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 1.78 by root, Tue Jul 5 14:38:53 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 form stdint.h, says idiot openbsd coder */
		66	#if HAVE_STDINT_H
		67	# include <stdint.h>
		68	#endif
		69
58	#ifndef EIO_FINISH	70	#ifndef EIO_FINISH
59	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	71	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60	#endif	72	#endif
61		73
62	#ifndef EIO_DESTROY	74	#ifndef EIO_DESTROY
…		…
70	#ifdef _WIN32	82	#ifdef _WIN32
71		83
72	/*doh*/	84	/*doh*/
73	#else	85	#else
74		86
75	# include "config.h"
76	# include <sys/time.h>	87	# include <sys/time.h>
77	# include <sys/select.h>	88	# include <sys/select.h>
78	# include <unistd.h>	89	# include <unistd.h>
79	# include <utime.h>	90	# include <utime.h>
80	# include <signal.h>	91	# include <signal.h>
81	# include <dirent.h>	92	# include <dirent.h>
82		93
		94	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		95	# include <sys/mman.h>
		96	#endif
		97
83	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */	98	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84	# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)	99	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
85	# define _DIRENT_HAVE_D_TYPE /* sigh */	100	# define _DIRENT_HAVE_D_TYPE /* sigh */
86	# define D_INO(de) (de)->d_fileno	101	# define D_INO(de) (de)->d_fileno
87	# define D_NAMLEN(de) (de)->d_namlen	102	# define D_NAMLEN(de) (de)->d_namlen
88	# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600	103	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
89	# define D_INO(de) (de)->d_ino	104	# define D_INO(de) (de)->d_ino
90	# endif	105	# endif
91		106
92	#ifdef _D_EXACT_NAMLEN	107	#ifdef _D_EXACT_NAMLEN
93	# undef D_NAMLEN	108	# undef D_NAMLEN
…		…
105	#endif	120	#endif
106		121
107	#if HAVE_SENDFILE	122	#if HAVE_SENDFILE
108	# if __linux	123	# if __linux
109	# include <sys/sendfile.h>	124	# include <sys/sendfile.h>
110	# elif __freebsd	125	# elif __FreeBSD__ || defined __APPLE__
111	# include <sys/socket.h>	126	# include <sys/socket.h>
112	# include <sys/uio.h>	127	# include <sys/uio.h>
113	# elif __hpux	128	# elif __hpux
114	# include <sys/socket.h>	129	# include <sys/socket.h>
115	# elif __solaris /* not yet */	130	# elif __solaris
116	# include <sys/sendfile.h>	131	# include <sys/sendfile.h>
117	# else	132	# else
118	# error sendfile support requested but not available	133	# error sendfile support requested but not available
119	# endif	134	# endif
120	#endif	135	#endif
…		…
127	#endif	142	#endif
128	#ifndef D_NAMLEN	143	#ifndef D_NAMLEN
129	# define D_NAMLEN(de) strlen ((de)->d_name)	144	# define D_NAMLEN(de) strlen ((de)->d_name)
130	#endif	145	#endif
131		146
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 */	147	/* used for struct dirent, AIX doesn't provide it */
136	#ifndef NAME_MAX	148	#ifndef NAME_MAX
137	# define NAME_MAX 4096	149	# define NAME_MAX 4096
		150	#endif
		151
		152	/* used for readlink etc. */
		153	#ifndef PATH_MAX
		154	# define PATH_MAX 4096
138	#endif	155	#endif
139		156
140	/* buffer size for various temporary buffers */	157	/* buffer size for various temporary buffers */
141	#define EIO_BUFSIZE 65536	158	#define EIO_BUFSIZE 65536
142		159
…		…
148	errno = ENOMEM; \	165	errno = ENOMEM; \
149	if (!eio_buf) \	166	if (!eio_buf) \
150	return -1;	167	return -1;
151		168
152	#define EIO_TICKS ((1000000 + 1023) >> 10)	169	#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		170
167	#define ETP_PRI_MIN EIO_PRI_MIN	171	#define ETP_PRI_MIN EIO_PRI_MIN
168	#define ETP_PRI_MAX EIO_PRI_MAX	172	#define ETP_PRI_MAX EIO_PRI_MAX
169		173
170	struct etp_worker;	174	struct etp_worker;
…		…
195		199
196	/*****************************************************************************/	200	/*****************************************************************************/
197		201
198	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	202	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
199		203
200	/* calculcate time difference in ~1/EIO_TICKS of a second */	204	/* calculate time difference in ~1/EIO_TICKS of a second */
		205	ecb_inline int
201	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	206	tvdiff (struct timeval *tv1, struct timeval *tv2)
202	{	207	{
203	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	208	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
204	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	209	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
205	}	210	}
206		211
…		…
213	static unsigned int max_poll_reqs; /* reslock */	218	static unsigned int max_poll_reqs; /* reslock */
214		219
215	static volatile unsigned int nreqs; /* reqlock */	220	static volatile unsigned int nreqs; /* reqlock */
216	static volatile unsigned int nready; /* reqlock */	221	static volatile unsigned int nready; /* reqlock */
217	static volatile unsigned int npending; /* reqlock */	222	static volatile unsigned int npending; /* reqlock */
218	static volatile unsigned int max_idle = 4;	223	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		224	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
219		225
220	static mutex_t wrklock = X_MUTEX_INIT;	226	static xmutex_t wrklock;
221	static mutex_t reslock = X_MUTEX_INIT;	227	static xmutex_t reslock;
222	static mutex_t reqlock = X_MUTEX_INIT;	228	static xmutex_t reqlock;
223	static cond_t reqwait = X_COND_INIT;	229	static xcond_t reqwait;
224		230
225	#if !HAVE_PREADWRITE	231	#if !HAVE_PREADWRITE
226	/*	232	/*
227	* make our pread/pwrite emulation safe against themselves, but not against	233	* make our pread/pwrite emulation safe against themselves, but not against
228	* normal read/write by using a mutex. slows down execution a lot,	234	* normal read/write by using a mutex. slows down execution a lot,
229	* but that's your problem, not mine.	235	* but that's your problem, not mine.
