ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/cvsroot/libeio/eio.c

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.33 by root, Sat Jun 6 19:44:17 2009 UTC vs.
Revision 1.85 by root, Wed Jul 13 00:53:03 2011 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libeio@schmorp.de>	4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5	* All rights reserved.	5	* All rights reserved.
6	*	6	*
7	* Redistribution and use in source and binary forms, with or without modifica-	7	* Redistribution and use in source and binary forms, with or without modifica-
8	* tion, are permitted provided that the following conditions are met:	8	* tion, are permitted provided that the following conditions are met:
9	*	9	*
…		…
35	* and other provisions required by the GPL. If you do not delete the	35	* and other provisions required by the GPL. If you do not delete the
36	* provisions above, a recipient may use your version of this file under	36	* provisions above, a recipient may use your version of this file under
37	* either the BSD or the GPL.	37	* either the BSD or the GPL.
38	*/	38	*/
39		39
		40	#ifndef _WIN32
		41	# include "config.h"
		42	#endif
		43
40	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
41		46
42	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
43	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
44	#endif	49	#endif
45	#include "xthread.h"	50	#include "xthread.h"
…		…
49	#include <stdlib.h>	54	#include <stdlib.h>
50	#include <string.h>	55	#include <string.h>
51	#include <errno.h>	56	#include <errno.h>
52	#include <sys/types.h>	57	#include <sys/types.h>
53	#include <sys/stat.h>	58	#include <sys/stat.h>
		59	#include <sys/statvfs.h>
54	#include <limits.h>	60	#include <limits.h>
55	#include <fcntl.h>	61	#include <fcntl.h>
56	#include <assert.h>	62	#include <assert.h>
57		63
		64	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
		65	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
		66	#if HAVE_STDINT_H
		67	# include <stdint.h>
		68	#endif
		69
		70	#ifndef ECANCELED
		71	# define ECANCELED EDOM
		72	#endif
		73
		74	static void eio_destroy (eio_req *req);
		75
58	#ifndef EIO_FINISH	76	#ifndef EIO_FINISH
59	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	77	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60	#endif	78	#endif
61		79
62	#ifndef EIO_DESTROY	80	#ifndef EIO_DESTROY
…		…
70	#ifdef _WIN32	88	#ifdef _WIN32
71		89
72	/*doh*/	90	/*doh*/
73	#else	91	#else
74		92
75	# include "config.h"
76	# include <sys/time.h>	93	# include <sys/time.h>
77	# include <sys/select.h>	94	# include <sys/select.h>
78	# include <unistd.h>	95	# include <unistd.h>
79	# include <utime.h>	96	# include <utime.h>
80	# include <signal.h>	97	# include <signal.h>
81	# include <dirent.h>	98	# include <dirent.h>
82		99
		100	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		101	# include <sys/mman.h>
		102	#endif
		103
83	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */	104	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84	# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)	105	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
85	# define _DIRENT_HAVE_D_TYPE /* sigh */	106	# define _DIRENT_HAVE_D_TYPE /* sigh */
86	# define D_INO(de) (de)->d_fileno	107	# define D_INO(de) (de)->d_fileno
87	# define D_NAMLEN(de) (de)->d_namlen	108	# define D_NAMLEN(de) (de)->d_namlen
88	# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600	109	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
89	# define D_INO(de) (de)->d_ino	110	# define D_INO(de) (de)->d_ino
90	# endif	111	# endif
91		112
92	#ifdef _D_EXACT_NAMLEN	113	#ifdef _D_EXACT_NAMLEN
93	# undef D_NAMLEN	114	# undef D_NAMLEN
…		…
105	#endif	126	#endif
106		127
107	#if HAVE_SENDFILE	128	#if HAVE_SENDFILE
108	# if __linux	129	# if __linux
109	# include <sys/sendfile.h>	130	# include <sys/sendfile.h>
110	# elif __freebsd	131	# elif __FreeBSD__ || defined __APPLE__
111	# include <sys/socket.h>	132	# include <sys/socket.h>
112	# include <sys/uio.h>	133	# include <sys/uio.h>
113	# elif __hpux	134	# elif __hpux
114	# include <sys/socket.h>	135	# include <sys/socket.h>
115	# elif __solaris /* not yet */	136	# elif __solaris
116	# include <sys/sendfile.h>	137	# include <sys/sendfile.h>
117	# else	138	# else
118	# error sendfile support requested but not available	139	# error sendfile support requested but not available
119	# endif	140	# endif
120	#endif	141	#endif
…		…
127	#endif	148	#endif
128	#ifndef D_NAMLEN	149	#ifndef D_NAMLEN
129	# define D_NAMLEN(de) strlen ((de)->d_name)	150	# define D_NAMLEN(de) strlen ((de)->d_name)
130	#endif	151	#endif
131		152
132	/* number of seconds after which an idle threads exit */
133	#define IDLE_TIMEOUT 10
134
135	/* used for struct dirent, AIX doesn't provide it */	153	/* used for struct dirent, AIX doesn't provide it */
136	#ifndef NAME_MAX	154	#ifndef NAME_MAX
137	# define NAME_MAX 4096	155	# define NAME_MAX 4096
		156	#endif
		157
		158	/* used for readlink etc. */
		159	#ifndef PATH_MAX
		160	# define PATH_MAX 4096
138	#endif	161	#endif
139		162
140	/* buffer size for various temporary buffers */	163	/* buffer size for various temporary buffers */
141	#define EIO_BUFSIZE 65536	164	#define EIO_BUFSIZE 65536
142		165
…		…
148	errno = ENOMEM; \	171	errno = ENOMEM; \
149	if (!eio_buf) \	172	if (!eio_buf) \
150	return -1;	173	return -1;
151		174
152	#define EIO_TICKS ((1000000 + 1023) >> 10)	175	#define EIO_TICKS ((1000000 + 1023) >> 10)
153
154	/*****************************************************************************/
155
156	#if __GNUC__ >= 3
157	# define expect(expr,value) __builtin_expect ((expr),(value))
158	#else
159	# define expect(expr,value) (expr)
160	#endif
161
162	#define expect_false(expr) expect ((expr) != 0, 0)
163	#define expect_true(expr) expect ((expr) != 0, 1)
164
165	/*****************************************************************************/
166		176
167	#define ETP_PRI_MIN EIO_PRI_MIN	177	#define ETP_PRI_MIN EIO_PRI_MIN
168	#define ETP_PRI_MAX EIO_PRI_MAX	178	#define ETP_PRI_MAX EIO_PRI_MAX
169		179
170	struct etp_worker;	180	struct etp_worker;
…		…
195		205
196	/*****************************************************************************/	206	/*****************************************************************************/
197		207
198	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	208	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
199		209
200	/* calculcate time difference in ~1/EIO_TICKS of a second */	210	/* calculate time difference in ~1/EIO_TICKS of a second */
		211	ecb_inline int
201	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	212	tvdiff (struct timeval *tv1, struct timeval *tv2)
202	{	213	{
203	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	214	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
204	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	215	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
205	}	216	}
206		217
…		…
213	static unsigned int max_poll_reqs; /* reslock */	224	static unsigned int max_poll_reqs; /* reslock */
214		225
215	static volatile unsigned int nreqs; /* reqlock */	226	static volatile unsigned int nreqs; /* reqlock */
216	static volatile unsigned int nready; /* reqlock */	227	static volatile unsigned int nready; /* reqlock */
217	static volatile unsigned int npending; /* reqlock */	228	static volatile unsigned int npending; /* reqlock */
218	static volatile unsigned int max_idle = 4;	229	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		230	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
219		231
220	static mutex_t wrklock = X_MUTEX_INIT;	232	static xmutex_t wrklock;
221	static mutex_t reslock = X_MUTEX_INIT;	233	static xmutex_t reslock;
222	static mutex_t reqlock = X_MUTEX_INIT;	234	static xmutex_t reqlock;
223	static cond_t reqwait = X_COND_INIT;	235	static xcond_t reqwait;
224		236
225	#if !HAVE_PREADWRITE	237	#if !HAVE_PREADWRITE
226	/*	238	/*
227	* make our pread/pwrite emulation safe against themselves, but not against	239	* make our pread/pwrite emulation safe against themselves, but not against
228	* normal read/write by using a mutex. slows down execution a lot,	240	* normal read/write by using a mutex. slows down execution a lot,
229	* but that's your problem, not mine.	241	* but that's your problem, not mine.
