ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/cvsroot/libeio/eio.c

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.19 by root, Tue Jun 17 23:34:20 2008 UTC vs.
Revision 1.87 by root, Thu Jul 14 19:19:22 2011 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008 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"
		46
		47	#ifdef EIO_STACKSIZE
		48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
		49	#endif
41	#include "xthread.h"	50	#include "xthread.h"
42		51
43	#include <errno.h>	52	#include <errno.h>
44	#include <stddef.h>	53	#include <stddef.h>
45	#include <stdlib.h>	54	#include <stdlib.h>
…		…
49	#include <sys/stat.h>	58	#include <sys/stat.h>
50	#include <limits.h>	59	#include <limits.h>
51	#include <fcntl.h>	60	#include <fcntl.h>
52	#include <assert.h>	61	#include <assert.h>
53		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
54	#ifndef EIO_FINISH	76	#ifndef EIO_FINISH
55	# 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
56	#endif	78	#endif
57		79
58	#ifndef EIO_DESTROY	80	#ifndef EIO_DESTROY
…		…
61		83
62	#ifndef EIO_FEED	84	#ifndef EIO_FEED
63	# 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)
64	#endif	86	#endif
65		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
66	#ifdef _WIN32	99	#ifdef _WIN32
67		100
68	/*doh*/	101	#define PAGESIZE 4096 /* GetSystemInfo? */
		102
		103	#ifdef EIO_STRUCT_STATI64
		104	#define stat(path,buf) _stati64 (path,buf)
		105	#define fstat(fd,buf) _fstati64 (path,buf)
		106	#endif
		107	#define lstat(path,buf) stat (path,buf)
		108	#define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		109	#define mkdir(path,mode) _mkdir (path)
		110	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		111
		112	#define chown(path,uid,gid) EIO_ENOSYS ()
		113	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		114	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		115	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		116	#define mknod(path,mode,dev) EIO_ENOSYS ()
		117	#define sync() EIO_ENOSYS ()
		118
		119	/* we could even stat and see if it exists */
		120	static int
		121	symlink (const char *old, const char *neu)
		122	{
		123	if (CreateSymbolicLink (neu, old, 1))
		124	return 0;
		125
		126	if (CreateSymbolicLink (neu, old, 0))
		127	return 0;
		128
		129	return EIO_ERRNO (ENOENT, -1);
		130	}
69		131
70	#else	132	#else
71		133
72	# include "config.h"
73	# include <sys/time.h>	134	#include <sys/time.h>
74	# include <sys/select.h>	135	#include <sys/select.h>
		136	#include <sys/statvfs.h>
75	# include <unistd.h>	137	#include <unistd.h>
76	# include <utime.h>	138	#include <utime.h>
77	# include <signal.h>	139	#include <signal.h>
78	# include <dirent.h>	140	#include <dirent.h>
79		141
		142	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		143	#include <sys/mman.h>
		144	#endif
		145
		146	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		147	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		148	#define _DIRENT_HAVE_D_TYPE /* sigh */
		149	#define D_INO(de) (de)->d_fileno
		150	#define D_NAMLEN(de) (de)->d_namlen
		151	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		152	#define D_INO(de) (de)->d_ino
		153	#endif
		154
		155	#ifdef _D_EXACT_NAMLEN
		156	#undef D_NAMLEN
		157	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		158	#endif
		159
		160	#ifdef _DIRENT_HAVE_D_TYPE
		161	#define D_TYPE(de) (de)->d_type
		162	#endif
		163
80	# ifndef EIO_STRUCT_DIRENT	164	#ifndef EIO_STRUCT_DIRENT
81	# define EIO_STRUCT_DIRENT struct dirent	165	#define EIO_STRUCT_DIRENT struct dirent
82	# endif	166	#endif
83		167
84	#endif	168	#endif
85		169
86	#if HAVE_SENDFILE	170	#if HAVE_SENDFILE
87	# if __linux	171	# if __linux
88	# include <sys/sendfile.h>	172	# include <sys/sendfile.h>
89	# elif __freebsd	173	# elif __FreeBSD__ || defined __APPLE__
90	# include <sys/socket.h>	174	# include <sys/socket.h>
91	# include <sys/uio.h>	175	# include <sys/uio.h>
92	# elif __hpux	176	# elif __hpux
93	# include <sys/socket.h>	177	# include <sys/socket.h>
94	# elif __solaris /* not yet */	178	# elif __solaris
95	# include <sys/sendfile.h>	179	# include <sys/sendfile.h>
96	# else	180	# else
97	# error sendfile support requested but not available	181	# error sendfile support requested but not available
98	# endif	182	# endif
99	#endif	183	#endif
100		184
101	/* number of seconds after which an idle threads exit */	185	#ifndef D_TYPE
102	#define IDLE_TIMEOUT 10	186	# define D_TYPE(de) 0
		187	#endif
		188	#ifndef D_INO
		189	# define D_INO(de) 0
		190	#endif
		191	#ifndef D_NAMLEN
		192	# define D_NAMLEN(de) strlen ((de)->d_name)
		193	#endif
103		194
104	/* used for struct dirent, AIX doesn't provide it */	195	/* used for struct dirent, AIX doesn't provide it */
105	#ifndef NAME_MAX	196	#ifndef NAME_MAX
106	# define NAME_MAX 4096	197	# define NAME_MAX 4096
		198	#endif
		199
		200	/* used for readlink etc. */
		201	#ifndef PATH_MAX
		202	# define PATH_MAX 4096
107	#endif	203	#endif
108		204
109	/* buffer size for various temporary buffers */	205	/* buffer size for various temporary buffers */
110	#define EIO_BUFSIZE 65536	206	#define EIO_BUFSIZE 65536
111		207
…		…
117	errno = ENOMEM; \	213	errno = ENOMEM; \
118	if (!eio_buf) \	214	if (!eio_buf) \
119	return -1;	215	return -1;
120		216
121	#define EIO_TICKS ((1000000 + 1023) >> 10)	217	#define EIO_TICKS ((1000000 + 1023) >> 10)
122
123	/*****************************************************************************/
124		218
125	#define ETP_PRI_MIN EIO_PRI_MIN	219	#define ETP_PRI_MIN EIO_PRI_MIN
126	#define ETP_PRI_MAX EIO_PRI_MAX	220	#define ETP_PRI_MAX EIO_PRI_MAX
127		221
128	struct etp_worker;	222	struct etp_worker;
…		…
144	if (wrk->dirp) \	238	if (wrk->dirp) \
145	{ \	239	{ \
146	closedir (wrk->dirp); \	240	closedir (wrk->dirp); \
147	wrk->dirp = 0; \	241	wrk->dirp = 0; \
148	}	242	}
		243
149	#define ETP_WORKER_COMMON \	244	#define ETP_WORKER_COMMON \
150	void *dbuf; \	245	void *dbuf; \
151	DIR *dirp;	246	DIR *dirp;
152		247
153	/*****************************************************************************/	248	/*****************************************************************************/
154		249
155	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	250	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
156		251
157	/* calculcate time difference in ~1/EIO_TICKS of a second */	252	/* calculate time difference in ~1/EIO_TICKS of a second */
		253	ecb_inline int
158	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	254	tvdiff (struct timeval *tv1, struct timeval *tv2)
159	{	255	{
160	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	256	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
161	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	257	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
162	}	258	}
163		259
…		…
170	static unsigned int max_poll_reqs; /* reslock */	266	static unsigned int max_poll_reqs; /* reslock */
171		267
172	static volatile unsigned int nreqs; /* reqlock */	268	static volatile unsigned int nreqs; /* reqlock */
173	static volatile unsigned int nready; /* reqlock */	269	static volatile unsigned int nready; /* reqlock */
174	static volatile unsigned int npending; /* reqlock */	270	static volatile unsigned int npending; /* reqlock */
175	static volatile unsigned int max_idle = 4;	271	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		272	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
176		273
		274	static xmutex_t wrklock;
		275	static xmutex_t reslock;
		276	static xmutex_t reqlock;
		277	static xcond_t reqwait;
		278
		279	#if !HAVE_PREADWRITE
		280	/*
		281	* make our pread/pwrite emulation safe against themselves, but not against
		282	* normal read/write by using a mutex. slows down execution a lot,
		283	* but that's your problem, not mine.
