ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.65 by root, Mon May 30 12:56:50 2011 UTC vs.
Revision 1.93 by root, Mon Jul 18 01:27:03 2011 UTC

…		…
40	#ifndef _WIN32	40	#ifndef _WIN32
41	# include "config.h"	41	# include "config.h"
42	#endif	42	#endif
43		43
44	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
45		46
46	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
47	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
48	#endif	49	#endif
49	#include "xthread.h"	50	#include "xthread.h"
…		…
53	#include <stdlib.h>	54	#include <stdlib.h>
54	#include <string.h>	55	#include <string.h>
55	#include <errno.h>	56	#include <errno.h>
56	#include <sys/types.h>	57	#include <sys/types.h>
57	#include <sys/stat.h>	58	#include <sys/stat.h>
58	#include <sys/statvfs.h>
59	#include <limits.h>	59	#include <limits.h>
60	#include <fcntl.h>	60	#include <fcntl.h>
61	#include <assert.h>	61	#include <assert.h>
62		62
		63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
		64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
		65	#if HAVE_STDINT_H
		66	# include <stdint.h>
		67	#endif
		68
		69	#ifndef ECANCELED
		70	# define ECANCELED EDOM
		71	#endif
		72	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	static void eio_destroy (eio_req *req);
		77
63	#ifndef EIO_FINISH	78	#ifndef EIO_FINISH
64	# 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
65	#endif	80	#endif
66		81
67	#ifndef EIO_DESTROY	82	#ifndef EIO_DESTROY
…		…
70		85
71	#ifndef EIO_FEED	86	#ifndef EIO_FEED
72	# 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)
73	#endif	88	#endif
74		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
75	#ifdef _WIN32	101	#ifdef _WIN32
76		102
77	/*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
78	#else	159	#else
79		160
80	# include <sys/time.h>	161	#include <sys/time.h>
81	# include <sys/select.h>	162	#include <sys/select.h>
		163	#include <sys/statvfs.h>
82	# include <unistd.h>	164	#include <unistd.h>
83	# include <utime.h>	165	#include <utime.h>
84	# include <signal.h>	166	#include <signal.h>
85	# include <dirent.h>	167	#include <dirent.h>
86		168
87	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES	169	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
88	# include <sys/mman.h>	170	#include <sys/mman.h>
89	#endif	171	#endif
90		172
		173	#define D_NAME(entp) entp->d_name
		174
91	/* 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 */
92	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__	176	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
93	# define _DIRENT_HAVE_D_TYPE /* sigh */	177	#define _DIRENT_HAVE_D_TYPE /* sigh */
94	# define D_INO(de) (de)->d_fileno	178	#define D_INO(de) (de)->d_fileno
95	# define D_NAMLEN(de) (de)->d_namlen	179	#define D_NAMLEN(de) (de)->d_namlen
96	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600	180	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
97	# define D_INO(de) (de)->d_ino	181	#define D_INO(de) (de)->d_ino
98	# endif	182	#endif
99		183
100	#ifdef _D_EXACT_NAMLEN	184	#ifdef _D_EXACT_NAMLEN
101	# undef D_NAMLEN	185	#undef D_NAMLEN
102	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)	186	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
103	#endif	187	#endif
104		188
105	# ifdef _DIRENT_HAVE_D_TYPE	189	#ifdef _DIRENT_HAVE_D_TYPE
106	# define D_TYPE(de) (de)->d_type	190	#define D_TYPE(de) (de)->d_type
107	# endif	191	#endif
108		192
109	# ifndef EIO_STRUCT_DIRENT	193	#ifndef EIO_STRUCT_DIRENT
110	# define EIO_STRUCT_DIRENT struct dirent	194	#define EIO_STRUCT_DIRENT struct dirent
111	# endif	195	#endif
112		196
113	#endif	197	#endif
114		198
115	#if HAVE_SENDFILE	199	#if HAVE_SENDFILE
116	# if __linux	200	# if __linux
…		…
132	#endif	216	#endif
133	#ifndef D_INO	217	#ifndef D_INO
134	# define D_INO(de) 0	218	# define D_INO(de) 0
135	#endif	219	#endif
136	#ifndef D_NAMLEN	220	#ifndef D_NAMLEN
137	# define D_NAMLEN(de) strlen ((de)->d_name)	221	# define D_NAMLEN(entp) strlen (D_NAME (entp))
138	#endif	222	#endif
139		223
140	/* used for struct dirent, AIX doesn't provide it */	224	/* used for struct dirent, AIX doesn't provide it */
141	#ifndef NAME_MAX	225	#ifndef NAME_MAX
142	# define NAME_MAX 4096	226	# define NAME_MAX 4096
…		…
159	if (!eio_buf) \	243	if (!eio_buf) \
160	return -1;	244	return -1;
161		245
162	#define EIO_TICKS ((1000000 + 1023) >> 10)	246	#define EIO_TICKS ((1000000 + 1023) >> 10)
163		247
164	/*****************************************************************************/
165
166	#if __GNUC__ >= 3
167	# define expect(expr,value) __builtin_expect ((expr),(value))
168	#else
169	# define expect(expr,value) (expr)
170	#endif
171
172	#define expect_false(expr) expect ((expr) != 0, 0)
173	#define expect_true(expr) expect ((expr) != 0, 1)
174
175	/*****************************************************************************/
176
177	#define ETP_PRI_MIN EIO_PRI_MIN	248	#define ETP_PRI_MIN EIO_PRI_MIN
178	#define ETP_PRI_MAX EIO_PRI_MAX	249	#define ETP_PRI_MAX EIO_PRI_MAX
179		250
180	struct etp_worker;	251	struct etp_worker;
181		252
…		…
189	#define ETP_WORKER_CLEAR(req) \	260	#define ETP_WORKER_CLEAR(req) \
190	if (wrk->dbuf) \	261	if (wrk->dbuf) \
191	{ \	262	{ \
192	free (wrk->dbuf); \	263	free (wrk->dbuf); \
193	wrk->dbuf = 0; \	264	wrk->dbuf = 0; \
194	} \
195	\
196	if (wrk->dirp) \
197	{ \
198	closedir (wrk->dirp); \
199	wrk->dirp = 0; \
200	}	265	}
201		266
202	#define ETP_WORKER_COMMON \	267	#define ETP_WORKER_COMMON \
203	void *dbuf; \	268	void *dbuf;
204	DIR *dirp;
205		269
206	/*****************************************************************************/	270	/*****************************************************************************/
207		271
208	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	272	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209		273
210	/* calculate time difference in ~1/EIO_TICKS of a second */	274	/* calculate time difference in ~1/EIO_TICKS of a second */
		275	ecb_inline int
211	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	276	tvdiff (struct timeval *tv1, struct timeval *tv2)
212	{	277	{
213	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	278	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
214	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	279	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
215	}	280	}
216		281
…		…
237	/*	302	/*
238	* make our pread/pwrite emulation safe against themselves, but not against	303	* make our pread/pwrite emulation safe against themselves, but not against
239	* normal read/write by using a mutex. slows down execution a lot,	304	* normal read/write by using a mutex. slows down execution a lot,
240	* but that's your problem, not mine.	305	* but that's your problem, not mine.