230	*/	236	*/
231	static mutex_t preadwritelock = X_MUTEX_INIT;	237	static xmutex_t preadwritelock = X_MUTEX_INIT;
232	#endif	238	#endif
233		239
234	typedef struct etp_worker	240	typedef struct etp_worker
235	{	241	{
236	/* locked by wrklock */	242	/* locked by wrklock */
237	struct etp_worker *prev, *next;	243	struct etp_worker *prev, *next;
238		244
239	thread_t tid;	245	xthread_t tid;
240		246
241	/* locked by reslock, reqlock or wrklock */	247	/* locked by reslock, reqlock or wrklock */
242	ETP_REQ *req; /* currently processed request */	248	ETP_REQ *req; /* currently processed request */
243		249
244	ETP_WORKER_COMMON	250	ETP_WORKER_COMMON
…		…
249	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	255	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
250	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	256	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
251		257
252	/* worker threads management */	258	/* worker threads management */
253		259
		260	static void ecb_cold
254	static void etp_worker_clear (etp_worker *wrk)	261	etp_worker_clear (etp_worker *wrk)
255	{	262	{
256	ETP_WORKER_CLEAR (wrk);	263	ETP_WORKER_CLEAR (wrk);
257	}	264	}
258		265
		266	static void ecb_cold
259	static void etp_worker_free (etp_worker *wrk)	267	etp_worker_free (etp_worker *wrk)
260	{	268	{
261	wrk->next->prev = wrk->prev;	269	wrk->next->prev = wrk->prev;
262	wrk->prev->next = wrk->next;	270	wrk->prev->next = wrk->next;
263		271
264	free (wrk);	272	free (wrk);
265	}	273	}
266		274
267	static unsigned int etp_nreqs (void)	275	static unsigned int
		276	etp_nreqs (void)
268	{	277	{
269	int retval;	278	int retval;
270	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	279	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271	retval = nreqs;	280	retval = nreqs;
272	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	281	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273	return retval;	282	return retval;
274	}	283	}
275		284
276	static unsigned int etp_nready (void)	285	static unsigned int
		286	etp_nready (void)
277	{	287	{
278	unsigned int retval;	288	unsigned int retval;
279		289
280	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	290	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281	retval = nready;	291	retval = nready;
282	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	292	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283		293
284	return retval;	294	return retval;
285	}	295	}
286		296
287	static unsigned int etp_npending (void)	297	static unsigned int
		298	etp_npending (void)
288	{	299	{
289	unsigned int retval;	300	unsigned int retval;
290		301
291	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	302	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
292	retval = npending;	303	retval = npending;
293	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	304	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
294		305
295	return retval;	306	return retval;
296	}	307	}
297		308
298	static unsigned int etp_nthreads (void)	309	static unsigned int
		310	etp_nthreads (void)
299	{	311	{
300	unsigned int retval;	312	unsigned int retval;
301		313
302	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	314	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303	retval = started;	315	retval = started;
…		…
317	} etp_reqq;	329	} etp_reqq;
318		330
319	static etp_reqq req_queue;	331	static etp_reqq req_queue;
320	static etp_reqq res_queue;	332	static etp_reqq res_queue;
321		333
		334	static int ecb_noinline
322	static int reqq_push (etp_reqq *q, ETP_REQ *req)	335	reqq_push (etp_reqq *q, ETP_REQ *req)
323	{	336	{
324	int pri = req->pri;	337	int pri = req->pri;
325	req->next = 0;	338	req->next = 0;
326		339
327	if (q->qe[pri])	340	if (q->qe[pri])
…		…
333	q->qe[pri] = q->qs[pri] = req;	346	q->qe[pri] = q->qs[pri] = req;
334		347
335	return q->size++;	348	return q->size++;
336	}	349	}
337		350
		351	static ETP_REQ * ecb_noinline
338	static ETP_REQ *reqq_shift (etp_reqq *q)	352	reqq_shift (etp_reqq *q)
339	{	353	{
340	int pri;	354	int pri;
341		355
342	if (!q->size)	356	if (!q->size)
343	return 0;	357	return 0;
…		…
358	}	372	}
359		373
360	abort ();	374	abort ();
361	}	375	}
362		376
		377	static void ecb_cold
		378	etp_thread_init (void)
		379	{
		380	X_MUTEX_CREATE (wrklock);
		381	X_MUTEX_CREATE (reslock);
		382	X_MUTEX_CREATE (reqlock);
		383	X_COND_CREATE (reqwait);
		384	}
		385
		386	static void ecb_cold
363	static void etp_atfork_prepare (void)	387	etp_atfork_prepare (void)
364	{	388	{
365	X_LOCK (wrklock);	389	X_LOCK (wrklock);
366	X_LOCK (reqlock);	390	X_LOCK (reqlock);
367	X_LOCK (reslock);	391	X_LOCK (reslock);
368	#if !HAVE_PREADWRITE	392	#if !HAVE_PREADWRITE
369	X_LOCK (preadwritelock);	393	X_LOCK (preadwritelock);
370	#endif	394	#endif
371	}	395	}
372		396
		397	static void ecb_cold
373	static void etp_atfork_parent (void)	398	etp_atfork_parent (void)
374	{	399	{
375	#if !HAVE_PREADWRITE	400	#if !HAVE_PREADWRITE
376	X_UNLOCK (preadwritelock);	401	X_UNLOCK (preadwritelock);
377	#endif	402	#endif
378	X_UNLOCK (reslock);	403	X_UNLOCK (reslock);
379	X_UNLOCK (reqlock);	404	X_UNLOCK (reqlock);
380	X_UNLOCK (wrklock);	405	X_UNLOCK (wrklock);
381	}	406	}
382		407
		408	static void ecb_cold
383	static void etp_atfork_child (void)	409	etp_atfork_child (void)
384	{	410	{
385	ETP_REQ *prv;	411	ETP_REQ *prv;
386		412
387	while ((prv = reqq_shift (&req_queue)))	413	while ((prv = reqq_shift (&req_queue)))
388	ETP_DESTROY (prv);	414	ETP_DESTROY (prv);
…		…
405	idle = 0;	431	idle = 0;
406	nreqs = 0;	432	nreqs = 0;
407	nready = 0;	433	nready = 0;
408	npending = 0;	434	npending = 0;
409		435
410	etp_atfork_parent ();	436	etp_thread_init ();
411	}	437	}
412		438
413	static void	439	static void ecb_cold
414	etp_once_init (void)	440	etp_once_init (void)
415	{	441	{
		442	etp_thread_init ();
416	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);	443	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
417	}	444	}
418		445
419	static int	446	static int ecb_cold
420	etp_init (void (*want_poll)(void), void (*done_poll)(void))	447	etp_init (void (*want_poll)(void), void (*done_poll)(void))
421	{	448	{
422	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	449	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
423		450
424	pthread_once (&doinit, etp_once_init);	451	pthread_once (&doinit, etp_once_init);
…		…
429	return 0;	456	return 0;
430	}	457	}
431		458
432	X_THREAD_PROC (etp_proc);	459	X_THREAD_PROC (etp_proc);
433		460
		461	static void ecb_cold
434	static void etp_start_thread (void)	462	etp_start_thread (void)
435	{	463	{
436	etp_worker *wrk = calloc (1, sizeof (etp_worker));	464	etp_worker *wrk = calloc (1, sizeof (etp_worker));
