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 1.80 by root, Tue Jul 5 18:59:28 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
		383	static void ecb_cold
		384	etp_thread_init (void)
		385	{
		386	X_MUTEX_CREATE (wrklock);
		387	X_MUTEX_CREATE (reslock);
		388	X_MUTEX_CREATE (reqlock);
		389	X_COND_CREATE (reqwait);
		390	}
		391
		392	static void ecb_cold
363	static void etp_atfork_prepare (void)	393	etp_atfork_prepare (void)
364	{	394	{
365	X_LOCK (wrklock);	395	X_LOCK (wrklock);
366	X_LOCK (reqlock);	396	X_LOCK (reqlock);
367	X_LOCK (reslock);	397	X_LOCK (reslock);
368	#if !HAVE_PREADWRITE	398	#if !HAVE_PREADWRITE
369	X_LOCK (preadwritelock);	399	X_LOCK (preadwritelock);
370	#endif	400	#endif
371	}	401	}
372		402
		403	static void ecb_cold
373	static void etp_atfork_parent (void)	404	etp_atfork_parent (void)
374	{	405	{
375	#if !HAVE_PREADWRITE	406	#if !HAVE_PREADWRITE
376	X_UNLOCK (preadwritelock);	407	X_UNLOCK (preadwritelock);
377	#endif	408	#endif
378	X_UNLOCK (reslock);	409	X_UNLOCK (reslock);
379	X_UNLOCK (reqlock);	410	X_UNLOCK (reqlock);
380	X_UNLOCK (wrklock);	411	X_UNLOCK (wrklock);
381	}	412	}
382		413
		414	static void ecb_cold
383	static void etp_atfork_child (void)	415	etp_atfork_child (void)
384	{	416	{
385	ETP_REQ *prv;	417	ETP_REQ *prv;
386		418
387	while ((prv = reqq_shift (&req_queue)))	419	while ((prv = reqq_shift (&req_queue)))
388	ETP_DESTROY (prv);	420	ETP_DESTROY (prv);
…		…
405	idle = 0;	437	idle = 0;
406	nreqs = 0;	438	nreqs = 0;
407	nready = 0;	439	nready = 0;
408	npending = 0;	440	npending = 0;
409		441
410	etp_atfork_parent ();	442	etp_thread_init ();
411	}	443	}
412		444
413	static void	445	static void ecb_cold
414	etp_once_init (void)	446	etp_once_init (void)
415	{	447	{
		448	etp_thread_init ();
416	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);	449	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
417	}	450	}
418		451
419	static int	452	static int ecb_cold
420	etp_init (void (*want_poll)(void), void (*done_poll)(void))	453	etp_init (void (*want_poll)(void), void (*done_poll)(void))
421	{	454	{
422	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	455	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
423		456
424	pthread_once (&doinit, etp_once_init);	457	pthread_once (&doinit, etp_once_init);
…		…
429	return 0;	462	return 0;
430	}	463	}
431		464
432	X_THREAD_PROC (etp_proc);	465	X_THREAD_PROC (etp_proc);
433		466
		467	static void ecb_cold
434	static void etp_start_thread (void)	468	etp_start_thread (void)
435	{	469	{
436	etp_worker *wrk = calloc (1, sizeof (etp_worker));	470	etp_worker *wrk = calloc (1, sizeof (etp_worker));
437		471
438	/*TODO*/	472	/*TODO*/
439	assert (("unable to allocate worker thread data", wrk));	473	assert (("unable to allocate worker thread data", wrk));
…		…
452	free (wrk);	486	free (wrk);
453		487
454	X_UNLOCK (wrklock);	488	X_UNLOCK (wrklock);
455	}	489	}
456		490
		491	static void
457	static void etp_maybe_start_thread (void)	492	etp_maybe_start_thread (void)
458	{	493	{
459	if (expect_true (etp_nthreads () >= wanted))	494	if (ecb_expect_true (etp_nthreads () >= wanted))
460	return;	495	return;
461		496
462	/* todo: maybe use idle here, but might be less exact */	497	/* todo: maybe use idle here, but might be less exact */
463	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	498	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
464	return;	499	return;
465		500
466	etp_start_thread ();	501	etp_start_thread ();
467	}	502	}
468		503
		504	static void ecb_cold
469	static void etp_end_thread (void)	505	etp_end_thread (void)
470	{	506	{
471	eio_req *req = calloc (1, sizeof (eio_req));	507	eio_req *req = calloc (1, sizeof (eio_req));
472		508
473	req->type = -1;	509	req->type = -1;
474	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	510	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
481	X_LOCK (wrklock);	517	X_LOCK (wrklock);
482	--started;	518	--started;
483	X_UNLOCK (wrklock);	519	X_UNLOCK (wrklock);
484	}	520	}
485		521
486	static int etp_poll (void)	522	static int
		523	etp_poll (void)
487	{	524	{
488	unsigned int maxreqs;	525	unsigned int maxreqs;
489	unsigned int maxtime;	526	unsigned int maxtime;
490	struct timeval tv_start, tv_now;	527	struct timeval tv_start, tv_now;
491		528
…		…
521		558
522	X_LOCK (reqlock);	559	X_LOCK (reqlock);
523	--nreqs;	560	--nreqs;
524	X_UNLOCK (reqlock);	561	X_UNLOCK (reqlock);
525		562
526	if (expect_false (req->type == EIO_GROUP && req->size))	563	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
527	{	564	{
528	req->int1 = 1; /* mark request as delayed */	565	req->int1 = 1; /* mark request as delayed */
529	continue;	566	continue;
530	}	567	}
531	else	568	else
532	{	569	{
533	int res = ETP_FINISH (req);	570	int res = ETP_FINISH (req);
534	if (expect_false (res))	571	if (ecb_expect_false (res))
535	return res;	572	return res;
536	}	573	}
537		574
538	if (expect_false (maxreqs && !--maxreqs))	575	if (ecb_expect_false (maxreqs && !--maxreqs))
539	break;	576	break;
540		577
541	if (maxtime)	578	if (maxtime)
542	{	579	{
543	gettimeofday (&tv_now, 0);	580	gettimeofday (&tv_now, 0);
…		…
549		586
550	errno = EAGAIN;	587	errno = EAGAIN;
551	return -1;	588	return -1;
552	}	589	}
553		590
		591	static void
554	static void etp_cancel (ETP_REQ *req)	592	etp_cancel (ETP_REQ *req)
