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 1.93 by root, Mon Jul 18 01:27:03 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	static void eio_destroy (eio_req *req);
		77
69	#ifndef EIO_FINISH	78	#ifndef EIO_FINISH
70	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	79	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
71	#endif	80	#endif
72		81
73	#ifndef EIO_DESTROY	82	#ifndef EIO_DESTROY
…		…
76		85
77	#ifndef EIO_FEED	86	#ifndef EIO_FEED
78	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	87	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
79	#endif	88	#endif
80		89
		90	#ifndef EIO_FD_TO_WIN32_HANDLE
		91	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		92	#endif
		93	#ifndef EIO_WIN32_HANDLE_TO_FD
		94	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		95	#endif
		96
		97	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		98
		99	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		100
81	#ifdef _WIN32	101	#ifdef _WIN32
82		102
83	/*doh*/	103	#undef PAGESIZE
		104	#define PAGESIZE 4096 /* GetSystemInfo? */
		105
		106	#ifdef EIO_STRUCT_STATI64
		107	#define stat(path,buf) _stati64 (path,buf)
		108	#define fstat(fd,buf) _fstati64 (path,buf)
		109	#endif
		110	#define lstat(path,buf) stat (path,buf)
		111	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		112	#define mkdir(path,mode) _mkdir (path)
		113	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		114
		115	#define chmod(path,mode) _chmod (path, mode)
		116	#define dup(fd) _dup (fd)
		117	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		118
		119	#define fchmod(fd,mode) EIO_ENOSYS ()
		120	#define chown(path,uid,gid) EIO_ENOSYS ()
		121	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		122	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		123	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		124	#define mknod(path,mode,dev) EIO_ENOSYS ()
		125	#define sync() EIO_ENOSYS ()
		126	#define readlink(path,buf,s) EIO_ENOSYS ()
		127	#define statvfs(path,buf) EIO_ENOSYS ()
		128	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		129
		130	/* we could even stat and see if it exists */
		131	static int
		132	symlink (const char *old, const char *neu)
		133	{
		134	#if WINVER >= 0x0600
		135	if (CreateSymbolicLink (neu, old, 1))
		136	return 0;
		137
		138	if (CreateSymbolicLink (neu, old, 0))
		139	return 0;
		140	#endif
		141
		142	return EIO_ERRNO (ENOENT, -1);
		143	}
		144
		145	/* POSIX API only */
		146	#define CreateHardLink(neu,old,flags) 0
		147	#define CreateSymbolicLink(neu,old,flags) 0
		148
		149	struct statvfs
		150	{
		151	int dummy;
		152	};
		153
		154	#define DT_DIR EIO_DT_DIR
		155	#define DT_REG EIO_DT_REG
		156	#define D_NAME(entp) entp.cFileName
		157	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
		158
84	#else	159	#else
85		160
86	# include <sys/time.h>	161	#include <sys/time.h>
87	# include <sys/select.h>	162	#include <sys/select.h>
		163	#include <sys/statvfs.h>
88	# include <unistd.h>	164	#include <unistd.h>
89	# include <utime.h>	165	#include <utime.h>
90	# include <signal.h>	166	#include <signal.h>
91	# include <dirent.h>	167	#include <dirent.h>
92		168
93	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES	169	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94	# include <sys/mman.h>	170	#include <sys/mman.h>
95	#endif	171	#endif
96		172
		173	#define D_NAME(entp) entp->d_name
		174
97	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */	175	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__	176	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99	# define _DIRENT_HAVE_D_TYPE /* sigh */	177	#define _DIRENT_HAVE_D_TYPE /* sigh */
100	# define D_INO(de) (de)->d_fileno	178	#define D_INO(de) (de)->d_fileno
101	# define D_NAMLEN(de) (de)->d_namlen	179	#define D_NAMLEN(de) (de)->d_namlen
102	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600	180	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103	# define D_INO(de) (de)->d_ino	181	#define D_INO(de) (de)->d_ino
104	# endif	182	#endif
105		183
106	#ifdef _D_EXACT_NAMLEN	184	#ifdef _D_EXACT_NAMLEN
107	# undef D_NAMLEN	185	#undef D_NAMLEN
108	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)	186	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109	#endif	187	#endif
110		188
111	# ifdef _DIRENT_HAVE_D_TYPE	189	#ifdef _DIRENT_HAVE_D_TYPE
112	# define D_TYPE(de) (de)->d_type	190	#define D_TYPE(de) (de)->d_type
113	# endif	191	#endif
114		192
115	# ifndef EIO_STRUCT_DIRENT	193	#ifndef EIO_STRUCT_DIRENT
116	# define EIO_STRUCT_DIRENT struct dirent	194	#define EIO_STRUCT_DIRENT struct dirent
117	# endif	195	#endif
118		196
119	#endif	197	#endif
120		198
121	#if HAVE_SENDFILE	199	#if HAVE_SENDFILE
122	# if __linux	200	# if __linux
…		…
138	#endif	216	#endif
139	#ifndef D_INO	217	#ifndef D_INO
140	# define D_INO(de) 0	218	# define D_INO(de) 0
141	#endif	219	#endif
142	#ifndef D_NAMLEN	220	#ifndef D_NAMLEN
143	# define D_NAMLEN(de) strlen ((de)->d_name)	221	# define D_NAMLEN(entp) strlen (D_NAME (entp))
144	#endif	222	#endif
145		223
146	/* used for struct dirent, AIX doesn't provide it */	224	/* used for struct dirent, AIX doesn't provide it */
147	#ifndef NAME_MAX	225	#ifndef NAME_MAX
148	# define NAME_MAX 4096	226	# define NAME_MAX 4096
…		…
165	if (!eio_buf) \	243	if (!eio_buf) \
166	return -1;	244	return -1;
167		245
168	#define EIO_TICKS ((1000000 + 1023) >> 10)	246	#define EIO_TICKS ((1000000 + 1023) >> 10)
169		247
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
183	#define ETP_PRI_MIN EIO_PRI_MIN	248	#define ETP_PRI_MIN EIO_PRI_MIN
184	#define ETP_PRI_MAX EIO_PRI_MAX	249	#define ETP_PRI_MAX EIO_PRI_MAX
185		250
186	struct etp_worker;	251	struct etp_worker;
187		252
…		…
195	#define ETP_WORKER_CLEAR(req) \	260	#define ETP_WORKER_CLEAR(req) \
196	if (wrk->dbuf) \	261	if (wrk->dbuf) \
197	{ \	262	{ \
198	free (wrk->dbuf); \	263	free (wrk->dbuf); \
199	wrk->dbuf = 0; \	264	wrk->dbuf = 0; \
200	} \
201	\
202	if (wrk->dirp) \
203	{ \
204	closedir (wrk->dirp); \
205	wrk->dirp = 0; \
206	}	265	}
207		266
208	#define ETP_WORKER_COMMON \	267	#define ETP_WORKER_COMMON \
209	void *dbuf; \	268	void *dbuf;
210	DIR *dirp;
211		269
212	/*****************************************************************************/	270	/*****************************************************************************/
213		271
214	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	272	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215		273
216	/* calculate time difference in ~1/EIO_TICKS of a second */	274	/* calculate time difference in ~1/EIO_TICKS of a second */
		275	ecb_inline int
217	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	276	tvdiff (struct timeval *tv1, struct timeval *tv2)
218	{	277	{
219	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	278	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	279	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
221	}	280	}
222		281
…		…
243	/*	302	/*
244	* make our pread/pwrite emulation safe against themselves, but not against	303	* make our pread/pwrite emulation safe against themselves, but not against
245	* normal read/write by using a mutex. slows down execution a lot,	304	* normal read/write by using a mutex. slows down execution a lot,
246	* but that's your problem, not mine.	305	* but that's your problem, not mine.
