1340	{	1884	{
1341	ETP_REQ *req;	1885	ETP_REQ *req;
1342	struct timespec ts;	1886	struct timespec ts;
1343	etp_worker *self = (etp_worker *)thr_arg;	1887	etp_worker *self = (etp_worker *)thr_arg;
1344		1888
1345	/* try to distribute timeouts somewhat randomly */	1889	/* try to distribute timeouts somewhat evenly */
1346	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1890	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1347		1891
1348	for (;;)	1892	for (;;)
1349	{	1893	{
		1894	ts.tv_sec = 0;
		1895
1350	X_LOCK (reqlock);	1896	X_LOCK (reqlock);
1351		1897
1352	for (;;)	1898	for (;;)
1353	{	1899	{
1354	self->req = req = reqq_shift (&req_queue);	1900	self->req = req = reqq_shift (&req_queue);
1355		1901
1356	if (req)	1902	if (req)
1357	break;	1903	break;
1358		1904
		1905	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1906	{
		1907	X_UNLOCK (reqlock);
		1908	X_LOCK (wrklock);
		1909	--started;
		1910	X_UNLOCK (wrklock);
		1911	goto quit;
		1912	}
		1913
1359	++idle;	1914	++idle;
1360		1915
1361	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1916	if (idle <= max_idle)
1362	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1917	/* we are allowed to idle, so do so without any timeout */
		1918	X_COND_WAIT (reqwait, reqlock);
		1919	else
1363	{	1920	{
1364	if (idle > max_idle)	1921	/* initialise timeout once */
1365	{	1922	if (!ts.tv_sec)
1366	--idle;	1923	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		1924
1374	/* we are allowed to idle, so do so without any timeout */
1375	X_COND_WAIT (reqwait, reqlock);	1925	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1926	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1376	}	1927	}
1377		1928
1378	--idle;	1929	--idle;
1379	}	1930	}
1380		1931
…		…
1383	X_UNLOCK (reqlock);	1934	X_UNLOCK (reqlock);
1384		1935
1385	if (req->type < 0)	1936	if (req->type < 0)
1386	goto quit;	1937	goto quit;
1387		1938
1388	if (!EIO_CANCELLED (req))
1389	ETP_EXECUTE (self, req);	1939	ETP_EXECUTE (self, req);
1390		1940
1391	X_LOCK (reslock);	1941	X_LOCK (reslock);
1392		1942
1393	++npending;	1943	++npending;
1394		1944
…		…
1409	return 0;	1959	return 0;
1410	}	1960	}
1411		1961
1412	/*****************************************************************************/	1962	/*****************************************************************************/
1413		1963
		1964	int ecb_cold
1414	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1965	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1415	{	1966	{
		1967	#if !HAVE_PREADWRITE
		1968	X_MUTEX_CREATE (preadwritelock);
		1969	#endif
		1970
1416	return etp_init (want_poll, done_poll);	1971	return etp_init (want_poll, done_poll);
1417	}	1972	}
1418		1973
		1974	ecb_inline void
1419	static void eio_api_destroy (eio_req *req)	1975	eio_api_destroy (eio_req *req)
1420	{	1976	{
1421	free (req);	1977	free (req);
1422	}	1978	}
1423		1979
1424	#define REQ(rtype) \	1980	#define REQ(rtype) \
…		…
1443	{ \	1999	{ \
1444	eio_api_destroy (req); \	2000	eio_api_destroy (req); \
1445	return 0; \	2001	return 0; \
1446	}	2002	}
1447		2003
		2004	static void
1448	static void eio_execute (etp_worker *self, eio_req *req)	2005	eio_execute (etp_worker *self, eio_req *req)
1449	{	2006	{
1450	errno = 0;	2007	if (ecb_expect_false (EIO_CANCELLED (req)))
		2008	{
		2009	req->result = -1;
		2010	req->errorno = ECANCELED;
		2011	return;
		2012	}
1451		2013
1452	switch (req->type)	2014	switch (req->type)
1453	{	2015	{
1454	case EIO_READ: ALLOC (req->size);	2016	case EIO_READ: ALLOC (req->size);
1455	req->result = req->offs >= 0	2017	req->result = req->offs >= 0
…		…
1458	case EIO_WRITE: req->result = req->offs >= 0	2020	case EIO_WRITE: req->result = req->offs >= 0
1459	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2021	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1460	: write (req->int1, req->ptr2, req->size); break;	2022	: write (req->int1, req->ptr2, req->size); break;
1461		2023
1462	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2024	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1463	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2025	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1464		2026
1465	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2027	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1466	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2028	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1467	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2029	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1468	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2030	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1469	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2031	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1470	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2032	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2033
		2034	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2035	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2036	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2037	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1471		2038
1472	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	2039	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1473	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	2040	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1474	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	2041	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1475	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	2042	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
