ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.34 by root, Sat Jun 6 19:57:22 2009 UTC vs.
Revision 1.86 by root, Thu Jul 14 18:30:10 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"
…		…
53	#include <sys/stat.h>	58	#include <sys/stat.h>
54	#include <limits.h>	59	#include <limits.h>
55	#include <fcntl.h>	60	#include <fcntl.h>
56	#include <assert.h>	61	#include <assert.h>
57		62
		63	#include <sys/statvfs.h>
		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
…		…
65		83
66	#ifndef EIO_FEED	84	#ifndef EIO_FEED
67	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	85	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68	#endif	86	#endif
69		87
		88	#ifndef EIO_FD_TO_WIN32_HANDLE
		89	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		90	#endif
		91	#ifndef EIO_WIN32_HANDLE_TO_FD
		92	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		93	#endif
		94
		95	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		96
		97	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		98
70	#ifdef _WIN32	99	#ifdef _WIN32
71		100
72	/*doh*/	101	#define PAGESIZE 4096 /* GetSystemInfo? */
		102
		103	#define stat(path,buf) _stati64 (path,buf)
		104	#define lstat(path,buf) stat (path,buf)
		105	#define fstat(fd,buf) _fstati64 (path,buf)
		106	#define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		107	#define mkdir(path,mode) _mkdir (path)
		108	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		109
		110	#define chown(path,uid,gid) EIO_ENOSYS ()
		111	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		112	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		113	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		114	#define mknod(path,mode,dev) EIO_ENOSYS ()
		115	#define sync() EIO_ENOSYS ()
		116
		117	/* we could even stat and see if it exists */
		118	static int
		119	symlink (const char *old, const char *neu)
		120	{
		121	if (CreateSymbolicLink (neu, old, 1))
		122	return 0;
		123
		124	if (CreateSymbolicLink (neu, old, 0))
		125	return 0;
		126
		127	return EIO_ERRNO (ENOENT, -1);
		128	}
		129
73	#else	130	#else
74		131
75	# include "config.h"
76	# include <sys/time.h>	132	#include <sys/time.h>
77	# include <sys/select.h>	133	#include <sys/select.h>
		134	#include <sys/statvfs.h>
78	# include <unistd.h>	135	#include <unistd.h>
79	# include <utime.h>	136	#include <utime.h>
80	# include <signal.h>	137	#include <signal.h>
81	# include <dirent.h>	138	#include <dirent.h>
82		139
		140	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		141	#include <sys/mman.h>
		142	#endif
		143
83	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */	144	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84	# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)	145	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
85	# define _DIRENT_HAVE_D_TYPE /* sigh */	146	#define _DIRENT_HAVE_D_TYPE /* sigh */
86	# define D_INO(de) (de)->d_fileno	147	#define D_INO(de) (de)->d_fileno
87	# define D_NAMLEN(de) (de)->d_namlen	148	#define D_NAMLEN(de) (de)->d_namlen
88	# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600	149	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
89	# define D_INO(de) (de)->d_ino	150	#define D_INO(de) (de)->d_ino
90	# endif	151	#endif
91		152
92	#ifdef _D_EXACT_NAMLEN	153	#ifdef _D_EXACT_NAMLEN
93	# undef D_NAMLEN	154	#undef D_NAMLEN
94	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)	155	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
95	#endif	156	#endif
96		157
97	# ifdef _DIRENT_HAVE_D_TYPE	158	#ifdef _DIRENT_HAVE_D_TYPE
98	# define D_TYPE(de) (de)->d_type	159	#define D_TYPE(de) (de)->d_type
99	# endif	160	#endif
100		161
101	# ifndef EIO_STRUCT_DIRENT	162	#ifndef EIO_STRUCT_DIRENT
102	# define EIO_STRUCT_DIRENT struct dirent	163	#define EIO_STRUCT_DIRENT struct dirent
103	# endif	164	#endif
104		165
105	#endif	166	#endif
106		167
107	#if HAVE_SENDFILE	168	#if HAVE_SENDFILE
108	# if __linux	169	# if __linux
109	# include <sys/sendfile.h>	170	# include <sys/sendfile.h>
110	# elif __freebsd	171	# elif __FreeBSD__ || defined __APPLE__
111	# include <sys/socket.h>	172	# include <sys/socket.h>
112	# include <sys/uio.h>	173	# include <sys/uio.h>
113	# elif __hpux	174	# elif __hpux
114	# include <sys/socket.h>	175	# include <sys/socket.h>
115	# elif __solaris /* not yet */	176	# elif __solaris
116	# include <sys/sendfile.h>	177	# include <sys/sendfile.h>
117	# else	178	# else
118	# error sendfile support requested but not available	179	# error sendfile support requested but not available
119	# endif	180	# endif
120	#endif	181	#endif
…		…
127	#endif	188	#endif
128	#ifndef D_NAMLEN	189	#ifndef D_NAMLEN
129	# define D_NAMLEN(de) strlen ((de)->d_name)	190	# define D_NAMLEN(de) strlen ((de)->d_name)
130	#endif	191	#endif
131		192
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 */	193	/* used for struct dirent, AIX doesn't provide it */
136	#ifndef NAME_MAX	194	#ifndef NAME_MAX
137	# define NAME_MAX 4096	195	# define NAME_MAX 4096
		196	#endif
		197
		198	/* used for readlink etc. */
		199	#ifndef PATH_MAX
		200	# define PATH_MAX 4096
138	#endif	201	#endif
139		202
140	/* buffer size for various temporary buffers */	203	/* buffer size for various temporary buffers */
141	#define EIO_BUFSIZE 65536	204	#define EIO_BUFSIZE 65536
142		205
…		…
148	errno = ENOMEM; \	211	errno = ENOMEM; \
149	if (!eio_buf) \	212	if (!eio_buf) \
150	return -1;	213	return -1;
151		214
152	#define EIO_TICKS ((1000000 + 1023) >> 10)	215	#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		216
167	#define ETP_PRI_MIN EIO_PRI_MIN	217	#define ETP_PRI_MIN EIO_PRI_MIN
168	#define ETP_PRI_MAX EIO_PRI_MAX	218	#define ETP_PRI_MAX EIO_PRI_MAX
169		219
170	struct etp_worker;	220	struct etp_worker;
…		…
195		245
196	/*****************************************************************************/	246	/*****************************************************************************/
197		247
198	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	248	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
199		249
200	/* calculcate time difference in ~1/EIO_TICKS of a second */	250	/* calculate time difference in ~1/EIO_TICKS of a second */
		251	ecb_inline int
201	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	252	tvdiff (struct timeval *tv1, struct timeval *tv2)
202	{	253	{
203	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	254	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
204	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	255	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
205	}	256	}
206		257
…		…
213	static unsigned int max_poll_reqs; /* reslock */	264	static unsigned int max_poll_reqs; /* reslock */
214		265
215	static volatile unsigned int nreqs; /* reqlock */	266	static volatile unsigned int nreqs; /* reqlock */
216	static volatile unsigned int nready; /* reqlock */	267	static volatile unsigned int nready; /* reqlock */
217	static volatile unsigned int npending; /* reqlock */	268	static volatile unsigned int npending; /* reqlock */
218	static volatile unsigned int max_idle = 4;	269	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		270	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
219		271
220	static mutex_t wrklock = X_MUTEX_INIT;	272	static xmutex_t wrklock;
221	static mutex_t reslock = X_MUTEX_INIT;	273	static xmutex_t reslock;
222	static mutex_t reqlock = X_MUTEX_INIT;	274	static xmutex_t reqlock;
223	static cond_t reqwait = X_COND_INIT;	275	static xcond_t reqwait;
224		276
225	#if !HAVE_PREADWRITE	277	#if !HAVE_PREADWRITE
226	/*	278	/*
227	* make our pread/pwrite emulation safe against themselves, but not against	279	* make our pread/pwrite emulation safe against themselves, but not against
228	* normal read/write by using a mutex. slows down execution a lot,	280	* normal read/write by using a mutex. slows down execution a lot,
229	* but that's your problem, not mine.	281	* but that's your problem, not mine.