247	*/	306	*/
248	static xmutex_t preadwritelock = X_MUTEX_INIT;	307	static xmutex_t preadwritelock;
249	#endif	308	#endif
250		309
251	typedef struct etp_worker	310	typedef struct etp_worker
252	{	311	{
253	/* locked by wrklock */	312	/* locked by wrklock */
…		…
259	ETP_REQ *req; /* currently processed request */	318	ETP_REQ *req; /* currently processed request */
260		319
261	ETP_WORKER_COMMON	320	ETP_WORKER_COMMON
262	} etp_worker;	321	} etp_worker;
263		322
264	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	323	static etp_worker wrk_first; /* NOT etp */
265		324
266	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	325	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	326	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268		327
269	/* worker threads management */	328	/* worker threads management */
270		329
		330	static void ecb_cold
271	static void etp_worker_clear (etp_worker *wrk)	331	etp_worker_clear (etp_worker *wrk)
272	{	332	{
273	ETP_WORKER_CLEAR (wrk);	333	ETP_WORKER_CLEAR (wrk);
274	}	334	}
275		335
		336	static void ecb_cold
276	static void etp_worker_free (etp_worker *wrk)	337	etp_worker_free (etp_worker *wrk)
277	{	338	{
278	wrk->next->prev = wrk->prev;	339	wrk->next->prev = wrk->prev;
279	wrk->prev->next = wrk->next;	340	wrk->prev->next = wrk->next;
280		341
281	free (wrk);	342	free (wrk);
282	}	343	}
283		344
284	static unsigned int etp_nreqs (void)	345	static unsigned int
		346	etp_nreqs (void)
285	{	347	{
286	int retval;	348	int retval;
287	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	349	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288	retval = nreqs;	350	retval = nreqs;
289	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	351	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290	return retval;	352	return retval;
291	}	353	}
292		354
293	static unsigned int etp_nready (void)	355	static unsigned int
		356	etp_nready (void)
294	{	357	{
295	unsigned int retval;	358	unsigned int retval;
296		359
297	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	360	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298	retval = nready;	361	retval = nready;
299	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	362	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300		363
301	return retval;	364	return retval;
302	}	365	}
303		366
304	static unsigned int etp_npending (void)	367	static unsigned int
		368	etp_npending (void)
305	{	369	{
306	unsigned int retval;	370	unsigned int retval;
307		371
308	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	372	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309	retval = npending;	373	retval = npending;
310	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	374	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311		375
312	return retval;	376	return retval;
313	}	377	}
314		378
315	static unsigned int etp_nthreads (void)	379	static unsigned int
		380	etp_nthreads (void)
316	{	381	{
317	unsigned int retval;	382	unsigned int retval;
318		383
319	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	384	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320	retval = started;	385	retval = started;
…		…
334	} etp_reqq;	399	} etp_reqq;
335		400
336	static etp_reqq req_queue;	401	static etp_reqq req_queue;
337	static etp_reqq res_queue;	402	static etp_reqq res_queue;
338		403
		404	static void ecb_noinline ecb_cold
		405	reqq_init (etp_reqq *q)
		406	{
		407	int pri;
		408
		409	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		410	q->qs[pri] = q->qe[pri] = 0;
		411
		412	q->size = 0;
		413	}
		414
		415	static int ecb_noinline
339	static int reqq_push (etp_reqq *q, ETP_REQ *req)	416	reqq_push (etp_reqq *q, ETP_REQ *req)
340	{	417	{
341	int pri = req->pri;	418	int pri = req->pri;
342	req->next = 0;	419	req->next = 0;
343		420
344	if (q->qe[pri])	421	if (q->qe[pri])
…		…
350	q->qe[pri] = q->qs[pri] = req;	427	q->qe[pri] = q->qs[pri] = req;
351		428
352	return q->size++;	429	return q->size++;
353	}	430	}
354		431
		432	static ETP_REQ * ecb_noinline
355	static ETP_REQ *reqq_shift (etp_reqq *q)	433	reqq_shift (etp_reqq *q)
356	{	434	{
357	int pri;	435	int pri;
358		436
359	if (!q->size)	437	if (!q->size)
360	return 0;	438	return 0;
…		…
375	}	453	}
376		454
377	abort ();	455	abort ();
378	}	456	}
379		457
380	static void etp_thread_init (void)	458	static int ecb_cold
		459	etp_init (void (*want_poll)(void), void (*done_poll)(void))
381	{	460	{
382	X_MUTEX_CREATE (wrklock);	461	X_MUTEX_CREATE (wrklock);
383	X_MUTEX_CREATE (reslock);	462	X_MUTEX_CREATE (reslock);
384	X_MUTEX_CREATE (reqlock);	463	X_MUTEX_CREATE (reqlock);
385	X_COND_CREATE (reqwait);	464	X_COND_CREATE (reqwait);
386	}
387		465
388	static void etp_atfork_prepare (void)	466	reqq_init (&req_queue);
389	{	467	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		468
398	static void etp_atfork_parent (void)	469	wrk_first.next =
399	{	470	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		471
429	started = 0;	472	started = 0;
430	idle = 0;	473	idle = 0;
431	nreqs = 0;	474	nreqs = 0;
432	nready = 0;	475	nready = 0;
433	npending = 0;	476	npending = 0;
434		477
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;	478	want_poll_cb = want_poll;
453	done_poll_cb = done_poll;	479	done_poll_cb = done_poll;
454		480
455	return 0;	481	return 0;
456	}	482	}
457		483
458	X_THREAD_PROC (etp_proc);	484	X_THREAD_PROC (etp_proc);
459		485
		486	static void ecb_cold
460	static void etp_start_thread (void)	487	etp_start_thread (void)
461	{	488	{
462	etp_worker *wrk = calloc (1, sizeof (etp_worker));	489	etp_worker *wrk = calloc (1, sizeof (etp_worker));
463		490
464	/*TODO*/	491	/*TODO*/
465	assert (("unable to allocate worker thread data", wrk));	492	assert (("unable to allocate worker thread data", wrk));
…		…
478	free (wrk);	505	free (wrk);
479		506
480	X_UNLOCK (wrklock);	507	X_UNLOCK (wrklock);
481	}	508	}
482		509
		510	static void
483	static void etp_maybe_start_thread (void)	511	etp_maybe_start_thread (void)
484	{	512	{
485	if (expect_true (etp_nthreads () >= wanted))	513	if (ecb_expect_true (etp_nthreads () >= wanted))
486	return;	514	return;
487		515
488	/* todo: maybe use idle here, but might be less exact */	516	/* todo: maybe use idle here, but might be less exact */
489	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	517	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
490	return;	518	return;
491		519
492	etp_start_thread ();	520	etp_start_thread ();
493	}	521	}
494		522
		523	static void ecb_cold
495	static void etp_end_thread (void)	524	etp_end_thread (void)
496	{	525	{
497	eio_req *req = calloc (1, sizeof (eio_req));	526	eio_req *req = calloc (1, sizeof (eio_req));
498		527
499	req->type = -1;	528	req->type = -1;
500	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	529	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
