ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.33 by root, Sat Jun 6 19:44:17 2009 UTC vs.
Revision 1.95 by root, Tue Jul 19 04:56:43 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>
		167	#include <sys/statvfs.h>
78	# include <unistd.h>	168	#include <unistd.h>
79	# include <utime.h>	169	#include <utime.h>
80	# include <signal.h>	170	#include <signal.h>
81	# include <dirent.h>	171	#include <dirent.h>
82		172
		173	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		174	#include <sys/mman.h>
		175	#endif
		176
		177	#define D_NAME(entp) entp->d_name
		178
83	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */	179	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84	# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)	180	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
85	# define _DIRENT_HAVE_D_TYPE /* sigh */	181	#define _DIRENT_HAVE_D_TYPE /* sigh */
86	# define D_INO(de) (de)->d_fileno	182	#define D_INO(de) (de)->d_fileno
87	# define D_NAMLEN(de) (de)->d_namlen	183	#define D_NAMLEN(de) (de)->d_namlen
88	# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600	184	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
89	# define D_INO(de) (de)->d_ino	185	#define D_INO(de) (de)->d_ino
90	# endif	186	#endif
91		187
92	#ifdef _D_EXACT_NAMLEN	188	#ifdef _D_EXACT_NAMLEN
93	# undef D_NAMLEN	189	#undef D_NAMLEN
94	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)	190	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
95	#endif	191	#endif
96		192
97	# ifdef _DIRENT_HAVE_D_TYPE	193	#ifdef _DIRENT_HAVE_D_TYPE
98	# define D_TYPE(de) (de)->d_type	194	#define D_TYPE(de) (de)->d_type
99	# endif	195	#endif
100		196
101	# ifndef EIO_STRUCT_DIRENT	197	#ifndef EIO_STRUCT_DIRENT
102	# define EIO_STRUCT_DIRENT struct dirent	198	#define EIO_STRUCT_DIRENT struct dirent
103	# endif	199	#endif
104		200
105	#endif	201	#endif
106		202
107	#if HAVE_SENDFILE	203	#if HAVE_SENDFILE
108	# if __linux	204	# if __linux
109	# include <sys/sendfile.h>	205	# include <sys/sendfile.h>
110	# elif __freebsd	206	# elif __FreeBSD__ || defined __APPLE__
111	# include <sys/socket.h>	207	# include <sys/socket.h>
112	# include <sys/uio.h>	208	# include <sys/uio.h>
113	# elif __hpux	209	# elif __hpux
114	# include <sys/socket.h>	210	# include <sys/socket.h>
115	# elif __solaris /* not yet */	211	# elif __solaris
116	# include <sys/sendfile.h>	212	# include <sys/sendfile.h>
117	# else	213	# else
118	# error sendfile support requested but not available	214	# error sendfile support requested but not available
119	# endif	215	# endif
120	#endif	216	#endif
…		…
124	#endif	220	#endif
125	#ifndef D_INO	221	#ifndef D_INO
126	# define D_INO(de) 0	222	# define D_INO(de) 0
127	#endif	223	#endif
128	#ifndef D_NAMLEN	224	#ifndef D_NAMLEN
129	# define D_NAMLEN(de) strlen ((de)->d_name)	225	# define D_NAMLEN(entp) strlen (D_NAME (entp))
130	#endif	226	#endif
131
132	/* number of seconds after which an idle threads exit */
133	#define IDLE_TIMEOUT 10
134		227
135	/* used for struct dirent, AIX doesn't provide it */	228	/* used for struct dirent, AIX doesn't provide it */
136	#ifndef NAME_MAX	229	#ifndef NAME_MAX
137	# define NAME_MAX 4096	230	# define NAME_MAX 4096
138	#endif	231	#endif
139		232
		233	/* used for readlink etc. */
		234	#ifndef PATH_MAX
		235	# define PATH_MAX 4096
		236	#endif
		237
140	/* buffer size for various temporary buffers */	238	/* buffer size for various temporary buffers */
141	#define EIO_BUFSIZE 65536	239	#define EIO_BUFSIZE 65536
142		240
143	#define dBUF \	241	#define dBUF \
144	char *eio_buf; \
145	ETP_WORKER_LOCK (self); \
146	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	242	char *eio_buf = malloc (EIO_BUFSIZE); \
147	ETP_WORKER_UNLOCK (self); \
148	errno = ENOMEM; \	243	errno = ENOMEM; \
149	if (!eio_buf) \	244	if (!eio_buf) \
150	return -1;	245	return -1
		246
		247	#define FUBd \
		248	free (eio_buf)
151		249
152	#define EIO_TICKS ((1000000 + 1023) >> 10)	250	#define EIO_TICKS ((1000000 + 1023) >> 10)
153
154	/*****************************************************************************/
155
156	#if __GNUC__ >= 3
157	# define expect(expr,value) __builtin_expect ((expr),(value))
158	#else
159	# define expect(expr,value) (expr)
160	#endif
161
162	#define expect_false(expr) expect ((expr) != 0, 0)
163	#define expect_true(expr) expect ((expr) != 0, 1)
164
165	/*****************************************************************************/
166		251
167	#define ETP_PRI_MIN EIO_PRI_MIN	252	#define ETP_PRI_MIN EIO_PRI_MIN
168	#define ETP_PRI_MAX EIO_PRI_MAX	253	#define ETP_PRI_MAX EIO_PRI_MAX
169		254
170	struct etp_worker;	255	struct etp_worker;
…		…
174	static int eio_finish (eio_req *req);	259	static int eio_finish (eio_req *req);
175	#define ETP_FINISH(req) eio_finish (req)	260	#define ETP_FINISH(req) eio_finish (req)
176	static void eio_execute (struct etp_worker *self, eio_req *req);	261	static void eio_execute (struct etp_worker *self, eio_req *req);
177	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	262	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
178		263
179	#define ETP_WORKER_CLEAR(req) \
180	if (wrk->dbuf) \
181	{ \
182	free (wrk->dbuf); \
183	wrk->dbuf = 0; \
184	} \
185	\
186	if (wrk->dirp) \
187	{ \
188	closedir (wrk->dirp); \
189	wrk->dirp = 0; \
190	}
191
192	#define ETP_WORKER_COMMON \
193	void *dbuf; \
194	DIR *dirp;
195
196	/*****************************************************************************/	264	/*****************************************************************************/
197		265
198	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	266	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
199		267
200	/* calculcate time difference in ~1/EIO_TICKS of a second */	268	/* calculate time difference in ~1/EIO_TICKS of a second */
		269	ecb_inline int
201	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	270	tvdiff (struct timeval *tv1, struct timeval *tv2)
202	{	271	{
203	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	272	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
204	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	273	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
205	}	274	}
206		275
…		…
213	static unsigned int max_poll_reqs; /* reslock */	282	static unsigned int max_poll_reqs; /* reslock */
214		283
215	static volatile unsigned int nreqs; /* reqlock */	284	static volatile unsigned int nreqs; /* reqlock */
216	static volatile unsigned int nready; /* reqlock */	285	static volatile unsigned int nready; /* reqlock */
217	static volatile unsigned int npending; /* reqlock */	286	static volatile unsigned int npending; /* reqlock */
218	static volatile unsigned int max_idle = 4;	287	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		288	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
219		289
220	static mutex_t wrklock = X_MUTEX_INIT;	290	static xmutex_t wrklock;
221	static mutex_t reslock = X_MUTEX_INIT;	291	static xmutex_t reslock;
222	static mutex_t reqlock = X_MUTEX_INIT;	292	static xmutex_t reqlock;
223	static cond_t reqwait = X_COND_INIT;	293	static xcond_t reqwait;
224		294
225	#if !HAVE_PREADWRITE	295	#if !HAVE_PREADWRITE
226	/*	296	/*
227	* make our pread/pwrite emulation safe against themselves, but not against	297	* make our pread/pwrite emulation safe against themselves, but not against
228	* normal read/write by using a mutex. slows down execution a lot,	298	* normal read/write by using a mutex. slows down execution a lot,
229	* but that's your problem, not mine.	299	* but that's your problem, not mine.
