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 1.98 by root, Sun Jul 24 05:53:34 2011 UTC

…		…
40	#ifndef _WIN32	40	#ifndef _WIN32
41	# include "config.h"	41	# include "config.h"
42	#endif	42	#endif
43		43
44	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
45		46
46	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
47	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
48	#endif	49	#endif
49	#include "xthread.h"	50	#include "xthread.h"
…		…
53	#include <stdlib.h>	54	#include <stdlib.h>
54	#include <string.h>	55	#include <string.h>
55	#include <errno.h>	56	#include <errno.h>
56	#include <sys/types.h>	57	#include <sys/types.h>
57	#include <sys/stat.h>	58	#include <sys/stat.h>
58	#include <sys/statvfs.h>
59	#include <limits.h>	59	#include <limits.h>
60	#include <fcntl.h>	60	#include <fcntl.h>
61	#include <assert.h>	61	#include <assert.h>
62		62
63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */	63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64	/* intptr_t only comes form stdint.h, says idiot openbsd coder */	64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65	#if HAVE_STDINT_H	65	#if HAVE_STDINT_H
66	# include <stdint.h>	66	# include <stdint.h>
67	#endif	67	#endif
68		68
		69	#ifndef ECANCELED
		70	# define ECANCELED EDOM
		71	#endif
		72	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
		77	# define ENOTSOCK WSAENOTSOCK
		78	#endif
		79
		80	static void eio_destroy (eio_req *req);
		81
69	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
70	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	83	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
71	#endif	84	#endif
72		85
73	#ifndef EIO_DESTROY	86	#ifndef EIO_DESTROY
…		…
76		89
77	#ifndef EIO_FEED	90	#ifndef EIO_FEED
78	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	91	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
79	#endif	92	#endif
80		93
		94	#ifndef EIO_FD_TO_WIN32_HANDLE
		95	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		96	#endif
		97	#ifndef EIO_WIN32_HANDLE_TO_FD
		98	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		99	#endif
		100
		101	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		102
		103	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		104
81	#ifdef _WIN32	105	#ifdef _WIN32
82		106
83	/*doh*/	107	#undef PAGESIZE
		108	#define PAGESIZE 4096 /* GetSystemInfo? */
		109
		110	#ifdef EIO_STRUCT_STATI64
		111	#define stat(path,buf) _stati64 (path,buf)
		112	#define fstat(fd,buf) _fstati64 (fd,buf)
		113	#endif
		114	#define lstat(path,buf) stat (path,buf)
		115	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		116	#define mkdir(path,mode) _mkdir (path)
		117	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		118
		119	#define chmod(path,mode) _chmod (path, mode)
		120	#define dup(fd) _dup (fd)
		121	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		122
		123	#define fchmod(fd,mode) EIO_ENOSYS ()
		124	#define chown(path,uid,gid) EIO_ENOSYS ()
		125	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		126	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		127	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		128	#define mknod(path,mode,dev) EIO_ENOSYS ()
		129	#define sync() EIO_ENOSYS ()
		130	#define readlink(path,buf,s) EIO_ENOSYS ()
		131	#define statvfs(path,buf) EIO_ENOSYS ()
		132	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		133
		134	/* we could even stat and see if it exists */
		135	static int
		136	symlink (const char *old, const char *neu)
		137	{
		138	#if WINVER >= 0x0600
		139	if (CreateSymbolicLink (neu, old, 1))
		140	return 0;
		141
		142	if (CreateSymbolicLink (neu, old, 0))
		143	return 0;
		144	#endif
		145
		146	return EIO_ERRNO (ENOENT, -1);
		147	}
		148
		149	/* POSIX API only */
		150	#define CreateHardLink(neu,old,flags) 0
		151	#define CreateSymbolicLink(neu,old,flags) 0
		152
		153	struct statvfs
		154	{
		155	int dummy;
		156	};
		157
		158	#define DT_DIR EIO_DT_DIR
		159	#define DT_REG EIO_DT_REG
		160	#define D_NAME(entp) entp.cFileName
		161	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
		162
84	#else	163	#else
85		164
86	# include <sys/time.h>	165	#include <sys/time.h>
87	# include <sys/select.h>	166	#include <sys/select.h>
		167	#include <sys/statvfs.h>
88	# include <unistd.h>	168	#include <unistd.h>
		169	#include <signal.h>
		170	#include <dirent.h>
		171
		172	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		173	#include <sys/mman.h>
		174	#endif
		175
		176	#define D_NAME(entp) entp->d_name
		177
		178	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		179	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		180	#define _DIRENT_HAVE_D_TYPE /* sigh */
		181	#define D_INO(de) (de)->d_fileno
		182	#define D_NAMLEN(de) (de)->d_namlen
		183	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		184	#define D_INO(de) (de)->d_ino
		185	#endif
		186
		187	#ifdef _D_EXACT_NAMLEN
		188	#undef D_NAMLEN
		189	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		190	#endif
		191
		192	#ifdef _DIRENT_HAVE_D_TYPE
		193	#define D_TYPE(de) (de)->d_type
		194	#endif
		195
		196	#ifndef EIO_STRUCT_DIRENT
		197	#define EIO_STRUCT_DIRENT struct dirent
		198	#endif
		199
		200	#endif
		201
		202	#if HAVE_UTIMES
89	# include <utime.h>	203	# include <utime.h>
90	# include <signal.h>
91	# include <dirent.h>
92
93	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94	# include <sys/mman.h>
95	#endif
96
97	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99	# define _DIRENT_HAVE_D_TYPE /* sigh */
100	# define D_INO(de) (de)->d_fileno
101	# define D_NAMLEN(de) (de)->d_namlen
102	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103	# define D_INO(de) (de)->d_ino
104	# endif	204	#endif
105		205
106	#ifdef _D_EXACT_NAMLEN	206	#if HAVE_SYS_SYSCALL_H
107	# undef D_NAMLEN	207	# include <sys/syscall.h>
108	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109	#endif
110
111	# ifdef _DIRENT_HAVE_D_TYPE
112	# define D_TYPE(de) (de)->d_type
113	# endif	208	#endif
114		209
115	# ifndef EIO_STRUCT_DIRENT	210	#if HAVE_SYS_PRCTL_H
116	# define EIO_STRUCT_DIRENT struct dirent	211	# include <sys/prctl.h>
117	# endif
118
119	#endif	212	#endif
120		213
121	#if HAVE_SENDFILE	214	#if HAVE_SENDFILE
122	# if __linux	215	# if __linux
123	# include <sys/sendfile.h>	216	# include <sys/sendfile.h>
…		…
138	#endif	231	#endif
139	#ifndef D_INO	232	#ifndef D_INO
140	# define D_INO(de) 0	233	# define D_INO(de) 0
141	#endif	234	#endif
142	#ifndef D_NAMLEN	235	#ifndef D_NAMLEN
143	# define D_NAMLEN(de) strlen ((de)->d_name)	236	# define D_NAMLEN(entp) strlen (D_NAME (entp))
144	#endif	237	#endif
145		238
146	/* used for struct dirent, AIX doesn't provide it */	239	/* used for struct dirent, AIX doesn't provide it */
147	#ifndef NAME_MAX	240	#ifndef NAME_MAX
148	# define NAME_MAX 4096	241	# define NAME_MAX 4096
…		…
155		248
156	/* buffer size for various temporary buffers */	249	/* buffer size for various temporary buffers */
157	#define EIO_BUFSIZE 65536	250	#define EIO_BUFSIZE 65536
158		251
159	#define dBUF \	252	#define dBUF \
160	char *eio_buf; \
161	ETP_WORKER_LOCK (self); \
162	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	253	char *eio_buf = malloc (EIO_BUFSIZE); \
163	ETP_WORKER_UNLOCK (self); \
164	errno = ENOMEM; \	254	errno = ENOMEM; \
165	if (!eio_buf) \	255	if (!eio_buf) \
166	return -1;	256	return -1
		257
		258	#define FUBd \
		259	free (eio_buf)
167		260
168	#define EIO_TICKS ((1000000 + 1023) >> 10)	261	#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		262
183	#define ETP_PRI_MIN EIO_PRI_MIN	263	#define ETP_PRI_MIN EIO_PRI_MIN
184	#define ETP_PRI_MAX EIO_PRI_MAX	264	#define ETP_PRI_MAX EIO_PRI_MAX
185		265
186	struct etp_worker;	266	struct etp_worker;
…		…
190	static int eio_finish (eio_req *req);	270	static int eio_finish (eio_req *req);
191	#define ETP_FINISH(req) eio_finish (req)	271	#define ETP_FINISH(req) eio_finish (req)
192	static void eio_execute (struct etp_worker *self, eio_req *req);	272	static void eio_execute (struct etp_worker *self, eio_req *req);
193	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	273	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194		274
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	/*****************************************************************************/	275	/*****************************************************************************/
213		276
214	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	277	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215		278
216	/* calculate time difference in ~1/EIO_TICKS of a second */	279	/* calculate time difference in ~1/EIO_TICKS of a second */
		280	ecb_inline int
217	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	281	tvdiff (struct timeval *tv1, struct timeval *tv2)
218	{	282	{
219	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	283	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	284	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
221	}	285	}
222		286
…		…
243	/*	307	/*
244	* make our pread/pwrite emulation safe against themselves, but not against	308	* make our pread/pwrite emulation safe against themselves, but not against
245	* normal read/write by using a mutex. slows down execution a lot,	309	* normal read/write by using a mutex. slows down execution a lot,
246	* but that's your problem, not mine.	310	* but that's your problem, not mine.