507	X_LOCK (wrklock);	536	X_LOCK (wrklock);
508	--started;	537	--started;
509	X_UNLOCK (wrklock);	538	X_UNLOCK (wrklock);
510	}	539	}
511		540
512	static int etp_poll (void)	541	static int
		542	etp_poll (void)
513	{	543	{
514	unsigned int maxreqs;	544	unsigned int maxreqs;
515	unsigned int maxtime;	545	unsigned int maxtime;
516	struct timeval tv_start, tv_now;	546	struct timeval tv_start, tv_now;
517		547
…		…
547		577
548	X_LOCK (reqlock);	578	X_LOCK (reqlock);
549	--nreqs;	579	--nreqs;
550	X_UNLOCK (reqlock);	580	X_UNLOCK (reqlock);
551		581
552	if (expect_false (req->type == EIO_GROUP && req->size))	582	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
553	{	583	{
554	req->int1 = 1; /* mark request as delayed */	584	req->int1 = 1; /* mark request as delayed */
555	continue;	585	continue;
556	}	586	}
557	else	587	else
558	{	588	{
559	int res = ETP_FINISH (req);	589	int res = ETP_FINISH (req);
560	if (expect_false (res))	590	if (ecb_expect_false (res))
561	return res;	591	return res;
562	}	592	}
563		593
564	if (expect_false (maxreqs && !--maxreqs))	594	if (ecb_expect_false (maxreqs && !--maxreqs))
565	break;	595	break;
566		596
567	if (maxtime)	597	if (maxtime)
568	{	598	{
569	gettimeofday (&tv_now, 0);	599	gettimeofday (&tv_now, 0);
…		…
575		605
576	errno = EAGAIN;	606	errno = EAGAIN;
577	return -1;	607	return -1;
578	}	608	}
579		609
		610	static void
580	static void etp_cancel (ETP_REQ *req)	611	etp_cancel (ETP_REQ *req)
581	{	612	{
582	X_LOCK (wrklock);	613	req->cancelled = 1;
583	req->flags |= EIO_FLAG_CANCELLED;
584	X_UNLOCK (wrklock);
585		614
586	eio_grp_cancel (req);	615	eio_grp_cancel (req);
587	}	616	}
588		617
		618	static void
589	static void etp_submit (ETP_REQ *req)	619	etp_submit (ETP_REQ *req)
590	{	620	{
591	req->pri -= ETP_PRI_MIN;	621	req->pri -= ETP_PRI_MIN;
592		622
593	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	623	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;	624	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595		625
596	if (expect_false (req->type == EIO_GROUP))	626	if (ecb_expect_false (req->type == EIO_GROUP))
597	{	627	{
598	/* I hope this is worth it :/ */	628	/* I hope this is worth it :/ */
599	X_LOCK (reqlock);	629	X_LOCK (reqlock);
600	++nreqs;	630	++nreqs;
601	X_UNLOCK (reqlock);	631	X_UNLOCK (reqlock);
…		…
620		650
621	etp_maybe_start_thread ();	651	etp_maybe_start_thread ();
622	}	652	}
623	}	653	}
624		654
		655	static void ecb_cold
625	static void etp_set_max_poll_time (double nseconds)	656	etp_set_max_poll_time (double nseconds)
626	{	657	{
627	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	658	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628	max_poll_time = nseconds * EIO_TICKS;	659	max_poll_time = nseconds * EIO_TICKS;
629	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	660	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630	}	661	}
631		662
		663	static void ecb_cold
632	static void etp_set_max_poll_reqs (unsigned int maxreqs)	664	etp_set_max_poll_reqs (unsigned int maxreqs)
633	{	665	{
634	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	666	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635	max_poll_reqs = maxreqs;	667	max_poll_reqs = maxreqs;
636	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	668	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637	}	669	}
638		670
		671	static void ecb_cold
639	static void etp_set_max_idle (unsigned int nthreads)	672	etp_set_max_idle (unsigned int nthreads)
640	{	673	{
641	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	674	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642	max_idle = nthreads;	675	max_idle = nthreads;
643	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	676	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644	}	677	}
645		678
		679	static void ecb_cold
646	static void etp_set_idle_timeout (unsigned int seconds)	680	etp_set_idle_timeout (unsigned int seconds)
647	{	681	{
648	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	682	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649	idle_timeout = seconds;	683	idle_timeout = seconds;
650	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	684	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651	}	685	}
652		686
		687	static void ecb_cold
653	static void etp_set_min_parallel (unsigned int nthreads)	688	etp_set_min_parallel (unsigned int nthreads)
654	{	689	{
655	if (wanted < nthreads)	690	if (wanted < nthreads)
656	wanted = nthreads;	691	wanted = nthreads;
657	}	692	}
658		693
		694	static void ecb_cold
659	static void etp_set_max_parallel (unsigned int nthreads)	695	etp_set_max_parallel (unsigned int nthreads)
660	{	696	{
661	if (wanted > nthreads)	697	if (wanted > nthreads)
662	wanted = nthreads;	698	wanted = nthreads;
663		699
664	while (started > wanted)	700	while (started > wanted)
665	etp_end_thread ();	701	etp_end_thread ();
666	}	702	}
667		703
668	/*****************************************************************************/	704	/*****************************************************************************/
669		705
		706	static void
670	static void grp_try_feed (eio_req *grp)	707	grp_try_feed (eio_req *grp)
671	{	708	{
672	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	709	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673	{	710	{
674	grp->flags &= ~EIO_FLAG_GROUPADD;	711	grp->flags &= ~EIO_FLAG_GROUPADD;
675		712
…		…
682	break;	719	break;
683	}	720	}
684	}	721	}
685	}	722	}
686		723
		724	static int
687	static int grp_dec (eio_req *grp)	725	grp_dec (eio_req *grp)
688	{	726	{
689	--grp->size;	727	--grp->size;
690		728
691	/* call feeder, if applicable */	729	/* call feeder, if applicable */
692	grp_try_feed (grp);	730	grp_try_feed (grp);
…		…
696	return eio_finish (grp);	734	return eio_finish (grp);
697	else	735	else
698	return 0;	736	return 0;
699	}	737	}
700		738
		739	static void
701	void eio_destroy (eio_req *req)	740	eio_destroy (eio_req *req)
702	{	741	{
703	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	742	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	743	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705		744
706	EIO_DESTROY (req);	745	EIO_DESTROY (req);
707	}	746	}
708		747
		748	static int
709	static int eio_finish (eio_req *req)	749	eio_finish (eio_req *req)
710	{	750	{
711	int res = EIO_FINISH (req);	751	int res = EIO_FINISH (req);
712		752
713	if (req->grp)	753	if (req->grp)
714	{	754	{
…		…
722	if (grp->grp_first == req)	762	if (grp->grp_first == req)
723	grp->grp_first = req->grp_next;	763	grp->grp_first = req->grp_next;
724		764
725	res2 = grp_dec (grp);	765	res2 = grp_dec (grp);
726		766
727	if (!res && res2)	767	if (!res)
728	res = res2;	768	res = res2;
729	}	769	}
730		770
731	eio_destroy (req);	771	eio_destroy (req);