230	*/	300	*/
231	static mutex_t preadwritelock = X_MUTEX_INIT;	301	static xmutex_t preadwritelock;
232	#endif	302	#endif
233		303
234	typedef struct etp_worker	304	typedef struct etp_worker
235	{	305	{
236	/* locked by wrklock */	306	/* locked by wrklock */
237	struct etp_worker *prev, *next;	307	struct etp_worker *prev, *next;
238		308
239	thread_t tid;	309	xthread_t tid;
240		310
241	/* locked by reslock, reqlock or wrklock */	311	/* locked by reslock, reqlock or wrklock */
242	ETP_REQ *req; /* currently processed request */	312	ETP_REQ *req; /* currently processed request */
243		313
		314	#ifdef ETP_WORKER_COMMON
244	ETP_WORKER_COMMON	315	ETP_WORKER_COMMON
		316	#endif
245	} etp_worker;	317	} etp_worker;
246		318
247	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	319	static etp_worker wrk_first; /* NOT etp */
248		320
249	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	321	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
250	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	322	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
251		323
252	/* worker threads management */	324	/* worker threads management */
253		325
		326	static void ecb_cold
254	static void etp_worker_clear (etp_worker *wrk)	327	etp_worker_clear (etp_worker *wrk)
255	{	328	{
256	ETP_WORKER_CLEAR (wrk);
257	}	329	}
258		330
		331	static void ecb_cold
259	static void etp_worker_free (etp_worker *wrk)	332	etp_worker_free (etp_worker *wrk)
260	{	333	{
261	wrk->next->prev = wrk->prev;	334	wrk->next->prev = wrk->prev;
262	wrk->prev->next = wrk->next;	335	wrk->prev->next = wrk->next;
263		336
264	free (wrk);	337	free (wrk);
265	}	338	}
266		339
267	static unsigned int etp_nreqs (void)	340	static unsigned int
		341	etp_nreqs (void)
268	{	342	{
269	int retval;	343	int retval;
270	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	344	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271	retval = nreqs;	345	retval = nreqs;
272	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	346	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273	return retval;	347	return retval;
274	}	348	}
275		349
276	static unsigned int etp_nready (void)	350	static unsigned int
		351	etp_nready (void)
277	{	352	{
278	unsigned int retval;	353	unsigned int retval;
279		354
280	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	355	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281	retval = nready;	356	retval = nready;
282	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	357	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283		358
284	return retval;	359	return retval;
285	}	360	}
286		361
287	static unsigned int etp_npending (void)	362	static unsigned int
		363	etp_npending (void)
288	{	364	{
289	unsigned int retval;	365	unsigned int retval;
290		366
291	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	367	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
292	retval = npending;	368	retval = npending;
293	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	369	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
294		370
295	return retval;	371	return retval;
296	}	372	}
297		373
298	static unsigned int etp_nthreads (void)	374	static unsigned int
		375	etp_nthreads (void)
299	{	376	{
300	unsigned int retval;	377	unsigned int retval;
301		378
302	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	379	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303	retval = started;	380	retval = started;
…		…
317	} etp_reqq;	394	} etp_reqq;
318		395
319	static etp_reqq req_queue;	396	static etp_reqq req_queue;
320	static etp_reqq res_queue;	397	static etp_reqq res_queue;
321		398
		399	static void ecb_noinline ecb_cold
		400	reqq_init (etp_reqq *q)
		401	{
		402	int pri;
		403
		404	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		405	q->qs[pri] = q->qe[pri] = 0;
		406
		407	q->size = 0;
		408	}
		409
		410	static int ecb_noinline
322	static int reqq_push (etp_reqq *q, ETP_REQ *req)	411	reqq_push (etp_reqq *q, ETP_REQ *req)
323	{	412	{
324	int pri = req->pri;	413	int pri = req->pri;
325	req->next = 0;	414	req->next = 0;
326		415
327	if (q->qe[pri])	416	if (q->qe[pri])
…		…
333	q->qe[pri] = q->qs[pri] = req;	422	q->qe[pri] = q->qs[pri] = req;
334		423
335	return q->size++;	424	return q->size++;
336	}	425	}
337		426
		427	static ETP_REQ * ecb_noinline
338	static ETP_REQ *reqq_shift (etp_reqq *q)	428	reqq_shift (etp_reqq *q)
339	{	429	{
340	int pri;	430	int pri;
341		431
342	if (!q->size)	432	if (!q->size)
343	return 0;	433	return 0;
…		…
358	}	448	}
359		449
360	abort ();	450	abort ();
361	}	451	}
362		452
363	static void etp_atfork_prepare (void)	453	static int ecb_cold
		454	etp_init (void (*want_poll)(void), void (*done_poll)(void))
364	{	455	{
365	X_LOCK (wrklock);	456	X_MUTEX_CREATE (wrklock);
366	X_LOCK (reqlock);	457	X_MUTEX_CREATE (reslock);
367	X_LOCK (reslock);	458	X_MUTEX_CREATE (reqlock);
368	#if !HAVE_PREADWRITE	459	X_COND_CREATE (reqwait);
369	X_LOCK (preadwritelock);
370	#endif
371	}
372		460
373	static void etp_atfork_parent (void)	461	reqq_init (&req_queue);
374	{	462	reqq_init (&res_queue);
375	#if !HAVE_PREADWRITE
376	X_UNLOCK (preadwritelock);
377	#endif
378	X_UNLOCK (reslock);
379	X_UNLOCK (reqlock);
380	X_UNLOCK (wrklock);
381	}
382		463
383	static void etp_atfork_child (void)	464	wrk_first.next =
384	{	465	wrk_first.prev = &wrk_first;
385	ETP_REQ *prv;
386
387	while ((prv = reqq_shift (&req_queue)))
388	ETP_DESTROY (prv);
389
390	while ((prv = reqq_shift (&res_queue)))
391	ETP_DESTROY (prv);
392
393	while (wrk_first.next != &wrk_first)
394	{
395	etp_worker *wrk = wrk_first.next;
396
397	if (wrk->req)
398	ETP_DESTROY (wrk->req);
399
400	etp_worker_clear (wrk);
401	etp_worker_free (wrk);
402	}
403		466
404	started = 0;	467	started = 0;
405	idle = 0;	468	idle = 0;
406	nreqs = 0;	469	nreqs = 0;
407	nready = 0;	470	nready = 0;
408	npending = 0;	471	npending = 0;
409		472
410	etp_atfork_parent ();
411	}
412
413	static void
414	etp_once_init (void)
415	{
416	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
417	}
418
419	static int
420	etp_init (void (*want_poll)(void), void (*done_poll)(void))
421	{
422	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
423
424	pthread_once (&doinit, etp_once_init);
425
426	want_poll_cb = want_poll;	473	want_poll_cb = want_poll;
427	done_poll_cb = done_poll;	474	done_poll_cb = done_poll;
428		475
429	return 0;	476	return 0;
430	}	477	}
431		478
432	X_THREAD_PROC (etp_proc);	479	X_THREAD_PROC (etp_proc);
433		480
		481	static void ecb_cold
434	static void etp_start_thread (void)	482	etp_start_thread (void)
435	{	483	{
436	etp_worker *wrk = calloc (1, sizeof (etp_worker));	484	etp_worker *wrk = calloc (1, sizeof (etp_worker));
437		485
438	/*TODO*/	486	/*TODO*/
439	assert (("unable to allocate worker thread data", wrk));	487	assert (("unable to allocate worker thread data", wrk));
…		…
452	free (wrk);	500	free (wrk);
453		501
454	X_UNLOCK (wrklock);	502	X_UNLOCK (wrklock);
455	}	503	}
456		504
		505	static void
457	static void etp_maybe_start_thread (void)	506	etp_maybe_start_thread (void)
458	{	507	{
459	if (expect_true (etp_nthreads () >= wanted))	508	if (ecb_expect_true (etp_nthreads () >= wanted))
460	return;	509	return;
461		510
462	/* todo: maybe use idle here, but might be less exact */	511	/* todo: maybe use idle here, but might be less exact */
463	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	512	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
464	return;	513	return;
465		514
466	etp_start_thread ();	515	etp_start_thread ();
467	}	516	}
468		517
		518	static void ecb_cold
469	static void etp_end_thread (void)	519	etp_end_thread (void)
470	{	520	{
471	eio_req *req = calloc (1, sizeof (eio_req));	521	eio_req *req = calloc (1, sizeof (eio_req));
472		522
473	req->type = -1;	523	req->type = -1;
474	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	524	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
481	X_LOCK (wrklock);	531	X_LOCK (wrklock);
482	--started;	532	--started;
483	X_UNLOCK (wrklock);	533	X_UNLOCK (wrklock);
484	}	534	}
485		535
486	static int etp_poll (void)	536	static int
		537	etp_poll (void)
487	{	538	{
488	unsigned int maxreqs;	539	unsigned int maxreqs;
489	unsigned int maxtime;	540	unsigned int maxtime;
490	struct timeval tv_start, tv_now;	541	struct timeval tv_start, tv_now;
491		542
…		…
521		572
522	X_LOCK (reqlock);	573	X_LOCK (reqlock);
523	--nreqs;	574	--nreqs;
524	X_UNLOCK (reqlock);	575	X_UNLOCK (reqlock);
525		576
526	if (expect_false (req->type == EIO_GROUP && req->size))	577	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
527	{	578	{
528	req->int1 = 1; /* mark request as delayed */	579	req->int1 = 1; /* mark request as delayed */
529	continue;	580	continue;
530	}	581	}
531	else	582	else
532	{	583	{
533	int res = ETP_FINISH (req);	584	int res = ETP_FINISH (req);
534	if (expect_false (res))	585	if (ecb_expect_false (res))
535	return res;	586	return res;
536	}	587	}
537		588
538	if (expect_false (maxreqs && !--maxreqs))	589	if (ecb_expect_false (maxreqs && !--maxreqs))
539	break;	590	break;
540		591
541	if (maxtime)	592	if (maxtime)
542	{	593	{
543	gettimeofday (&tv_now, 0);	594	gettimeofday (&tv_now, 0);
…		…
549		600
550	errno = EAGAIN;	601	errno = EAGAIN;
551	return -1;	602	return -1;
552	}	603	}
553		604
		605	static void