247	*/	311	*/
248	static xmutex_t preadwritelock = X_MUTEX_INIT;	312	static xmutex_t preadwritelock;
249	#endif	313	#endif
250		314
251	typedef struct etp_worker	315	typedef struct etp_worker
252	{	316	{
253	/* locked by wrklock */	317	/* locked by wrklock */
…		…
256	xthread_t tid;	320	xthread_t tid;
257		321
258	/* locked by reslock, reqlock or wrklock */	322	/* locked by reslock, reqlock or wrklock */
259	ETP_REQ *req; /* currently processed request */	323	ETP_REQ *req; /* currently processed request */
260		324
		325	#ifdef ETP_WORKER_COMMON
261	ETP_WORKER_COMMON	326	ETP_WORKER_COMMON
		327	#endif
262	} etp_worker;	328	} etp_worker;
263		329
264	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	330	static etp_worker wrk_first; /* NOT etp */
265		331
266	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	332	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	333	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268		334
269	/* worker threads management */	335	/* worker threads management */
270		336
		337	static void ecb_cold
271	static void etp_worker_clear (etp_worker *wrk)	338	etp_worker_clear (etp_worker *wrk)
272	{	339	{
273	ETP_WORKER_CLEAR (wrk);
274	}	340	}
275		341
		342	static void ecb_cold
276	static void etp_worker_free (etp_worker *wrk)	343	etp_worker_free (etp_worker *wrk)
277	{	344	{
278	wrk->next->prev = wrk->prev;	345	wrk->next->prev = wrk->prev;
279	wrk->prev->next = wrk->next;	346	wrk->prev->next = wrk->next;
280		347
281	free (wrk);	348	free (wrk);
282	}	349	}
283		350
284	static unsigned int etp_nreqs (void)	351	static unsigned int
		352	etp_nreqs (void)
285	{	353	{
286	int retval;	354	int retval;
287	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	355	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288	retval = nreqs;	356	retval = nreqs;
289	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	357	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290	return retval;	358	return retval;
291	}	359	}
292		360
293	static unsigned int etp_nready (void)	361	static unsigned int
		362	etp_nready (void)
294	{	363	{
295	unsigned int retval;	364	unsigned int retval;
296		365
297	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	366	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298	retval = nready;	367	retval = nready;
299	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	368	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300		369
301	return retval;	370	return retval;
302	}	371	}
303		372
304	static unsigned int etp_npending (void)	373	static unsigned int
		374	etp_npending (void)
305	{	375	{
306	unsigned int retval;	376	unsigned int retval;
307		377
308	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	378	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309	retval = npending;	379	retval = npending;
310	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	380	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311		381
312	return retval;	382	return retval;
313	}	383	}
314		384
315	static unsigned int etp_nthreads (void)	385	static unsigned int
		386	etp_nthreads (void)
316	{	387	{
317	unsigned int retval;	388	unsigned int retval;
318		389
319	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	390	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320	retval = started;	391	retval = started;
…		…
334	} etp_reqq;	405	} etp_reqq;
335		406
336	static etp_reqq req_queue;	407	static etp_reqq req_queue;
337	static etp_reqq res_queue;	408	static etp_reqq res_queue;
338		409
		410	static void ecb_noinline ecb_cold
		411	reqq_init (etp_reqq *q)
		412	{
		413	int pri;
		414
		415	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		416	q->qs[pri] = q->qe[pri] = 0;
		417
		418	q->size = 0;
		419	}
		420
		421	static int ecb_noinline
339	static int reqq_push (etp_reqq *q, ETP_REQ *req)	422	reqq_push (etp_reqq *q, ETP_REQ *req)
340	{	423	{
341	int pri = req->pri;	424	int pri = req->pri;
342	req->next = 0;	425	req->next = 0;
343		426
344	if (q->qe[pri])	427	if (q->qe[pri])
…		…
350	q->qe[pri] = q->qs[pri] = req;	433	q->qe[pri] = q->qs[pri] = req;
351		434
352	return q->size++;	435	return q->size++;
353	}	436	}
354		437
		438	static ETP_REQ * ecb_noinline
355	static ETP_REQ *reqq_shift (etp_reqq *q)	439	reqq_shift (etp_reqq *q)
356	{	440	{
357	int pri;	441	int pri;
358		442
359	if (!q->size)	443	if (!q->size)
360	return 0;	444	return 0;
…		…
375	}	459	}
376		460
377	abort ();	461	abort ();
378	}	462	}
379		463
380	static void etp_thread_init (void)	464	static int ecb_cold
		465	etp_init (void (*want_poll)(void), void (*done_poll)(void))
381	{	466	{
382	X_MUTEX_CREATE (wrklock);	467	X_MUTEX_CREATE (wrklock);
383	X_MUTEX_CREATE (reslock);	468	X_MUTEX_CREATE (reslock);
384	X_MUTEX_CREATE (reqlock);	469	X_MUTEX_CREATE (reqlock);
385	X_COND_CREATE (reqwait);	470	X_COND_CREATE (reqwait);
386	}
387		471
388	static void etp_atfork_prepare (void)	472	reqq_init (&req_queue);
389	{	473	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		474
398	static void etp_atfork_parent (void)	475	wrk_first.next =
399	{	476	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		477
429	started = 0;	478	started = 0;
430	idle = 0;	479	idle = 0;
431	nreqs = 0;	480	nreqs = 0;
432	nready = 0;	481	nready = 0;
433	npending = 0;	482	npending = 0;
434		483
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;	484	want_poll_cb = want_poll;
453	done_poll_cb = done_poll;	485	done_poll_cb = done_poll;
454		486
455	return 0;	487	return 0;
456	}	488	}
457		489
458	X_THREAD_PROC (etp_proc);	490	X_THREAD_PROC (etp_proc);
459		491
		492	static void ecb_cold
460	static void etp_start_thread (void)	493	etp_start_thread (void)
461	{	494	{
462	etp_worker *wrk = calloc (1, sizeof (etp_worker));	495	etp_worker *wrk = calloc (1, sizeof (etp_worker));
463		496
464	/*TODO*/	497	/*TODO*/
465	assert (("unable to allocate worker thread data", wrk));	498	assert (("unable to allocate worker thread data", wrk));
…		…
478	free (wrk);	511	free (wrk);
479		512
480	X_UNLOCK (wrklock);	513	X_UNLOCK (wrklock);
481	}	514	}
482		515
		516	static void
483	static void etp_maybe_start_thread (void)	517	etp_maybe_start_thread (void)
484	{	518	{
485	if (expect_true (etp_nthreads () >= wanted))	519	if (ecb_expect_true (etp_nthreads () >= wanted))
486	return;	520	return;
487		521
488	/* todo: maybe use idle here, but might be less exact */	522	/* todo: maybe use idle here, but might be less exact */
489	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	523	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
490	return;	524	return;
491		525
492	etp_start_thread ();	526	etp_start_thread ();
493	}	527	}
494		528
		529	static void ecb_cold
495	static void etp_end_thread (void)	530	etp_end_thread (void)
496	{	531	{
497	eio_req *req = calloc (1, sizeof (eio_req));	532	eio_req *req = calloc (1, sizeof (eio_req));
498		533
499	req->type = -1;	534	req->type = -1;
500	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	535	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
507	X_LOCK (wrklock);	542	X_LOCK (wrklock);