241	*/	306	*/
242	static xmutex_t preadwritelock = X_MUTEX_INIT;	307	static xmutex_t preadwritelock;
243	#endif	308	#endif
244		309
245	typedef struct etp_worker	310	typedef struct etp_worker
246	{	311	{
247	/* locked by wrklock */	312	/* locked by wrklock */
…		…
253	ETP_REQ *req; /* currently processed request */	318	ETP_REQ *req; /* currently processed request */
254		319
255	ETP_WORKER_COMMON	320	ETP_WORKER_COMMON
256	} etp_worker;	321	} etp_worker;
257		322
258	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	323	static etp_worker wrk_first; /* NOT etp */
259		324
260	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	325	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
261	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	326	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
262		327
263	/* worker threads management */	328	/* worker threads management */
264		329
		330	static void ecb_cold
265	static void etp_worker_clear (etp_worker *wrk)	331	etp_worker_clear (etp_worker *wrk)
266	{	332	{
267	ETP_WORKER_CLEAR (wrk);	333	ETP_WORKER_CLEAR (wrk);
268	}	334	}
269		335
		336	static void ecb_cold
270	static void etp_worker_free (etp_worker *wrk)	337	etp_worker_free (etp_worker *wrk)
271	{	338	{
272	wrk->next->prev = wrk->prev;	339	wrk->next->prev = wrk->prev;
273	wrk->prev->next = wrk->next;	340	wrk->prev->next = wrk->next;
274		341
275	free (wrk);	342	free (wrk);
276	}	343	}
277		344
278	static unsigned int etp_nreqs (void)	345	static unsigned int
		346	etp_nreqs (void)
279	{	347	{
280	int retval;	348	int retval;
281	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	349	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282	retval = nreqs;	350	retval = nreqs;
283	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	351	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284	return retval;	352	return retval;
285	}	353	}
286		354
287	static unsigned int etp_nready (void)	355	static unsigned int
		356	etp_nready (void)
288	{	357	{
289	unsigned int retval;	358	unsigned int retval;
290		359
291	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	360	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
292	retval = nready;	361	retval = nready;
293	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	362	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
294		363
295	return retval;	364	return retval;
296	}	365	}
297		366
298	static unsigned int etp_npending (void)	367	static unsigned int
		368	etp_npending (void)
299	{	369	{
300	unsigned int retval;	370	unsigned int retval;
301		371
302	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	372	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303	retval = npending;	373	retval = npending;
304	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	374	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
305		375
306	return retval;	376	return retval;
307	}	377	}
308		378
309	static unsigned int etp_nthreads (void)	379	static unsigned int
		380	etp_nthreads (void)
310	{	381	{
311	unsigned int retval;	382	unsigned int retval;
312		383
313	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	384	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
314	retval = started;	385	retval = started;
…		…
328	} etp_reqq;	399	} etp_reqq;
329		400
330	static etp_reqq req_queue;	401	static etp_reqq req_queue;
331	static etp_reqq res_queue;	402	static etp_reqq res_queue;
332		403
		404	static void ecb_noinline ecb_cold
		405	reqq_init (etp_reqq *q)
		406	{
		407	int pri;
		408
		409	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		410	q->qs[pri] = q->qe[pri] = 0;
		411
		412	q->size = 0;
		413	}
		414
		415	static int ecb_noinline
333	static int reqq_push (etp_reqq *q, ETP_REQ *req)	416	reqq_push (etp_reqq *q, ETP_REQ *req)
334	{	417	{
335	int pri = req->pri;	418	int pri = req->pri;
336	req->next = 0;	419	req->next = 0;
337		420
338	if (q->qe[pri])	421	if (q->qe[pri])
…		…
344	q->qe[pri] = q->qs[pri] = req;	427	q->qe[pri] = q->qs[pri] = req;
345		428
346	return q->size++;	429	return q->size++;
347	}	430	}
348		431
		432	static ETP_REQ * ecb_noinline
349	static ETP_REQ *reqq_shift (etp_reqq *q)	433	reqq_shift (etp_reqq *q)
350	{	434	{
351	int pri;	435	int pri;
352		436
353	if (!q->size)	437	if (!q->size)
354	return 0;	438	return 0;
…		…
369	}	453	}
370		454
371	abort ();	455	abort ();
372	}	456	}
373		457
374	static void etp_thread_init (void)	458	static int ecb_cold
		459	etp_init (void (*want_poll)(void), void (*done_poll)(void))
375	{	460	{
376	X_MUTEX_CREATE (wrklock);	461	X_MUTEX_CREATE (wrklock);
377	X_MUTEX_CREATE (reslock);	462	X_MUTEX_CREATE (reslock);
378	X_MUTEX_CREATE (reqlock);	463	X_MUTEX_CREATE (reqlock);
379	X_COND_CREATE (reqwait);	464	X_COND_CREATE (reqwait);
380	}
381		465
382	static void etp_atfork_prepare (void)	466	reqq_init (&req_queue);
383	{	467	reqq_init (&res_queue);
384	X_LOCK (wrklock);
385	X_LOCK (reqlock);
386	X_LOCK (reslock);
387	#if !HAVE_PREADWRITE
388	X_LOCK (preadwritelock);
389	#endif
390	}
391		468
392	static void etp_atfork_parent (void)	469	wrk_first.next =
393	{	470	wrk_first.prev = &wrk_first;
394	#if !HAVE_PREADWRITE
395	X_UNLOCK (preadwritelock);
396	#endif
397	X_UNLOCK (reslock);
398	X_UNLOCK (reqlock);
399	X_UNLOCK (wrklock);
400	}
401
402	static void etp_atfork_child (void)
403	{
404	ETP_REQ *prv;
405
406	while ((prv = reqq_shift (&req_queue)))
407	ETP_DESTROY (prv);
408
409	while ((prv = reqq_shift (&res_queue)))
410	ETP_DESTROY (prv);
411
412	while (wrk_first.next != &wrk_first)
413	{
414	etp_worker *wrk = wrk_first.next;
415
416	if (wrk->req)
417	ETP_DESTROY (wrk->req);
418
419	etp_worker_clear (wrk);
420	etp_worker_free (wrk);
421	}
422		471
423	started = 0;	472	started = 0;
424	idle = 0;	473	idle = 0;
425	nreqs = 0;	474	nreqs = 0;
426	nready = 0;	475	nready = 0;
427	npending = 0;	476	npending = 0;
428		477
429	etp_thread_init ();
430	}
431
432	static void
433	etp_once_init (void)
434	{
435	etp_thread_init ();
436	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
437	}
438
439	static int
440	etp_init (void (*want_poll)(void), void (*done_poll)(void))
441	{
442	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
443
444	pthread_once (&doinit, etp_once_init);
445
446	want_poll_cb = want_poll;	478	want_poll_cb = want_poll;
447	done_poll_cb = done_poll;	479	done_poll_cb = done_poll;
448		480
449	return 0;	481	return 0;
450	}	482	}
451		483
452	X_THREAD_PROC (etp_proc);	484	X_THREAD_PROC (etp_proc);
453		485
		486	static void ecb_cold
454	static void etp_start_thread (void)	487	etp_start_thread (void)
455	{	488	{
456	etp_worker *wrk = calloc (1, sizeof (etp_worker));	489	etp_worker *wrk = calloc (1, sizeof (etp_worker));
457		490
458	/*TODO*/	491	/*TODO*/
459	assert (("unable to allocate worker thread data", wrk));	492	assert (("unable to allocate worker thread data", wrk));
…		…
472	free (wrk);	505	free (wrk);
473		506
474	X_UNLOCK (wrklock);	507	X_UNLOCK (wrklock);
475	}	508	}
476		509
		510	static void
477	static void etp_maybe_start_thread (void)	511	etp_maybe_start_thread (void)
478	{	512	{
479	if (expect_true (etp_nthreads () >= wanted))	513	if (ecb_expect_true (etp_nthreads () >= wanted))
480	return;	514	return;
481		515
482	/* todo: maybe use idle here, but might be less exact */	516	/* todo: maybe use idle here, but might be less exact */
