ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.70 by root, Fri Jun 10 06:55:10 2011 UTC vs.
Revision 1.99 by root, Tue Jul 26 11:07:08 2011 UTC

…		…
40	#ifndef _WIN32	40	#ifndef _WIN32
41	# include "config.h"	41	# include "config.h"
42	#endif	42	#endif
43		43
44	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
45		46
46	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
47	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
48	#endif	49	#endif
49	#include "xthread.h"	50	#include "xthread.h"
…		…
53	#include <stdlib.h>	54	#include <stdlib.h>
54	#include <string.h>	55	#include <string.h>
55	#include <errno.h>	56	#include <errno.h>
56	#include <sys/types.h>	57	#include <sys/types.h>
57	#include <sys/stat.h>	58	#include <sys/stat.h>
58	#include <sys/statvfs.h>
59	#include <limits.h>	59	#include <limits.h>
60	#include <fcntl.h>	60	#include <fcntl.h>
61	#include <assert.h>	61	#include <assert.h>
62		62
63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */	63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64	/* intptr_t only comes form stdint.h, says idiot openbsd coder */	64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65	#if HAVE_STDINT_H	65	#if HAVE_STDINT_H
66	# include <stdint.h>	66	# include <stdint.h>
67	#endif	67	#endif
68		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
69	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
70	# 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
71	#endif	84	#endif
72		85
73	#ifndef EIO_DESTROY	86	#ifndef EIO_DESTROY
…		…
76		89
77	#ifndef EIO_FEED	90	#ifndef EIO_FEED
78	# 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)
79	#endif	92	#endif
80		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
81	#ifdef _WIN32	105	#ifdef _WIN32
82		106
83	/*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	/* rename() uses MoveFile, which fails to overwrite */
		135	#define rename(old,neu) eio__rename (old, neu)
		136
		137	static int
		138	eio__rename (const char *old, const char *neu)
		139	{
		140	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		141	return 0;
		142
		143	/* should steal _dosmaperr */
		144	switch (GetLastError ())
		145	{
		146	case ERROR_FILE_NOT_FOUND:
		147	case ERROR_PATH_NOT_FOUND:
		148	case ERROR_INVALID_DRIVE:
		149	case ERROR_NO_MORE_FILES:
		150	case ERROR_BAD_NETPATH:
		151	case ERROR_BAD_NET_NAME:
		152	case ERROR_BAD_PATHNAME:
		153	case ERROR_FILENAME_EXCED_RANGE:
		154	errno = ENOENT;
		155	break;
		156
		157	default:
		158	errno = EACCES;
		159	break;
		160	}
		161
		162	return -1;
		163	}
		164
		165	/* we could even stat and see if it exists */
		166	static int
		167	symlink (const char *old, const char *neu)
		168	{
		169	#if WINVER >= 0x0600
		170	if (CreateSymbolicLink (neu, old, 1))
		171	return 0;
		172
		173	if (CreateSymbolicLink (neu, old, 0))
		174	return 0;
		175	#endif
		176
		177	return EIO_ERRNO (ENOENT, -1);
		178	}
		179
		180	/* POSIX API only */
		181	#define CreateHardLink(neu,old,flags) 0
		182	#define CreateSymbolicLink(neu,old,flags) 0
		183
		184	struct statvfs
		185	{
		186	int dummy;
		187	};
		188
		189	#define DT_DIR EIO_DT_DIR
		190	#define DT_REG EIO_DT_REG
		191	#define D_NAME(entp) entp.cFileName
		192	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
		193
84	#else	194	#else
85		195
86	# include <sys/time.h>	196	#include <sys/time.h>
87	# include <sys/select.h>	197	#include <sys/select.h>
		198	#include <sys/statvfs.h>
88	# include <unistd.h>	199	#include <unistd.h>
		200	#include <signal.h>
		201	#include <dirent.h>
		202
		203	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		204	#include <sys/mman.h>
		205	#endif
		206
		207	#define D_NAME(entp) entp->d_name
		208
		209	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		210	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		211	#define _DIRENT_HAVE_D_TYPE /* sigh */
		212	#define D_INO(de) (de)->d_fileno
		213	#define D_NAMLEN(de) (de)->d_namlen
		214	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		215	#define D_INO(de) (de)->d_ino
		216	#endif
		217
		218	#ifdef _D_EXACT_NAMLEN
		219	#undef D_NAMLEN
		220	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		221	#endif
		222
		223	#ifdef _DIRENT_HAVE_D_TYPE
		224	#define D_TYPE(de) (de)->d_type
		225	#endif
		226
		227	#ifndef EIO_STRUCT_DIRENT
		228	#define EIO_STRUCT_DIRENT struct dirent
		229	#endif
		230
		231	#endif
		232
		233	#if HAVE_UTIMES
89	# include <utime.h>	234	# include <utime.h>
90	# include <signal.h>
91	# include <dirent.h>
92
93	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94	# include <sys/mman.h>
95	#endif
96
97	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99	# define _DIRENT_HAVE_D_TYPE /* sigh */
100	# define D_INO(de) (de)->d_fileno
101	# define D_NAMLEN(de) (de)->d_namlen
102	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103	# define D_INO(de) (de)->d_ino
104	# endif	235	#endif
105		236
106	#ifdef _D_EXACT_NAMLEN	237	#if HAVE_SYS_SYSCALL_H
107	# undef D_NAMLEN	238	# include <sys/syscall.h>
108	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109	#endif
110
111	# ifdef _DIRENT_HAVE_D_TYPE
112	# define D_TYPE(de) (de)->d_type
113	# endif	239	#endif
114		240
115	# ifndef EIO_STRUCT_DIRENT	241	#if HAVE_SYS_PRCTL_H
116	# define EIO_STRUCT_DIRENT struct dirent	242	# include <sys/prctl.h>
117	# endif
118
119	#endif	243	#endif
120		244
121	#if HAVE_SENDFILE	245	#if HAVE_SENDFILE
122	# if __linux	246	# if __linux
123	# include <sys/sendfile.h>	247	# include <sys/sendfile.h>
…		…
138	#endif	262	#endif
139	#ifndef D_INO	263	#ifndef D_INO
140	# define D_INO(de) 0	264	# define D_INO(de) 0
141	#endif	265	#endif
142	#ifndef D_NAMLEN	266	#ifndef D_NAMLEN
143	# define D_NAMLEN(de) strlen ((de)->d_name)	267	# define D_NAMLEN(entp) strlen (D_NAME (entp))
144	#endif	268	#endif
145		269
146	/* used for struct dirent, AIX doesn't provide it */	270	/* used for struct dirent, AIX doesn't provide it */
147	#ifndef NAME_MAX	271	#ifndef NAME_MAX
148	# define NAME_MAX 4096	272	# define NAME_MAX 4096
…		…
155		279
156	/* buffer size for various temporary buffers */	280	/* buffer size for various temporary buffers */
157	#define EIO_BUFSIZE 65536	281	#define EIO_BUFSIZE 65536
158		282
159	#define dBUF \	283	#define dBUF \
160	char *eio_buf; \
161	ETP_WORKER_LOCK (self); \
162	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	284	char *eio_buf = malloc (EIO_BUFSIZE); \
163	ETP_WORKER_UNLOCK (self); \
164	errno = ENOMEM; \	285	errno = ENOMEM; \
165	if (!eio_buf) \	286	if (!eio_buf) \
166	return -1;	287	return -1
		288
		289	#define FUBd \
		290	free (eio_buf)
167		291
168	#define EIO_TICKS ((1000000 + 1023) >> 10)	292	#define EIO_TICKS ((1000000 + 1023) >> 10)
169
170	/*****************************************************************************/
171
172	#if __GNUC__ >= 3
173	# define expect(expr,value) __builtin_expect ((expr),(value))
174	#else
175	# define expect(expr,value) (expr)
176	#endif
177
178	#define expect_false(expr) expect ((expr) != 0, 0)
179	#define expect_true(expr) expect ((expr) != 0, 1)
180
181	/*****************************************************************************/
182		293
183	#define ETP_PRI_MIN EIO_PRI_MIN	294	#define ETP_PRI_MIN EIO_PRI_MIN
184	#define ETP_PRI_MAX EIO_PRI_MAX	295	#define ETP_PRI_MAX EIO_PRI_MAX
185		296
186	struct etp_worker;	297	struct etp_worker;
…		…
190	static int eio_finish (eio_req *req);	301	static int eio_finish (eio_req *req);
191	#define ETP_FINISH(req) eio_finish (req)	302	#define ETP_FINISH(req) eio_finish (req)
192	static void eio_execute (struct etp_worker *self, eio_req *req);	303	static void eio_execute (struct etp_worker *self, eio_req *req);
193	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	304	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194		305
195	#define ETP_WORKER_CLEAR(req) \
196	if (wrk->dbuf) \
197	{ \
198	free (wrk->dbuf); \
199	wrk->dbuf = 0; \
200	} \
201	\
202	if (wrk->dirp) \
203	{ \
204	closedir (wrk->dirp); \
205	wrk->dirp = 0; \
206	}
207
208	#define ETP_WORKER_COMMON \
209	void *dbuf; \
210	DIR *dirp;
211
212	/*****************************************************************************/	306	/*****************************************************************************/
213		307
214	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	308	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215		309
216	/* calculate time difference in ~1/EIO_TICKS of a second */	310	/* calculate time difference in ~1/EIO_TICKS of a second */
		311	ecb_inline int
217	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	312	tvdiff (struct timeval *tv1, struct timeval *tv2)
218	{	313	{
219	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	314	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	315	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
221	}	316	}
222		317
…		…
243	/*	338	/*
244	* make our pread/pwrite emulation safe against themselves, but not against	339	* make our pread/pwrite emulation safe against themselves, but not against
245	* normal read/write by using a mutex. slows down execution a lot,	340	* normal read/write by using a mutex. slows down execution a lot,
246	* but that's your problem, not mine.	341	* but that's your problem, not mine.