		284	*/
177	static mutex_t wrklock = X_MUTEX_INIT;	285	static xmutex_t preadwritelock = X_MUTEX_INIT;
178	static mutex_t reslock = X_MUTEX_INIT;	286	#endif
179	static mutex_t reqlock = X_MUTEX_INIT;
180	static cond_t reqwait = X_COND_INIT;
181		287
182	typedef struct etp_worker	288	typedef struct etp_worker
183	{	289	{
184	/* locked by wrklock */	290	/* locked by wrklock */
185	struct etp_worker *prev, *next;	291	struct etp_worker *prev, *next;
186		292
187	thread_t tid;	293	xthread_t tid;
188		294
189	/* locked by reslock, reqlock or wrklock */	295	/* locked by reslock, reqlock or wrklock */
190	ETP_REQ *req; /* currently processed request */	296	ETP_REQ *req; /* currently processed request */
191		297
192	ETP_WORKER_COMMON	298	ETP_WORKER_COMMON
…		…
197	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	303	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
198	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	304	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
199		305
200	/* worker threads management */	306	/* worker threads management */
201		307
		308	static void ecb_cold
202	static void etp_worker_clear (etp_worker *wrk)	309	etp_worker_clear (etp_worker *wrk)
203	{	310	{
204	ETP_WORKER_CLEAR (wrk);	311	ETP_WORKER_CLEAR (wrk);
205	}	312	}
206		313
		314	static void ecb_cold
207	static void etp_worker_free (etp_worker *wrk)	315	etp_worker_free (etp_worker *wrk)
208	{	316	{
209	wrk->next->prev = wrk->prev;	317	wrk->next->prev = wrk->prev;
210	wrk->prev->next = wrk->next;	318	wrk->prev->next = wrk->next;
211		319
212	free (wrk);	320	free (wrk);
213	}	321	}
214		322
215	static unsigned int etp_nreqs (void)	323	static unsigned int
		324	etp_nreqs (void)
216	{	325	{
217	int retval;	326	int retval;
218	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	327	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
219	retval = nreqs;	328	retval = nreqs;
220	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	329	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
221	return retval;	330	return retval;
222	}	331	}
223		332
224	static unsigned int etp_nready (void)	333	static unsigned int
		334	etp_nready (void)
225	{	335	{
226	unsigned int retval;	336	unsigned int retval;
227		337
228	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	338	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
229	retval = nready;	339	retval = nready;
230	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	340	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
231		341
232	return retval;	342	return retval;
233	}	343	}
234		344
235	static unsigned int etp_npending (void)	345	static unsigned int
		346	etp_npending (void)
236	{	347	{
237	unsigned int retval;	348	unsigned int retval;
238		349
239	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	350	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
240	retval = npending;	351	retval = npending;
241	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	352	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
242		353
243	return retval;	354	return retval;
244	}	355	}
245		356
246	static unsigned int etp_nthreads (void)	357	static unsigned int
		358	etp_nthreads (void)
247	{	359	{
248	unsigned int retval;	360	unsigned int retval;
249		361
250	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	362	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
251	retval = started;	363	retval = started;
…		…
265	} etp_reqq;	377	} etp_reqq;
266		378
267	static etp_reqq req_queue;	379	static etp_reqq req_queue;
268	static etp_reqq res_queue;	380	static etp_reqq res_queue;
269		381
		382	static int ecb_noinline
270	static int reqq_push (etp_reqq *q, ETP_REQ *req)	383	reqq_push (etp_reqq *q, ETP_REQ *req)
271	{	384	{
272	int pri = req->pri;	385	int pri = req->pri;
273	req->next = 0;	386	req->next = 0;
274		387
275	if (q->qe[pri])	388	if (q->qe[pri])
…		…
281	q->qe[pri] = q->qs[pri] = req;	394	q->qe[pri] = q->qs[pri] = req;
282		395
283	return q->size++;	396	return q->size++;
284	}	397	}
285		398
		399	static ETP_REQ * ecb_noinline
286	static ETP_REQ *reqq_shift (etp_reqq *q)	400	reqq_shift (etp_reqq *q)
287	{	401	{
288	int pri;	402	int pri;
289		403
290	if (!q->size)	404	if (!q->size)
291	return 0;	405	return 0;
…		…
306	}	420	}
307		421
308	abort ();	422	abort ();
309	}	423	}
310		424
311	static void etp_atfork_prepare (void)	425	static void ecb_cold
		426	etp_thread_init (void)
312	{	427	{
313	X_LOCK (wrklock);
314	X_LOCK (reqlock);
315	X_LOCK (reslock);
316	#if !HAVE_PREADWRITE	428	#if !HAVE_PREADWRITE
317	X_LOCK (preadwritelock);	429	X_MUTEX_CREATE (preadwritelock);
318	#endif	430	#endif
		431	X_MUTEX_CREATE (wrklock);
		432	X_MUTEX_CREATE (reslock);
		433	X_MUTEX_CREATE (reqlock);
		434	X_COND_CREATE (reqwait);
319	}	435	}
320		436
		437	static void ecb_cold
		438	etp_atfork_prepare (void)
		439	{
		440	}
		441
		442	static void ecb_cold
321	static void etp_atfork_parent (void)	443	etp_atfork_parent (void)
322	{	444	{
323	#if !HAVE_PREADWRITE
324	X_UNLOCK (preadwritelock);
325	#endif
326	X_UNLOCK (reslock);
327	X_UNLOCK (reqlock);
328	X_UNLOCK (wrklock);
329	}	445	}
330		446
		447	static void ecb_cold
331	static void etp_atfork_child (void)	448	etp_atfork_child (void)
332	{	449	{
333	ETP_REQ *prv;	450	ETP_REQ *prv;
334		451
335	while (prv = reqq_shift (&req_queue))	452	while ((prv = reqq_shift (&req_queue)))
336	ETP_DESTROY (prv);	453	ETP_DESTROY (prv);
337		454
338	while (prv = reqq_shift (&res_queue))	455	while ((prv = reqq_shift (&res_queue)))
339	ETP_DESTROY (prv);	456	ETP_DESTROY (prv);
340		457
341	while (wrk_first.next != &wrk_first)	458	while (wrk_first.next != &wrk_first)
342	{	459	{
343	etp_worker *wrk = wrk_first.next;	460	etp_worker *wrk = wrk_first.next;
…		…
353	idle = 0;	470	idle = 0;
354	nreqs = 0;	471	nreqs = 0;
355	nready = 0;	472	nready = 0;
356	npending = 0;	473	npending = 0;
357		474
358	etp_atfork_parent ();	475	etp_thread_init ();
359	}	476	}
360		477
361	static void	478	static void ecb_cold
362	etp_once_init (void)	479	etp_once_init (void)
363	{	480	{
		481	etp_thread_init ();
364	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);	482	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
365	}	483	}
366		484
367	static int	485	static int ecb_cold
