ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/cvsroot/libeio/eio.c

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.70 by root, Fri Jun 10 06:55:10 2011 UTC vs.
Revision 1.94 by root, Mon Jul 18 02:59:58 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
…		…
155		233
156	/* buffer size for various temporary buffers */	234	/* buffer size for various temporary buffers */
157	#define EIO_BUFSIZE 65536	235	#define EIO_BUFSIZE 65536
158		236
159	#define dBUF \	237	#define dBUF \
160	char *eio_buf; \
161	ETP_WORKER_LOCK (self); \
162	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	238	char *eio_buf = malloc (EIO_BUFSIZE); \
163	ETP_WORKER_UNLOCK (self); \
164	errno = ENOMEM; \	239	errno = ENOMEM; \
165	if (!eio_buf) \	240	if (!eio_buf) \
166	return -1;	241	return -1
		242
		243	#define FUBd \
		244	free (eio_buf)
167		245
168	#define EIO_TICKS ((1000000 + 1023) >> 10)	246	#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		247
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;
…		…
190	static int eio_finish (eio_req *req);	255	static int eio_finish (eio_req *req);
191	#define ETP_FINISH(req) eio_finish (req)	256	#define ETP_FINISH(req) eio_finish (req)
192	static void eio_execute (struct etp_worker *self, eio_req *req);	257	static void eio_execute (struct etp_worker *self, eio_req *req);
193	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	258	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194		259
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	/*****************************************************************************/	260	/*****************************************************************************/
213		261
214	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	262	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215		263
216	/* calculate time difference in ~1/EIO_TICKS of a second */	264	/* calculate time difference in ~1/EIO_TICKS of a second */
		265	ecb_inline int
217	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	266	tvdiff (struct timeval *tv1, struct timeval *tv2)
218	{	267	{
219	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	268	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	269	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
221	}	270	}
222		271
…		…
243	/*	292	/*
244	* make our pread/pwrite emulation safe against themselves, but not against	293	* make our pread/pwrite emulation safe against themselves, but not against
245	* normal read/write by using a mutex. slows down execution a lot,	294	* normal read/write by using a mutex. slows down execution a lot,
246	* but that's your problem, not mine.	295	* but that's your problem, not mine.
247	*/	296	*/
248	static xmutex_t preadwritelock = X_MUTEX_INIT;	297	static xmutex_t preadwritelock;
249	#endif	298	#endif
250		299
251	typedef struct etp_worker	300	typedef struct etp_worker
252	{	301	{
253	/* locked by wrklock */	302	/* locked by wrklock */
…		…
256	xthread_t tid;	305	xthread_t tid;
257		306
258	/* locked by reslock, reqlock or wrklock */	307	/* locked by reslock, reqlock or wrklock */
259	ETP_REQ *req; /* currently processed request */	308	ETP_REQ *req; /* currently processed request */
260		309
		310	#ifdef ETP_WORKER_COMMON
261	ETP_WORKER_COMMON	311	ETP_WORKER_COMMON
		312	#endif
262	} etp_worker;	313	} etp_worker;
263		314
264	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	315	static etp_worker wrk_first; /* NOT etp */
265		316
266	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	317	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	318	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268		319
269	/* worker threads management */	320	/* worker threads management */
270		321
		322	static void ecb_cold
271	static void etp_worker_clear (etp_worker *wrk)	323	etp_worker_clear (etp_worker *wrk)
272	{	324	{
273	ETP_WORKER_CLEAR (wrk);
274	}	325	}
275		326
		327	static void ecb_cold
276	static void etp_worker_free (etp_worker *wrk)	328	etp_worker_free (etp_worker *wrk)
277	{	329	{
278	wrk->next->prev = wrk->prev;	330	wrk->next->prev = wrk->prev;
279	wrk->prev->next = wrk->next;	331	wrk->prev->next = wrk->next;
280		332
281	free (wrk);	333	free (wrk);
282	}	334	}
283		335
284	static unsigned int etp_nreqs (void)	336	static unsigned int
		337	etp_nreqs (void)
285	{	338	{
286	int retval;	339	int retval;
287	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	340	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288	retval = nreqs;	341	retval = nreqs;
289	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	342	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290	return retval;	343	return retval;
291	}	344	}
292		345
293	static unsigned int etp_nready (void)	346	static unsigned int
		347	etp_nready (void)
294	{	348	{
295	unsigned int retval;	349	unsigned int retval;
296		350
297	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	351	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298	retval = nready;	352	retval = nready;
299	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	353	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300		354
301	return retval;	355	return retval;
302	}	356	}
303		357
304	static unsigned int etp_npending (void)	358	static unsigned int
		359	etp_npending (void)
305	{	360	{
306	unsigned int retval;	361	unsigned int retval;
307		362
308	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	363	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309	retval = npending;	364	retval = npending;
310	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	365	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311		366
312	return retval;	367	return retval;
313	}	368	}
314		369
315	static unsigned int etp_nthreads (void)	370	static unsigned int
		371	etp_nthreads (void)
316	{	372	{
317	unsigned int retval;	373	unsigned int retval;
318		374
319	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	375	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320	retval = started;	376	retval = started;
…		…
334	} etp_reqq;	390	} etp_reqq;
335		391
336	static etp_reqq req_queue;	392	static etp_reqq req_queue;