230	*/	282	*/
231	static mutex_t preadwritelock = X_MUTEX_INIT;	283	static xmutex_t preadwritelock = X_MUTEX_INIT;
232	#endif	284	#endif
233		285
234	typedef struct etp_worker	286	typedef struct etp_worker
235	{	287	{
236	/* locked by wrklock */	288	/* locked by wrklock */
237	struct etp_worker *prev, *next;	289	struct etp_worker *prev, *next;
238		290
239	thread_t tid;	291	xthread_t tid;
240		292
241	/* locked by reslock, reqlock or wrklock */	293	/* locked by reslock, reqlock or wrklock */
242	ETP_REQ *req; /* currently processed request */	294	ETP_REQ *req; /* currently processed request */
243		295
244	ETP_WORKER_COMMON	296	ETP_WORKER_COMMON
…		…
249	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	301	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
250	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	302	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
251		303
252	/* worker threads management */	304	/* worker threads management */
253		305
		306	static void ecb_cold
254	static void etp_worker_clear (etp_worker *wrk)	307	etp_worker_clear (etp_worker *wrk)
255	{	308	{
256	ETP_WORKER_CLEAR (wrk);	309	ETP_WORKER_CLEAR (wrk);
257	}	310	}
258		311
		312	static void ecb_cold
259	static void etp_worker_free (etp_worker *wrk)	313	etp_worker_free (etp_worker *wrk)
260	{	314	{
261	wrk->next->prev = wrk->prev;	315	wrk->next->prev = wrk->prev;
262	wrk->prev->next = wrk->next;	316	wrk->prev->next = wrk->next;
263		317
264	free (wrk);	318	free (wrk);
265	}	319	}
266		320
267	static unsigned int etp_nreqs (void)	321	static unsigned int
		322	etp_nreqs (void)
268	{	323	{
269	int retval;	324	int retval;
270	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	325	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271	retval = nreqs;	326	retval = nreqs;
272	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	327	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273	return retval;	328	return retval;
274	}	329	}
275		330
276	static unsigned int etp_nready (void)	331	static unsigned int
		332	etp_nready (void)
277	{	333	{
278	unsigned int retval;	334	unsigned int retval;
279		335
280	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	336	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281	retval = nready;	337	retval = nready;
282	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	338	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283		339
284	return retval;	340	return retval;
285	}	341	}
286		342
287	static unsigned int etp_npending (void)	343	static unsigned int
		344	etp_npending (void)
288	{	345	{
289	unsigned int retval;	346	unsigned int retval;
290		347
291	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	348	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
292	retval = npending;	349	retval = npending;
293	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	350	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
294		351
295	return retval;	352	return retval;
296	}	353	}
297		354
298	static unsigned int etp_nthreads (void)	355	static unsigned int
		356	etp_nthreads (void)
299	{	357	{
300	unsigned int retval;	358	unsigned int retval;
301		359
302	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	360	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303	retval = started;	361	retval = started;
…		…
317	} etp_reqq;	375	} etp_reqq;
318		376
319	static etp_reqq req_queue;	377	static etp_reqq req_queue;
320	static etp_reqq res_queue;	378	static etp_reqq res_queue;
321		379
		380	static int ecb_noinline
322	static int reqq_push (etp_reqq *q, ETP_REQ *req)	381	reqq_push (etp_reqq *q, ETP_REQ *req)
323	{	382	{
324	int pri = req->pri;	383	int pri = req->pri;
325	req->next = 0;	384	req->next = 0;
326		385
327	if (q->qe[pri])	386	if (q->qe[pri])
…		…
333	q->qe[pri] = q->qs[pri] = req;	392	q->qe[pri] = q->qs[pri] = req;
334		393
335	return q->size++;	394	return q->size++;
336	}	395	}
337		396
		397	static ETP_REQ * ecb_noinline
338	static ETP_REQ *reqq_shift (etp_reqq *q)	398	reqq_shift (etp_reqq *q)
339	{	399	{
340	int pri;	400	int pri;
341		401
342	if (!q->size)	402	if (!q->size)
343	return 0;	403	return 0;
…		…
358	}	418	}
359		419
360	abort ();	420	abort ();
361	}	421	}
362		422
363	static void etp_atfork_prepare (void)	423	static void ecb_cold
		424	etp_thread_init (void)
364	{	425	{
365	X_LOCK (wrklock);
366	X_LOCK (reqlock);
367	X_LOCK (reslock);
368	#if !HAVE_PREADWRITE	426	#if !HAVE_PREADWRITE
369	X_LOCK (preadwritelock);	427	X_MUTEX_CREATE (preadwritelock);
370	#endif	428	#endif
		429	X_MUTEX_CREATE (wrklock);
		430	X_MUTEX_CREATE (reslock);
		431	X_MUTEX_CREATE (reqlock);
		432	X_COND_CREATE (reqwait);
371	}	433	}
372		434
		435	static void ecb_cold
		436	etp_atfork_prepare (void)
		437	{
		438	}
		439
		440	static void ecb_cold
373	static void etp_atfork_parent (void)	441	etp_atfork_parent (void)
374	{	442	{
375	#if !HAVE_PREADWRITE
376	X_UNLOCK (preadwritelock);
377	#endif
378	X_UNLOCK (reslock);
379	X_UNLOCK (reqlock);
380	X_UNLOCK (wrklock);
381	}	443	}
382		444
		445	static void ecb_cold
383	static void etp_atfork_child (void)	446	etp_atfork_child (void)
384	{	447	{
385	ETP_REQ *prv;	448	ETP_REQ *prv;
386		449
387	while ((prv = reqq_shift (&req_queue)))	450	while ((prv = reqq_shift (&req_queue)))
388	ETP_DESTROY (prv);	451	ETP_DESTROY (prv);
…		…
405	idle = 0;	468	idle = 0;
406	nreqs = 0;	469	nreqs = 0;
407	nready = 0;	470	nready = 0;
408	npending = 0;	471	npending = 0;
409		472
410	etp_atfork_parent ();	473	etp_thread_init ();
411	}	474	}
412		475
413	static void	476	static void ecb_cold
414	etp_once_init (void)	477	etp_once_init (void)
415	{	478	{
		479	etp_thread_init ();
416	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);	480	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
417	}	481	}
418		482
419	static int	483	static int ecb_cold
420	etp_init (void (*want_poll)(void), void (*done_poll)(void))	484	etp_init (void (*want_poll)(void), void (*done_poll)(void))
421	{	485	{
422	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	486	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
423		487
424	pthread_once (&doinit, etp_once_init);	488	pthread_once (&doinit, etp_once_init);
…		…
429	return 0;	493	return 0;
430	}	494	}
431		495
432	X_THREAD_PROC (etp_proc);	496	X_THREAD_PROC (etp_proc);
433		497
		498	static void ecb_cold
434	static void etp_start_thread (void)	499	etp_start_thread (void)
435	{	500	{
436	etp_worker *wrk = calloc (1, sizeof (etp_worker));	501	etp_worker *wrk = calloc (1, sizeof (etp_worker));
437		502
438	/*TODO*/	503	/*TODO*/
439	assert (("unable to allocate worker thread data", wrk));	504	assert (("unable to allocate worker thread data", wrk));
…		…
452	free (wrk);	517	free (wrk);
453		518
454	X_UNLOCK (wrklock);	519	X_UNLOCK (wrklock);
455	}	520	}
456		521
		522	static void
457	static void etp_maybe_start_thread (void)	523	etp_maybe_start_thread (void)