368	etp_init (void (*want_poll)(void), void (*done_poll)(void))	486	etp_init (void (*want_poll)(void), void (*done_poll)(void))
369	{	487	{
370	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	488	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
371		489
372	pthread_once (&doinit, etp_once_init);	490	pthread_once (&doinit, etp_once_init);
…		…
377	return 0;	495	return 0;
378	}	496	}
379		497
380	X_THREAD_PROC (etp_proc);	498	X_THREAD_PROC (etp_proc);
381		499
		500	static void ecb_cold
382	static void etp_start_thread (void)	501	etp_start_thread (void)
383	{	502	{
384	etp_worker *wrk = calloc (1, sizeof (etp_worker));	503	etp_worker *wrk = calloc (1, sizeof (etp_worker));
385		504
386	/*TODO*/	505	/*TODO*/
387	assert (("unable to allocate worker thread data", wrk));	506	assert (("unable to allocate worker thread data", wrk));
…		…
400	free (wrk);	519	free (wrk);
401		520
402	X_UNLOCK (wrklock);	521	X_UNLOCK (wrklock);
403	}	522	}
404		523
		524	static void
405	static void etp_maybe_start_thread (void)	525	etp_maybe_start_thread (void)
406	{	526	{
407	if (etp_nthreads () >= wanted)	527	if (ecb_expect_true (etp_nthreads () >= wanted))
408	return;	528	return;
409		529
410	/* todo: maybe use idle here, but might be less exact */	530	/* todo: maybe use idle here, but might be less exact */
411	if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())	531	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
412	return;	532	return;
413		533
414	etp_start_thread ();	534	etp_start_thread ();
415	}	535	}
416		536
		537	static void ecb_cold
417	static void etp_end_thread (void)	538	etp_end_thread (void)
418	{	539	{
419	eio_req *req = calloc (1, sizeof (eio_req));	540	eio_req *req = calloc (1, sizeof (eio_req));
420		541
421	req->type = -1;	542	req->type = -1;
422	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	543	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
429	X_LOCK (wrklock);	550	X_LOCK (wrklock);
430	--started;	551	--started;
431	X_UNLOCK (wrklock);	552	X_UNLOCK (wrklock);
432	}	553	}
433		554
434	static int etp_poll (void)	555	static int
		556	etp_poll (void)
435	{	557	{
436	unsigned int maxreqs;	558	unsigned int maxreqs;
437	unsigned int maxtime;	559	unsigned int maxtime;
438	struct timeval tv_start, tv_now;	560	struct timeval tv_start, tv_now;
439		561
…		…
469		591
470	X_LOCK (reqlock);	592	X_LOCK (reqlock);
471	--nreqs;	593	--nreqs;
472	X_UNLOCK (reqlock);	594	X_UNLOCK (reqlock);
473		595
474	if (req->type == EIO_GROUP && req->size)	596	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
475	{	597	{
476	req->int1 = 1; /* mark request as delayed */	598	req->int1 = 1; /* mark request as delayed */
477	continue;	599	continue;
478	}	600	}
479	else	601	else
480	{	602	{
481	int res = ETP_FINISH (req);	603	int res = ETP_FINISH (req);
482	if (res)	604	if (ecb_expect_false (res))
483	return res;	605	return res;
484	}	606	}
485		607
486	if (maxreqs && !--maxreqs)	608	if (ecb_expect_false (maxreqs && !--maxreqs))
487	break;	609	break;
488		610
489	if (maxtime)	611	if (maxtime)
490	{	612	{
491	gettimeofday (&tv_now, 0);	613	gettimeofday (&tv_now, 0);
…		…
497		619
498	errno = EAGAIN;	620	errno = EAGAIN;
499	return -1;	621	return -1;
500	}	622	}
501		623
		624	static void
502	static void etp_cancel (ETP_REQ *req)	625	etp_cancel (ETP_REQ *req)
503	{	626	{
504	X_LOCK (wrklock);	627	req->cancelled = 1;
505	req->flags |= EIO_FLAG_CANCELLED;
506	X_UNLOCK (wrklock);
507		628
508	eio_grp_cancel (req);	629	eio_grp_cancel (req);
509	}	630	}
510		631
		632	static void
511	static void etp_submit (ETP_REQ *req)	633	etp_submit (ETP_REQ *req)
512	{	634	{
513	req->pri -= ETP_PRI_MIN;	635	req->pri -= ETP_PRI_MIN;
514		636
515	if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	637	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
516	if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	638	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
517		639
		640	if (ecb_expect_false (req->type == EIO_GROUP))
		641	{
		642	/* I hope this is worth it :/ */
518	X_LOCK (reqlock);	643	X_LOCK (reqlock);
519	++nreqs;	644	++nreqs;
		645	X_UNLOCK (reqlock);
		646
		647	X_LOCK (reslock);
		648
		649	++npending;
		650
		651	if (!reqq_push (&res_queue, req) && want_poll_cb)
		652	want_poll_cb ();
		653
		654	X_UNLOCK (reslock);
		655	}
		656	else
		657	{
		658	X_LOCK (reqlock);
		659	++nreqs;
520	++nready;	660	++nready;
521	reqq_push (&req_queue, req);	661	reqq_push (&req_queue, req);
522	X_COND_SIGNAL (reqwait);	662	X_COND_SIGNAL (reqwait);
523	X_UNLOCK (reqlock);	663	X_UNLOCK (reqlock);
524		664
525	etp_maybe_start_thread ();	665	etp_maybe_start_thread ();
		666	}
526	}	667	}
527		668
		669	static void ecb_cold
528	static void etp_set_max_poll_time (double nseconds)	670	etp_set_max_poll_time (double nseconds)
529	{	671	{
530	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	672	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
531	max_poll_time = nseconds;	673	max_poll_time = nseconds * EIO_TICKS;
532	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	674	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
533	}	675	}
534		676
		677	static void ecb_cold
535	static void etp_set_max_poll_reqs (unsigned int maxreqs)	678	etp_set_max_poll_reqs (unsigned int maxreqs)
536	{	679	{
537	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	680	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
538	max_poll_reqs = maxreqs;	681	max_poll_reqs = maxreqs;
539	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	682	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
540	}	683	}
541		684
		685	static void ecb_cold
542	static void etp_set_max_idle (unsigned int nthreads)	686	etp_set_max_idle (unsigned int nthreads)
543	{	687	{
544	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	688	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
545	max_idle = nthreads <= 0 ? 1 : nthreads;	689	max_idle = nthreads;
546	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	690	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
547	}	691	}
548		692
		693	static void ecb_cold
		694	etp_set_idle_timeout (unsigned int seconds)
		695	{
		696	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		697	idle_timeout = seconds;