337	static etp_reqq res_queue;	393	static etp_reqq res_queue;
338		394
		395	static void ecb_noinline ecb_cold
		396	reqq_init (etp_reqq *q)
		397	{
		398	int pri;
		399
		400	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		401	q->qs[pri] = q->qe[pri] = 0;
		402
		403	q->size = 0;
		404	}
		405
		406	static int ecb_noinline
339	static int reqq_push (etp_reqq *q, ETP_REQ *req)	407	reqq_push (etp_reqq *q, ETP_REQ *req)
340	{	408	{
341	int pri = req->pri;	409	int pri = req->pri;
342	req->next = 0;	410	req->next = 0;
343		411
344	if (q->qe[pri])	412	if (q->qe[pri])
…		…
350	q->qe[pri] = q->qs[pri] = req;	418	q->qe[pri] = q->qs[pri] = req;
351		419
352	return q->size++;	420	return q->size++;
353	}	421	}
354		422
		423	static ETP_REQ * ecb_noinline
355	static ETP_REQ *reqq_shift (etp_reqq *q)	424	reqq_shift (etp_reqq *q)
356	{	425	{
357	int pri;	426	int pri;
358		427
359	if (!q->size)	428	if (!q->size)
360	return 0;	429	return 0;
…		…
375	}	444	}
376		445
377	abort ();	446	abort ();
378	}	447	}
379		448
380	static void etp_thread_init (void)	449	static int ecb_cold
		450	etp_init (void (*want_poll)(void), void (*done_poll)(void))
381	{	451	{
382	X_MUTEX_CREATE (wrklock);	452	X_MUTEX_CREATE (wrklock);
383	X_MUTEX_CREATE (reslock);	453	X_MUTEX_CREATE (reslock);
384	X_MUTEX_CREATE (reqlock);	454	X_MUTEX_CREATE (reqlock);
385	X_COND_CREATE (reqwait);	455	X_COND_CREATE (reqwait);
386	}
387		456
388	static void etp_atfork_prepare (void)	457	reqq_init (&req_queue);
389	{	458	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		459
398	static void etp_atfork_parent (void)	460	wrk_first.next =
399	{	461	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		462
429	started = 0;	463	started = 0;
430	idle = 0;	464	idle = 0;
431	nreqs = 0;	465	nreqs = 0;
432	nready = 0;	466	nready = 0;
433	npending = 0;	467	npending = 0;
434		468
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;	469	want_poll_cb = want_poll;
453	done_poll_cb = done_poll;	470	done_poll_cb = done_poll;
454		471
455	return 0;	472	return 0;
456	}	473	}
457		474
458	X_THREAD_PROC (etp_proc);	475	X_THREAD_PROC (etp_proc);
459		476
		477	static void ecb_cold
460	static void etp_start_thread (void)	478	etp_start_thread (void)
461	{	479	{
462	etp_worker *wrk = calloc (1, sizeof (etp_worker));	480	etp_worker *wrk = calloc (1, sizeof (etp_worker));
463		481
464	/*TODO*/	482	/*TODO*/
465	assert (("unable to allocate worker thread data", wrk));	483	assert (("unable to allocate worker thread data", wrk));
…		…
478	free (wrk);	496	free (wrk);
479		497
480	X_UNLOCK (wrklock);	498	X_UNLOCK (wrklock);
481	}	499	}
482		500
		501	static void
483	static void etp_maybe_start_thread (void)	502	etp_maybe_start_thread (void)
484	{	503	{
485	if (expect_true (etp_nthreads () >= wanted))	504	if (ecb_expect_true (etp_nthreads () >= wanted))
486	return;	505	return;
487		506
488	/* todo: maybe use idle here, but might be less exact */	507	/* todo: maybe use idle here, but might be less exact */
489	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	508	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
490	return;	509	return;
491		510
492	etp_start_thread ();	511	etp_start_thread ();
493	}	512	}
494		513
		514	static void ecb_cold
495	static void etp_end_thread (void)	515	etp_end_thread (void)
496	{	516	{
497	eio_req *req = calloc (1, sizeof (eio_req));	517	eio_req *req = calloc (1, sizeof (eio_req));
498		518
499	req->type = -1;	519	req->type = -1;
500	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	520	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
507	X_LOCK (wrklock);	527	X_LOCK (wrklock);
508	--started;	528	--started;
509	X_UNLOCK (wrklock);	529	X_UNLOCK (wrklock);
510	}	530	}
511		531
512	static int etp_poll (void)	532	static int
		533	etp_poll (void)
513	{	534	{
514	unsigned int maxreqs;	535	unsigned int maxreqs;
515	unsigned int maxtime;	536	unsigned int maxtime;
516	struct timeval tv_start, tv_now;	537	struct timeval tv_start, tv_now;
517		538
…		…
547		568
548	X_LOCK (reqlock);	569	X_LOCK (reqlock);
549	--nreqs;	570	--nreqs;
550	X_UNLOCK (reqlock);	571	X_UNLOCK (reqlock);
551		572
552	if (expect_false (req->type == EIO_GROUP && req->size))	573	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
553	{	574	{
554	req->int1 = 1; /* mark request as delayed */	575	req->int1 = 1; /* mark request as delayed */
555	continue;	576	continue;
556	}	577	}
557	else	578	else
558	{	579	{
559	int res = ETP_FINISH (req);	580	int res = ETP_FINISH (req);
560	if (expect_false (res))	581	if (ecb_expect_false (res))
561	return res;	582	return res;
562	}	583	}
563		584
564	if (expect_false (maxreqs && !--maxreqs))	585	if (ecb_expect_false (maxreqs && !--maxreqs))
565	break;	586	break;
566		587
567	if (maxtime)	588	if (maxtime)
568	{	589	{
569	gettimeofday (&tv_now, 0);	590	gettimeofday (&tv_now, 0);
…		…
575		596
576	errno = EAGAIN;	597	errno = EAGAIN;
577	return -1;	598	return -1;
578	}	599	}
579		600
		601	static void
580	static void etp_cancel (ETP_REQ *req)	602	etp_cancel (ETP_REQ *req)
581	{	603	{
582	X_LOCK (wrklock);	604	req->cancelled = 1;
583	req->flags |= EIO_FLAG_CANCELLED;
584	X_UNLOCK (wrklock);
585		605
586	eio_grp_cancel (req);	606	eio_grp_cancel (req);
587	}	607	}
588		608
		609	static void
589	static void etp_submit (ETP_REQ *req)	610	etp_submit (ETP_REQ *req)
590	{	611	{
591	req->pri -= ETP_PRI_MIN;	612	req->pri -= ETP_PRI_MIN;
592		613
593	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	614	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
594	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	615	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595		616
596	if (expect_false (req->type == EIO_GROUP))	617	if (ecb_expect_false (req->type == EIO_GROUP))
597	{	618	{
598	/* I hope this is worth it :/ */	619	/* I hope this is worth it :/ */
599	X_LOCK (reqlock);	620	X_LOCK (reqlock);
600	++nreqs;	621	++nreqs;
601	X_UNLOCK (reqlock);	622	X_UNLOCK (reqlock);
…		…
620		641
621	etp_maybe_start_thread ();	642	etp_maybe_start_thread ();
622	}	643	}
623	}	644	}
624		645
		646	static void ecb_cold