247	*/	342	*/
248	static xmutex_t preadwritelock = X_MUTEX_INIT;	343	static xmutex_t preadwritelock;
249	#endif	344	#endif
250		345
251	typedef struct etp_worker	346	typedef struct etp_worker
252	{	347	{
253	/* locked by wrklock */	348	/* locked by wrklock */
…		…
256	xthread_t tid;	351	xthread_t tid;
257		352
258	/* locked by reslock, reqlock or wrklock */	353	/* locked by reslock, reqlock or wrklock */
259	ETP_REQ *req; /* currently processed request */	354	ETP_REQ *req; /* currently processed request */
260		355
		356	#ifdef ETP_WORKER_COMMON
261	ETP_WORKER_COMMON	357	ETP_WORKER_COMMON
		358	#endif
262	} etp_worker;	359	} etp_worker;
263		360
264	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	361	static etp_worker wrk_first; /* NOT etp */
265		362
266	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	363	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	364	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268		365
269	/* worker threads management */	366	/* worker threads management */
270		367
		368	static void ecb_cold
271	static void etp_worker_clear (etp_worker *wrk)	369	etp_worker_clear (etp_worker *wrk)
272	{	370	{
273	ETP_WORKER_CLEAR (wrk);
274	}	371	}
275		372
		373	static void ecb_cold
276	static void etp_worker_free (etp_worker *wrk)	374	etp_worker_free (etp_worker *wrk)
277	{	375	{
278	wrk->next->prev = wrk->prev;	376	wrk->next->prev = wrk->prev;
279	wrk->prev->next = wrk->next;	377	wrk->prev->next = wrk->next;
280		378
281	free (wrk);	379	free (wrk);
282	}	380	}
283		381
284	static unsigned int etp_nreqs (void)	382	static unsigned int
		383	etp_nreqs (void)
285	{	384	{
286	int retval;	385	int retval;
287	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	386	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288	retval = nreqs;	387	retval = nreqs;
289	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	388	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290	return retval;	389	return retval;
291	}	390	}
292		391
293	static unsigned int etp_nready (void)	392	static unsigned int
		393	etp_nready (void)
294	{	394	{
295	unsigned int retval;	395	unsigned int retval;
296		396
297	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	397	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298	retval = nready;	398	retval = nready;
299	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	399	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300		400
301	return retval;	401	return retval;
302	}	402	}
303		403
304	static unsigned int etp_npending (void)	404	static unsigned int
		405	etp_npending (void)
305	{	406	{
306	unsigned int retval;	407	unsigned int retval;
307		408
308	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	409	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309	retval = npending;	410	retval = npending;
310	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	411	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311		412
312	return retval;	413	return retval;
313	}	414	}
314		415
315	static unsigned int etp_nthreads (void)	416	static unsigned int
		417	etp_nthreads (void)
316	{	418	{
317	unsigned int retval;	419	unsigned int retval;
318		420
319	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	421	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320	retval = started;	422	retval = started;
…		…
334	} etp_reqq;	436	} etp_reqq;
335		437
336	static etp_reqq req_queue;	438	static etp_reqq req_queue;
337	static etp_reqq res_queue;	439	static etp_reqq res_queue;
338		440
		441	static void ecb_noinline ecb_cold
		442	reqq_init (etp_reqq *q)
		443	{
		444	int pri;
		445
		446	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		447	q->qs[pri] = q->qe[pri] = 0;
		448
		449	q->size = 0;
		450	}
		451
		452	static int ecb_noinline
339	static int reqq_push (etp_reqq *q, ETP_REQ *req)	453	reqq_push (etp_reqq *q, ETP_REQ *req)
340	{	454	{
341	int pri = req->pri;	455	int pri = req->pri;
342	req->next = 0;	456	req->next = 0;
343		457
344	if (q->qe[pri])	458	if (q->qe[pri])
…		…
350	q->qe[pri] = q->qs[pri] = req;	464	q->qe[pri] = q->qs[pri] = req;
351		465
352	return q->size++;	466	return q->size++;
353	}	467	}
354		468
		469	static ETP_REQ * ecb_noinline
355	static ETP_REQ *reqq_shift (etp_reqq *q)	470	reqq_shift (etp_reqq *q)
356	{	471	{
357	int pri;	472	int pri;
358		473
359	if (!q->size)	474	if (!q->size)
360	return 0;	475	return 0;
…		…
375	}	490	}
376		491
377	abort ();	492	abort ();
378	}	493	}
379		494
380	static void etp_thread_init (void)	495	static int ecb_cold
		496	etp_init (void (*want_poll)(void), void (*done_poll)(void))
381	{	497	{
382	X_MUTEX_CREATE (wrklock);	498	X_MUTEX_CREATE (wrklock);
383	X_MUTEX_CREATE (reslock);	499	X_MUTEX_CREATE (reslock);
384	X_MUTEX_CREATE (reqlock);	500	X_MUTEX_CREATE (reqlock);
385	X_COND_CREATE (reqwait);	501	X_COND_CREATE (reqwait);
386	}
387		502
388	static void etp_atfork_prepare (void)	503	reqq_init (&req_queue);
389	{	504	reqq_init (&res_queue);
390	X_LOCK (wrklock);
391	X_LOCK (reqlock);
392	X_LOCK (reslock);
393	#if !HAVE_PREADWRITE
394	X_LOCK (preadwritelock);
395	#endif
396	}
397		505
398	static void etp_atfork_parent (void)	506	wrk_first.next =
399	{	507	wrk_first.prev = &wrk_first;
400	#if !HAVE_PREADWRITE
401	X_UNLOCK (preadwritelock);
402	#endif
403	X_UNLOCK (reslock);
404	X_UNLOCK (reqlock);
405	X_UNLOCK (wrklock);
406	}
407
408	static void etp_atfork_child (void)
409	{
410	ETP_REQ *prv;
411
412	while ((prv = reqq_shift (&req_queue)))
413	ETP_DESTROY (prv);
414
415	while ((prv = reqq_shift (&res_queue)))
416	ETP_DESTROY (prv);
417
418	while (wrk_first.next != &wrk_first)
419	{
420	etp_worker *wrk = wrk_first.next;
421
422	if (wrk->req)
423	ETP_DESTROY (wrk->req);
424
425	etp_worker_clear (wrk);
426	etp_worker_free (wrk);
427	}
428		508
429	started = 0;	509	started = 0;
430	idle = 0;	510	idle = 0;
431	nreqs = 0;	511	nreqs = 0;
432	nready = 0;	512	nready = 0;
433	npending = 0;	513	npending = 0;
434		514
435	etp_thread_init ();
436	}
437
438	static void
439	etp_once_init (void)
440	{
441	etp_thread_init ();
442	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
443	}
444
445	static int
446	etp_init (void (*want_poll)(void), void (*done_poll)(void))
447	{
448	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
449
450	pthread_once (&doinit, etp_once_init);
451
452	want_poll_cb = want_poll;	515	want_poll_cb = want_poll;
453	done_poll_cb = done_poll;	516	done_poll_cb = done_poll;
454		517
455	return 0;	518	return 0;
456	}	519	}
457		520
458	X_THREAD_PROC (etp_proc);	521	X_THREAD_PROC (etp_proc);
459		522
		523	static void ecb_cold
460	static void etp_start_thread (void)	524	etp_start_thread (void)
461	{	525	{
462	etp_worker *wrk = calloc (1, sizeof (etp_worker));	526	etp_worker *wrk = calloc (1, sizeof (etp_worker));
463		527
464	/*TODO*/	528	/*TODO*/
465	assert (("unable to allocate worker thread data", wrk));	529	assert (("unable to allocate worker thread data", wrk));
…		…
478	free (wrk);	542	free (wrk);
479		543
480	X_UNLOCK (wrklock);	544	X_UNLOCK (wrklock);
481	}	545	}
482		546
		547	static void
483	static void etp_maybe_start_thread (void)	548	etp_maybe_start_thread (void)
484	{	549	{
485	if (expect_true (etp_nthreads () >= wanted))	550	if (ecb_expect_true (etp_nthreads () >= wanted))
486	return;	551	return;
487		552
488	/* todo: maybe use idle here, but might be less exact */	553	/* todo: maybe use idle here, but might be less exact */