230	*/	242	*/
231	static mutex_t preadwritelock = X_MUTEX_INIT;	243	static xmutex_t preadwritelock = X_MUTEX_INIT;
232	#endif	244	#endif
233		245
234	typedef struct etp_worker	246	typedef struct etp_worker
235	{	247	{
236	/* locked by wrklock */	248	/* locked by wrklock */
237	struct etp_worker *prev, *next;	249	struct etp_worker *prev, *next;
238		250
239	thread_t tid;	251	xthread_t tid;
240		252
241	/* locked by reslock, reqlock or wrklock */	253	/* locked by reslock, reqlock or wrklock */
242	ETP_REQ *req; /* currently processed request */	254	ETP_REQ *req; /* currently processed request */
243		255
244	ETP_WORKER_COMMON	256	ETP_WORKER_COMMON
…		…
249	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	261	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
250	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	262	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
251		263
252	/* worker threads management */	264	/* worker threads management */
253		265
		266	static void ecb_cold
254	static void etp_worker_clear (etp_worker *wrk)	267	etp_worker_clear (etp_worker *wrk)
255	{	268	{
256	ETP_WORKER_CLEAR (wrk);	269	ETP_WORKER_CLEAR (wrk);
257	}	270	}
258		271
		272	static void ecb_cold
259	static void etp_worker_free (etp_worker *wrk)	273	etp_worker_free (etp_worker *wrk)
260	{	274	{
261	wrk->next->prev = wrk->prev;	275	wrk->next->prev = wrk->prev;
262	wrk->prev->next = wrk->next;	276	wrk->prev->next = wrk->next;
263		277
264	free (wrk);	278	free (wrk);
265	}	279	}
266		280
267	static unsigned int etp_nreqs (void)	281	static unsigned int
		282	etp_nreqs (void)
268	{	283	{
269	int retval;	284	int retval;
270	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	285	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271	retval = nreqs;	286	retval = nreqs;
272	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	287	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273	return retval;	288	return retval;
274	}	289	}
275		290
276	static unsigned int etp_nready (void)	291	static unsigned int
		292	etp_nready (void)
277	{	293	{
278	unsigned int retval;	294	unsigned int retval;
279		295
280	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	296	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281	retval = nready;	297	retval = nready;
282	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	298	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283		299
284	return retval;	300	return retval;
285	}	301	}
286		302
287	static unsigned int etp_npending (void)	303	static unsigned int
		304	etp_npending (void)
288	{	305	{
289	unsigned int retval;	306	unsigned int retval;
290		307
291	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	308	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
292	retval = npending;	309	retval = npending;
293	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	310	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
294		311
295	return retval;	312	return retval;
296	}	313	}
297		314
298	static unsigned int etp_nthreads (void)	315	static unsigned int
		316	etp_nthreads (void)
299	{	317	{
300	unsigned int retval;	318	unsigned int retval;
301		319
302	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	320	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303	retval = started;	321	retval = started;
…		…
317	} etp_reqq;	335	} etp_reqq;
318		336
319	static etp_reqq req_queue;	337	static etp_reqq req_queue;
320	static etp_reqq res_queue;	338	static etp_reqq res_queue;
321		339
		340	static int ecb_noinline
322	static int reqq_push (etp_reqq *q, ETP_REQ *req)	341	reqq_push (etp_reqq *q, ETP_REQ *req)
323	{	342	{
324	int pri = req->pri;	343	int pri = req->pri;
325	req->next = 0;	344	req->next = 0;
326		345
327	if (q->qe[pri])	346	if (q->qe[pri])
…		…
333	q->qe[pri] = q->qs[pri] = req;	352	q->qe[pri] = q->qs[pri] = req;
334		353
335	return q->size++;	354	return q->size++;
336	}	355	}
337		356
		357	static ETP_REQ * ecb_noinline
338	static ETP_REQ *reqq_shift (etp_reqq *q)	358	reqq_shift (etp_reqq *q)
339	{	359	{
340	int pri;	360	int pri;
341		361
342	if (!q->size)	362	if (!q->size)
343	return 0;	363	return 0;
…		…
358	}	378	}
359		379
360	abort ();	380	abort ();
361	}	381	}
362		382
363	static void etp_atfork_prepare (void)	383	static void ecb_cold
		384	etp_thread_init (void)
364	{	385	{
365	X_LOCK (wrklock);
366	X_LOCK (reqlock);
367	X_LOCK (reslock);
368	#if !HAVE_PREADWRITE	386	#if !HAVE_PREADWRITE
369	X_LOCK (preadwritelock);	387	X_MUTEX_CREATE (preadwritelock);
370	#endif	388	#endif
		389	X_MUTEX_CREATE (wrklock);
		390	X_MUTEX_CREATE (reslock);
		391	X_MUTEX_CREATE (reqlock);
		392	X_COND_CREATE (reqwait);
371	}	393	}
372		394
		395	static void ecb_cold
		396	etp_atfork_prepare (void)
		397	{
		398	}
		399
		400	static void ecb_cold
373	static void etp_atfork_parent (void)	401	etp_atfork_parent (void)
374	{	402	{
375	#if !HAVE_PREADWRITE
376	X_UNLOCK (preadwritelock);
377	#endif
378	X_UNLOCK (reslock);
379	X_UNLOCK (reqlock);
380	X_UNLOCK (wrklock);
381	}	403	}
382		404
		405	static void ecb_cold
383	static void etp_atfork_child (void)	406	etp_atfork_child (void)
384	{	407	{
385	ETP_REQ *prv;	408	ETP_REQ *prv;
386		409
387	while ((prv = reqq_shift (&req_queue)))	410	while ((prv = reqq_shift (&req_queue)))
388	ETP_DESTROY (prv);	411	ETP_DESTROY (prv);
…		…
405	idle = 0;	428	idle = 0;
406	nreqs = 0;	429	nreqs = 0;
407	nready = 0;	430	nready = 0;
408	npending = 0;	431	npending = 0;
409		432
410	etp_atfork_parent ();	433	etp_thread_init ();
411	}	434	}
412		435
413	static void	436	static void ecb_cold
414	etp_once_init (void)	437	etp_once_init (void)
415	{	438	{
		439	etp_thread_init ();
416	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);	440	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
417	}	441	}
418		442
419	static int	443	static int ecb_cold
420	etp_init (void (*want_poll)(void), void (*done_poll)(void))	444	etp_init (void (*want_poll)(void), void (*done_poll)(void))
421	{	445	{
422	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	446	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
423		447
424	pthread_once (&doinit, etp_once_init);	448	pthread_once (&doinit, etp_once_init);
…		…
429	return 0;	453	return 0;
430	}	454	}
431		455
432	X_THREAD_PROC (etp_proc);	456	X_THREAD_PROC (etp_proc);
433		457
		458	static void ecb_cold
434	static void etp_start_thread (void)	459	etp_start_thread (void)
435	{	460	{
436	etp_worker *wrk = calloc (1, sizeof (etp_worker));	461	etp_worker *wrk = calloc (1, sizeof (etp_worker));
437		462
438	/*TODO*/	463	/*TODO*/
439	assert (("unable to allocate worker thread data", wrk));	464	assert (("unable to allocate worker thread data", wrk));
…		…
452	free (wrk);	477	free (wrk);
453		478
454	X_UNLOCK (wrklock);	479	X_UNLOCK (wrklock);
455	}	480	}
456		481
		482	static void
457	static void etp_maybe_start_thread (void)	483	etp_maybe_start_thread (void)
458	{	484	{
459	if (expect_true (etp_nthreads () >= wanted))	485	if (ecb_expect_true (etp_nthreads () >= wanted))
460	return;	486	return;
461		487
462	/* todo: maybe use idle here, but might be less exact */	488	/* todo: maybe use idle here, but might be less exact */
463	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	489	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
464	return;	490	return;
465		491
466	etp_start_thread ();	492	etp_start_thread ();
467	}	493	}
468		494
		495	static void ecb_cold
469	static void etp_end_thread (void)	496	etp_end_thread (void)
470	{	497	{
471	eio_req *req = calloc (1, sizeof (eio_req));	498	eio_req *req = calloc (1, sizeof (eio_req));
472		499
473	req->type = -1;	500	req->type = -1;
474	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	501	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
481	X_LOCK (wrklock);	508	X_LOCK (wrklock);
482	--started;	509	--started;
483	X_UNLOCK (wrklock);	510	X_UNLOCK (wrklock);
484	}	511	}
485		512
486	static int etp_poll (void)	513	static int
		514	etp_poll (void)
487	{	515	{
488	unsigned int maxreqs;	516	unsigned int maxreqs;
489	unsigned int maxtime;	517	unsigned int maxtime;
490	struct timeval tv_start, tv_now;	518	struct timeval tv_start, tv_now;
491		519
…		…
521		549
522	X_LOCK (reqlock);	550	X_LOCK (reqlock);
523	--nreqs;	551	--nreqs;
524	X_UNLOCK (reqlock);	552	X_UNLOCK (reqlock);
525		553
526	if (expect_false (req->type == EIO_GROUP && req->size))	554	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
527	{	555	{
528	req->int1 = 1; /* mark request as delayed */	556	req->int1 = 1; /* mark request as delayed */