625	static void etp_set_max_poll_time (double nseconds)	647	etp_set_max_poll_time (double nseconds)
626	{	648	{
627	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	649	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628	max_poll_time = nseconds * EIO_TICKS;	650	max_poll_time = nseconds * EIO_TICKS;
629	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	651	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630	}	652	}
631		653
		654	static void ecb_cold
632	static void etp_set_max_poll_reqs (unsigned int maxreqs)	655	etp_set_max_poll_reqs (unsigned int maxreqs)
633	{	656	{
634	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	657	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635	max_poll_reqs = maxreqs;	658	max_poll_reqs = maxreqs;
636	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	659	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637	}	660	}
638		661
		662	static void ecb_cold
639	static void etp_set_max_idle (unsigned int nthreads)	663	etp_set_max_idle (unsigned int nthreads)
640	{	664	{
641	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	665	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642	max_idle = nthreads;	666	max_idle = nthreads;
643	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	667	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644	}	668	}
645		669
		670	static void ecb_cold
646	static void etp_set_idle_timeout (unsigned int seconds)	671	etp_set_idle_timeout (unsigned int seconds)
647	{	672	{
648	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	673	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649	idle_timeout = seconds;	674	idle_timeout = seconds;
650	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	675	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651	}	676	}
652		677
		678	static void ecb_cold
653	static void etp_set_min_parallel (unsigned int nthreads)	679	etp_set_min_parallel (unsigned int nthreads)
654	{	680	{
655	if (wanted < nthreads)	681	if (wanted < nthreads)
656	wanted = nthreads;	682	wanted = nthreads;
657	}	683	}
658		684
		685	static void ecb_cold
659	static void etp_set_max_parallel (unsigned int nthreads)	686	etp_set_max_parallel (unsigned int nthreads)
660	{	687	{
661	if (wanted > nthreads)	688	if (wanted > nthreads)
662	wanted = nthreads;	689	wanted = nthreads;
663		690
664	while (started > wanted)	691	while (started > wanted)
665	etp_end_thread ();	692	etp_end_thread ();
666	}	693	}
667		694
668	/*****************************************************************************/	695	/*****************************************************************************/
669		696
		697	static void
670	static void grp_try_feed (eio_req *grp)	698	grp_try_feed (eio_req *grp)
671	{	699	{
672	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	700	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673	{	701	{
674	grp->flags &= ~EIO_FLAG_GROUPADD;	702	grp->flags &= ~EIO_FLAG_GROUPADD;
675		703
…		…
682	break;	710	break;
683	}	711	}
684	}	712	}
685	}	713	}
686		714
		715	static int
687	static int grp_dec (eio_req *grp)	716	grp_dec (eio_req *grp)
688	{	717	{
689	--grp->size;	718	--grp->size;
690		719
691	/* call feeder, if applicable */	720	/* call feeder, if applicable */
692	grp_try_feed (grp);	721	grp_try_feed (grp);
…		…
696	return eio_finish (grp);	725	return eio_finish (grp);
697	else	726	else
698	return 0;	727	return 0;
699	}	728	}
700		729
		730	static void
701	void eio_destroy (eio_req *req)	731	eio_destroy (eio_req *req)
702	{	732	{
703	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	733	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	734	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705		735
706	EIO_DESTROY (req);	736	EIO_DESTROY (req);
707	}	737	}
708		738
		739	static int
709	static int eio_finish (eio_req *req)	740	eio_finish (eio_req *req)
710	{	741	{
711	int res = EIO_FINISH (req);	742	int res = EIO_FINISH (req);
712		743
713	if (req->grp)	744	if (req->grp)
714	{	745	{
…		…
722	if (grp->grp_first == req)	753	if (grp->grp_first == req)
723	grp->grp_first = req->grp_next;	754	grp->grp_first = req->grp_next;
724		755
725	res2 = grp_dec (grp);	756	res2 = grp_dec (grp);
726		757
727	if (!res && res2)	758	if (!res)
728	res = res2;	759	res = res2;
729	}	760	}
730		761
731	eio_destroy (req);	762	eio_destroy (req);
732		763
733	return res;	764	return res;
734	}	765	}
735		766
		767	void
736	void eio_grp_cancel (eio_req *grp)	768	eio_grp_cancel (eio_req *grp)
737	{	769	{
738	for (grp = grp->grp_first; grp; grp = grp->grp_next)	770	for (grp = grp->grp_first; grp; grp = grp->grp_next)
739	eio_cancel (grp);	771	eio_cancel (grp);
740	}	772	}
741		773
		774	void
742	void eio_cancel (eio_req *req)	775	eio_cancel (eio_req *req)
743	{	776	{
744	etp_cancel (req);	777	etp_cancel (req);
745	}	778	}
746		779
		780	void
747	void eio_submit (eio_req *req)	781	eio_submit (eio_req *req)
748	{	782	{
749	etp_submit (req);	783	etp_submit (req);
750	}	784	}
751		785
752	unsigned int eio_nreqs (void)	786	unsigned int
		787	eio_nreqs (void)
753	{	788	{
754	return etp_nreqs ();	789	return etp_nreqs ();
755	}	790	}
756		791
757	unsigned int eio_nready (void)	792	unsigned int
		793	eio_nready (void)
758	{	794	{
759	return etp_nready ();	795	return etp_nready ();
760	}	796	}
761		797
762	unsigned int eio_npending (void)	798	unsigned int
		799	eio_npending (void)
763	{	800	{
764	return etp_npending ();	801	return etp_npending ();
765	}	802	}
766		803
767	unsigned int eio_nthreads (void)	804	unsigned int ecb_cold
		805	eio_nthreads (void)
768	{	806	{
769	return etp_nthreads ();	807	return etp_nthreads ();
770	}	808	}
771		809
		810	void ecb_cold
772	void eio_set_max_poll_time (double nseconds)	811	eio_set_max_poll_time (double nseconds)
773	{	812	{
774	etp_set_max_poll_time (nseconds);	813	etp_set_max_poll_time (nseconds);
775	}	814	}
776		815
		816	void ecb_cold
777	void eio_set_max_poll_reqs (unsigned int maxreqs)	817	eio_set_max_poll_reqs (unsigned int maxreqs)
778	{	818	{
779	etp_set_max_poll_reqs (maxreqs);	819	etp_set_max_poll_reqs (maxreqs);
780	}	820	}
781		821
		822	void ecb_cold
782	void eio_set_max_idle (unsigned int nthreads)	823	eio_set_max_idle (unsigned int nthreads)