489	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	554	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
490	return;	555	return;
491		556
492	etp_start_thread ();	557	etp_start_thread ();
493	}	558	}
494		559
		560	static void ecb_cold
495	static void etp_end_thread (void)	561	etp_end_thread (void)
496	{	562	{
497	eio_req *req = calloc (1, sizeof (eio_req));	563	eio_req *req = calloc (1, sizeof (eio_req));
498		564
499	req->type = -1;	565	req->type = -1;
500	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	566	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
507	X_LOCK (wrklock);	573	X_LOCK (wrklock);
508	--started;	574	--started;
509	X_UNLOCK (wrklock);	575	X_UNLOCK (wrklock);
510	}	576	}
511		577
512	static int etp_poll (void)	578	static int
		579	etp_poll (void)
513	{	580	{
514	unsigned int maxreqs;	581	unsigned int maxreqs;
515	unsigned int maxtime;	582	unsigned int maxtime;
516	struct timeval tv_start, tv_now;	583	struct timeval tv_start, tv_now;
517		584
…		…
547		614
548	X_LOCK (reqlock);	615	X_LOCK (reqlock);
549	--nreqs;	616	--nreqs;
550	X_UNLOCK (reqlock);	617	X_UNLOCK (reqlock);
551		618
552	if (expect_false (req->type == EIO_GROUP && req->size))	619	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
553	{	620	{
554	req->int1 = 1; /* mark request as delayed */	621	req->int1 = 1; /* mark request as delayed */
555	continue;	622	continue;
556	}	623	}
557	else	624	else
558	{	625	{
559	int res = ETP_FINISH (req);	626	int res = ETP_FINISH (req);
560	if (expect_false (res))	627	if (ecb_expect_false (res))
561	return res;	628	return res;
562	}	629	}
563		630
564	if (expect_false (maxreqs && !--maxreqs))	631	if (ecb_expect_false (maxreqs && !--maxreqs))
565	break;	632	break;
566		633
567	if (maxtime)	634	if (maxtime)
568	{	635	{
569	gettimeofday (&tv_now, 0);	636	gettimeofday (&tv_now, 0);
…		…
575		642
576	errno = EAGAIN;	643	errno = EAGAIN;
577	return -1;	644	return -1;
578	}	645	}
579		646
		647	static void
580	static void etp_cancel (ETP_REQ *req)	648	etp_cancel (ETP_REQ *req)
581	{	649	{
582	X_LOCK (wrklock);	650	req->cancelled = 1;
583	req->flags |= EIO_FLAG_CANCELLED;
584	X_UNLOCK (wrklock);
585		651
586	eio_grp_cancel (req);	652	eio_grp_cancel (req);
587	}	653	}
588		654
		655	static void
589	static void etp_submit (ETP_REQ *req)	656	etp_submit (ETP_REQ *req)
590	{	657	{
591	req->pri -= ETP_PRI_MIN;	658	req->pri -= ETP_PRI_MIN;
592		659
593	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	660	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
594	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	661	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595		662
596	if (expect_false (req->type == EIO_GROUP))	663	if (ecb_expect_false (req->type == EIO_GROUP))
597	{	664	{
598	/* I hope this is worth it :/ */	665	/* I hope this is worth it :/ */
599	X_LOCK (reqlock);	666	X_LOCK (reqlock);
600	++nreqs;	667	++nreqs;
601	X_UNLOCK (reqlock);	668	X_UNLOCK (reqlock);
…		…
620		687
621	etp_maybe_start_thread ();	688	etp_maybe_start_thread ();
622	}	689	}
623	}	690	}
624		691
		692	static void ecb_cold
625	static void etp_set_max_poll_time (double nseconds)	693	etp_set_max_poll_time (double nseconds)
626	{	694	{
627	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	695	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628	max_poll_time = nseconds * EIO_TICKS;	696	max_poll_time = nseconds * EIO_TICKS;
629	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	697	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630	}	698	}
631		699
		700	static void ecb_cold
632	static void etp_set_max_poll_reqs (unsigned int maxreqs)	701	etp_set_max_poll_reqs (unsigned int maxreqs)
633	{	702	{
634	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	703	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635	max_poll_reqs = maxreqs;	704	max_poll_reqs = maxreqs;
636	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	705	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637	}	706	}
638		707
		708	static void ecb_cold
639	static void etp_set_max_idle (unsigned int nthreads)	709	etp_set_max_idle (unsigned int nthreads)
640	{	710	{
641	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	711	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642	max_idle = nthreads;	712	max_idle = nthreads;
643	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	713	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644	}	714	}
645		715
		716	static void ecb_cold
646	static void etp_set_idle_timeout (unsigned int seconds)	717	etp_set_idle_timeout (unsigned int seconds)
647	{	718	{
648	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	719	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649	idle_timeout = seconds;	720	idle_timeout = seconds;
650	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	721	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651	}	722	}
652		723
		724	static void ecb_cold
653	static void etp_set_min_parallel (unsigned int nthreads)	725	etp_set_min_parallel (unsigned int nthreads)
654	{	726	{
655	if (wanted < nthreads)	727	if (wanted < nthreads)
656	wanted = nthreads;	728	wanted = nthreads;
657	}	729	}
658		730
		731	static void ecb_cold
659	static void etp_set_max_parallel (unsigned int nthreads)	732	etp_set_max_parallel (unsigned int nthreads)
660	{	733	{
661	if (wanted > nthreads)	734	if (wanted > nthreads)
662	wanted = nthreads;	735	wanted = nthreads;
663		736
664	while (started > wanted)	737	while (started > wanted)
665	etp_end_thread ();	738	etp_end_thread ();
666	}	739	}
667		740
668	/*****************************************************************************/	741	/*****************************************************************************/
669		742
		743	static void
670	static void grp_try_feed (eio_req *grp)	744	grp_try_feed (eio_req *grp)
671	{	745	{
672	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	746	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673	{	747	{
674	grp->flags &= ~EIO_FLAG_GROUPADD;	748	grp->flags &= ~EIO_FLAG_GROUPADD;
675		749
…		…
682	break;	756	break;
683	}	757	}
684	}	758	}
685	}	759	}
686		760
		761	static int
687	static int grp_dec (eio_req *grp)	762	grp_dec (eio_req *grp)
688	{	763	{
689	--grp->size;	764	--grp->size;
690		765
691	/* call feeder, if applicable */	766	/* call feeder, if applicable */
692	grp_try_feed (grp);	767	grp_try_feed (grp);
…		…
696	return eio_finish (grp);	771	return eio_finish (grp);
697	else	772	else
698	return 0;	773	return 0;
699	}	774	}
700		775
		776	static void
701	void eio_destroy (eio_req *req)	777	eio_destroy (eio_req *req)
702	{	778	{
703	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	779	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	780	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705		781
706	EIO_DESTROY (req);	782	EIO_DESTROY (req);
707	}	783	}
708		784
		785	static int
709	static int eio_finish (eio_req *req)	786	eio_finish (eio_req *req)
710	{	787	{
711	int res = EIO_FINISH (req);	788	int res = EIO_FINISH (req);
712		789
713	if (req->grp)	790	if (req->grp)
714	{	791	{
…		…
722	if (grp->grp_first == req)	799	if (grp->grp_first == req)
723	grp->grp_first = req->grp_next;	800	grp->grp_first = req->grp_next;
724		801
725	res2 = grp_dec (grp);	802	res2 = grp_dec (grp);
726		803
727	if (!res && res2)	804	if (!res)
728	res = res2;	805	res = res2;