529	continue;	557	continue;
530	}	558	}
531	else	559	else
532	{	560	{
533	int res = ETP_FINISH (req);	561	int res = ETP_FINISH (req);
534	if (expect_false (res))	562	if (ecb_expect_false (res))
535	return res;	563	return res;
536	}	564	}
537		565
538	if (expect_false (maxreqs && !--maxreqs))	566	if (ecb_expect_false (maxreqs && !--maxreqs))
539	break;	567	break;
540		568
541	if (maxtime)	569	if (maxtime)
542	{	570	{
543	gettimeofday (&tv_now, 0);	571	gettimeofday (&tv_now, 0);
…		…
549		577
550	errno = EAGAIN;	578	errno = EAGAIN;
551	return -1;	579	return -1;
552	}	580	}
553		581
		582	static void
554	static void etp_cancel (ETP_REQ *req)	583	etp_cancel (ETP_REQ *req)
555	{	584	{
556	X_LOCK (wrklock);	585	req->cancelled = 1;
557	req->flags |= EIO_FLAG_CANCELLED;
558	X_UNLOCK (wrklock);
559		586
560	eio_grp_cancel (req);	587	eio_grp_cancel (req);
561	}	588	}
562		589
		590	static void
563	static void etp_submit (ETP_REQ *req)	591	etp_submit (ETP_REQ *req)
564	{	592	{
565	req->pri -= ETP_PRI_MIN;	593	req->pri -= ETP_PRI_MIN;
566		594
567	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	595	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
568	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	596	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
569		597
570	if (expect_false (req->type == EIO_GROUP))	598	if (ecb_expect_false (req->type == EIO_GROUP))
571	{	599	{
572	/* I hope this is worth it :/ */	600	/* I hope this is worth it :/ */
573	X_LOCK (reqlock);	601	X_LOCK (reqlock);
574	++nreqs;	602	++nreqs;
575	X_UNLOCK (reqlock);	603	X_UNLOCK (reqlock);
…		…
594		622
595	etp_maybe_start_thread ();	623	etp_maybe_start_thread ();
596	}	624	}
597	}	625	}
598		626
		627	static void ecb_cold
599	static void etp_set_max_poll_time (double nseconds)	628	etp_set_max_poll_time (double nseconds)
600	{	629	{
601	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	630	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
602	max_poll_time = nseconds;	631	max_poll_time = nseconds * EIO_TICKS;
603	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	632	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
604	}	633	}
605		634
		635	static void ecb_cold
606	static void etp_set_max_poll_reqs (unsigned int maxreqs)	636	etp_set_max_poll_reqs (unsigned int maxreqs)
607	{	637	{
608	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	638	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609	max_poll_reqs = maxreqs;	639	max_poll_reqs = maxreqs;
610	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	640	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611	}	641	}
612		642
		643	static void ecb_cold
613	static void etp_set_max_idle (unsigned int nthreads)	644	etp_set_max_idle (unsigned int nthreads)
614	{	645	{
615	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	646	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
616	max_idle = nthreads <= 0 ? 1 : nthreads;	647	max_idle = nthreads;
617	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	648	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
618	}	649	}
619		650
		651	static void ecb_cold
		652	etp_set_idle_timeout (unsigned int seconds)
		653	{
		654	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		655	idle_timeout = seconds;
		656	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		657	}
		658
		659	static void ecb_cold
620	static void etp_set_min_parallel (unsigned int nthreads)	660	etp_set_min_parallel (unsigned int nthreads)
621	{	661	{
622	if (wanted < nthreads)	662	if (wanted < nthreads)
623	wanted = nthreads;	663	wanted = nthreads;
624	}	664	}
625		665
		666	static void ecb_cold
626	static void etp_set_max_parallel (unsigned int nthreads)	667	etp_set_max_parallel (unsigned int nthreads)
627	{	668	{
628	if (wanted > nthreads)	669	if (wanted > nthreads)
629	wanted = nthreads;	670	wanted = nthreads;
630		671
631	while (started > wanted)	672	while (started > wanted)
632	etp_end_thread ();	673	etp_end_thread ();
633	}	674	}
634		675
635	/*****************************************************************************/	676	/*****************************************************************************/
636		677
		678	static void
637	static void grp_try_feed (eio_req *grp)	679	grp_try_feed (eio_req *grp)
638	{	680	{
639	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	681	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
640	{	682	{
641	grp->flags &= ~EIO_FLAG_GROUPADD;	683	grp->flags &= ~EIO_FLAG_GROUPADD;
642		684
…		…
649	break;	691	break;
650	}	692	}
651	}	693	}
652	}	694	}
653		695
		696	static int
654	static int grp_dec (eio_req *grp)	697	grp_dec (eio_req *grp)
655	{	698	{
656	--grp->size;	699	--grp->size;
657		700
658	/* call feeder, if applicable */	701	/* call feeder, if applicable */
659	grp_try_feed (grp);	702	grp_try_feed (grp);
…		…
663	return eio_finish (grp);	706	return eio_finish (grp);
664	else	707	else
665	return 0;	708	return 0;
666	}	709	}
667		710
		711	static void
668	void eio_destroy (eio_req *req)	712	eio_destroy (eio_req *req)
669	{	713	{
670	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	714	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
671	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	715	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
672		716
673	EIO_DESTROY (req);	717	EIO_DESTROY (req);
674	}	718	}
675		719
		720	static int
676	static int eio_finish (eio_req *req)	721	eio_finish (eio_req *req)
677	{	722	{
678	int res = EIO_FINISH (req);	723	int res = EIO_FINISH (req);
679		724
680	if (req->grp)	725	if (req->grp)
681	{	726	{
…		…
689	if (grp->grp_first == req)	734	if (grp->grp_first == req)
690	grp->grp_first = req->grp_next;	735	grp->grp_first = req->grp_next;
691		736
692	res2 = grp_dec (grp);	737	res2 = grp_dec (grp);
693		738
694	if (!res && res2)	739	if (!res)
695	res = res2;	740	res = res2;
696	}	741	}
697		742
698	eio_destroy (req);	743	eio_destroy (req);
699		744
700	return res;	745	return res;
701	}	746	}
702		747
		748	void
703	void eio_grp_cancel (eio_req *grp)	749	eio_grp_cancel (eio_req *grp)
704	{	750	{
705	for (grp = grp->grp_first; grp; grp = grp->grp_next)	751	for (grp = grp->grp_first; grp; grp = grp->grp_next)
706	eio_cancel (grp);	752	eio_cancel (grp);
707	}	753	}
708		754
		755	void
709	void eio_cancel (eio_req *req)	756	eio_cancel (eio_req *req)
710	{	757	{
711	etp_cancel (req);	758	etp_cancel (req);
712	}	759	}
713		760
		761	void
714	void eio_submit (eio_req *req)	762	eio_submit (eio_req *req)
715	{	763	{
716	etp_submit (req);	764	etp_submit (req);
717	}	765	}
718		766
719	unsigned int eio_nreqs (void)	767	unsigned int
		768	eio_nreqs (void)
720	{	769	{
721	return etp_nreqs ();	770	return etp_nreqs ();
722	}	771	}
723		772
724	unsigned int eio_nready (void)	773	unsigned int
		774	eio_nready (void)
725	{	775	{
726	return etp_nready ();	776	return etp_nready ();
727	}	777	}
728		778
729	unsigned int eio_npending (void)	779	unsigned int
		780	eio_npending (void)
730	{	781	{
731	return etp_npending ();	782	return etp_npending ();
732	}	783	}
733		784
734	unsigned int eio_nthreads (void)	785	unsigned int ecb_cold
		786	eio_nthreads (void)
735	{	787	{
736	return etp_nthreads ();	788	return etp_nthreads ();
737	}	789	}
738		790
		791	void ecb_cold
739	void eio_set_max_poll_time (double nseconds)	792	eio_set_max_poll_time (double nseconds)
740	{	793	{
741	etp_set_max_poll_time (nseconds);	794	etp_set_max_poll_time (nseconds);
742	}	795	}
743		796
		797	void ecb_cold
744	void eio_set_max_poll_reqs (unsigned int maxreqs)	798	eio_set_max_poll_reqs (unsigned int maxreqs)
745	{	799	{
746	etp_set_max_poll_reqs (maxreqs);	800	etp_set_max_poll_reqs (maxreqs);
747	}	801	}
748		802
		803	void ecb_cold
749	void eio_set_max_idle (unsigned int nthreads)	804	eio_set_max_idle (unsigned int nthreads)
750	{	805	{
751	etp_set_max_idle (nthreads);	806	etp_set_max_idle (nthreads);
752	}	807	}
753		808
		809	void ecb_cold
		810	eio_set_idle_timeout (unsigned int seconds)
		811	{
		812	etp_set_idle_timeout (seconds);
		813	}
		814
		815	void ecb_cold
754	void eio_set_min_parallel (unsigned int nthreads)	816	eio_set_min_parallel (unsigned int nthreads)
755	{	817	{
756	etp_set_min_parallel (nthreads);	818	etp_set_min_parallel (nthreads);
757	}	819	}
758		820
		821	void ecb_cold
759	void eio_set_max_parallel (unsigned int nthreads)	822	eio_set_max_parallel (unsigned int nthreads)
760	{	823	{
761	etp_set_max_parallel (nthreads);	824	etp_set_max_parallel (nthreads);