483	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	517	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
484	return;	518	return;
485		519
486	etp_start_thread ();	520	etp_start_thread ();
487	}	521	}
488		522
		523	static void ecb_cold
489	static void etp_end_thread (void)	524	etp_end_thread (void)
490	{	525	{
491	eio_req *req = calloc (1, sizeof (eio_req));	526	eio_req *req = calloc (1, sizeof (eio_req));
492		527
493	req->type = -1;	528	req->type = -1;
494	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	529	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
501	X_LOCK (wrklock);	536	X_LOCK (wrklock);
502	--started;	537	--started;
503	X_UNLOCK (wrklock);	538	X_UNLOCK (wrklock);
504	}	539	}
505		540
506	static int etp_poll (void)	541	static int
		542	etp_poll (void)
507	{	543	{
508	unsigned int maxreqs;	544	unsigned int maxreqs;
509	unsigned int maxtime;	545	unsigned int maxtime;
510	struct timeval tv_start, tv_now;	546	struct timeval tv_start, tv_now;
511		547
…		…
541		577
542	X_LOCK (reqlock);	578	X_LOCK (reqlock);
543	--nreqs;	579	--nreqs;
544	X_UNLOCK (reqlock);	580	X_UNLOCK (reqlock);
545		581
546	if (expect_false (req->type == EIO_GROUP && req->size))	582	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
547	{	583	{
548	req->int1 = 1; /* mark request as delayed */	584	req->int1 = 1; /* mark request as delayed */
549	continue;	585	continue;
550	}	586	}
551	else	587	else
552	{	588	{
553	int res = ETP_FINISH (req);	589	int res = ETP_FINISH (req);
554	if (expect_false (res))	590	if (ecb_expect_false (res))
555	return res;	591	return res;
556	}	592	}
557		593
558	if (expect_false (maxreqs && !--maxreqs))	594	if (ecb_expect_false (maxreqs && !--maxreqs))
559	break;	595	break;
560		596
561	if (maxtime)	597	if (maxtime)
562	{	598	{
563	gettimeofday (&tv_now, 0);	599	gettimeofday (&tv_now, 0);
…		…
569		605
570	errno = EAGAIN;	606	errno = EAGAIN;
571	return -1;	607	return -1;
572	}	608	}
573		609
		610	static void
574	static void etp_cancel (ETP_REQ *req)	611	etp_cancel (ETP_REQ *req)
575	{	612	{
576	X_LOCK (wrklock);	613	req->cancelled = 1;
577	req->flags |= EIO_FLAG_CANCELLED;
578	X_UNLOCK (wrklock);
579		614
580	eio_grp_cancel (req);	615	eio_grp_cancel (req);
581	}	616	}
582		617
		618	static void
583	static void etp_submit (ETP_REQ *req)	619	etp_submit (ETP_REQ *req)
584	{	620	{
585	req->pri -= ETP_PRI_MIN;	621	req->pri -= ETP_PRI_MIN;
586		622
587	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	623	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
588	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	624	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
589		625
590	if (expect_false (req->type == EIO_GROUP))	626	if (ecb_expect_false (req->type == EIO_GROUP))
591	{	627	{
592	/* I hope this is worth it :/ */	628	/* I hope this is worth it :/ */
593	X_LOCK (reqlock);	629	X_LOCK (reqlock);
594	++nreqs;	630	++nreqs;
595	X_UNLOCK (reqlock);	631	X_UNLOCK (reqlock);
…		…
614		650
615	etp_maybe_start_thread ();	651	etp_maybe_start_thread ();
616	}	652	}
617	}	653	}
618		654
		655	static void ecb_cold
619	static void etp_set_max_poll_time (double nseconds)	656	etp_set_max_poll_time (double nseconds)
620	{	657	{
621	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	658	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
622	max_poll_time = nseconds * EIO_TICKS;	659	max_poll_time = nseconds * EIO_TICKS;
623	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	660	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
624	}	661	}
625		662
		663	static void ecb_cold
626	static void etp_set_max_poll_reqs (unsigned int maxreqs)	664	etp_set_max_poll_reqs (unsigned int maxreqs)
627	{	665	{
628	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	666	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
629	max_poll_reqs = maxreqs;	667	max_poll_reqs = maxreqs;
630	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	668	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
631	}	669	}
632		670
		671	static void ecb_cold
633	static void etp_set_max_idle (unsigned int nthreads)	672	etp_set_max_idle (unsigned int nthreads)
634	{	673	{
635	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	674	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
636	max_idle = nthreads;	675	max_idle = nthreads;
637	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	676	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
638	}	677	}
639		678
		679	static void ecb_cold
640	static void etp_set_idle_timeout (unsigned int seconds)	680	etp_set_idle_timeout (unsigned int seconds)
641	{	681	{
642	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	682	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
643	idle_timeout = seconds;	683	idle_timeout = seconds;
644	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	684	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
645	}	685	}
646		686
		687	static void ecb_cold
647	static void etp_set_min_parallel (unsigned int nthreads)	688	etp_set_min_parallel (unsigned int nthreads)
648	{	689	{
649	if (wanted < nthreads)	690	if (wanted < nthreads)
650	wanted = nthreads;	691	wanted = nthreads;
651	}	692	}
652		693
		694	static void ecb_cold
653	static void etp_set_max_parallel (unsigned int nthreads)	695	etp_set_max_parallel (unsigned int nthreads)
654	{	696	{
655	if (wanted > nthreads)	697	if (wanted > nthreads)
656	wanted = nthreads;	698	wanted = nthreads;
657		699
658	while (started > wanted)	700	while (started > wanted)
659	etp_end_thread ();	701	etp_end_thread ();
660	}	702	}
661		703
662	/*****************************************************************************/	704	/*****************************************************************************/
663		705
		706	static void
664	static void grp_try_feed (eio_req *grp)	707	grp_try_feed (eio_req *grp)
665	{	708	{
666	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	709	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
667	{	710	{
668	grp->flags &= ~EIO_FLAG_GROUPADD;	711	grp->flags &= ~EIO_FLAG_GROUPADD;
669		712
…		…
676	break;	719	break;
677	}	720	}
678	}	721	}
679	}	722	}
680		723
		724	static int
681	static int grp_dec (eio_req *grp)	725	grp_dec (eio_req *grp)
682	{	726	{
683	--grp->size;	727	--grp->size;
684		728
685	/* call feeder, if applicable */	729	/* call feeder, if applicable */
686	grp_try_feed (grp);	730	grp_try_feed (grp);
…		…
690	return eio_finish (grp);	734	return eio_finish (grp);
691	else	735	else
692	return 0;	736	return 0;
693	}	737	}
694		738
		739	static void
695	void eio_destroy (eio_req *req)	740	eio_destroy (eio_req *req)
696	{	741	{
697	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	742	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
698	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	743	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
699		744
700	EIO_DESTROY (req);	745	EIO_DESTROY (req);
701	}	746	}
702		747
		748	static int
703	static int eio_finish (eio_req *req)	749	eio_finish (eio_req *req)
704	{	750	{
705	int res = EIO_FINISH (req);	751	int res = EIO_FINISH (req);
706		752
707	if (req->grp)	753	if (req->grp)
708	{	754	{
…		…
716	if (grp->grp_first == req)	762	if (grp->grp_first == req)
717	grp->grp_first = req->grp_next;	763	grp->grp_first = req->grp_next;
718		764
719	res2 = grp_dec (grp);	765	res2 = grp_dec (grp);