		698	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		699	}
		700
		701	static void ecb_cold
549	static void etp_set_min_parallel (unsigned int nthreads)	702	etp_set_min_parallel (unsigned int nthreads)
550	{	703	{
551	if (wanted < nthreads)	704	if (wanted < nthreads)
552	wanted = nthreads;	705	wanted = nthreads;
553	}	706	}
554		707
		708	static void ecb_cold
555	static void etp_set_max_parallel (unsigned int nthreads)	709	etp_set_max_parallel (unsigned int nthreads)
556	{	710	{
557	if (wanted > nthreads)	711	if (wanted > nthreads)
558	wanted = nthreads;	712	wanted = nthreads;
559		713
560	while (started > wanted)	714	while (started > wanted)
561	etp_end_thread ();	715	etp_end_thread ();
562	}	716	}
563		717
564	/*****************************************************************************/	718	/*****************************************************************************/
565		719
		720	static void
566	static void grp_try_feed (eio_req *grp)	721	grp_try_feed (eio_req *grp)
567	{	722	{
568	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	723	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
569	{	724	{
570	int old_len = grp->size;	725	grp->flags &= ~EIO_FLAG_GROUPADD;
571		726
572	EIO_FEED (grp);	727	EIO_FEED (grp);
573		728
574	/* stop if no progress has been made */	729	/* stop if no progress has been made */
575	if (old_len == grp->size)	730	if (!(grp->flags & EIO_FLAG_GROUPADD))
576	{	731	{
577	grp->feed = 0;	732	grp->feed = 0;
578	break;	733	break;
579	}	734	}
580	}	735	}
581	}	736	}
582		737
		738	static int
583	static int grp_dec (eio_req *grp)	739	grp_dec (eio_req *grp)
584	{	740	{
585	--grp->size;	741	--grp->size;
586		742
587	/* call feeder, if applicable */	743	/* call feeder, if applicable */
588	grp_try_feed (grp);	744	grp_try_feed (grp);
…		…
592	return eio_finish (grp);	748	return eio_finish (grp);
593	else	749	else
594	return 0;	750	return 0;
595	}	751	}
596		752
		753	static void
597	void eio_destroy (eio_req *req)	754	eio_destroy (eio_req *req)
598	{	755	{
599	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	756	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
600	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	757	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
601		758
602	EIO_DESTROY (req);	759	EIO_DESTROY (req);
603	}	760	}
604		761
		762	static int
605	static int eio_finish (eio_req *req)	763	eio_finish (eio_req *req)
606	{	764	{
607	int res = EIO_FINISH (req);	765	int res = EIO_FINISH (req);
608		766
609	if (req->grp)	767	if (req->grp)
610	{	768	{
…		…
618	if (grp->grp_first == req)	776	if (grp->grp_first == req)
619	grp->grp_first = req->grp_next;	777	grp->grp_first = req->grp_next;
620		778
621	res2 = grp_dec (grp);	779	res2 = grp_dec (grp);
622		780
623	if (!res && res2)	781	if (!res)
624	res = res2;	782	res = res2;
625	}	783	}
626		784
627	eio_destroy (req);	785	eio_destroy (req);
628		786
629	return res;	787	return res;
630	}	788	}
631		789
		790	void
632	void eio_grp_cancel (eio_req *grp)	791	eio_grp_cancel (eio_req *grp)
633	{	792	{
634	for (grp = grp->grp_first; grp; grp = grp->grp_next)	793	for (grp = grp->grp_first; grp; grp = grp->grp_next)
635	eio_cancel (grp);	794	eio_cancel (grp);
636	}	795	}
637		796
		797	void
638	void eio_cancel (eio_req *req)	798	eio_cancel (eio_req *req)
639	{	799	{
640	etp_cancel (req);	800	etp_cancel (req);
641	}	801	}
642		802
		803	void
643	void eio_submit (eio_req *req)	804	eio_submit (eio_req *req)
644	{	805	{
645	etp_submit (req);	806	etp_submit (req);
646	}	807	}
647		808
648	unsigned int eio_nreqs (void)	809	unsigned int
		810	eio_nreqs (void)
649	{	811	{
650	return etp_nreqs ();	812	return etp_nreqs ();
651	}	813	}
652		814
653	unsigned int eio_nready (void)	815	unsigned int
		816	eio_nready (void)
654	{	817	{
655	return etp_nready ();	818	return etp_nready ();
656	}	819	}
657		820
658	unsigned int eio_npending (void)	821	unsigned int
		822	eio_npending (void)
659	{	823	{
660	return etp_npending ();	824	return etp_npending ();
661	}	825	}
662		826
663	unsigned int eio_nthreads (void)	827	unsigned int ecb_cold
		828	eio_nthreads (void)
664	{	829	{
665	return etp_nthreads ();	830	return etp_nthreads ();
666	}	831	}
667		832
		833	void ecb_cold
668	void eio_set_max_poll_time (double nseconds)	834	eio_set_max_poll_time (double nseconds)
669	{	835	{
670	etp_set_max_poll_time (nseconds);	836	etp_set_max_poll_time (nseconds);
671	}	837	}
672		838
		839	void ecb_cold
673	void eio_set_max_poll_reqs (unsigned int maxreqs)	840	eio_set_max_poll_reqs (unsigned int maxreqs)
674	{	841	{
675	etp_set_max_poll_reqs (maxreqs);	842	etp_set_max_poll_reqs (maxreqs);
676	}	843	}
677		844
		845	void ecb_cold
678	void eio_set_max_idle (unsigned int nthreads)	846	eio_set_max_idle (unsigned int nthreads)
679	{	847	{
680	etp_set_max_idle (nthreads);	848	etp_set_max_idle (nthreads);
681	}	849	}
682		850
		851	void ecb_cold
		852	eio_set_idle_timeout (unsigned int seconds)
		853	{
		854	etp_set_idle_timeout (seconds);
		855	}
		856
		857	void ecb_cold
683	void eio_set_min_parallel (unsigned int nthreads)	858	eio_set_min_parallel (unsigned int nthreads)
684	{	859	{
685	etp_set_min_parallel (nthreads);	860	etp_set_min_parallel (nthreads);
686	}	861	}
687		862
		863	void ecb_cold
688	void eio_set_max_parallel (unsigned int nthreads)	864	eio_set_max_parallel (unsigned int nthreads)
689	{	865	{
690	etp_set_max_parallel (nthreads);	866	etp_set_max_parallel (nthreads);
691	}	867	}
692		868
693	int eio_poll (void)	869	int eio_poll (void)
…		…
697		873
698	/*****************************************************************************/	874	/*****************************************************************************/
699	/* work around various missing functions */	875	/* work around various missing functions */
700		876
701	#if !HAVE_PREADWRITE	877	#if !HAVE_PREADWRITE
		878	# undef pread
		879	# undef pwrite
702	# define pread eio__pread	880	# define pread eio__pread
703	# define pwrite eio__pwrite	881	# define pwrite eio__pwrite
704
705	/*
706	* make our pread/pwrite safe against themselves, but not against
707	* normal read/write by using a mutex. slows down execution a lot,
708	* but that's your problem, not mine.