555	{	593	{
556	X_LOCK (wrklock);	594	X_LOCK (wrklock);
557	req->flags |= EIO_FLAG_CANCELLED;	595	req->flags |= EIO_FLAG_CANCELLED;
558	X_UNLOCK (wrklock);	596	X_UNLOCK (wrklock);
559		597
560	eio_grp_cancel (req);	598	eio_grp_cancel (req);
561	}	599	}
562		600
		601	static void
563	static void etp_submit (ETP_REQ *req)	602	etp_submit (ETP_REQ *req)
564	{	603	{
565	req->pri -= ETP_PRI_MIN;	604	req->pri -= ETP_PRI_MIN;
566		605
567	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	606	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;	607	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
569		608
570	if (expect_false (req->type == EIO_GROUP))	609	if (ecb_expect_false (req->type == EIO_GROUP))
571	{	610	{
572	/* I hope this is worth it :/ */	611	/* I hope this is worth it :/ */
573	X_LOCK (reqlock);	612	X_LOCK (reqlock);
574	++nreqs;	613	++nreqs;
575	X_UNLOCK (reqlock);	614	X_UNLOCK (reqlock);
…		…
594		633
595	etp_maybe_start_thread ();	634	etp_maybe_start_thread ();
596	}	635	}
597	}	636	}
598		637
		638	static void ecb_cold
599	static void etp_set_max_poll_time (double nseconds)	639	etp_set_max_poll_time (double nseconds)
600	{	640	{
601	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	641	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
602	max_poll_time = nseconds;	642	max_poll_time = nseconds * EIO_TICKS;
603	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	643	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
604	}	644	}
605		645
		646	static void ecb_cold
606	static void etp_set_max_poll_reqs (unsigned int maxreqs)	647	etp_set_max_poll_reqs (unsigned int maxreqs)
607	{	648	{
608	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	649	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609	max_poll_reqs = maxreqs;	650	max_poll_reqs = maxreqs;
610	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	651	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611	}	652	}
612		653
		654	static void ecb_cold
613	static void etp_set_max_idle (unsigned int nthreads)	655	etp_set_max_idle (unsigned int nthreads)
614	{	656	{
615	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	657	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
616	max_idle = nthreads <= 0 ? 1 : nthreads;	658	max_idle = nthreads;
617	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	659	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
618	}	660	}
619		661
		662	static void ecb_cold
		663	etp_set_idle_timeout (unsigned int seconds)
		664	{
		665	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		666	idle_timeout = seconds;
		667	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		668	}
		669
		670	static void ecb_cold
620	static void etp_set_min_parallel (unsigned int nthreads)	671	etp_set_min_parallel (unsigned int nthreads)
621	{	672	{
622	if (wanted < nthreads)	673	if (wanted < nthreads)
623	wanted = nthreads;	674	wanted = nthreads;
624	}	675	}
625		676
		677	static void ecb_cold
626	static void etp_set_max_parallel (unsigned int nthreads)	678	etp_set_max_parallel (unsigned int nthreads)
627	{	679	{
628	if (wanted > nthreads)	680	if (wanted > nthreads)
629	wanted = nthreads;	681	wanted = nthreads;
630		682
631	while (started > wanted)	683	while (started > wanted)
632	etp_end_thread ();	684	etp_end_thread ();
633	}	685	}
634		686
635	/*****************************************************************************/	687	/*****************************************************************************/
636		688
		689	static void
637	static void grp_try_feed (eio_req *grp)	690	grp_try_feed (eio_req *grp)
638	{	691	{
639	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	692	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
640	{	693	{
641	grp->flags &= ~EIO_FLAG_GROUPADD;	694	grp->flags &= ~EIO_FLAG_GROUPADD;
642		695
…		…
649	break;	702	break;
650	}	703	}
651	}	704	}
652	}	705	}
653		706
		707	static int
654	static int grp_dec (eio_req *grp)	708	grp_dec (eio_req *grp)
655	{	709	{
656	--grp->size;	710	--grp->size;
657		711
658	/* call feeder, if applicable */	712	/* call feeder, if applicable */
659	grp_try_feed (grp);	713	grp_try_feed (grp);
…		…
663	return eio_finish (grp);	717	return eio_finish (grp);
664	else	718	else
665	return 0;	719	return 0;
666	}	720	}
667		721
		722	static void
668	void eio_destroy (eio_req *req)	723	eio_destroy (eio_req *req)
669	{	724	{
670	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	725	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
671	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	726	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
672		727
673	EIO_DESTROY (req);	728	EIO_DESTROY (req);
674	}	729	}
675		730
		731	static int
676	static int eio_finish (eio_req *req)	732	eio_finish (eio_req *req)
677	{	733	{
678	int res = EIO_FINISH (req);	734	int res = EIO_FINISH (req);
679		735
680	if (req->grp)	736	if (req->grp)
681	{	737	{
…		…
689	if (grp->grp_first == req)	745	if (grp->grp_first == req)
690	grp->grp_first = req->grp_next;	746	grp->grp_first = req->grp_next;
691		747
692	res2 = grp_dec (grp);	748	res2 = grp_dec (grp);
693		749
694	if (!res && res2)	750	if (!res)
695	res = res2;	751	res = res2;
696	}	752	}
697		753
698	eio_destroy (req);	754	eio_destroy (req);
699		755