783	{	824	{
784	etp_set_max_idle (nthreads);	825	etp_set_max_idle (nthreads);
785	}	826	}
786		827
		828	void ecb_cold
787	void eio_set_idle_timeout (unsigned int seconds)	829	eio_set_idle_timeout (unsigned int seconds)
788	{	830	{
789	etp_set_idle_timeout (seconds);	831	etp_set_idle_timeout (seconds);
790	}	832	}
791		833
		834	void ecb_cold
792	void eio_set_min_parallel (unsigned int nthreads)	835	eio_set_min_parallel (unsigned int nthreads)
793	{	836	{
794	etp_set_min_parallel (nthreads);	837	etp_set_min_parallel (nthreads);
795	}	838	}
796		839
		840	void ecb_cold
797	void eio_set_max_parallel (unsigned int nthreads)	841	eio_set_max_parallel (unsigned int nthreads)
798	{	842	{
799	etp_set_max_parallel (nthreads);	843	etp_set_max_parallel (nthreads);
800	}	844	}
801		845
802	int eio_poll (void)	846	int eio_poll (void)
…		…
811	# undef pread	855	# undef pread
812	# undef pwrite	856	# undef pwrite
813	# define pread eio__pread	857	# define pread eio__pread
814	# define pwrite eio__pwrite	858	# define pwrite eio__pwrite
815		859
816	static ssize_t	860	static eio_ssize_t
817	eio__pread (int fd, void *buf, size_t count, off_t offset)	861	eio__pread (int fd, void *buf, size_t count, off_t offset)
818	{	862	{
819	ssize_t res;	863	eio_ssize_t res;
820	off_t ooffset;	864	off_t ooffset;
821		865
822	X_LOCK (preadwritelock);	866	X_LOCK (preadwritelock);
823	ooffset = lseek (fd, 0, SEEK_CUR);	867	ooffset = lseek (fd, 0, SEEK_CUR);
824	lseek (fd, offset, SEEK_SET);	868	lseek (fd, offset, SEEK_SET);
…		…
827	X_UNLOCK (preadwritelock);	871	X_UNLOCK (preadwritelock);
828		872
829	return res;	873	return res;
830	}	874	}
831		875
832	static ssize_t	876	static eio_ssize_t
833	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	877	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834	{	878	{
835	ssize_t res;	879	eio_ssize_t res;
836	off_t ooffset;	880	off_t ooffset;
837		881
838	X_LOCK (preadwritelock);	882	X_LOCK (preadwritelock);
839	ooffset = lseek (fd, 0, SEEK_CUR);	883	ooffset = lseek (fd, 0, SEEK_CUR);
840	lseek (fd, offset, SEEK_SET);	884	lseek (fd, offset, SEEK_SET);
…		…
872	#ifndef HAVE_FUTIMES	916	#ifndef HAVE_FUTIMES
873		917
874	# undef futimes	918	# undef futimes
875	# define futimes(fd,times) eio__futimes (fd, times)	919	# define futimes(fd,times) eio__futimes (fd, times)
876		920
		921	static int
877	static int eio__futimes (int fd, const struct timeval tv[2])	922	eio__futimes (int fd, const struct timeval tv[2])
878	{	923	{
879	errno = ENOSYS;	924	errno = ENOSYS;
880	return -1;	925	return -1;
881	}	926	}
882		927
…		…
886	# undef fdatasync	931	# undef fdatasync
887	# define fdatasync(fd) fsync (fd)	932	# define fdatasync(fd) fsync (fd)
888	#endif	933	#endif
889		934
890	/* sync_file_range always needs emulation */	935	/* sync_file_range always needs emulation */
891	int	936	static int
892	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	937	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893	{	938	{
894	#if HAVE_SYNC_FILE_RANGE	939	#if HAVE_SYNC_FILE_RANGE
895	int res;	940	int res;
896		941
…		…
913	/* even though we could play tricks with the flags, it's better to always	958	/* even though we could play tricks with the flags, it's better to always
914	* call fdatasync, as that matches the expectation of its users best */	959	* call fdatasync, as that matches the expectation of its users best */
915	return fdatasync (fd);	960	return fdatasync (fd);
916	}	961	}
917		962
		963	static int
		964	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		965	{
		966	#if HAVE_FALLOCATE
		967	return fallocate (fd, mode, offset, len);
		968	#else
		969	errno = ENOSYS;
		970	return -1;
		971	#endif
		972	}
		973
918	#if !HAVE_READAHEAD	974	#if !HAVE_READAHEAD
919	# undef readahead	975	# undef readahead
920	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	976	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921		977
922	static ssize_t	978	static eio_ssize_t
923	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	979	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
924	{	980	{
925	size_t todo = count;	981	size_t todo = count;
926	dBUF;	982	dBUF;
927		983
…		…
932	pread (fd, eio_buf, len, offset);	988	pread (fd, eio_buf, len, offset);
933	offset += len;	989	offset += len;
934	todo -= len;	990	todo -= len;
935	}	991	}
936		992
		993	FUBd;
		994
937	errno = 0;	995	errno = 0;
938	return count;	996	return count;
939	}	997	}
940		998
941	#endif	999	#endif
942		1000
943	/* sendfile always needs emulation */	1001	/* sendfile always needs emulation */
944	static ssize_t	1002	static eio_ssize_t
945	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1003	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
946	{	1004	{
947	ssize_t written = 0;	1005	eio_ssize_t written = 0;
948	ssize_t res;	1006	eio_ssize_t res;
949		1007
950	if (!count)	1008	if (!count)
951	return 0;	1009	return 0;
952		1010
953	for (;;)	1011	for (;;)
954	{	1012	{
		1013	#ifdef __APPLE__
		1014	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1015	#endif
955	#if HAVE_SENDFILE	1016	#if HAVE_SENDFILE
956	# if __linux	1017	# if __linux
957	off_t soffset = offset;	1018	off_t soffset = offset;
958	res = sendfile (ofd, ifd, &soffset, count);	1019	res = sendfile (ofd, ifd, &soffset, count);
959		1020
…		…
1002	if (res < 0 && sbytes)	1063	if (res < 0 && sbytes)
1003	res = sbytes;	1064	res = sbytes;
1004		1065
1005	# endif	1066	# endif
1006		1067
1007	#elif defined (_WIN32)	1068	#elif defined (_WIN32) && 0
1008	/* does not work, just for documentation of what would need to be done */	1069	/* 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, */	1070	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
1010	/* libeio guarantees that the file offset does not change, and windows */	1071	/* 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 */	1072	/* has no way to get an independent handle to the same file description */
1012	HANDLE h = TO_SOCKET (ifd);	1073	HANDLE h = TO_SOCKET (ifd);
…		…
1045	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1106	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1046	/* BSDs */	1107	/* BSDs */