554	static void etp_cancel (ETP_REQ *req)	606	etp_cancel (ETP_REQ *req)
555	{	607	{
556	X_LOCK (wrklock);	608	req->cancelled = 1;
557	req->flags |= EIO_FLAG_CANCELLED;
558	X_UNLOCK (wrklock);
559		609
560	eio_grp_cancel (req);	610	eio_grp_cancel (req);
561	}	611	}
562		612
		613	static void
563	static void etp_submit (ETP_REQ *req)	614	etp_submit (ETP_REQ *req)
564	{	615	{
565	req->pri -= ETP_PRI_MIN;	616	req->pri -= ETP_PRI_MIN;
566		617
567	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	618	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
568	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	619	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
569		620
570	if (expect_false (req->type == EIO_GROUP))	621	if (ecb_expect_false (req->type == EIO_GROUP))
571	{	622	{
572	/* I hope this is worth it :/ */	623	/* I hope this is worth it :/ */
573	X_LOCK (reqlock);	624	X_LOCK (reqlock);
574	++nreqs;	625	++nreqs;
575	X_UNLOCK (reqlock);	626	X_UNLOCK (reqlock);
…		…
594		645
595	etp_maybe_start_thread ();	646	etp_maybe_start_thread ();
596	}	647	}
597	}	648	}
598		649
		650	static void ecb_cold
599	static void etp_set_max_poll_time (double nseconds)	651	etp_set_max_poll_time (double nseconds)
600	{	652	{
601	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	653	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
602	max_poll_time = nseconds;	654	max_poll_time = nseconds * EIO_TICKS;
603	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	655	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
604	}	656	}
605		657
		658	static void ecb_cold
606	static void etp_set_max_poll_reqs (unsigned int maxreqs)	659	etp_set_max_poll_reqs (unsigned int maxreqs)
607	{	660	{
608	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	661	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609	max_poll_reqs = maxreqs;	662	max_poll_reqs = maxreqs;
610	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	663	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611	}	664	}
612		665
		666	static void ecb_cold
613	static void etp_set_max_idle (unsigned int nthreads)	667	etp_set_max_idle (unsigned int nthreads)
614	{	668	{
615	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	669	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
616	max_idle = nthreads <= 0 ? 1 : nthreads;	670	max_idle = nthreads;
617	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	671	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
618	}	672	}
619		673
		674	static void ecb_cold
		675	etp_set_idle_timeout (unsigned int seconds)
		676	{
		677	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		678	idle_timeout = seconds;
		679	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		680	}
		681
		682	static void ecb_cold
620	static void etp_set_min_parallel (unsigned int nthreads)	683	etp_set_min_parallel (unsigned int nthreads)
621	{	684	{
622	if (wanted < nthreads)	685	if (wanted < nthreads)
623	wanted = nthreads;	686	wanted = nthreads;
624	}	687	}
625		688
		689	static void ecb_cold
626	static void etp_set_max_parallel (unsigned int nthreads)	690	etp_set_max_parallel (unsigned int nthreads)
627	{	691	{
628	if (wanted > nthreads)	692	if (wanted > nthreads)
629	wanted = nthreads;	693	wanted = nthreads;
630		694
631	while (started > wanted)	695	while (started > wanted)
632	etp_end_thread ();	696	etp_end_thread ();
633	}	697	}
634		698
635	/*****************************************************************************/	699	/*****************************************************************************/
636		700
		701	static void
637	static void grp_try_feed (eio_req *grp)	702	grp_try_feed (eio_req *grp)
638	{	703	{
639	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	704	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
640	{	705	{
641	grp->flags &= ~EIO_FLAG_GROUPADD;	706	grp->flags &= ~EIO_FLAG_GROUPADD;
642		707
…		…
649	break;	714	break;
650	}	715	}
651	}	716	}
652	}	717	}
653		718
		719	static int
654	static int grp_dec (eio_req *grp)	720	grp_dec (eio_req *grp)
655	{	721	{
656	--grp->size;	722	--grp->size;
657		723
658	/* call feeder, if applicable */	724	/* call feeder, if applicable */
659	grp_try_feed (grp);	725	grp_try_feed (grp);
…		…
663	return eio_finish (grp);	729	return eio_finish (grp);
664	else	730	else
665	return 0;	731	return 0;
666	}	732	}
667		733
		734	static void
668	void eio_destroy (eio_req *req)	735	eio_destroy (eio_req *req)
669	{	736	{
670	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	737	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
671	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	738	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
672		739
673	EIO_DESTROY (req);	740	EIO_DESTROY (req);
674	}	741	}
675		742
		743	static int
676	static int eio_finish (eio_req *req)	744	eio_finish (eio_req *req)
677	{	745	{
678	int res = EIO_FINISH (req);	746	int res = EIO_FINISH (req);
679		747
680	if (req->grp)	748	if (req->grp)
681	{	749	{
…		…
689	if (grp->grp_first == req)	757	if (grp->grp_first == req)
690	grp->grp_first = req->grp_next;	758	grp->grp_first = req->grp_next;
691		759
692	res2 = grp_dec (grp);	760	res2 = grp_dec (grp);
693		761
694	if (!res && res2)	762	if (!res)
695	res = res2;	763	res = res2;
696	}	764	}
697		765
698	eio_destroy (req);	766	eio_destroy (req);
699		767
700	return res;	768	return res;
701	}	769	}
702		770
		771	void
703	void eio_grp_cancel (eio_req *grp)	772	eio_grp_cancel (eio_req *grp)
704	{	773	{
705	for (grp = grp->grp_first; grp; grp = grp->grp_next)	774	for (grp = grp->grp_first; grp; grp = grp->grp_next)
706	eio_cancel (grp);	775	eio_cancel (grp);
707	}	776	}
708		777
		778	void
709	void eio_cancel (eio_req *req)	779	eio_cancel (eio_req *req)
710	{	780	{
711	etp_cancel (req);	781	etp_cancel (req);
712	}	782	}
713		783
		784	void
714	void eio_submit (eio_req *req)	785	eio_submit (eio_req *req)
715	{	786	{
716	etp_submit (req);	787	etp_submit (req);
717	}	788	}
718		789
719	unsigned int eio_nreqs (void)	790	unsigned int
		791	eio_nreqs (void)
720	{	792	{
721	return etp_nreqs ();	793	return etp_nreqs ();
722	}	794	}
723		795
724	unsigned int eio_nready (void)	796	unsigned int
		797	eio_nready (void)
725	{	798	{
726	return etp_nready ();	799	return etp_nready ();
727	}	800	}
728		801
729	unsigned int eio_npending (void)	802	unsigned int
		803	eio_npending (void)
730	{	804	{
731	return etp_npending ();	805	return etp_npending ();
732	}	806	}
733		807
734	unsigned int eio_nthreads (void)	808	unsigned int ecb_cold
		809	eio_nthreads (void)
735	{	810	{
736	return etp_nthreads ();	811	return etp_nthreads ();
737	}	812	}
738		813
		814	void ecb_cold
739	void eio_set_max_poll_time (double nseconds)	815	eio_set_max_poll_time (double nseconds)
740	{	816	{
741	etp_set_max_poll_time (nseconds);	817	etp_set_max_poll_time (nseconds);
742	}	818	}
743		819
		820	void ecb_cold
744	void eio_set_max_poll_reqs (unsigned int maxreqs)	821	eio_set_max_poll_reqs (unsigned int maxreqs)
745	{	822	{
746	etp_set_max_poll_reqs (maxreqs);	823	etp_set_max_poll_reqs (maxreqs);
747	}	824	}
748		825
		826	void ecb_cold
749	void eio_set_max_idle (unsigned int nthreads)	827	eio_set_max_idle (unsigned int nthreads)
750	{	828	{
751	etp_set_max_idle (nthreads);	829	etp_set_max_idle (nthreads);
752	}	830	}
753		831
		832	void ecb_cold
		833	eio_set_idle_timeout (unsigned int seconds)
		834	{
		835	etp_set_idle_timeout (seconds);
		836	}
		837
		838	void ecb_cold
754	void eio_set_min_parallel (unsigned int nthreads)	839	eio_set_min_parallel (unsigned int nthreads)
755	{	840	{
756	etp_set_min_parallel (nthreads);	841	etp_set_min_parallel (nthreads);
757	}	842	}
758		843
		844	void ecb_cold
759	void eio_set_max_parallel (unsigned int nthreads)	845	eio_set_max_parallel (unsigned int nthreads)
760	{	846	{
761	etp_set_max_parallel (nthreads);	847	etp_set_max_parallel (nthreads);
762	}	848	}
763		849
764	int eio_poll (void)	850	int eio_poll (void)
…		…
773	# undef pread	859	# undef pread
774	# undef pwrite	860	# undef pwrite
775	# define pread eio__pread	861	# define pread eio__pread
776	# define pwrite eio__pwrite	862	# define pwrite eio__pwrite
777		863
778	static ssize_t	864	static eio_ssize_t
779	eio__pread (int fd, void *buf, size_t count, off_t offset)	865	eio__pread (int fd, void *buf, size_t count, off_t offset)
780	{	866	{
781	ssize_t res;	867	eio_ssize_t res;
782	off_t ooffset;	868	off_t ooffset;
783		869
784	X_LOCK (preadwritelock);	870	X_LOCK (preadwritelock);
785	ooffset = lseek (fd, 0, SEEK_CUR);	871	ooffset = lseek (fd, 0, SEEK_CUR);
786	lseek (fd, offset, SEEK_SET);	872	lseek (fd, offset, SEEK_SET);
…		…
789	X_UNLOCK (preadwritelock);	875	X_UNLOCK (preadwritelock);
790		876
791	return res;	877	return res;
792	}	878	}
793		879
794	static ssize_t	880	static eio_ssize_t
795	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	881	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
796	{	882	{
797	ssize_t res;	883	eio_ssize_t res;
798	off_t ooffset;	884	off_t ooffset;
799		885
800	X_LOCK (preadwritelock);	886	X_LOCK (preadwritelock);
801	ooffset = lseek (fd, 0, SEEK_CUR);	887	ooffset = lseek (fd, 0, SEEK_CUR);
802	lseek (fd, offset, SEEK_SET);	888	lseek (fd, offset, SEEK_SET);
…		…
806		892
807	return res;	893	return res;
808	}	894	}
809	#endif	895	#endif
810		896
811	#ifndef HAVE_FUTIMES	897	#ifndef HAVE_UTIMES
812		898
813	# undef utimes	899	# undef utimes