…		…
1483	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	2050	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1484	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	2051	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1485	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2052	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1486	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2053	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1487	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2054	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;	2055	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1489		2056
		2057	case EIO_REALPATH: eio__realpath (req, self); break;
		2058
1490	case EIO_READLINK: ALLOC (NAME_MAX);	2059	case EIO_READLINK: ALLOC (PATH_MAX);
1491	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	2060	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1492		2061
1493	case EIO_SYNC: req->result = 0; sync (); break;	2062	case EIO_SYNC: req->result = 0; sync (); break;
1494	case EIO_FSYNC: req->result = fsync (req->int1); break;	2063	case EIO_FSYNC: req->result = fsync (req->int1); break;
1495	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2064	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1496	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;	2065	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2066	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1497	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	2067	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2068	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1498	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	2069	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2070	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1499		2071
1500	case EIO_READDIR: eio__scandir (req, self); break;	2072	case EIO_READDIR: eio__scandir (req, self); break;
1501		2073
1502	case EIO_BUSY:	2074	case EIO_BUSY:
1503	#ifdef _WIN32	2075	#ifdef _WIN32
1504	Sleep (req->nv1 * 1000.);	2076	Sleep (req->nv1 * 1e3);
1505	#else	2077	#else
1506	{	2078	{
1507	struct timeval tv;	2079	struct timeval tv;
1508		2080
1509	tv.tv_sec = req->nv1;	2081	tv.tv_sec = req->nv1;
1510	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	2082	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1511		2083
1512	req->result = select (0, 0, 0, 0, &tv);	2084	req->result = select (0, 0, 0, 0, &tv);
1513	}	2085	}
1514	#endif	2086	#endif
1515	break;	2087	break;
…		…
1530	times = tv;	2102	times = tv;
1531	}	2103	}
1532	else	2104	else
1533	times = 0;	2105	times = 0;
1534		2106
1535
1536	req->result = req->type == EIO_FUTIME	2107	req->result = req->type == EIO_FUTIME
1537	? futimes (req->int1, times)	2108	? futimes (req->int1, times)
1538	: utimes (req->ptr1, times);	2109	: utimes (req->ptr1, times);
1539	}	2110	}
1540	break;	2111	break;
…		…
1545	case EIO_NOP:	2116	case EIO_NOP:
1546	req->result = 0;	2117	req->result = 0;
1547	break;	2118	break;
1548		2119
1549	case EIO_CUSTOM:	2120	case EIO_CUSTOM:
1550	((void (*)(eio_req *))req->feed) (req);	2121	req->feed (req);
1551	break;	2122	break;
1552		2123
1553	default:	2124	default:
		2125	errno = ENOSYS;
1554	req->result = -1;	2126	req->result = -1;
1555	break;	2127	break;
1556	}	2128	}
1557		2129
1558	req->errorno = errno;	2130	req->errorno = errno;
…		…
1588	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2160	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1589	{	2161	{
1590	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2162	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1591	}	2163	}
1592		2164
		2165	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2166	{
		2167	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2168	}
		2169
		2170	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2171	{
		2172	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2173	}
		2174
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)	2175	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1594	{	2176	{
1595	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2177	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1596	}	2178	}
1597		2179
		2180	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2181	{
		2182	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2183	}
		2184
1598	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2185	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1599	{	2186	{
1600	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2187	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1601	}	2188	}
1602		2189
…		…
1623	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2210	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1624	{	2211	{
1625	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2212	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1626	}	2213	}
1627		2214
		2215	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2216	{
		2217	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2218	}
		2219
1628	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2220	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1629	{	2221	{
1630	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2222	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1631	}	2223	}
1632		2224
…		…
1638	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2230	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1639	{	2231	{
1640	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2232	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1641	}	2233	}
1642		2234