762	}	825	}
763		826
764	int eio_poll (void)	827	int eio_poll (void)
…		…
806		869
807	return res;	870	return res;
808	}	871	}
809	#endif	872	#endif
810		873
811	#ifndef HAVE_FUTIMES	874	#ifndef HAVE_UTIMES
812		875
813	# undef utimes	876	# undef utimes
814	# undef futimes
815	# define utimes(path,times) eio__utimes (path, times)	877	# define utimes(path,times) eio__utimes (path, times)
816	# define futimes(fd,times) eio__futimes (fd, times)
817		878
818	static int	879	static int
819	eio__utimes (const char *filename, const struct timeval times[2])	880	eio__utimes (const char *filename, const struct timeval times[2])
820	{	881	{
821	if (times)	882	if (times)
…		…
829	}	890	}
830	else	891	else
831	return utime (filename, 0);	892	return utime (filename, 0);
832	}	893	}
833		894
		895	#endif
		896
		897	#ifndef HAVE_FUTIMES
		898
		899	# undef futimes
		900	# define futimes(fd,times) eio__futimes (fd, times)
		901
		902	static int
834	static int eio__futimes (int fd, const struct timeval tv[2])	903	eio__futimes (int fd, const struct timeval tv[2])
835	{	904	{
836	errno = ENOSYS;	905	errno = ENOSYS;
837	return -1;	906	return -1;
838	}	907	}
839		908
…		…
843	# undef fdatasync	912	# undef fdatasync
844	# define fdatasync(fd) fsync (fd)	913	# define fdatasync(fd) fsync (fd)
845	#endif	914	#endif
846		915
847	/* sync_file_range always needs emulation */	916	/* sync_file_range always needs emulation */
848	int	917	static int
849	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	918	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
850	{	919	{
851	#if HAVE_SYNC_FILE_RANGE	920	#if HAVE_SYNC_FILE_RANGE
852	int res;	921	int res;
853		922
…		…
866	if (!res || errno != ENOSYS)	935	if (!res || errno != ENOSYS)
867	return res;	936	return res;
868	#endif	937	#endif
869		938
870	/* even though we could play tricks with the flags, it's better to always	939	/* even though we could play tricks with the flags, it's better to always
871	* call fdatasync, as thta matches the expectation of it's users best */	940	* call fdatasync, as that matches the expectation of its users best */
872	return fdatasync (fd);	941	return fdatasync (fd);
		942	}
		943
		944	static int
		945	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		946	{
		947	#if HAVE_FALLOCATE
		948	return fallocate (fd, mode, offset, len);
		949	#else
		950	errno = ENOSYS;
		951	return -1;
		952	#endif
873	}	953	}
874		954
875	#if !HAVE_READAHEAD	955	#if !HAVE_READAHEAD
876	# undef readahead	956	# undef readahead
877	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	957	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
…		…
899		979
900	/* sendfile always needs emulation */	980	/* sendfile always needs emulation */
901	static ssize_t	981	static ssize_t
902	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	982	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
903	{	983	{
		984	ssize_t written = 0;
904	ssize_t res;	985	ssize_t res;
905		986
906	if (!count)	987	if (!count)
907	return 0;	988	return 0;
908		989
		990	for (;;)
		991	{
		992	#ifdef __APPLE__
		993	# undef HAVE_SENDFILE /* broken, as everything on os x */
		994	#endif
909	#if HAVE_SENDFILE	995	#if HAVE_SENDFILE
910	# if __linux	996	# if __linux
		997	off_t soffset = offset;
911	res = sendfile (ofd, ifd, &offset, count);	998	res = sendfile (ofd, ifd, &soffset, count);
912		999
913	# elif __freebsd	1000	# elif __FreeBSD__
914	/*	1001	/*
915	* Of course, the freebsd sendfile is a dire hack with no thoughts	1002	* Of course, the freebsd sendfile is a dire hack with no thoughts
916	* wasted on making it similar to other I/O functions.	1003	* wasted on making it similar to other I/O functions.
917	*/	1004	*/
918	{
919	off_t sbytes;	1005	off_t sbytes;
920	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1006	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
921		1007
922	if (res < 0 && sbytes)	1008	#if 0 /* according to the manpage, this is correct, but broken behaviour */
923	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1009	/* freebsd' sendfile will return 0 on success */
		1010	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1011	/* not on e.g. EIO or EPIPE - sounds broken */
		1012	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
924	res = sbytes;	1013	res = sbytes;
925	}	1014	#endif
		1015
		1016	/* according to source inspection, this is correct, and useful behaviour */
		1017	if (sbytes)
		1018	res = sbytes;
		1019
		1020	# elif defined (__APPLE__)
		1021	off_t sbytes = count;
		1022	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1023
		1024	/* according to the manpage, sbytes is always valid */
		1025	if (sbytes)
		1026	res = sbytes;
926		1027
927	# elif __hpux	1028	# elif __hpux
928	res = sendfile (ofd, ifd, offset, count, 0, 0);	1029	res = sendfile (ofd, ifd, offset, count, 0, 0);
929		1030
930	# elif __solaris	1031	# elif __solaris
931	{
932	struct sendfilevec vec;	1032	struct sendfilevec vec;
933	size_t sbytes;	1033	size_t sbytes;
934		1034
935	vec.sfv_fd = ifd;	1035	vec.sfv_fd = ifd;
936	vec.sfv_flag = 0;	1036	vec.sfv_flag = 0;
937	vec.sfv_off = offset;	1037	vec.sfv_off = offset;
938	vec.sfv_len = count;	1038	vec.sfv_len = count;
939		1039
940	res = sendfilev (ofd, &vec, 1, &sbytes);	1040	res = sendfilev (ofd, &vec, 1, &sbytes);
941		1041
942	if (res < 0 && sbytes)	1042	if (res < 0 && sbytes)
943	res = sbytes;	1043	res = sbytes;
944	}
945		1044
946	# endif	1045	# endif
		1046
		1047	#elif defined (_WIN32)
		1048	/* does not work, just for documentation of what would need to be done */
		1049	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1050	/* libeio guarantees that the file offset does not change, and windows */
		1051	/* has no way to get an independent handle to the same file description */
		1052	HANDLE h = TO_SOCKET (ifd);
		1053	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1054	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1055
947	#else	1056	#else
948	res = -1;	1057	res = -1;
949	errno = ENOSYS;	1058	errno = ENOSYS;
950	#endif	1059	#endif
951		1060
		1061	/* we assume sendfile can copy at least 128mb in one go */
		1062	if (res <= 128 * 1024 * 1024)
		1063	{
		1064	if (res > 0)
		1065	written += res;
		1066
		1067	if (written)
		1068	return written;
		1069
		1070	break;
		1071	}
		1072	else
		1073	{
		1074	/* if we requested more, then probably the kernel was lazy */
		1075	written += res;
		1076	offset += res;
		1077	count -= res;
		1078
		1079	if (!count)
		1080	return written;
		1081	}
		1082	}
		1083
952	if (res < 0	1084	if (res < 0
953	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1085	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1086	/* BSDs */
		1087	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1088	|| errno == ENOTSUP
		1089	#endif
		1090	|| errno == EOPNOTSUPP /* BSDs */
954	#if __solaris	1091	#if __solaris
955	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1092	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
956	#endif	1093	#endif
957	)	1094	)
958	)	1095	)
…		…
989	}	1126	}
990		1127
991	return res;	1128	return res;
992	}	1129	}
993		1130
		1131	#ifdef PAGESIZE
		1132	# define eio_pagesize() PAGESIZE
		1133	#else
		1134	static intptr_t
		1135	eio_pagesize (void)
		1136	{
		1137	static intptr_t page;
		1138
		1139	if (!page)
		1140	page = sysconf (_SC_PAGESIZE);
		1141
		1142	return page;
		1143	}
		1144	#endif
		1145
		1146	static void
		1147	eio_page_align (void **addr, size_t *length)
		1148	{
		1149	intptr_t mask = eio_pagesize () - 1;
		1150
		1151	/* round down addr */
		1152	intptr_t adj = mask & (intptr_t)*addr;
		1153
		1154	*addr = (void *)((intptr_t)*addr - adj);
		1155	*length += adj;
		1156
		1157	/* round up length */
		1158	*length = (*length + mask) & ~mask;
		1159	}
		1160
		1161	#if !_POSIX_MEMLOCK
		1162	# define eio__mlockall(a) ((errno = ENOSYS), -1)
		1163	#else
		1164
994	static int	1165	static int
		1166	eio__mlockall (int flags)
		1167	{
		1168	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1169	extern int mallopt (int, int);
		1170	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1171	#endif
		1172
		1173	if (EIO_MCL_CURRENT != MCL_CURRENT
		1174	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1175	{
		1176	flags = 0
		1177	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1178	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1179	}
		1180
		1181	return mlockall (flags);
		1182	}
		1183	#endif
		1184
		1185	#if !_POSIX_MEMLOCK_RANGE