1047	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1108	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1048	|| errno == ENOTSUP	1109	|| errno == ENOTSUP
1049	#endif	1110	#endif
		1111	#ifdef EOPNOTSUPP /* windows */
1050	|| errno == EOPNOTSUPP /* BSDs */	1112	|| errno == EOPNOTSUPP /* BSDs */
		1113	#endif
1051	#if __solaris	1114	#if __solaris
1052	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1115	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1053	#endif	1116	#endif
1054	)	1117	)
1055	)	1118	)
…		…
1059		1122
1060	res = 0;	1123	res = 0;
1061		1124
1062	while (count)	1125	while (count)
1063	{	1126	{
1064	ssize_t cnt;	1127	eio_ssize_t cnt;
1065		1128
1066	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1129	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1067		1130
1068	if (cnt <= 0)	1131	if (cnt <= 0)
1069	{	1132	{
…		…
1081		1144
1082	offset += cnt;	1145	offset += cnt;
1083	res += cnt;	1146	res += cnt;
1084	count -= cnt;	1147	count -= cnt;
1085	}	1148	}
		1149
		1150	FUBd;
1086	}	1151	}
1087		1152
1088	return res;	1153	return res;
		1154	}
		1155
		1156	#ifdef PAGESIZE
		1157	# define eio_pagesize() PAGESIZE
		1158	#else
		1159	static intptr_t
		1160	eio_pagesize (void)
		1161	{
		1162	static intptr_t page;
		1163
		1164	if (!page)
		1165	page = sysconf (_SC_PAGESIZE);
		1166
		1167	return page;
		1168	}
		1169	#endif
		1170
		1171	static void
		1172	eio_page_align (void **addr, size_t *length)
		1173	{
		1174	intptr_t mask = eio_pagesize () - 1;
		1175
		1176	/* round down addr */
		1177	intptr_t adj = mask & (intptr_t)*addr;
		1178
		1179	*addr = (void *)((intptr_t)*addr - adj);
		1180	*length += adj;
		1181
		1182	/* round up length */
		1183	*length = (*length + mask) & ~mask;
		1184	}
		1185
		1186	#if !_POSIX_MEMLOCK
		1187	# define eio__mlockall(a) EIO_ENOSYS ()
		1188	#else
		1189
		1190	static int
		1191	eio__mlockall (int flags)
		1192	{
		1193	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1194	extern int mallopt (int, int);
		1195	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1196	#endif
		1197
		1198	if (EIO_MCL_CURRENT != MCL_CURRENT
		1199	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1200	{
		1201	flags = 0
		1202	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1203	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1204	}
		1205
		1206	return mlockall (flags);
		1207	}
		1208	#endif
		1209
		1210	#if !_POSIX_MEMLOCK_RANGE
		1211	# define eio__mlock(a,b) EIO_ENOSYS ()
		1212	#else
		1213
		1214	static int
		1215	eio__mlock (void *addr, size_t length)
		1216	{
		1217	eio_page_align (&addr, &length);
		1218
		1219	return mlock (addr, length);
		1220	}
		1221
		1222	#endif
		1223
		1224	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1225	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1226	#else
		1227
		1228	static int
		1229	eio__msync (void *mem, size_t len, int flags)
		1230	{
		1231	eio_page_align (&mem, &len);
		1232
		1233	if (EIO_MS_ASYNC != MS_SYNC
		1234	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1235	|| EIO_MS_SYNC != MS_SYNC)
		1236	{
		1237	flags = 0
		1238	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1239	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1240	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1241	}
		1242
		1243	return msync (mem, len, flags);
		1244	}
		1245
		1246	#endif
		1247
		1248	static int
		1249	eio__mtouch (eio_req *req)
		1250	{
		1251	void *mem = req->ptr2;
		1252	size_t len = req->size;
		1253	int flags = req->int1;
		1254
		1255	eio_page_align (&mem, &len);
		1256
		1257	{
		1258	intptr_t addr = (intptr_t)mem;
		1259	intptr_t end = addr + len;
		1260	intptr_t page = eio_pagesize ();
		1261
		1262	if (addr < end)
		1263	if (flags & EIO_MT_MODIFY) /* modify */
		1264	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1265	else
		1266	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1267	}
		1268
		1269	return 0;
		1270	}
		1271
		1272	/*****************************************************************************/
		1273	/* requests implemented outside eio_execute, because they are so large */
		1274
		1275	static void
		1276	eio__realpath (eio_req *req, etp_worker *self)
		1277	{
		1278	char *rel = req->ptr1;
		1279	char *res;
		1280	char *tmp1, *tmp2;
		1281	#if SYMLOOP_MAX > 32
		1282	int symlinks = SYMLOOP_MAX;
		1283	#else
		1284	int symlinks = 32;
		1285	#endif
		1286
		1287	req->result = -1;
		1288
		1289	errno = EINVAL;
		1290	if (!rel)
		1291	return;
		1292
		1293	errno = ENOENT;
		1294	if (!*rel)
		1295	return;
		1296
		1297	if (!req->ptr2)
		1298	{
		1299	X_LOCK (wrklock);
		1300	req->flags |= EIO_FLAG_PTR2_FREE;
		1301	X_UNLOCK (wrklock);
		1302	req->ptr2 = malloc (PATH_MAX * 3);
		1303
		1304	errno = ENOMEM;
		1305	if (!req->ptr2)
		1306	return;
		1307	}
		1308
		1309	res = req->ptr2;
		1310	tmp1 = res + PATH_MAX;
		1311	tmp2 = tmp1 + PATH_MAX;
		1312
		1313	#if 0 /* disabled, the musl way to do things is just too racy */
		1314	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1315	/* on linux we may be able to ask the kernel */
		1316	{
		1317	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1318
		1319	if (fd >= 0)
		1320	{
		1321	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1322	req->result = readlink (tmp1, res, PATH_MAX);
		1323	close (fd);
		1324
		1325	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1326
		1327	if (req->result > 0)
		1328	goto done;
		1329	}
		1330	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1331	return;
		1332	}
		1333	#endif
		1334	#endif
		1335
		1336	if (*rel != '/')
		1337	{
		1338	if (!getcwd (res, PATH_MAX))
		1339	return;
		1340
		1341	if (res [1]) /* only use if not / */
		1342	res += strlen (res);
		1343	}
		1344
		1345	while (*rel)
		1346	{
		1347	eio_ssize_t len, linklen;
		1348	char *beg = rel;
		1349
		1350	while (*rel && *rel != '/')
		1351	++rel;
		1352
		1353	len = rel - beg;
		1354
		1355	if (!len) /* skip slashes */
		1356	{
		1357	++rel;
		1358	continue;
		1359	}
		1360
		1361	if (beg [0] == '.')