709	*/
710	static mutex_t preadwritelock = X_MUTEX_INIT;
711		882
712	static ssize_t	883	static ssize_t
713	eio__pread (int fd, void *buf, size_t count, off_t offset)	884	eio__pread (int fd, void *buf, size_t count, off_t offset)
714	{	885	{
715	ssize_t res;	886	ssize_t res;
…		…
733		904
734	X_LOCK (preadwritelock);	905	X_LOCK (preadwritelock);
735	ooffset = lseek (fd, 0, SEEK_CUR);	906	ooffset = lseek (fd, 0, SEEK_CUR);
736	lseek (fd, offset, SEEK_SET);	907	lseek (fd, offset, SEEK_SET);
737	res = write (fd, buf, count);	908	res = write (fd, buf, count);
738	lseek (fd, offset, SEEK_SET);	909	lseek (fd, ooffset, SEEK_SET);
739	X_UNLOCK (preadwritelock);	910	X_UNLOCK (preadwritelock);
740		911
741	return res;	912	return res;
742	}	913	}
743	#endif	914	#endif
744		915
745	#ifndef HAVE_FUTIMES	916	#ifndef HAVE_UTIMES
746		917
		918	# undef utimes
747	# define utimes(path,times) eio__utimes (path, times)	919	# define utimes(path,times) eio__utimes (path, times)
748	# define futimes(fd,times) eio__futimes (fd, times)
749		920
750	static int	921	static int
751	eio__utimes (const char *filename, const struct timeval times[2])	922	eio__utimes (const char *filename, const struct timeval times[2])
752	{	923	{
753	if (times)	924	if (times)
…		…
761	}	932	}
762	else	933	else
763	return utime (filename, 0);	934	return utime (filename, 0);
764	}	935	}
765		936
		937	#endif
		938
		939	#ifndef HAVE_FUTIMES
		940
		941	# undef futimes
		942	# define futimes(fd,times) eio__futimes (fd, times)
		943
		944	static int
766	static int eio__futimes (int fd, const struct timeval tv[2])	945	eio__futimes (int fd, const struct timeval tv[2])
767	{	946	{
768	errno = ENOSYS;	947	errno = ENOSYS;
769	return -1;	948	return -1;
770	}	949	}
771		950
772	#endif	951	#endif
773		952
774	#if !HAVE_FDATASYNC	953	#if !HAVE_FDATASYNC
		954	# undef fdatasync
775	# define fdatasync fsync	955	# define fdatasync(fd) fsync (fd)
776	#endif	956	#endif
		957
		958	/* sync_file_range always needs emulation */
		959	static int
		960	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		961	{
		962	#if HAVE_SYNC_FILE_RANGE
		963	int res;
		964
		965	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		966	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		967	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		968	{
		969	flags = 0
		970	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		971	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		972	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		973	}
		974
		975	res = sync_file_range (fd, offset, nbytes, flags);
		976
		977	if (!res || errno != ENOSYS)
		978	return res;
		979	#endif
		980
		981	/* even though we could play tricks with the flags, it's better to always
		982	* call fdatasync, as that matches the expectation of its users best */
		983	return fdatasync (fd);
		984	}
		985
		986	static int
		987	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		988	{
		989	#if HAVE_FALLOCATE
		990	return fallocate (fd, mode, offset, len);
		991	#else
		992	errno = ENOSYS;
		993	return -1;
		994	#endif
		995	}
777		996
778	#if !HAVE_READAHEAD	997	#if !HAVE_READAHEAD
		998	# undef readahead
779	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	999	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
780		1000
781	static ssize_t	1001	static ssize_t
782	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1002	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
783	{	1003	{
…		…
801		1021
802	/* sendfile always needs emulation */	1022	/* sendfile always needs emulation */
803	static ssize_t	1023	static ssize_t
804	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1024	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
805	{	1025	{
		1026	ssize_t written = 0;
806	ssize_t res;	1027	ssize_t res;
807		1028
808	if (!count)	1029	if (!count)
809	return 0;	1030	return 0;
810		1031
		1032	for (;;)
		1033	{
		1034	#ifdef __APPLE__
		1035	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1036	#endif
811	#if HAVE_SENDFILE	1037	#if HAVE_SENDFILE
812	# if __linux	1038	# if __linux
		1039	off_t soffset = offset;
813	res = sendfile (ofd, ifd, &offset, count);	1040	res = sendfile (ofd, ifd, &soffset, count);
814		1041
815	# elif __freebsd	1042	# elif __FreeBSD__
816	/*	1043	/*
817	* Of course, the freebsd sendfile is a dire hack with no thoughts	1044	* Of course, the freebsd sendfile is a dire hack with no thoughts
818	* wasted on making it similar to other I/O functions.	1045	* wasted on making it similar to other I/O functions.
819	*/	1046	*/
820	{
821	off_t sbytes;	1047	off_t sbytes;
822	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1048	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
823		1049
824	if (res < 0 && sbytes)	1050	#if 0 /* according to the manpage, this is correct, but broken behaviour */
825	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1051	/* freebsd' sendfile will return 0 on success */
		1052	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1053	/* not on e.g. EIO or EPIPE - sounds broken */
		1054	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
826	res = sbytes;	1055	res = sbytes;
827	}	1056	#endif
		1057
		1058	/* according to source inspection, this is correct, and useful behaviour */
		1059	if (sbytes)
		1060	res = sbytes;
		1061
		1062	# elif defined (__APPLE__)
		1063	off_t sbytes = count;
		1064	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1065
		1066	/* according to the manpage, sbytes is always valid */
		1067	if (sbytes)
		1068	res = sbytes;
828		1069
829	# elif __hpux	1070	# elif __hpux
830	res = sendfile (ofd, ifd, offset, count, 0, 0);	1071	res = sendfile (ofd, ifd, offset, count, 0, 0);
831		1072
832	# elif __solaris	1073	# elif __solaris
833	{
834	struct sendfilevec vec;	1074	struct sendfilevec vec;
835	size_t sbytes;	1075	size_t sbytes;
836		1076
837	vec.sfv_fd = ifd;	1077	vec.sfv_fd = ifd;
838	vec.sfv_flag = 0;	1078	vec.sfv_flag = 0;
839	vec.sfv_off = offset;	1079	vec.sfv_off = offset;
840	vec.sfv_len = count;	1080	vec.sfv_len = count;
841		1081
842	res = sendfilev (ofd, &vec, 1, &sbytes);	1082	res = sendfilev (ofd, &vec, 1, &sbytes);
843		1083
844	if (res < 0 && sbytes)	1084	if (res < 0 && sbytes)
845	res = sbytes;	1085	res = sbytes;
846	}
847		1086
848	# endif	1087	# endif
		1088
		1089	#elif defined (_WIN32) && 0
		1090	/* does not work, just for documentation of what would need to be done */
		1091	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1092	/* libeio guarantees that the file offset does not change, and windows */
		1093	/* has no way to get an independent handle to the same file description */
		1094	HANDLE h = TO_SOCKET (ifd);
		1095	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1096	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1097
849	#else	1098	#else
850	res = -1;	1099	res = -1;
851	errno = ENOSYS;	1100	errno = ENOSYS;
852	#endif	1101	#endif
853		1102
		1103	/* we assume sendfile can copy at least 128mb in one go */
		1104	if (res <= 128 * 1024 * 1024)
		1105	{
		1106	if (res > 0)
		1107	written += res;
		1108
		1109	if (written)
		1110	return written;
		1111
		1112	break;
		1113	}
		1114	else
		1115	{
		1116	/* if we requested more, then probably the kernel was lazy */
		1117	written += res;
		1118	offset += res;
		1119	count -= res;
		1120
		1121	if (!count)
		1122	return written;
		1123	}
		1124	}
		1125
854	if (res < 0	1126	if (res < 0
855	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1127	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1128	/* BSDs */
		1129	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1130	|| errno == ENOTSUP
		1131	#endif
		1132	|| errno == EOPNOTSUPP /* BSDs */
856	#if __solaris	1133	#if __solaris
857	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1134	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
858	#endif	1135	#endif
859	)	1136	)
860	)	1137	)
…		…
891	}	1168	}
892		1169
893	return res;	1170	return res;
894	}	1171	}
895		1172
		1173	#ifdef PAGESIZE
		1174	# define eio_pagesize() PAGESIZE
		1175	#else
		1176	static intptr_t
		1177	eio_pagesize (void)
		1178	{
		1179	static intptr_t page;
		1180
		1181	if (!page)
		1182	page = sysconf (_SC_PAGESIZE);
		1183
		1184	return page;
		1185	}