814	# undef futimes
815	# define utimes(path,times) eio__utimes (path, times)	900	# define utimes(path,times) eio__utimes (path, times)
816	# define futimes(fd,times) eio__futimes (fd, times)
817		901
818	static int	902	static int
819	eio__utimes (const char *filename, const struct timeval times[2])	903	eio__utimes (const char *filename, const struct timeval times[2])
820	{	904	{
821	if (times)	905	if (times)
…		…
829	}	913	}
830	else	914	else
831	return utime (filename, 0);	915	return utime (filename, 0);
832	}	916	}
833		917
		918	#endif
		919
		920	#ifndef HAVE_FUTIMES
		921
		922	# undef futimes
		923	# define futimes(fd,times) eio__futimes (fd, times)
		924
		925	static int
834	static int eio__futimes (int fd, const struct timeval tv[2])	926	eio__futimes (int fd, const struct timeval tv[2])
835	{	927	{
836	errno = ENOSYS;	928	errno = ENOSYS;
837	return -1;	929	return -1;
838	}	930	}
839		931
…		…
843	# undef fdatasync	935	# undef fdatasync
844	# define fdatasync(fd) fsync (fd)	936	# define fdatasync(fd) fsync (fd)
845	#endif	937	#endif
846		938
847	/* sync_file_range always needs emulation */	939	/* sync_file_range always needs emulation */
848	int	940	static int
849	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	941	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
850	{	942	{
851	#if HAVE_SYNC_FILE_RANGE	943	#if HAVE_SYNC_FILE_RANGE
852	int res;	944	int res;
853		945
…		…
866	if (!res || errno != ENOSYS)	958	if (!res || errno != ENOSYS)
867	return res;	959	return res;
868	#endif	960	#endif
869		961
870	/* even though we could play tricks with the flags, it's better to always	962	/* even though we could play tricks with the flags, it's better to always
871	* call fdatasync, as thta matches the expectation of it's users best */	963	* call fdatasync, as that matches the expectation of its users best */
872	return fdatasync (fd);	964	return fdatasync (fd);
		965	}
		966
		967	static int
		968	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		969	{
		970	#if HAVE_FALLOCATE
		971	return fallocate (fd, mode, offset, len);
		972	#else
		973	errno = ENOSYS;
		974	return -1;
		975	#endif
873	}	976	}
874		977
875	#if !HAVE_READAHEAD	978	#if !HAVE_READAHEAD
876	# undef readahead	979	# undef readahead
877	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	980	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
878		981
879	static ssize_t	982	static eio_ssize_t
880	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	983	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
881	{	984	{
882	size_t todo = count;	985	size_t todo = count;
883	dBUF;	986	dBUF;
884		987
…		…
889	pread (fd, eio_buf, len, offset);	992	pread (fd, eio_buf, len, offset);
890	offset += len;	993	offset += len;
891	todo -= len;	994	todo -= len;
892	}	995	}
893		996
		997	FUBd;
		998
894	errno = 0;	999	errno = 0;
895	return count;	1000	return count;
896	}	1001	}
897		1002
898	#endif	1003	#endif
899		1004
900	/* sendfile always needs emulation */	1005	/* sendfile always needs emulation */
901	static ssize_t	1006	static eio_ssize_t
902	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1007	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
903	{	1008	{
		1009	eio_ssize_t written = 0;
904	ssize_t res;	1010	eio_ssize_t res;
905		1011
906	if (!count)	1012	if (!count)
907	return 0;	1013	return 0;
908		1014
		1015	for (;;)
		1016	{
		1017	#ifdef __APPLE__
		1018	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1019	#endif
909	#if HAVE_SENDFILE	1020	#if HAVE_SENDFILE
910	# if __linux	1021	# if __linux
		1022	off_t soffset = offset;
911	res = sendfile (ofd, ifd, &offset, count);	1023	res = sendfile (ofd, ifd, &soffset, count);
912		1024
913	# elif __freebsd	1025	# elif __FreeBSD__
914	/*	1026	/*
915	* Of course, the freebsd sendfile is a dire hack with no thoughts	1027	* Of course, the freebsd sendfile is a dire hack with no thoughts
916	* wasted on making it similar to other I/O functions.	1028	* wasted on making it similar to other I/O functions.
917	*/	1029	*/
918	{
919	off_t sbytes;	1030	off_t sbytes;
920	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1031	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
921		1032
922	if (res < 0 && sbytes)	1033	#if 0 /* according to the manpage, this is correct, but broken behaviour */
923	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1034	/* freebsd' sendfile will return 0 on success */
		1035	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1036	/* not on e.g. EIO or EPIPE - sounds broken */
		1037	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
924	res = sbytes;	1038	res = sbytes;
925	}	1039	#endif
		1040
		1041	/* according to source inspection, this is correct, and useful behaviour */
		1042	if (sbytes)
		1043	res = sbytes;
		1044
		1045	# elif defined (__APPLE__)
		1046	off_t sbytes = count;
		1047	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1048
		1049	/* according to the manpage, sbytes is always valid */
		1050	if (sbytes)
		1051	res = sbytes;
926		1052
927	# elif __hpux	1053	# elif __hpux
928	res = sendfile (ofd, ifd, offset, count, 0, 0);	1054	res = sendfile (ofd, ifd, offset, count, 0, 0);
929		1055
930	# elif __solaris	1056	# elif __solaris
931	{
932	struct sendfilevec vec;	1057	struct sendfilevec vec;
933	size_t sbytes;	1058	size_t sbytes;
934		1059
935	vec.sfv_fd = ifd;	1060	vec.sfv_fd = ifd;
936	vec.sfv_flag = 0;	1061	vec.sfv_flag = 0;
937	vec.sfv_off = offset;	1062	vec.sfv_off = offset;
938	vec.sfv_len = count;	1063	vec.sfv_len = count;
939		1064
940	res = sendfilev (ofd, &vec, 1, &sbytes);	1065	res = sendfilev (ofd, &vec, 1, &sbytes);
941		1066
942	if (res < 0 && sbytes)	1067	if (res < 0 && sbytes)
943	res = sbytes;	1068	res = sbytes;
944	}
945		1069
946	# endif	1070	# endif
		1071
		1072	#elif defined (_WIN32) && 0
		1073	/* does not work, just for documentation of what would need to be done */
		1074	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1075	/* libeio guarantees that the file offset does not change, and windows */
		1076	/* has no way to get an independent handle to the same file description */
		1077	HANDLE h = TO_SOCKET (ifd);
		1078	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1079	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1080
947	#else	1081	#else
948	res = -1;	1082	res = -1;
949	errno = ENOSYS;	1083	errno = ENOSYS;
950	#endif	1084	#endif
951		1085
		1086	/* we assume sendfile can copy at least 128mb in one go */
		1087	if (res <= 128 * 1024 * 1024)
		1088	{
		1089	if (res > 0)
		1090	written += res;
		1091
		1092	if (written)
		1093	return written;
		1094
		1095	break;
		1096	}
		1097	else
		1098	{
		1099	/* if we requested more, then probably the kernel was lazy */
		1100	written += res;
		1101	offset += res;
		1102	count -= res;
		1103
		1104	if (!count)
		1105	return written;
		1106	}
		1107	}
		1108
952	if (res < 0	1109	if (res < 0
953	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1110	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1111	/* BSDs */
		1112	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1113	|| errno == ENOTSUP
		1114	#endif
		1115	#ifdef EOPNOTSUPP /* windows */
		1116	|| errno == EOPNOTSUPP /* BSDs */
		1117	#endif
954	#if __solaris	1118	#if __solaris
955	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1119	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
956	#endif	1120	#endif
957	)	1121	)
958	)	1122	)
…		…
962		1126
963	res = 0;	1127	res = 0;
964		1128
965	while (count)	1129	while (count)
966	{	1130	{
967	ssize_t cnt;	1131	eio_ssize_t cnt;
968		1132
969	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1133	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
970		1134
971	if (cnt <= 0)	1135	if (cnt <= 0)
972	{	1136	{
…		…
984		1148
985	offset += cnt;	1149	offset += cnt;
986	res += cnt;	1150	res += cnt;
987	count -= cnt;	1151	count -= cnt;
988	}	1152	}
		1153
		1154	FUBd;
989	}	1155	}
990		1156
991	return res;	1157	return res;
992	}	1158	}
993		1159
		1160	#ifdef PAGESIZE
		1161	# define eio_pagesize() PAGESIZE
		1162	#else
		1163	static intptr_t
		1164	eio_pagesize (void)
		1165	{
		1166	static intptr_t page;
		1167
		1168	if (!page)
		1169	page = sysconf (_SC_PAGESIZE);
		1170
		1171	return page;
		1172	}
		1173	#endif
		1174
		1175	static void
		1176	eio_page_align (void **addr, size_t *length)
		1177	{
		1178	intptr_t mask = eio_pagesize () - 1;
		1179
		1180	/* round down addr */
		1181	intptr_t adj = mask & (intptr_t)*addr;
		1182
		1183	*addr = (void *)((intptr_t)*addr - adj);
		1184	*length += adj;
		1185
		1186	/* round up length */
		1187	*length = (*length + mask) & ~mask;
		1188	}
		1189
		1190	#if !_POSIX_MEMLOCK
		1191	# define eio__mlockall(a) EIO_ENOSYS ()
		1192	#else
		1193
994	static int	1194	static int
		1195	eio__mlockall (int flags)
		1196	{
		1197	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1198	extern int mallopt (int, int);
		1199	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1200	#endif
		1201
		1202	if (EIO_MCL_CURRENT != MCL_CURRENT
		1203	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1204	{
		1205	flags = 0
		1206	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1207	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1208	}
		1209
		1210	return mlockall (flags);
		1211	}
		1212	#endif
		1213
		1214	#if !_POSIX_MEMLOCK_RANGE