700	return res;	756	return res;
701	}	757	}
702		758
		759	void
703	void eio_grp_cancel (eio_req *grp)	760	eio_grp_cancel (eio_req *grp)
704	{	761	{
705	for (grp = grp->grp_first; grp; grp = grp->grp_next)	762	for (grp = grp->grp_first; grp; grp = grp->grp_next)
706	eio_cancel (grp);	763	eio_cancel (grp);
707	}	764	}
708		765
		766	void
709	void eio_cancel (eio_req *req)	767	eio_cancel (eio_req *req)
710	{	768	{
711	etp_cancel (req);	769	etp_cancel (req);
712	}	770	}
713		771
		772	void
714	void eio_submit (eio_req *req)	773	eio_submit (eio_req *req)
715	{	774	{
716	etp_submit (req);	775	etp_submit (req);
717	}	776	}
718		777
719	unsigned int eio_nreqs (void)	778	unsigned int
		779	eio_nreqs (void)
720	{	780	{
721	return etp_nreqs ();	781	return etp_nreqs ();
722	}	782	}
723		783
724	unsigned int eio_nready (void)	784	unsigned int
		785	eio_nready (void)
725	{	786	{
726	return etp_nready ();	787	return etp_nready ();
727	}	788	}
728		789
729	unsigned int eio_npending (void)	790	unsigned int
		791	eio_npending (void)
730	{	792	{
731	return etp_npending ();	793	return etp_npending ();
732	}	794	}
733		795
734	unsigned int eio_nthreads (void)	796	unsigned int ecb_cold
		797	eio_nthreads (void)
735	{	798	{
736	return etp_nthreads ();	799	return etp_nthreads ();
737	}	800	}
738		801
		802	void ecb_cold
739	void eio_set_max_poll_time (double nseconds)	803	eio_set_max_poll_time (double nseconds)
740	{	804	{
741	etp_set_max_poll_time (nseconds);	805	etp_set_max_poll_time (nseconds);
742	}	806	}
743		807
		808	void ecb_cold
744	void eio_set_max_poll_reqs (unsigned int maxreqs)	809	eio_set_max_poll_reqs (unsigned int maxreqs)
745	{	810	{
746	etp_set_max_poll_reqs (maxreqs);	811	etp_set_max_poll_reqs (maxreqs);
747	}	812	}
748		813
		814	void ecb_cold
749	void eio_set_max_idle (unsigned int nthreads)	815	eio_set_max_idle (unsigned int nthreads)
750	{	816	{
751	etp_set_max_idle (nthreads);	817	etp_set_max_idle (nthreads);
752	}	818	}
753		819
		820	void ecb_cold
		821	eio_set_idle_timeout (unsigned int seconds)
		822	{
		823	etp_set_idle_timeout (seconds);
		824	}
		825
		826	void ecb_cold
754	void eio_set_min_parallel (unsigned int nthreads)	827	eio_set_min_parallel (unsigned int nthreads)
755	{	828	{
756	etp_set_min_parallel (nthreads);	829	etp_set_min_parallel (nthreads);
757	}	830	}
758		831
		832	void ecb_cold
759	void eio_set_max_parallel (unsigned int nthreads)	833	eio_set_max_parallel (unsigned int nthreads)
760	{	834	{
761	etp_set_max_parallel (nthreads);	835	etp_set_max_parallel (nthreads);
762	}	836	}
763		837
764	int eio_poll (void)	838	int eio_poll (void)
…		…
806		880
807	return res;	881	return res;
808	}	882	}
809	#endif	883	#endif
810		884
811	#ifndef HAVE_FUTIMES	885	#ifndef HAVE_UTIMES
812		886
813	# undef utimes	887	# undef utimes
814	# undef futimes
815	# define utimes(path,times) eio__utimes (path, times)	888	# define utimes(path,times) eio__utimes (path, times)
816	# define futimes(fd,times) eio__futimes (fd, times)
817		889
818	static int	890	static int
819	eio__utimes (const char *filename, const struct timeval times[2])	891	eio__utimes (const char *filename, const struct timeval times[2])
820	{	892	{
821	if (times)	893	if (times)
…		…
829	}	901	}
830	else	902	else
831	return utime (filename, 0);	903	return utime (filename, 0);
832	}	904	}
833		905
		906	#endif
		907
		908	#ifndef HAVE_FUTIMES
		909
		910	# undef futimes
		911	# define futimes(fd,times) eio__futimes (fd, times)
		912
		913	static int
834	static int eio__futimes (int fd, const struct timeval tv[2])	914	eio__futimes (int fd, const struct timeval tv[2])
835	{	915	{
836	errno = ENOSYS;	916	errno = ENOSYS;
837	return -1;	917	return -1;
838	}	918	}
839		919
…		…
843	# undef fdatasync	923	# undef fdatasync
844	# define fdatasync(fd) fsync (fd)	924	# define fdatasync(fd) fsync (fd)
845	#endif	925	#endif
846		926
847	/* sync_file_range always needs emulation */	927	/* sync_file_range always needs emulation */
848	int	928	static int
849	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	929	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
850	{	930	{
851	#if HAVE_SYNC_FILE_RANGE	931	#if HAVE_SYNC_FILE_RANGE
852	int res;	932	int res;
853		933
…		…
866	if (!res || errno != ENOSYS)	946	if (!res || errno != ENOSYS)
867	return res;	947	return res;
868	#endif	948	#endif
869		949
870	/* even though we could play tricks with the flags, it's better to always	950	/* 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 */	951	* call fdatasync, as that matches the expectation of its users best */
872	return fdatasync (fd);	952	return fdatasync (fd);
873	}	953	}
874		954
875	#if !HAVE_READAHEAD	955	#if !HAVE_READAHEAD
876	# undef readahead	956	# undef readahead
…		…
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	{
909	#if HAVE_SENDFILE	992	#if HAVE_SENDFILE
910	# if __linux	993	# if __linux
		994	off_t soffset = offset;
911	res = sendfile (ofd, ifd, &offset, count);	995	res = sendfile (ofd, ifd, &soffset, count);
912		996
913	# elif __freebsd	997	# elif __FreeBSD__
914	/*	998	/*
915	* Of course, the freebsd sendfile is a dire hack with no thoughts	999	* Of course, the freebsd sendfile is a dire hack with no thoughts
916	* wasted on making it similar to other I/O functions.	1000	* wasted on making it similar to other I/O functions.