732		772
733	return res;	773	return res;
734	}	774	}
735		775
		776	void
736	void eio_grp_cancel (eio_req *grp)	777	eio_grp_cancel (eio_req *grp)
737	{	778	{
738	for (grp = grp->grp_first; grp; grp = grp->grp_next)	779	for (grp = grp->grp_first; grp; grp = grp->grp_next)
739	eio_cancel (grp);	780	eio_cancel (grp);
740	}	781	}
741		782
		783	void
742	void eio_cancel (eio_req *req)	784	eio_cancel (eio_req *req)
743	{	785	{
744	etp_cancel (req);	786	etp_cancel (req);
745	}	787	}
746		788
		789	void
747	void eio_submit (eio_req *req)	790	eio_submit (eio_req *req)
748	{	791	{
749	etp_submit (req);	792	etp_submit (req);
750	}	793	}
751		794
752	unsigned int eio_nreqs (void)	795	unsigned int
		796	eio_nreqs (void)
753	{	797	{
754	return etp_nreqs ();	798	return etp_nreqs ();
755	}	799	}
756		800
757	unsigned int eio_nready (void)	801	unsigned int
		802	eio_nready (void)
758	{	803	{
759	return etp_nready ();	804	return etp_nready ();
760	}	805	}
761		806
762	unsigned int eio_npending (void)	807	unsigned int
		808	eio_npending (void)
763	{	809	{
764	return etp_npending ();	810	return etp_npending ();
765	}	811	}
766		812
767	unsigned int eio_nthreads (void)	813	unsigned int ecb_cold
		814	eio_nthreads (void)
768	{	815	{
769	return etp_nthreads ();	816	return etp_nthreads ();
770	}	817	}
771		818
		819	void ecb_cold
772	void eio_set_max_poll_time (double nseconds)	820	eio_set_max_poll_time (double nseconds)
773	{	821	{
774	etp_set_max_poll_time (nseconds);	822	etp_set_max_poll_time (nseconds);
775	}	823	}
776		824
		825	void ecb_cold
777	void eio_set_max_poll_reqs (unsigned int maxreqs)	826	eio_set_max_poll_reqs (unsigned int maxreqs)
778	{	827	{
779	etp_set_max_poll_reqs (maxreqs);	828	etp_set_max_poll_reqs (maxreqs);
780	}	829	}
781		830
		831	void ecb_cold
782	void eio_set_max_idle (unsigned int nthreads)	832	eio_set_max_idle (unsigned int nthreads)
783	{	833	{
784	etp_set_max_idle (nthreads);	834	etp_set_max_idle (nthreads);
785	}	835	}
786		836
		837	void ecb_cold
787	void eio_set_idle_timeout (unsigned int seconds)	838	eio_set_idle_timeout (unsigned int seconds)
788	{	839	{
789	etp_set_idle_timeout (seconds);	840	etp_set_idle_timeout (seconds);
790	}	841	}
791		842
		843	void ecb_cold
792	void eio_set_min_parallel (unsigned int nthreads)	844	eio_set_min_parallel (unsigned int nthreads)
793	{	845	{
794	etp_set_min_parallel (nthreads);	846	etp_set_min_parallel (nthreads);
795	}	847	}
796		848
		849	void ecb_cold
797	void eio_set_max_parallel (unsigned int nthreads)	850	eio_set_max_parallel (unsigned int nthreads)
798	{	851	{
799	etp_set_max_parallel (nthreads);	852	etp_set_max_parallel (nthreads);
800	}	853	}
801		854
802	int eio_poll (void)	855	int eio_poll (void)
…		…
811	# undef pread	864	# undef pread
812	# undef pwrite	865	# undef pwrite
813	# define pread eio__pread	866	# define pread eio__pread
814	# define pwrite eio__pwrite	867	# define pwrite eio__pwrite
815		868
816	static ssize_t	869	static eio_ssize_t
817	eio__pread (int fd, void *buf, size_t count, off_t offset)	870	eio__pread (int fd, void *buf, size_t count, off_t offset)
818	{	871	{
819	ssize_t res;	872	eio_ssize_t res;
820	off_t ooffset;	873	off_t ooffset;
821		874
822	X_LOCK (preadwritelock);	875	X_LOCK (preadwritelock);
823	ooffset = lseek (fd, 0, SEEK_CUR);	876	ooffset = lseek (fd, 0, SEEK_CUR);
824	lseek (fd, offset, SEEK_SET);	877	lseek (fd, offset, SEEK_SET);
…		…
827	X_UNLOCK (preadwritelock);	880	X_UNLOCK (preadwritelock);
828		881
829	return res;	882	return res;
830	}	883	}
831		884
832	static ssize_t	885	static eio_ssize_t
833	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	886	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834	{	887	{
835	ssize_t res;	888	eio_ssize_t res;
836	off_t ooffset;	889	off_t ooffset;
837		890
838	X_LOCK (preadwritelock);	891	X_LOCK (preadwritelock);
839	ooffset = lseek (fd, 0, SEEK_CUR);	892	ooffset = lseek (fd, 0, SEEK_CUR);
840	lseek (fd, offset, SEEK_SET);	893	lseek (fd, offset, SEEK_SET);
…		…
872	#ifndef HAVE_FUTIMES	925	#ifndef HAVE_FUTIMES
873		926
874	# undef futimes	927	# undef futimes
875	# define futimes(fd,times) eio__futimes (fd, times)	928	# define futimes(fd,times) eio__futimes (fd, times)
876		929
		930	static int
877	static int eio__futimes (int fd, const struct timeval tv[2])	931	eio__futimes (int fd, const struct timeval tv[2])
878	{	932	{
879	errno = ENOSYS;	933	errno = ENOSYS;
880	return -1;	934	return -1;
881	}	935	}
882		936
…		…
886	# undef fdatasync	940	# undef fdatasync
887	# define fdatasync(fd) fsync (fd)	941	# define fdatasync(fd) fsync (fd)
888	#endif	942	#endif
889		943
890	/* sync_file_range always needs emulation */	944	/* sync_file_range always needs emulation */
891	int	945	static int
892	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	946	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893	{	947	{
894	#if HAVE_SYNC_FILE_RANGE	948	#if HAVE_SYNC_FILE_RANGE
895	int res;	949	int res;
896		950
…		…
913	/* even though we could play tricks with the flags, it's better to always	967	/* 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 */	968	* call fdatasync, as that matches the expectation of its users best */
915	return fdatasync (fd);	969	return fdatasync (fd);
916	}	970	}
917		971
		972	static int
		973	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		974	{
		975	#if HAVE_FALLOCATE
		976	return fallocate (fd, mode, offset, len);
		977	#else
		978	errno = ENOSYS;
		979	return -1;
		980	#endif
		981	}
		982
918	#if !HAVE_READAHEAD	983	#if !HAVE_READAHEAD
919	# undef readahead	984	# undef readahead
920	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	985	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921		986
922	static ssize_t	987	static eio_ssize_t
923	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	988	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
924	{	989	{
925	size_t todo = count;	990	size_t todo = count;
926	dBUF;	991	dBUF;
927		992
…		…
939	}	1004	}
940		1005
941	#endif	1006	#endif
942		1007
943	/* sendfile always needs emulation */	1008	/* sendfile always needs emulation */
944	static ssize_t	1009	static eio_ssize_t
945	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1010	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
946	{	1011	{
947	ssize_t written = 0;	1012	eio_ssize_t written = 0;
948	ssize_t res;	1013	eio_ssize_t res;
949		1014
950	if (!count)	1015	if (!count)
951	return 0;	1016	return 0;
952		1017
953	for (;;)	1018	for (;;)
954	{	1019	{
		1020	#ifdef __APPLE__
		1021	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1022	#endif
955	#if HAVE_SENDFILE	1023	#if HAVE_SENDFILE
956	# if __linux	1024	# if __linux
957	off_t soffset = offset;	1025	off_t soffset = offset;
958	res = sendfile (ofd, ifd, &soffset, count);	1026	res = sendfile (ofd, ifd, &soffset, count);