		1215	# define eio__mlock(a,b) EIO_ENOSYS ()
		1216	#else
		1217
		1218	static int
		1219	eio__mlock (void *addr, size_t length)
		1220	{
		1221	eio_page_align (&addr, &length);
		1222
		1223	return mlock (addr, length);
		1224	}
		1225
		1226	#endif
		1227
		1228	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1229	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1230	#else
		1231
		1232	static int
		1233	eio__msync (void *mem, size_t len, int flags)
		1234	{
		1235	eio_page_align (&mem, &len);
		1236
		1237	if (EIO_MS_ASYNC != MS_SYNC
		1238	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1239	|| EIO_MS_SYNC != MS_SYNC)
		1240	{
		1241	flags = 0
		1242	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1243	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1244	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1245	}
		1246
		1247	return msync (mem, len, flags);
		1248	}
		1249
		1250	#endif
		1251
		1252	static int
		1253	eio__mtouch (eio_req *req)
		1254	{
		1255	void *mem = req->ptr2;
		1256	size_t len = req->size;
		1257	int flags = req->int1;
		1258
		1259	eio_page_align (&mem, &len);
		1260
		1261	{
		1262	intptr_t addr = (intptr_t)mem;
		1263	intptr_t end = addr + len;
		1264	intptr_t page = eio_pagesize ();
		1265
		1266	if (addr < end)
		1267	if (flags & EIO_MT_MODIFY) /* modify */
		1268	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1269	else
		1270	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1271	}
		1272
		1273	return 0;
		1274	}
		1275
		1276	/*****************************************************************************/
		1277	/* requests implemented outside eio_execute, because they are so large */
		1278
		1279	static void
		1280	eio__realpath (eio_req *req, etp_worker *self)
		1281	{
		1282	char *rel = req->ptr1;
		1283	char *res;
		1284	char *tmp1, *tmp2;
		1285	#if SYMLOOP_MAX > 32
		1286	int symlinks = SYMLOOP_MAX;
		1287	#else
		1288	int symlinks = 32;
		1289	#endif
		1290
		1291	req->result = -1;
		1292
		1293	errno = EINVAL;
		1294	if (!rel)
		1295	return;
		1296
		1297	errno = ENOENT;
		1298	if (!*rel)
		1299	return;
		1300
		1301	if (!req->ptr2)
		1302	{
		1303	X_LOCK (wrklock);
		1304	req->flags |= EIO_FLAG_PTR2_FREE;
		1305	X_UNLOCK (wrklock);
		1306	req->ptr2 = malloc (PATH_MAX * 3);
		1307
		1308	errno = ENOMEM;
		1309	if (!req->ptr2)
		1310	return;
		1311	}
		1312
		1313	res = req->ptr2;
		1314	tmp1 = res + PATH_MAX;
		1315	tmp2 = tmp1 + PATH_MAX;
		1316
		1317	#if 0 /* disabled, the musl way to do things is just too racy */
		1318	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1319	/* on linux we may be able to ask the kernel */
		1320	{
		1321	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1322
		1323	if (fd >= 0)
		1324	{
		1325	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1326	req->result = readlink (tmp1, res, PATH_MAX);
		1327	close (fd);
		1328
		1329	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1330
		1331	if (req->result > 0)
		1332	goto done;
		1333	}
		1334	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1335	return;
		1336	}
		1337	#endif
		1338	#endif
		1339
		1340	if (*rel != '/')
		1341	{
		1342	if (!getcwd (res, PATH_MAX))
		1343	return;
		1344
		1345	if (res [1]) /* only use if not / */
		1346	res += strlen (res);
		1347	}
		1348
		1349	while (*rel)
		1350	{
		1351	eio_ssize_t len, linklen;
		1352	char *beg = rel;
		1353
		1354	while (*rel && *rel != '/')
		1355	++rel;
		1356
		1357	len = rel - beg;
		1358
		1359	if (!len) /* skip slashes */
		1360	{
		1361	++rel;
		1362	continue;
		1363	}
		1364
		1365	if (beg [0] == '.')
		1366	{
		1367	if (len == 1)
		1368	continue; /* . - nop */
		1369
		1370	if (beg [1] == '.' && len == 2)
		1371	{
		1372	/* .. - back up one component, if possible */
		1373
		1374	while (res != req->ptr2)
		1375	if (*--res == '/')
		1376	break;
		1377
		1378	continue;
		1379	}
		1380	}
		1381
		1382	errno = ENAMETOOLONG;
		1383	if (res + 1 + len + 1 >= tmp1)
		1384	return;
		1385
		1386	/* copy one component */
		1387	*res = '/';
		1388	memcpy (res + 1, beg, len);
		1389
		1390	/* zero-terminate, for readlink */
		1391	res [len + 1] = 0;
		1392
		1393	/* now check if it's a symlink */
		1394	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1395
		1396	if (linklen < 0)
		1397	{
		1398	if (errno != EINVAL)
		1399	return;
		1400
		1401	/* it's a normal directory. hopefully */
		1402	res += len + 1;
		1403	}
		1404	else
		1405	{
		1406	/* yay, it was a symlink - build new path in tmp2 */
		1407	int rellen = strlen (rel);
		1408
		1409	errno = ENAMETOOLONG;
		1410	if (linklen + 1 + rellen >= PATH_MAX)
		1411	return;
		1412
		1413	errno = ELOOP;
		1414	if (!--symlinks)
		1415	return;
		1416
		1417	if (*tmp1 == '/')
		1418	res = req->ptr2; /* symlink resolves to an absolute path */
		1419
		1420	/* we need to be careful, as rel might point into tmp2 already */
		1421	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1422	tmp2 [linklen] = '/';
		1423	memcpy (tmp2, tmp1, linklen);
		1424
		1425	rel = tmp2;
		1426	}
		1427	}
		1428
		1429	/* special case for the lone root path */
		1430	if (res == req->ptr2)
		1431	*res++ = '/';
		1432
		1433	req->result = res - (char *)req->ptr2;
		1434
		1435	done:
		1436	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1437	}
		1438
		1439	static signed char
995	eio_dent_cmp (const void *a_, const void *b_)	1440	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
996	{	1441	{
997	const eio_dirent *a = (const eio_dirent *)a_;	1442	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
998	const eio_dirent *b = (const eio_dirent *)b_;	1443	: a->inode < b->inode ? -1
		1444	: a->inode > b->inode ? 1
		1445	: 0;
		1446	}
999		1447
1000	return (int)b->score - (int)a->score ? (int)b->score - (int)a->score	1448	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1001	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */	1449
		1450	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1451	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1452
		1453	static void
		1454	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1455	{
		1456	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
		1457	unsigned char *bit = bits;
		1458
		1459	assert (CHAR_BIT == 8);
		1460	assert (sizeof (eio_dirent) * 8 < 256);
		1461	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1462	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1463
		1464	if (size <= EIO_SORT_FAST)
		1465	return;
		1466
		1467	/* first prepare an array of bits to test in our radix sort */
		1468	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
		1469	/* inode_bits must contain all inodes ORed together */
		1470	/* which is used to skip bits that are 0 everywhere, which is very common */
		1471	{
		1472	eio_ino_t endianness;
		1473	int i, j;
		1474
		1475	/* we store the byte offset of byte n into byte n of "endianness" */
		1476	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1477	((unsigned char *)&endianness)[i] = i;
		1478
		1479	*bit++ = 0;
		1480
		1481	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1482	{
		1483	/* shifting off the byte offsets out of "endianness" */
		1484	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1485	endianness >>= 8;
		1486
		1487	for (j = 0; j < 8; ++j)
		1488	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
		1489	*bit++ = offs + j;
		1490	}
		1491
		1492	for (j = 0; j < 8; ++j)
		1493	if (score_bits & (1 << j))
		1494	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1495	}
		1496
		1497	/* now actually do the sorting (a variant of MSD radix sort) */
		1498	{
		1499	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
		1500	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
		1501	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
		1502	int stk_idx = 0;
		1503
		1504	base_stk [stk_idx] = dents;
		1505	end_stk [stk_idx] = dents + size;
		1506	bit_stk [stk_idx] = bit - 1;
		1507
		1508	do
		1509	{
		1510	base = base_stk [stk_idx];
		1511	end = end_stk [stk_idx];
		1512	bit = bit_stk [stk_idx];
		1513
		1514	for (;;)
		1515	{
		1516	unsigned char O = *bit >> 3;
		1517	unsigned char M = 1 << (*bit & 7);
		1518
		1519	eio_dirent *a = base;
		1520	eio_dirent *b = end;
		1521
		1522	if (b - a < EIO_SORT_CUTOFF)
		1523	break;
		1524
		1525	/* now bit-partition the array on the bit */
		1526	/* this ugly asymmetric loop seems to perform much better than typical */
		1527	/* partition algos found in the literature */
		1528	do
		1529	if (!(((unsigned char *)a)[O] & M))
		1530	++a;
		1531	else if (!(((unsigned char *)--b)[O] & M))
		1532	{
		1533	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1534	++a;
		1535	}
		1536	while (b > a);
		1537
		1538	/* next bit, or stop, if no bits left in this path */
		1539	if (!*--bit)
		1540	break;
		1541
		1542	base_stk [stk_idx] = a;
		1543	end_stk [stk_idx] = end;
		1544	bit_stk [stk_idx] = bit;
		1545	++stk_idx;
		1546
		1547	end = a;
		1548	}
		1549	}
		1550	while (stk_idx--);
		1551	}
		1552	}
		1553
		1554	static void
		1555	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1556	{
		1557	/* first move the smallest element to the front, to act as a sentinel */
		1558	{
		1559	int i;
		1560	eio_dirent *min = dents;
		1561
		1562	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1563	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1564	if (EIO_DENT_CMP (dents [i], <, *min))