508	--started;	543	--started;
509	X_UNLOCK (wrklock);	544	X_UNLOCK (wrklock);
510	}	545	}
511		546
512	static int etp_poll (void)	547	static int
		548	etp_poll (void)
513	{	549	{
514	unsigned int maxreqs;	550	unsigned int maxreqs;
515	unsigned int maxtime;	551	unsigned int maxtime;
516	struct timeval tv_start, tv_now;	552	struct timeval tv_start, tv_now;
517		553
…		…
547		583
548	X_LOCK (reqlock);	584	X_LOCK (reqlock);
549	--nreqs;	585	--nreqs;
550	X_UNLOCK (reqlock);	586	X_UNLOCK (reqlock);
551		587
552	if (expect_false (req->type == EIO_GROUP && req->size))	588	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
553	{	589	{
554	req->int1 = 1; /* mark request as delayed */	590	req->int1 = 1; /* mark request as delayed */
555	continue;	591	continue;
556	}	592	}
557	else	593	else
558	{	594	{
559	int res = ETP_FINISH (req);	595	int res = ETP_FINISH (req);
560	if (expect_false (res))	596	if (ecb_expect_false (res))
561	return res;	597	return res;
562	}	598	}
563		599
564	if (expect_false (maxreqs && !--maxreqs))	600	if (ecb_expect_false (maxreqs && !--maxreqs))
565	break;	601	break;
566		602
567	if (maxtime)	603	if (maxtime)
568	{	604	{
569	gettimeofday (&tv_now, 0);	605	gettimeofday (&tv_now, 0);
…		…
575		611
576	errno = EAGAIN;	612	errno = EAGAIN;
577	return -1;	613	return -1;
578	}	614	}
579		615
		616	static void
580	static void etp_cancel (ETP_REQ *req)	617	etp_cancel (ETP_REQ *req)
581	{	618	{
582	X_LOCK (wrklock);	619	req->cancelled = 1;
583	req->flags |= EIO_FLAG_CANCELLED;
584	X_UNLOCK (wrklock);
585		620
586	eio_grp_cancel (req);	621	eio_grp_cancel (req);
587	}	622	}
588		623
		624	static void
589	static void etp_submit (ETP_REQ *req)	625	etp_submit (ETP_REQ *req)
590	{	626	{
591	req->pri -= ETP_PRI_MIN;	627	req->pri -= ETP_PRI_MIN;
592		628
593	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	629	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;	630	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595		631
596	if (expect_false (req->type == EIO_GROUP))	632	if (ecb_expect_false (req->type == EIO_GROUP))
597	{	633	{
598	/* I hope this is worth it :/ */	634	/* I hope this is worth it :/ */
599	X_LOCK (reqlock);	635	X_LOCK (reqlock);
600	++nreqs;	636	++nreqs;
601	X_UNLOCK (reqlock);	637	X_UNLOCK (reqlock);
…		…
620		656
621	etp_maybe_start_thread ();	657	etp_maybe_start_thread ();
622	}	658	}
623	}	659	}
624		660
		661	static void ecb_cold
625	static void etp_set_max_poll_time (double nseconds)	662	etp_set_max_poll_time (double nseconds)
626	{	663	{
627	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	664	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628	max_poll_time = nseconds * EIO_TICKS;	665	max_poll_time = nseconds * EIO_TICKS;
629	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	666	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630	}	667	}
631		668
		669	static void ecb_cold
632	static void etp_set_max_poll_reqs (unsigned int maxreqs)	670	etp_set_max_poll_reqs (unsigned int maxreqs)
633	{	671	{
634	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	672	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635	max_poll_reqs = maxreqs;	673	max_poll_reqs = maxreqs;
636	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	674	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637	}	675	}
638		676
		677	static void ecb_cold
639	static void etp_set_max_idle (unsigned int nthreads)	678	etp_set_max_idle (unsigned int nthreads)
640	{	679	{
641	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	680	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642	max_idle = nthreads;	681	max_idle = nthreads;
643	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	682	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644	}	683	}
645		684
		685	static void ecb_cold
646	static void etp_set_idle_timeout (unsigned int seconds)	686	etp_set_idle_timeout (unsigned int seconds)
647	{	687	{
648	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	688	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649	idle_timeout = seconds;	689	idle_timeout = seconds;
650	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	690	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651	}	691	}
652		692
		693	static void ecb_cold
653	static void etp_set_min_parallel (unsigned int nthreads)	694	etp_set_min_parallel (unsigned int nthreads)
654	{	695	{
655	if (wanted < nthreads)	696	if (wanted < nthreads)
656	wanted = nthreads;	697	wanted = nthreads;
657	}	698	}
658		699
		700	static void ecb_cold
659	static void etp_set_max_parallel (unsigned int nthreads)	701	etp_set_max_parallel (unsigned int nthreads)
660	{	702	{
661	if (wanted > nthreads)	703	if (wanted > nthreads)
662	wanted = nthreads;	704	wanted = nthreads;
663		705
664	while (started > wanted)	706	while (started > wanted)
665	etp_end_thread ();	707	etp_end_thread ();
666	}	708	}
667		709
668	/*****************************************************************************/	710	/*****************************************************************************/
669		711
		712	static void
670	static void grp_try_feed (eio_req *grp)	713	grp_try_feed (eio_req *grp)
671	{	714	{
672	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	715	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673	{	716	{
674	grp->flags &= ~EIO_FLAG_GROUPADD;	717	grp->flags &= ~EIO_FLAG_GROUPADD;
675		718
…		…
682	break;	725	break;
683	}	726	}
684	}	727	}
685	}	728	}
686		729
		730	static int
687	static int grp_dec (eio_req *grp)	731	grp_dec (eio_req *grp)
688	{	732	{
689	--grp->size;	733	--grp->size;
690		734
691	/* call feeder, if applicable */	735	/* call feeder, if applicable */
692	grp_try_feed (grp);	736	grp_try_feed (grp);
…		…
696	return eio_finish (grp);	740	return eio_finish (grp);
697	else	741	else
698	return 0;	742	return 0;
699	}	743	}
700		744
		745	static void
701	void eio_destroy (eio_req *req)	746	eio_destroy (eio_req *req)
702	{	747	{
703	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	748	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	749	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705		750
706	EIO_DESTROY (req);	751	EIO_DESTROY (req);
707	}	752	}
708		753
		754	static int
709	static int eio_finish (eio_req *req)	755	eio_finish (eio_req *req)
710	{	756	{
711	int res = EIO_FINISH (req);	757	int res = EIO_FINISH (req);
712		758
713	if (req->grp)	759	if (req->grp)
714	{	760	{
…		…
722	if (grp->grp_first == req)	768	if (grp->grp_first == req)
723	grp->grp_first = req->grp_next;	769	grp->grp_first = req->grp_next;
724		770
725	res2 = grp_dec (grp);	771	res2 = grp_dec (grp);
726		772
727	if (!res && res2)	773	if (!res)
728	res = res2;	774	res = res2;
729	}	775	}
730		776
731	eio_destroy (req);	777	eio_destroy (req);
732		778
733	return res;	779	return res;
734	}	780	}
735		781
		782	void
736	void eio_grp_cancel (eio_req *grp)	783	eio_grp_cancel (eio_req *grp)
737	{	784	{
738	for (grp = grp->grp_first; grp; grp = grp->grp_next)	785	for (grp = grp->grp_first; grp; grp = grp->grp_next)
739	eio_cancel (grp);	786	eio_cancel (grp);