437		465
438	/*TODO*/	466	/*TODO*/
439	assert (("unable to allocate worker thread data", wrk));	467	assert (("unable to allocate worker thread data", wrk));
…		…
452	free (wrk);	480	free (wrk);
453		481
454	X_UNLOCK (wrklock);	482	X_UNLOCK (wrklock);
455	}	483	}
456		484
		485	static void
457	static void etp_maybe_start_thread (void)	486	etp_maybe_start_thread (void)
458	{	487	{
459	if (expect_true (etp_nthreads () >= wanted))	488	if (ecb_expect_true (etp_nthreads () >= wanted))
460	return;	489	return;
461		490
462	/* todo: maybe use idle here, but might be less exact */	491	/* todo: maybe use idle here, but might be less exact */
463	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	492	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
464	return;	493	return;
465		494
466	etp_start_thread ();	495	etp_start_thread ();
467	}	496	}
468		497
		498	static void ecb_cold
469	static void etp_end_thread (void)	499	etp_end_thread (void)
470	{	500	{
471	eio_req *req = calloc (1, sizeof (eio_req));	501	eio_req *req = calloc (1, sizeof (eio_req));
472		502
473	req->type = -1;	503	req->type = -1;
474	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	504	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
481	X_LOCK (wrklock);	511	X_LOCK (wrklock);
482	--started;	512	--started;
483	X_UNLOCK (wrklock);	513	X_UNLOCK (wrklock);
484	}	514	}
485		515
486	static int etp_poll (void)	516	static int
		517	etp_poll (void)
487	{	518	{
488	unsigned int maxreqs;	519	unsigned int maxreqs;
489	unsigned int maxtime;	520	unsigned int maxtime;
490	struct timeval tv_start, tv_now;	521	struct timeval tv_start, tv_now;
491		522
…		…
521		552
522	X_LOCK (reqlock);	553	X_LOCK (reqlock);
523	--nreqs;	554	--nreqs;
524	X_UNLOCK (reqlock);	555	X_UNLOCK (reqlock);
525		556
526	if (expect_false (req->type == EIO_GROUP && req->size))	557	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
527	{	558	{
528	req->int1 = 1; /* mark request as delayed */	559	req->int1 = 1; /* mark request as delayed */
529	continue;	560	continue;
530	}	561	}
531	else	562	else
532	{	563	{
533	int res = ETP_FINISH (req);	564	int res = ETP_FINISH (req);
534	if (expect_false (res))	565	if (ecb_expect_false (res))
535	return res;	566	return res;
536	}	567	}
537		568
538	if (expect_false (maxreqs && !--maxreqs))	569	if (ecb_expect_false (maxreqs && !--maxreqs))
539	break;	570	break;
540		571
541	if (maxtime)	572	if (maxtime)
542	{	573	{
543	gettimeofday (&tv_now, 0);	574	gettimeofday (&tv_now, 0);
…		…
549		580
550	errno = EAGAIN;	581	errno = EAGAIN;
551	return -1;	582	return -1;
552	}	583	}
553		584
		585	static void
554	static void etp_cancel (ETP_REQ *req)	586	etp_cancel (ETP_REQ *req)
555	{	587	{
556	X_LOCK (wrklock);	588	X_LOCK (wrklock);
557	req->flags |= EIO_FLAG_CANCELLED;	589	req->flags |= EIO_FLAG_CANCELLED;
558	X_UNLOCK (wrklock);	590	X_UNLOCK (wrklock);
559		591
560	eio_grp_cancel (req);	592	eio_grp_cancel (req);
561	}	593	}
562		594
		595	static void
563	static void etp_submit (ETP_REQ *req)	596	etp_submit (ETP_REQ *req)
564	{	597	{
565	req->pri -= ETP_PRI_MIN;	598	req->pri -= ETP_PRI_MIN;
566		599
567	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	600	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;	601	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
569		602
570	if (expect_false (req->type == EIO_GROUP))	603	if (ecb_expect_false (req->type == EIO_GROUP))
571	{	604	{
572	/* I hope this is worth it :/ */	605	/* I hope this is worth it :/ */
573	X_LOCK (reqlock);	606	X_LOCK (reqlock);
574	++nreqs;	607	++nreqs;
575	X_UNLOCK (reqlock);	608	X_UNLOCK (reqlock);
…		…
594		627
595	etp_maybe_start_thread ();	628	etp_maybe_start_thread ();
596	}	629	}
597	}	630	}
598		631
		632	static void ecb_cold
599	static void etp_set_max_poll_time (double nseconds)	633	etp_set_max_poll_time (double nseconds)
600	{	634	{
601	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	635	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
602	max_poll_time = nseconds;	636	max_poll_time = nseconds * EIO_TICKS;
603	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	637	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
604	}	638	}
605		639
		640	static void ecb_cold
606	static void etp_set_max_poll_reqs (unsigned int maxreqs)	641	etp_set_max_poll_reqs (unsigned int maxreqs)
607	{	642	{
608	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	643	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609	max_poll_reqs = maxreqs;	644	max_poll_reqs = maxreqs;
610	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	645	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611	}	646	}
612		647
		648	static void ecb_cold
613	static void etp_set_max_idle (unsigned int nthreads)	649	etp_set_max_idle (unsigned int nthreads)
614	{	650	{
615	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	651	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
616	max_idle = nthreads <= 0 ? 1 : nthreads;	652	max_idle = nthreads;
617	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	653	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
618	}	654	}
619		655
		656	static void ecb_cold
		657	etp_set_idle_timeout (unsigned int seconds)
		658	{
		659	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		660	idle_timeout = seconds;
		661	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		662	}
		663
		664	static void ecb_cold
620	static void etp_set_min_parallel (unsigned int nthreads)	665	etp_set_min_parallel (unsigned int nthreads)
621	{	666	{
622	if (wanted < nthreads)	667	if (wanted < nthreads)
623	wanted = nthreads;	668	wanted = nthreads;
624	}	669	}
625		670
		671	static void ecb_cold
626	static void etp_set_max_parallel (unsigned int nthreads)	672	etp_set_max_parallel (unsigned int nthreads)
627	{	673	{
628	if (wanted > nthreads)	674	if (wanted > nthreads)
629	wanted = nthreads;	675	wanted = nthreads;
630		676
631	while (started > wanted)	677	while (started > wanted)
632	etp_end_thread ();	678	etp_end_thread ();
633	}	679	}
634		680
635	/*****************************************************************************/	681	/*****************************************************************************/
636		682
		683	static void
637	static void grp_try_feed (eio_req *grp)	684	grp_try_feed (eio_req *grp)
638	{	685	{
639	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	686	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
640	{	687	{
641	grp->flags &= ~EIO_FLAG_GROUPADD;	688	grp->flags &= ~EIO_FLAG_GROUPADD;
642		689
…		…
649	break;	696	break;
650	}	697	}
651	}	698	}
652	}	699	}
653		700
		701	static int
654	static int grp_dec (eio_req *grp)	702	grp_dec (eio_req *grp)