		1186	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
		1187	#else
		1188
		1189	static int
		1190	eio__mlock (void *addr, size_t length)
		1191	{
		1192	eio_page_align (&addr, &length);
		1193
		1194	return mlock (addr, length);
		1195	}
		1196
		1197	#endif
		1198
		1199	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1200	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
		1201	#else
		1202
		1203	static int
		1204	eio__msync (void *mem, size_t len, int flags)
		1205	{
		1206	eio_page_align (&mem, &len);
		1207
		1208	if (EIO_MS_ASYNC != MS_SYNC
		1209	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1210	|| EIO_MS_SYNC != MS_SYNC)
		1211	{
		1212	flags = 0
		1213	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1214	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1215	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1216	}
		1217
		1218	return msync (mem, len, flags);
		1219	}
		1220
		1221	#endif
		1222
		1223	static int
		1224	eio__mtouch (eio_req *req)
		1225	{
		1226	void *mem = req->ptr2;
		1227	size_t len = req->size;
		1228	int flags = req->int1;
		1229
		1230	eio_page_align (&mem, &len);
		1231
		1232	{
		1233	intptr_t addr = (intptr_t)mem;
		1234	intptr_t end = addr + len;
		1235	intptr_t page = eio_pagesize ();
		1236
		1237	if (addr < end)
		1238	if (flags & EIO_MT_MODIFY) /* modify */
		1239	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1240	else
		1241	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1242	}
		1243
		1244	return 0;
		1245	}
		1246
		1247	/*****************************************************************************/
		1248	/* requests implemented outside eio_execute, because they are so large */
		1249
		1250	static void
		1251	eio__realpath (eio_req *req, etp_worker *self)
		1252	{
		1253	char *rel = req->ptr1;
		1254	char *res;
		1255	char *tmp1, *tmp2;
		1256	#if SYMLOOP_MAX > 32
		1257	int symlinks = SYMLOOP_MAX;
		1258	#else
		1259	int symlinks = 32;
		1260	#endif
		1261
		1262	req->result = -1;
		1263
		1264	errno = EINVAL;
		1265	if (!rel)
		1266	return;
		1267
		1268	errno = ENOENT;
		1269	if (!*rel)
		1270	return;
		1271
		1272	if (!req->ptr2)
		1273	{
		1274	X_LOCK (wrklock);
		1275	req->flags |= EIO_FLAG_PTR2_FREE;
		1276	X_UNLOCK (wrklock);
		1277	req->ptr2 = malloc (PATH_MAX * 3);
		1278
		1279	errno = ENOMEM;
		1280	if (!req->ptr2)
		1281	return;
		1282	}
		1283
		1284	res = req->ptr2;
		1285	tmp1 = res + PATH_MAX;
		1286	tmp2 = tmp1 + PATH_MAX;
		1287
		1288	#if 0 /* disabled, the musl way to do things is just too racy */
		1289	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1290	/* on linux we may be able to ask the kernel */
		1291	{
		1292	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1293
		1294	if (fd >= 0)
		1295	{
		1296	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1297	req->result = readlink (tmp1, res, PATH_MAX);
		1298	close (fd);
		1299
		1300	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1301
		1302	if (req->result > 0)
		1303	goto done;
		1304	}
		1305	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1306	return;
		1307	}
		1308	#endif
		1309	#endif
		1310
		1311	if (*rel != '/')
		1312	{
		1313	if (!getcwd (res, PATH_MAX))
		1314	return;
		1315
		1316	if (res [1]) /* only use if not / */
		1317	res += strlen (res);
		1318	}
		1319
		1320	while (*rel)
		1321	{
		1322	ssize_t len, linklen;
		1323	char *beg = rel;
		1324
		1325	while (*rel && *rel != '/')
		1326	++rel;
		1327
		1328	len = rel - beg;
		1329
		1330	if (!len) /* skip slashes */
		1331	{
		1332	++rel;
		1333	continue;
		1334	}
		1335
		1336	if (beg [0] == '.')
		1337	{
		1338	if (len == 1)
		1339	continue; /* . - nop */
		1340
		1341	if (beg [1] == '.' && len == 2)
		1342	{
		1343	/* .. - back up one component, if possible */
		1344
		1345	while (res != req->ptr2)
		1346	if (*--res == '/')
		1347	break;
		1348
		1349	continue;
		1350	}
		1351	}
		1352
		1353	errno = ENAMETOOLONG;
		1354	if (res + 1 + len + 1 >= tmp1)
		1355	return;
		1356
		1357	/* copy one component */
		1358	*res = '/';
		1359	memcpy (res + 1, beg, len);
		1360
		1361	/* zero-terminate, for readlink */
		1362	res [len + 1] = 0;
		1363
		1364	/* now check if it's a symlink */
		1365	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1366
		1367	if (linklen < 0)
		1368	{
		1369	if (errno != EINVAL)
		1370	return;
		1371
		1372	/* it's a normal directory. hopefully */
		1373	res += len + 1;
		1374	}
		1375	else
		1376	{
		1377	/* yay, it was a symlink - build new path in tmp2 */
		1378	int rellen = strlen (rel);
		1379
		1380	errno = ENAMETOOLONG;
		1381	if (linklen + 1 + rellen >= PATH_MAX)
		1382	return;
		1383
		1384	errno = ELOOP;
		1385	if (!--symlinks)
		1386	return;
		1387
		1388	if (*tmp1 == '/')
		1389	res = req->ptr2; /* symlink resolves to an absolute path */
		1390
		1391	/* we need to be careful, as rel might point into tmp2 already */
		1392	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1393	tmp2 [linklen] = '/';
		1394	memcpy (tmp2, tmp1, linklen);
		1395
		1396	rel = tmp2;
		1397	}
		1398	}
		1399
		1400	/* special case for the lone root path */
		1401	if (res == req->ptr2)
		1402	*res++ = '/';
		1403
		1404	req->result = res - (char *)req->ptr2;
		1405
		1406	done:
		1407	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1408	}
		1409
		1410	static signed char
995	eio_dent_cmp (const void *a_, const void *b_)	1411	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
996	{	1412	{
997	const eio_dirent *a = (const eio_dirent *)a_;	1413	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
998	const eio_dirent *b = (const eio_dirent *)b_;	1414	: a->inode < b->inode ? -1
		1415	: a->inode > b->inode ? 1
		1416	: 0;
		1417	}
999		1418
1000	return (int)b->score - (int)a->score ? (int)b->score - (int)a->score	1419	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1001	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */	1420
		1421	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1422	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1423
		1424	static void
		1425	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1426	{
		1427	unsigned char bits [9 + sizeof (ino_t) * 8];
		1428	unsigned char *bit = bits;
		1429
		1430	assert (CHAR_BIT == 8);
		1431	assert (sizeof (eio_dirent) * 8 < 256);
		1432	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1433	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1434
		1435	if (size <= EIO_SORT_FAST)
		1436	return;
		1437
		1438	/* first prepare an array of bits to test in our radix sort */
		1439	/* try to take endianness into account, as well as differences in ino_t sizes */
		1440	/* inode_bits must contain all inodes ORed together */
		1441	/* which is used to skip bits that are 0 everywhere, which is very common */
		1442	{
		1443	ino_t endianness;
		1444	int i, j;
		1445
		1446	/* we store the byte offset of byte n into byte n of "endianness" */
		1447	for (i = 0; i < sizeof (ino_t); ++i)
		1448	((unsigned char *)&endianness)[i] = i;
		1449
		1450	*bit++ = 0;
		1451
		1452	for (i = 0; i < sizeof (ino_t); ++i)
		1453	{
		1454	/* shifting off the byte offsets out of "endianness" */
		1455	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1456	endianness >>= 8;
		1457
		1458	for (j = 0; j < 8; ++j)
		1459	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1460	*bit++ = offs + j;
		1461	}
		1462
		1463	for (j = 0; j < 8; ++j)
		1464	if (score_bits & (1 << j))
		1465	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1466	}
		1467
		1468	/* now actually do the sorting (a variant of MSD radix sort) */
		1469	{
		1470	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1471	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1472	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1473	int stk_idx = 0;
		1474
		1475	base_stk [stk_idx] = dents;
		1476	end_stk [stk_idx] = dents + size;
		1477	bit_stk [stk_idx] = bit - 1;
		1478
		1479	do
		1480	{
		1481	base = base_stk [stk_idx];
		1482	end = end_stk [stk_idx];
		1483	bit = bit_stk [stk_idx];
		1484
		1485	for (;;)
		1486	{
		1487	unsigned char O = *bit >> 3;
		1488	unsigned char M = 1 << (*bit & 7);
		1489
		1490	eio_dirent *a = base;
		1491	eio_dirent *b = end;
		1492
		1493	if (b - a < EIO_SORT_CUTOFF)
		1494	break;
		1495
		1496	/* now bit-partition the array on the bit */
		1497	/* this ugly asymmetric loop seems to perform much better than typical */
		1498	/* partition algos found in the literature */
		1499	do
		1500	if (!(((unsigned char *)a)[O] & M))