740	}	787	}
741		788
		789	void
742	void eio_cancel (eio_req *req)	790	eio_cancel (eio_req *req)
743	{	791	{
744	etp_cancel (req);	792	etp_cancel (req);
745	}	793	}
746		794
		795	void
747	void eio_submit (eio_req *req)	796	eio_submit (eio_req *req)
748	{	797	{
749	etp_submit (req);	798	etp_submit (req);
750	}	799	}
751		800
752	unsigned int eio_nreqs (void)	801	unsigned int
		802	eio_nreqs (void)
753	{	803	{
754	return etp_nreqs ();	804	return etp_nreqs ();
755	}	805	}
756		806
757	unsigned int eio_nready (void)	807	unsigned int
		808	eio_nready (void)
758	{	809	{
759	return etp_nready ();	810	return etp_nready ();
760	}	811	}
761		812
762	unsigned int eio_npending (void)	813	unsigned int
		814	eio_npending (void)
763	{	815	{
764	return etp_npending ();	816	return etp_npending ();
765	}	817	}
766		818
767	unsigned int eio_nthreads (void)	819	unsigned int ecb_cold
		820	eio_nthreads (void)
768	{	821	{
769	return etp_nthreads ();	822	return etp_nthreads ();
770	}	823	}
771		824
		825	void ecb_cold
772	void eio_set_max_poll_time (double nseconds)	826	eio_set_max_poll_time (double nseconds)
773	{	827	{
774	etp_set_max_poll_time (nseconds);	828	etp_set_max_poll_time (nseconds);
775	}	829	}
776		830
		831	void ecb_cold
777	void eio_set_max_poll_reqs (unsigned int maxreqs)	832	eio_set_max_poll_reqs (unsigned int maxreqs)
778	{	833	{
779	etp_set_max_poll_reqs (maxreqs);	834	etp_set_max_poll_reqs (maxreqs);
780	}	835	}
781		836
		837	void ecb_cold
782	void eio_set_max_idle (unsigned int nthreads)	838	eio_set_max_idle (unsigned int nthreads)
783	{	839	{
784	etp_set_max_idle (nthreads);	840	etp_set_max_idle (nthreads);
785	}	841	}
786		842
		843	void ecb_cold
787	void eio_set_idle_timeout (unsigned int seconds)	844	eio_set_idle_timeout (unsigned int seconds)
788	{	845	{
789	etp_set_idle_timeout (seconds);	846	etp_set_idle_timeout (seconds);
790	}	847	}
791		848
		849	void ecb_cold
792	void eio_set_min_parallel (unsigned int nthreads)	850	eio_set_min_parallel (unsigned int nthreads)
793	{	851	{
794	etp_set_min_parallel (nthreads);	852	etp_set_min_parallel (nthreads);
795	}	853	}
796		854
		855	void ecb_cold
797	void eio_set_max_parallel (unsigned int nthreads)	856	eio_set_max_parallel (unsigned int nthreads)
798	{	857	{
799	etp_set_max_parallel (nthreads);	858	etp_set_max_parallel (nthreads);
800	}	859	}
801		860
802	int eio_poll (void)	861	int eio_poll (void)
…		…
811	# undef pread	870	# undef pread
812	# undef pwrite	871	# undef pwrite
813	# define pread eio__pread	872	# define pread eio__pread
814	# define pwrite eio__pwrite	873	# define pwrite eio__pwrite
815		874
816	static ssize_t	875	static eio_ssize_t
817	eio__pread (int fd, void *buf, size_t count, off_t offset)	876	eio__pread (int fd, void *buf, size_t count, off_t offset)
818	{	877	{
819	ssize_t res;	878	eio_ssize_t res;
820	off_t ooffset;	879	off_t ooffset;
821		880
822	X_LOCK (preadwritelock);	881	X_LOCK (preadwritelock);
823	ooffset = lseek (fd, 0, SEEK_CUR);	882	ooffset = lseek (fd, 0, SEEK_CUR);
824	lseek (fd, offset, SEEK_SET);	883	lseek (fd, offset, SEEK_SET);
…		…
827	X_UNLOCK (preadwritelock);	886	X_UNLOCK (preadwritelock);
828		887
829	return res;	888	return res;
830	}	889	}
831		890
832	static ssize_t	891	static eio_ssize_t
833	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	892	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834	{	893	{
835	ssize_t res;	894	eio_ssize_t res;
836	off_t ooffset;	895	off_t ooffset;
837		896
838	X_LOCK (preadwritelock);	897	X_LOCK (preadwritelock);
839	ooffset = lseek (fd, 0, SEEK_CUR);	898	ooffset = lseek (fd, 0, SEEK_CUR);
840	lseek (fd, offset, SEEK_SET);	899	lseek (fd, offset, SEEK_SET);
…		…
872	#ifndef HAVE_FUTIMES	931	#ifndef HAVE_FUTIMES
873		932
874	# undef futimes	933	# undef futimes
875	# define futimes(fd,times) eio__futimes (fd, times)	934	# define futimes(fd,times) eio__futimes (fd, times)
876		935
		936	static int
877	static int eio__futimes (int fd, const struct timeval tv[2])	937	eio__futimes (int fd, const struct timeval tv[2])
878	{	938	{
879	errno = ENOSYS;	939	errno = ENOSYS;
880	return -1;	940	return -1;
881	}	941	}
882		942
…		…
885	#if !HAVE_FDATASYNC	945	#if !HAVE_FDATASYNC
886	# undef fdatasync	946	# undef fdatasync
887	# define fdatasync(fd) fsync (fd)	947	# define fdatasync(fd) fsync (fd)
888	#endif	948	#endif
889		949
		950	static int
		951	eio__syncfs (int fd)
		952	{
		953	int res;
		954
		955	#if HAVE_SYS_SYNCFS
		956	res = (int)syscall (__NR_syncfs, (int)(fd));
		957	#else
		958	res = -1;
		959	errno = ENOSYS;
		960	#endif
		961
		962	if (res < 0 && errno == ENOSYS && fd >= 0)
		963	sync ();
		964
		965	return res;
		966	}
		967
890	/* sync_file_range always needs emulation */	968	/* sync_file_range always needs emulation */
891	int	969	static int
892	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	970	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893	{	971	{
894	#if HAVE_SYNC_FILE_RANGE	972	#if HAVE_SYNC_FILE_RANGE
895	int res;	973	int res;
896		974
…		…
913	/* even though we could play tricks with the flags, it's better to always	991	/* 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 */	992	* call fdatasync, as that matches the expectation of its users best */
915	return fdatasync (fd);	993	return fdatasync (fd);
916	}	994	}
917		995
		996	static int
		997	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		998	{
		999	#if HAVE_FALLOCATE
		1000	return fallocate (fd, mode, offset, len);
		1001	#else
		1002	errno = ENOSYS;
		1003	return -1;
		1004	#endif
		1005	}
		1006
918	#if !HAVE_READAHEAD	1007	#if !HAVE_READAHEAD
919	# undef readahead	1008	# undef readahead
920	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1009	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921		1010
922	static ssize_t	1011	static eio_ssize_t
923	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1012	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
924	{	1013	{
925	size_t todo = count;	1014	size_t todo = count;
926	dBUF;	1015	dBUF;
927		1016
…		…
932	pread (fd, eio_buf, len, offset);	1021	pread (fd, eio_buf, len, offset);
933	offset += len;	1022	offset += len;
934	todo -= len;	1023	todo -= len;
935	}	1024	}
936		1025
		1026	FUBd;
		1027
937	errno = 0;	1028	errno = 0;
938	return count;	1029	return count;
939	}	1030	}
940		1031
941	#endif	1032	#endif
942		1033
943	/* sendfile always needs emulation */	1034	/* sendfile always needs emulation */
944	static ssize_t	1035	static eio_ssize_t
945	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1036	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
946	{	1037	{
947	ssize_t written = 0;	1038	eio_ssize_t written = 0;
948	ssize_t res;	1039	eio_ssize_t res;
949		1040
950	if (!count)	1041	if (!count)
951	return 0;	1042	return 0;
952		1043
953	for (;;)	1044	for (;;)
954	{	1045	{
		1046	#ifdef __APPLE__
		1047	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1048	#endif
955	#if HAVE_SENDFILE	1049	#if HAVE_SENDFILE
956	# if __linux	1050	# if __linux
957	off_t soffset = offset;	1051	off_t soffset = offset;