917	*/	1001	*/
918	{
919	off_t sbytes;	1002	off_t sbytes;
920	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1003	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
921		1004
922	if (res < 0 && sbytes)	1005	#if 0 /* according to the manpage, this is correct, but broken behaviour */
923	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1006	/* freebsd' sendfile will return 0 on success */
		1007	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1008	/* not on e.g. EIO or EPIPE - sounds broken */
		1009	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
924	res = sbytes;	1010	res = sbytes;
925	}	1011	#endif
		1012
		1013	/* according to source inspection, this is correct, and useful behaviour */
		1014	if (sbytes)
		1015	res = sbytes;
		1016
		1017	# elif defined (__APPLE__) && 0 /* broken, as everything on os x */
		1018	off_t sbytes = count;
		1019	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1020
		1021	/* according to the manpage, sbytes is always valid */
		1022	if (sbytes)
		1023	res = sbytes;
926		1024
927	# elif __hpux	1025	# elif __hpux
928	res = sendfile (ofd, ifd, offset, count, 0, 0);	1026	res = sendfile (ofd, ifd, offset, count, 0, 0);
929		1027
930	# elif __solaris	1028	# elif __solaris
931	{
932	struct sendfilevec vec;	1029	struct sendfilevec vec;
933	size_t sbytes;	1030	size_t sbytes;
934		1031
935	vec.sfv_fd = ifd;	1032	vec.sfv_fd = ifd;
936	vec.sfv_flag = 0;	1033	vec.sfv_flag = 0;
937	vec.sfv_off = offset;	1034	vec.sfv_off = offset;
938	vec.sfv_len = count;	1035	vec.sfv_len = count;
939		1036
940	res = sendfilev (ofd, &vec, 1, &sbytes);	1037	res = sendfilev (ofd, &vec, 1, &sbytes);
941		1038
942	if (res < 0 && sbytes)	1039	if (res < 0 && sbytes)
943	res = sbytes;	1040	res = sbytes;
944	}
945		1041
946	# endif	1042	# endif
		1043
		1044	#elif defined (_WIN32)
		1045	/* does not work, just for documentation of what would need to be done */
		1046	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1047	/* libeio guarantees that the file offset does not change, and windows */
		1048	/* has no way to get an independent handle to the same file description */
		1049	HANDLE h = TO_SOCKET (ifd);
		1050	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1051	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1052
947	#else	1053	#else
948	res = -1;	1054	res = -1;
949	errno = ENOSYS;	1055	errno = ENOSYS;
950	#endif	1056	#endif
951		1057
		1058	/* we assume sendfile can copy at least 128mb in one go */
		1059	if (res <= 128 * 1024 * 1024)
		1060	{
		1061	if (res > 0)
		1062	written += res;
		1063
		1064	if (written)
		1065	return written;
		1066
		1067	break;
		1068	}
		1069	else
		1070	{
		1071	/* if we requested more, then probably the kernel was lazy */
		1072	written += res;
		1073	offset += res;
		1074	count -= res;
		1075
		1076	if (!count)
		1077	return written;
		1078	}
		1079	}
		1080
952	if (res < 0	1081	if (res < 0
953	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1082	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1083	/* BSDs */
		1084	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1085	|| errno == ENOTSUP
		1086	#endif
		1087	|| errno == EOPNOTSUPP /* BSDs */
954	#if __solaris	1088	#if __solaris
955	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1089	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
956	#endif	1090	#endif
957	)	1091	)
958	)	1092	)
…		…
989	}	1123	}
990		1124
991	return res;	1125	return res;
992	}	1126	}
993		1127
		1128	#ifdef PAGESIZE
		1129	# define eio_pagesize() PAGESIZE
		1130	#else
		1131	static intptr_t
		1132	eio_pagesize (void)
		1133	{
		1134	static intptr_t page;
		1135
		1136	if (!page)
		1137	page = sysconf (_SC_PAGESIZE);
		1138
		1139	return page;
		1140	}
		1141	#endif
		1142
		1143	static void
		1144	eio_page_align (void **addr, size_t *length)
		1145	{
		1146	intptr_t mask = eio_pagesize () - 1;
		1147
		1148	/* round down addr */
		1149	intptr_t adj = mask & (intptr_t)*addr;
		1150
		1151	*addr = (void *)((intptr_t)*addr - adj);
		1152	*length += adj;
		1153
		1154	/* round up length */
		1155	*length = (*length + mask) & ~mask;
		1156	}
		1157
		1158	#if !_POSIX_MEMLOCK
		1159	# define eio__mlockall(a) ((errno = ENOSYS), -1)
		1160	#else
		1161
994	static int	1162	static int
		1163	eio__mlockall (int flags)
		1164	{
		1165	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1166	extern int mallopt (int, int);
		1167	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1168	#endif
		1169
		1170	if (EIO_MCL_CURRENT != MCL_CURRENT
		1171	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1172	{
		1173	flags = 0
		1174	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1175	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1176	}
		1177
		1178	return mlockall (flags);
		1179	}
		1180	#endif
		1181
		1182	#if !_POSIX_MEMLOCK_RANGE
		1183	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
		1184	#else
		1185
		1186	static int
		1187	eio__mlock (void *addr, size_t length)
		1188	{
		1189	eio_page_align (&addr, &length);
		1190
		1191	return mlock (addr, length);
		1192	}
		1193
		1194	#endif
		1195
		1196	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1197	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
		1198	#else
		1199
		1200	static int
		1201	eio__msync (void *mem, size_t len, int flags)
		1202	{
		1203	eio_page_align (&mem, &len);
		1204
		1205	if (EIO_MS_ASYNC != MS_SYNC
		1206	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1207	|| EIO_MS_SYNC != MS_SYNC)
		1208	{
		1209	flags = 0
		1210	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1211	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1212	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1213	}
		1214
		1215	return msync (mem, len, flags);
		1216	}
		1217
		1218	#endif
		1219
		1220	static int
		1221	eio__mtouch (eio_req *req)
		1222	{
		1223	void *mem = req->ptr2;
		1224	size_t len = req->size;
		1225	int flags = req->int1;
		1226
		1227	eio_page_align (&mem, &len);
		1228
		1229	{
		1230	intptr_t addr = (intptr_t)mem;
		1231	intptr_t end = addr + len;
		1232	intptr_t page = eio_pagesize ();
		1233
		1234	if (addr < end)
		1235	if (flags & EIO_MT_MODIFY) /* modify */
		1236	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1237	else
		1238	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1239	}
		1240
		1241	return 0;
		1242	}
		1243
		1244	/*****************************************************************************/
		1245	/* requests implemented outside eio_execute, because they are so large */
		1246
		1247	static void
		1248	eio__realpath (eio_req *req, etp_worker *self)
		1249	{
		1250	char *rel = req->ptr1;
		1251	char *res;
		1252	char *tmp1, *tmp2;
		1253	#if SYMLOOP_MAX > 32
		1254	int symlinks = SYMLOOP_MAX;
		1255	#else
		1256	int symlinks = 32;
		1257	#endif
		1258
		1259	req->result = -1;
		1260
		1261	errno = EINVAL;
		1262	if (!rel)
		1263	return;
		1264
		1265	errno = ENOENT;
		1266	if (!*rel)
		1267	return;
		1268
		1269	if (!req->ptr2)
		1270	{
		1271	X_LOCK (wrklock);
		1272	req->flags |= EIO_FLAG_PTR2_FREE;
		1273	X_UNLOCK (wrklock);
		1274	req->ptr2 = malloc (PATH_MAX * 3);
		1275
		1276	errno = ENOMEM;
		1277	if (!req->ptr2)
		1278	return;
		1279	}
		1280
		1281	res = req->ptr2;
		1282	tmp1 = res + PATH_MAX;
		1283	tmp2 = tmp1 + PATH_MAX;
		1284
		1285	#if 0 /* disabled, the musl way to do things is just too racy */
		1286	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1287	/* on linux we may be able to ask the kernel */
		1288	{
		1289	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1290
		1291	if (fd >= 0)
		1292	{
		1293	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1294	req->result = readlink (tmp1, res, PATH_MAX);
		1295	close (fd);
		1296
		1297	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1298
		1299	if (req->result > 0)
		1300	goto done;
		1301	}
		1302	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1303	return;
		1304	}
		1305	#endif
		1306	#endif
		1307
		1308	if (*rel != '/')
		1309	{
		1310	if (!getcwd (res, PATH_MAX))
		1311	return;
		1312
		1313	if (res [1]) /* only use if not / */
		1314	res += strlen (res);
		1315	}
		1316
		1317	while (*rel)
		1318	{
		1319	ssize_t len, linklen;
		1320	char *beg = rel;
		1321
		1322	while (*rel && *rel != '/')
		1323	++rel;
		1324
		1325	len = rel - beg;
		1326
		1327	if (!len) /* skip slashes */
		1328	{
		1329	++rel;
		1330	continue;
		1331	}
		1332
		1333	if (beg [0] == '.')