		1362	{
		1363	if (len == 1)
		1364	continue; /* . - nop */
		1365
		1366	if (beg [1] == '.' && len == 2)
		1367	{
		1368	/* .. - back up one component, if possible */
		1369
		1370	while (res != req->ptr2)
		1371	if (*--res == '/')
		1372	break;
		1373
		1374	continue;
		1375	}
		1376	}
		1377
		1378	errno = ENAMETOOLONG;
		1379	if (res + 1 + len + 1 >= tmp1)
		1380	return;
		1381
		1382	/* copy one component */
		1383	*res = '/';
		1384	memcpy (res + 1, beg, len);
		1385
		1386	/* zero-terminate, for readlink */
		1387	res [len + 1] = 0;
		1388
		1389	/* now check if it's a symlink */
		1390	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1391
		1392	if (linklen < 0)
		1393	{
		1394	if (errno != EINVAL)
		1395	return;
		1396
		1397	/* it's a normal directory. hopefully */
		1398	res += len + 1;
		1399	}
		1400	else
		1401	{
		1402	/* yay, it was a symlink - build new path in tmp2 */
		1403	int rellen = strlen (rel);
		1404
		1405	errno = ENAMETOOLONG;
		1406	if (linklen + 1 + rellen >= PATH_MAX)
		1407	return;
		1408
		1409	errno = ELOOP;
		1410	if (!--symlinks)
		1411	return;
		1412
		1413	if (*tmp1 == '/')
		1414	res = req->ptr2; /* symlink resolves to an absolute path */
		1415
		1416	/* we need to be careful, as rel might point into tmp2 already */
		1417	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1418	tmp2 [linklen] = '/';
		1419	memcpy (tmp2, tmp1, linklen);
		1420
		1421	rel = tmp2;
		1422	}
		1423	}
		1424
		1425	/* special case for the lone root path */
		1426	if (res == req->ptr2)
		1427	*res++ = '/';
		1428
		1429	req->result = res - (char *)req->ptr2;
		1430
		1431	done:
		1432	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1089	}	1433	}
1090		1434
1091	static signed char	1435	static signed char
1092	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1436	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1093	{	1437	{
…		…
1101		1445
1102	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1446	#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 */	1447	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1104		1448
1105	static void	1449	static void
1106	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1450	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1107	{	1451	{
1108	unsigned char bits [9 + sizeof (ino_t) * 8];	1452	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1109	unsigned char *bit = bits;	1453	unsigned char *bit = bits;
1110		1454
1111	assert (CHAR_BIT == 8);	1455	assert (CHAR_BIT == 8);
1112	assert (sizeof (eio_dirent) * 8 < 256);	1456	assert (sizeof (eio_dirent) * 8 < 256);
1113	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1457	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1115		1459
1116	if (size <= EIO_SORT_FAST)	1460	if (size <= EIO_SORT_FAST)
1117	return;	1461	return;
1118		1462
1119	/* first prepare an array of bits to test in our radix sort */	1463	/* 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 */	1464	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1121	/* inode_bits must contain all inodes ORed together */	1465	/* inode_bits must contain all inodes ORed together */
1122	/* which is used to skip bits that are 0 everywhere, which is very common */	1466	/* which is used to skip bits that are 0 everywhere, which is very common */
1123	{	1467	{
1124	ino_t endianness;	1468	eio_ino_t endianness;
1125	int i, j;	1469	int i, j;
1126		1470
1127	/* we store the byte offset of byte n into byte n of "endianness" */	1471	/* we store the byte offset of byte n into byte n of "endianness" */
1128	for (i = 0; i < sizeof (ino_t); ++i)	1472	for (i = 0; i < sizeof (eio_ino_t); ++i)
1129	((unsigned char *)&endianness)[i] = i;	1473	((unsigned char *)&endianness)[i] = i;
1130		1474
1131	*bit++ = 0;	1475	*bit++ = 0;
1132		1476
1133	for (i = 0; i < sizeof (ino_t); ++i)	1477	for (i = 0; i < sizeof (eio_ino_t); ++i)
1134	{	1478	{
1135	/* shifting off the byte offsets out of "endianness" */	1479	/* shifting off the byte offsets out of "endianness" */
1136	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1480	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1137	endianness >>= 8;	1481	endianness >>= 8;
1138		1482
1139	for (j = 0; j < 8; ++j)	1483	for (j = 0; j < 8; ++j)
1140	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1484	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1141	*bit++ = offs + j;	1485	*bit++ = offs + j;
1142	}	1486	}
1143		1487
1144	for (j = 0; j < 8; ++j)	1488	for (j = 0; j < 8; ++j)
1145	if (score_bits & (1 << j))	1489	if (score_bits & (1 << j))
1146	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1490	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1147	}	1491	}
1148		1492
1149	/* now actually do the sorting (a variant of MSD radix sort) */	1493	/* now actually do the sorting (a variant of MSD radix sort) */
1150	{	1494	{
1151	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1495	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1152	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1496	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1153	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1497	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1154	int stk_idx = 0;	1498	int stk_idx = 0;
1155		1499
1156	base_stk [stk_idx] = dents;	1500	base_stk [stk_idx] = dents;
1157	end_stk [stk_idx] = dents + size;	1501	end_stk [stk_idx] = dents + size;
1158	bit_stk [stk_idx] = bit - 1;	1502	bit_stk [stk_idx] = bit - 1;
…		…
1237	}	1581	}
1238	}	1582	}
1239	}	1583	}
1240		1584
1241	static void	1585	static void
1242	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1586	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1243	{	1587	{
1244	if (size <= 1)	1588	if (size <= 1)
1245	return; /* our insertion sort relies on size > 0 */	1589	return; /* our insertion sort relies on size > 0 */
1246		1590
1247	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1591	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1255		1599
1256	/* read a full directory */	1600	/* read a full directory */
1257	static void	1601	static void
1258	eio__scandir (eio_req *req, etp_worker *self)	1602	eio__scandir (eio_req *req, etp_worker *self)
1259	{	1603	{
1260	DIR *dirp;
1261	EIO_STRUCT_DIRENT *entp;
1262	char *name, *names;	1604	char *name, *names;
1263	int namesalloc = 4096;	1605	int namesalloc = 4096 - sizeof (void *) * 4;
1264	int namesoffs = 0;	1606	int namesoffs = 0;
1265	int flags = req->int1;	1607	int flags = req->int1;
1266	eio_dirent *dents = 0;	1608	eio_dirent *dents = 0;
1267	int dentalloc = 128;	1609	int dentalloc = 128;
1268	int dentoffs = 0;	1610	int dentoffs = 0;
1269	ino_t inode_bits = 0;	1611	eio_ino_t inode_bits = 0;
		1612	#ifdef _WIN32
		1613	HANDLE dirp;
		1614	WIN32_FIND_DATA entp;
		1615	#else
		1616	DIR *dirp;
		1617	EIO_STRUCT_DIRENT *entp;
		1618	#endif
1270		1619
1271	req->result = -1;	1620	req->result = -1;
1272		1621
1273	if (!(flags & EIO_READDIR_DENTS))	1622	if (!(flags & EIO_READDIR_DENTS))
1274	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1623	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1275		1624
1276	X_LOCK (wrklock);	1625	#ifdef _WIN32
1277	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1626	{
		1627	int len = strlen ((const char *)req->ptr1);
		1628	char *path = malloc (MAX_PATH);
		1629	const char *fmt;
		1630
		1631	if (!len)
		1632	fmt = "./*";
		1633	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1634	fmt = "%s*";
		1635	else
		1636	fmt = "%s/*";
		1637
		1638	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1639	dirp = FindFirstFile (path, &entp);
		1640	free (path);
		1641
		1642	if (dirp == INVALID_HANDLE_VALUE)
		1643	{
		1644	dirp = 0;
		1645
		1646	switch (GetLastError ())