1643	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2235	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1644	{	2236	{
1645	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2237	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1646	}	2238	}
1647		2239
1648	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2240	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1668	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2260	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1669	{	2261	{
1670	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2262	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1671	}	2263	}
1672		2264
1673	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2265	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1674	{	2266	{
1675	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2267	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1676	}	2268	}
1677		2269
1678	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2270	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1694	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2286	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1695	{	2287	{
1696	return eio__1path (EIO_READLINK, path, pri, cb, data);	2288	return eio__1path (EIO_READLINK, path, pri, cb, data);
1697	}	2289	}
1698		2290
		2291	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2292	{
		2293	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2294	}
		2295
1699	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2296	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1700	{	2297	{
1701	return eio__1path (EIO_STAT, path, pri, cb, data);	2298	return eio__1path (EIO_STAT, path, pri, cb, data);
1702	}	2299	}
1703		2300
1704	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2301	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1705	{	2302	{
1706	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2303	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1707	}	2304	}
1708		2305
		2306	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2307	{
		2308	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2309	}
		2310
1709	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2311	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1710	{	2312	{
1711	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2313	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1712	}	2314	}
1713		2315
…		…
1721	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2323	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1722	}	2324	}
1723		2325
1724	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2326	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1725	{	2327	{
1726	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2328	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1727	}	2329	}
1728		2330
1729	static eio_req *	2331	static eio_req *
1730	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2332	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1731	{	2333	{
…		…
1755	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2357	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1756	{	2358	{
1757	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2359	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1758	}	2360	}
1759		2361
1760	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2362	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1761	{	2363	{
1762	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2364	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1763	}	2365	}
1764		2366
1765	#endif	2367	#endif
1766		2368
1767	eio_req *eio_grp (eio_cb cb, void *data)	2369	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1776	#undef SEND	2378	#undef SEND
1777		2379
1778	/*****************************************************************************/	2380	/*****************************************************************************/
1779	/* grp functions */	2381	/* grp functions */
1780		2382
		2383	void
1781	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2384	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1782	{	2385	{
1783	grp->int2 = limit;	2386	grp->int2 = limit;
1784	grp->feed = feed;	2387	grp->feed = feed;
1785		2388
1786	grp_try_feed (grp);	2389	grp_try_feed (grp);
1787	}	2390	}
1788		2391
		2392	void
1789	void eio_grp_limit (eio_req *grp, int limit)	2393	eio_grp_limit (eio_req *grp, int limit)
1790	{	2394	{
1791	grp->int2 = limit;	2395	grp->int2 = limit;
1792		2396
1793	grp_try_feed (grp);	2397	grp_try_feed (grp);
1794	}	2398	}
1795		2399
		2400	void
1796	void eio_grp_add (eio_req *grp, eio_req *req)	2401	eio_grp_add (eio_req *grp, eio_req *req)
1797	{	2402	{
1798	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2403	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1799		2404
1800	grp->flags |= EIO_FLAG_GROUPADD;	2405	grp->flags |= EIO_FLAG_GROUPADD;
1801		2406
…		…
1812	}	2417	}
1813		2418
1814	/*****************************************************************************/	2419	/*****************************************************************************/
1815	/* misc garbage */	2420	/* misc garbage */
1816		2421
		2422	eio_ssize_t
1817	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2423	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1818	{	2424	{
1819	etp_worker wrk;
1820	ssize_t ret;
1821
1822	wrk.dbuf = 0;
1823
1824	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2425	return eio__sendfile (ofd, ifd, offset, count);
1825
1826	if (wrk.dbuf)
1827	free (wrk.dbuf);
1828
1829	return ret;
1830	}	2426	}
1831		2427

Diff Legend

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