		1565	min = &dents [i];
		1566
		1567	/* swap elements 0 and j (minimum) */
		1568	{
		1569	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1570	}
		1571	}
		1572
		1573	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1574	{
		1575	eio_dirent *i, *j;
		1576
		1577	for (i = dents + 1; i < dents + size; ++i)
		1578	{
		1579	eio_dirent value = *i;
		1580
		1581	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1582	j [1] = j [0];
		1583
		1584	j [1] = value;
		1585	}
		1586	}
		1587	}
		1588
		1589	static void
		1590	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1591	{
		1592	if (size <= 1)
		1593	return; /* our insertion sort relies on size > 0 */
		1594
		1595	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1596	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1597	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1598
		1599	/* use an insertion sort at the end, or for small arrays, */
		1600	/* as insertion sort is more efficient for small partitions */
		1601	eio_dent_insertion_sort (dents, size);
1002	}	1602	}
1003		1603
1004	/* read a full directory */	1604	/* read a full directory */
1005	static void	1605	static void
1006	eio__scandir (eio_req *req, etp_worker *self)	1606	eio__scandir (eio_req *req, etp_worker *self)
1007	{	1607	{
1008	DIR *dirp;
1009	EIO_STRUCT_DIRENT *entp;
1010	unsigned char *name, *names;	1608	char *name, *names;
1011	int namesalloc = 4096;	1609	int namesalloc = 4096 - sizeof (void *) * 4;
1012	int namesoffs = 0;	1610	int namesoffs = 0;
1013	int flags = req->int1;	1611	int flags = req->int1;
1014	eio_dirent *dents = 0;	1612	eio_dirent *dents = 0;
1015	int dentalloc = 128;	1613	int dentalloc = 128;
1016	int dentoffs = 0;	1614	int dentoffs = 0;
		1615	eio_ino_t inode_bits = 0;
		1616	#ifdef _WIN32
		1617	HANDLE dirp;
		1618	WIN32_FIND_DATA entp;
		1619	#else
		1620	DIR *dirp;
		1621	EIO_STRUCT_DIRENT *entp;
		1622	#endif
1017		1623
1018	req->result = -1;	1624	req->result = -1;
1019		1625
1020	if (!(flags & EIO_READDIR_DENTS))	1626	if (!(flags & EIO_READDIR_DENTS))
1021	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1627	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1022		1628
1023	X_LOCK (wrklock);	1629	#ifdef _WIN32
1024	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1630	{
		1631	int len = strlen ((const char *)req->ptr1);
		1632	char *path = malloc (MAX_PATH);
		1633	const char *fmt;
		1634
		1635	if (!len)
		1636	fmt = "./*";
		1637	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1638	fmt = "%s*";
		1639	else
		1640	fmt = "%s/*";
		1641
		1642	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1643	dirp = FindFirstFile (path, &entp);
		1644	free (path);
		1645
		1646	if (dirp == INVALID_HANDLE_VALUE)
		1647	{
		1648	dirp = 0;
		1649
		1650	switch (GetLastError ())
		1651	{
		1652	case ERROR_FILE_NOT_FOUND:
		1653	req->result = 0;
		1654	break;
		1655
		1656	case ERROR_INVALID_NAME:
		1657	case ERROR_PATH_NOT_FOUND:
		1658	case ERROR_NO_MORE_FILES:
		1659	errno = ENOENT;
		1660	break;
		1661
		1662	case ERROR_NOT_ENOUGH_MEMORY:
		1663	errno = ENOMEM;
		1664	break;
		1665
		1666	default:
		1667	errno = EINVAL;
		1668	break;
		1669	}
		1670	}
		1671	}
		1672	#else
1025	self->dirp = dirp = opendir (req->ptr1);	1673	dirp = opendir (req->ptr1);
		1674	#endif
		1675
		1676	if (req->flags & EIO_FLAG_PTR1_FREE)
		1677	free (req->ptr1);
		1678
1026	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1679	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1027	req->ptr1 = names = malloc (namesalloc);
1028	req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1680	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1029	X_UNLOCK (wrklock);	1681	req->ptr2 = names = malloc (namesalloc);
1030		1682
1031	if (dirp && names && (!flags || dents))	1683	if (dirp && names && (!flags || dents))
1032	for (;;)	1684	for (;;)
1033	{	1685	{
		1686	int done;
		1687
		1688	#ifdef _WIN32
		1689	done = !dirp;
		1690	#else
1034	errno = 0;	1691	errno = 0;
1035	entp = readdir (dirp);	1692	entp = readdir (dirp);
		1693	done = !entp;
		1694	#endif
1036		1695
1037	if (!entp)	1696	if (done)
1038	{	1697	{
		1698	#ifndef _WIN32
		1699	int old_errno = errno;
		1700	closedir (dirp);
		1701	errno = old_errno;
		1702
1039	if (errno)	1703	if (errno)
1040	break;	1704	break;
		1705	#endif
1041		1706
1042	/* sort etc. */	1707	/* sort etc. */
1043	req->int1 = flags;	1708	req->int1 = flags;
1044	req->result = dentoffs;	1709	req->result = dentoffs;
1045		1710
1046	if (dents)
1047	{
1048	eio_dirent *ent = dents + dentoffs;
1049
1050	while (ent > dents)
1051	(--ent)->name = names + (size_t)ent->name;
1052	}
1053
1054	if (flags & EIO_READDIR_STAT_ORDER	1711	if (flags & EIO_READDIR_STAT_ORDER)
1055	|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))	1712	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1056	{
1057	/* pray your qsort doesn't use quicksort */
1058	qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1059	}
1060	else if (flags & EIO_READDIR_DIRS_FIRST)	1713	else if (flags & EIO_READDIR_DIRS_FIRST)
		1714	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1715	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1716	else
1061	{	1717	{
1062	/* in this case, all is known, and we just put dirs first and sort them */	1718	/* in this case, all is known, and we just put dirs first and sort them */
1063	eio_dirent *ent = dents + dentoffs;	1719	eio_dirent *oth = dents + dentoffs;
1064	eio_dirent *dir = dents;	1720	eio_dirent *dir = dents;
1065		1721
		1722	/* now partition dirs to the front, and non-dirs to the back */
		1723	/* by walking from both sides and swapping if necessary */
1066	while (ent > dir)	1724	while (oth > dir)
1067	{	1725	{
1068	if (dir->type == DT_DIR)	1726	if (dir->type == EIO_DT_DIR)
1069	++dir;	1727	++dir;
1070	else
1071	{
1072	--ent;
1073
1074	if (ent->type == DT_DIR)	1728	else if ((--oth)->type == EIO_DT_DIR)
1075	{	1729	{
1076	eio_dirent tmp = *dir;	1730	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1077	*dir = *ent;
1078	*ent = tmp;
1079		1731
1080	++dir;	1732	++dir;
1081	}	1733	}
1082	}	1734	}
		1735
		1736	/* now sort the dirs only (dirs all have the same score) */
		1737	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1083	}	1738	}
1084
1085	/* now sort the dirs only */
1086	qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1087	}
1088
1089	/* only provide the names array unless DENTS is specified */
1090	if (!(flags & EIO_READDIR_DENTS))
1091	{
1092	X_LOCK (wrklock);
1093	assert (!dents);
1094	req->ptr1 = 0;
1095	req->ptr2 = names;
1096	X_UNLOCK (wrklock);
1097	}
1098		1739
1099	break;	1740	break;
1100	}	1741	}
1101		1742
1102	/* now add the entry to our list(s) */	1743	/* now add the entry to our list(s) */
1103	name = entp->d_name;	1744	name = D_NAME (entp);
1104		1745
1105	/* skip . and .. entries */	1746	/* skip . and .. entries */
1106	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1747	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1107	{	1748	{
1108	int len = D_NAMLEN (entp) + 1;	1749	int len = D_NAMLEN (entp) + 1;
1109		1750
1110	while (expect_false (namesoffs + len > namesalloc))	1751	while (ecb_expect_false (namesoffs + len > namesalloc))
1111	{	1752	{
1112	namesalloc *= 2;	1753	namesalloc *= 2;
1113	X_LOCK (wrklock);
1114	req->ptr1 = names = realloc (names, namesalloc);	1754	req->ptr2 = names = realloc (names, namesalloc);
1115	X_UNLOCK (wrklock);
1116		1755
1117	if (!names)	1756	if (!names)
1118	break;	1757	break;
1119	}	1758	}
1120		1759
…		…
1122		1761
1123	if (dents)	1762	if (dents)
1124	{	1763	{
1125	struct eio_dirent *ent;	1764	struct eio_dirent *ent;
1126		1765
1127	if (expect_false (dentoffs == dentalloc))	1766	if (ecb_expect_false (dentoffs == dentalloc))
1128	{	1767	{
1129	dentalloc *= 2;	1768	dentalloc *= 2;
1130	X_LOCK (wrklock);
1131	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1769	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1132	X_UNLOCK (wrklock);
1133		1770
1134	if (!dents)	1771	if (!dents)
1135	break;	1772	break;
1136	}	1773	}
1137		1774
1138	ent = dents + dentoffs;	1775	ent = dents + dentoffs;
1139		1776
1140	ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */	1777	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1141	ent->namelen = len - 1;	1778	ent->namelen = len - 1;
1142	ent->inode = D_INO (entp);	1779	ent->inode = D_INO (entp);
		1780
		1781	inode_bits |= ent->inode;
1143		1782
1144	switch (D_TYPE (entp))	1783	switch (D_TYPE (entp))
1145	{	1784	{
1146	default:	1785	default:
1147	ent->type = EIO_DT_UNKNOWN;	1786	ent->type = EIO_DT_UNKNOWN;
…		…
1190	#ifdef DT_WHT	1829	#ifdef DT_WHT
1191	case DT_WHT: ent->type = EIO_DT_WHT; break;	1830	case DT_WHT: ent->type = EIO_DT_WHT; break;
1192	#endif	1831	#endif
1193	}	1832	}
1194		1833
1195	ent->score = 0;	1834	ent->score = 7;
1196		1835
1197	if (flags & EIO_READDIR_DIRS_FIRST)	1836	if (flags & EIO_READDIR_DIRS_FIRST)
1198	{	1837	{
1199	if (ent->type == EIO_DT_UNKNOWN)	1838	if (ent->type == EIO_DT_UNKNOWN)
1200	{	1839	{
1201	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */	1840	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1202	ent->score = 98;	1841	ent->score = 1;