655	{	703	{
656	--grp->size;	704	--grp->size;
657		705
658	/* call feeder, if applicable */	706	/* call feeder, if applicable */
659	grp_try_feed (grp);	707	grp_try_feed (grp);
…		…
663	return eio_finish (grp);	711	return eio_finish (grp);
664	else	712	else
665	return 0;	713	return 0;
666	}	714	}
667		715
		716	void
668	void eio_destroy (eio_req *req)	717	eio_destroy (eio_req *req)
669	{	718	{
670	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	719	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
671	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	720	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
672		721
673	EIO_DESTROY (req);	722	EIO_DESTROY (req);
674	}	723	}
675		724
		725	static int
676	static int eio_finish (eio_req *req)	726	eio_finish (eio_req *req)
677	{	727	{
678	int res = EIO_FINISH (req);	728	int res = EIO_FINISH (req);
679		729
680	if (req->grp)	730	if (req->grp)
681	{	731	{
…		…
698	eio_destroy (req);	748	eio_destroy (req);
699		749
700	return res;	750	return res;
701	}	751	}
702		752
		753	void
703	void eio_grp_cancel (eio_req *grp)	754	eio_grp_cancel (eio_req *grp)
704	{	755	{
705	for (grp = grp->grp_first; grp; grp = grp->grp_next)	756	for (grp = grp->grp_first; grp; grp = grp->grp_next)
706	eio_cancel (grp);	757	eio_cancel (grp);
707	}	758	}
708		759
		760	void
709	void eio_cancel (eio_req *req)	761	eio_cancel (eio_req *req)
710	{	762	{
711	etp_cancel (req);	763	etp_cancel (req);
712	}	764	}
713		765
		766	void
714	void eio_submit (eio_req *req)	767	eio_submit (eio_req *req)
715	{	768	{
716	etp_submit (req);	769	etp_submit (req);
717	}	770	}
718		771
719	unsigned int eio_nreqs (void)	772	unsigned int
		773	eio_nreqs (void)
720	{	774	{
721	return etp_nreqs ();	775	return etp_nreqs ();
722	}	776	}
723		777
724	unsigned int eio_nready (void)	778	unsigned int
		779	eio_nready (void)
725	{	780	{
726	return etp_nready ();	781	return etp_nready ();
727	}	782	}
728		783
729	unsigned int eio_npending (void)	784	unsigned int
		785	eio_npending (void)
730	{	786	{
731	return etp_npending ();	787	return etp_npending ();
732	}	788	}
733		789
734	unsigned int eio_nthreads (void)	790	unsigned int ecb_cold
		791	eio_nthreads (void)
735	{	792	{
736	return etp_nthreads ();	793	return etp_nthreads ();
737	}	794	}
738		795
		796	void ecb_cold
739	void eio_set_max_poll_time (double nseconds)	797	eio_set_max_poll_time (double nseconds)
740	{	798	{
741	etp_set_max_poll_time (nseconds);	799	etp_set_max_poll_time (nseconds);
742	}	800	}
743		801
		802	void ecb_cold
744	void eio_set_max_poll_reqs (unsigned int maxreqs)	803	eio_set_max_poll_reqs (unsigned int maxreqs)
745	{	804	{
746	etp_set_max_poll_reqs (maxreqs);	805	etp_set_max_poll_reqs (maxreqs);
747	}	806	}
748		807
		808	void ecb_cold
749	void eio_set_max_idle (unsigned int nthreads)	809	eio_set_max_idle (unsigned int nthreads)
750	{	810	{
751	etp_set_max_idle (nthreads);	811	etp_set_max_idle (nthreads);
752	}	812	}
753		813
		814	void ecb_cold
		815	eio_set_idle_timeout (unsigned int seconds)
		816	{
		817	etp_set_idle_timeout (seconds);
		818	}
		819
		820	void ecb_cold
754	void eio_set_min_parallel (unsigned int nthreads)	821	eio_set_min_parallel (unsigned int nthreads)
755	{	822	{
756	etp_set_min_parallel (nthreads);	823	etp_set_min_parallel (nthreads);
757	}	824	}
758		825
		826	void ecb_cold
759	void eio_set_max_parallel (unsigned int nthreads)	827	eio_set_max_parallel (unsigned int nthreads)
760	{	828	{
761	etp_set_max_parallel (nthreads);	829	etp_set_max_parallel (nthreads);
762	}	830	}
763		831
764	int eio_poll (void)	832	int eio_poll (void)
…		…
806		874
807	return res;	875	return res;
808	}	876	}
809	#endif	877	#endif
810		878
811	#ifndef HAVE_FUTIMES	879	#ifndef HAVE_UTIMES
812		880
813	# undef utimes	881	# undef utimes
814	# undef futimes
815	# define utimes(path,times) eio__utimes (path, times)	882	# define utimes(path,times) eio__utimes (path, times)
816	# define futimes(fd,times) eio__futimes (fd, times)
817		883
818	static int	884	static int
819	eio__utimes (const char *filename, const struct timeval times[2])	885	eio__utimes (const char *filename, const struct timeval times[2])
820	{	886	{
821	if (times)	887	if (times)
…		…
829	}	895	}
830	else	896	else
831	return utime (filename, 0);	897	return utime (filename, 0);
832	}	898	}
833		899
		900	#endif
		901
		902	#ifndef HAVE_FUTIMES
		903
		904	# undef futimes
		905	# define futimes(fd,times) eio__futimes (fd, times)
		906
		907	static int
834	static int eio__futimes (int fd, const struct timeval tv[2])	908	eio__futimes (int fd, const struct timeval tv[2])
835	{	909	{
836	errno = ENOSYS;	910	errno = ENOSYS;
837	return -1;	911	return -1;
838	}	912	}
839		913
…		…
843	# undef fdatasync	917	# undef fdatasync
844	# define fdatasync(fd) fsync (fd)	918	# define fdatasync(fd) fsync (fd)
845	#endif	919	#endif
846		920
847	/* sync_file_range always needs emulation */	921	/* sync_file_range always needs emulation */
848	int	922	static int
849	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	923	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
850	{	924	{
851	#if HAVE_SYNC_FILE_RANGE	925	#if HAVE_SYNC_FILE_RANGE
852	int res;	926	int res;
853		927
…		…
866	if (!res || errno != ENOSYS)	940	if (!res || errno != ENOSYS)
867	return res;	941	return res;
868	#endif	942	#endif
869		943
870	/* even though we could play tricks with the flags, it's better to always	944	/* 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 */	945	* call fdatasync, as that matches the expectation of its users best */
872	return fdatasync (fd);	946	return fdatasync (fd);
873	}	947	}
874		948
875	#if !HAVE_READAHEAD	949	#if !HAVE_READAHEAD
876	# undef readahead	950	# undef readahead
…		…
899		973
900	/* sendfile always needs emulation */	974	/* sendfile always needs emulation */
901	static ssize_t	975	static ssize_t
902	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	976	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
903	{	977	{
		978	ssize_t written = 0;
904	ssize_t res;	979	ssize_t res;
905		980
906	if (!count)	981	if (!count)
907	return 0;	982	return 0;
908		983
		984	for (;;)
		985	{
909	#if HAVE_SENDFILE	986	#if HAVE_SENDFILE
910	# if __linux	987	# if __linux
		988	off_t soffset = offset;
911	res = sendfile (ofd, ifd, &offset, count);	989	res = sendfile (ofd, ifd, &soffset, count);
912		990
913	# elif __freebsd	991	# elif __FreeBSD__
914	/*	992	/*
915	* Of course, the freebsd sendfile is a dire hack with no thoughts	993	* Of course, the freebsd sendfile is a dire hack with no thoughts
916	* wasted on making it similar to other I/O functions.	994	* wasted on making it similar to other I/O functions.