958	res = sendfile (ofd, ifd, &soffset, count);	1052	res = sendfile (ofd, ifd, &soffset, count);
959		1053
…		…
1002	if (res < 0 && sbytes)	1096	if (res < 0 && sbytes)
1003	res = sbytes;	1097	res = sbytes;
1004		1098
1005	# endif	1099	# endif
1006		1100
1007	#elif defined (_WIN32)	1101	#elif defined (_WIN32) && 0
1008	/* does not work, just for documentation of what would need to be done */	1102	/* 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, */	1103	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
1010	/* libeio guarantees that the file offset does not change, and windows */	1104	/* 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 */	1105	/* has no way to get an independent handle to the same file description */
1012	HANDLE h = TO_SOCKET (ifd);	1106	HANDLE h = TO_SOCKET (ifd);
…		…
1045	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1139	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1046	/* BSDs */	1140	/* BSDs */
1047	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1141	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1048	|| errno == ENOTSUP	1142	|| errno == ENOTSUP
1049	#endif	1143	#endif
		1144	#ifdef EOPNOTSUPP /* windows */
1050	|| errno == EOPNOTSUPP /* BSDs */	1145	|| errno == EOPNOTSUPP /* BSDs */
		1146	#endif
1051	#if __solaris	1147	#if __solaris
1052	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1148	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1053	#endif	1149	#endif
1054	)	1150	)
1055	)	1151	)
…		…
1059		1155
1060	res = 0;	1156	res = 0;
1061		1157
1062	while (count)	1158	while (count)
1063	{	1159	{
1064	ssize_t cnt;	1160	eio_ssize_t cnt;
1065		1161
1066	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1162	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1067		1163
1068	if (cnt <= 0)	1164	if (cnt <= 0)
1069	{	1165	{
…		…
1081		1177
1082	offset += cnt;	1178	offset += cnt;
1083	res += cnt;	1179	res += cnt;
1084	count -= cnt;	1180	count -= cnt;
1085	}	1181	}
		1182
		1183	FUBd;
1086	}	1184	}
1087		1185
1088	return res;	1186	return res;
		1187	}
		1188
		1189	#ifdef PAGESIZE
		1190	# define eio_pagesize() PAGESIZE
		1191	#else
		1192	static intptr_t
		1193	eio_pagesize (void)
		1194	{
		1195	static intptr_t page;
		1196
		1197	if (!page)
		1198	page = sysconf (_SC_PAGESIZE);
		1199
		1200	return page;
		1201	}
		1202	#endif
		1203
		1204	static void
		1205	eio_page_align (void **addr, size_t *length)
		1206	{
		1207	intptr_t mask = eio_pagesize () - 1;
		1208
		1209	/* round down addr */
		1210	intptr_t adj = mask & (intptr_t)*addr;
		1211
		1212	*addr = (void *)((intptr_t)*addr - adj);
		1213	*length += adj;
		1214
		1215	/* round up length */
		1216	*length = (*length + mask) & ~mask;
		1217	}
		1218
		1219	#if !_POSIX_MEMLOCK
		1220	# define eio__mlockall(a) EIO_ENOSYS ()
		1221	#else
		1222
		1223	static int
		1224	eio__mlockall (int flags)
		1225	{
		1226	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1227	extern int mallopt (int, int);
		1228	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1229	#endif
		1230
		1231	if (EIO_MCL_CURRENT != MCL_CURRENT
		1232	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1233	{
		1234	flags = 0
		1235	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1236	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1237	}
		1238
		1239	return mlockall (flags);
		1240	}
		1241	#endif
		1242
		1243	#if !_POSIX_MEMLOCK_RANGE
		1244	# define eio__mlock(a,b) EIO_ENOSYS ()
		1245	#else
		1246
		1247	static int
		1248	eio__mlock (void *addr, size_t length)
		1249	{
		1250	eio_page_align (&addr, &length);
		1251
		1252	return mlock (addr, length);
		1253	}
		1254
		1255	#endif
		1256
		1257	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1258	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1259	#else
		1260
		1261	static int
		1262	eio__msync (void *mem, size_t len, int flags)
		1263	{
		1264	eio_page_align (&mem, &len);
		1265
		1266	if (EIO_MS_ASYNC != MS_SYNC
		1267	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1268	|| EIO_MS_SYNC != MS_SYNC)
		1269	{
		1270	flags = 0
		1271	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1272	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1273	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1274	}
		1275
		1276	return msync (mem, len, flags);
		1277	}
		1278
		1279	#endif
		1280
		1281	static int
		1282	eio__mtouch (eio_req *req)
		1283	{
		1284	void *mem = req->ptr2;
		1285	size_t len = req->size;
		1286	int flags = req->int1;
		1287
		1288	eio_page_align (&mem, &len);
		1289
		1290	{
		1291	intptr_t addr = (intptr_t)mem;
		1292	intptr_t end = addr + len;
		1293	intptr_t page = eio_pagesize ();
		1294
		1295	if (addr < end)
		1296	if (flags & EIO_MT_MODIFY) /* modify */
		1297	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1298	else
		1299	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1300	}
		1301
		1302	return 0;
		1303	}
		1304
		1305	/*****************************************************************************/
		1306	/* requests implemented outside eio_execute, because they are so large */
		1307
		1308	static void
		1309	eio__realpath (eio_req *req, etp_worker *self)
		1310	{
		1311	char *rel = req->ptr1;
		1312	char *res;
		1313	char *tmp1, *tmp2;
		1314	#if SYMLOOP_MAX > 32
		1315	int symlinks = SYMLOOP_MAX;
		1316	#else
		1317	int symlinks = 32;
		1318	#endif
		1319
		1320	req->result = -1;
		1321
		1322	errno = EINVAL;
		1323	if (!rel)
		1324	return;
		1325
		1326	errno = ENOENT;
		1327	if (!*rel)
		1328	return;
		1329
		1330	if (!req->ptr2)
		1331	{
		1332	X_LOCK (wrklock);
		1333	req->flags |= EIO_FLAG_PTR2_FREE;
		1334	X_UNLOCK (wrklock);
		1335	req->ptr2 = malloc (PATH_MAX * 3);
		1336
		1337	errno = ENOMEM;
		1338	if (!req->ptr2)
		1339	return;
		1340	}
		1341
		1342	res = req->ptr2;
		1343	tmp1 = res + PATH_MAX;
		1344	tmp2 = tmp1 + PATH_MAX;
		1345
		1346	#if 0 /* disabled, the musl way to do things is just too racy */
		1347	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1348	/* on linux we may be able to ask the kernel */
		1349	{
		1350	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1351
		1352	if (fd >= 0)
		1353	{
		1354	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1355	req->result = readlink (tmp1, res, PATH_MAX);
		1356	close (fd);
		1357
		1358	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1359
		1360	if (req->result > 0)
		1361	goto done;
		1362	}
		1363	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1364	return;
		1365	}
		1366	#endif
		1367	#endif
		1368
		1369	if (*rel != '/')
		1370	{
		1371	if (!getcwd (res, PATH_MAX))
		1372	return;
		1373
		1374	if (res [1]) /* only use if not / */
		1375	res += strlen (res);
		1376	}
		1377
		1378	while (*rel)
		1379	{
		1380	eio_ssize_t len, linklen;
		1381	char *beg = rel;
		1382
		1383	while (*rel && *rel != '/')
		1384	++rel;
		1385
		1386	len = rel - beg;
		1387
		1388	if (!len) /* skip slashes */
		1389	{
		1390	++rel;
		1391	continue;
		1392	}
		1393
		1394	if (beg [0] == '.')