		1647	{
		1648	case ERROR_FILE_NOT_FOUND:
		1649	req->result = 0;
		1650	break;
		1651
		1652	case ERROR_INVALID_NAME:
		1653	case ERROR_PATH_NOT_FOUND:
		1654	case ERROR_NO_MORE_FILES:
		1655	errno = ENOENT;
		1656	break;
		1657
		1658	case ERROR_NOT_ENOUGH_MEMORY:
		1659	errno = ENOMEM;
		1660	break;
		1661
		1662	default:
		1663	errno = EINVAL;
		1664	break;
		1665	}
		1666	}
		1667	}
		1668	#else
1278	self->dirp = dirp = opendir (req->ptr1);	1669	dirp = opendir (req->ptr1);
		1670	#endif
		1671
		1672	if (req->flags & EIO_FLAG_PTR1_FREE)
		1673	free (req->ptr1);
1279		1674
1280	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1675	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1281	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1676	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1282	req->ptr2 = names = malloc (namesalloc);	1677	req->ptr2 = names = malloc (namesalloc);
1283	X_UNLOCK (wrklock);
1284		1678
1285	if (dirp && names && (!flags || dents))	1679	if (dirp && names && (!flags || dents))
1286	for (;;)	1680	for (;;)
1287	{	1681	{
		1682	int done;
		1683
		1684	#ifdef _WIN32
		1685	done = !dirp;
		1686	#else
1288	errno = 0;	1687	errno = 0;
1289	entp = readdir (dirp);	1688	entp = readdir (dirp);
		1689	done = !entp;
		1690	#endif
1290		1691
1291	if (!entp)	1692	if (done)
1292	{	1693	{
		1694	#ifndef _WIN32
		1695	int old_errno = errno;
		1696	closedir (dirp);
		1697	errno = old_errno;
		1698
1293	if (errno)	1699	if (errno)
1294	break;	1700	break;
		1701	#endif
1295		1702
1296	/* sort etc. */	1703	/* sort etc. */
1297	req->int1 = flags;	1704	req->int1 = flags;
1298	req->result = dentoffs;	1705	req->result = dentoffs;
1299		1706
…		…
1328		1735
1329	break;	1736	break;
1330	}	1737	}
1331		1738
1332	/* now add the entry to our list(s) */	1739	/* now add the entry to our list(s) */
1333	name = entp->d_name;	1740	name = D_NAME (entp);
1334		1741
1335	/* skip . and .. entries */	1742	/* skip . and .. entries */
1336	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1743	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1337	{	1744	{
1338	int len = D_NAMLEN (entp) + 1;	1745	int len = D_NAMLEN (entp) + 1;
1339		1746
1340	while (expect_false (namesoffs + len > namesalloc))	1747	while (ecb_expect_false (namesoffs + len > namesalloc))
1341	{	1748	{
1342	namesalloc *= 2;	1749	namesalloc *= 2;
1343	X_LOCK (wrklock);
1344	req->ptr2 = names = realloc (names, namesalloc);	1750	req->ptr2 = names = realloc (names, namesalloc);
1345	X_UNLOCK (wrklock);
1346		1751
1347	if (!names)	1752	if (!names)
1348	break;	1753	break;
1349	}	1754	}
1350		1755
…		…
1352		1757
1353	if (dents)	1758	if (dents)
1354	{	1759	{
1355	struct eio_dirent *ent;	1760	struct eio_dirent *ent;
1356		1761
1357	if (expect_false (dentoffs == dentalloc))	1762	if (ecb_expect_false (dentoffs == dentalloc))
1358	{	1763	{
1359	dentalloc *= 2;	1764	dentalloc *= 2;
1360	X_LOCK (wrklock);
1361	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1765	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1362	X_UNLOCK (wrklock);
1363		1766
1364	if (!dents)	1767	if (!dents)
1365	break;	1768	break;
1366	}	1769	}
1367		1770
…		…
1447	if (EIO_CANCELLED (req))	1850	if (EIO_CANCELLED (req))
1448	{	1851	{
1449	errno = ECANCELED;	1852	errno = ECANCELED;
1450	break;	1853	break;
1451	}	1854	}
		1855
		1856	#ifdef _WIN32
		1857	if (!FindNextFile (dirp, &entp))
		1858	{
		1859	FindClose (dirp);
		1860	dirp = 0;
		1861	}
		1862	#endif
1452	}	1863	}
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	}	1864	}
1570		1865
1571	/*****************************************************************************/	1866	/*****************************************************************************/
1572		1867
1573	#define ALLOC(len) \	1868	#define ALLOC(len) \
…		…
1589	{	1884	{
1590	ETP_REQ *req;	1885	ETP_REQ *req;
1591	struct timespec ts;	1886	struct timespec ts;
1592	etp_worker *self = (etp_worker *)thr_arg;	1887	etp_worker *self = (etp_worker *)thr_arg;
1593		1888
1594	/* try to distribute timeouts somewhat randomly */	1889	/* try to distribute timeouts somewhat evenly */
1595	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1890	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1596		1891
1597	for (;;)	1892	for (;;)
1598	{	1893	{
		1894	ts.tv_sec = 0;
		1895
1599	X_LOCK (reqlock);	1896	X_LOCK (reqlock);
1600		1897
1601	for (;;)	1898	for (;;)
1602	{	1899	{
1603	self->req = req = reqq_shift (&req_queue);	1900	self->req = req = reqq_shift (&req_queue);
1604		1901
1605	if (req)	1902	if (req)
1606	break;	1903	break;
1607		1904
		1905	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1906	{
		1907	X_UNLOCK (reqlock);
		1908	X_LOCK (wrklock);
		1909	--started;
		1910	X_UNLOCK (wrklock);
		1911	goto quit;
		1912	}
		1913
1608	++idle;	1914	++idle;
1609		1915
1610	ts.tv_sec = time (0) + idle_timeout;	1916	if (idle <= max_idle)
1611	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1917	/* we are allowed to idle, so do so without any timeout */
		1918	X_COND_WAIT (reqwait, reqlock);
		1919	else
1612	{	1920	{
1613	if (idle > max_idle)	1921	/* initialise timeout once */
1614	{	1922	if (!ts.tv_sec)
1615	--idle;	1923	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		1924
1623	/* we are allowed to idle, so do so without any timeout */
1624	X_COND_WAIT (reqwait, reqlock);	1925	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1926	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1625	}	1927	}
1626		1928
1627	--idle;	1929	--idle;
1628	}	1930	}
1629		1931
…		…
1632	X_UNLOCK (reqlock);	1934	X_UNLOCK (reqlock);
1633		1935
1634	if (req->type < 0)	1936	if (req->type < 0)
1635	goto quit;	1937	goto quit;
1636		1938
1637	if (!EIO_CANCELLED (req))
1638	ETP_EXECUTE (self, req);	1939	ETP_EXECUTE (self, req);
1639		1940
1640	X_LOCK (reslock);	1941	X_LOCK (reslock);
1641		1942
1642	++npending;	1943	++npending;
1643		1944
…		…
1658	return 0;	1959	return 0;
1659	}	1960	}
1660		1961
1661	/*****************************************************************************/	1962	/*****************************************************************************/
1662		1963
		1964	int ecb_cold
1663	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1965	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1664	{	1966	{
		1967	#if !HAVE_PREADWRITE
		1968	X_MUTEX_CREATE (preadwritelock);
		1969	#endif
		1970
1665	return etp_init (want_poll, done_poll);	1971	return etp_init (want_poll, done_poll);
1666	}	1972	}
1667		1973
		1974	ecb_inline void
1668	static void eio_api_destroy (eio_req *req)	1975	eio_api_destroy (eio_req *req)
1669	{	1976	{
1670	free (req);	1977	free (req);
1671	}	1978	}
1672		1979
1673	#define REQ(rtype) \	1980	#define REQ(rtype) \
…		…
1692	{ \	1999	{ \
1693	eio_api_destroy (req); \	2000	eio_api_destroy (req); \
1694	return 0; \	2001	return 0; \
1695	}	2002	}
1696		2003
		2004	static void
1697	static void eio_execute (etp_worker *self, eio_req *req)	2005	eio_execute (etp_worker *self, eio_req *req)
1698	{	2006	{
		2007	if (ecb_expect_false (EIO_CANCELLED (req)))
		2008	{
		2009	req->result = -1;
		2010	req->errorno = ECANCELED;
		2011	return;
		2012	}
		2013
1699	switch (req->type)	2014	switch (req->type)
1700	{	2015	{
1701	case EIO_READ: ALLOC (req->size);	2016	case EIO_READ: ALLOC (req->size);
1702	req->result = req->offs >= 0	2017	req->result = req->offs >= 0
1703	? pread (req->int1, req->ptr2, req->size, req->offs)	2018	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1705	case EIO_WRITE: req->result = req->offs >= 0	2020	case EIO_WRITE: req->result = req->offs >= 0
1706	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2021	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1707	: write (req->int1, req->ptr2, req->size); break;	2022	: write (req->int1, req->ptr2, req->size); break;