729	}	806	}
730		807
731	eio_destroy (req);	808	eio_destroy (req);
732		809
733	return res;	810	return res;
734	}	811	}
735		812
		813	void
736	void eio_grp_cancel (eio_req *grp)	814	eio_grp_cancel (eio_req *grp)
737	{	815	{
738	for (grp = grp->grp_first; grp; grp = grp->grp_next)	816	for (grp = grp->grp_first; grp; grp = grp->grp_next)
739	eio_cancel (grp);	817	eio_cancel (grp);
740	}	818	}
741		819
		820	void
742	void eio_cancel (eio_req *req)	821	eio_cancel (eio_req *req)
743	{	822	{
744	etp_cancel (req);	823	etp_cancel (req);
745	}	824	}
746		825
		826	void
747	void eio_submit (eio_req *req)	827	eio_submit (eio_req *req)
748	{	828	{
749	etp_submit (req);	829	etp_submit (req);
750	}	830	}
751		831
752	unsigned int eio_nreqs (void)	832	unsigned int
		833	eio_nreqs (void)
753	{	834	{
754	return etp_nreqs ();	835	return etp_nreqs ();
755	}	836	}
756		837
757	unsigned int eio_nready (void)	838	unsigned int
		839	eio_nready (void)
758	{	840	{
759	return etp_nready ();	841	return etp_nready ();
760	}	842	}
761		843
762	unsigned int eio_npending (void)	844	unsigned int
		845	eio_npending (void)
763	{	846	{
764	return etp_npending ();	847	return etp_npending ();
765	}	848	}
766		849
767	unsigned int eio_nthreads (void)	850	unsigned int ecb_cold
		851	eio_nthreads (void)
768	{	852	{
769	return etp_nthreads ();	853	return etp_nthreads ();
770	}	854	}
771		855
		856	void ecb_cold
772	void eio_set_max_poll_time (double nseconds)	857	eio_set_max_poll_time (double nseconds)
773	{	858	{
774	etp_set_max_poll_time (nseconds);	859	etp_set_max_poll_time (nseconds);
775	}	860	}
776		861
		862	void ecb_cold
777	void eio_set_max_poll_reqs (unsigned int maxreqs)	863	eio_set_max_poll_reqs (unsigned int maxreqs)
778	{	864	{
779	etp_set_max_poll_reqs (maxreqs);	865	etp_set_max_poll_reqs (maxreqs);
780	}	866	}
781		867
		868	void ecb_cold
782	void eio_set_max_idle (unsigned int nthreads)	869	eio_set_max_idle (unsigned int nthreads)
783	{	870	{
784	etp_set_max_idle (nthreads);	871	etp_set_max_idle (nthreads);
785	}	872	}
786		873
		874	void ecb_cold
787	void eio_set_idle_timeout (unsigned int seconds)	875	eio_set_idle_timeout (unsigned int seconds)
788	{	876	{
789	etp_set_idle_timeout (seconds);	877	etp_set_idle_timeout (seconds);
790	}	878	}
791		879
		880	void ecb_cold
792	void eio_set_min_parallel (unsigned int nthreads)	881	eio_set_min_parallel (unsigned int nthreads)
793	{	882	{
794	etp_set_min_parallel (nthreads);	883	etp_set_min_parallel (nthreads);
795	}	884	}
796		885
		886	void ecb_cold
797	void eio_set_max_parallel (unsigned int nthreads)	887	eio_set_max_parallel (unsigned int nthreads)
798	{	888	{
799	etp_set_max_parallel (nthreads);	889	etp_set_max_parallel (nthreads);
800	}	890	}
801		891
802	int eio_poll (void)	892	int eio_poll (void)
…		…
811	# undef pread	901	# undef pread
812	# undef pwrite	902	# undef pwrite
813	# define pread eio__pread	903	# define pread eio__pread
814	# define pwrite eio__pwrite	904	# define pwrite eio__pwrite
815		905
816	static ssize_t	906	static eio_ssize_t
817	eio__pread (int fd, void *buf, size_t count, off_t offset)	907	eio__pread (int fd, void *buf, size_t count, off_t offset)
818	{	908	{
819	ssize_t res;	909	eio_ssize_t res;
820	off_t ooffset;	910	off_t ooffset;
821		911
822	X_LOCK (preadwritelock);	912	X_LOCK (preadwritelock);
823	ooffset = lseek (fd, 0, SEEK_CUR);	913	ooffset = lseek (fd, 0, SEEK_CUR);
824	lseek (fd, offset, SEEK_SET);	914	lseek (fd, offset, SEEK_SET);
…		…
827	X_UNLOCK (preadwritelock);	917	X_UNLOCK (preadwritelock);
828		918
829	return res;	919	return res;
830	}	920	}
831		921
832	static ssize_t	922	static eio_ssize_t
833	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	923	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834	{	924	{
835	ssize_t res;	925	eio_ssize_t res;
836	off_t ooffset;	926	off_t ooffset;
837		927
838	X_LOCK (preadwritelock);	928	X_LOCK (preadwritelock);
839	ooffset = lseek (fd, 0, SEEK_CUR);	929	ooffset = lseek (fd, 0, SEEK_CUR);
840	lseek (fd, offset, SEEK_SET);	930	lseek (fd, offset, SEEK_SET);
…		…
872	#ifndef HAVE_FUTIMES	962	#ifndef HAVE_FUTIMES
873		963
874	# undef futimes	964	# undef futimes
875	# define futimes(fd,times) eio__futimes (fd, times)	965	# define futimes(fd,times) eio__futimes (fd, times)
876		966
		967	static int
877	static int eio__futimes (int fd, const struct timeval tv[2])	968	eio__futimes (int fd, const struct timeval tv[2])
878	{	969	{
879	errno = ENOSYS;	970	errno = ENOSYS;
880	return -1;	971	return -1;
881	}	972	}
882		973
…		…
885	#if !HAVE_FDATASYNC	976	#if !HAVE_FDATASYNC
886	# undef fdatasync	977	# undef fdatasync
887	# define fdatasync(fd) fsync (fd)	978	# define fdatasync(fd) fsync (fd)
888	#endif	979	#endif
889		980
		981	static int
		982	eio__syncfs (int fd)
		983	{
		984	int res;
		985
		986	#if HAVE_SYS_SYNCFS
		987	res = (int)syscall (__NR_syncfs, (int)(fd));
		988	#else
		989	res = -1;
		990	errno = ENOSYS;
		991	#endif
		992
		993	if (res < 0 && errno == ENOSYS && fd >= 0)
		994	sync ();
		995
		996	return res;
		997	}
		998
890	/* sync_file_range always needs emulation */	999	/* sync_file_range always needs emulation */
891	int	1000	static int
892	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	1001	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893	{	1002	{
894	#if HAVE_SYNC_FILE_RANGE	1003	#if HAVE_SYNC_FILE_RANGE
895	int res;	1004	int res;
896		1005
…		…
913	/* even though we could play tricks with the flags, it's better to always	1022	/* even though we could play tricks with the flags, it's better to always
914	* call fdatasync, as that matches the expectation of its users best */	1023	* call fdatasync, as that matches the expectation of its users best */
915	return fdatasync (fd);	1024	return fdatasync (fd);
916	}	1025	}
917		1026
		1027	static int
		1028	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		1029	{
		1030	#if HAVE_FALLOCATE
		1031	return fallocate (fd, mode, offset, len);
		1032	#else
		1033	errno = ENOSYS;
		1034	return -1;
		1035	#endif
		1036	}
		1037
918	#if !HAVE_READAHEAD	1038	#if !HAVE_READAHEAD
919	# undef readahead	1039	# undef readahead
920	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1040	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921		1041
922	static ssize_t	1042	static eio_ssize_t
923	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1043	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
924	{	1044	{
925	size_t todo = count;	1045	size_t todo = count;
926	dBUF;	1046	dBUF;
927		1047
…		…
932	pread (fd, eio_buf, len, offset);	1052	pread (fd, eio_buf, len, offset);
933	offset += len;	1053	offset += len;
934	todo -= len;	1054	todo -= len;
935	}	1055	}
936		1056
		1057	FUBd;
		1058
937	errno = 0;	1059	errno = 0;
938	return count;	1060	return count;
939	}	1061	}
940		1062
941	#endif	1063	#endif
942		1064
943	/* sendfile always needs emulation */	1065	/* sendfile always needs emulation */
944	static ssize_t	1066	static eio_ssize_t
945	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1067	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
946	{	1068	{
947	ssize_t written = 0;	1069	eio_ssize_t written = 0;
948	ssize_t res;	1070	eio_ssize_t res;
949		1071
950	if (!count)	1072	if (!count)
951	return 0;	1073	return 0;
952		1074
953	for (;;)	1075	for (;;)
954	{	1076	{
		1077	#ifdef __APPLE__
		1078	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1079	#endif