458	{	524	{
459	if (expect_true (etp_nthreads () >= wanted))	525	if (ecb_expect_true (etp_nthreads () >= wanted))
460	return;	526	return;
461		527
462	/* todo: maybe use idle here, but might be less exact */	528	/* todo: maybe use idle here, but might be less exact */
463	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	529	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
464	return;	530	return;
465		531
466	etp_start_thread ();	532	etp_start_thread ();
467	}	533	}
468		534
		535	static void ecb_cold
469	static void etp_end_thread (void)	536	etp_end_thread (void)
470	{	537	{
471	eio_req *req = calloc (1, sizeof (eio_req));	538	eio_req *req = calloc (1, sizeof (eio_req));
472		539
473	req->type = -1;	540	req->type = -1;
474	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	541	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
481	X_LOCK (wrklock);	548	X_LOCK (wrklock);
482	--started;	549	--started;
483	X_UNLOCK (wrklock);	550	X_UNLOCK (wrklock);
484	}	551	}
485		552
486	static int etp_poll (void)	553	static int
		554	etp_poll (void)
487	{	555	{
488	unsigned int maxreqs;	556	unsigned int maxreqs;
489	unsigned int maxtime;	557	unsigned int maxtime;
490	struct timeval tv_start, tv_now;	558	struct timeval tv_start, tv_now;
491		559
…		…
521		589
522	X_LOCK (reqlock);	590	X_LOCK (reqlock);
523	--nreqs;	591	--nreqs;
524	X_UNLOCK (reqlock);	592	X_UNLOCK (reqlock);
525		593
526	if (expect_false (req->type == EIO_GROUP && req->size))	594	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
527	{	595	{
528	req->int1 = 1; /* mark request as delayed */	596	req->int1 = 1; /* mark request as delayed */
529	continue;	597	continue;
530	}	598	}
531	else	599	else
532	{	600	{
533	int res = ETP_FINISH (req);	601	int res = ETP_FINISH (req);
534	if (expect_false (res))	602	if (ecb_expect_false (res))
535	return res;	603	return res;
536	}	604	}
537		605
538	if (expect_false (maxreqs && !--maxreqs))	606	if (ecb_expect_false (maxreqs && !--maxreqs))
539	break;	607	break;
540		608
541	if (maxtime)	609	if (maxtime)
542	{	610	{
543	gettimeofday (&tv_now, 0);	611	gettimeofday (&tv_now, 0);
…		…
549		617
550	errno = EAGAIN;	618	errno = EAGAIN;
551	return -1;	619	return -1;
552	}	620	}
553		621
		622	static void
554	static void etp_cancel (ETP_REQ *req)	623	etp_cancel (ETP_REQ *req)
555	{	624	{
556	X_LOCK (wrklock);	625	req->cancelled = 1;
557	req->flags |= EIO_FLAG_CANCELLED;
558	X_UNLOCK (wrklock);
559		626
560	eio_grp_cancel (req);	627	eio_grp_cancel (req);
561	}	628	}
562		629
		630	static void
563	static void etp_submit (ETP_REQ *req)	631	etp_submit (ETP_REQ *req)
564	{	632	{
565	req->pri -= ETP_PRI_MIN;	633	req->pri -= ETP_PRI_MIN;
566		634
567	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	635	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;	636	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
569		637
570	if (expect_false (req->type == EIO_GROUP))	638	if (ecb_expect_false (req->type == EIO_GROUP))
571	{	639	{
572	/* I hope this is worth it :/ */	640	/* I hope this is worth it :/ */
573	X_LOCK (reqlock);	641	X_LOCK (reqlock);
574	++nreqs;	642	++nreqs;
575	X_UNLOCK (reqlock);	643	X_UNLOCK (reqlock);
…		…
594		662
595	etp_maybe_start_thread ();	663	etp_maybe_start_thread ();
596	}	664	}
597	}	665	}
598		666
		667	static void ecb_cold
599	static void etp_set_max_poll_time (double nseconds)	668	etp_set_max_poll_time (double nseconds)
600	{	669	{
601	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	670	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
602	max_poll_time = nseconds;	671	max_poll_time = nseconds * EIO_TICKS;
603	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	672	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
604	}	673	}
605		674
		675	static void ecb_cold
606	static void etp_set_max_poll_reqs (unsigned int maxreqs)	676	etp_set_max_poll_reqs (unsigned int maxreqs)
607	{	677	{
608	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	678	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609	max_poll_reqs = maxreqs;	679	max_poll_reqs = maxreqs;
610	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	680	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611	}	681	}
612		682
		683	static void ecb_cold
613	static void etp_set_max_idle (unsigned int nthreads)	684	etp_set_max_idle (unsigned int nthreads)
614	{	685	{
615	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	686	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
616	max_idle = nthreads <= 0 ? 1 : nthreads;	687	max_idle = nthreads;
617	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	688	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
618	}	689	}
619		690
		691	static void ecb_cold
		692	etp_set_idle_timeout (unsigned int seconds)
		693	{
		694	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		695	idle_timeout = seconds;
		696	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		697	}
		698
		699	static void ecb_cold
620	static void etp_set_min_parallel (unsigned int nthreads)	700	etp_set_min_parallel (unsigned int nthreads)
621	{	701	{
622	if (wanted < nthreads)	702	if (wanted < nthreads)
623	wanted = nthreads;	703	wanted = nthreads;
624	}	704	}
625		705
		706	static void ecb_cold
626	static void etp_set_max_parallel (unsigned int nthreads)	707	etp_set_max_parallel (unsigned int nthreads)
627	{	708	{
628	if (wanted > nthreads)	709	if (wanted > nthreads)
629	wanted = nthreads;	710	wanted = nthreads;
630		711
631	while (started > wanted)	712	while (started > wanted)
632	etp_end_thread ();	713	etp_end_thread ();
633	}	714	}
634		715
635	/*****************************************************************************/	716	/*****************************************************************************/
636		717
		718	static void
637	static void grp_try_feed (eio_req *grp)	719	grp_try_feed (eio_req *grp)
638	{	720	{
639	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	721	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
640	{	722	{
641	grp->flags &= ~EIO_FLAG_GROUPADD;	723	grp->flags &= ~EIO_FLAG_GROUPADD;
642		724
…		…
649	break;	731	break;
650	}	732	}
651	}	733	}
652	}	734	}
653		735
		736	static int
654	static int grp_dec (eio_req *grp)	737	grp_dec (eio_req *grp)
655	{	738	{
656	--grp->size;	739	--grp->size;
657		740
658	/* call feeder, if applicable */	741	/* call feeder, if applicable */
659	grp_try_feed (grp);	742	grp_try_feed (grp);
…		…
663	return eio_finish (grp);	746	return eio_finish (grp);
664	else	747	else
665	return 0;	748	return 0;
666	}	749	}
667		750
		751	static void
668	void eio_destroy (eio_req *req)	752	eio_destroy (eio_req *req)
669	{	753	{
670	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	754	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
671	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	755	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