		1334	{
		1335	if (len == 1)
		1336	continue; /* . - nop */
		1337
		1338	if (beg [1] == '.' && len == 2)
		1339	{
		1340	/* .. - back up one component, if possible */
		1341
		1342	while (res != req->ptr2)
		1343	if (*--res == '/')
		1344	break;
		1345
		1346	continue;
		1347	}
		1348	}
		1349
		1350	errno = ENAMETOOLONG;
		1351	if (res + 1 + len + 1 >= tmp1)
		1352	return;
		1353
		1354	/* copy one component */
		1355	*res = '/';
		1356	memcpy (res + 1, beg, len);
		1357
		1358	/* zero-terminate, for readlink */
		1359	res [len + 1] = 0;
		1360
		1361	/* now check if it's a symlink */
		1362	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1363
		1364	if (linklen < 0)
		1365	{
		1366	if (errno != EINVAL)
		1367	return;
		1368
		1369	/* it's a normal directory. hopefully */
		1370	res += len + 1;
		1371	}
		1372	else
		1373	{
		1374	/* yay, it was a symlink - build new path in tmp2 */
		1375	int rellen = strlen (rel);
		1376
		1377	errno = ENAMETOOLONG;
		1378	if (linklen + 1 + rellen >= PATH_MAX)
		1379	return;
		1380
		1381	errno = ELOOP;
		1382	if (!--symlinks)
		1383	return;
		1384
		1385	if (*tmp1 == '/')
		1386	res = req->ptr2; /* symlink resolves to an absolute path */
		1387
		1388	/* we need to be careful, as rel might point into tmp2 already */
		1389	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1390	tmp2 [linklen] = '/';
		1391	memcpy (tmp2, tmp1, linklen);
		1392
		1393	rel = tmp2;
		1394	}
		1395	}
		1396
		1397	/* special case for the lone root path */
		1398	if (res == req->ptr2)
		1399	*res++ = '/';
		1400
		1401	req->result = res - (char *)req->ptr2;
		1402
		1403	done:
		1404	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1405	}
		1406
		1407	static signed char
995	eio_dent_cmp (const void *a_, const void *b_)	1408	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
996	{	1409	{
997	const eio_dirent *a = (const eio_dirent *)a_;	1410	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_;	1411	: a->inode < b->inode ? -1
		1412	: a->inode > b->inode ? 1
		1413	: 0;
		1414	}
999		1415
1000	return (int)b->score - (int)a->score ? (int)b->score - (int)a->score	1416	#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 */	1417
		1418	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1419	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1420
		1421	static void
		1422	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1423	{
		1424	unsigned char bits [9 + sizeof (ino_t) * 8];
		1425	unsigned char *bit = bits;
		1426
		1427	assert (CHAR_BIT == 8);
		1428	assert (sizeof (eio_dirent) * 8 < 256);
		1429	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1430	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1431
		1432	if (size <= EIO_SORT_FAST)
		1433	return;
		1434
		1435	/* first prepare an array of bits to test in our radix sort */
		1436	/* try to take endianness into account, as well as differences in ino_t sizes */
		1437	/* inode_bits must contain all inodes ORed together */
		1438	/* which is used to skip bits that are 0 everywhere, which is very common */
		1439	{
		1440	ino_t endianness;
		1441	int i, j;
		1442
		1443	/* we store the byte offset of byte n into byte n of "endianness" */
		1444	for (i = 0; i < sizeof (ino_t); ++i)
		1445	((unsigned char *)&endianness)[i] = i;
		1446
		1447	*bit++ = 0;
		1448
		1449	for (i = 0; i < sizeof (ino_t); ++i)
		1450	{
		1451	/* shifting off the byte offsets out of "endianness" */
		1452	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1453	endianness >>= 8;
		1454
		1455	for (j = 0; j < 8; ++j)
		1456	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1457	*bit++ = offs + j;
		1458	}
		1459
		1460	for (j = 0; j < 8; ++j)
		1461	if (score_bits & (1 << j))
		1462	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1463	}
		1464
		1465	/* now actually do the sorting (a variant of MSD radix sort) */
		1466	{
		1467	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1468	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1469	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1470	int stk_idx = 0;
		1471
		1472	base_stk [stk_idx] = dents;
		1473	end_stk [stk_idx] = dents + size;
		1474	bit_stk [stk_idx] = bit - 1;
		1475
		1476	do
		1477	{
		1478	base = base_stk [stk_idx];
		1479	end = end_stk [stk_idx];
		1480	bit = bit_stk [stk_idx];
		1481
		1482	for (;;)
		1483	{
		1484	unsigned char O = *bit >> 3;
		1485	unsigned char M = 1 << (*bit & 7);
		1486
		1487	eio_dirent *a = base;
		1488	eio_dirent *b = end;
		1489
		1490	if (b - a < EIO_SORT_CUTOFF)
		1491	break;
		1492
		1493	/* now bit-partition the array on the bit */
		1494	/* this ugly asymmetric loop seems to perform much better than typical */
		1495	/* partition algos found in the literature */
		1496	do
		1497	if (!(((unsigned char *)a)[O] & M))
		1498	++a;
		1499	else if (!(((unsigned char *)--b)[O] & M))
		1500	{
		1501	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1502	++a;
		1503	}
		1504	while (b > a);
		1505
		1506	/* next bit, or stop, if no bits left in this path */
		1507	if (!*--bit)
		1508	break;
		1509
		1510	base_stk [stk_idx] = a;
		1511	end_stk [stk_idx] = end;
		1512	bit_stk [stk_idx] = bit;
		1513	++stk_idx;
		1514
		1515	end = a;
		1516	}
		1517	}
		1518	while (stk_idx--);
		1519	}
		1520	}
		1521
		1522	static void
		1523	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1524	{
		1525	/* first move the smallest element to the front, to act as a sentinel */
		1526	{
		1527	int i;
		1528	eio_dirent *min = dents;
		1529
		1530	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1531	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1532	if (EIO_DENT_CMP (dents [i], <, *min))
		1533	min = &dents [i];
		1534
		1535	/* swap elements 0 and j (minimum) */
		1536	{
		1537	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1538	}
		1539	}
		1540
		1541	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1542	{
		1543	eio_dirent *i, *j;
		1544
		1545	for (i = dents + 1; i < dents + size; ++i)
		1546	{
		1547	eio_dirent value = *i;
		1548
		1549	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1550	j [1] = j [0];
		1551
		1552	j [1] = value;
		1553	}
		1554	}
		1555	}
		1556
		1557	static void
		1558	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1559	{
		1560	if (size <= 1)
		1561	return; /* our insertion sort relies on size > 0 */
		1562
		1563	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1564	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1565	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1566