955	#if HAVE_SENDFILE	1080	#if HAVE_SENDFILE
956	# if __linux	1081	# if __linux
957	off_t soffset = offset;	1082	off_t soffset = offset;
958	res = sendfile (ofd, ifd, &soffset, count);	1083	res = sendfile (ofd, ifd, &soffset, count);
959		1084
…		…
1002	if (res < 0 && sbytes)	1127	if (res < 0 && sbytes)
1003	res = sbytes;	1128	res = sbytes;
1004		1129
1005	# endif	1130	# endif
1006		1131
1007	#elif defined (_WIN32)	1132	#elif defined (_WIN32) && 0
1008	/* does not work, just for documentation of what would need to be done */	1133	/* does not work, just for documentation of what would need to be done */
1009	/* actually, cannot be done like this, as TransmitFile changes the file offset, */	1134	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
1010	/* libeio guarantees that the file offset does not change, and windows */	1135	/* libeio guarantees that the file offset does not change, and windows */
1011	/* has no way to get an independent handle to the same file description */	1136	/* has no way to get an independent handle to the same file description */
1012	HANDLE h = TO_SOCKET (ifd);	1137	HANDLE h = TO_SOCKET (ifd);
…		…
1045	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1170	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1046	/* BSDs */	1171	/* BSDs */
1047	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1172	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1048	|| errno == ENOTSUP	1173	|| errno == ENOTSUP
1049	#endif	1174	#endif
		1175	#ifdef EOPNOTSUPP /* windows */
1050	|| errno == EOPNOTSUPP /* BSDs */	1176	|| errno == EOPNOTSUPP /* BSDs */
		1177	#endif
1051	#if __solaris	1178	#if __solaris
1052	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1179	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1053	#endif	1180	#endif
1054	)	1181	)
1055	)	1182	)
…		…
1059		1186
1060	res = 0;	1187	res = 0;
1061		1188
1062	while (count)	1189	while (count)
1063	{	1190	{
1064	ssize_t cnt;	1191	eio_ssize_t cnt;
1065		1192
1066	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1193	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1067		1194
1068	if (cnt <= 0)	1195	if (cnt <= 0)
1069	{	1196	{
…		…
1081		1208
1082	offset += cnt;	1209	offset += cnt;
1083	res += cnt;	1210	res += cnt;
1084	count -= cnt;	1211	count -= cnt;
1085	}	1212	}
		1213
		1214	FUBd;
1086	}	1215	}
1087		1216
1088	return res;	1217	return res;
		1218	}
		1219
		1220	#ifdef PAGESIZE
		1221	# define eio_pagesize() PAGESIZE
		1222	#else
		1223	static intptr_t
		1224	eio_pagesize (void)
		1225	{
		1226	static intptr_t page;
		1227
		1228	if (!page)
		1229	page = sysconf (_SC_PAGESIZE);
		1230
		1231	return page;
		1232	}
		1233	#endif
		1234
		1235	static void
		1236	eio_page_align (void **addr, size_t *length)
		1237	{
		1238	intptr_t mask = eio_pagesize () - 1;
		1239
		1240	/* round down addr */
		1241	intptr_t adj = mask & (intptr_t)*addr;
		1242
		1243	*addr = (void *)((intptr_t)*addr - adj);
		1244	*length += adj;
		1245
		1246	/* round up length */
		1247	*length = (*length + mask) & ~mask;
		1248	}
		1249
		1250	#if !_POSIX_MEMLOCK
		1251	# define eio__mlockall(a) EIO_ENOSYS ()
		1252	#else
		1253
		1254	static int
		1255	eio__mlockall (int flags)
		1256	{
		1257	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1258	extern int mallopt (int, int);
		1259	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1260	#endif
		1261
		1262	if (EIO_MCL_CURRENT != MCL_CURRENT
		1263	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1264	{
		1265	flags = 0
		1266	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1267	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1268	}
		1269
		1270	return mlockall (flags);
		1271	}
		1272	#endif
		1273
		1274	#if !_POSIX_MEMLOCK_RANGE
		1275	# define eio__mlock(a,b) EIO_ENOSYS ()
		1276	#else
		1277
		1278	static int
		1279	eio__mlock (void *addr, size_t length)
		1280	{
		1281	eio_page_align (&addr, &length);
		1282
		1283	return mlock (addr, length);
		1284	}
		1285
		1286	#endif
		1287
		1288	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1289	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1290	#else
		1291
		1292	static int
		1293	eio__msync (void *mem, size_t len, int flags)
		1294	{
		1295	eio_page_align (&mem, &len);
		1296
		1297	if (EIO_MS_ASYNC != MS_SYNC
		1298	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1299	|| EIO_MS_SYNC != MS_SYNC)
		1300	{
		1301	flags = 0
		1302	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1303	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1304	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1305	}
		1306
		1307	return msync (mem, len, flags);
		1308	}
		1309
		1310	#endif
		1311
		1312	static int
		1313	eio__mtouch (eio_req *req)
		1314	{
		1315	void *mem = req->ptr2;
		1316	size_t len = req->size;
		1317	int flags = req->int1;
		1318
		1319	eio_page_align (&mem, &len);
		1320
		1321	{
		1322	intptr_t addr = (intptr_t)mem;
		1323	intptr_t end = addr + len;
		1324	intptr_t page = eio_pagesize ();
		1325
		1326	if (addr < end)
		1327	if (flags & EIO_MT_MODIFY) /* modify */
		1328	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1329	else
		1330	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1331	}
		1332
		1333	return 0;
		1334	}
		1335
		1336	/*****************************************************************************/
		1337	/* requests implemented outside eio_execute, because they are so large */
		1338
		1339	static void
		1340	eio__realpath (eio_req *req, etp_worker *self)
		1341	{
		1342	char *rel = req->ptr1;
		1343	char *res;
		1344	char *tmp1, *tmp2;
		1345	#if SYMLOOP_MAX > 32
		1346	int symlinks = SYMLOOP_MAX;
		1347	#else
		1348	int symlinks = 32;
		1349	#endif
		1350
		1351	req->result = -1;
		1352
		1353	errno = EINVAL;
		1354	if (!rel)
		1355	return;
		1356
		1357	errno = ENOENT;
		1358	if (!*rel)
		1359	return;
		1360
		1361	if (!req->ptr2)
		1362	{
		1363	X_LOCK (wrklock);
		1364	req->flags |= EIO_FLAG_PTR2_FREE;
		1365	X_UNLOCK (wrklock);
		1366	req->ptr2 = malloc (PATH_MAX * 3);
		1367
		1368	errno = ENOMEM;
		1369	if (!req->ptr2)
		1370	return;
		1371	}
		1372
		1373	res = req->ptr2;
		1374	tmp1 = res + PATH_MAX;
		1375	tmp2 = tmp1 + PATH_MAX;
		1376
		1377	#if 0 /* disabled, the musl way to do things is just too racy */
		1378	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1379	/* on linux we may be able to ask the kernel */
		1380	{
		1381	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1382
		1383	if (fd >= 0)
		1384	{
		1385	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1386	req->result = readlink (tmp1, res, PATH_MAX);
		1387	close (fd);
		1388
		1389	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1390
		1391	if (req->result > 0)
		1392	goto done;
		1393	}
		1394	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1395	return;
		1396	}
		1397	#endif
		1398	#endif
		1399
		1400	if (*rel != '/')
		1401	{
		1402	if (!getcwd (res, PATH_MAX))
		1403	return;
		1404
		1405	if (res [1]) /* only use if not / */
		1406	res += strlen (res);
		1407	}
		1408
		1409	while (*rel)
		1410	{
		1411	eio_ssize_t len, linklen;
		1412	char *beg = rel;
		1413
		1414	while (*rel && *rel != '/')
		1415	++rel;
		1416
		1417	len = rel - beg;
		1418
		1419	if (!len) /* skip slashes */
		1420	{
		1421	++rel;
		1422	continue;
		1423	}
		1424
		1425	if (beg [0] == '.')