1203	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */	1842	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1204	ent->score = len <= 2 ? len + 60 : len <= 4 ? 50 : len <= 7 ? 40 : 10; /* shorter == more likely dir, but avoid too many classes */	1843	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1205	}	1844	}
1206	else if (ent->type == EIO_DT_DIR)	1845	else if (ent->type == EIO_DT_DIR)
1207	ent->score = 100;	1846	ent->score = 0;
1208	}	1847	}
1209	}	1848	}
1210		1849
1211	namesoffs += len;	1850	namesoffs += len;
1212	++dentoffs;	1851	++dentoffs;
1213	}	1852	}
		1853
		1854	if (EIO_CANCELLED (req))
		1855	{
		1856	errno = ECANCELED;
		1857	break;
		1858	}
		1859
		1860	#ifdef _WIN32
		1861	if (!FindNextFile (dirp, &entp))
		1862	{
		1863	FindClose (dirp);
		1864	dirp = 0;
		1865	}
		1866	#endif
1214	}	1867	}
1215	else
1216	req->result = -1;
1217	}
1218
1219	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1220	# undef msync
1221	# define msync(a,b,c) ((errno = ENOSYS), -1)
1222	#endif
1223
1224	int
1225	eio__mtouch (void *mem, size_t len, int flags)
1226	{
1227	intptr_t addr = (intptr_t)mem;
1228	intptr_t end = addr + len;
1229	#ifdef PAGESIZE
1230	const intptr_t page = PAGESIZE;
1231	#else
1232	static intptr_t page;
1233
1234	if (!page)
1235	page = sysconf (_SC_PAGESIZE);
1236	#endif
1237
1238	addr &= ~(page - 1); /* assume page size is always a power of two */
1239
1240	if (addr < end)
1241	if (flags) /* modify */
1242	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1243	else
1244	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1245
1246	return 0;
1247	}	1868	}
1248		1869
1249	/*****************************************************************************/	1870	/*****************************************************************************/
1250		1871
1251	#define ALLOC(len) \	1872	#define ALLOC(len) \
…		…
1267	{	1888	{
1268	ETP_REQ *req;	1889	ETP_REQ *req;
1269	struct timespec ts;	1890	struct timespec ts;
1270	etp_worker *self = (etp_worker *)thr_arg;	1891	etp_worker *self = (etp_worker *)thr_arg;
1271		1892
1272	/* try to distribute timeouts somewhat randomly */	1893	/* try to distribute timeouts somewhat evenly */
1273	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1894	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1274		1895
1275	for (;;)	1896	for (;;)
1276	{	1897	{
		1898	ts.tv_sec = 0;
		1899
1277	X_LOCK (reqlock);	1900	X_LOCK (reqlock);
1278		1901
1279	for (;;)	1902	for (;;)
1280	{	1903	{
1281	self->req = req = reqq_shift (&req_queue);	1904	self->req = req = reqq_shift (&req_queue);
1282		1905
1283	if (req)	1906	if (req)
1284	break;	1907	break;
1285		1908
		1909	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1910	{
		1911	X_UNLOCK (reqlock);
		1912	X_LOCK (wrklock);
		1913	--started;
		1914	X_UNLOCK (wrklock);
		1915	goto quit;
		1916	}
		1917
1286	++idle;	1918	++idle;
1287		1919
1288	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1920	if (idle <= max_idle)
1289	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1921	/* we are allowed to idle, so do so without any timeout */
		1922	X_COND_WAIT (reqwait, reqlock);
		1923	else
1290	{	1924	{
1291	if (idle > max_idle)	1925	/* initialise timeout once */
1292	{	1926	if (!ts.tv_sec)
1293	--idle;	1927	ts.tv_sec = time (0) + idle_timeout;
1294	X_UNLOCK (reqlock);
1295	X_LOCK (wrklock);
1296	--started;
1297	X_UNLOCK (wrklock);
1298	goto quit;
1299	}
1300		1928
1301	/* we are allowed to idle, so do so without any timeout */
1302	X_COND_WAIT (reqwait, reqlock);	1929	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1930	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1303	}	1931	}
1304		1932
1305	--idle;	1933	--idle;
1306	}	1934	}
1307		1935
…		…
1310	X_UNLOCK (reqlock);	1938	X_UNLOCK (reqlock);
1311		1939
1312	if (req->type < 0)	1940	if (req->type < 0)
1313	goto quit;	1941	goto quit;
1314		1942
1315	if (!EIO_CANCELLED (req))
1316	ETP_EXECUTE (self, req);	1943	ETP_EXECUTE (self, req);
1317		1944
1318	X_LOCK (reslock);	1945	X_LOCK (reslock);
1319		1946
1320	++npending;	1947	++npending;
1321		1948
…		…
1336	return 0;	1963	return 0;
1337	}	1964	}
1338		1965
1339	/*****************************************************************************/	1966	/*****************************************************************************/
1340		1967
		1968	int ecb_cold
1341	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1969	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1342	{	1970	{
		1971	#if !HAVE_PREADWRITE
		1972	X_MUTEX_CREATE (preadwritelock);
		1973	#endif
		1974
1343	return etp_init (want_poll, done_poll);	1975	return etp_init (want_poll, done_poll);
1344	}	1976	}
1345		1977
		1978	ecb_inline void
1346	static void eio_api_destroy (eio_req *req)	1979	eio_api_destroy (eio_req *req)
1347	{	1980	{
1348	free (req);	1981	free (req);
1349	}	1982	}
1350		1983
1351	#define REQ(rtype) \	1984	#define REQ(rtype) \
…		…
1370	{ \	2003	{ \
1371	eio_api_destroy (req); \	2004	eio_api_destroy (req); \
1372	return 0; \	2005	return 0; \
1373	}	2006	}
1374		2007
		2008	static void
1375	static void eio_execute (etp_worker *self, eio_req *req)	2009	eio_execute (etp_worker *self, eio_req *req)
1376	{	2010	{
1377	errno = 0;	2011	if (ecb_expect_false (EIO_CANCELLED (req)))
		2012	{
		2013	req->result = -1;
		2014	req->errorno = ECANCELED;
		2015	return;
		2016	}
1378		2017
1379	switch (req->type)	2018	switch (req->type)
1380	{	2019	{
1381	case EIO_READ: ALLOC (req->size);	2020	case EIO_READ: ALLOC (req->size);
1382	req->result = req->offs >= 0	2021	req->result = req->offs >= 0
…		…
1385	case EIO_WRITE: req->result = req->offs >= 0	2024	case EIO_WRITE: req->result = req->offs >= 0
1386	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2025	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1387	: write (req->int1, req->ptr2, req->size); break;	2026	: write (req->int1, req->ptr2, req->size); break;
1388		2027
1389	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2028	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1390	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2029	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1391		2030
1392	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2031	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1393	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2032	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1394	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2033	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1395	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2034	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1396	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2035	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1397	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2036	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2037
		2038	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2039	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2040	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2041	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1398		2042
1399	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	2043	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1400	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	2044	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1401	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	2045	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1402	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	2046	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
…		…
1410	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	2054	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1411	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	2055	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1412	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2056	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1413	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2057	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1414	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2058	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1415	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;	2059	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1416		2060
		2061	case EIO_REALPATH: eio__realpath (req, self); break;
		2062
1417	case EIO_READLINK: ALLOC (NAME_MAX);	2063	case EIO_READLINK: ALLOC (PATH_MAX);
1418	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	2064	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1419		2065
1420	case EIO_SYNC: req->result = 0; sync (); break;	2066	case EIO_SYNC: req->result = 0; sync (); break;
1421	case EIO_FSYNC: req->result = fsync (req->int1); break;	2067	case EIO_FSYNC: req->result = fsync (req->int1); break;
1422	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2068	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1423	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;	2069	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2070	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1424	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	2071	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2072	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1425	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	2073	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2074	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1426		2075