672		756
673	EIO_DESTROY (req);	757	EIO_DESTROY (req);
674	}	758	}
675		759
		760	static int
676	static int eio_finish (eio_req *req)	761	eio_finish (eio_req *req)
677	{	762	{
678	int res = EIO_FINISH (req);	763	int res = EIO_FINISH (req);
679		764
680	if (req->grp)	765	if (req->grp)
681	{	766	{
…		…
689	if (grp->grp_first == req)	774	if (grp->grp_first == req)
690	grp->grp_first = req->grp_next;	775	grp->grp_first = req->grp_next;
691		776
692	res2 = grp_dec (grp);	777	res2 = grp_dec (grp);
693		778
694	if (!res && res2)	779	if (!res)
695	res = res2;	780	res = res2;
696	}	781	}
697		782
698	eio_destroy (req);	783	eio_destroy (req);
699		784
700	return res;	785	return res;
701	}	786	}
702		787
		788	void
703	void eio_grp_cancel (eio_req *grp)	789	eio_grp_cancel (eio_req *grp)
704	{	790	{
705	for (grp = grp->grp_first; grp; grp = grp->grp_next)	791	for (grp = grp->grp_first; grp; grp = grp->grp_next)
706	eio_cancel (grp);	792	eio_cancel (grp);
707	}	793	}
708		794
		795	void
709	void eio_cancel (eio_req *req)	796	eio_cancel (eio_req *req)
710	{	797	{
711	etp_cancel (req);	798	etp_cancel (req);
712	}	799	}
713		800
		801	void
714	void eio_submit (eio_req *req)	802	eio_submit (eio_req *req)
715	{	803	{
716	etp_submit (req);	804	etp_submit (req);
717	}	805	}
718		806
719	unsigned int eio_nreqs (void)	807	unsigned int
		808	eio_nreqs (void)
720	{	809	{
721	return etp_nreqs ();	810	return etp_nreqs ();
722	}	811	}
723		812
724	unsigned int eio_nready (void)	813	unsigned int
		814	eio_nready (void)
725	{	815	{
726	return etp_nready ();	816	return etp_nready ();
727	}	817	}
728		818
729	unsigned int eio_npending (void)	819	unsigned int
		820	eio_npending (void)
730	{	821	{
731	return etp_npending ();	822	return etp_npending ();
732	}	823	}
733		824
734	unsigned int eio_nthreads (void)	825	unsigned int ecb_cold
		826	eio_nthreads (void)
735	{	827	{
736	return etp_nthreads ();	828	return etp_nthreads ();
737	}	829	}
738		830
		831	void ecb_cold
739	void eio_set_max_poll_time (double nseconds)	832	eio_set_max_poll_time (double nseconds)
740	{	833	{
741	etp_set_max_poll_time (nseconds);	834	etp_set_max_poll_time (nseconds);
742	}	835	}
743		836
		837	void ecb_cold
744	void eio_set_max_poll_reqs (unsigned int maxreqs)	838	eio_set_max_poll_reqs (unsigned int maxreqs)
745	{	839	{
746	etp_set_max_poll_reqs (maxreqs);	840	etp_set_max_poll_reqs (maxreqs);
747	}	841	}
748		842
		843	void ecb_cold
749	void eio_set_max_idle (unsigned int nthreads)	844	eio_set_max_idle (unsigned int nthreads)
750	{	845	{
751	etp_set_max_idle (nthreads);	846	etp_set_max_idle (nthreads);
752	}	847	}
753		848
		849	void ecb_cold
		850	eio_set_idle_timeout (unsigned int seconds)
		851	{
		852	etp_set_idle_timeout (seconds);
		853	}
		854
		855	void ecb_cold
754	void eio_set_min_parallel (unsigned int nthreads)	856	eio_set_min_parallel (unsigned int nthreads)
755	{	857	{
756	etp_set_min_parallel (nthreads);	858	etp_set_min_parallel (nthreads);
757	}	859	}
758		860
		861	void ecb_cold
759	void eio_set_max_parallel (unsigned int nthreads)	862	eio_set_max_parallel (unsigned int nthreads)
760	{	863	{
761	etp_set_max_parallel (nthreads);	864	etp_set_max_parallel (nthreads);
762	}	865	}
763		866
764	int eio_poll (void)	867	int eio_poll (void)
…		…
806		909
807	return res;	910	return res;
808	}	911	}
809	#endif	912	#endif
810		913
811	#ifndef HAVE_FUTIMES	914	#ifndef HAVE_UTIMES
812		915
813	# undef utimes	916	# undef utimes
814	# undef futimes
815	# define utimes(path,times) eio__utimes (path, times)	917	# define utimes(path,times) eio__utimes (path, times)
816	# define futimes(fd,times) eio__futimes (fd, times)
817		918
818	static int	919	static int
819	eio__utimes (const char *filename, const struct timeval times[2])	920	eio__utimes (const char *filename, const struct timeval times[2])
820	{	921	{
821	if (times)	922	if (times)
…		…
829	}	930	}
830	else	931	else
831	return utime (filename, 0);	932	return utime (filename, 0);
832	}	933	}
833		934
		935	#endif
		936
		937	#ifndef HAVE_FUTIMES
		938
		939	# undef futimes
		940	# define futimes(fd,times) eio__futimes (fd, times)
		941
		942	static int
834	static int eio__futimes (int fd, const struct timeval tv[2])	943	eio__futimes (int fd, const struct timeval tv[2])
835	{	944	{
836	errno = ENOSYS;	945	errno = ENOSYS;
837	return -1;	946	return -1;
838	}	947	}
839		948
…		…
843	# undef fdatasync	952	# undef fdatasync
844	# define fdatasync(fd) fsync (fd)	953	# define fdatasync(fd) fsync (fd)
845	#endif	954	#endif
846		955
847	/* sync_file_range always needs emulation */	956	/* sync_file_range always needs emulation */
848	int	957	static int
849	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	958	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
850	{	959	{
851	#if HAVE_SYNC_FILE_RANGE	960	#if HAVE_SYNC_FILE_RANGE
852	int res;	961	int res;
853		962
…		…
866	if (!res || errno != ENOSYS)	975	if (!res || errno != ENOSYS)
867	return res;	976	return res;
868	#endif	977	#endif
869		978
870	/* even though we could play tricks with the flags, it's better to always	979	/* 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 */	980	* call fdatasync, as that matches the expectation of its users best */
872	return fdatasync (fd);	981	return fdatasync (fd);
		982	}
		983
		984	static int
		985	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		986	{
		987	#if HAVE_FALLOCATE
		988	return fallocate (fd, mode, offset, len);
		989	#else
		990	errno = ENOSYS;
		991	return -1;
		992	#endif
873	}	993	}
874		994
875	#if !HAVE_READAHEAD	995	#if !HAVE_READAHEAD
876	# undef readahead	996	# undef readahead
877	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	997	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
…		…
899		1019
900	/* sendfile always needs emulation */	1020	/* sendfile always needs emulation */
901	static ssize_t	1021	static ssize_t
902	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1022	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
903	{	1023	{
		1024	ssize_t written = 0;
904	ssize_t res;	1025	ssize_t res;
905		1026
906	if (!count)	1027	if (!count)
907	return 0;	1028	return 0;
908		1029
		1030	for (;;)
		1031	{
		1032	#ifdef __APPLE__
		1033	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1034	#endif
909	#if HAVE_SENDFILE	1035	#if HAVE_SENDFILE
910	# if __linux	1036	# if __linux
		1037	off_t soffset = offset;
911	res = sendfile (ofd, ifd, &offset, count);	1038	res = sendfile (ofd, ifd, &soffset, count);
912		1039
913	# elif __freebsd	1040	# elif __FreeBSD__
914	/*	1041	/*
915	* Of course, the freebsd sendfile is a dire hack with no thoughts	1042	* Of course, the freebsd sendfile is a dire hack with no thoughts
916	* wasted on making it similar to other I/O functions.	1043	* wasted on making it similar to other I/O functions.