		1426	{
		1427	if (len == 1)
		1428	continue; /* . - nop */
		1429
		1430	if (beg [1] == '.' && len == 2)
		1431	{
		1432	/* .. - back up one component, if possible */
		1433
		1434	while (res != req->ptr2)
		1435	if (*--res == '/')
		1436	break;
		1437
		1438	continue;
		1439	}
		1440	}
		1441
		1442	errno = ENAMETOOLONG;
		1443	if (res + 1 + len + 1 >= tmp1)
		1444	return;
		1445
		1446	/* copy one component */
		1447	*res = '/';
		1448	memcpy (res + 1, beg, len);
		1449
		1450	/* zero-terminate, for readlink */
		1451	res [len + 1] = 0;
		1452
		1453	/* now check if it's a symlink */
		1454	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1455
		1456	if (linklen < 0)
		1457	{
		1458	if (errno != EINVAL)
		1459	return;
		1460
		1461	/* it's a normal directory. hopefully */
		1462	res += len + 1;
		1463	}
		1464	else
		1465	{
		1466	/* yay, it was a symlink - build new path in tmp2 */
		1467	int rellen = strlen (rel);
		1468
		1469	errno = ENAMETOOLONG;
		1470	if (linklen + 1 + rellen >= PATH_MAX)
		1471	return;
		1472
		1473	errno = ELOOP;
		1474	if (!--symlinks)
		1475	return;
		1476
		1477	if (*tmp1 == '/')
		1478	res = req->ptr2; /* symlink resolves to an absolute path */
		1479
		1480	/* we need to be careful, as rel might point into tmp2 already */
		1481	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1482	tmp2 [linklen] = '/';
		1483	memcpy (tmp2, tmp1, linklen);
		1484
		1485	rel = tmp2;
		1486	}
		1487	}
		1488
		1489	/* special case for the lone root path */
		1490	if (res == req->ptr2)
		1491	*res++ = '/';
		1492
		1493	req->result = res - (char *)req->ptr2;
		1494
		1495	done:
		1496	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1089	}	1497	}
1090		1498
1091	static signed char	1499	static signed char
1092	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1500	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1093	{	1501	{
…		…
1101		1509
1102	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1510	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1103	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1511	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1104		1512
1105	static void	1513	static void
1106	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1514	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1107	{	1515	{
1108	unsigned char bits [9 + sizeof (ino_t) * 8];	1516	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1109	unsigned char *bit = bits;	1517	unsigned char *bit = bits;
1110		1518
1111	assert (CHAR_BIT == 8);	1519	assert (CHAR_BIT == 8);
1112	assert (sizeof (eio_dirent) * 8 < 256);	1520	assert (sizeof (eio_dirent) * 8 < 256);
1113	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1521	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1115		1523
1116	if (size <= EIO_SORT_FAST)	1524	if (size <= EIO_SORT_FAST)
1117	return;	1525	return;
1118		1526
1119	/* first prepare an array of bits to test in our radix sort */	1527	/* first prepare an array of bits to test in our radix sort */
1120	/* try to take endianness into account, as well as differences in ino_t sizes */	1528	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1121	/* inode_bits must contain all inodes ORed together */	1529	/* inode_bits must contain all inodes ORed together */
1122	/* which is used to skip bits that are 0 everywhere, which is very common */	1530	/* which is used to skip bits that are 0 everywhere, which is very common */
1123	{	1531	{
1124	ino_t endianness;	1532	eio_ino_t endianness;
1125	int i, j;	1533	int i, j;
1126		1534
1127	/* we store the byte offset of byte n into byte n of "endianness" */	1535	/* we store the byte offset of byte n into byte n of "endianness" */
1128	for (i = 0; i < sizeof (ino_t); ++i)	1536	for (i = 0; i < sizeof (eio_ino_t); ++i)
1129	((unsigned char *)&endianness)[i] = i;	1537	((unsigned char *)&endianness)[i] = i;
1130		1538
1131	*bit++ = 0;	1539	*bit++ = 0;
1132		1540
1133	for (i = 0; i < sizeof (ino_t); ++i)	1541	for (i = 0; i < sizeof (eio_ino_t); ++i)
1134	{	1542	{
1135	/* shifting off the byte offsets out of "endianness" */	1543	/* shifting off the byte offsets out of "endianness" */
1136	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1544	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1137	endianness >>= 8;	1545	endianness >>= 8;
1138		1546
1139	for (j = 0; j < 8; ++j)	1547	for (j = 0; j < 8; ++j)
1140	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1548	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1141	*bit++ = offs + j;	1549	*bit++ = offs + j;
1142	}	1550	}
1143		1551
1144	for (j = 0; j < 8; ++j)	1552	for (j = 0; j < 8; ++j)
1145	if (score_bits & (1 << j))	1553	if (score_bits & (1 << j))
1146	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1554	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1147	}	1555	}
1148		1556
1149	/* now actually do the sorting (a variant of MSD radix sort) */	1557	/* now actually do the sorting (a variant of MSD radix sort) */
1150	{	1558	{
1151	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1559	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1152	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1560	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1153	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1561	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1154	int stk_idx = 0;	1562	int stk_idx = 0;
1155		1563
1156	base_stk [stk_idx] = dents;	1564	base_stk [stk_idx] = dents;
1157	end_stk [stk_idx] = dents + size;	1565	end_stk [stk_idx] = dents + size;
1158	bit_stk [stk_idx] = bit - 1;	1566	bit_stk [stk_idx] = bit - 1;
…		…
1237	}	1645	}
1238	}	1646	}
1239	}	1647	}
1240		1648
1241	static void	1649	static void
1242	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1650	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1243	{	1651	{
1244	if (size <= 1)	1652	if (size <= 1)
1245	return; /* our insertion sort relies on size > 0 */	1653	return; /* our insertion sort relies on size > 0 */
1246		1654
1247	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1655	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1255		1663
1256	/* read a full directory */	1664	/* read a full directory */
1257	static void	1665	static void
1258	eio__scandir (eio_req *req, etp_worker *self)	1666	eio__scandir (eio_req *req, etp_worker *self)
1259	{	1667	{
1260	DIR *dirp;
1261	EIO_STRUCT_DIRENT *entp;
1262	char *name, *names;	1668	char *name, *names;
1263	int namesalloc = 4096;	1669	int namesalloc = 4096 - sizeof (void *) * 4;
1264	int namesoffs = 0;	1670	int namesoffs = 0;
1265	int flags = req->int1;	1671	int flags = req->int1;
1266	eio_dirent *dents = 0;	1672	eio_dirent *dents = 0;
1267	int dentalloc = 128;	1673	int dentalloc = 128;
1268	int dentoffs = 0;	1674	int dentoffs = 0;
1269	ino_t inode_bits = 0;	1675	eio_ino_t inode_bits = 0;
		1676	#ifdef _WIN32
		1677	HANDLE dirp;
		1678	WIN32_FIND_DATA entp;
		1679	#else
		1680	DIR *dirp;
		1681	EIO_STRUCT_DIRENT *entp;
		1682	#endif
1270		1683
1271	req->result = -1;	1684	req->result = -1;
1272		1685
1273	if (!(flags & EIO_READDIR_DENTS))	1686	if (!(flags & EIO_READDIR_DENTS))
1274	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1687	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1275		1688
1276	X_LOCK (wrklock);	1689	#ifdef _WIN32
1277	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1690	{
		1691	int len = strlen ((const char *)req->ptr1);
		1692	char *path = malloc (MAX_PATH);
		1693	const char *fmt;
		1694
		1695	if (!len)
		1696	fmt = "./*";
		1697	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1698	fmt = "%s*";
		1699	else
		1700	fmt = "%s/*";
		1701
		1702	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1703	dirp = FindFirstFile (path, &entp);
		1704	free (path);
		1705
		1706	if (dirp == INVALID_HANDLE_VALUE)
		1707	{
		1708	dirp = 0;
		1709
		1710	/* should steal _dosmaperr */
		1711	switch (GetLastError ())
		1712	{
		1713	case ERROR_FILE_NOT_FOUND:
		1714	req->result = 0;
		1715	break;
		1716
		1717	case ERROR_INVALID_NAME:
		1718	case ERROR_PATH_NOT_FOUND:
		1719	case ERROR_NO_MORE_FILES:
		1720	errno = ENOENT;
		1721	break;
		1722
		1723	case ERROR_NOT_ENOUGH_MEMORY:
		1724	errno = ENOMEM;
		1725	break;
		1726
		1727	default:
		1728	errno = EINVAL;
		1729	break;
		1730	}
		1731	}
		1732	}
		1733	#else
1278	self->dirp = dirp = opendir (req->ptr1);	1734	dirp = opendir (req->ptr1);
		1735	#endif
		1736
		1737	if (req->flags & EIO_FLAG_PTR1_FREE)
		1738	free (req->ptr1);
1279		1739
1280	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1740	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1281	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1741	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1282	req->ptr2 = names = malloc (namesalloc);	1742	req->ptr2 = names = malloc (namesalloc);
1283	X_UNLOCK (wrklock);
1284		1743
1285	if (dirp && names && (!flags || dents))	1744	if (dirp && names && (!flags || dents))
1286	for (;;)	1745	for (;;)
1287	{	1746	{
		1747	int done;
		1748
		1749	#ifdef _WIN32
		1750	done = !dirp;
		1751	#else
1288	errno = 0;	1752	errno = 0;
1289	entp = readdir (dirp);	1753	entp = readdir (dirp);
		1754	done = !entp;
		1755	#endif
1290		1756
1291	if (!entp)	1757	if (done)