959		1027
…		…
1002	if (res < 0 && sbytes)	1070	if (res < 0 && sbytes)
1003	res = sbytes;	1071	res = sbytes;
1004		1072
1005	# endif	1073	# endif
1006		1074
1007	#elif defined (_WIN32)	1075	#elif defined (_WIN32) && 0
1008	/* does not work, just for documentation of what would need to be done */	1076	/* 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, */	1077	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
1010	/* libeio guarantees that the file offset does not change, and windows */	1078	/* 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 */	1079	/* has no way to get an independent handle to the same file description */
1012	HANDLE h = TO_SOCKET (ifd);	1080	HANDLE h = TO_SOCKET (ifd);
…		…
1045	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1113	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1046	/* BSDs */	1114	/* BSDs */
1047	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1115	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1048	|| errno == ENOTSUP	1116	|| errno == ENOTSUP
1049	#endif	1117	#endif
		1118	#ifdef EOPNOTSUPP /* windows */
1050	|| errno == EOPNOTSUPP /* BSDs */	1119	|| errno == EOPNOTSUPP /* BSDs */
		1120	#endif
1051	#if __solaris	1121	#if __solaris
1052	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1122	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1053	#endif	1123	#endif
1054	)	1124	)
1055	)	1125	)
…		…
1059		1129
1060	res = 0;	1130	res = 0;
1061		1131
1062	while (count)	1132	while (count)
1063	{	1133	{
1064	ssize_t cnt;	1134	eio_ssize_t cnt;
1065		1135
1066	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1136	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1067		1137
1068	if (cnt <= 0)	1138	if (cnt <= 0)
1069	{	1139	{
…		…
1086	}	1156	}
1087		1157
1088	return res;	1158	return res;
1089	}	1159	}
1090		1160
		1161	#ifdef PAGESIZE
		1162	# define eio_pagesize() PAGESIZE
		1163	#else
		1164	static intptr_t
		1165	eio_pagesize (void)
		1166	{
		1167	static intptr_t page;
		1168
		1169	if (!page)
		1170	page = sysconf (_SC_PAGESIZE);
		1171
		1172	return page;
		1173	}
		1174	#endif
		1175
		1176	static void
		1177	eio_page_align (void **addr, size_t *length)
		1178	{
		1179	intptr_t mask = eio_pagesize () - 1;
		1180
		1181	/* round down addr */
		1182	intptr_t adj = mask & (intptr_t)*addr;
		1183
		1184	*addr = (void *)((intptr_t)*addr - adj);
		1185	*length += adj;
		1186
		1187	/* round up length */
		1188	*length = (*length + mask) & ~mask;
		1189	}
		1190
		1191	#if !_POSIX_MEMLOCK
		1192	# define eio__mlockall(a) EIO_ENOSYS ()
		1193	#else
		1194
		1195	static int
		1196	eio__mlockall (int flags)
		1197	{
		1198	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1199	extern int mallopt (int, int);
		1200	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1201	#endif
		1202
		1203	if (EIO_MCL_CURRENT != MCL_CURRENT
		1204	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1205	{
		1206	flags = 0
		1207	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1208	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1209	}
		1210
		1211	return mlockall (flags);
		1212	}
		1213	#endif
		1214
		1215	#if !_POSIX_MEMLOCK_RANGE
		1216	# define eio__mlock(a,b) EIO_ENOSYS ()
		1217	#else
		1218
		1219	static int
		1220	eio__mlock (void *addr, size_t length)
		1221	{
		1222	eio_page_align (&addr, &length);
		1223
		1224	return mlock (addr, length);
		1225	}
		1226
		1227	#endif
		1228
		1229	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1230	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1231	#else
		1232
		1233	static int
		1234	eio__msync (void *mem, size_t len, int flags)
		1235	{
		1236	eio_page_align (&mem, &len);
		1237
		1238	if (EIO_MS_ASYNC != MS_SYNC
		1239	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1240	|| EIO_MS_SYNC != MS_SYNC)
		1241	{
		1242	flags = 0
		1243	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1244	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1245	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1246	}
		1247
		1248	return msync (mem, len, flags);
		1249	}
		1250
		1251	#endif
		1252
		1253	static int
		1254	eio__mtouch (eio_req *req)
		1255	{
		1256	void *mem = req->ptr2;
		1257	size_t len = req->size;
		1258	int flags = req->int1;
		1259
		1260	eio_page_align (&mem, &len);
		1261
		1262	{
		1263	intptr_t addr = (intptr_t)mem;
		1264	intptr_t end = addr + len;
		1265	intptr_t page = eio_pagesize ();
		1266
		1267	if (addr < end)
		1268	if (flags & EIO_MT_MODIFY) /* modify */
		1269	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1270	else
		1271	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1272	}
		1273
		1274	return 0;
		1275	}
		1276
		1277	/*****************************************************************************/
		1278	/* requests implemented outside eio_execute, because they are so large */
		1279
		1280	static void
		1281	eio__realpath (eio_req *req, etp_worker *self)
		1282	{
		1283	char *rel = req->ptr1;
		1284	char *res;
		1285	char *tmp1, *tmp2;
		1286	#if SYMLOOP_MAX > 32
		1287	int symlinks = SYMLOOP_MAX;
		1288	#else
		1289	int symlinks = 32;
		1290	#endif
		1291
		1292	req->result = -1;
		1293
		1294	errno = EINVAL;
		1295	if (!rel)
		1296	return;
		1297
		1298	errno = ENOENT;
		1299	if (!*rel)
		1300	return;
		1301
		1302	if (!req->ptr2)
		1303	{
		1304	X_LOCK (wrklock);
		1305	req->flags |= EIO_FLAG_PTR2_FREE;
		1306	X_UNLOCK (wrklock);
		1307	req->ptr2 = malloc (PATH_MAX * 3);
		1308
		1309	errno = ENOMEM;
		1310	if (!req->ptr2)
		1311	return;
		1312	}
		1313
		1314	res = req->ptr2;
		1315	tmp1 = res + PATH_MAX;
		1316	tmp2 = tmp1 + PATH_MAX;
		1317
		1318	#if 0 /* disabled, the musl way to do things is just too racy */
		1319	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1320	/* on linux we may be able to ask the kernel */
		1321	{
		1322	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1323
		1324	if (fd >= 0)
		1325	{
		1326	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1327	req->result = readlink (tmp1, res, PATH_MAX);
		1328	close (fd);
		1329
		1330	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1331
		1332	if (req->result > 0)
		1333	goto done;
		1334	}
		1335	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1336	return;
		1337	}
		1338	#endif
		1339	#endif
		1340
		1341	if (*rel != '/')
		1342	{
		1343	if (!getcwd (res, PATH_MAX))
		1344	return;
		1345
		1346	if (res [1]) /* only use if not / */
		1347	res += strlen (res);
		1348	}
		1349
		1350	while (*rel)
		1351	{
		1352	eio_ssize_t len, linklen;
		1353	char *beg = rel;
		1354
		1355	while (*rel && *rel != '/')
		1356	++rel;
		1357
		1358	len = rel - beg;
		1359
		1360	if (!len) /* skip slashes */
		1361	{
		1362	++rel;
		1363	continue;
		1364	}
		1365
		1366	if (beg [0] == '.')