917	*/	1044	*/
918	{
919	off_t sbytes;	1045	off_t sbytes;
920	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1046	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
921		1047
922	if (res < 0 && sbytes)	1048	#if 0 /* according to the manpage, this is correct, but broken behaviour */
923	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1049	/* freebsd' sendfile will return 0 on success */
		1050	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1051	/* not on e.g. EIO or EPIPE - sounds broken */
		1052	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
924	res = sbytes;	1053	res = sbytes;
925	}	1054	#endif
		1055
		1056	/* according to source inspection, this is correct, and useful behaviour */
		1057	if (sbytes)
		1058	res = sbytes;
		1059
		1060	# elif defined (__APPLE__)
		1061	off_t sbytes = count;
		1062	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1063
		1064	/* according to the manpage, sbytes is always valid */
		1065	if (sbytes)
		1066	res = sbytes;
926		1067
927	# elif __hpux	1068	# elif __hpux
928	res = sendfile (ofd, ifd, offset, count, 0, 0);	1069	res = sendfile (ofd, ifd, offset, count, 0, 0);
929		1070
930	# elif __solaris	1071	# elif __solaris
931	{
932	struct sendfilevec vec;	1072	struct sendfilevec vec;
933	size_t sbytes;	1073	size_t sbytes;
934		1074
935	vec.sfv_fd = ifd;	1075	vec.sfv_fd = ifd;
936	vec.sfv_flag = 0;	1076	vec.sfv_flag = 0;
937	vec.sfv_off = offset;	1077	vec.sfv_off = offset;
938	vec.sfv_len = count;	1078	vec.sfv_len = count;
939		1079
940	res = sendfilev (ofd, &vec, 1, &sbytes);	1080	res = sendfilev (ofd, &vec, 1, &sbytes);
941		1081
942	if (res < 0 && sbytes)	1082	if (res < 0 && sbytes)
943	res = sbytes;	1083	res = sbytes;
944	}
945		1084
946	# endif	1085	# endif
		1086
		1087	#elif defined (_WIN32) && 0
		1088	/* does not work, just for documentation of what would need to be done */
		1089	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1090	/* libeio guarantees that the file offset does not change, and windows */
		1091	/* has no way to get an independent handle to the same file description */
		1092	HANDLE h = TO_SOCKET (ifd);
		1093	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1094	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1095
947	#else	1096	#else
948	res = -1;	1097	res = -1;
949	errno = ENOSYS;	1098	errno = ENOSYS;
950	#endif	1099	#endif
951		1100
		1101	/* we assume sendfile can copy at least 128mb in one go */
		1102	if (res <= 128 * 1024 * 1024)
		1103	{
		1104	if (res > 0)
		1105	written += res;
		1106
		1107	if (written)
		1108	return written;
		1109
		1110	break;
		1111	}
		1112	else
		1113	{
		1114	/* if we requested more, then probably the kernel was lazy */
		1115	written += res;
		1116	offset += res;
		1117	count -= res;
		1118
		1119	if (!count)
		1120	return written;
		1121	}
		1122	}
		1123
952	if (res < 0	1124	if (res < 0
953	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1125	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1126	/* BSDs */
		1127	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1128	|| errno == ENOTSUP
		1129	#endif
		1130	|| errno == EOPNOTSUPP /* BSDs */
954	#if __solaris	1131	#if __solaris
955	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1132	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
956	#endif	1133	#endif
957	)	1134	)
958	)	1135	)
…		…
989	}	1166	}
990		1167
991	return res;	1168	return res;
992	}	1169	}
993		1170
		1171	#ifdef PAGESIZE
		1172	# define eio_pagesize() PAGESIZE
		1173	#else
		1174	static intptr_t
		1175	eio_pagesize (void)
		1176	{
		1177	static intptr_t page;
		1178
		1179	if (!page)
		1180	page = sysconf (_SC_PAGESIZE);
		1181
		1182	return page;
		1183	}
		1184	#endif
		1185
		1186	static void
		1187	eio_page_align (void **addr, size_t *length)
		1188	{
		1189	intptr_t mask = eio_pagesize () - 1;
		1190
		1191	/* round down addr */
		1192	intptr_t adj = mask & (intptr_t)*addr;
		1193
		1194	*addr = (void *)((intptr_t)*addr - adj);
		1195	*length += adj;
		1196
		1197	/* round up length */
		1198	*length = (*length + mask) & ~mask;
		1199	}
		1200
		1201	#if !_POSIX_MEMLOCK
		1202	# define eio__mlockall(a) eio_nosyscall()
		1203	#else
		1204
994	static int	1205	static int
		1206	eio__mlockall (int flags)
		1207	{
		1208	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1209	extern int mallopt (int, int);
		1210	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1211	#endif
		1212
		1213	if (EIO_MCL_CURRENT != MCL_CURRENT
		1214	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1215	{
		1216	flags = 0
		1217	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1218	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1219	}
		1220
		1221	return mlockall (flags);
		1222	}
		1223	#endif
		1224
		1225	#if !_POSIX_MEMLOCK_RANGE
		1226	# define eio__mlock(a,b) EIO_ENOSYS ()
		1227	#else
		1228
		1229	static int
		1230	eio__mlock (void *addr, size_t length)
		1231	{
		1232	eio_page_align (&addr, &length);
		1233
		1234	return mlock (addr, length);
		1235	}
		1236
		1237	#endif
		1238
		1239	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1240	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1241	#else
		1242
		1243	static int
		1244	eio__msync (void *mem, size_t len, int flags)
		1245	{
		1246	eio_page_align (&mem, &len);
		1247
		1248	if (EIO_MS_ASYNC != MS_SYNC
		1249	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1250	|| EIO_MS_SYNC != MS_SYNC)
		1251	{
		1252	flags = 0
		1253	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1254	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1255	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1256	}
		1257
		1258	return msync (mem, len, flags);
		1259	}
		1260
		1261	#endif
		1262
		1263	static int
		1264	eio__mtouch (eio_req *req)
		1265	{
		1266	void *mem = req->ptr2;
		1267	size_t len = req->size;
		1268	int flags = req->int1;
		1269
		1270	eio_page_align (&mem, &len);
		1271
		1272	{
		1273	intptr_t addr = (intptr_t)mem;
		1274	intptr_t end = addr + len;
		1275	intptr_t page = eio_pagesize ();
		1276
		1277	if (addr < end)
		1278	if (flags & EIO_MT_MODIFY) /* modify */
		1279	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1280	else
		1281	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1282	}
		1283
		1284	return 0;
		1285	}
		1286
		1287	/*****************************************************************************/
		1288	/* requests implemented outside eio_execute, because they are so large */
		1289
		1290	static void
		1291	eio__realpath (eio_req *req, etp_worker *self)
		1292	{
		1293	char *rel = req->ptr1;
		1294	char *res;
		1295	char *tmp1, *tmp2;
		1296	#if SYMLOOP_MAX > 32
		1297	int symlinks = SYMLOOP_MAX;
		1298	#else
		1299	int symlinks = 32;
		1300	#endif
		1301
		1302	req->result = -1;
		1303
		1304	errno = EINVAL;
		1305	if (!rel)
		1306	return;
		1307
		1308	errno = ENOENT;
		1309	if (!*rel)
		1310	return;
		1311
		1312	if (!req->ptr2)
		1313	{
		1314	X_LOCK (wrklock);
		1315	req->flags |= EIO_FLAG_PTR2_FREE;
		1316	X_UNLOCK (wrklock);
		1317	req->ptr2 = malloc (PATH_MAX * 3);
		1318
		1319	errno = ENOMEM;
		1320	if (!req->ptr2)
		1321	return;
		1322	}
		1323
		1324	res = req->ptr2;
		1325	tmp1 = res + PATH_MAX;
		1326	tmp2 = tmp1 + PATH_MAX;
		1327
		1328	#if 0 /* disabled, the musl way to do things is just too racy */
		1329	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1330	/* on linux we may be able to ask the kernel */
		1331	{
		1332	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1333
		1334	if (fd >= 0)
		1335	{
		1336	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1337	req->result = readlink (tmp1, res, PATH_MAX);
		1338	close (fd);
		1339
		1340	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1341
		1342	if (req->result > 0)
		1343	goto done;
		1344	}
		1345	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1346	return;
		1347	}
		1348	#endif
		1349	#endif
		1350
		1351	if (*rel != '/')
		1352	{
		1353	if (!getcwd (res, PATH_MAX))
		1354	return;
		1355
		1356	if (res [1]) /* only use if not / */
		1357	res += strlen (res);
		1358	}
		1359
		1360	while (*rel)
		1361	{
		1362	ssize_t len, linklen;
		1363	char *beg = rel;
		1364
		1365	while (*rel && *rel != '/')
		1366	++rel;
		1367
		1368	len = rel - beg;
		1369
		1370	if (!len) /* skip slashes */
		1371	{
		1372	++rel;
		1373	continue;
		1374	}
		1375
		1376	if (beg [0] == '.')