		1186	#endif
		1187
		1188	static void
		1189	eio_page_align (void **addr, size_t *length)
		1190	{
		1191	intptr_t mask = eio_pagesize () - 1;
		1192
		1193	/* round down addr */
		1194	intptr_t adj = mask & (intptr_t)*addr;
		1195
		1196	*addr = (void *)((intptr_t)*addr - adj);
		1197	*length += adj;
		1198
		1199	/* round up length */
		1200	*length = (*length + mask) & ~mask;
		1201	}
		1202
		1203	#if !_POSIX_MEMLOCK
		1204	# define eio__mlockall(a) eio_nosyscall()
		1205	#else
		1206
		1207	static int
		1208	eio__mlockall (int flags)
		1209	{
		1210	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1211	extern int mallopt (int, int);
		1212	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1213	#endif
		1214
		1215	if (EIO_MCL_CURRENT != MCL_CURRENT
		1216	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1217	{
		1218	flags = 0
		1219	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1220	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1221	}
		1222
		1223	return mlockall (flags);
		1224	}
		1225	#endif
		1226
		1227	#if !_POSIX_MEMLOCK_RANGE
		1228	# define eio__mlock(a,b) EIO_ENOSYS ()
		1229	#else
		1230
		1231	static int
		1232	eio__mlock (void *addr, size_t length)
		1233	{
		1234	eio_page_align (&addr, &length);
		1235
		1236	return mlock (addr, length);
		1237	}
		1238
		1239	#endif
		1240
		1241	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1242	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1243	#else
		1244
		1245	static int
		1246	eio__msync (void *mem, size_t len, int flags)
		1247	{
		1248	eio_page_align (&mem, &len);
		1249
		1250	if (EIO_MS_ASYNC != MS_SYNC
		1251	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1252	|| EIO_MS_SYNC != MS_SYNC)
		1253	{
		1254	flags = 0
		1255	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1256	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1257	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1258	}
		1259
		1260	return msync (mem, len, flags);
		1261	}
		1262
		1263	#endif
		1264
		1265	static int
		1266	eio__mtouch (eio_req *req)
		1267	{
		1268	void *mem = req->ptr2;
		1269	size_t len = req->size;
		1270	int flags = req->int1;
		1271
		1272	eio_page_align (&mem, &len);
		1273
		1274	{
		1275	intptr_t addr = (intptr_t)mem;
		1276	intptr_t end = addr + len;
		1277	intptr_t page = eio_pagesize ();
		1278
		1279	if (addr < end)
		1280	if (flags & EIO_MT_MODIFY) /* modify */
		1281	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1282	else
		1283	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1284	}
		1285
		1286	return 0;
		1287	}
		1288
		1289	/*****************************************************************************/
		1290	/* requests implemented outside eio_execute, because they are so large */
		1291
		1292	static void
		1293	eio__realpath (eio_req *req, etp_worker *self)
		1294	{
		1295	char *rel = req->ptr1;
		1296	char *res;
		1297	char *tmp1, *tmp2;
		1298	#if SYMLOOP_MAX > 32
		1299	int symlinks = SYMLOOP_MAX;
		1300	#else
		1301	int symlinks = 32;
		1302	#endif
		1303
		1304	req->result = -1;
		1305
		1306	errno = EINVAL;
		1307	if (!rel)
		1308	return;
		1309
		1310	errno = ENOENT;
		1311	if (!*rel)
		1312	return;
		1313
		1314	if (!req->ptr2)
		1315	{
		1316	X_LOCK (wrklock);
		1317	req->flags |= EIO_FLAG_PTR2_FREE;
		1318	X_UNLOCK (wrklock);
		1319	req->ptr2 = malloc (PATH_MAX * 3);
		1320
		1321	errno = ENOMEM;
		1322	if (!req->ptr2)
		1323	return;
		1324	}
		1325
		1326	res = req->ptr2;
		1327	tmp1 = res + PATH_MAX;
		1328	tmp2 = tmp1 + PATH_MAX;
		1329
		1330	#if 0 /* disabled, the musl way to do things is just too racy */
		1331	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1332	/* on linux we may be able to ask the kernel */
		1333	{
		1334	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1335
		1336	if (fd >= 0)
		1337	{
		1338	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1339	req->result = readlink (tmp1, res, PATH_MAX);
		1340	close (fd);
		1341
		1342	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1343
		1344	if (req->result > 0)
		1345	goto done;
		1346	}
		1347	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1348	return;
		1349	}
		1350	#endif
		1351	#endif
		1352
		1353	if (*rel != '/')
		1354	{
		1355	if (!getcwd (res, PATH_MAX))
		1356	return;
		1357
		1358	if (res [1]) /* only use if not / */
		1359	res += strlen (res);
		1360	}
		1361
		1362	while (*rel)
		1363	{
		1364	ssize_t len, linklen;
		1365	char *beg = rel;
		1366
		1367	while (*rel && *rel != '/')
		1368	++rel;
		1369
		1370	len = rel - beg;
		1371
		1372	if (!len) /* skip slashes */
		1373	{
		1374	++rel;
		1375	continue;
		1376	}
		1377
		1378	if (beg [0] == '.')
		1379	{
		1380	if (len == 1)
		1381	continue; /* . - nop */
		1382
		1383	if (beg [1] == '.' && len == 2)
		1384	{
		1385	/* .. - back up one component, if possible */
		1386
		1387	while (res != req->ptr2)
		1388	if (*--res == '/')
		1389	break;
		1390
		1391	continue;
		1392	}
		1393	}
		1394
		1395	errno = ENAMETOOLONG;
		1396	if (res + 1 + len + 1 >= tmp1)
		1397	return;
		1398
		1399	/* copy one component */
		1400	*res = '/';
		1401	memcpy (res + 1, beg, len);
		1402
		1403	/* zero-terminate, for readlink */
		1404	res [len + 1] = 0;
		1405
		1406	/* now check if it's a symlink */
		1407	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1408
		1409	if (linklen < 0)
		1410	{
		1411	if (errno != EINVAL)
		1412	return;
		1413
		1414	/* it's a normal directory. hopefully */
		1415	res += len + 1;
		1416	}
		1417	else
		1418	{
		1419	/* yay, it was a symlink - build new path in tmp2 */
		1420	int rellen = strlen (rel);
		1421
		1422	errno = ENAMETOOLONG;
		1423	if (linklen + 1 + rellen >= PATH_MAX)
		1424	return;
		1425
		1426	errno = ELOOP;
		1427	if (!--symlinks)
		1428	return;
		1429
		1430	if (*tmp1 == '/')
		1431	res = req->ptr2; /* symlink resolves to an absolute path */
		1432
		1433	/* we need to be careful, as rel might point into tmp2 already */
		1434	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1435	tmp2 [linklen] = '/';
		1436	memcpy (tmp2, tmp1, linklen);
		1437
		1438	rel = tmp2;
		1439	}
		1440	}
		1441
		1442	/* special case for the lone root path */
		1443	if (res == req->ptr2)
		1444	*res++ = '/';
		1445
		1446	req->result = res - (char *)req->ptr2;
		1447
		1448	done:
		1449	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1450	}
		1451
		1452	static signed char
		1453	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1454	{
		1455	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1456	: a->inode < b->inode ? -1
		1457	: a->inode > b->inode ? 1
		1458	: 0;
		1459	}
		1460
		1461	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1462
		1463	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1464	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1465
		1466	static void
		1467	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1468	{
		1469	unsigned char bits [9 + sizeof (ino_t) * 8];
		1470	unsigned char *bit = bits;
		1471
		1472	assert (CHAR_BIT == 8);
		1473	assert (sizeof (eio_dirent) * 8 < 256);
		1474	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1475	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1476
		1477	if (size <= EIO_SORT_FAST)
		1478	return;
		1479
		1480	/* first prepare an array of bits to test in our radix sort */
		1481	/* try to take endianness into account, as well as differences in ino_t sizes */
		1482	/* inode_bits must contain all inodes ORed together */
		1483	/* which is used to skip bits that are 0 everywhere, which is very common */
		1484	{
		1485	ino_t endianness;
		1486	int i, j;
		1487
		1488	/* we store the byte offset of byte n into byte n of "endianness" */
		1489	for (i = 0; i < sizeof (ino_t); ++i)
		1490	((unsigned char *)&endianness)[i] = i;
		1491
		1492	*bit++ = 0;
		1493
		1494	for (i = 0; i < sizeof (ino_t); ++i)
		1495	{
		1496	/* shifting off the byte offsets out of "endianness" */
		1497	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1498	endianness >>= 8;
		1499
		1500	for (j = 0; j < 8; ++j)
		1501	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1502	*bit++ = offs + j;
		1503	}
		1504
		1505	for (j = 0; j < 8; ++j)
		1506	if (score_bits & (1 << j))