917	*/	995	*/
918	{
919	off_t sbytes;	996	off_t sbytes;
920	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	997	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
921		998
922	if (res < 0 && sbytes)	999	#if 0 /* according to the manpage, this is correct, but broken behaviour */
923	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1000	/* freebsd' sendfile will return 0 on success */
		1001	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1002	/* not on e.g. EIO or EPIPE - sounds broken */
		1003	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
924	res = sbytes;	1004	res = sbytes;
925	}	1005	#endif
		1006
		1007	/* according to source inspection, this is correct, and useful behaviour */
		1008	if (sbytes)
		1009	res = sbytes;
		1010
		1011	# elif defined (__APPLE__) && 0 /* broken, as everything on os x */
		1012	off_t sbytes = count;
		1013	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1014
		1015	/* according to the manpage, sbytes is always valid */
		1016	if (sbytes)
		1017	res = sbytes;
926		1018
927	# elif __hpux	1019	# elif __hpux
928	res = sendfile (ofd, ifd, offset, count, 0, 0);	1020	res = sendfile (ofd, ifd, offset, count, 0, 0);
929		1021
930	# elif __solaris	1022	# elif __solaris
931	{
932	struct sendfilevec vec;	1023	struct sendfilevec vec;
933	size_t sbytes;	1024	size_t sbytes;
934		1025
935	vec.sfv_fd = ifd;	1026	vec.sfv_fd = ifd;
936	vec.sfv_flag = 0;	1027	vec.sfv_flag = 0;
937	vec.sfv_off = offset;	1028	vec.sfv_off = offset;
938	vec.sfv_len = count;	1029	vec.sfv_len = count;
939		1030
940	res = sendfilev (ofd, &vec, 1, &sbytes);	1031	res = sendfilev (ofd, &vec, 1, &sbytes);
941		1032
942	if (res < 0 && sbytes)	1033	if (res < 0 && sbytes)
943	res = sbytes;	1034	res = sbytes;
944	}
945		1035
946	# endif	1036	# endif
		1037
		1038	#elif defined (_WIN32)
		1039	/* does not work, just for documentation of what would need to be done */
		1040	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1041	/* libeio guarantees that the file offset does not change, and windows */
		1042	/* has no way to get an independent handle to the same file description */
		1043	HANDLE h = TO_SOCKET (ifd);
		1044	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1045	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1046
947	#else	1047	#else
948	res = -1;	1048	res = -1;
949	errno = ENOSYS;	1049	errno = ENOSYS;
950	#endif	1050	#endif
951		1051
		1052	/* we assume sendfile can copy at least 128mb in one go */
		1053	if (res <= 128 * 1024 * 1024)
		1054	{
		1055	if (res > 0)
		1056	written += res;
		1057
		1058	if (written)
		1059	return written;
		1060
		1061	break;
		1062	}
		1063	else
		1064	{
		1065	/* if we requested more, then probably the kernel was lazy */
		1066	written += res;
		1067	offset += res;
		1068	count -= res;
		1069
		1070	if (!count)
		1071	return written;
		1072	}
		1073	}
		1074
952	if (res < 0	1075	if (res < 0
953	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1076	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1077	/* BSDs */
		1078	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1079	|| errno == ENOTSUP
		1080	#endif
		1081	|| errno == EOPNOTSUPP /* BSDs */
954	#if __solaris	1082	#if __solaris
955	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1083	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
956	#endif	1084	#endif
957	)	1085	)
958	)	1086	)
…		…
989	}	1117	}
990		1118
991	return res;	1119	return res;
992	}	1120	}
993		1121
		1122	#ifdef PAGESIZE
		1123	# define eio_pagesize() PAGESIZE
		1124	#else
		1125	static intptr_t
		1126	eio_pagesize (void)
		1127	{
		1128	static intptr_t page;
		1129
		1130	if (!page)
		1131	page = sysconf (_SC_PAGESIZE);
		1132
		1133	return page;
		1134	}
		1135	#endif
		1136
		1137	static void
		1138	eio_page_align (void **addr, size_t *length)
		1139	{
		1140	intptr_t mask = eio_pagesize () - 1;
		1141
		1142	/* round down addr */
		1143	intptr_t adj = mask & (intptr_t)*addr;
		1144
		1145	*addr = (void *)((intptr_t)*addr - adj);
		1146	*length += adj;
		1147
		1148	/* round up length */
		1149	*length = (*length + mask) & ~mask;
		1150	}
		1151
		1152	#if !_POSIX_MEMLOCK
		1153	# define eio__mlockall(a) ((errno = ENOSYS), -1)
		1154	#else
		1155
994	static int	1156	static int
		1157	eio__mlockall (int flags)
		1158	{
		1159	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1160	extern int mallopt (int, int);
		1161	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1162	#endif
		1163
		1164	if (EIO_MCL_CURRENT != MCL_CURRENT
		1165	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1166	{
		1167	flags = 0
		1168	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1169	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1170	}
		1171
		1172	return mlockall (flags);
		1173	}
		1174	#endif
		1175
		1176	#if !_POSIX_MEMLOCK_RANGE
		1177	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
		1178	#else
		1179
		1180	static int
		1181	eio__mlock (void *addr, size_t length)
		1182	{
		1183	eio_page_align (&addr, &length);
		1184
		1185	return mlock (addr, length);
		1186	}
		1187
		1188	#endif
		1189
		1190	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1191	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
		1192	#else
		1193
		1194	static int
		1195	eio__msync (void *mem, size_t len, int flags)
		1196	{
		1197	eio_page_align (&mem, &len);
		1198
		1199	if (EIO_MS_ASYNC != MS_SYNC
		1200	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1201	|| EIO_MS_SYNC != MS_SYNC)
		1202	{
		1203	flags = 0
		1204	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1205	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1206	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1207	}
		1208
		1209	return msync (mem, len, flags);
		1210	}
		1211
		1212	#endif
		1213
		1214	static int
		1215	eio__mtouch (eio_req *req)
		1216	{
		1217	void *mem = req->ptr2;
		1218	size_t len = req->size;
		1219	int flags = req->int1;
		1220
		1221	eio_page_align (&mem, &len);
		1222
		1223	{
		1224	intptr_t addr = (intptr_t)mem;
		1225	intptr_t end = addr + len;
		1226	intptr_t page = eio_pagesize ();
		1227
		1228	if (addr < end)
		1229	if (flags & EIO_MT_MODIFY) /* modify */
		1230	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1231	else
		1232	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1233	}
		1234
		1235	return 0;
		1236	}
		1237
		1238	/*****************************************************************************/
		1239	/* requests implemented outside eio_execute, because they are so large */
		1240
		1241	static void
		1242	eio__realpath (eio_req *req, etp_worker *self)
		1243	{
		1244	char *rel = req->ptr1;
		1245	char *res;
		1246	char *tmp1, *tmp2;
		1247	#if SYMLOOP_MAX > 32
		1248	int symlinks = SYMLOOP_MAX;
		1249	#else
		1250	int symlinks = 32;
		1251	#endif
		1252
		1253	req->result = -1;
		1254
		1255	errno = EINVAL;
		1256	if (!rel)
		1257	return;
		1258
		1259	errno = ENOENT;
		1260	if (!*rel)
		1261	return;
		1262
		1263	if (!req->ptr2)
		1264	{
		1265	X_LOCK (wrklock);
		1266	req->flags |= EIO_FLAG_PTR2_FREE;
		1267	X_UNLOCK (wrklock);
		1268	req->ptr2 = malloc (PATH_MAX * 3);
		1269
		1270	errno = ENOMEM;
		1271	if (!req->ptr2)
		1272	return;
		1273	}
		1274
		1275	res = req->ptr2;
		1276	tmp1 = res + PATH_MAX;
		1277	tmp2 = tmp1 + PATH_MAX;
		1278
		1279	#if 0 /* disabled, the musl way to do things is just too racy */
		1280	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1281	/* on linux we may be able to ask the kernel */
		1282	{
		1283	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1284
		1285	if (fd >= 0)
		1286	{
		1287	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1288	req->result = readlink (tmp1, res, PATH_MAX);
		1289	close (fd);
		1290
		1291	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1292
		1293	if (req->result > 0)
		1294	goto done;
		1295	}
		1296	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1297	return;
		1298	}
		1299	#endif
		1300	#endif
		1301
		1302	if (*rel != '/')
		1303	{
		1304	if (!getcwd (res, PATH_MAX))
		1305	return;
		1306
		1307	if (res [1]) /* only use if not / */
		1308	res += strlen (res);
		1309	}
		1310
		1311	while (*rel)
		1312	{
		1313	ssize_t len, linklen;
		1314	char *beg = rel;
		1315
		1316	while (*rel && *rel != '/')
		1317	++rel;
		1318
		1319	len = rel - beg;
		1320
		1321	if (!len) /* skip slashes */
		1322	{
		1323	++rel;
		1324	continue;
		1325	}
		1326
		1327	if (beg [0] == '.')