		1395	{
		1396	if (len == 1)
		1397	continue; /* . - nop */
		1398
		1399	if (beg [1] == '.' && len == 2)
		1400	{
		1401	/* .. - back up one component, if possible */
		1402
		1403	while (res != req->ptr2)
		1404	if (*--res == '/')
		1405	break;
		1406
		1407	continue;
		1408	}
		1409	}
		1410
		1411	errno = ENAMETOOLONG;
		1412	if (res + 1 + len + 1 >= tmp1)
		1413	return;
		1414
		1415	/* copy one component */
		1416	*res = '/';
		1417	memcpy (res + 1, beg, len);
		1418
		1419	/* zero-terminate, for readlink */
		1420	res [len + 1] = 0;
		1421
		1422	/* now check if it's a symlink */
		1423	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1424
		1425	if (linklen < 0)
		1426	{
		1427	if (errno != EINVAL)
		1428	return;
		1429
		1430	/* it's a normal directory. hopefully */
		1431	res += len + 1;
		1432	}
		1433	else
		1434	{
		1435	/* yay, it was a symlink - build new path in tmp2 */
		1436	int rellen = strlen (rel);
		1437
		1438	errno = ENAMETOOLONG;
		1439	if (linklen + 1 + rellen >= PATH_MAX)
		1440	return;
		1441
		1442	errno = ELOOP;
		1443	if (!--symlinks)
		1444	return;
		1445
		1446	if (*tmp1 == '/')
		1447	res = req->ptr2; /* symlink resolves to an absolute path */
		1448
		1449	/* we need to be careful, as rel might point into tmp2 already */
		1450	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1451	tmp2 [linklen] = '/';
		1452	memcpy (tmp2, tmp1, linklen);
		1453
		1454	rel = tmp2;
		1455	}
		1456	}
		1457
		1458	/* special case for the lone root path */
		1459	if (res == req->ptr2)
		1460	*res++ = '/';
		1461
		1462	req->result = res - (char *)req->ptr2;
		1463
		1464	done:
		1465	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1089	}	1466	}
1090		1467
1091	static signed char	1468	static signed char
1092	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1469	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1093	{	1470	{
…		…
1101		1478
1102	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1479	#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 */	1480	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1104		1481
1105	static void	1482	static void
1106	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1483	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1107	{	1484	{
1108	unsigned char bits [9 + sizeof (ino_t) * 8];	1485	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1109	unsigned char *bit = bits;	1486	unsigned char *bit = bits;
1110		1487
1111	assert (CHAR_BIT == 8);	1488	assert (CHAR_BIT == 8);
1112	assert (sizeof (eio_dirent) * 8 < 256);	1489	assert (sizeof (eio_dirent) * 8 < 256);
1113	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1490	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1115		1492
1116	if (size <= EIO_SORT_FAST)	1493	if (size <= EIO_SORT_FAST)
1117	return;	1494	return;
1118		1495
1119	/* first prepare an array of bits to test in our radix sort */	1496	/* 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 */	1497	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1121	/* inode_bits must contain all inodes ORed together */	1498	/* inode_bits must contain all inodes ORed together */
1122	/* which is used to skip bits that are 0 everywhere, which is very common */	1499	/* which is used to skip bits that are 0 everywhere, which is very common */
1123	{	1500	{
1124	ino_t endianness;	1501	eio_ino_t endianness;
1125	int i, j;	1502	int i, j;
1126		1503
1127	/* we store the byte offset of byte n into byte n of "endianness" */	1504	/* we store the byte offset of byte n into byte n of "endianness" */
1128	for (i = 0; i < sizeof (ino_t); ++i)	1505	for (i = 0; i < sizeof (eio_ino_t); ++i)
1129	((unsigned char *)&endianness)[i] = i;	1506	((unsigned char *)&endianness)[i] = i;
1130		1507
1131	*bit++ = 0;	1508	*bit++ = 0;
1132		1509
1133	for (i = 0; i < sizeof (ino_t); ++i)	1510	for (i = 0; i < sizeof (eio_ino_t); ++i)
1134	{	1511	{
1135	/* shifting off the byte offsets out of "endianness" */	1512	/* shifting off the byte offsets out of "endianness" */
1136	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1513	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1137	endianness >>= 8;	1514	endianness >>= 8;
1138		1515
1139	for (j = 0; j < 8; ++j)	1516	for (j = 0; j < 8; ++j)
1140	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1517	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1141	*bit++ = offs + j;	1518	*bit++ = offs + j;
1142	}	1519	}
1143		1520
1144	for (j = 0; j < 8; ++j)	1521	for (j = 0; j < 8; ++j)
1145	if (score_bits & (1 << j))	1522	if (score_bits & (1 << j))
1146	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1523	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1147	}	1524	}
1148		1525
1149	/* now actually do the sorting (a variant of MSD radix sort) */	1526	/* now actually do the sorting (a variant of MSD radix sort) */
1150	{	1527	{
1151	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1528	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1152	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1529	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1153	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1530	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1154	int stk_idx = 0;	1531	int stk_idx = 0;
1155		1532
1156	base_stk [stk_idx] = dents;	1533	base_stk [stk_idx] = dents;
1157	end_stk [stk_idx] = dents + size;	1534	end_stk [stk_idx] = dents + size;
1158	bit_stk [stk_idx] = bit - 1;	1535	bit_stk [stk_idx] = bit - 1;
…		…
1237	}	1614	}
1238	}	1615	}
1239	}	1616	}
1240		1617
1241	static void	1618	static void
1242	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1619	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1243	{	1620	{
1244	if (size <= 1)	1621	if (size <= 1)
1245	return; /* our insertion sort relies on size > 0 */	1622	return; /* our insertion sort relies on size > 0 */
1246		1623
1247	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1624	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1255		1632
1256	/* read a full directory */	1633	/* read a full directory */
1257	static void	1634	static void
1258	eio__scandir (eio_req *req, etp_worker *self)	1635	eio__scandir (eio_req *req, etp_worker *self)
1259	{	1636	{
1260	DIR *dirp;
1261	EIO_STRUCT_DIRENT *entp;
1262	char *name, *names;	1637	char *name, *names;
1263	int namesalloc = 4096;	1638	int namesalloc = 4096 - sizeof (void *) * 4;
1264	int namesoffs = 0;	1639	int namesoffs = 0;
1265	int flags = req->int1;	1640	int flags = req->int1;
1266	eio_dirent *dents = 0;	1641	eio_dirent *dents = 0;
1267	int dentalloc = 128;	1642	int dentalloc = 128;
1268	int dentoffs = 0;	1643	int dentoffs = 0;
1269	ino_t inode_bits = 0;	1644	eio_ino_t inode_bits = 0;
		1645	#ifdef _WIN32
		1646	HANDLE dirp;
		1647	WIN32_FIND_DATA entp;
		1648	#else
		1649	DIR *dirp;
		1650	EIO_STRUCT_DIRENT *entp;
		1651	#endif
1270		1652
1271	req->result = -1;	1653	req->result = -1;
1272		1654
1273	if (!(flags & EIO_READDIR_DENTS))	1655	if (!(flags & EIO_READDIR_DENTS))
1274	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1656	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1275		1657
1276	X_LOCK (wrklock);	1658	#ifdef _WIN32
1277	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1659	{
		1660	int len = strlen ((const char *)req->ptr1);
		1661	char *path = malloc (MAX_PATH);
		1662	const char *fmt;
		1663
		1664	if (!len)
		1665	fmt = "./*";
		1666	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1667	fmt = "%s*";
		1668	else
		1669	fmt = "%s/*";
		1670
		1671	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1672	dirp = FindFirstFile (path, &entp);
		1673	free (path);
		1674
		1675	if (dirp == INVALID_HANDLE_VALUE)
		1676	{
		1677	dirp = 0;
		1678
		1679	switch (GetLastError ())
		1680	{
		1681	case ERROR_FILE_NOT_FOUND:
		1682	req->result = 0;
		1683	break;
		1684
		1685	case ERROR_INVALID_NAME:
		1686	case ERROR_PATH_NOT_FOUND:
		1687	case ERROR_NO_MORE_FILES:
		1688	errno = ENOENT;
		1689	break;
		1690
		1691	case ERROR_NOT_ENOUGH_MEMORY:
		1692	errno = ENOMEM;
		1693	break;
		1694
		1695	default:
		1696	errno = EINVAL;
		1697	break;
		1698	}
		1699	}
		1700	}
		1701	#else
1278	self->dirp = dirp = opendir (req->ptr1);	1702	dirp = opendir (req->ptr1);