		1567	/* use an insertion sort at the end, or for small arrays, */
		1568	/* as insertion sort is more efficient for small partitions */
		1569	eio_dent_insertion_sort (dents, size);
1002	}	1570	}
1003		1571
1004	/* read a full directory */	1572	/* read a full directory */
1005	static void	1573	static void
1006	eio__scandir (eio_req *req, etp_worker *self)	1574	eio__scandir (eio_req *req, etp_worker *self)
1007	{	1575	{
1008	DIR *dirp;	1576	DIR *dirp;
1009	EIO_STRUCT_DIRENT *entp;	1577	EIO_STRUCT_DIRENT *entp;
1010	unsigned char *name, *names;	1578	char *name, *names;
1011	int namesalloc = 4096;	1579	int namesalloc = 4096;
1012	int namesoffs = 0;	1580	int namesoffs = 0;
1013	int flags = req->int1;	1581	int flags = req->int1;
1014	eio_dirent *dents = 0;	1582	eio_dirent *dents = 0;
1015	int dentalloc = 128;	1583	int dentalloc = 128;
1016	int dentoffs = 0;	1584	int dentoffs = 0;
		1585	ino_t inode_bits = 0;
1017		1586
1018	req->result = -1;	1587	req->result = -1;
1019		1588
1020	if (!(flags & EIO_READDIR_DENTS))	1589	if (!(flags & EIO_READDIR_DENTS))
1021	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1590	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1022		1591
1023	X_LOCK (wrklock);	1592	X_LOCK (wrklock);
1024	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1593	/* the corresponding closedir is in ETP_WORKER_CLEAR */
1025	self->dirp = dirp = opendir (req->ptr1);	1594	self->dirp = dirp = opendir (req->ptr1);
		1595
		1596	if (req->flags & EIO_FLAG_PTR1_FREE)
		1597	free (req->ptr1);
		1598
1026	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1599	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;	1600	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
		1601	req->ptr2 = names = malloc (namesalloc);
1029	X_UNLOCK (wrklock);	1602	X_UNLOCK (wrklock);
1030		1603
1031	if (dirp && names && (!flags || dents))	1604	if (dirp && names && (!flags || dents))
1032	for (;;)	1605	for (;;)
1033	{	1606	{
…		…
1041		1614
1042	/* sort etc. */	1615	/* sort etc. */
1043	req->int1 = flags;	1616	req->int1 = flags;
1044	req->result = dentoffs;	1617	req->result = dentoffs;
1045		1618
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	1619	if (flags & EIO_READDIR_STAT_ORDER)
1055	|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))	1620	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)	1621	else if (flags & EIO_READDIR_DIRS_FIRST)
		1622	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1623	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1624	else
1061	{	1625	{
1062	/* in this case, all is known, and we just put dirs first and sort them */	1626	/* in this case, all is known, and we just put dirs first and sort them */
1063	eio_dirent *ent = dents + dentoffs;	1627	eio_dirent *oth = dents + dentoffs;
1064	eio_dirent *dir = dents;	1628	eio_dirent *dir = dents;
1065		1629
		1630	/* now partition dirs to the front, and non-dirs to the back */
		1631	/* by walking from both sides and swapping if necessary */
1066	while (ent > dir)	1632	while (oth > dir)
1067	{	1633	{
1068	if (dir->type == DT_DIR)	1634	if (dir->type == EIO_DT_DIR)
1069	++dir;	1635	++dir;
1070	else
1071	{
1072	--ent;
1073
1074	if (ent->type == DT_DIR)	1636	else if ((--oth)->type == EIO_DT_DIR)
1075	{	1637	{
1076	eio_dirent tmp = *dir;	1638	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1077	*dir = *ent;
1078	*ent = tmp;
1079		1639
1080	++dir;	1640	++dir;
1081	}	1641	}
1082	}	1642	}
		1643
		1644	/* now sort the dirs only (dirs all have the same score) */
		1645	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1083	}	1646	}
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		1647
1099	break;	1648	break;
1100	}	1649	}
1101		1650
1102	/* now add the entry to our list(s) */	1651	/* now add the entry to our list(s) */
…		…
1105	/* skip . and .. entries */	1654	/* skip . and .. entries */
1106	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1655	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1107	{	1656	{
1108	int len = D_NAMLEN (entp) + 1;	1657	int len = D_NAMLEN (entp) + 1;
1109		1658
1110	while (expect_false (namesoffs + len > namesalloc))	1659	while (ecb_expect_false (namesoffs + len > namesalloc))
1111	{	1660	{
1112	namesalloc *= 2;	1661	namesalloc *= 2;
1113	X_LOCK (wrklock);	1662	X_LOCK (wrklock);
1114	req->ptr1 = names = realloc (names, namesalloc);	1663	req->ptr2 = names = realloc (names, namesalloc);
1115	X_UNLOCK (wrklock);	1664	X_UNLOCK (wrklock);
1116		1665
1117	if (!names)	1666	if (!names)
1118	break;	1667	break;
1119	}	1668	}
…		…
1122		1671
1123	if (dents)	1672	if (dents)
1124	{	1673	{
1125	struct eio_dirent *ent;	1674	struct eio_dirent *ent;
1126		1675
1127	if (expect_false (dentoffs == dentalloc))	1676	if (ecb_expect_false (dentoffs == dentalloc))
1128	{	1677	{
1129	dentalloc *= 2;	1678	dentalloc *= 2;
1130	X_LOCK (wrklock);	1679	X_LOCK (wrklock);
1131	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1680	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1132	X_UNLOCK (wrklock);	1681	X_UNLOCK (wrklock);
1133		1682
1134	if (!dents)	1683	if (!dents)
1135	break;	1684	break;
1136	}	1685	}
1137		1686
1138	ent = dents + dentoffs;	1687	ent = dents + dentoffs;
1139		1688
1140	ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */	1689	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1141	ent->namelen = len - 1;	1690	ent->namelen = len - 1;
1142	ent->inode = D_INO (entp);	1691	ent->inode = D_INO (entp);
		1692
		1693	inode_bits |= ent->inode;
1143		1694
1144	switch (D_TYPE (entp))	1695	switch (D_TYPE (entp))
1145	{	1696	{
1146	default:	1697	default:
1147	ent->type = EIO_DT_UNKNOWN;	1698	ent->type = EIO_DT_UNKNOWN;
…		…
1190	#ifdef DT_WHT	1741	#ifdef DT_WHT
1191	case DT_WHT: ent->type = EIO_DT_WHT; break;	1742	case DT_WHT: ent->type = EIO_DT_WHT; break;
1192	#endif	1743	#endif
1193	}	1744	}
1194		1745
1195	ent->score = 0;	1746	ent->score = 7;
1196		1747
1197	if (flags & EIO_READDIR_DIRS_FIRST)	1748	if (flags & EIO_READDIR_DIRS_FIRST)
1198	{	1749	{
1199	if (ent->type == EIO_DT_UNKNOWN)	1750	if (ent->type == EIO_DT_UNKNOWN)
1200	{	1751	{
1201	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */	1752	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1202	ent->score = 98;	1753	ent->score = 1;
1203	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */	1754	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 */	1755	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1205	}	1756	}
1206	else if (ent->type == EIO_DT_DIR)	1757	else if (ent->type == EIO_DT_DIR)
1207	ent->score = 100;	1758	ent->score = 0;
1208	}	1759	}
1209	}	1760	}