1708		2023
1709	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2024	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1710	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2025	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1711		2026
1712	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2027	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1713	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2028	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1714	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2029	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1715	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2030	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
…		…
1737	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2052	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1738	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2053	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1739	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2054	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;	2055	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1741		2056
		2057	case EIO_REALPATH: eio__realpath (req, self); break;
		2058
1742	case EIO_READLINK: ALLOC (PATH_MAX);	2059	case EIO_READLINK: ALLOC (PATH_MAX);
1743	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;	2060	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1744		2061
1745	case EIO_SYNC: req->result = 0; sync (); break;	2062	case EIO_SYNC: req->result = 0; sync (); break;
1746	case EIO_FSYNC: req->result = fsync (req->int1); break;	2063	case EIO_FSYNC: req->result = fsync (req->int1); break;
…		…
1748	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2065	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1749	case EIO_MTOUCH: req->result = eio__mtouch (req); break;	2066	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1750	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2067	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1751	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2068	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;	2069	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2070	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1753		2071
1754	case EIO_READDIR: eio__scandir (req, self); break;	2072	case EIO_READDIR: eio__scandir (req, self); break;
1755		2073
1756	case EIO_BUSY:	2074	case EIO_BUSY:
1757	#ifdef _WIN32	2075	#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)	2175	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1858	{	2176	{
1859	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2177	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1860	}	2178	}
1861		2179
		2180	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2181	{
		2182	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2183	}
		2184
1862	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2185	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1863	{	2186	{
1864	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2187	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1865	}	2188	}
1866		2189
…		…
1907	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2230	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1908	{	2231	{
1909	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2232	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1910	}	2233	}
1911		2234
1912	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2235	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1913	{	2236	{
1914	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2237	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1915	}	2238	}
1916		2239
1917	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2240	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1937	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2260	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1938	{	2261	{
1939	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2262	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1940	}	2263	}
1941		2264
1942	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2265	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1943	{	2266	{
1944	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2267	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1945	}	2268	}
1946		2269
1947	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2270	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1961	}	2284	}
1962		2285
1963	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2286	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1964	{	2287	{
1965	return eio__1path (EIO_READLINK, path, pri, cb, data);	2288	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2289	}
		2290
		2291	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2292	{
		2293	return eio__1path (EIO_REALPATH, path, pri, cb, data);
1966	}	2294	}
1967		2295
1968	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2296	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1969	{	2297	{
1970	return eio__1path (EIO_STAT, path, pri, cb, data);	2298	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)	2357	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2030	{	2358	{
2031	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2359	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2032	}	2360	}
2033		2361
2034	eio_req *eio_custom (void (*)(eio_req *) execute, int pri, eio_cb cb, void *data);	2362	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2035	{	2363	{
2036	REQ (EIO_CUSTOM); req->feed = execute; SEND;	2364	REQ (EIO_CUSTOM); req->feed = execute; SEND;
2037	}	2365	}
2038		2366
2039	#endif	2367	#endif
…		…
2050	#undef SEND	2378	#undef SEND
2051		2379
2052	/*****************************************************************************/	2380	/*****************************************************************************/
2053	/* grp functions */	2381	/* grp functions */
2054		2382
		2383	void
2055	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2384	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2056	{	2385	{
2057	grp->int2 = limit;	2386	grp->int2 = limit;
2058	grp->feed = feed;	2387	grp->feed = feed;
2059		2388
2060	grp_try_feed (grp);	2389	grp_try_feed (grp);
2061	}	2390	}
2062		2391
		2392	void
2063	void eio_grp_limit (eio_req *grp, int limit)	2393	eio_grp_limit (eio_req *grp, int limit)
2064	{	2394	{
2065	grp->int2 = limit;	2395	grp->int2 = limit;
2066		2396
2067	grp_try_feed (grp);	2397	grp_try_feed (grp);
2068	}	2398	}
2069		2399
		2400	void
2070	void eio_grp_add (eio_req *grp, eio_req *req)	2401	eio_grp_add (eio_req *grp, eio_req *req)
2071	{	2402	{
2072	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2403	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2073		2404
2074	grp->flags |= EIO_FLAG_GROUPADD;	2405	grp->flags |= EIO_FLAG_GROUPADD;
2075		2406
…		…
2086	}	2417	}
2087		2418
2088	/*****************************************************************************/	2419	/*****************************************************************************/
2089	/* misc garbage */	2420	/* misc garbage */
2090		2421
		2422	eio_ssize_t
2091	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2423	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2092	{	2424	{
2093	etp_worker wrk;
2094	ssize_t ret;
2095
2096	wrk.dbuf = 0;
2097
2098	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2425	return eio__sendfile (ofd, ifd, offset, count);
2099
2100	if (wrk.dbuf)
2101	free (wrk.dbuf);
2102
2103	return ret;
2104	}	2426	}
2105		2427

Diff Legend

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