1292	{	1758	{
		1759	#ifndef _WIN32
		1760	int old_errno = errno;
		1761	closedir (dirp);
		1762	errno = old_errno;
		1763
1293	if (errno)	1764	if (errno)
1294	break;	1765	break;
		1766	#endif
1295		1767
1296	/* sort etc. */	1768	/* sort etc. */
1297	req->int1 = flags;	1769	req->int1 = flags;
1298	req->result = dentoffs;	1770	req->result = dentoffs;
1299		1771
…		…
1328		1800
1329	break;	1801	break;
1330	}	1802	}
1331		1803
1332	/* now add the entry to our list(s) */	1804	/* now add the entry to our list(s) */
1333	name = entp->d_name;	1805	name = D_NAME (entp);
1334		1806
1335	/* skip . and .. entries */	1807	/* skip . and .. entries */
1336	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1808	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1337	{	1809	{
1338	int len = D_NAMLEN (entp) + 1;	1810	int len = D_NAMLEN (entp) + 1;
1339		1811
1340	while (expect_false (namesoffs + len > namesalloc))	1812	while (ecb_expect_false (namesoffs + len > namesalloc))
1341	{	1813	{
1342	namesalloc *= 2;	1814	namesalloc *= 2;
1343	X_LOCK (wrklock);
1344	req->ptr2 = names = realloc (names, namesalloc);	1815	req->ptr2 = names = realloc (names, namesalloc);
1345	X_UNLOCK (wrklock);
1346		1816
1347	if (!names)	1817	if (!names)
1348	break;	1818	break;
1349	}	1819	}
1350		1820
…		…
1352		1822
1353	if (dents)	1823	if (dents)
1354	{	1824	{
1355	struct eio_dirent *ent;	1825	struct eio_dirent *ent;
1356		1826
1357	if (expect_false (dentoffs == dentalloc))	1827	if (ecb_expect_false (dentoffs == dentalloc))
1358	{	1828	{
1359	dentalloc *= 2;	1829	dentalloc *= 2;
1360	X_LOCK (wrklock);
1361	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1830	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1362	X_UNLOCK (wrklock);
1363		1831
1364	if (!dents)	1832	if (!dents)
1365	break;	1833	break;
1366	}	1834	}
1367		1835
…		…
1447	if (EIO_CANCELLED (req))	1915	if (EIO_CANCELLED (req))
1448	{	1916	{
1449	errno = ECANCELED;	1917	errno = ECANCELED;
1450	break;	1918	break;
1451	}	1919	}
		1920
		1921	#ifdef _WIN32
		1922	if (!FindNextFile (dirp, &entp))
		1923	{
		1924	FindClose (dirp);
		1925	dirp = 0;
		1926	}
		1927	#endif
1452	}	1928	}
1453	}
1454
1455	#ifdef PAGESIZE
1456	# define eio_pagesize() PAGESIZE
1457	#else
1458	static intptr_t
1459	eio_pagesize (void)
1460	{
1461	static intptr_t page;
1462
1463	if (!page)
1464	page = sysconf (_SC_PAGESIZE);
1465
1466	return page;
1467	}
1468	#endif
1469
1470	static void
1471	eio_page_align (void **addr, size_t *length)
1472	{
1473	intptr_t mask = eio_pagesize () - 1;
1474
1475	/* round down addr */
1476	intptr_t adj = mask & (intptr_t)*addr;
1477
1478	*addr = (void *)((intptr_t)*addr - adj);
1479	*length += adj;
1480
1481	/* round up length */
1482	*length = (*length + mask) & ~mask;
1483	}
1484
1485	#if !_POSIX_MEMLOCK
1486	# define eio__mlockall(a) ((errno = ENOSYS), -1)
1487	#else
1488
1489	static int
1490	eio__mlockall (int flags)
1491	{
1492	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1493	extern int mallopt (int, int);
1494	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1495	#endif
1496
1497	if (EIO_MCL_CURRENT != MCL_CURRENT
1498	|| EIO_MCL_FUTURE != MCL_FUTURE)
1499	{
1500	flags = 0
1501	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1502	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1503	}
1504
1505	return mlockall (flags);
1506	}
1507	#endif
1508
1509	#if !_POSIX_MEMLOCK_RANGE
1510	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1511	#else
1512
1513	static int
1514	eio__mlock (void *addr, size_t length)
1515	{
1516	eio_page_align (&addr, &length);
1517
1518	return mlock (addr, length);
1519	}
1520
1521	#endif
1522
1523	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1524	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1525	#else
1526
1527	int
1528	eio__msync (void *mem, size_t len, int flags)
1529	{
1530	eio_page_align (&mem, &len);
1531
1532	if (EIO_MS_ASYNC != MS_SYNC
1533	|| EIO_MS_INVALIDATE != MS_INVALIDATE
1534	|| EIO_MS_SYNC != MS_SYNC)
1535	{
1536	flags = 0
1537	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1538	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1539	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1540	}
1541
1542	return msync (mem, len, flags);
1543	}
1544
1545	#endif
1546
1547	int
1548	eio__mtouch (eio_req *req)
1549	{
1550	void *mem = req->ptr2;
1551	size_t len = req->size;
1552	int flags = req->int1;
1553
1554	eio_page_align (&mem, &len);
1555
1556	{
1557	intptr_t addr = (intptr_t)mem;
1558	intptr_t end = addr + len;
1559	intptr_t page = eio_pagesize ();
1560
1561	if (addr < end)
1562	if (flags & EIO_MT_MODIFY) /* modify */
1563	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1564	else
1565	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1566	}
1567
1568	return 0;
1569	}	1929	}
1570		1930
1571	/*****************************************************************************/	1931	/*****************************************************************************/
1572		1932
1573	#define ALLOC(len) \	1933	#define ALLOC(len) \
…		…
1589	{	1949	{
1590	ETP_REQ *req;	1950	ETP_REQ *req;
1591	struct timespec ts;	1951	struct timespec ts;
1592	etp_worker *self = (etp_worker *)thr_arg;	1952	etp_worker *self = (etp_worker *)thr_arg;
1593		1953
		1954	#if HAVE_PRCTL_SET_NAME
		1955	prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
		1956	#endif
		1957
1594	/* try to distribute timeouts somewhat randomly */	1958	/* try to distribute timeouts somewhat evenly */
1595	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1959	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1596		1960
1597	for (;;)	1961	for (;;)
1598	{	1962	{
		1963	ts.tv_sec = 0;
		1964
1599	X_LOCK (reqlock);	1965	X_LOCK (reqlock);
1600		1966
1601	for (;;)	1967	for (;;)
1602	{	1968	{
1603	self->req = req = reqq_shift (&req_queue);	1969	self->req = req = reqq_shift (&req_queue);
1604		1970
1605	if (req)	1971	if (req)
1606	break;	1972	break;
1607		1973
		1974	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1975	{
		1976	X_UNLOCK (reqlock);
		1977	X_LOCK (wrklock);
		1978	--started;
		1979	X_UNLOCK (wrklock);
		1980	goto quit;
		1981	}
		1982
1608	++idle;	1983	++idle;
1609		1984
1610	ts.tv_sec = time (0) + idle_timeout;	1985	if (idle <= max_idle)
1611	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1986	/* we are allowed to idle, so do so without any timeout */
		1987	X_COND_WAIT (reqwait, reqlock);
		1988	else
1612	{	1989	{
1613	if (idle > max_idle)	1990	/* initialise timeout once */
1614	{	1991	if (!ts.tv_sec)
1615	--idle;	1992	ts.tv_sec = time (0) + idle_timeout;
1616	X_UNLOCK (reqlock);
1617	X_LOCK (wrklock);
1618	--started;
1619	X_UNLOCK (wrklock);
1620	goto quit;
1621	}
1622		1993
1623	/* we are allowed to idle, so do so without any timeout */
1624	X_COND_WAIT (reqwait, reqlock);	1994	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1995	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1625	}	1996	}
1626		1997
1627	--idle;	1998	--idle;
1628	}	1999	}
1629		2000
…		…
1632	X_UNLOCK (reqlock);	2003	X_UNLOCK (reqlock);
1633		2004
1634	if (req->type < 0)	2005	if (req->type < 0)
1635	goto quit;	2006	goto quit;
1636		2007
1637	if (!EIO_CANCELLED (req))
1638	ETP_EXECUTE (self, req);	2008	ETP_EXECUTE (self, req);
1639		2009
1640	X_LOCK (reslock);	2010	X_LOCK (reslock);
1641		2011
1642	++npending;	2012	++npending;
1643		2013
…		…
1658	return 0;	2028	return 0;
1659	}	2029	}
1660		2030
1661	/*****************************************************************************/	2031	/*****************************************************************************/
1662		2032
		2033	int ecb_cold
1663	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	2034	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1664	{	2035	{
		2036	#if !HAVE_PREADWRITE
		2037	X_MUTEX_CREATE (preadwritelock);
		2038	#endif
		2039
1665	return etp_init (want_poll, done_poll);	2040	return etp_init (want_poll, done_poll);
1666	}	2041	}
1667		2042
		2043	ecb_inline void
1668	static void eio_api_destroy (eio_req *req)	2044	eio_api_destroy (eio_req *req)
1669	{	2045	{
1670	free (req);	2046	free (req);
1671	}	2047	}
1672		2048
1673	#define REQ(rtype) \	2049	#define REQ(rtype) \
…		…
1692	{ \	2068	{ \
1693	eio_api_destroy (req); \	2069	eio_api_destroy (req); \
1694	return 0; \	2070	return 0; \
1695	}	2071	}
1696		2072
		2073	static void
1697	static void eio_execute (etp_worker *self, eio_req *req)	2074	eio_execute (etp_worker *self, eio_req *req)
1698	{	2075	{
		2076	if (ecb_expect_false (EIO_CANCELLED (req)))
		2077	{
		2078	req->result = -1;
		2079	req->errorno = ECANCELED;
		2080	return;
		2081	}
		2082
1699	switch (req->type)	2083	switch (req->type)
1700	{	2084	{
1701	case EIO_READ: ALLOC (req->size);	2085	case EIO_READ: ALLOC (req->size);
1702	req->result = req->offs >= 0	2086	req->result = req->offs >= 0
1703	? pread (req->int1, req->ptr2, req->size, req->offs)	2087	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1705	case EIO_WRITE: req->result = req->offs >= 0	2089	case EIO_WRITE: req->result = req->offs >= 0
1706	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2090	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1707	: write (req->int1, req->ptr2, req->size); break;	2091	: write (req->int1, req->ptr2, req->size); break;