		1703	#endif
		1704
		1705	if (req->flags & EIO_FLAG_PTR1_FREE)
		1706	free (req->ptr1);
1279		1707
1280	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1708	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1281	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1709	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1282	req->ptr2 = names = malloc (namesalloc);	1710	req->ptr2 = names = malloc (namesalloc);
1283	X_UNLOCK (wrklock);
1284		1711
1285	if (dirp && names && (!flags || dents))	1712	if (dirp && names && (!flags || dents))
1286	for (;;)	1713	for (;;)
1287	{	1714	{
		1715	int done;
		1716
		1717	#ifdef _WIN32
		1718	done = !dirp;
		1719	#else
1288	errno = 0;	1720	errno = 0;
1289	entp = readdir (dirp);	1721	entp = readdir (dirp);
		1722	done = !entp;
		1723	#endif
1290		1724
1291	if (!entp)	1725	if (done)
1292	{	1726	{
		1727	#ifndef _WIN32
		1728	int old_errno = errno;
		1729	closedir (dirp);
		1730	errno = old_errno;
		1731
1293	if (errno)	1732	if (errno)
1294	break;	1733	break;
		1734	#endif
1295		1735
1296	/* sort etc. */	1736	/* sort etc. */
1297	req->int1 = flags;	1737	req->int1 = flags;
1298	req->result = dentoffs;	1738	req->result = dentoffs;
1299		1739
…		…
1328		1768
1329	break;	1769	break;
1330	}	1770	}
1331		1771
1332	/* now add the entry to our list(s) */	1772	/* now add the entry to our list(s) */
1333	name = entp->d_name;	1773	name = D_NAME (entp);
1334		1774
1335	/* skip . and .. entries */	1775	/* skip . and .. entries */
1336	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1776	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1337	{	1777	{
1338	int len = D_NAMLEN (entp) + 1;	1778	int len = D_NAMLEN (entp) + 1;
1339		1779
1340	while (expect_false (namesoffs + len > namesalloc))	1780	while (ecb_expect_false (namesoffs + len > namesalloc))
1341	{	1781	{
1342	namesalloc *= 2;	1782	namesalloc *= 2;
1343	X_LOCK (wrklock);
1344	req->ptr2 = names = realloc (names, namesalloc);	1783	req->ptr2 = names = realloc (names, namesalloc);
1345	X_UNLOCK (wrklock);
1346		1784
1347	if (!names)	1785	if (!names)
1348	break;	1786	break;
1349	}	1787	}
1350		1788
…		…
1352		1790
1353	if (dents)	1791	if (dents)
1354	{	1792	{
1355	struct eio_dirent *ent;	1793	struct eio_dirent *ent;
1356		1794
1357	if (expect_false (dentoffs == dentalloc))	1795	if (ecb_expect_false (dentoffs == dentalloc))
1358	{	1796	{
1359	dentalloc *= 2;	1797	dentalloc *= 2;
1360	X_LOCK (wrklock);
1361	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1798	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1362	X_UNLOCK (wrklock);
1363		1799
1364	if (!dents)	1800	if (!dents)
1365	break;	1801	break;
1366	}	1802	}
1367		1803
…		…
1447	if (EIO_CANCELLED (req))	1883	if (EIO_CANCELLED (req))
1448	{	1884	{
1449	errno = ECANCELED;	1885	errno = ECANCELED;
1450	break;	1886	break;
1451	}	1887	}
		1888
		1889	#ifdef _WIN32
		1890	if (!FindNextFile (dirp, &entp))
		1891	{
		1892	FindClose (dirp);
		1893	dirp = 0;
		1894	}
		1895	#endif
1452	}	1896	}
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	}	1897	}
1570		1898
1571	/*****************************************************************************/	1899	/*****************************************************************************/
1572		1900
1573	#define ALLOC(len) \	1901	#define ALLOC(len) \
…		…
1589	{	1917	{
1590	ETP_REQ *req;	1918	ETP_REQ *req;
1591	struct timespec ts;	1919	struct timespec ts;
1592	etp_worker *self = (etp_worker *)thr_arg;	1920	etp_worker *self = (etp_worker *)thr_arg;
1593		1921
		1922	#if HAVE_PRCTL_SET_NAME
		1923	prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
		1924	#endif
		1925
1594	/* try to distribute timeouts somewhat randomly */	1926	/* try to distribute timeouts somewhat evenly */
1595	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1927	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1596		1928
1597	for (;;)	1929	for (;;)
1598	{	1930	{
		1931	ts.tv_sec = 0;
		1932
1599	X_LOCK (reqlock);	1933	X_LOCK (reqlock);
1600		1934
1601	for (;;)	1935	for (;;)
1602	{	1936	{
1603	self->req = req = reqq_shift (&req_queue);	1937	self->req = req = reqq_shift (&req_queue);
1604		1938
1605	if (req)	1939	if (req)
1606	break;	1940	break;
1607		1941
		1942	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1943	{
		1944	X_UNLOCK (reqlock);
		1945	X_LOCK (wrklock);
		1946	--started;
		1947	X_UNLOCK (wrklock);
		1948	goto quit;
		1949	}
		1950
1608	++idle;	1951	++idle;
1609		1952
1610	ts.tv_sec = time (0) + idle_timeout;	1953	if (idle <= max_idle)
1611	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1954	/* we are allowed to idle, so do so without any timeout */
		1955	X_COND_WAIT (reqwait, reqlock);
		1956	else
1612	{	1957	{
1613	if (idle > max_idle)	1958	/* initialise timeout once */
1614	{	1959	if (!ts.tv_sec)
1615	--idle;	1960	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		1961
1623	/* we are allowed to idle, so do so without any timeout */
1624	X_COND_WAIT (reqwait, reqlock);	1962	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1963	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1625	}	1964	}
1626		1965
1627	--idle;	1966	--idle;
1628	}	1967	}
1629		1968
…		…
1632	X_UNLOCK (reqlock);	1971	X_UNLOCK (reqlock);
1633		1972
1634	if (req->type < 0)	1973	if (req->type < 0)
1635	goto quit;	1974	goto quit;
1636		1975
1637	if (!EIO_CANCELLED (req))
1638	ETP_EXECUTE (self, req);	1976	ETP_EXECUTE (self, req);
1639		1977
1640	X_LOCK (reslock);	1978	X_LOCK (reslock);
1641		1979
1642	++npending;	1980	++npending;
1643		1981
…		…
1658	return 0;	1996	return 0;
1659	}	1997	}
1660		1998
1661	/*****************************************************************************/	1999	/*****************************************************************************/
1662		2000
		2001	int ecb_cold
1663	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	2002	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1664	{	2003	{
		2004	#if !HAVE_PREADWRITE
		2005	X_MUTEX_CREATE (preadwritelock);
		2006	#endif
		2007
1665	return etp_init (want_poll, done_poll);	2008	return etp_init (want_poll, done_poll);
1666	}	2009	}
1667		2010
		2011	ecb_inline void
1668	static void eio_api_destroy (eio_req *req)	2012	eio_api_destroy (eio_req *req)
1669	{	2013	{
1670	free (req);	2014	free (req);
1671	}	2015	}
1672		2016
1673	#define REQ(rtype) \	2017	#define REQ(rtype) \
…		…
1692	{ \	2036	{ \
1693	eio_api_destroy (req); \	2037	eio_api_destroy (req); \
1694	return 0; \	2038	return 0; \
1695	}	2039	}
1696		2040
		2041	static void
1697	static void eio_execute (etp_worker *self, eio_req *req)	2042	eio_execute (etp_worker *self, eio_req *req)
1698	{	2043	{
		2044	if (ecb_expect_false (EIO_CANCELLED (req)))
		2045	{
		2046	req->result = -1;
		2047	req->errorno = ECANCELED;
		2048	return;
		2049	}
		2050
1699	switch (req->type)	2051	switch (req->type)
1700	{	2052	{
1701	case EIO_READ: ALLOC (req->size);	2053	case EIO_READ: ALLOC (req->size);
1702	req->result = req->offs >= 0	2054	req->result = req->offs >= 0
1703	? pread (req->int1, req->ptr2, req->size, req->offs)	2055	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1705	case EIO_WRITE: req->result = req->offs >= 0	2057	case EIO_WRITE: req->result = req->offs >= 0
1706	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2058	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1707	: write (req->int1, req->ptr2, req->size); break;	2059	: write (req->int1, req->ptr2, req->size); break;
1708		2060
1709	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2061	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;	2062	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1711		2063
1712	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2064	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1713	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2065	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1714	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2066	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1715	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2067	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
…		…