1427	case EIO_READDIR: eio__scandir (req, self); break;	2076	case EIO_READDIR: eio__scandir (req, self); break;
1428		2077
1429	case EIO_BUSY:	2078	case EIO_BUSY:
1430	#ifdef _WIN32	2079	#ifdef _WIN32
1431	Sleep (req->nv1 * 1000.);	2080	Sleep (req->nv1 * 1e3);
1432	#else	2081	#else
1433	{	2082	{
1434	struct timeval tv;	2083	struct timeval tv;
1435		2084
1436	tv.tv_sec = req->nv1;	2085	tv.tv_sec = req->nv1;
1437	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	2086	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1438		2087
1439	req->result = select (0, 0, 0, 0, &tv);	2088	req->result = select (0, 0, 0, 0, &tv);
1440	}	2089	}
1441	#endif	2090	#endif
1442	break;	2091	break;
…		…
1457	times = tv;	2106	times = tv;
1458	}	2107	}
1459	else	2108	else
1460	times = 0;	2109	times = 0;
1461		2110
1462
1463	req->result = req->type == EIO_FUTIME	2111	req->result = req->type == EIO_FUTIME
1464	? futimes (req->int1, times)	2112	? futimes (req->int1, times)
1465	: utimes (req->ptr1, times);	2113	: utimes (req->ptr1, times);
1466	}	2114	}
1467	break;	2115	break;
…		…
1472	case EIO_NOP:	2120	case EIO_NOP:
1473	req->result = 0;	2121	req->result = 0;
1474	break;	2122	break;
1475		2123
1476	case EIO_CUSTOM:	2124	case EIO_CUSTOM:
1477	((void (*)(eio_req *))req->feed) (req);	2125	req->feed (req);
1478	break;	2126	break;
1479		2127
1480	default:	2128	default:
		2129	errno = ENOSYS;
1481	req->result = -1;	2130	req->result = -1;
1482	break;	2131	break;
1483	}	2132	}
1484		2133
1485	req->errorno = errno;	2134	req->errorno = errno;
…		…
1515	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2164	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1516	{	2165	{
1517	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2166	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1518	}	2167	}
1519		2168
		2169	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2170	{
		2171	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2172	}
		2173
		2174	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2175	{
		2176	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2177	}
		2178
1520	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2179	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1521	{	2180	{
1522	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2181	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1523	}	2182	}
1524		2183
		2184	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2185	{
		2186	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2187	}
		2188
1525	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2189	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1526	{	2190	{
1527	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2191	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1528	}	2192	}
1529		2193
…		…
1550	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2214	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1551	{	2215	{
1552	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2216	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1553	}	2217	}
1554		2218
		2219	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2220	{
		2221	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2222	}
		2223
1555	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2224	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1556	{	2225	{
1557	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2226	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1558	}	2227	}
1559		2228
…		…
1565	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2234	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1566	{	2235	{
1567	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2236	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1568	}	2237	}
1569		2238
1570	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2239	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1571	{	2240	{
1572	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2241	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1573	}	2242	}
1574		2243
1575	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2244	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1595	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2264	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1596	{	2265	{
1597	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2266	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1598	}	2267	}
1599		2268
1600	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2269	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1601	{	2270	{
1602	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2271	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1603	}	2272	}
1604		2273
1605	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2274	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1621	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2290	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1622	{	2291	{
1623	return eio__1path (EIO_READLINK, path, pri, cb, data);	2292	return eio__1path (EIO_READLINK, path, pri, cb, data);
1624	}	2293	}
1625		2294
		2295	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2296	{
		2297	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2298	}
		2299
1626	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2300	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1627	{	2301	{
1628	return eio__1path (EIO_STAT, path, pri, cb, data);	2302	return eio__1path (EIO_STAT, path, pri, cb, data);
1629	}	2303	}
1630		2304
1631	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2305	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1632	{	2306	{
1633	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2307	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1634	}	2308	}
1635		2309
		2310	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2311	{
		2312	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2313	}
		2314
1636	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2315	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1637	{	2316	{
1638	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2317	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1639	}	2318	}
1640		2319
…		…
1648	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2327	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1649	}	2328	}
1650		2329
1651	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2330	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1652	{	2331	{
1653	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2332	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1654	}	2333	}
1655		2334
1656	static eio_req *	2335	static eio_req *
1657	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2336	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1658	{	2337	{
…		…
1682	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2361	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1683	{	2362	{
1684	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2363	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1685	}	2364	}
1686		2365
1687	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2366	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1688	{	2367	{
1689	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2368	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1690	}	2369	}
1691		2370
1692	#endif	2371	#endif
1693		2372
1694	eio_req *eio_grp (eio_cb cb, void *data)	2373	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1703	#undef SEND	2382	#undef SEND
1704		2383
1705	/*****************************************************************************/	2384	/*****************************************************************************/
1706	/* grp functions */	2385	/* grp functions */
1707		2386
		2387	void
1708	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2388	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1709	{	2389	{
1710	grp->int2 = limit;	2390	grp->int2 = limit;
1711	grp->feed = feed;	2391	grp->feed = feed;
1712		2392
1713	grp_try_feed (grp);	2393	grp_try_feed (grp);
1714	}	2394	}
1715		2395
		2396	void
1716	void eio_grp_limit (eio_req *grp, int limit)	2397	eio_grp_limit (eio_req *grp, int limit)
1717	{	2398	{
1718	grp->int2 = limit;	2399	grp->int2 = limit;
1719		2400
1720	grp_try_feed (grp);	2401	grp_try_feed (grp);
1721	}	2402	}
1722		2403
		2404	void
1723	void eio_grp_add (eio_req *grp, eio_req *req)	2405	eio_grp_add (eio_req *grp, eio_req *req)
1724	{	2406	{
1725	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2407	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1726		2408
1727	grp->flags |= EIO_FLAG_GROUPADD;	2409	grp->flags |= EIO_FLAG_GROUPADD;
1728		2410
…		…
1739	}	2421	}
1740		2422
1741	/*****************************************************************************/	2423	/*****************************************************************************/
1742	/* misc garbage */	2424	/* misc garbage */
1743		2425
		2426	eio_ssize_t
1744	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2427	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1745	{	2428	{
1746	etp_worker wrk;
1747
1748	wrk.dbuf = 0;
1749
1750	eio__sendfile (ofd, ifd, offset, count, &wrk);	2429	return eio__sendfile (ofd, ifd, offset, count);
1751
1752	if (wrk.dbuf)
1753	free (wrk.dbuf);
1754	}	2430	}
1755		2431

Diff Legend

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