720		766
721	if (!res && res2)	767	if (!res)
722	res = res2;	768	res = res2;
723	}	769	}
724		770
725	eio_destroy (req);	771	eio_destroy (req);
726		772
727	return res;	773	return res;
728	}	774	}
729		775
		776	void
730	void eio_grp_cancel (eio_req *grp)	777	eio_grp_cancel (eio_req *grp)
731	{	778	{
732	for (grp = grp->grp_first; grp; grp = grp->grp_next)	779	for (grp = grp->grp_first; grp; grp = grp->grp_next)
733	eio_cancel (grp);	780	eio_cancel (grp);
734	}	781	}
735		782
		783	void
736	void eio_cancel (eio_req *req)	784	eio_cancel (eio_req *req)
737	{	785	{
738	etp_cancel (req);	786	etp_cancel (req);
739	}	787	}
740		788
		789	void
741	void eio_submit (eio_req *req)	790	eio_submit (eio_req *req)
742	{	791	{
743	etp_submit (req);	792	etp_submit (req);
744	}	793	}
745		794
746	unsigned int eio_nreqs (void)	795	unsigned int
		796	eio_nreqs (void)
747	{	797	{
748	return etp_nreqs ();	798	return etp_nreqs ();
749	}	799	}
750		800
751	unsigned int eio_nready (void)	801	unsigned int
		802	eio_nready (void)
752	{	803	{
753	return etp_nready ();	804	return etp_nready ();
754	}	805	}
755		806
756	unsigned int eio_npending (void)	807	unsigned int
		808	eio_npending (void)
757	{	809	{
758	return etp_npending ();	810	return etp_npending ();
759	}	811	}
760		812
761	unsigned int eio_nthreads (void)	813	unsigned int ecb_cold
		814	eio_nthreads (void)
762	{	815	{
763	return etp_nthreads ();	816	return etp_nthreads ();
764	}	817	}
765		818
		819	void ecb_cold
766	void eio_set_max_poll_time (double nseconds)	820	eio_set_max_poll_time (double nseconds)
767	{	821	{
768	etp_set_max_poll_time (nseconds);	822	etp_set_max_poll_time (nseconds);
769	}	823	}
770		824
		825	void ecb_cold
771	void eio_set_max_poll_reqs (unsigned int maxreqs)	826	eio_set_max_poll_reqs (unsigned int maxreqs)
772	{	827	{
773	etp_set_max_poll_reqs (maxreqs);	828	etp_set_max_poll_reqs (maxreqs);
774	}	829	}
775		830
		831	void ecb_cold
776	void eio_set_max_idle (unsigned int nthreads)	832	eio_set_max_idle (unsigned int nthreads)
777	{	833	{
778	etp_set_max_idle (nthreads);	834	etp_set_max_idle (nthreads);
779	}	835	}
780		836
		837	void ecb_cold
781	void eio_set_idle_timeout (unsigned int seconds)	838	eio_set_idle_timeout (unsigned int seconds)
782	{	839	{
783	etp_set_idle_timeout (seconds);	840	etp_set_idle_timeout (seconds);
784	}	841	}
785		842
		843	void ecb_cold
786	void eio_set_min_parallel (unsigned int nthreads)	844	eio_set_min_parallel (unsigned int nthreads)
787	{	845	{
788	etp_set_min_parallel (nthreads);	846	etp_set_min_parallel (nthreads);
789	}	847	}
790		848
		849	void ecb_cold
791	void eio_set_max_parallel (unsigned int nthreads)	850	eio_set_max_parallel (unsigned int nthreads)
792	{	851	{
793	etp_set_max_parallel (nthreads);	852	etp_set_max_parallel (nthreads);
794	}	853	}
795		854
796	int eio_poll (void)	855	int eio_poll (void)
…		…
805	# undef pread	864	# undef pread
806	# undef pwrite	865	# undef pwrite
807	# define pread eio__pread	866	# define pread eio__pread
808	# define pwrite eio__pwrite	867	# define pwrite eio__pwrite
809		868
810	static ssize_t	869	static eio_ssize_t
811	eio__pread (int fd, void *buf, size_t count, off_t offset)	870	eio__pread (int fd, void *buf, size_t count, off_t offset)
812	{	871	{
813	ssize_t res;	872	eio_ssize_t res;
814	off_t ooffset;	873	off_t ooffset;
815		874
816	X_LOCK (preadwritelock);	875	X_LOCK (preadwritelock);
817	ooffset = lseek (fd, 0, SEEK_CUR);	876	ooffset = lseek (fd, 0, SEEK_CUR);
818	lseek (fd, offset, SEEK_SET);	877	lseek (fd, offset, SEEK_SET);
…		…
821	X_UNLOCK (preadwritelock);	880	X_UNLOCK (preadwritelock);
822		881
823	return res;	882	return res;
824	}	883	}
825		884
826	static ssize_t	885	static eio_ssize_t
827	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	886	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
828	{	887	{
829	ssize_t res;	888	eio_ssize_t res;
830	off_t ooffset;	889	off_t ooffset;
831		890
832	X_LOCK (preadwritelock);	891	X_LOCK (preadwritelock);
833	ooffset = lseek (fd, 0, SEEK_CUR);	892	ooffset = lseek (fd, 0, SEEK_CUR);
834	lseek (fd, offset, SEEK_SET);	893	lseek (fd, offset, SEEK_SET);
…		…
866	#ifndef HAVE_FUTIMES	925	#ifndef HAVE_FUTIMES
867		926
868	# undef futimes	927	# undef futimes
869	# define futimes(fd,times) eio__futimes (fd, times)	928	# define futimes(fd,times) eio__futimes (fd, times)
870		929
		930	static int
871	static int eio__futimes (int fd, const struct timeval tv[2])	931	eio__futimes (int fd, const struct timeval tv[2])
872	{	932	{
873	errno = ENOSYS;	933	errno = ENOSYS;
874	return -1;	934	return -1;
875	}	935	}
876		936
…		…
880	# undef fdatasync	940	# undef fdatasync
881	# define fdatasync(fd) fsync (fd)	941	# define fdatasync(fd) fsync (fd)
882	#endif	942	#endif
883		943
884	/* sync_file_range always needs emulation */	944	/* sync_file_range always needs emulation */
885	int	945	static int
886	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	946	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
887	{	947	{
888	#if HAVE_SYNC_FILE_RANGE	948	#if HAVE_SYNC_FILE_RANGE
889	int res;	949	int res;
890		950
…		…
907	/* even though we could play tricks with the flags, it's better to always	967	/* even though we could play tricks with the flags, it's better to always
908	* call fdatasync, as that matches the expectation of its users best */	968	* call fdatasync, as that matches the expectation of its users best */
909	return fdatasync (fd);	969	return fdatasync (fd);
910	}	970	}
911		971
		972	static int
		973	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		974	{
		975	#if HAVE_FALLOCATE
		976	return fallocate (fd, mode, offset, len);
		977	#else
		978	errno = ENOSYS;
		979	return -1;
		980	#endif
		981	}
		982
912	#if !HAVE_READAHEAD	983	#if !HAVE_READAHEAD
913	# undef readahead	984	# undef readahead
914	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	985	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
915		986
916	static ssize_t	987	static eio_ssize_t
917	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	988	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
918	{	989	{
919	size_t todo = count;	990	size_t todo = count;
920	dBUF;	991	dBUF;
921		992
…		…
933	}	1004	}
934		1005
935	#endif	1006	#endif
936		1007
937	/* sendfile always needs emulation */	1008	/* sendfile always needs emulation */
938	static ssize_t	1009	static eio_ssize_t
939	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1010	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
940	{	1011	{
		1012	eio_ssize_t written = 0;
941	ssize_t res;	1013	eio_ssize_t res;
942		1014
943	if (!count)	1015	if (!count)
944	return 0;	1016	return 0;
945		1017
		1018	for (;;)
		1019	{
		1020	#ifdef __APPLE__
		1021	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1022	#endif
946	#if HAVE_SENDFILE	1023	#if HAVE_SENDFILE
947	# if __linux	1024	# if __linux
		1025	off_t soffset = offset;
948	res = sendfile (ofd, ifd, &offset, count);	1026	res = sendfile (ofd, ifd, &soffset, count);
949		1027
950	# elif __FreeBSD__	1028	# elif __FreeBSD__
951	/*	1029	/*
952	* Of course, the freebsd sendfile is a dire hack with no thoughts	1030	* Of course, the freebsd sendfile is a dire hack with no thoughts
953	* wasted on making it similar to other I/O functions.	1031	* wasted on making it similar to other I/O functions.