		1507	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1508	}
		1509
		1510	/* now actually do the sorting (a variant of MSD radix sort) */
		1511	{
		1512	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1513	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1514	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1515	int stk_idx = 0;
		1516
		1517	base_stk [stk_idx] = dents;
		1518	end_stk [stk_idx] = dents + size;
		1519	bit_stk [stk_idx] = bit - 1;
		1520
		1521	do
		1522	{
		1523	base = base_stk [stk_idx];
		1524	end = end_stk [stk_idx];
		1525	bit = bit_stk [stk_idx];
		1526
		1527	for (;;)
		1528	{
		1529	unsigned char O = *bit >> 3;
		1530	unsigned char M = 1 << (*bit & 7);
		1531
		1532	eio_dirent *a = base;
		1533	eio_dirent *b = end;
		1534
		1535	if (b - a < EIO_SORT_CUTOFF)
		1536	break;
		1537
		1538	/* now bit-partition the array on the bit */
		1539	/* this ugly asymmetric loop seems to perform much better than typical */
		1540	/* partition algos found in the literature */
		1541	do
		1542	if (!(((unsigned char *)a)[O] & M))
		1543	++a;
		1544	else if (!(((unsigned char *)--b)[O] & M))
		1545	{
		1546	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1547	++a;
		1548	}
		1549	while (b > a);
		1550
		1551	/* next bit, or stop, if no bits left in this path */
		1552	if (!*--bit)
		1553	break;
		1554
		1555	base_stk [stk_idx] = a;
		1556	end_stk [stk_idx] = end;
		1557	bit_stk [stk_idx] = bit;
		1558	++stk_idx;
		1559
		1560	end = a;
		1561	}
		1562	}
		1563	while (stk_idx--);
		1564	}
		1565	}
		1566
		1567	static void
		1568	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1569	{
		1570	/* first move the smallest element to the front, to act as a sentinel */
		1571	{
		1572	int i;
		1573	eio_dirent *min = dents;
		1574
		1575	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1576	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1577	if (EIO_DENT_CMP (dents [i], <, *min))
		1578	min = &dents [i];
		1579
		1580	/* swap elements 0 and j (minimum) */
		1581	{
		1582	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1583	}
		1584	}
		1585
		1586	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1587	{
		1588	eio_dirent *i, *j;
		1589
		1590	for (i = dents + 1; i < dents + size; ++i)
		1591	{
		1592	eio_dirent value = *i;
		1593
		1594	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1595	j [1] = j [0];
		1596
		1597	j [1] = value;
		1598	}
		1599	}
		1600	}
		1601
		1602	static void
		1603	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1604	{
		1605	if (size <= 1)
		1606	return; /* our insertion sort relies on size > 0 */
		1607
		1608	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1609	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1610	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1611
		1612	/* use an insertion sort at the end, or for small arrays, */
		1613	/* as insertion sort is more efficient for small partitions */
		1614	eio_dent_insertion_sort (dents, size);
		1615	}
		1616
896	/* read a full directory */	1617	/* read a full directory */
897	static void	1618	static void
898	eio__scandir (eio_req *req, etp_worker *self)	1619	eio__scandir (eio_req *req, etp_worker *self)
899	{	1620	{
900	DIR *dirp;	1621	DIR *dirp;
901	EIO_STRUCT_DIRENT *entp;	1622	EIO_STRUCT_DIRENT *entp;
902	char *name, *names;	1623	char *name, *names;
903	int memlen = 4096;	1624	int namesalloc = 4096;
904	int memofs = 0;	1625	int namesoffs = 0;
		1626	int flags = req->int1;
		1627	eio_dirent *dents = 0;
		1628	int dentalloc = 128;
905	int res = 0;	1629	int dentoffs = 0;
		1630	ino_t inode_bits = 0;
		1631
		1632	req->result = -1;
		1633
		1634	if (!(flags & EIO_READDIR_DENTS))
		1635	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
906		1636
907	X_LOCK (wrklock);	1637	X_LOCK (wrklock);
		1638	/* the corresponding closedir is in ETP_WORKER_CLEAR */
908	self->dirp = dirp = opendir (req->ptr1);	1639	self->dirp = dirp = opendir (req->ptr1);
		1640
909	req->flags |= EIO_FLAG_PTR2_FREE;	1641	if (req->flags & EIO_FLAG_PTR1_FREE)
		1642	free (req->ptr1);
		1643
		1644	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1645	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
910	req->ptr2 = names = malloc (memlen);	1646	req->ptr2 = names = malloc (namesalloc);
911	X_UNLOCK (wrklock);	1647	X_UNLOCK (wrklock);
912		1648
913	if (dirp && names)	1649	if (dirp && names && (!flags || dents))
914	for (;;)	1650	for (;;)
915	{	1651	{
916	errno = 0;	1652	errno = 0;
917	entp = readdir (dirp);	1653	entp = readdir (dirp);
918		1654
919	if (!entp)	1655	if (!entp)
		1656	{
		1657	if (errno)
		1658	break;
		1659
		1660	/* sort etc. */
		1661	req->int1 = flags;
		1662	req->result = dentoffs;
		1663
		1664	if (flags & EIO_READDIR_STAT_ORDER)
		1665	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1666	else if (flags & EIO_READDIR_DIRS_FIRST)
		1667	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1668	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1669	else
		1670	{
		1671	/* in this case, all is known, and we just put dirs first and sort them */
		1672	eio_dirent *oth = dents + dentoffs;
		1673	eio_dirent *dir = dents;
		1674
		1675	/* now partition dirs to the front, and non-dirs to the back */
		1676	/* by walking from both sides and swapping if necessary */
		1677	while (oth > dir)
		1678	{
		1679	if (dir->type == EIO_DT_DIR)
		1680	++dir;
		1681	else if ((--oth)->type == EIO_DT_DIR)
		1682	{
		1683	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1684
		1685	++dir;
		1686	}
		1687	}
		1688
		1689	/* now sort the dirs only (dirs all have the same score) */
		1690	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1691	}
		1692
920	break;	1693	break;
		1694	}
921		1695
		1696	/* now add the entry to our list(s) */
922	name = entp->d_name;	1697	name = entp->d_name;
923		1698
		1699	/* skip . and .. entries */
924	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1700	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
925	{	1701	{
926	int len = strlen (name) + 1;	1702	int len = D_NAMLEN (entp) + 1;
927		1703
928	res++;	1704	while (ecb_expect_false (namesoffs + len > namesalloc))
929
930	while (memofs + len > memlen)
931	{	1705	{
932	memlen *= 2;	1706	namesalloc *= 2;
933	X_LOCK (wrklock);	1707	X_LOCK (wrklock);
934	req->ptr2 = names = realloc (names, memlen);	1708	req->ptr2 = names = realloc (names, namesalloc);
935	X_UNLOCK (wrklock);	1709	X_UNLOCK (wrklock);
936		1710
937	if (!names)	1711	if (!names)
938	break;	1712	break;
939	}	1713	}
940		1714
941	memcpy (names + memofs, name, len);	1715	memcpy (names + namesoffs, name, len);
		1716
		1717	if (dents)
		1718	{
		1719	struct eio_dirent *ent;
		1720
		1721	if (ecb_expect_false (dentoffs == dentalloc))
		1722	{
		1723	dentalloc *= 2;
		1724	X_LOCK (wrklock);
		1725	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1726	X_UNLOCK (wrklock);
		1727
		1728	if (!dents)
		1729	break;
		1730	}
		1731
		1732	ent = dents + dentoffs;
		1733
		1734	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1735	ent->namelen = len - 1;
		1736	ent->inode = D_INO (entp);
		1737
		1738	inode_bits |= ent->inode;
		1739
		1740	switch (D_TYPE (entp))
		1741	{
		1742	default:
		1743	ent->type = EIO_DT_UNKNOWN;
		1744	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1745	break;
		1746
		1747	#ifdef DT_FIFO
		1748	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1749	#endif
		1750	#ifdef DT_CHR
		1751	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1752	#endif
		1753	#ifdef DT_MPC
		1754	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1755	#endif
		1756	#ifdef DT_DIR
		1757	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1758	#endif
		1759	#ifdef DT_NAM
		1760	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1761	#endif
		1762	#ifdef DT_BLK
		1763	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1764	#endif
		1765	#ifdef DT_MPB
		1766	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1767	#endif
		1768	#ifdef DT_REG
		1769	case DT_REG: ent->type = EIO_DT_REG; break;
		1770	#endif
		1771	#ifdef DT_NWK
		1772	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1773	#endif
		1774	#ifdef DT_CMP
		1775	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1776	#endif