		1328	{
		1329	if (len == 1)
		1330	continue; /* . - nop */
		1331
		1332	if (beg [1] == '.' && len == 2)
		1333	{
		1334	/* .. - back up one component, if possible */
		1335
		1336	while (res != req->ptr2)
		1337	if (*--res == '/')
		1338	break;
		1339
		1340	continue;
		1341	}
		1342	}
		1343
		1344	errno = ENAMETOOLONG;
		1345	if (res + 1 + len + 1 >= tmp1)
		1346	return;
		1347
		1348	/* copy one component */
		1349	*res = '/';
		1350	memcpy (res + 1, beg, len);
		1351
		1352	/* zero-terminate, for readlink */
		1353	res [len + 1] = 0;
		1354
		1355	/* now check if it's a symlink */
		1356	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1357
		1358	if (linklen < 0)
		1359	{
		1360	if (errno != EINVAL)
		1361	return;
		1362
		1363	/* it's a normal directory. hopefully */
		1364	res += len + 1;
		1365	}
		1366	else
		1367	{
		1368	/* yay, it was a symlink - build new path in tmp2 */
		1369	int rellen = strlen (rel);
		1370
		1371	errno = ENAMETOOLONG;
		1372	if (linklen + 1 + rellen >= PATH_MAX)
		1373	return;
		1374
		1375	errno = ELOOP;
		1376	if (!--symlinks)
		1377	return;
		1378
		1379	if (*tmp1 == '/')
		1380	res = req->ptr2; /* symlink resolves to an absolute path */
		1381
		1382	/* we need to be careful, as rel might point into tmp2 already */
		1383	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1384	tmp2 [linklen] = '/';
		1385	memcpy (tmp2, tmp1, linklen);
		1386
		1387	rel = tmp2;
		1388	}
		1389	}
		1390
		1391	/* special case for the lone root path */
		1392	if (res == req->ptr2)
		1393	*res++ = '/';
		1394
		1395	req->result = res - (char *)req->ptr2;
		1396
		1397	done:
		1398	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1399	}
		1400
		1401	static signed char
995	eio_dent_cmp (const void *a_, const void *b_)	1402	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
996	{	1403	{
997	const eio_dirent *a = (const eio_dirent *)a_;	1404	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_;	1405	: a->inode < b->inode ? -1
		1406	: a->inode > b->inode ? 1
		1407	: 0;
		1408	}
999		1409
1000	return (int)b->score - (int)a->score ? (int)b->score - (int)a->score	1410	#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 */	1411
		1412	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1413	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1414
		1415	static void
		1416	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1417	{
		1418	unsigned char bits [9 + sizeof (ino_t) * 8];
		1419	unsigned char *bit = bits;
		1420
		1421	assert (CHAR_BIT == 8);
		1422	assert (sizeof (eio_dirent) * 8 < 256);
		1423	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1424	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1425
		1426	if (size <= EIO_SORT_FAST)
		1427	return;
		1428
		1429	/* first prepare an array of bits to test in our radix sort */
		1430	/* try to take endianness into account, as well as differences in ino_t sizes */
		1431	/* inode_bits must contain all inodes ORed together */
		1432	/* which is used to skip bits that are 0 everywhere, which is very common */
		1433	{
		1434	ino_t endianness;
		1435	int i, j;
		1436
		1437	/* we store the byte offset of byte n into byte n of "endianness" */
		1438	for (i = 0; i < sizeof (ino_t); ++i)
		1439	((unsigned char *)&endianness)[i] = i;
		1440
		1441	*bit++ = 0;
		1442
		1443	for (i = 0; i < sizeof (ino_t); ++i)
		1444	{
		1445	/* shifting off the byte offsets out of "endianness" */
		1446	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1447	endianness >>= 8;
		1448
		1449	for (j = 0; j < 8; ++j)
		1450	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1451	*bit++ = offs + j;
		1452	}
		1453
		1454	for (j = 0; j < 8; ++j)
		1455	if (score_bits & (1 << j))
		1456	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1457	}
		1458
		1459	/* now actually do the sorting (a variant of MSD radix sort) */
		1460	{
		1461	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1462	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1463	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1464	int stk_idx = 0;
		1465
		1466	base_stk [stk_idx] = dents;
		1467	end_stk [stk_idx] = dents + size;
		1468	bit_stk [stk_idx] = bit - 1;
		1469
		1470	do
		1471	{
		1472	base = base_stk [stk_idx];
		1473	end = end_stk [stk_idx];
		1474	bit = bit_stk [stk_idx];
		1475
		1476	for (;;)
		1477	{
		1478	unsigned char O = *bit >> 3;
		1479	unsigned char M = 1 << (*bit & 7);
		1480
		1481	eio_dirent *a = base;
		1482	eio_dirent *b = end;
		1483
		1484	if (b - a < EIO_SORT_CUTOFF)
		1485	break;
		1486
		1487	/* now bit-partition the array on the bit */
		1488	/* this ugly asymmetric loop seems to perform much better than typical */
		1489	/* partition algos found in the literature */
		1490	do
		1491	if (!(((unsigned char *)a)[O] & M))
		1492	++a;
		1493	else if (!(((unsigned char *)--b)[O] & M))
		1494	{
		1495	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1496	++a;
		1497	}
		1498	while (b > a);
		1499
		1500	/* next bit, or stop, if no bits left in this path */
		1501	if (!*--bit)
		1502	break;
		1503
		1504	base_stk [stk_idx] = a;
		1505	end_stk [stk_idx] = end;
		1506	bit_stk [stk_idx] = bit;
		1507	++stk_idx;
		1508
		1509	end = a;
		1510	}
		1511	}
		1512	while (stk_idx--);
		1513	}
		1514	}
		1515
		1516	static void
		1517	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1518	{
		1519	/* first move the smallest element to the front, to act as a sentinel */
		1520	{
		1521	int i;
		1522	eio_dirent *min = dents;
		1523
		1524	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1525	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1526	if (EIO_DENT_CMP (dents [i], <, *min))
		1527	min = &dents [i];
		1528
		1529	/* swap elements 0 and j (minimum) */
		1530	{
		1531	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1532	}
		1533	}
		1534
		1535	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1536	{
		1537	eio_dirent *i, *j;
		1538
		1539	for (i = dents + 1; i < dents + size; ++i)
		1540	{
		1541	eio_dirent value = *i;
		1542
		1543	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1544	j [1] = j [0];
		1545
		1546	j [1] = value;
		1547	}
		1548	}
		1549	}
		1550
		1551	static void
		1552	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1553	{
		1554	if (size <= 1)
		1555	return; /* our insertion sort relies on size > 0 */
		1556
		1557	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1558	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1559	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1560
		1561	/* use an insertion sort at the end, or for small arrays, */
		1562	/* as insertion sort is more efficient for small partitions */
		1563	eio_dent_insertion_sort (dents, size);
1002	}	1564	}
1003		1565
1004	/* read a full directory */	1566	/* read a full directory */
1005	static void	1567	static void
1006	eio__scandir (eio_req *req, etp_worker *self)	1568	eio__scandir (eio_req *req, etp_worker *self)
1007	{	1569	{
1008	DIR *dirp;	1570	DIR *dirp;
1009	EIO_STRUCT_DIRENT *entp;	1571	EIO_STRUCT_DIRENT *entp;
1010	unsigned char *name, *names;	1572	char *name, *names;
1011	int namesalloc = 4096;	1573	int namesalloc = 4096;
1012	int namesoffs = 0;	1574	int namesoffs = 0;
1013	int flags = req->int1;	1575	int flags = req->int1;
1014	eio_dirent *dents = 0;	1576	eio_dirent *dents = 0;
1015	int dentalloc = 128;	1577	int dentalloc = 128;
1016	int dentoffs = 0;	1578	int dentoffs = 0;
		1579	ino_t inode_bits = 0;
1017		1580
1018	req->result = -1;	1581	req->result = -1;
1019		1582
1020	if (!(flags & EIO_READDIR_DENTS))	1583	if (!(flags & EIO_READDIR_DENTS))
1021	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1584	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1022		1585
1023	X_LOCK (wrklock);	1586	X_LOCK (wrklock);
1024	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1587	/* the corresponding closedir is in ETP_WORKER_CLEAR */
1025	self->dirp = dirp = opendir (req->ptr1);	1588	self->dirp = dirp = opendir (req->ptr1);
		1589
		1590	if (req->flags & EIO_FLAG_PTR1_FREE)
		1591	free (req->ptr1);
		1592
1026	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1593	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;	1594	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
		1595	req->ptr2 = names = malloc (namesalloc);
1029	X_UNLOCK (wrklock);	1596	X_UNLOCK (wrklock);
1030		1597
1031	if (dirp && names && (!flags || dents))	1598	if (dirp && names && (!flags || dents))
1032	for (;;)	1599	for (;;)
1033	{	1600	{
…		…
1041		1608
1042	/* sort etc. */	1609	/* sort etc. */
1043	req->int1 = flags;	1610	req->int1 = flags;