954	*/	1032	*/
955	{
956	off_t sbytes;	1033	off_t sbytes;
957	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1034	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
958		1035
959	#if 0 /* according to the manpage, this is correct, but broken behaviour */	1036	#if 0 /* according to the manpage, this is correct, but broken behaviour */
960	/* freebsd' sendfile will return 0 on success */	1037	/* freebsd' sendfile will return 0 on success */
961	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */	1038	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
962	/* not on e.g. EIO or EPIPE - sounds broken */	1039	/* not on e.g. EIO or EPIPE - sounds broken */
963	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)	1040	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
964	res = sbytes;	1041	res = sbytes;
965	#endif	1042	#endif
966		1043
967	/* according to source inspection, this is correct, and useful behaviour */	1044	/* according to source inspection, this is correct, and useful behaviour */
968	if (sbytes)	1045	if (sbytes)
969	res = sbytes;	1046	res = sbytes;
970	}
971		1047
972	# elif defined (__APPLE__)	1048	# elif defined (__APPLE__)
973
974	{
975	off_t sbytes = count;	1049	off_t sbytes = count;
976	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	1050	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
977		1051
978	/* according to the manpage, sbytes is always valid */	1052	/* according to the manpage, sbytes is always valid */
979	if (sbytes)	1053	if (sbytes)
980	res = sbytes;	1054	res = sbytes;
981	}
982		1055
983	# elif __hpux	1056	# elif __hpux
984	res = sendfile (ofd, ifd, offset, count, 0, 0);	1057	res = sendfile (ofd, ifd, offset, count, 0, 0);
985		1058
986	# elif __solaris	1059	# elif __solaris
987	{
988	struct sendfilevec vec;	1060	struct sendfilevec vec;
989	size_t sbytes;	1061	size_t sbytes;
990		1062
991	vec.sfv_fd = ifd;	1063	vec.sfv_fd = ifd;
992	vec.sfv_flag = 0;	1064	vec.sfv_flag = 0;
993	vec.sfv_off = offset;	1065	vec.sfv_off = offset;
994	vec.sfv_len = count;	1066	vec.sfv_len = count;
995		1067
996	res = sendfilev (ofd, &vec, 1, &sbytes);	1068	res = sendfilev (ofd, &vec, 1, &sbytes);
997		1069
998	if (res < 0 && sbytes)	1070	if (res < 0 && sbytes)
999	res = sbytes;	1071	res = sbytes;
1000	}
1001		1072
1002	# endif	1073	# endif
1003		1074
1004	#elif defined (_WIN32)	1075	#elif defined (_WIN32) && 0
1005
1006	/* does not work, just for documentation of what would need to be done */	1076	/* does not work, just for documentation of what would need to be done */
1007	{	1077	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1078	/* libeio guarantees that the file offset does not change, and windows */
		1079	/* has no way to get an independent handle to the same file description */
1008	HANDLE h = TO_SOCKET (ifd);	1080	HANDLE h = TO_SOCKET (ifd);
1009	SetFilePointer (h, offset, 0, FILE_BEGIN);	1081	SetFilePointer (h, offset, 0, FILE_BEGIN);
1010	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);	1082	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1011	}
1012		1083
1013	#else	1084	#else
1014	res = -1;	1085	res = -1;
1015	errno = ENOSYS;	1086	errno = ENOSYS;
1016	#endif	1087	#endif
1017		1088
		1089	/* we assume sendfile can copy at least 128mb in one go */
		1090	if (res <= 128 * 1024 * 1024)
		1091	{
		1092	if (res > 0)
		1093	written += res;
		1094
		1095	if (written)
		1096	return written;
		1097
		1098	break;
		1099	}
		1100	else
		1101	{
		1102	/* if we requested more, then probably the kernel was lazy */
		1103	written += res;
		1104	offset += res;
		1105	count -= res;
		1106
		1107	if (!count)
		1108	return written;
		1109	}
		1110	}
		1111
1018	if (res < 0	1112	if (res < 0
1019	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1113	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1020	/* BSDs */	1114	/* BSDs */
1021	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1115	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1022	|| errno == ENOTSUP	1116	|| errno == ENOTSUP
1023	#endif	1117	#endif
		1118	#ifdef EOPNOTSUPP /* windows */
1024	|| errno == EOPNOTSUPP /* BSDs */	1119	|| errno == EOPNOTSUPP /* BSDs */
		1120	#endif
1025	#if __solaris	1121	#if __solaris
1026	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1122	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1027	#endif	1123	#endif
1028	)	1124	)
1029	)	1125	)
…		…
1033		1129
1034	res = 0;	1130	res = 0;
1035		1131
1036	while (count)	1132	while (count)
1037	{	1133	{
1038	ssize_t cnt;	1134	eio_ssize_t cnt;
1039		1135
1040	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1136	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1041		1137
1042	if (cnt <= 0)	1138	if (cnt <= 0)
1043	{	1139	{
…		…
1060	}	1156	}
1061		1157
1062	return res;	1158	return res;
1063	}	1159	}
1064		1160
		1161	#ifdef PAGESIZE
		1162	# define eio_pagesize() PAGESIZE
		1163	#else
		1164	static intptr_t
		1165	eio_pagesize (void)
		1166	{
		1167	static intptr_t page;
		1168
		1169	if (!page)
		1170	page = sysconf (_SC_PAGESIZE);
		1171
		1172	return page;
		1173	}
		1174	#endif
		1175
		1176	static void
		1177	eio_page_align (void **addr, size_t *length)
		1178	{
		1179	intptr_t mask = eio_pagesize () - 1;
		1180
		1181	/* round down addr */
		1182	intptr_t adj = mask & (intptr_t)*addr;
		1183
		1184	*addr = (void *)((intptr_t)*addr - adj);
		1185	*length += adj;
		1186
		1187	/* round up length */
		1188	*length = (*length + mask) & ~mask;
		1189	}
		1190
		1191	#if !_POSIX_MEMLOCK
		1192	# define eio__mlockall(a) EIO_ENOSYS ()
		1193	#else
		1194
		1195	static int
		1196	eio__mlockall (int flags)
		1197	{
		1198	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1199	extern int mallopt (int, int);
		1200	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1201	#endif
		1202
		1203	if (EIO_MCL_CURRENT != MCL_CURRENT
		1204	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1205	{
		1206	flags = 0
		1207	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1208	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1209	}
		1210
		1211	return mlockall (flags);
		1212	}
		1213	#endif
		1214
		1215	#if !_POSIX_MEMLOCK_RANGE
		1216	# define eio__mlock(a,b) EIO_ENOSYS ()
		1217	#else
		1218
		1219	static int
		1220	eio__mlock (void *addr, size_t length)
		1221	{
		1222	eio_page_align (&addr, &length);
		1223
		1224	return mlock (addr, length);
		1225	}
		1226
		1227	#endif
		1228
		1229	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1230	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1231	#else
		1232
		1233	static int
		1234	eio__msync (void *mem, size_t len, int flags)
		1235	{
		1236	eio_page_align (&mem, &len);
		1237
		1238	if (EIO_MS_ASYNC != MS_SYNC
		1239	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1240	|| EIO_MS_SYNC != MS_SYNC)
		1241	{
		1242	flags = 0
		1243	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1244	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1245	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1246	}
		1247
		1248	return msync (mem, len, flags);
		1249	}
		1250
		1251	#endif
		1252
		1253	static int
		1254	eio__mtouch (eio_req *req)
		1255	{
		1256	void *mem = req->ptr2;
		1257	size_t len = req->size;
		1258	int flags = req->int1;
		1259
		1260	eio_page_align (&mem, &len);
		1261
		1262	{
		1263	intptr_t addr = (intptr_t)mem;
		1264	intptr_t end = addr + len;
		1265	intptr_t page = eio_pagesize ();
		1266
		1267	if (addr < end)
		1268	if (flags & EIO_MT_MODIFY) /* modify */
		1269	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1270	else
		1271	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1272	}
		1273
		1274	return 0;
		1275	}
		1276
		1277	/*****************************************************************************/
		1278	/* requests implemented outside eio_execute, because they are so large */
		1279
		1280	static void
		1281	eio__realpath (eio_req *req, etp_worker *self)
		1282	{
		1283	char *rel = req->ptr1;
		1284	char *res;
		1285	char *tmp1, *tmp2;
		1286	#if SYMLOOP_MAX > 32
		1287	int symlinks = SYMLOOP_MAX;
		1288	#else
		1289	int symlinks = 32;
		1290	#endif
		1291
		1292	req->result = -1;
		1293
		1294	errno = EINVAL;
		1295	if (!rel)
		1296	return;
		1297
		1298	errno = ENOENT;
		1299	if (!*rel)
		1300	return;
		1301
		1302	if (!req->ptr2)
		1303	{
		1304	X_LOCK (wrklock);
		1305	req->flags |= EIO_FLAG_PTR2_FREE;
		1306	X_UNLOCK (wrklock);
		1307	req->ptr2 = malloc (PATH_MAX * 3);
		1308
		1309	errno = ENOMEM;
		1310	if (!req->ptr2)
		1311	return;
		1312	}
		1313
		1314	res = req->ptr2;
		1315	tmp1 = res + PATH_MAX;
		1316	tmp2 = tmp1 + PATH_MAX;
		1317
		1318	#if 0 /* disabled, the musl way to do things is just too racy */
		1319	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1320	/* on linux we may be able to ask the kernel */
		1321	{
		1322	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1323
		1324	if (fd >= 0)
		1325	{
		1326	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1327	req->result = readlink (tmp1, res, PATH_MAX);
		1328	close (fd);
		1329
		1330	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1331
		1332	if (req->result > 0)
		1333	goto done;
		1334	}
		1335	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1336	return;
		1337	}
		1338	#endif
		1339	#endif
		1340
		1341	if (*rel != '/')
		1342	{
		1343	if (!getcwd (res, PATH_MAX))
		1344	return;
		1345
		1346	if (res [1]) /* only use if not / */
		1347	res += strlen (res);
		1348	}
		1349
		1350	while (*rel)
		1351	{
		1352	eio_ssize_t len, linklen;
		1353	char *beg = rel;
		1354
		1355	while (*rel && *rel != '/')
		1356	++rel;
		1357
		1358	len = rel - beg;
		1359
		1360	if (!len) /* skip slashes */
		1361	{
		1362	++rel;
		1363	continue;
		1364	}
		1365
		1366	if (beg [0] == '.')