		1377	{
		1378	if (len == 1)
		1379	continue; /* . - nop */
		1380
		1381	if (beg [1] == '.' && len == 2)
		1382	{
		1383	/* .. - back up one component, if possible */
		1384
		1385	while (res != req->ptr2)
		1386	if (*--res == '/')
		1387	break;
		1388
		1389	continue;
		1390	}
		1391	}
		1392
		1393	errno = ENAMETOOLONG;
		1394	if (res + 1 + len + 1 >= tmp1)
		1395	return;
		1396
		1397	/* copy one component */
		1398	*res = '/';
		1399	memcpy (res + 1, beg, len);
		1400
		1401	/* zero-terminate, for readlink */
		1402	res [len + 1] = 0;
		1403
		1404	/* now check if it's a symlink */
		1405	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1406
		1407	if (linklen < 0)
		1408	{
		1409	if (errno != EINVAL)
		1410	return;
		1411
		1412	/* it's a normal directory. hopefully */
		1413	res += len + 1;
		1414	}
		1415	else
		1416	{
		1417	/* yay, it was a symlink - build new path in tmp2 */
		1418	int rellen = strlen (rel);
		1419
		1420	errno = ENAMETOOLONG;
		1421	if (linklen + 1 + rellen >= PATH_MAX)
		1422	return;
		1423
		1424	errno = ELOOP;
		1425	if (!--symlinks)
		1426	return;
		1427
		1428	if (*tmp1 == '/')
		1429	res = req->ptr2; /* symlink resolves to an absolute path */
		1430
		1431	/* we need to be careful, as rel might point into tmp2 already */
		1432	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1433	tmp2 [linklen] = '/';
		1434	memcpy (tmp2, tmp1, linklen);
		1435
		1436	rel = tmp2;
		1437	}
		1438	}
		1439
		1440	/* special case for the lone root path */
		1441	if (res == req->ptr2)
		1442	*res++ = '/';
		1443
		1444	req->result = res - (char *)req->ptr2;
		1445
		1446	done:
		1447	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1448	}
		1449
		1450	static signed char
995	eio_dent_cmp (const void *a_, const void *b_)	1451	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
996	{	1452	{
997	const eio_dirent *a = (const eio_dirent *)a_;	1453	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_;	1454	: a->inode < b->inode ? -1
		1455	: a->inode > b->inode ? 1
		1456	: 0;
		1457	}
999		1458
1000	return (int)b->score - (int)a->score ? (int)b->score - (int)a->score	1459	#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 */	1460
		1461	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1462	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1463
		1464	static void
		1465	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1466	{
		1467	unsigned char bits [9 + sizeof (ino_t) * 8];
		1468	unsigned char *bit = bits;
		1469
		1470	assert (CHAR_BIT == 8);
		1471	assert (sizeof (eio_dirent) * 8 < 256);
		1472	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1473	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1474
		1475	if (size <= EIO_SORT_FAST)
		1476	return;
		1477
		1478	/* first prepare an array of bits to test in our radix sort */
		1479	/* try to take endianness into account, as well as differences in ino_t sizes */
		1480	/* inode_bits must contain all inodes ORed together */
		1481	/* which is used to skip bits that are 0 everywhere, which is very common */
		1482	{
		1483	ino_t endianness;
		1484	int i, j;
		1485
		1486	/* we store the byte offset of byte n into byte n of "endianness" */
		1487	for (i = 0; i < sizeof (ino_t); ++i)
		1488	((unsigned char *)&endianness)[i] = i;
		1489
		1490	*bit++ = 0;
		1491
		1492	for (i = 0; i < sizeof (ino_t); ++i)
		1493	{
		1494	/* shifting off the byte offsets out of "endianness" */
		1495	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1496	endianness >>= 8;
		1497
		1498	for (j = 0; j < 8; ++j)
		1499	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1500	*bit++ = offs + j;
		1501	}
		1502
		1503	for (j = 0; j < 8; ++j)
		1504	if (score_bits & (1 << j))
		1505	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1506	}
		1507
		1508	/* now actually do the sorting (a variant of MSD radix sort) */
		1509	{
		1510	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1511	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1512	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1513	int stk_idx = 0;
		1514
		1515	base_stk [stk_idx] = dents;
		1516	end_stk [stk_idx] = dents + size;
		1517	bit_stk [stk_idx] = bit - 1;
		1518
		1519	do
		1520	{
		1521	base = base_stk [stk_idx];
		1522	end = end_stk [stk_idx];
		1523	bit = bit_stk [stk_idx];
		1524
		1525	for (;;)
		1526	{
		1527	unsigned char O = *bit >> 3;
		1528	unsigned char M = 1 << (*bit & 7);
		1529
		1530	eio_dirent *a = base;
		1531	eio_dirent *b = end;
		1532
		1533	if (b - a < EIO_SORT_CUTOFF)
		1534	break;
		1535
		1536	/* now bit-partition the array on the bit */
		1537	/* this ugly asymmetric loop seems to perform much better than typical */
		1538	/* partition algos found in the literature */
		1539	do
		1540	if (!(((unsigned char *)a)[O] & M))
		1541	++a;
		1542	else if (!(((unsigned char *)--b)[O] & M))
		1543	{
		1544	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1545	++a;
		1546	}
		1547	while (b > a);
		1548
		1549	/* next bit, or stop, if no bits left in this path */
		1550	if (!*--bit)
		1551	break;
		1552
		1553	base_stk [stk_idx] = a;
		1554	end_stk [stk_idx] = end;
		1555	bit_stk [stk_idx] = bit;
		1556	++stk_idx;
		1557
		1558	end = a;
		1559	}
		1560	}
		1561	while (stk_idx--);
		1562	}
		1563	}
		1564
		1565	static void
		1566	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1567	{
		1568	/* first move the smallest element to the front, to act as a sentinel */
		1569	{
		1570	int i;
		1571	eio_dirent *min = dents;
		1572
		1573	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1574	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1575	if (EIO_DENT_CMP (dents [i], <, *min))
		1576	min = &dents [i];
		1577
		1578	/* swap elements 0 and j (minimum) */
		1579	{
		1580	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1581	}
		1582	}
		1583
		1584	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1585	{
		1586	eio_dirent *i, *j;
		1587
		1588	for (i = dents + 1; i < dents + size; ++i)
		1589	{
		1590	eio_dirent value = *i;
		1591
		1592	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1593	j [1] = j [0];
		1594
		1595	j [1] = value;
		1596	}
		1597	}
		1598	}
		1599
		1600	static void
		1601	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1602	{
		1603	if (size <= 1)
		1604	return; /* our insertion sort relies on size > 0 */
		1605
		1606	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1607	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1608	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1609
		1610	/* use an insertion sort at the end, or for small arrays, */
		1611	/* as insertion sort is more efficient for small partitions */
		1612	eio_dent_insertion_sort (dents, size);
1002	}	1613	}
1003		1614
1004	/* read a full directory */	1615	/* read a full directory */
1005	static void	1616	static void
1006	eio__scandir (eio_req *req, etp_worker *self)	1617	eio__scandir (eio_req *req, etp_worker *self)
1007	{	1618	{
1008	DIR *dirp;	1619	DIR *dirp;
1009	EIO_STRUCT_DIRENT *entp;	1620	EIO_STRUCT_DIRENT *entp;
1010	unsigned char *name, *names;	1621	char *name, *names;
1011	int namesalloc = 4096;	1622	int namesalloc = 4096;
1012	int namesoffs = 0;	1623	int namesoffs = 0;
1013	int flags = req->int1;	1624	int flags = req->int1;
1014	eio_dirent *dents = 0;	1625	eio_dirent *dents = 0;
1015	int dentalloc = 128;	1626	int dentalloc = 128;
1016	int dentoffs = 0;	1627	int dentoffs = 0;
		1628	ino_t inode_bits = 0;
1017		1629
1018	req->result = -1;	1630	req->result = -1;
1019		1631
1020	if (!(flags & EIO_READDIR_DENTS))	1632	if (!(flags & EIO_READDIR_DENTS))
1021	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1633	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1022		1634
1023	X_LOCK (wrklock);	1635	X_LOCK (wrklock);
1024	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1636	/* the corresponding closedir is in ETP_WORKER_CLEAR */
1025	self->dirp = dirp = opendir (req->ptr1);	1637	self->dirp = dirp = opendir (req->ptr1);
		1638
		1639	if (req->flags & EIO_FLAG_PTR1_FREE)
		1640	free (req->ptr1);
		1641
1026	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1642	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;	1643	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
		1644	req->ptr2 = names = malloc (namesalloc);
1029	X_UNLOCK (wrklock);	1645	X_UNLOCK (wrklock);
1030		1646
1031	if (dirp && names && (!flags || dents))	1647	if (dirp && names && (!flags || dents))
1032	for (;;)	1648	for (;;)
1033	{	1649	{
…		…
1041		1657
1042	/* sort etc. */	1658	/* sort etc. */
1043	req->int1 = flags;	1659	req->int1 = flags;
1044	req->result = dentoffs;	1660	req->result = dentoffs;
1045		1661
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	1662	if (flags & EIO_READDIR_STAT_ORDER)
1055	|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))	1663	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)	1664	else if (flags & EIO_READDIR_DIRS_FIRST)
		1665	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1666	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1667	else
1061	{	1668	{
1062	/* in this case, all is known, and we just put dirs first and sort them */	1669	/* in this case, all is known, and we just put dirs first and sort them */
1063	eio_dirent *ent = dents + dentoffs;	1670	eio_dirent *oth = dents + dentoffs;
1064	eio_dirent *dir = dents;	1671	eio_dirent *dir = dents;
1065		1672
		1673	/* now partition dirs to the front, and non-dirs to the back */