1044	req->result = dentoffs;	1611	req->result = dentoffs;
1045		1612
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	1613	if (flags & EIO_READDIR_STAT_ORDER)
1055	|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))	1614	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)	1615	else if (flags & EIO_READDIR_DIRS_FIRST)
		1616	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1617	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1618	else
1061	{	1619	{
1062	/* in this case, all is known, and we just put dirs first and sort them */	1620	/* in this case, all is known, and we just put dirs first and sort them */
1063	eio_dirent *ent = dents + dentoffs;	1621	eio_dirent *oth = dents + dentoffs;
1064	eio_dirent *dir = dents;	1622	eio_dirent *dir = dents;
1065		1623
		1624	/* now partition dirs to the front, and non-dirs to the back */
		1625	/* by walking from both sides and swapping if necessary */
1066	while (ent > dir)	1626	while (oth > dir)
1067	{	1627	{
1068	if (dir->type == DT_DIR)	1628	if (dir->type == EIO_DT_DIR)
1069	++dir;	1629	++dir;
1070	else
1071	{
1072	--ent;
1073
1074	if (ent->type == DT_DIR)	1630	else if ((--oth)->type == EIO_DT_DIR)
1075	{	1631	{
1076	eio_dirent tmp = *dir;	1632	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1077	*dir = *ent;
1078	*ent = tmp;
1079		1633
1080	++dir;	1634	++dir;
1081	}	1635	}
1082	}	1636	}
		1637
		1638	/* now sort the dirs only (dirs all have the same score) */
		1639	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1083	}	1640	}
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		1641
1099	break;	1642	break;
1100	}	1643	}
1101		1644
1102	/* now add the entry to our list(s) */	1645	/* now add the entry to our list(s) */
…		…
1105	/* skip . and .. entries */	1648	/* skip . and .. entries */
1106	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1649	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1107	{	1650	{
1108	int len = D_NAMLEN (entp) + 1;	1651	int len = D_NAMLEN (entp) + 1;
1109		1652
1110	while (expect_false (namesoffs + len > namesalloc))	1653	while (ecb_expect_false (namesoffs + len > namesalloc))
1111	{	1654	{
1112	namesalloc *= 2;	1655	namesalloc *= 2;
1113	X_LOCK (wrklock);	1656	X_LOCK (wrklock);
1114	req->ptr1 = names = realloc (names, namesalloc);	1657	req->ptr2 = names = realloc (names, namesalloc);
1115	X_UNLOCK (wrklock);	1658	X_UNLOCK (wrklock);
1116		1659
1117	if (!names)	1660	if (!names)
1118	break;	1661	break;
1119	}	1662	}
…		…
1122		1665
1123	if (dents)	1666	if (dents)
1124	{	1667	{
1125	struct eio_dirent *ent;	1668	struct eio_dirent *ent;
1126		1669
1127	if (expect_false (dentoffs == dentalloc))	1670	if (ecb_expect_false (dentoffs == dentalloc))
1128	{	1671	{
1129	dentalloc *= 2;	1672	dentalloc *= 2;
1130	X_LOCK (wrklock);	1673	X_LOCK (wrklock);
1131	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1674	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1132	X_UNLOCK (wrklock);	1675	X_UNLOCK (wrklock);
1133		1676
1134	if (!dents)	1677	if (!dents)
1135	break;	1678	break;
1136	}	1679	}
1137		1680
1138	ent = dents + dentoffs;	1681	ent = dents + dentoffs;
1139		1682
1140	ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */	1683	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1141	ent->namelen = len - 1;	1684	ent->namelen = len - 1;
1142	ent->inode = D_INO (entp);	1685	ent->inode = D_INO (entp);
		1686
		1687	inode_bits |= ent->inode;
1143		1688
1144	switch (D_TYPE (entp))	1689	switch (D_TYPE (entp))
1145	{	1690	{
1146	default:	1691	default:
1147	ent->type = EIO_DT_UNKNOWN;	1692	ent->type = EIO_DT_UNKNOWN;
…		…
1190	#ifdef DT_WHT	1735	#ifdef DT_WHT
1191	case DT_WHT: ent->type = EIO_DT_WHT; break;	1736	case DT_WHT: ent->type = EIO_DT_WHT; break;
1192	#endif	1737	#endif
1193	}	1738	}
1194		1739
1195	ent->score = 0;	1740	ent->score = 7;
1196		1741
1197	if (flags & EIO_READDIR_DIRS_FIRST)	1742	if (flags & EIO_READDIR_DIRS_FIRST)
1198	{	1743	{
1199	if (ent->type == EIO_DT_UNKNOWN)	1744	if (ent->type == EIO_DT_UNKNOWN)
1200	{	1745	{
1201	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */	1746	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1202	ent->score = 98;	1747	ent->score = 1;
1203	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */	1748	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 */	1749	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1205	}	1750	}
1206	else if (ent->type == EIO_DT_DIR)	1751	else if (ent->type == EIO_DT_DIR)
1207	ent->score = 100;	1752	ent->score = 0;
1208	}	1753	}
1209	}	1754	}
1210		1755
1211	namesoffs += len;	1756	namesoffs += len;
1212	++dentoffs;	1757	++dentoffs;
1213	}	1758	}
		1759
		1760	if (EIO_CANCELLED (req))
		1761	{
		1762	errno = ECANCELED;
		1763	break;
		1764	}
1214	}	1765	}
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	}	1766	}
1248		1767
1249	/*****************************************************************************/	1768	/*****************************************************************************/
1250		1769
1251	#define ALLOC(len) \	1770	#define ALLOC(len) \
…		…
1283	if (req)	1802	if (req)
1284	break;	1803	break;
1285		1804
1286	++idle;	1805	++idle;
1287		1806
1288	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1807	ts.tv_sec = time (0) + idle_timeout;
1289	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1808	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1290	{	1809	{
1291	if (idle > max_idle)	1810	if (idle > max_idle)
1292	{	1811	{
1293	--idle;	1812	--idle;
…		…
1336	return 0;	1855	return 0;
1337	}	1856	}
1338		1857
1339	/*****************************************************************************/	1858	/*****************************************************************************/
1340		1859
		1860	int ecb_cold
1341	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1861	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1342	{	1862	{
1343	return etp_init (want_poll, done_poll);	1863	return etp_init (want_poll, done_poll);
1344	}	1864	}
1345		1865
		1866	ecb_inline void
1346	static void eio_api_destroy (eio_req *req)	1867	eio_api_destroy (eio_req *req)
1347	{	1868	{
1348	free (req);	1869	free (req);
1349	}	1870	}
1350		1871
1351	#define REQ(rtype) \	1872	#define REQ(rtype) \
…		…
1370	{ \	1891	{ \
1371	eio_api_destroy (req); \	1892	eio_api_destroy (req); \
1372	return 0; \	1893	return 0; \
1373	}	1894	}
1374		1895
		1896	static void
1375	static void eio_execute (etp_worker *self, eio_req *req)	1897	eio_execute (etp_worker *self, eio_req *req)
1376	{	1898	{
1377	errno = 0;
1378
1379	switch (req->type)	1899	switch (req->type)
1380	{	1900	{
1381	case EIO_READ: ALLOC (req->size);	1901	case EIO_READ: ALLOC (req->size);
1382	req->result = req->offs >= 0	1902	req->result = req->offs >= 0
1383	? pread (req->int1, req->ptr2, req->size, req->offs)	1903	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1394	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1914	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1395	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1915	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1396	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1916	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1397	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1917	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1398		1918
		1919	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1920	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1921	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1922	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1923
1399	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	1924	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;	1925	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;	1926	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;	1927	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;	1928	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
…		…
1410	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	1935	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1411	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	1936	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;	1937	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1413	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	1938	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1414	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	1939	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;	1940	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1416		1941
		1942	case EIO_REALPATH: eio__realpath (req, self); break;
		1943
1417	case EIO_READLINK: ALLOC (NAME_MAX);	1944	case EIO_READLINK: ALLOC (PATH_MAX);