		1367	{
		1368	if (len == 1)
		1369	continue; /* . - nop */
		1370
		1371	if (beg [1] == '.' && len == 2)
		1372	{
		1373	/* .. - back up one component, if possible */
		1374
		1375	while (res != req->ptr2)
		1376	if (*--res == '/')
		1377	break;
		1378
		1379	continue;
		1380	}
		1381	}
		1382
		1383	errno = ENAMETOOLONG;
		1384	if (res + 1 + len + 1 >= tmp1)
		1385	return;
		1386
		1387	/* copy one component */
		1388	*res = '/';
		1389	memcpy (res + 1, beg, len);
		1390
		1391	/* zero-terminate, for readlink */
		1392	res [len + 1] = 0;
		1393
		1394	/* now check if it's a symlink */
		1395	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1396
		1397	if (linklen < 0)
		1398	{
		1399	if (errno != EINVAL)
		1400	return;
		1401
		1402	/* it's a normal directory. hopefully */
		1403	res += len + 1;
		1404	}
		1405	else
		1406	{
		1407	/* yay, it was a symlink - build new path in tmp2 */
		1408	int rellen = strlen (rel);
		1409
		1410	errno = ENAMETOOLONG;
		1411	if (linklen + 1 + rellen >= PATH_MAX)
		1412	return;
		1413
		1414	errno = ELOOP;
		1415	if (!--symlinks)
		1416	return;
		1417
		1418	if (*tmp1 == '/')
		1419	res = req->ptr2; /* symlink resolves to an absolute path */
		1420
		1421	/* we need to be careful, as rel might point into tmp2 already */
		1422	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1423	tmp2 [linklen] = '/';
		1424	memcpy (tmp2, tmp1, linklen);
		1425
		1426	rel = tmp2;
		1427	}
		1428	}
		1429
		1430	/* special case for the lone root path */
		1431	if (res == req->ptr2)
		1432	*res++ = '/';
		1433
		1434	req->result = res - (char *)req->ptr2;
		1435
		1436	done:
		1437	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1438	}
		1439
1065	static signed char	1440	static signed char
1066	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1441	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1067	{	1442	{
1068	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */	1443	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1069	: a->inode < b->inode ? -1	1444	: a->inode < b->inode ? -1
…		…
1075		1450
1076	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1451	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1077	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1452	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1078		1453
1079	static void	1454	static void
1080	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1455	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1081	{	1456	{
1082	unsigned char bits [9 + sizeof (ino_t) * 8];	1457	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1083	unsigned char *bit = bits;	1458	unsigned char *bit = bits;
1084		1459
1085	assert (CHAR_BIT == 8);	1460	assert (CHAR_BIT == 8);
1086	assert (sizeof (eio_dirent) * 8 < 256);	1461	assert (sizeof (eio_dirent) * 8 < 256);
1087	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1462	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1089		1464
1090	if (size <= EIO_SORT_FAST)	1465	if (size <= EIO_SORT_FAST)
1091	return;	1466	return;
1092		1467
1093	/* first prepare an array of bits to test in our radix sort */	1468	/* first prepare an array of bits to test in our radix sort */
1094	/* try to take endianness into account, as well as differences in ino_t sizes */	1469	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1095	/* inode_bits must contain all inodes ORed together */	1470	/* inode_bits must contain all inodes ORed together */
1096	/* which is used to skip bits that are 0 everywhere, which is very common */	1471	/* which is used to skip bits that are 0 everywhere, which is very common */
1097	{	1472	{
1098	ino_t endianness;	1473	eio_ino_t endianness;
1099	int i, j;	1474	int i, j;
1100		1475
1101	/* we store the byte offset of byte n into byte n of "endianness" */	1476	/* we store the byte offset of byte n into byte n of "endianness" */
1102	for (i = 0; i < sizeof (ino_t); ++i)	1477	for (i = 0; i < sizeof (eio_ino_t); ++i)
1103	((unsigned char *)&endianness)[i] = i;	1478	((unsigned char *)&endianness)[i] = i;
1104		1479
1105	*bit++ = 0;	1480	*bit++ = 0;
1106		1481
1107	for (i = 0; i < sizeof (ino_t); ++i)	1482	for (i = 0; i < sizeof (eio_ino_t); ++i)
1108	{	1483	{
1109	/* shifting off the byte offsets out of "endianness" */	1484	/* shifting off the byte offsets out of "endianness" */
1110	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1485	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1111	endianness >>= 8;	1486	endianness >>= 8;
1112		1487
1113	for (j = 0; j < 8; ++j)	1488	for (j = 0; j < 8; ++j)
1114	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1489	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1115	*bit++ = offs + j;	1490	*bit++ = offs + j;
1116	}	1491	}
1117		1492
1118	for (j = 0; j < 8; ++j)	1493	for (j = 0; j < 8; ++j)
1119	if (score_bits & (1 << j))	1494	if (score_bits & (1 << j))
1120	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1495	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1121	}	1496	}
1122		1497
1123	/* now actually do the sorting (a variant of MSD radix sort) */	1498	/* now actually do the sorting (a variant of MSD radix sort) */
1124	{	1499	{
1125	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1500	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1126	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1501	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1127	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1502	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1128	int stk_idx = 0;	1503	int stk_idx = 0;
1129		1504
1130	base_stk [stk_idx] = dents;	1505	base_stk [stk_idx] = dents;
1131	end_stk [stk_idx] = dents + size;	1506	end_stk [stk_idx] = dents + size;
1132	bit_stk [stk_idx] = bit - 1;	1507	bit_stk [stk_idx] = bit - 1;
…		…
1211	}	1586	}
1212	}	1587	}
1213	}	1588	}
1214		1589
1215	static void	1590	static void
1216	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1591	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1217	{	1592	{
1218	if (size <= 1)	1593	if (size <= 1)
1219	return; /* our insertion sort relies on size > 0 */	1594	return; /* our insertion sort relies on size > 0 */
1220		1595
1221	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1596	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1229		1604
1230	/* read a full directory */	1605	/* read a full directory */
1231	static void	1606	static void
1232	eio__scandir (eio_req *req, etp_worker *self)	1607	eio__scandir (eio_req *req, etp_worker *self)
1233	{	1608	{
1234	DIR *dirp;
1235	EIO_STRUCT_DIRENT *entp;
1236	char *name, *names;	1609	char *name, *names;
1237	int namesalloc = 4096;	1610	int namesalloc = 4096 - sizeof (void *) * 4;
1238	int namesoffs = 0;	1611	int namesoffs = 0;
1239	int flags = req->int1;	1612	int flags = req->int1;
1240	eio_dirent *dents = 0;	1613	eio_dirent *dents = 0;
1241	int dentalloc = 128;	1614	int dentalloc = 128;
1242	int dentoffs = 0;	1615	int dentoffs = 0;
1243	ino_t inode_bits = 0;	1616	eio_ino_t inode_bits = 0;
		1617	#ifdef _WIN32
		1618	HANDLE dirp;
		1619	WIN32_FIND_DATA entp;
		1620	#else
		1621	DIR *dirp;
		1622	EIO_STRUCT_DIRENT *entp;
		1623	#endif
1244		1624
1245	req->result = -1;	1625	req->result = -1;
1246		1626
1247	if (!(flags & EIO_READDIR_DENTS))	1627	if (!(flags & EIO_READDIR_DENTS))
1248	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1628	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1249		1629
1250	X_LOCK (wrklock);	1630	#ifdef _WIN32
1251	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1631	{
		1632	int len = strlen ((const char *)req->ptr1);
		1633	char *path = malloc (MAX_PATH);
		1634	const char *fmt;
		1635
		1636	if (!len)
		1637	fmt = "./*";
		1638	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1639	fmt = "%s*";
		1640	else
		1641	fmt = "%s/*";
		1642
		1643	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1644	dirp = FindFirstFile (path, &entp);
		1645	free (path);
		1646
		1647	if (dirp == INVALID_HANDLE_VALUE)
		1648	{
		1649	dirp = 0;
		1650
		1651	switch (GetLastError ())
		1652	{
		1653	case ERROR_FILE_NOT_FOUND:
		1654	req->result = 0;
		1655	break;
		1656
		1657	case ERROR_INVALID_NAME:
		1658	case ERROR_PATH_NOT_FOUND:
		1659	case ERROR_NO_MORE_FILES:
		1660	errno = ENOENT;
		1661	break;
		1662
		1663	case ERROR_NOT_ENOUGH_MEMORY:
		1664	errno = ENOMEM;
		1665	break;
		1666
		1667	default:
		1668	errno = EINVAL;
		1669	break;
		1670	}
		1671	}
		1672	}
		1673	#else
1252	self->dirp = dirp = opendir (req->ptr1);	1674	dirp = opendir (req->ptr1);
		1675	#endif
		1676
		1677	if (req->flags & EIO_FLAG_PTR1_FREE)