1737	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2089	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1738	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2090	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1739	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2091	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;	2092	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1741		2093
		2094	case EIO_REALPATH: eio__realpath (req, self); break;
		2095
1742	case EIO_READLINK: ALLOC (PATH_MAX);	2096	case EIO_READLINK: ALLOC (PATH_MAX);
1743	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;	2097	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1744		2098
1745	case EIO_SYNC: req->result = 0; sync (); break;	2099	case EIO_SYNC: req->result = 0; sync (); break;
1746	case EIO_FSYNC: req->result = fsync (req->int1); break;	2100	case EIO_FSYNC: req->result = fsync (req->int1); break;
1747	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2101	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2102	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2103	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1748	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2104	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1749	case EIO_MTOUCH: req->result = eio__mtouch (req); break;	2105	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1750	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2106	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1751	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2107	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;	2108	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1753		2109
1754	case EIO_READDIR: eio__scandir (req, self); break;	2110	case EIO_READDIR: eio__scandir (req, self); break;
1755		2111
1756	case EIO_BUSY:	2112	case EIO_BUSY:
1757	#ifdef _WIN32	2113	#ifdef _WIN32
…		…
1837	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2193	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1838	{	2194	{
1839	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2195	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1840	}	2196	}
1841		2197
		2198	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		2199	{
		2200	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2201	}
		2202
		2203	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2204	{
		2205	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2206	}
		2207
		2208	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2209	{
		2210	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2211	}
		2212
1842	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2213	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1843	{	2214	{
1844	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2215	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1845	}	2216	}
1846		2217
…		…
1852	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)	2223	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1853	{	2224	{
1854	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;	2225	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1855	}	2226	}
1856		2227
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)	2228	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1858	{	2229	{
1859	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2230	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1860	}
1861
1862	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1863	{
1864	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1865	}	2231	}
1866		2232
1867	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2233	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1868	{	2234	{
1869	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2235	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
1907	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2273	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1908	{	2274	{
1909	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2275	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1910	}	2276	}
1911		2277
1912	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2278	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1913	{	2279	{
1914	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2280	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1915	}	2281	}
1916		2282
1917	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2283	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)	2303	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1938	{	2304	{
1939	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2305	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1940	}	2306	}
1941		2307
1942	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2308	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1943	{	2309	{
1944	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2310	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1945	}	2311	}
1946		2312
1947	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2313	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1961	}	2327	}
1962		2328
1963	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2329	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1964	{	2330	{
1965	return eio__1path (EIO_READLINK, path, pri, cb, data);	2331	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2332	}
		2333
		2334	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2335	{
		2336	return eio__1path (EIO_REALPATH, path, pri, cb, data);
1966	}	2337	}
1967		2338
1968	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2339	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1969	{	2340	{
1970	return eio__1path (EIO_STAT, path, pri, cb, data);	2341	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)	2400	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2030	{	2401	{
2031	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2402	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2032	}	2403	}
2033		2404
2034	eio_req *eio_custom (void (*)(eio_req *) execute, int pri, eio_cb cb, void *data);	2405	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2035	{	2406	{
2036	REQ (EIO_CUSTOM); req->feed = execute; SEND;	2407	REQ (EIO_CUSTOM); req->feed = execute; SEND;
2037	}	2408	}
2038		2409
2039	#endif	2410	#endif
…		…
2050	#undef SEND	2421	#undef SEND
2051		2422
2052	/*****************************************************************************/	2423	/*****************************************************************************/
2053	/* grp functions */	2424	/* grp functions */
2054		2425
		2426	void
2055	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2427	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2056	{	2428	{
2057	grp->int2 = limit;	2429	grp->int2 = limit;
2058	grp->feed = feed;	2430	grp->feed = feed;
2059		2431
2060	grp_try_feed (grp);	2432	grp_try_feed (grp);
2061	}	2433	}
2062		2434
		2435	void
2063	void eio_grp_limit (eio_req *grp, int limit)	2436	eio_grp_limit (eio_req *grp, int limit)
2064	{	2437	{
2065	grp->int2 = limit;	2438	grp->int2 = limit;
2066		2439
2067	grp_try_feed (grp);	2440	grp_try_feed (grp);
2068	}	2441	}
2069		2442
		2443	void
2070	void eio_grp_add (eio_req *grp, eio_req *req)	2444	eio_grp_add (eio_req *grp, eio_req *req)
2071	{	2445	{
2072	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2446	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2073		2447
2074	grp->flags |= EIO_FLAG_GROUPADD;	2448	grp->flags |= EIO_FLAG_GROUPADD;
2075		2449
…		…
2086	}	2460	}
2087		2461
2088	/*****************************************************************************/	2462	/*****************************************************************************/
2089	/* misc garbage */	2463	/* misc garbage */
2090		2464
		2465	eio_ssize_t
2091	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2466	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2092	{	2467	{
2093	etp_worker wrk;
2094	ssize_t ret;
2095
2096	wrk.dbuf = 0;
2097
2098	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2468	return eio__sendfile (ofd, ifd, offset, count);
2099
2100	if (wrk.dbuf)
2101	free (wrk.dbuf);
2102
2103	return ret;
2104	}	2469	}
2105		2470

Diff Legend

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