		1674	/* by walking from both sides and swapping if necessary */
1066	while (ent > dir)	1675	while (oth > dir)
1067	{	1676	{
1068	if (dir->type == DT_DIR)	1677	if (dir->type == EIO_DT_DIR)
1069	++dir;	1678	++dir;
1070	else
1071	{
1072	--ent;
1073
1074	if (ent->type == DT_DIR)	1679	else if ((--oth)->type == EIO_DT_DIR)
1075	{	1680	{
1076	eio_dirent tmp = *dir;	1681	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1077	*dir = *ent;
1078	*ent = tmp;
1079		1682
1080	++dir;	1683	++dir;
1081	}	1684	}
1082	}	1685	}
		1686
		1687	/* now sort the dirs only (dirs all have the same score) */
		1688	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1083	}	1689	}
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		1690
1099	break;	1691	break;
1100	}	1692	}
1101		1693
1102	/* now add the entry to our list(s) */	1694	/* now add the entry to our list(s) */
…		…
1105	/* skip . and .. entries */	1697	/* skip . and .. entries */
1106	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1698	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1107	{	1699	{
1108	int len = D_NAMLEN (entp) + 1;	1700	int len = D_NAMLEN (entp) + 1;
1109		1701
1110	while (expect_false (namesoffs + len > namesalloc))	1702	while (ecb_expect_false (namesoffs + len > namesalloc))
1111	{	1703	{
1112	namesalloc *= 2;	1704	namesalloc *= 2;
1113	X_LOCK (wrklock);	1705	X_LOCK (wrklock);
1114	req->ptr1 = names = realloc (names, namesalloc);	1706	req->ptr2 = names = realloc (names, namesalloc);
1115	X_UNLOCK (wrklock);	1707	X_UNLOCK (wrklock);
1116		1708
1117	if (!names)	1709	if (!names)
1118	break;	1710	break;
1119	}	1711	}
…		…
1122		1714
1123	if (dents)	1715	if (dents)
1124	{	1716	{
1125	struct eio_dirent *ent;	1717	struct eio_dirent *ent;
1126		1718
1127	if (expect_false (dentoffs == dentalloc))	1719	if (ecb_expect_false (dentoffs == dentalloc))
1128	{	1720	{
1129	dentalloc *= 2;	1721	dentalloc *= 2;
1130	X_LOCK (wrklock);	1722	X_LOCK (wrklock);
1131	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1723	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1132	X_UNLOCK (wrklock);	1724	X_UNLOCK (wrklock);
1133		1725
1134	if (!dents)	1726	if (!dents)
1135	break;	1727	break;
1136	}	1728	}
1137		1729
1138	ent = dents + dentoffs;	1730	ent = dents + dentoffs;
1139		1731
1140	ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */	1732	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1141	ent->namelen = len - 1;	1733	ent->namelen = len - 1;
1142	ent->inode = D_INO (entp);	1734	ent->inode = D_INO (entp);
		1735
		1736	inode_bits |= ent->inode;
1143		1737
1144	switch (D_TYPE (entp))	1738	switch (D_TYPE (entp))
1145	{	1739	{
1146	default:	1740	default:
1147	ent->type = EIO_DT_UNKNOWN;	1741	ent->type = EIO_DT_UNKNOWN;
…		…
1190	#ifdef DT_WHT	1784	#ifdef DT_WHT
1191	case DT_WHT: ent->type = EIO_DT_WHT; break;	1785	case DT_WHT: ent->type = EIO_DT_WHT; break;
1192	#endif	1786	#endif
1193	}	1787	}
1194		1788
1195	ent->score = 0;	1789	ent->score = 7;
1196		1790
1197	if (flags & EIO_READDIR_DIRS_FIRST)	1791	if (flags & EIO_READDIR_DIRS_FIRST)
1198	{	1792	{
1199	if (ent->type == EIO_DT_UNKNOWN)	1793	if (ent->type == EIO_DT_UNKNOWN)
1200	{	1794	{
1201	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */	1795	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1202	ent->score = 98;	1796	ent->score = 1;
1203	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */	1797	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1204	ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */	1798	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1205	}	1799	}
1206	else if (ent->type == EIO_DT_DIR)	1800	else if (ent->type == EIO_DT_DIR)
1207	ent->score = 100;	1801	ent->score = 0;
1208	}	1802	}
1209	}	1803	}
1210		1804
1211	namesoffs += len;	1805	namesoffs += len;
1212	++dentoffs;	1806	++dentoffs;
1213	}	1807	}
		1808
		1809	if (EIO_CANCELLED (req))
		1810	{
		1811	errno = ECANCELED;
		1812	break;
		1813	}
1214	}	1814	}
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	}	1815	}
1248		1816
1249	/*****************************************************************************/	1817	/*****************************************************************************/
1250		1818
1251	#define ALLOC(len) \	1819	#define ALLOC(len) \
…		…
1266	X_THREAD_PROC (etp_proc)	1834	X_THREAD_PROC (etp_proc)
1267	{	1835	{
1268	ETP_REQ *req;	1836	ETP_REQ *req;
1269	struct timespec ts;	1837	struct timespec ts;
1270	etp_worker *self = (etp_worker *)thr_arg;	1838	etp_worker *self = (etp_worker *)thr_arg;
		1839	int timeout;
1271		1840
1272	/* try to distribute timeouts somewhat randomly */	1841	/* try to distribute timeouts somewhat evenly */
1273	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1842	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1274		1843
1275	for (;;)	1844	for (;;)
1276	{	1845	{
		1846	ts.tv_sec = 0;
		1847
1277	X_LOCK (reqlock);	1848	X_LOCK (reqlock);
1278		1849
1279	for (;;)	1850	for (;;)
1280	{	1851	{
1281	self->req = req = reqq_shift (&req_queue);	1852	self->req = req = reqq_shift (&req_queue);
1282		1853
1283	if (req)	1854	if (req)
1284	break;	1855	break;
1285		1856
		1857	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1858	{
		1859	X_UNLOCK (reqlock);
		1860	X_LOCK (wrklock);
		1861	--started;
		1862	X_UNLOCK (wrklock);
		1863	goto quit;
		1864	}
		1865
1286	++idle;	1866	++idle;
1287		1867
1288	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1868	if (idle <= max_idle)
1289	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1869	/* we are allowed to idle, so do so without any timeout */
		1870	X_COND_WAIT (reqwait, reqlock);
		1871	else
1290	{	1872	{
1291	if (idle > max_idle)	1873	/* initialise timeout once */
1292	{	1874	if (!ts.tv_sec)
1293	--idle;	1875	ts.tv_sec = time (0) + idle_timeout;
1294	X_UNLOCK (reqlock);
1295	X_LOCK (wrklock);
1296	--started;
1297	X_UNLOCK (wrklock);
1298	goto quit;
1299	}
1300		1876
1301	/* we are allowed to idle, so do so without any timeout */
1302	X_COND_WAIT (reqwait, reqlock);	1877	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1878	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1303	}	1879	}
1304		1880
1305	--idle;	1881	--idle;
1306	}	1882	}
1307		1883
…		…
1310	X_UNLOCK (reqlock);	1886	X_UNLOCK (reqlock);
1311		1887
1312	if (req->type < 0)	1888	if (req->type < 0)
1313	goto quit;	1889	goto quit;
1314		1890
1315	if (!EIO_CANCELLED (req))
1316	ETP_EXECUTE (self, req);	1891	ETP_EXECUTE (self, req);
1317		1892
1318	X_LOCK (reslock);	1893	X_LOCK (reslock);
1319		1894
1320	++npending;	1895	++npending;
1321		1896
…		…
1336	return 0;	1911	return 0;
1337	}	1912	}
1338		1913
1339	/*****************************************************************************/	1914	/*****************************************************************************/
1340		1915
		1916	int ecb_cold
1341	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1917	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1342	{	1918	{
1343	return etp_init (want_poll, done_poll);	1919	return etp_init (want_poll, done_poll);
1344	}	1920	}
1345		1921
		1922	ecb_inline void
1346	static void eio_api_destroy (eio_req *req)	1923	eio_api_destroy (eio_req *req)
1347	{	1924	{
1348	free (req);	1925	free (req);
1349	}	1926	}
1350		1927
1351	#define REQ(rtype) \	1928	#define REQ(rtype) \
…		…
1370	{ \	1947	{ \
1371	eio_api_destroy (req); \	1948	eio_api_destroy (req); \
1372	return 0; \	1949	return 0; \
1373	}	1950	}
1374		1951
		1952	static void
1375	static void eio_execute (etp_worker *self, eio_req *req)	1953	eio_execute (etp_worker *self, eio_req *req)
1376	{	1954	{
1377	errno = 0;	1955	if (ecb_expect_false (EIO_CANCELLED (req)))
		1956	{
		1957	req->result = -1;
		1958	req->errorno = ECANCELED;
		1959	return;
		1960	}
1378		1961
1379	switch (req->type)	1962	switch (req->type)
1380	{	1963	{
1381	case EIO_READ: ALLOC (req->size);	1964	case EIO_READ: ALLOC (req->size);
1382	req->result = req->offs >= 0	1965	req->result = req->offs >= 0
…		…
1394	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1977	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1395	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1978	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1396	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1979	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1397	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1980	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1398		1981
		1982	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1983	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1984	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1985	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1986
1399	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	1987	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;	1988	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;	1989	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;	1990	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;	1991	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
…		…
1410	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	1998	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1411	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	1999	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;	2000	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1413	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2001	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1414	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2002	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;	2003	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1416		2004