		1501	++a;
		1502	else if (!(((unsigned char *)--b)[O] & M))
		1503	{
		1504	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1505	++a;
		1506	}
		1507	while (b > a);
		1508
		1509	/* next bit, or stop, if no bits left in this path */
		1510	if (!*--bit)
		1511	break;
		1512
		1513	base_stk [stk_idx] = a;
		1514	end_stk [stk_idx] = end;
		1515	bit_stk [stk_idx] = bit;
		1516	++stk_idx;
		1517
		1518	end = a;
		1519	}
		1520	}
		1521	while (stk_idx--);
		1522	}
		1523	}
		1524
		1525	static void
		1526	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1527	{
		1528	/* first move the smallest element to the front, to act as a sentinel */
		1529	{
		1530	int i;
		1531	eio_dirent *min = dents;
		1532
		1533	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1534	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1535	if (EIO_DENT_CMP (dents [i], <, *min))
		1536	min = &dents [i];
		1537
		1538	/* swap elements 0 and j (minimum) */
		1539	{
		1540	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1541	}
		1542	}
		1543
		1544	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1545	{
		1546	eio_dirent *i, *j;
		1547
		1548	for (i = dents + 1; i < dents + size; ++i)
		1549	{
		1550	eio_dirent value = *i;
		1551
		1552	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1553	j [1] = j [0];
		1554
		1555	j [1] = value;
		1556	}
		1557	}
		1558	}
		1559
		1560	static void
		1561	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1562	{
		1563	if (size <= 1)
		1564	return; /* our insertion sort relies on size > 0 */
		1565
		1566	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1567	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1568	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1569
		1570	/* use an insertion sort at the end, or for small arrays, */
		1571	/* as insertion sort is more efficient for small partitions */
		1572	eio_dent_insertion_sort (dents, size);
1002	}	1573	}
1003		1574
1004	/* read a full directory */	1575	/* read a full directory */
1005	static void	1576	static void
1006	eio__scandir (eio_req *req, etp_worker *self)	1577	eio__scandir (eio_req *req, etp_worker *self)
1007	{	1578	{
1008	DIR *dirp;	1579	DIR *dirp;
1009	EIO_STRUCT_DIRENT *entp;	1580	EIO_STRUCT_DIRENT *entp;
1010	unsigned char *name, *names;	1581	char *name, *names;
1011	int namesalloc = 4096;	1582	int namesalloc = 4096;
1012	int namesoffs = 0;	1583	int namesoffs = 0;
1013	int flags = req->int1;	1584	int flags = req->int1;
1014	eio_dirent *dents = 0;	1585	eio_dirent *dents = 0;
1015	int dentalloc = 128;	1586	int dentalloc = 128;
1016	int dentoffs = 0;	1587	int dentoffs = 0;
		1588	ino_t inode_bits = 0;
1017		1589
1018	req->result = -1;	1590	req->result = -1;
1019		1591
1020	if (!(flags & EIO_READDIR_DENTS))	1592	if (!(flags & EIO_READDIR_DENTS))
1021	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1593	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1022		1594
1023	X_LOCK (wrklock);	1595	X_LOCK (wrklock);
1024	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1596	/* the corresponding closedir is in ETP_WORKER_CLEAR */
1025	self->dirp = dirp = opendir (req->ptr1);	1597	self->dirp = dirp = opendir (req->ptr1);
		1598
		1599	if (req->flags & EIO_FLAG_PTR1_FREE)
		1600	free (req->ptr1);
		1601
1026	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1602	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1027	req->ptr1 = names = malloc (namesalloc);
1028	req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1603	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
		1604	req->ptr2 = names = malloc (namesalloc);
1029	X_UNLOCK (wrklock);	1605	X_UNLOCK (wrklock);
1030		1606
1031	if (dirp && names && (!flags || dents))	1607	if (dirp && names && (!flags || dents))
1032	for (;;)	1608	for (;;)
1033	{	1609	{
…		…
1041		1617
1042	/* sort etc. */	1618	/* sort etc. */
1043	req->int1 = flags;	1619	req->int1 = flags;
1044	req->result = dentoffs;	1620	req->result = dentoffs;
1045		1621
1046	if (dents)
1047	{
1048	eio_dirent *ent = dents + dentoffs;
1049
1050	while (ent > dents)
1051	(--ent)->name = names + (size_t)ent->name;
1052	}
1053
1054	if (flags & EIO_READDIR_STAT_ORDER	1622	if (flags & EIO_READDIR_STAT_ORDER)
1055	|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))	1623	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1056	{
1057	/* pray your qsort doesn't use quicksort */
1058	qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1059	}
1060	else if (flags & EIO_READDIR_DIRS_FIRST)	1624	else if (flags & EIO_READDIR_DIRS_FIRST)
		1625	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1626	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1627	else
1061	{	1628	{
1062	/* in this case, all is known, and we just put dirs first and sort them */	1629	/* in this case, all is known, and we just put dirs first and sort them */
1063	eio_dirent *ent = dents + dentoffs;	1630	eio_dirent *oth = dents + dentoffs;
1064	eio_dirent *dir = dents;	1631	eio_dirent *dir = dents;
1065		1632
		1633	/* now partition dirs to the front, and non-dirs to the back */
		1634	/* by walking from both sides and swapping if necessary */
1066	while (ent > dir)	1635	while (oth > dir)
1067	{	1636	{
1068	if (dir->type == DT_DIR)	1637	if (dir->type == EIO_DT_DIR)
1069	++dir;	1638	++dir;
1070	else
1071	{
1072	--ent;
1073
1074	if (ent->type == DT_DIR)	1639	else if ((--oth)->type == EIO_DT_DIR)
1075	{	1640	{
1076	eio_dirent tmp = *dir;	1641	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1077	*dir = *ent;
1078	*ent = tmp;
1079		1642
1080	++dir;	1643	++dir;
1081	}	1644	}
1082	}	1645	}
		1646
		1647	/* now sort the dirs only (dirs all have the same score) */
		1648	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1083	}	1649	}
1084
1085	/* now sort the dirs only */
1086	qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1087	}
1088
1089	/* only provide the names array unless DENTS is specified */
1090	if (!(flags & EIO_READDIR_DENTS))
1091	{
1092	X_LOCK (wrklock);
1093	assert (!dents);
1094	req->ptr1 = 0;
1095	req->ptr2 = names;
1096	X_UNLOCK (wrklock);
1097	}
1098		1650
1099	break;	1651	break;
1100	}	1652	}
1101		1653
1102	/* now add the entry to our list(s) */	1654	/* now add the entry to our list(s) */
…		…
1105	/* skip . and .. entries */	1657	/* skip . and .. entries */
1106	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1658	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1107	{	1659	{
1108	int len = D_NAMLEN (entp) + 1;	1660	int len = D_NAMLEN (entp) + 1;
1109		1661
1110	while (expect_false (namesoffs + len > namesalloc))	1662	while (ecb_expect_false (namesoffs + len > namesalloc))
1111	{	1663	{
1112	namesalloc *= 2;	1664	namesalloc *= 2;
1113	X_LOCK (wrklock);	1665	X_LOCK (wrklock);
1114	req->ptr1 = names = realloc (names, namesalloc);	1666	req->ptr2 = names = realloc (names, namesalloc);
1115	X_UNLOCK (wrklock);	1667	X_UNLOCK (wrklock);
1116		1668
1117	if (!names)	1669	if (!names)
1118	break;	1670	break;
1119	}	1671	}
…		…
1122		1674
1123	if (dents)	1675	if (dents)
1124	{	1676	{
1125	struct eio_dirent *ent;	1677	struct eio_dirent *ent;
1126		1678
1127	if (expect_false (dentoffs == dentalloc))	1679	if (ecb_expect_false (dentoffs == dentalloc))
1128	{	1680	{
1129	dentalloc *= 2;	1681	dentalloc *= 2;
1130	X_LOCK (wrklock);	1682	X_LOCK (wrklock);
1131	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1683	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1132	X_UNLOCK (wrklock);	1684	X_UNLOCK (wrklock);
1133		1685
1134	if (!dents)	1686	if (!dents)
1135	break;	1687	break;
1136	}	1688	}
1137		1689
1138	ent = dents + dentoffs;	1690	ent = dents + dentoffs;
1139		1691
1140	ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */	1692	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1141	ent->namelen = len - 1;	1693	ent->namelen = len - 1;
1142	ent->inode = D_INO (entp);	1694	ent->inode = D_INO (entp);
		1695
		1696	inode_bits |= ent->inode;
1143		1697
1144	switch (D_TYPE (entp))	1698	switch (D_TYPE (entp))
1145	{	1699	{
1146	default:	1700	default:
1147	ent->type = EIO_DT_UNKNOWN;	1701	ent->type = EIO_DT_UNKNOWN;
…		…
1190	#ifdef DT_WHT	1744	#ifdef DT_WHT
1191	case DT_WHT: ent->type = EIO_DT_WHT; break;	1745	case DT_WHT: ent->type = EIO_DT_WHT; break;
1192	#endif	1746	#endif
1193	}	1747	}
1194		1748
1195	ent->score = 0;	1749	ent->score = 7;
1196		1750
1197	if (flags & EIO_READDIR_DIRS_FIRST)	1751	if (flags & EIO_READDIR_DIRS_FIRST)
1198	{	1752	{
1199	if (ent->type == EIO_DT_UNKNOWN)	1753	if (ent->type == EIO_DT_UNKNOWN)