		1367	{
		1368	if (len == 1)
		1369	continue; /* . - nop */
		1370
		1371	if (beg [1] == '.' && len == 2)
		1372	{
		1373	/* .. - back up one component, if possible */
		1374
		1375	while (res != req->ptr2)
		1376	if (*--res == '/')
		1377	break;
		1378
		1379	continue;
		1380	}
		1381	}
		1382
		1383	errno = ENAMETOOLONG;
		1384	if (res + 1 + len + 1 >= tmp1)
		1385	return;
		1386
		1387	/* copy one component */
		1388	*res = '/';
		1389	memcpy (res + 1, beg, len);
		1390
		1391	/* zero-terminate, for readlink */
		1392	res [len + 1] = 0;
		1393
		1394	/* now check if it's a symlink */
		1395	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1396
		1397	if (linklen < 0)
		1398	{
		1399	if (errno != EINVAL)
		1400	return;
		1401
		1402	/* it's a normal directory. hopefully */
		1403	res += len + 1;
		1404	}
		1405	else
		1406	{
		1407	/* yay, it was a symlink - build new path in tmp2 */
		1408	int rellen = strlen (rel);
		1409
		1410	errno = ENAMETOOLONG;
		1411	if (linklen + 1 + rellen >= PATH_MAX)
		1412	return;
		1413
		1414	errno = ELOOP;
		1415	if (!--symlinks)
		1416	return;
		1417
		1418	if (*tmp1 == '/')
		1419	res = req->ptr2; /* symlink resolves to an absolute path */
		1420
		1421	/* we need to be careful, as rel might point into tmp2 already */
		1422	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1423	tmp2 [linklen] = '/';
		1424	memcpy (tmp2, tmp1, linklen);
		1425
		1426	rel = tmp2;
		1427	}
		1428	}
		1429
		1430	/* special case for the lone root path */
		1431	if (res == req->ptr2)
		1432	*res++ = '/';
		1433
		1434	req->result = res - (char *)req->ptr2;
		1435
		1436	done:
		1437	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1438	}
		1439
1091	static signed char	1440	static signed char
1092	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1441	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1093	{	1442	{
1094	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */	1443	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1095	: a->inode < b->inode ? -1	1444	: a->inode < b->inode ? -1
…		…
1101		1450
1102	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1451	#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 */	1452	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1104		1453
1105	static void	1454	static void
1106	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1455	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1107	{	1456	{
1108	unsigned char bits [9 + sizeof (ino_t) * 8];	1457	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1109	unsigned char *bit = bits;	1458	unsigned char *bit = bits;
1110		1459
1111	assert (CHAR_BIT == 8);	1460	assert (CHAR_BIT == 8);
1112	assert (sizeof (eio_dirent) * 8 < 256);	1461	assert (sizeof (eio_dirent) * 8 < 256);
1113	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1462	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1115		1464
1116	if (size <= EIO_SORT_FAST)	1465	if (size <= EIO_SORT_FAST)
1117	return;	1466	return;
1118		1467
1119	/* first prepare an array of bits to test in our radix sort */	1468	/* 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 */	1469	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1121	/* inode_bits must contain all inodes ORed together */	1470	/* inode_bits must contain all inodes ORed together */
1122	/* which is used to skip bits that are 0 everywhere, which is very common */	1471	/* which is used to skip bits that are 0 everywhere, which is very common */
1123	{	1472	{
1124	ino_t endianness;	1473	eio_ino_t endianness;
1125	int i, j;	1474	int i, j;
1126		1475
1127	/* we store the byte offset of byte n into byte n of "endianness" */	1476	/* we store the byte offset of byte n into byte n of "endianness" */
1128	for (i = 0; i < sizeof (ino_t); ++i)	1477	for (i = 0; i < sizeof (eio_ino_t); ++i)
1129	((unsigned char *)&endianness)[i] = i;	1478	((unsigned char *)&endianness)[i] = i;
1130		1479
1131	*bit++ = 0;	1480	*bit++ = 0;
1132		1481
1133	for (i = 0; i < sizeof (ino_t); ++i)	1482	for (i = 0; i < sizeof (eio_ino_t); ++i)
1134	{	1483	{
1135	/* shifting off the byte offsets out of "endianness" */	1484	/* shifting off the byte offsets out of "endianness" */
1136	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1485	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1137	endianness >>= 8;	1486	endianness >>= 8;
1138		1487
1139	for (j = 0; j < 8; ++j)	1488	for (j = 0; j < 8; ++j)
1140	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1489	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1141	*bit++ = offs + j;	1490	*bit++ = offs + j;
1142	}	1491	}
1143		1492
1144	for (j = 0; j < 8; ++j)	1493	for (j = 0; j < 8; ++j)
1145	if (score_bits & (1 << j))	1494	if (score_bits & (1 << j))
1146	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1495	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1147	}	1496	}
1148		1497
1149	/* now actually do the sorting (a variant of MSD radix sort) */	1498	/* now actually do the sorting (a variant of MSD radix sort) */
1150	{	1499	{
1151	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1500	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1152	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1501	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1153	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1502	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1154	int stk_idx = 0;	1503	int stk_idx = 0;
1155		1504
1156	base_stk [stk_idx] = dents;	1505	base_stk [stk_idx] = dents;
1157	end_stk [stk_idx] = dents + size;	1506	end_stk [stk_idx] = dents + size;
1158	bit_stk [stk_idx] = bit - 1;	1507	bit_stk [stk_idx] = bit - 1;
…		…
1237	}	1586	}
1238	}	1587	}
1239	}	1588	}
1240		1589
1241	static void	1590	static void
1242	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1591	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1243	{	1592	{
1244	if (size <= 1)	1593	if (size <= 1)
1245	return; /* our insertion sort relies on size > 0 */	1594	return; /* our insertion sort relies on size > 0 */
1246		1595
1247	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1596	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1255		1604
1256	/* read a full directory */	1605	/* read a full directory */
1257	static void	1606	static void
1258	eio__scandir (eio_req *req, etp_worker *self)	1607	eio__scandir (eio_req *req, etp_worker *self)
1259	{	1608	{
1260	DIR *dirp;
1261	EIO_STRUCT_DIRENT *entp;
1262	char *name, *names;	1609	char *name, *names;
1263	int namesalloc = 4096;	1610	int namesalloc = 4096 - sizeof (void *) * 4;
1264	int namesoffs = 0;	1611	int namesoffs = 0;
1265	int flags = req->int1;	1612	int flags = req->int1;
1266	eio_dirent *dents = 0;	1613	eio_dirent *dents = 0;
1267	int dentalloc = 128;	1614	int dentalloc = 128;
1268	int dentoffs = 0;	1615	int dentoffs = 0;
1269	ino_t inode_bits = 0;	1616	eio_ino_t inode_bits = 0;
		1617	#ifdef _WIN32
		1618	HANDLE dirp;
		1619	WIN32_FIND_DATA entp;
		1620	#else
		1621	DIR *dirp;
		1622	EIO_STRUCT_DIRENT *entp;
		1623	#endif
1270		1624
1271	req->result = -1;	1625	req->result = -1;
1272		1626
1273	if (!(flags & EIO_READDIR_DENTS))	1627	if (!(flags & EIO_READDIR_DENTS))
1274	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1628	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1275		1629
1276	X_LOCK (wrklock);	1630	#ifdef _WIN32
1277	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1631	{