1708		2092
1709	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2093	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1710	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2094	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1711		2095
1712	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2096	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1713	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2097	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1714	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2098	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1715	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2099	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
…		…
1737	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2121	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1738	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2122	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1739	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2123	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1740	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;	2124	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1741		2125
		2126	case EIO_REALPATH: eio__realpath (req, self); break;
		2127
1742	case EIO_READLINK: ALLOC (PATH_MAX);	2128	case EIO_READLINK: ALLOC (PATH_MAX);
1743	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;	2129	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1744		2130
1745	case EIO_SYNC: req->result = 0; sync (); break;	2131	case EIO_SYNC: req->result = 0; sync (); break;
1746	case EIO_FSYNC: req->result = fsync (req->int1); break;	2132	case EIO_FSYNC: req->result = fsync (req->int1); break;
1747	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2133	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2134	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2135	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1748	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2136	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1749	case EIO_MTOUCH: req->result = eio__mtouch (req); break;	2137	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1750	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2138	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1751	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2139	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1752	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	2140	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1753		2141
1754	case EIO_READDIR: eio__scandir (req, self); break;	2142	case EIO_READDIR: eio__scandir (req, self); break;
1755		2143
1756	case EIO_BUSY:	2144	case EIO_BUSY:
1757	#ifdef _WIN32	2145	#ifdef _WIN32
…		…
1837	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2225	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1838	{	2226	{
1839	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2227	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1840	}	2228	}
1841		2229
		2230	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		2231	{
		2232	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2233	}
		2234
		2235	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2236	{
		2237	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2238	}
		2239
		2240	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2241	{
		2242	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2243	}
		2244
1842	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2245	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1843	{	2246	{
1844	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2247	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1845	}	2248	}
1846		2249
…		…
1852	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)	2255	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1853	{	2256	{
1854	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;	2257	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1855	}	2258	}
1856		2259
1857	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2260	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1858	{	2261	{
1859	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2262	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1860	}
1861
1862	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1863	{
1864	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1865	}	2263	}
1866		2264
1867	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2265	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1868	{	2266	{
1869	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2267	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
1907	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2305	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1908	{	2306	{
1909	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2307	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1910	}	2308	}
1911		2309
1912	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2310	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1913	{	2311	{
1914	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2312	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1915	}	2313	}
1916		2314
1917	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2315	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1937	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2335	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1938	{	2336	{
1939	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2337	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1940	}	2338	}
1941		2339
1942	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2340	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1943	{	2341	{
1944	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2342	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1945	}	2343	}
1946		2344
1947	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2345	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1961	}	2359	}
1962		2360
1963	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2361	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1964	{	2362	{
1965	return eio__1path (EIO_READLINK, path, pri, cb, data);	2363	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2364	}
		2365
		2366	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2367	{
		2368	return eio__1path (EIO_REALPATH, path, pri, cb, data);
1966	}	2369	}
1967		2370
1968	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2371	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1969	{	2372	{
1970	return eio__1path (EIO_STAT, path, pri, cb, data);	2373	return eio__1path (EIO_STAT, path, pri, cb, data);
…		…
2029	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2432	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2030	{	2433	{
2031	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2434	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2032	}	2435	}
2033		2436
2034	eio_req *eio_custom (void (*)(eio_req *) execute, int pri, eio_cb cb, void *data);	2437	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2035	{	2438	{
2036	REQ (EIO_CUSTOM); req->feed = execute; SEND;	2439	REQ (EIO_CUSTOM); req->feed = execute; SEND;
2037	}	2440	}
2038		2441
2039	#endif	2442	#endif
…		…
2050	#undef SEND	2453	#undef SEND
2051		2454
2052	/*****************************************************************************/	2455	/*****************************************************************************/
2053	/* grp functions */	2456	/* grp functions */
2054		2457
		2458	void
2055	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2459	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2056	{	2460	{
2057	grp->int2 = limit;	2461	grp->int2 = limit;
2058	grp->feed = feed;	2462	grp->feed = feed;
2059		2463
2060	grp_try_feed (grp);	2464	grp_try_feed (grp);
2061	}	2465	}
2062		2466
		2467	void
2063	void eio_grp_limit (eio_req *grp, int limit)	2468	eio_grp_limit (eio_req *grp, int limit)
2064	{	2469	{
2065	grp->int2 = limit;	2470	grp->int2 = limit;
2066		2471
2067	grp_try_feed (grp);	2472	grp_try_feed (grp);
2068	}	2473	}
2069		2474
		2475	void
2070	void eio_grp_add (eio_req *grp, eio_req *req)	2476	eio_grp_add (eio_req *grp, eio_req *req)
2071	{	2477	{
2072	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2478	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2073		2479
2074	grp->flags |= EIO_FLAG_GROUPADD;	2480	grp->flags |= EIO_FLAG_GROUPADD;
2075		2481
…		…
2086	}	2492	}
2087		2493
2088	/*****************************************************************************/	2494	/*****************************************************************************/
2089	/* misc garbage */	2495	/* misc garbage */
2090		2496
		2497	eio_ssize_t
2091	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2498	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2092	{	2499	{
2093	etp_worker wrk;
2094	ssize_t ret;
2095
2096	wrk.dbuf = 0;
2097
2098	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2500	return eio__sendfile (ofd, ifd, offset, count);
2099
2100	if (wrk.dbuf)
2101	free (wrk.dbuf);
2102
2103	return ret;
2104	}	2501	}
2105		2502

Diff Legend

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