1200	{	1754	{
1201	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */	1755	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1202	ent->score = 98;	1756	ent->score = 1;
1203	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */	1757	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1204	ent->score = len <= 2 ? len + 60 : len <= 4 ? 50 : len <= 7 ? 40 : 10; /* shorter == more likely dir, but avoid too many classes */	1758	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1205	}	1759	}
1206	else if (ent->type == EIO_DT_DIR)	1760	else if (ent->type == EIO_DT_DIR)
1207	ent->score = 100;	1761	ent->score = 0;
1208	}	1762	}
1209	}	1763	}
1210		1764
1211	namesoffs += len;	1765	namesoffs += len;
1212	++dentoffs;	1766	++dentoffs;
1213	}	1767	}
		1768
		1769	if (EIO_CANCELLED (req))
		1770	{
		1771	errno = ECANCELED;
		1772	break;
		1773	}
1214	}	1774	}
1215	else
1216	req->result = -1;
1217	}
1218
1219	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1220	# undef msync
1221	# define msync(a,b,c) ((errno = ENOSYS), -1)
1222	#endif
1223
1224	int
1225	eio__mtouch (void *mem, size_t len, int flags)
1226	{
1227	intptr_t addr = (intptr_t)mem;
1228	intptr_t end = addr + len;
1229	#ifdef PAGESIZE
1230	const intptr_t page = PAGESIZE;
1231	#else
1232	static intptr_t page;
1233
1234	if (!page)
1235	page = sysconf (_SC_PAGESIZE);
1236	#endif
1237
1238	addr &= ~(page - 1); /* assume page size is always a power of two */
1239
1240	if (addr < end)
1241	if (flags) /* modify */
1242	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1243	else
1244	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1245
1246	return 0;
1247	}	1775	}
1248		1776
1249	/*****************************************************************************/	1777	/*****************************************************************************/
1250		1778
1251	#define ALLOC(len) \	1779	#define ALLOC(len) \
…		…
1266	X_THREAD_PROC (etp_proc)	1794	X_THREAD_PROC (etp_proc)
1267	{	1795	{
1268	ETP_REQ *req;	1796	ETP_REQ *req;
1269	struct timespec ts;	1797	struct timespec ts;
1270	etp_worker *self = (etp_worker *)thr_arg;	1798	etp_worker *self = (etp_worker *)thr_arg;
		1799	int timeout;
1271		1800
1272	/* try to distribute timeouts somewhat randomly */	1801	/* try to distribute timeouts somewhat evenly */
1273	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1802	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1274		1803
1275	for (;;)	1804	for (;;)
1276	{	1805	{
		1806	ts.tv_sec = 0;
		1807
1277	X_LOCK (reqlock);	1808	X_LOCK (reqlock);
1278		1809
1279	for (;;)	1810	for (;;)
1280	{	1811	{
1281	self->req = req = reqq_shift (&req_queue);	1812	self->req = req = reqq_shift (&req_queue);
1282		1813
1283	if (req)	1814	if (req)
1284	break;	1815	break;
1285		1816
		1817	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1818	{
		1819	X_UNLOCK (reqlock);
		1820	X_LOCK (wrklock);
		1821	--started;
		1822	X_UNLOCK (wrklock);
		1823	goto quit;
		1824	}
		1825
1286	++idle;	1826	++idle;
1287		1827
1288	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1828	if (idle <= max_idle)
1289	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1829	/* we are allowed to idle, so do so without any timeout */
		1830	X_COND_WAIT (reqwait, reqlock);
		1831	else
1290	{	1832	{
1291	if (idle > max_idle)	1833	/* initialise timeout once */
1292	{	1834	if (!ts.tv_sec)
1293	--idle;	1835	ts.tv_sec = time (0) + idle_timeout;
1294	X_UNLOCK (reqlock);
1295	X_LOCK (wrklock);
1296	--started;
1297	X_UNLOCK (wrklock);
1298	goto quit;
1299	}
1300		1836
1301	/* we are allowed to idle, so do so without any timeout */
1302	X_COND_WAIT (reqwait, reqlock);	1837	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1838	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1303	}	1839	}
1304		1840
1305	--idle;	1841	--idle;
1306	}	1842	}
1307		1843
…		…
1310	X_UNLOCK (reqlock);	1846	X_UNLOCK (reqlock);
1311		1847
1312	if (req->type < 0)	1848	if (req->type < 0)
1313	goto quit;	1849	goto quit;
1314		1850
1315	if (!EIO_CANCELLED (req))
1316	ETP_EXECUTE (self, req);	1851	ETP_EXECUTE (self, req);
1317		1852
1318	X_LOCK (reslock);	1853	X_LOCK (reslock);
1319		1854
1320	++npending;	1855	++npending;
1321		1856
…		…
1336	return 0;	1871	return 0;
1337	}	1872	}
1338		1873
1339	/*****************************************************************************/	1874	/*****************************************************************************/
1340		1875
		1876	int ecb_cold
1341	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1877	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1342	{	1878	{
1343	return etp_init (want_poll, done_poll);	1879	return etp_init (want_poll, done_poll);
1344	}	1880	}
1345		1881
		1882	ecb_inline void
1346	static void eio_api_destroy (eio_req *req)	1883	eio_api_destroy (eio_req *req)
1347	{	1884	{
1348	free (req);	1885	free (req);
1349	}	1886	}
1350		1887
1351	#define REQ(rtype) \	1888	#define REQ(rtype) \
…		…
1370	{ \	1907	{ \
1371	eio_api_destroy (req); \	1908	eio_api_destroy (req); \
1372	return 0; \	1909	return 0; \
1373	}	1910	}
1374		1911
		1912	static void
1375	static void eio_execute (etp_worker *self, eio_req *req)	1913	eio_execute (etp_worker *self, eio_req *req)
1376	{	1914	{
1377	errno = 0;	1915	if (ecb_expect_false (EIO_CANCELLED (req)))
		1916	{
		1917	req->result = -1;
		1918	req->errorno = ECANCELED;
		1919	return;
		1920	}
1378		1921
1379	switch (req->type)	1922	switch (req->type)
1380	{	1923	{
1381	case EIO_READ: ALLOC (req->size);	1924	case EIO_READ: ALLOC (req->size);
1382	req->result = req->offs >= 0	1925	req->result = req->offs >= 0
…		…
1394	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1937	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1395	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1938	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1396	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1939	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1397	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1940	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1398		1941
		1942	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1943	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1944	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1945	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1946
1399	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	1947	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1400	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	1948	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1401	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	1949	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1402	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	1950	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1403	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;	1951	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
…		…
1410	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	1958	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1411	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	1959	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1412	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	1960	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1413	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	1961	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1414	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	1962	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1415	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;	1963	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1416		1964
		1965	case EIO_REALPATH: eio__realpath (req, self); break;
		1966
1417	case EIO_READLINK: ALLOC (NAME_MAX);	1967	case EIO_READLINK: ALLOC (PATH_MAX);
1418	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	1968	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1419		1969
1420	case EIO_SYNC: req->result = 0; sync (); break;	1970	case EIO_SYNC: req->result = 0; sync (); break;
1421	case EIO_FSYNC: req->result = fsync (req->int1); break;	1971	case EIO_FSYNC: req->result = fsync (req->int1); break;
1422	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	1972	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1423	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;	1973	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1974	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1424	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	1975	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		1976	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1425	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	1977	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1978	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1426		1979