		2005	case EIO_REALPATH: eio__realpath (req, self); break;
		2006
1417	case EIO_READLINK: ALLOC (NAME_MAX);	2007	case EIO_READLINK: ALLOC (PATH_MAX);
1418	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	2008	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1419		2009
1420	case EIO_SYNC: req->result = 0; sync (); break;	2010	case EIO_SYNC: req->result = 0; sync (); break;
1421	case EIO_FSYNC: req->result = fsync (req->int1); break;	2011	case EIO_FSYNC: req->result = fsync (req->int1); break;
1422	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2012	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1423	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;	2013	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2014	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1424	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	2015	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2016	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;	2017	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2018	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1426		2019
1427	case EIO_READDIR: eio__scandir (req, self); break;	2020	case EIO_READDIR: eio__scandir (req, self); break;
1428		2021
1429	case EIO_BUSY:	2022	case EIO_BUSY:
1430	#ifdef _WIN32	2023	#ifdef _WIN32
1431	Sleep (req->nv1 * 1000.);	2024	Sleep (req->nv1 * 1e3);
1432	#else	2025	#else
1433	{	2026	{
1434	struct timeval tv;	2027	struct timeval tv;
1435		2028
1436	tv.tv_sec = req->nv1;	2029	tv.tv_sec = req->nv1;
1437	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	2030	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1438		2031
1439	req->result = select (0, 0, 0, 0, &tv);	2032	req->result = select (0, 0, 0, 0, &tv);
1440	}	2033	}
1441	#endif	2034	#endif
1442	break;	2035	break;
…		…
1457	times = tv;	2050	times = tv;
1458	}	2051	}
1459	else	2052	else
1460	times = 0;	2053	times = 0;
1461		2054
1462
1463	req->result = req->type == EIO_FUTIME	2055	req->result = req->type == EIO_FUTIME
1464	? futimes (req->int1, times)	2056	? futimes (req->int1, times)
1465	: utimes (req->ptr1, times);	2057	: utimes (req->ptr1, times);
1466	}	2058	}
1467	break;	2059	break;
…		…
1472	case EIO_NOP:	2064	case EIO_NOP:
1473	req->result = 0;	2065	req->result = 0;
1474	break;	2066	break;
1475		2067
1476	case EIO_CUSTOM:	2068	case EIO_CUSTOM:
1477	((void (*)(eio_req *))req->feed) (req);	2069	req->feed (req);
1478	break;	2070	break;
1479		2071
1480	default:	2072	default:
		2073	errno = ENOSYS;
1481	req->result = -1;	2074	req->result = -1;
1482	break;	2075	break;
1483	}	2076	}
1484		2077
1485	req->errorno = errno;	2078	req->errorno = errno;
…		…
1515	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2108	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1516	{	2109	{
1517	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2110	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1518	}	2111	}
1519		2112
		2113	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2114	{
		2115	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2116	}
		2117
		2118	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2119	{
		2120	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2121	}
		2122
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)	2123	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	{	2124	{
1522	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2125	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1523	}	2126	}
1524		2127
		2128	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2129	{
		2130	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2131	}
		2132
1525	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2133	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1526	{	2134	{
1527	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2135	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1528	}	2136	}
1529		2137
…		…
1548	}	2156	}
1549		2157
1550	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2158	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1551	{	2159	{
1552	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2160	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		2161	}
		2162
		2163	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2164	{
		2165	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1553	}	2166	}
1554		2167
1555	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2168	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1556	{	2169	{
1557	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2170	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)	2234	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1622	{	2235	{
1623	return eio__1path (EIO_READLINK, path, pri, cb, data);	2236	return eio__1path (EIO_READLINK, path, pri, cb, data);
1624	}	2237	}
1625		2238
		2239	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2240	{
		2241	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2242	}
		2243
1626	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2244	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1627	{	2245	{
1628	return eio__1path (EIO_STAT, path, pri, cb, data);	2246	return eio__1path (EIO_STAT, path, pri, cb, data);
1629	}	2247	}
1630		2248
1631	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2249	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1632	{	2250	{
1633	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2251	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1634	}	2252	}
1635		2253
		2254	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2255	{
		2256	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2257	}
		2258
1636	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2259	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1637	{	2260	{
1638	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2261	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1639	}	2262	}
1640		2263
…		…
1648	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2271	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1649	}	2272	}
1650		2273
1651	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2274	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1652	{	2275	{
1653	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2276	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1654	}	2277	}
1655		2278
1656	static eio_req *	2279	static eio_req *
1657	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2280	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1658	{	2281	{
…		…
1682	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2305	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1683	{	2306	{
1684	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2307	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1685	}	2308	}
1686		2309
1687	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2310	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1688	{	2311	{
1689	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2312	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1690	}	2313	}
1691		2314
1692	#endif	2315	#endif
1693		2316
1694	eio_req *eio_grp (eio_cb cb, void *data)	2317	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1703	#undef SEND	2326	#undef SEND
1704		2327
1705	/*****************************************************************************/	2328	/*****************************************************************************/
1706	/* grp functions */	2329	/* grp functions */
1707		2330
		2331	void
1708	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2332	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1709	{	2333	{
1710	grp->int2 = limit;	2334	grp->int2 = limit;
1711	grp->feed = feed;	2335	grp->feed = feed;
1712		2336
1713	grp_try_feed (grp);	2337	grp_try_feed (grp);
1714	}	2338	}
1715		2339
		2340	void
1716	void eio_grp_limit (eio_req *grp, int limit)	2341	eio_grp_limit (eio_req *grp, int limit)
1717	{	2342	{
1718	grp->int2 = limit;	2343	grp->int2 = limit;
1719		2344
1720	grp_try_feed (grp);	2345	grp_try_feed (grp);
1721	}	2346	}
1722		2347
		2348	void
1723	void eio_grp_add (eio_req *grp, eio_req *req)	2349	eio_grp_add (eio_req *grp, eio_req *req)
1724	{	2350	{
1725	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2351	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1726		2352
1727	grp->flags |= EIO_FLAG_GROUPADD;	2353	grp->flags |= EIO_FLAG_GROUPADD;
1728		2354
…		…
1739	}	2365	}
1740		2366
1741	/*****************************************************************************/	2367	/*****************************************************************************/
1742	/* misc garbage */	2368	/* misc garbage */
1743		2369
		2370	ssize_t
1744	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2371	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1745	{	2372	{
1746	etp_worker wrk;	2373	etp_worker wrk;
		2374	ssize_t ret;
1747		2375
1748	wrk.dbuf = 0;	2376	wrk.dbuf = 0;
1749		2377
1750	eio__sendfile (ofd, ifd, offset, count, &wrk);	2378	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1751		2379
1752	if (wrk.dbuf)	2380	if (wrk.dbuf)
1753	free (wrk.dbuf);	2381	free (wrk.dbuf);
1754	}
1755		2382
		2383	return ret;
		2384	}
		2385

Diff Legend

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