		1678	free (req->ptr1);
1253		1679
1254	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1680	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1255	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1681	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1256	req->ptr2 = names = malloc (namesalloc);	1682	req->ptr2 = names = malloc (namesalloc);
1257	X_UNLOCK (wrklock);
1258		1683
1259	if (dirp && names && (!flags || dents))	1684	if (dirp && names && (!flags || dents))
1260	for (;;)	1685	for (;;)
1261	{	1686	{
		1687	int done;
		1688
		1689	#ifdef _WIN32
		1690	done = !dirp;
		1691	#else
1262	errno = 0;	1692	errno = 0;
1263	entp = readdir (dirp);	1693	entp = readdir (dirp);
		1694	done = !entp;
		1695	#endif
1264		1696
1265	if (!entp)	1697	if (done)
1266	{	1698	{
		1699	#ifndef _WIN32
		1700	int old_errno = errno;
		1701	closedir (dirp);
		1702	errno = old_errno;
		1703
1267	if (errno)	1704	if (errno)
1268	break;	1705	break;
		1706	#endif
1269		1707
1270	/* sort etc. */	1708	/* sort etc. */
1271	req->int1 = flags;	1709	req->int1 = flags;
1272	req->result = dentoffs;	1710	req->result = dentoffs;
1273		1711
…		…
1302		1740
1303	break;	1741	break;
1304	}	1742	}
1305		1743
1306	/* now add the entry to our list(s) */	1744	/* now add the entry to our list(s) */
1307	name = entp->d_name;	1745	name = D_NAME (entp);
1308		1746
1309	/* skip . and .. entries */	1747	/* skip . and .. entries */
1310	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1748	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1311	{	1749	{
1312	int len = D_NAMLEN (entp) + 1;	1750	int len = D_NAMLEN (entp) + 1;
1313		1751
1314	while (expect_false (namesoffs + len > namesalloc))	1752	while (ecb_expect_false (namesoffs + len > namesalloc))
1315	{	1753	{
1316	namesalloc *= 2;	1754	namesalloc *= 2;
1317	X_LOCK (wrklock);
1318	req->ptr2 = names = realloc (names, namesalloc);	1755	req->ptr2 = names = realloc (names, namesalloc);
1319	X_UNLOCK (wrklock);
1320		1756
1321	if (!names)	1757	if (!names)
1322	break;	1758	break;
1323	}	1759	}
1324		1760
…		…
1326		1762
1327	if (dents)	1763	if (dents)
1328	{	1764	{
1329	struct eio_dirent *ent;	1765	struct eio_dirent *ent;
1330		1766
1331	if (expect_false (dentoffs == dentalloc))	1767	if (ecb_expect_false (dentoffs == dentalloc))
1332	{	1768	{
1333	dentalloc *= 2;	1769	dentalloc *= 2;
1334	X_LOCK (wrklock);
1335	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1770	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1336	X_UNLOCK (wrklock);
1337		1771
1338	if (!dents)	1772	if (!dents)
1339	break;	1773	break;
1340	}	1774	}
1341		1775
…		…
1421	if (EIO_CANCELLED (req))	1855	if (EIO_CANCELLED (req))
1422	{	1856	{
1423	errno = ECANCELED;	1857	errno = ECANCELED;
1424	break;	1858	break;
1425	}	1859	}
		1860
		1861	#ifdef _WIN32
		1862	if (!FindNextFile (dirp, &entp))
		1863	{
		1864	FindClose (dirp);
		1865	dirp = 0;
		1866	}
		1867	#endif
1426	}	1868	}
1427	}
1428
1429	#ifdef PAGESIZE
1430	# define eio_pagesize() PAGESIZE
1431	#else
1432	static intptr_t
1433	eio_pagesize (void)
1434	{
1435	static intptr_t page;
1436
1437	if (!page)
1438	page = sysconf (_SC_PAGESIZE);
1439
1440	return page;
1441	}
1442	#endif
1443
1444	static void
1445	eio_page_align (void **addr, size_t *length)
1446	{
1447	intptr_t mask = eio_pagesize () - 1;
1448
1449	/* round down addr */
1450	intptr_t adj = mask & (intptr_t)*addr;
1451
1452	*addr = (void *)((intptr_t)*addr - adj);
1453	*length += adj;
1454
1455	/* round up length */
1456	*length = (*length + mask) & ~mask;
1457	}
1458
1459	#if !_POSIX_MEMLOCK
1460	# define eio__mlockall(a) ((errno = ENOSYS), -1)
1461	#else
1462
1463	static int
1464	eio__mlockall (int flags)
1465	{
1466	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1467	extern int mallopt (int, int);
1468	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1469	#endif
1470
1471	if (EIO_MCL_CURRENT != MCL_CURRENT
1472	|| EIO_MCL_FUTURE != MCL_FUTURE)
1473	{
1474	flags = 0
1475	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1476	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1477	}
1478
1479	return mlockall (flags);
1480	}
1481	#endif
1482
1483	#if !_POSIX_MEMLOCK_RANGE
1484	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1485	#else
1486
1487	static int
1488	eio__mlock (void *addr, size_t length)
1489	{
1490	eio_page_align (&addr, &length);
1491
1492	return mlock (addr, length);
1493	}
1494
1495	#endif
1496
1497	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1498	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1499	#else
1500
1501	int
1502	eio__msync (void *mem, size_t len, int flags)
1503	{
1504	eio_page_align (&mem, &len);
1505
1506	if (EIO_MS_ASYNC != MS_SYNC
1507	|| EIO_MS_INVALIDATE != MS_INVALIDATE
1508	|| EIO_MS_SYNC != MS_SYNC)
1509	{
1510	flags = 0
1511	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1512	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1513	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1514	}
1515
1516	return msync (mem, len, flags);
1517	}
1518
1519	#endif
1520
1521	int
1522	eio__mtouch (void *mem, size_t len, int flags)
1523	{
1524	eio_page_align (&mem, &len);
1525
1526	{
1527	intptr_t addr = (intptr_t)mem;
1528	intptr_t end = addr + len;
1529	intptr_t page = eio_pagesize ();
1530
1531	if (addr < end)
1532	if (flags & EIO_MT_MODIFY) /* modify */
1533	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1534	else
1535	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1536	}
1537
1538	return 0;
1539	}	1869	}
1540		1870
1541	/*****************************************************************************/	1871	/*****************************************************************************/
1542		1872
1543	#define ALLOC(len) \	1873	#define ALLOC(len) \
…		…
1559	{	1889	{
1560	ETP_REQ *req;	1890	ETP_REQ *req;
1561	struct timespec ts;	1891	struct timespec ts;
1562	etp_worker *self = (etp_worker *)thr_arg;	1892	etp_worker *self = (etp_worker *)thr_arg;
1563		1893
1564	/* try to distribute timeouts somewhat randomly */	1894	/* try to distribute timeouts somewhat evenly */
1565	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1895	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1566		1896
1567	for (;;)	1897	for (;;)
1568	{	1898	{
		1899	ts.tv_sec = 0;
		1900
1569	X_LOCK (reqlock);	1901	X_LOCK (reqlock);
1570		1902
1571	for (;;)	1903	for (;;)
1572	{	1904	{
1573	self->req = req = reqq_shift (&req_queue);	1905	self->req = req = reqq_shift (&req_queue);
1574		1906
1575	if (req)	1907	if (req)
1576	break;	1908	break;
1577		1909
		1910	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1911	{
		1912	X_UNLOCK (reqlock);
		1913	X_LOCK (wrklock);
		1914	--started;
		1915	X_UNLOCK (wrklock);
		1916	goto quit;
		1917	}
		1918
1578	++idle;	1919	++idle;
1579		1920
1580	ts.tv_sec = time (0) + idle_timeout;	1921	if (idle <= max_idle)
1581	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1922	/* we are allowed to idle, so do so without any timeout */
		1923	X_COND_WAIT (reqwait, reqlock);
		1924	else
1582	{	1925	{
1583	if (idle > max_idle)	1926	/* initialise timeout once */
1584	{	1927	if (!ts.tv_sec)
1585	--idle;	1928	ts.tv_sec = time (0) + idle_timeout;
1586	X_UNLOCK (reqlock);
1587	X_LOCK (wrklock);
1588	--started;
1589	X_UNLOCK (wrklock);
1590	goto quit;
1591	}
1592		1929
1593	/* we are allowed to idle, so do so without any timeout */
1594	X_COND_WAIT (reqwait, reqlock);	1930	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1931	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1595	}	1932	}
1596		1933
1597	--idle;	1934	--idle;
1598	}	1935	}
1599		1936
…		…
1602	X_UNLOCK (reqlock);	1939	X_UNLOCK (reqlock);
1603		1940
1604	if (req->type < 0)	1941	if (req->type < 0)
1605	goto quit;	1942	goto quit;
1606		1943
1607	if (!EIO_CANCELLED (req))
1608	ETP_EXECUTE (self, req);	1944	ETP_EXECUTE (self, req);
1609		1945
1610	X_LOCK (reslock);	1946	X_LOCK (reslock);
1611		1947
1612	++npending;	1948	++npending;
1613		1949
…		…
1628	return 0;	1964	return 0;
1629	}	1965	}
1630		1966
1631	/*****************************************************************************/	1967	/*****************************************************************************/
1632		1968
		1969	int ecb_cold
1633	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1970	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1634	{	1971	{
		1972	#if !HAVE_PREADWRITE
		1973	X_MUTEX_CREATE (preadwritelock);
		1974	#endif
		1975
1635	return etp_init (want_poll, done_poll);	1976	return etp_init (want_poll, done_poll);
1636	}	1977	}
1637		1978
		1979	ecb_inline void