1427	case EIO_READDIR: eio__scandir (req, self); break;	1980	case EIO_READDIR: eio__scandir (req, self); break;
1428		1981
1429	case EIO_BUSY:	1982	case EIO_BUSY:
1430	#ifdef _WIN32	1983	#ifdef _WIN32
1431	Sleep (req->nv1 * 1000.);	1984	Sleep (req->nv1 * 1e3);
1432	#else	1985	#else
1433	{	1986	{
1434	struct timeval tv;	1987	struct timeval tv;
1435		1988
1436	tv.tv_sec = req->nv1;	1989	tv.tv_sec = req->nv1;
1437	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	1990	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1438		1991
1439	req->result = select (0, 0, 0, 0, &tv);	1992	req->result = select (0, 0, 0, 0, &tv);
1440	}	1993	}
1441	#endif	1994	#endif
1442	break;	1995	break;
…		…
1457	times = tv;	2010	times = tv;
1458	}	2011	}
1459	else	2012	else
1460	times = 0;	2013	times = 0;
1461		2014
1462
1463	req->result = req->type == EIO_FUTIME	2015	req->result = req->type == EIO_FUTIME
1464	? futimes (req->int1, times)	2016	? futimes (req->int1, times)
1465	: utimes (req->ptr1, times);	2017	: utimes (req->ptr1, times);
1466	}	2018	}
1467	break;	2019	break;
…		…
1472	case EIO_NOP:	2024	case EIO_NOP:
1473	req->result = 0;	2025	req->result = 0;
1474	break;	2026	break;
1475		2027
1476	case EIO_CUSTOM:	2028	case EIO_CUSTOM:
1477	((void (*)(eio_req *))req->feed) (req);	2029	req->feed (req);
1478	break;	2030	break;
1479		2031
1480	default:	2032	default:
		2033	errno = ENOSYS;
1481	req->result = -1;	2034	req->result = -1;
1482	break;	2035	break;
1483	}	2036	}
1484		2037
1485	req->errorno = errno;	2038	req->errorno = errno;
…		…
1515	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2068	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1516	{	2069	{
1517	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2070	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1518	}	2071	}
1519		2072
		2073	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2074	{
		2075	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2076	}
		2077
		2078	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2079	{
		2080	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2081	}
		2082
1520	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2083	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1521	{	2084	{
1522	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2085	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1523	}	2086	}
1524		2087
		2088	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2089	{
		2090	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2091	}
		2092
1525	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2093	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1526	{	2094	{
1527	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2095	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1528	}	2096	}
1529		2097
…		…
1548	}	2116	}
1549		2117
1550	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2118	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1551	{	2119	{
1552	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2120	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		2121	}
		2122
		2123	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2124	{
		2125	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1553	}	2126	}
1554		2127
1555	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2128	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1556	{	2129	{
1557	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2130	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
…		…
1621	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2194	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1622	{	2195	{
1623	return eio__1path (EIO_READLINK, path, pri, cb, data);	2196	return eio__1path (EIO_READLINK, path, pri, cb, data);
1624	}	2197	}
1625		2198
		2199	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2200	{
		2201	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2202	}
		2203
1626	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2204	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1627	{	2205	{
1628	return eio__1path (EIO_STAT, path, pri, cb, data);	2206	return eio__1path (EIO_STAT, path, pri, cb, data);
1629	}	2207	}
1630		2208
1631	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2209	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1632	{	2210	{
1633	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2211	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1634	}	2212	}
1635		2213
		2214	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2215	{
		2216	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2217	}
		2218
1636	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2219	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1637	{	2220	{
1638	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2221	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1639	}	2222	}
1640		2223
…		…
1648	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2231	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1649	}	2232	}
1650		2233
1651	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2234	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1652	{	2235	{
1653	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2236	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1654	}	2237	}
1655		2238
1656	static eio_req *	2239	static eio_req *
1657	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2240	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1658	{	2241	{
…		…
1682	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2265	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1683	{	2266	{
1684	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2267	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1685	}	2268	}
1686		2269
1687	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2270	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1688	{	2271	{
1689	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2272	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1690	}	2273	}
1691		2274
1692	#endif	2275	#endif
1693		2276
1694	eio_req *eio_grp (eio_cb cb, void *data)	2277	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1703	#undef SEND	2286	#undef SEND
1704		2287
1705	/*****************************************************************************/	2288	/*****************************************************************************/
1706	/* grp functions */	2289	/* grp functions */
1707		2290
		2291	void
1708	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2292	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1709	{	2293	{
1710	grp->int2 = limit;	2294	grp->int2 = limit;
1711	grp->feed = feed;	2295	grp->feed = feed;
1712		2296
1713	grp_try_feed (grp);	2297	grp_try_feed (grp);
1714	}	2298	}
1715		2299
		2300	void
1716	void eio_grp_limit (eio_req *grp, int limit)	2301	eio_grp_limit (eio_req *grp, int limit)
1717	{	2302	{
1718	grp->int2 = limit;	2303	grp->int2 = limit;
1719		2304
1720	grp_try_feed (grp);	2305	grp_try_feed (grp);
1721	}	2306	}
1722		2307
		2308	void
1723	void eio_grp_add (eio_req *grp, eio_req *req)	2309	eio_grp_add (eio_req *grp, eio_req *req)
1724	{	2310	{
1725	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2311	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1726		2312
1727	grp->flags |= EIO_FLAG_GROUPADD;	2313	grp->flags |= EIO_FLAG_GROUPADD;
1728		2314
…		…
1739	}	2325	}
1740		2326
1741	/*****************************************************************************/	2327	/*****************************************************************************/
1742	/* misc garbage */	2328	/* misc garbage */
1743		2329
		2330	ssize_t
1744	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2331	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1745	{	2332	{
1746	etp_worker wrk;	2333	etp_worker wrk;
		2334	ssize_t ret;
1747		2335
1748	wrk.dbuf = 0;	2336	wrk.dbuf = 0;
1749		2337
1750	eio__sendfile (ofd, ifd, offset, count, &wrk);	2338	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1751		2339
1752	if (wrk.dbuf)	2340	if (wrk.dbuf)
1753	free (wrk.dbuf);	2341	free (wrk.dbuf);
1754	}
1755		2342
		2343	return ret;
		2344	}
		2345

Diff Legend

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