		1632	int len = strlen ((const char *)req->ptr1);
		1633	char *path = malloc (MAX_PATH);
		1634	const char *fmt;
		1635
		1636	if (!len)
		1637	fmt = "./*";
		1638	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1639	fmt = "%s*";
		1640	else
		1641	fmt = "%s/*";
		1642
		1643	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1644	dirp = FindFirstFile (path, &entp);
		1645	free (path);
		1646
		1647	if (dirp == INVALID_HANDLE_VALUE)
		1648	{
		1649	dirp = 0;
		1650
		1651	switch (GetLastError ())
		1652	{
		1653	case ERROR_FILE_NOT_FOUND:
		1654	req->result = 0;
		1655	break;
		1656
		1657	case ERROR_INVALID_NAME:
		1658	case ERROR_PATH_NOT_FOUND:
		1659	case ERROR_NO_MORE_FILES:
		1660	errno = ENOENT;
		1661	break;
		1662
		1663	case ERROR_NOT_ENOUGH_MEMORY:
		1664	errno = ENOMEM;
		1665	break;
		1666
		1667	default:
		1668	errno = EINVAL;
		1669	break;
		1670	}
		1671	}
		1672	}
		1673	#else
1278	self->dirp = dirp = opendir (req->ptr1);	1674	dirp = opendir (req->ptr1);
		1675	#endif
		1676
		1677	if (req->flags & EIO_FLAG_PTR1_FREE)
		1678	free (req->ptr1);
1279		1679
1280	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1680	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1281	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1681	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1282	req->ptr2 = names = malloc (namesalloc);	1682	req->ptr2 = names = malloc (namesalloc);
1283	X_UNLOCK (wrklock);
1284		1683
1285	if (dirp && names && (!flags || dents))	1684	if (dirp && names && (!flags || dents))
1286	for (;;)	1685	for (;;)
1287	{	1686	{
		1687	int done;
		1688
		1689	#ifdef _WIN32
		1690	done = !dirp;
		1691	#else
1288	errno = 0;	1692	errno = 0;
1289	entp = readdir (dirp);	1693	entp = readdir (dirp);
		1694	done = !entp;
		1695	#endif
1290		1696
1291	if (!entp)	1697	if (done)
1292	{	1698	{
		1699	#ifndef _WIN32
		1700	int old_errno = errno;
		1701	closedir (dirp);
		1702	errno = old_errno;
		1703
1293	if (errno)	1704	if (errno)
1294	break;	1705	break;
		1706	#endif
1295		1707
1296	/* sort etc. */	1708	/* sort etc. */
1297	req->int1 = flags;	1709	req->int1 = flags;
1298	req->result = dentoffs;	1710	req->result = dentoffs;
1299		1711
…		…
1328		1740
1329	break;	1741	break;
1330	}	1742	}
1331		1743
1332	/* now add the entry to our list(s) */	1744	/* now add the entry to our list(s) */
1333	name = entp->d_name;	1745	name = D_NAME (entp);
1334		1746
1335	/* skip . and .. entries */	1747	/* skip . and .. entries */
1336	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1748	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1337	{	1749	{
1338	int len = D_NAMLEN (entp) + 1;	1750	int len = D_NAMLEN (entp) + 1;
1339		1751
1340	while (expect_false (namesoffs + len > namesalloc))	1752	while (ecb_expect_false (namesoffs + len > namesalloc))
1341	{	1753	{
1342	namesalloc *= 2;	1754	namesalloc *= 2;
1343	X_LOCK (wrklock);
1344	req->ptr2 = names = realloc (names, namesalloc);	1755	req->ptr2 = names = realloc (names, namesalloc);
1345	X_UNLOCK (wrklock);
1346		1756
1347	if (!names)	1757	if (!names)
1348	break;	1758	break;
1349	}	1759	}
1350		1760
…		…
1352		1762
1353	if (dents)	1763	if (dents)
1354	{	1764	{
1355	struct eio_dirent *ent;	1765	struct eio_dirent *ent;
1356		1766
1357	if (expect_false (dentoffs == dentalloc))	1767	if (ecb_expect_false (dentoffs == dentalloc))
1358	{	1768	{
1359	dentalloc *= 2;	1769	dentalloc *= 2;
1360	X_LOCK (wrklock);
1361	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1770	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1362	X_UNLOCK (wrklock);
1363		1771
1364	if (!dents)	1772	if (!dents)
1365	break;	1773	break;
1366	}	1774	}
1367		1775
…		…
1447	if (EIO_CANCELLED (req))	1855	if (EIO_CANCELLED (req))
1448	{	1856	{
1449	errno = ECANCELED;	1857	errno = ECANCELED;
1450	break;	1858	break;
1451	}	1859	}
		1860
		1861	#ifdef _WIN32
		1862	if (!FindNextFile (dirp, &entp))
		1863	{
		1864	FindClose (dirp);
		1865	dirp = 0;
		1866	}
		1867	#endif
1452	}	1868	}
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	}	1869	}
1570		1870
1571	/*****************************************************************************/	1871	/*****************************************************************************/
1572		1872
1573	#define ALLOC(len) \	1873	#define ALLOC(len) \
…		…
1589	{	1889	{
1590	ETP_REQ *req;	1890	ETP_REQ *req;
1591	struct timespec ts;	1891	struct timespec ts;
1592	etp_worker *self = (etp_worker *)thr_arg;	1892	etp_worker *self = (etp_worker *)thr_arg;
1593		1893
1594	/* try to distribute timeouts somewhat randomly */	1894	/* try to distribute timeouts somewhat evenly */
1595	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1895	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1596		1896
1597	for (;;)	1897	for (;;)
1598	{	1898	{
		1899	ts.tv_sec = 0;
		1900
1599	X_LOCK (reqlock);	1901	X_LOCK (reqlock);
1600		1902
1601	for (;;)	1903	for (;;)
1602	{	1904	{
1603	self->req = req = reqq_shift (&req_queue);	1905	self->req = req = reqq_shift (&req_queue);
1604		1906
1605	if (req)	1907	if (req)
1606	break;	1908	break;
1607		1909
		1910	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1911	{
		1912	X_UNLOCK (reqlock);
		1913	X_LOCK (wrklock);
		1914	--started;
		1915	X_UNLOCK (wrklock);
		1916	goto quit;
		1917	}
		1918
1608	++idle;	1919	++idle;
1609		1920
1610	ts.tv_sec = time (0) + idle_timeout;	1921	if (idle <= max_idle)
1611	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1922	/* we are allowed to idle, so do so without any timeout */
		1923	X_COND_WAIT (reqwait, reqlock);
		1924	else
1612	{	1925	{
1613	if (idle > max_idle)	1926	/* initialise timeout once */
1614	{	1927	if (!ts.tv_sec)
1615	--idle;	1928	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		1929
1623	/* we are allowed to idle, so do so without any timeout */
1624	X_COND_WAIT (reqwait, reqlock);	1930	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1931	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1625	}	1932	}
1626		1933
1627	--idle;	1934	--idle;
1628	}	1935	}
1629		1936
…		…
1632	X_UNLOCK (reqlock);	1939	X_UNLOCK (reqlock);
1633		1940
1634	if (req->type < 0)	1941	if (req->type < 0)
1635	goto quit;	1942	goto quit;
1636		1943
1637	if (!EIO_CANCELLED (req))
1638	ETP_EXECUTE (self, req);	1944	ETP_EXECUTE (self, req);
1639		1945
1640	X_LOCK (reslock);	1946	X_LOCK (reslock);
1641		1947
1642	++npending;	1948	++npending;
1643		1949
…		…
1658	return 0;	1964	return 0;
1659	}	1965	}
1660		1966
1661	/*****************************************************************************/	1967	/*****************************************************************************/
1662		1968
		1969	int ecb_cold
1663	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1970	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1664	{	1971	{
		1972	#if !HAVE_PREADWRITE
		1973	X_MUTEX_CREATE (preadwritelock);
		1974	#endif
		1975
1665	return etp_init (want_poll, done_poll);	1976	return etp_init (want_poll, done_poll);
1666	}	1977	}
1667		1978
		1979	ecb_inline void
1668	static void eio_api_destroy (eio_req *req)	1980	eio_api_destroy (eio_req *req)
1669	{	1981	{
1670	free (req);	1982	free (req);
1671	}	1983	}
1672		1984
1673	#define REQ(rtype) \	1985	#define REQ(rtype) \
…		…
1692	{ \	2004	{ \
1693	eio_api_destroy (req); \	2005	eio_api_destroy (req); \
1694	return 0; \	2006	return 0; \
1695	}	2007	}
1696		2008
		2009	static void