1210		1761
1211	namesoffs += len;	1762	namesoffs += len;
1212	++dentoffs;	1763	++dentoffs;
1213	}	1764	}
		1765
		1766	if (EIO_CANCELLED (req))
		1767	{
		1768	errno = ECANCELED;
		1769	break;
		1770	}
1214	}	1771	}
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	}	1772	}
1248		1773
1249	/*****************************************************************************/	1774	/*****************************************************************************/
1250		1775
1251	#define ALLOC(len) \	1776	#define ALLOC(len) \
…		…
1283	if (req)	1808	if (req)
1284	break;	1809	break;
1285		1810
1286	++idle;	1811	++idle;
1287		1812
1288	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1813	ts.tv_sec = time (0) + idle_timeout;
1289	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1814	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1290	{	1815	{
1291	if (idle > max_idle)	1816	if (idle > max_idle)
1292	{	1817	{
1293	--idle;	1818	--idle;
…		…
1310	X_UNLOCK (reqlock);	1835	X_UNLOCK (reqlock);
1311		1836
1312	if (req->type < 0)	1837	if (req->type < 0)
1313	goto quit;	1838	goto quit;
1314		1839
1315	if (!EIO_CANCELLED (req))
1316	ETP_EXECUTE (self, req);	1840	ETP_EXECUTE (self, req);
1317		1841
1318	X_LOCK (reslock);	1842	X_LOCK (reslock);
1319		1843
1320	++npending;	1844	++npending;
1321		1845
…		…
1336	return 0;	1860	return 0;
1337	}	1861	}
1338		1862
1339	/*****************************************************************************/	1863	/*****************************************************************************/
1340		1864
		1865	int ecb_cold
1341	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1866	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1342	{	1867	{
1343	return etp_init (want_poll, done_poll);	1868	return etp_init (want_poll, done_poll);
1344	}	1869	}
1345		1870
		1871	ecb_inline void
1346	static void eio_api_destroy (eio_req *req)	1872	eio_api_destroy (eio_req *req)
1347	{	1873	{
1348	free (req);	1874	free (req);
1349	}	1875	}
1350		1876
1351	#define REQ(rtype) \	1877	#define REQ(rtype) \
…		…
1370	{ \	1896	{ \
1371	eio_api_destroy (req); \	1897	eio_api_destroy (req); \
1372	return 0; \	1898	return 0; \
1373	}	1899	}
1374		1900
		1901	static void
1375	static void eio_execute (etp_worker *self, eio_req *req)	1902	eio_execute (etp_worker *self, eio_req *req)
1376	{	1903	{
1377	errno = 0;	1904	if (ecb_expect_false (EIO_CANCELLED (req)))
		1905	{
		1906	req->result = -1;
		1907	req->errorno = ECANCELED;
		1908	return;
		1909	}
1378		1910
1379	switch (req->type)	1911	switch (req->type)
1380	{	1912	{
1381	case EIO_READ: ALLOC (req->size);	1913	case EIO_READ: ALLOC (req->size);
1382	req->result = req->offs >= 0	1914	req->result = req->offs >= 0
…		…
1394	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1926	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1395	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1927	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1396	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1928	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1397	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1929	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1398		1930
		1931	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1932	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1933	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1934	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1935
1399	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	1936	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;	1937	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;	1938	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;	1939	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;	1940	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
…		…
1410	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	1947	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1411	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	1948	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;	1949	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1413	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	1950	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1414	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	1951	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;	1952	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1416		1953
		1954	case EIO_REALPATH: eio__realpath (req, self); break;
		1955
1417	case EIO_READLINK: ALLOC (NAME_MAX);	1956	case EIO_READLINK: ALLOC (PATH_MAX);
1418	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	1957	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1419		1958
1420	case EIO_SYNC: req->result = 0; sync (); break;	1959	case EIO_SYNC: req->result = 0; sync (); break;
1421	case EIO_FSYNC: req->result = fsync (req->int1); break;	1960	case EIO_FSYNC: req->result = fsync (req->int1); break;
1422	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	1961	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1423	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;	1962	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1963	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1424	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	1964	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		1965	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;	1966	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1426		1967
1427	case EIO_READDIR: eio__scandir (req, self); break;	1968	case EIO_READDIR: eio__scandir (req, self); break;
1428		1969
1429	case EIO_BUSY:	1970	case EIO_BUSY:
1430	#ifdef _WIN32	1971	#ifdef _WIN32
1431	Sleep (req->nv1 * 1000.);	1972	Sleep (req->nv1 * 1e3);
1432	#else	1973	#else
1433	{	1974	{
1434	struct timeval tv;	1975	struct timeval tv;
1435		1976
1436	tv.tv_sec = req->nv1;	1977	tv.tv_sec = req->nv1;
1437	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	1978	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1438		1979
1439	req->result = select (0, 0, 0, 0, &tv);	1980	req->result = select (0, 0, 0, 0, &tv);
1440	}	1981	}
1441	#endif	1982	#endif
1442	break;	1983	break;
…		…
1457	times = tv;	1998	times = tv;
1458	}	1999	}
1459	else	2000	else
1460	times = 0;	2001	times = 0;
1461		2002
1462
1463	req->result = req->type == EIO_FUTIME	2003	req->result = req->type == EIO_FUTIME
1464	? futimes (req->int1, times)	2004	? futimes (req->int1, times)
1465	: utimes (req->ptr1, times);	2005	: utimes (req->ptr1, times);
1466	}	2006	}
1467	break;	2007	break;
…		…
1472	case EIO_NOP:	2012	case EIO_NOP:
1473	req->result = 0;	2013	req->result = 0;
1474	break;	2014	break;
1475		2015
1476	case EIO_CUSTOM:	2016	case EIO_CUSTOM:
1477	((void (*)(eio_req *))req->feed) (req);	2017	req->feed (req);
1478	break;	2018	break;
1479		2019
1480	default:	2020	default:
		2021	errno = ENOSYS;
1481	req->result = -1;	2022	req->result = -1;
1482	break;	2023	break;
1483	}	2024	}
1484		2025
1485	req->errorno = errno;	2026	req->errorno = errno;
…		…
1515	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2056	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1516	{	2057	{
1517	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2058	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1518	}	2059	}
1519		2060
		2061	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2062	{
		2063	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2064	}
		2065
		2066	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2067	{
		2068	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2069	}
		2070
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)	2071	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	{	2072	{
1522	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2073	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1523	}	2074	}
1524		2075
…		…
1548	}	2099	}
1549		2100
1550	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2101	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1551	{	2102	{
1552	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2103	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		2104	}
		2105
		2106	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2107	{
		2108	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1553	}	2109	}
1554		2110
1555	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2111	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1556	{	2112	{
1557	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2113	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)	2177	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1622	{	2178	{
1623	return eio__1path (EIO_READLINK, path, pri, cb, data);	2179	return eio__1path (EIO_READLINK, path, pri, cb, data);
1624	}	2180	}
1625		2181
		2182	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2183	{
		2184	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2185	}
		2186
1626	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2187	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1627	{	2188	{
1628	return eio__1path (EIO_STAT, path, pri, cb, data);	2189	return eio__1path (EIO_STAT, path, pri, cb, data);
1629	}	2190	}
1630		2191
1631	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2192	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1632	{	2193	{
1633	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2194	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1634	}	2195	}
1635		2196
		2197	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2198	{
		2199	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2200	}
		2201
1636	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2202	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1637	{	2203	{
1638	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2204	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1639	}	2205	}
1640		2206
…		…
1648	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2214	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1649	}	2215	}
1650		2216
1651	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2217	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1652	{	2218	{
1653	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2219	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1654	}	2220	}
1655		2221
1656	static eio_req *	2222	static eio_req *
1657	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2223	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1658	{	2224	{
…		…
1682	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2248	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1683	{	2249	{
1684	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2250	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1685	}	2251	}
1686		2252
1687	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2253	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1688	{	2254	{
1689	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2255	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1690	}	2256	}
1691		2257
1692	#endif	2258	#endif
1693		2259
1694	eio_req *eio_grp (eio_cb cb, void *data)	2260	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1703	#undef SEND	2269	#undef SEND
1704		2270
1705	/*****************************************************************************/	2271	/*****************************************************************************/
1706	/* grp functions */	2272	/* grp functions */
1707		2273
		2274	void
1708	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2275	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1709	{	2276	{
1710	grp->int2 = limit;	2277	grp->int2 = limit;
1711	grp->feed = feed;	2278	grp->feed = feed;
1712		2279
1713	grp_try_feed (grp);	2280	grp_try_feed (grp);
1714	}	2281	}
1715		2282
		2283	void
1716	void eio_grp_limit (eio_req *grp, int limit)	2284	eio_grp_limit (eio_req *grp, int limit)
1717	{	2285	{
1718	grp->int2 = limit;	2286	grp->int2 = limit;
1719		2287
1720	grp_try_feed (grp);	2288	grp_try_feed (grp);
1721	}	2289	}
1722		2290
		2291	void
1723	void eio_grp_add (eio_req *grp, eio_req *req)	2292	eio_grp_add (eio_req *grp, eio_req *req)
1724	{	2293	{
1725	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2294	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1726		2295
1727	grp->flags |= EIO_FLAG_GROUPADD;	2296	grp->flags |= EIO_FLAG_GROUPADD;
1728		2297
…		…
1739	}	2308	}
1740		2309
1741	/*****************************************************************************/	2310	/*****************************************************************************/
1742	/* misc garbage */	2311	/* misc garbage */
1743		2312
		2313	ssize_t
1744	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2314	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1745	{	2315	{
1746	etp_worker wrk;	2316	etp_worker wrk;
		2317	ssize_t ret;
1747		2318
1748	wrk.dbuf = 0;	2319	wrk.dbuf = 0;
1749		2320
1750	eio__sendfile (ofd, ifd, offset, count, &wrk);	2321	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1751		2322
1752	if (wrk.dbuf)	2323	if (wrk.dbuf)
1753	free (wrk.dbuf);	2324	free (wrk.dbuf);
1754	}
1755		2325
		2326	return ret;
		2327	}
		2328

Diff Legend

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