1418	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	1945	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1419		1946
1420	case EIO_SYNC: req->result = 0; sync (); break;	1947	case EIO_SYNC: req->result = 0; sync (); break;
1421	case EIO_FSYNC: req->result = fsync (req->int1); break;	1948	case EIO_FSYNC: req->result = fsync (req->int1); break;
1422	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	1949	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1423	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;	1950	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1951	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1424	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	1952	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		1953	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;	1954	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1426		1955
1427	case EIO_READDIR: eio__scandir (req, self); break;	1956	case EIO_READDIR: eio__scandir (req, self); break;
1428		1957
1429	case EIO_BUSY:	1958	case EIO_BUSY:
1430	#ifdef _WIN32	1959	#ifdef _WIN32
1431	Sleep (req->nv1 * 1000.);	1960	Sleep (req->nv1 * 1e3);
1432	#else	1961	#else
1433	{	1962	{
1434	struct timeval tv;	1963	struct timeval tv;
1435		1964
1436	tv.tv_sec = req->nv1;	1965	tv.tv_sec = req->nv1;
1437	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	1966	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1438		1967
1439	req->result = select (0, 0, 0, 0, &tv);	1968	req->result = select (0, 0, 0, 0, &tv);
1440	}	1969	}
1441	#endif	1970	#endif
1442	break;	1971	break;
…		…
1457	times = tv;	1986	times = tv;
1458	}	1987	}
1459	else	1988	else
1460	times = 0;	1989	times = 0;
1461		1990
1462
1463	req->result = req->type == EIO_FUTIME	1991	req->result = req->type == EIO_FUTIME
1464	? futimes (req->int1, times)	1992	? futimes (req->int1, times)
1465	: utimes (req->ptr1, times);	1993	: utimes (req->ptr1, times);
1466	}	1994	}
1467	break;	1995	break;
…		…
1472	case EIO_NOP:	2000	case EIO_NOP:
1473	req->result = 0;	2001	req->result = 0;
1474	break;	2002	break;
1475		2003
1476	case EIO_CUSTOM:	2004	case EIO_CUSTOM:
1477	((void (*)(eio_req *))req->feed) (req);	2005	req->feed (req);
1478	break;	2006	break;
1479		2007
1480	default:	2008	default:
		2009	errno = ENOSYS;
1481	req->result = -1;	2010	req->result = -1;
1482	break;	2011	break;
1483	}	2012	}
1484		2013
1485	req->errorno = errno;	2014	req->errorno = errno;
…		…
1515	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2044	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1516	{	2045	{
1517	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2046	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1518	}	2047	}
1519		2048
		2049	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2050	{
		2051	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2052	}
		2053
		2054	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2055	{
		2056	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2057	}
		2058
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)	2059	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	{	2060	{
1522	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2061	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1523	}	2062	}
1524		2063
…		…
1548	}	2087	}
1549		2088
1550	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2089	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1551	{	2090	{
1552	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2091	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		2092	}
		2093
		2094	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2095	{
		2096	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1553	}	2097	}
1554		2098
1555	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2099	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1556	{	2100	{
1557	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2101	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)	2165	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1622	{	2166	{
1623	return eio__1path (EIO_READLINK, path, pri, cb, data);	2167	return eio__1path (EIO_READLINK, path, pri, cb, data);
1624	}	2168	}
1625		2169
		2170	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2171	{
		2172	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2173	}
		2174
1626	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2175	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1627	{	2176	{
1628	return eio__1path (EIO_STAT, path, pri, cb, data);	2177	return eio__1path (EIO_STAT, path, pri, cb, data);
1629	}	2178	}
1630		2179
1631	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2180	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1632	{	2181	{
1633	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2182	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1634	}	2183	}
1635		2184
		2185	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2186	{
		2187	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2188	}
		2189
1636	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2190	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1637	{	2191	{
1638	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2192	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1639	}	2193	}
1640		2194
…		…
1648	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2202	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1649	}	2203	}
1650		2204
1651	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2205	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1652	{	2206	{
1653	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2207	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1654	}	2208	}
1655		2209
1656	static eio_req *	2210	static eio_req *
1657	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2211	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1658	{	2212	{
…		…
1682	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2236	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1683	{	2237	{
1684	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2238	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1685	}	2239	}
1686		2240
1687	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2241	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1688	{	2242	{
1689	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2243	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1690	}	2244	}
1691		2245
1692	#endif	2246	#endif
1693		2247
1694	eio_req *eio_grp (eio_cb cb, void *data)	2248	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1703	#undef SEND	2257	#undef SEND
1704		2258
1705	/*****************************************************************************/	2259	/*****************************************************************************/
1706	/* grp functions */	2260	/* grp functions */
1707		2261
		2262	void
1708	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2263	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1709	{	2264	{
1710	grp->int2 = limit;	2265	grp->int2 = limit;
1711	grp->feed = feed;	2266	grp->feed = feed;
1712		2267
1713	grp_try_feed (grp);	2268	grp_try_feed (grp);
1714	}	2269	}
1715		2270
		2271	void
1716	void eio_grp_limit (eio_req *grp, int limit)	2272	eio_grp_limit (eio_req *grp, int limit)
1717	{	2273	{
1718	grp->int2 = limit;	2274	grp->int2 = limit;
1719		2275
1720	grp_try_feed (grp);	2276	grp_try_feed (grp);
1721	}	2277	}
1722		2278
		2279	void
1723	void eio_grp_add (eio_req *grp, eio_req *req)	2280	eio_grp_add (eio_req *grp, eio_req *req)
1724	{	2281	{
1725	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2282	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1726		2283
1727	grp->flags |= EIO_FLAG_GROUPADD;	2284	grp->flags |= EIO_FLAG_GROUPADD;
1728		2285
…		…
1739	}	2296	}
1740		2297
1741	/*****************************************************************************/	2298	/*****************************************************************************/
1742	/* misc garbage */	2299	/* misc garbage */
1743		2300
		2301	ssize_t
1744	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2302	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1745	{	2303	{
1746	etp_worker wrk;	2304	etp_worker wrk;
		2305	ssize_t ret;
1747		2306
1748	wrk.dbuf = 0;	2307	wrk.dbuf = 0;
1749		2308
1750	eio__sendfile (ofd, ifd, offset, count, &wrk);	2309	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1751		2310
1752	if (wrk.dbuf)	2311	if (wrk.dbuf)
1753	free (wrk.dbuf);	2312	free (wrk.dbuf);
1754	}
1755		2313
		2314	return ret;
		2315	}
		2316

Diff Legend

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