1697	static void eio_execute (etp_worker *self, eio_req *req)	2010	eio_execute (etp_worker *self, eio_req *req)
1698	{	2011	{
		2012	if (ecb_expect_false (EIO_CANCELLED (req)))
		2013	{
		2014	req->result = -1;
		2015	req->errorno = ECANCELED;
		2016	return;
		2017	}
		2018
1699	switch (req->type)	2019	switch (req->type)
1700	{	2020	{
1701	case EIO_READ: ALLOC (req->size);	2021	case EIO_READ: ALLOC (req->size);
1702	req->result = req->offs >= 0	2022	req->result = req->offs >= 0
1703	? pread (req->int1, req->ptr2, req->size, req->offs)	2023	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1737	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2057	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1738	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2058	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1739	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2059	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;	2060	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1741		2061
		2062	case EIO_REALPATH: eio__realpath (req, self); break;
		2063
1742	case EIO_READLINK: ALLOC (PATH_MAX);	2064	case EIO_READLINK: ALLOC (PATH_MAX);
1743	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;	2065	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1744		2066
1745	case EIO_SYNC: req->result = 0; sync (); break;	2067	case EIO_SYNC: req->result = 0; sync (); break;
1746	case EIO_FSYNC: req->result = fsync (req->int1); break;	2068	case EIO_FSYNC: req->result = fsync (req->int1); break;
…		…
1748	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2070	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1749	case EIO_MTOUCH: req->result = eio__mtouch (req); break;	2071	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1750	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2072	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1751	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2073	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;	2074	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2075	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1753		2076
1754	case EIO_READDIR: eio__scandir (req, self); break;	2077	case EIO_READDIR: eio__scandir (req, self); break;
1755		2078
1756	case EIO_BUSY:	2079	case EIO_BUSY:
1757	#ifdef _WIN32	2080	#ifdef _WIN32
…		…
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)	2180	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1858	{	2181	{
1859	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2182	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1860	}	2183	}
1861		2184
		2185	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2186	{
		2187	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2188	}
		2189
1862	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2190	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1863	{	2191	{
1864	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2192	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1865	}	2193	}
1866		2194
…		…
1907	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2235	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1908	{	2236	{
1909	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2237	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1910	}	2238	}
1911		2239
1912	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2240	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1913	{	2241	{
1914	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2242	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1915	}	2243	}
1916		2244
1917	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2245	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)	2265	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1938	{	2266	{
1939	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2267	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1940	}	2268	}
1941		2269
1942	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2270	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1943	{	2271	{
1944	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2272	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1945	}	2273	}
1946		2274
1947	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2275	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1961	}	2289	}
1962		2290
1963	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2291	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1964	{	2292	{
1965	return eio__1path (EIO_READLINK, path, pri, cb, data);	2293	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2294	}
		2295
		2296	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2297	{
		2298	return eio__1path (EIO_REALPATH, path, pri, cb, data);
1966	}	2299	}
1967		2300
1968	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2301	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1969	{	2302	{
1970	return eio__1path (EIO_STAT, path, pri, cb, data);	2303	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)	2362	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2030	{	2363	{
2031	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2364	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2032	}	2365	}
2033		2366
2034	eio_req *eio_custom (void (*)(eio_req *) execute, int pri, eio_cb cb, void *data);	2367	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2035	{	2368	{
2036	REQ (EIO_CUSTOM); req->feed = execute; SEND;	2369	REQ (EIO_CUSTOM); req->feed = execute; SEND;
2037	}	2370	}
2038		2371
2039	#endif	2372	#endif
…		…
2050	#undef SEND	2383	#undef SEND
2051		2384
2052	/*****************************************************************************/	2385	/*****************************************************************************/
2053	/* grp functions */	2386	/* grp functions */
2054		2387
		2388	void
2055	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2389	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2056	{	2390	{
2057	grp->int2 = limit;	2391	grp->int2 = limit;
2058	grp->feed = feed;	2392	grp->feed = feed;
2059		2393
2060	grp_try_feed (grp);	2394	grp_try_feed (grp);
2061	}	2395	}
2062		2396
		2397	void
2063	void eio_grp_limit (eio_req *grp, int limit)	2398	eio_grp_limit (eio_req *grp, int limit)
2064	{	2399	{
2065	grp->int2 = limit;	2400	grp->int2 = limit;
2066		2401
2067	grp_try_feed (grp);	2402	grp_try_feed (grp);
2068	}	2403	}
2069		2404
		2405	void
2070	void eio_grp_add (eio_req *grp, eio_req *req)	2406	eio_grp_add (eio_req *grp, eio_req *req)
2071	{	2407	{
2072	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2408	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2073		2409
2074	grp->flags |= EIO_FLAG_GROUPADD;	2410	grp->flags |= EIO_FLAG_GROUPADD;
2075		2411
…		…
2086	}	2422	}
2087		2423
2088	/*****************************************************************************/	2424	/*****************************************************************************/
2089	/* misc garbage */	2425	/* misc garbage */
2090		2426
		2427	eio_ssize_t
2091	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2428	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2092	{	2429	{
2093	etp_worker wrk;	2430	etp_worker wrk;
2094	ssize_t ret;	2431	eio_ssize_t ret;
2095		2432
2096	wrk.dbuf = 0;	2433	wrk.dbuf = 0;
2097		2434
2098	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2435	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2099		2436

Diff Legend

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