1638	static void eio_api_destroy (eio_req *req)	1980	eio_api_destroy (eio_req *req)
1639	{	1981	{
1640	free (req);	1982	free (req);
1641	}	1983	}
1642		1984
1643	#define REQ(rtype) \	1985	#define REQ(rtype) \
…		…
1662	{ \	2004	{ \
1663	eio_api_destroy (req); \	2005	eio_api_destroy (req); \
1664	return 0; \	2006	return 0; \
1665	}	2007	}
1666		2008
		2009	static void
1667	static void eio_execute (etp_worker *self, eio_req *req)	2010	eio_execute (etp_worker *self, eio_req *req)
1668	{	2011	{
		2012	if (ecb_expect_false (EIO_CANCELLED (req)))
		2013	{
		2014	req->result = -1;
		2015	req->errorno = ECANCELED;
		2016	return;
		2017	}
		2018
1669	switch (req->type)	2019	switch (req->type)
1670	{	2020	{
1671	case EIO_READ: ALLOC (req->size);	2021	case EIO_READ: ALLOC (req->size);
1672	req->result = req->offs >= 0	2022	req->result = req->offs >= 0
1673	? pread (req->int1, req->ptr2, req->size, req->offs)	2023	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1707	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2057	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1708	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2058	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1709	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2059	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1710	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;	2060	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1711		2061
		2062	case EIO_REALPATH: eio__realpath (req, self); break;
		2063
1712	case EIO_READLINK: ALLOC (PATH_MAX);	2064	case EIO_READLINK: ALLOC (PATH_MAX);
1713	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;	2065	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1714		2066
1715	case EIO_SYNC: req->result = 0; sync (); break;	2067	case EIO_SYNC: req->result = 0; sync (); break;
1716	case EIO_FSYNC: req->result = fsync (req->int1); break;	2068	case EIO_FSYNC: req->result = fsync (req->int1); break;
1717	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2069	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1718	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2070	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1719	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	2071	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1720	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2072	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1721	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2073	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1722	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	2074	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2075	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1723		2076
1724	case EIO_READDIR: eio__scandir (req, self); break;	2077	case EIO_READDIR: eio__scandir (req, self); break;
1725		2078
1726	case EIO_BUSY:	2079	case EIO_BUSY:
1727	#ifdef _WIN32	2080	#ifdef _WIN32
…		…
1768	case EIO_NOP:	2121	case EIO_NOP:
1769	req->result = 0;	2122	req->result = 0;
1770	break;	2123	break;
1771		2124
1772	case EIO_CUSTOM:	2125	case EIO_CUSTOM:
1773	((void (*)(eio_req *))req->feed) (req);	2126	req->feed (req);
1774	break;	2127	break;
1775		2128
1776	default:	2129	default:
1777	errno = ENOSYS;	2130	errno = ENOSYS;
1778	req->result = -1;	2131	req->result = -1;
…		…
1827	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2180	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1828	{	2181	{
1829	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2182	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1830	}	2183	}
1831		2184
		2185	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2186	{
		2187	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2188	}
		2189
1832	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2190	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1833	{	2191	{
1834	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2192	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1835	}	2193	}
1836		2194
…		…
1877	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2235	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1878	{	2236	{
1879	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2237	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1880	}	2238	}
1881		2239
1882	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2240	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1883	{	2241	{
1884	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2242	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1885	}	2243	}
1886		2244
1887	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2245	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1907	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2265	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1908	{	2266	{
1909	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2267	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1910	}	2268	}
1911		2269
1912	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2270	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1913	{	2271	{
1914	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2272	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1915	}	2273	}
1916		2274
1917	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2275	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1931	}	2289	}
1932		2290
1933	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2291	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1934	{	2292	{
1935	return eio__1path (EIO_READLINK, path, pri, cb, data);	2293	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2294	}
		2295
		2296	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2297	{
		2298	return eio__1path (EIO_REALPATH, path, pri, cb, data);
1936	}	2299	}
1937		2300
1938	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2301	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1939	{	2302	{
1940	return eio__1path (EIO_STAT, path, pri, cb, data);	2303	return eio__1path (EIO_STAT, path, pri, cb, data);
…		…
1999	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2362	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2000	{	2363	{
2001	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2364	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2002	}	2365	}
2003		2366
2004	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2367	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2005	{	2368	{
2006	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2369	REQ (EIO_CUSTOM); req->feed = execute; SEND;
2007	}	2370	}
2008		2371
2009	#endif	2372	#endif
2010		2373
2011	eio_req *eio_grp (eio_cb cb, void *data)	2374	eio_req *eio_grp (eio_cb cb, void *data)
…		…
2020	#undef SEND	2383	#undef SEND
2021		2384
2022	/*****************************************************************************/	2385	/*****************************************************************************/
2023	/* grp functions */	2386	/* grp functions */
2024		2387
		2388	void
2025	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2389	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2026	{	2390	{
2027	grp->int2 = limit;	2391	grp->int2 = limit;
2028	grp->feed = feed;	2392	grp->feed = feed;
2029		2393
2030	grp_try_feed (grp);	2394	grp_try_feed (grp);
2031	}	2395	}
2032		2396
		2397	void
2033	void eio_grp_limit (eio_req *grp, int limit)	2398	eio_grp_limit (eio_req *grp, int limit)
2034	{	2399	{
2035	grp->int2 = limit;	2400	grp->int2 = limit;
2036		2401
2037	grp_try_feed (grp);	2402	grp_try_feed (grp);
2038	}	2403	}
2039		2404
		2405	void
2040	void eio_grp_add (eio_req *grp, eio_req *req)	2406	eio_grp_add (eio_req *grp, eio_req *req)
2041	{	2407	{
2042	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2408	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2043		2409
2044	grp->flags |= EIO_FLAG_GROUPADD;	2410	grp->flags |= EIO_FLAG_GROUPADD;
2045		2411
…		…
2056	}	2422	}
2057		2423
2058	/*****************************************************************************/	2424	/*****************************************************************************/
2059	/* misc garbage */	2425	/* misc garbage */
2060		2426
		2427	eio_ssize_t
2061	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2428	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2062	{	2429	{
2063	etp_worker wrk;	2430	etp_worker wrk;
2064	ssize_t ret;	2431	eio_ssize_t ret;
2065		2432
2066	wrk.dbuf = 0;	2433	wrk.dbuf = 0;
2067		2434
2068	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2435	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2069		2436

Diff Legend

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