		1777	#ifdef DT_LNK
		1778	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1779	#endif
		1780	#ifdef DT_SOCK
		1781	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1782	#endif
		1783	#ifdef DT_DOOR
		1784	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1785	#endif
		1786	#ifdef DT_WHT
		1787	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1788	#endif
		1789	}
		1790
		1791	ent->score = 7;
		1792
		1793	if (flags & EIO_READDIR_DIRS_FIRST)
		1794	{
		1795	if (ent->type == EIO_DT_UNKNOWN)
		1796	{
		1797	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1798	ent->score = 1;
		1799	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1800	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1801	}
		1802	else if (ent->type == EIO_DT_DIR)
		1803	ent->score = 0;
		1804	}
		1805	}
		1806
942	memofs += len;	1807	namesoffs += len;
		1808	++dentoffs;
		1809	}
		1810
		1811	if (EIO_CANCELLED (req))
		1812	{
		1813	errno = ECANCELED;
		1814	break;
943	}	1815	}
944	}	1816	}
945
946	if (errno)
947	res = -1;
948
949	req->result = res;
950	}	1817	}
951		1818
952	/*****************************************************************************/	1819	/*****************************************************************************/
953		1820
954	#define ALLOC(len) \	1821	#define ALLOC(len) \
…		…
969	X_THREAD_PROC (etp_proc)	1836	X_THREAD_PROC (etp_proc)
970	{	1837	{
971	ETP_REQ *req;	1838	ETP_REQ *req;
972	struct timespec ts;	1839	struct timespec ts;
973	etp_worker *self = (etp_worker *)thr_arg;	1840	etp_worker *self = (etp_worker *)thr_arg;
		1841	int timeout;
974		1842
975	/* try to distribute timeouts somewhat randomly */	1843	/* try to distribute timeouts somewhat evenly */
976	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1844	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
977		1845
978	for (;;)	1846	for (;;)
979	{	1847	{
		1848	ts.tv_sec = 0;
		1849
980	X_LOCK (reqlock);	1850	X_LOCK (reqlock);
981		1851
982	for (;;)	1852	for (;;)
983	{	1853	{
984	self->req = req = reqq_shift (&req_queue);	1854	self->req = req = reqq_shift (&req_queue);
985		1855
986	if (req)	1856	if (req)
987	break;	1857	break;
988		1858
		1859	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1860	{
		1861	X_UNLOCK (reqlock);
		1862	X_LOCK (wrklock);
		1863	--started;
		1864	X_UNLOCK (wrklock);
		1865	goto quit;
		1866	}
		1867
989	++idle;	1868	++idle;
990		1869
991	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1870	if (idle <= max_idle)
992	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1871	/* we are allowed to idle, so do so without any timeout */
		1872	X_COND_WAIT (reqwait, reqlock);
		1873	else
993	{	1874	{
994	if (idle > max_idle)	1875	/* initialise timeout once */
995	{	1876	if (!ts.tv_sec)
996	--idle;	1877	ts.tv_sec = time (0) + idle_timeout;
997	X_UNLOCK (reqlock);
998	X_LOCK (wrklock);
999	--started;
1000	X_UNLOCK (wrklock);
1001	goto quit;
1002	}
1003		1878
1004	/* we are allowed to idle, so do so without any timeout */
1005	X_COND_WAIT (reqwait, reqlock);	1879	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1880	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1006	}	1881	}
1007		1882
1008	--idle;	1883	--idle;
1009	}	1884	}
1010		1885
…		…
1013	X_UNLOCK (reqlock);	1888	X_UNLOCK (reqlock);
1014		1889
1015	if (req->type < 0)	1890	if (req->type < 0)
1016	goto quit;	1891	goto quit;
1017		1892
1018	if (!EIO_CANCELLED (req))
1019	ETP_EXECUTE (self, req);	1893	ETP_EXECUTE (self, req);
1020		1894
1021	X_LOCK (reslock);	1895	X_LOCK (reslock);
1022		1896
1023	++npending;	1897	++npending;
1024		1898
…		…
1039	return 0;	1913	return 0;
1040	}	1914	}
1041		1915
1042	/*****************************************************************************/	1916	/*****************************************************************************/
1043		1917
		1918	int ecb_cold
1044	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1919	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1045	{	1920	{
1046	return etp_init (want_poll, done_poll);	1921	return etp_init (want_poll, done_poll);
1047	}	1922	}
1048		1923
		1924	ecb_inline void
1049	static void eio_api_destroy (eio_req *req)	1925	eio_api_destroy (eio_req *req)
1050	{	1926	{
1051	free (req);	1927	free (req);
1052	}	1928	}
1053		1929
1054	#define REQ(rtype) \	1930	#define REQ(rtype) \
…		…
1073	{ \	1949	{ \
1074	eio_api_destroy (req); \	1950	eio_api_destroy (req); \
1075	return 0; \	1951	return 0; \
1076	}	1952	}
1077		1953
		1954	static void
1078	static void eio_execute (etp_worker *self, eio_req *req)	1955	eio_execute (etp_worker *self, eio_req *req)
1079	{	1956	{
1080	errno = 0;	1957	if (ecb_expect_false (EIO_CANCELLED (req)))
		1958	{
		1959	req->result = -1;
		1960	req->errorno = ECANCELED;
		1961	return;
		1962	}
1081		1963
1082	switch (req->type)	1964	switch (req->type)
1083	{	1965	{
1084	case EIO_READ: ALLOC (req->size);	1966	case EIO_READ: ALLOC (req->size);
1085	req->result = req->offs >= 0	1967	req->result = req->offs >= 0
…		…
1097	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1979	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1098	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1980	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1099	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1981	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1100	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1982	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1101		1983
		1984	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1985	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1986	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1987	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1988
1102	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	1989	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1103	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	1990	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1104	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	1991	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1105	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	1992	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1106	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;	1993	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
…		…
1113	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	2000	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1114	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	2001	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1115	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2002	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1116	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2003	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1117	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2004	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1118	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;	2005	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1119		2006
		2007	case EIO_REALPATH: eio__realpath (req, self); break;
		2008
1120	case EIO_READLINK: ALLOC (NAME_MAX);	2009	case EIO_READLINK: ALLOC (PATH_MAX);
1121	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	2010	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1122		2011
1123	case EIO_SYNC: req->result = 0; sync (); break;	2012	case EIO_SYNC: req->result = 0; sync (); break;
1124	case EIO_FSYNC: req->result = fsync (req->int1); break;	2013	case EIO_FSYNC: req->result = fsync (req->int1); break;
1125	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2014	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2015	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2016	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		2017	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2018	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		2019	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2020	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1126		2021
1127	case EIO_READDIR: eio__scandir (req, self); break;	2022	case EIO_READDIR: eio__scandir (req, self); break;
1128		2023
1129	case EIO_BUSY:	2024	case EIO_BUSY:
1130	#ifdef _WIN32	2025	#ifdef _WIN32
1131	Sleep (req->nv1 * 1000.);	2026	Sleep (req->nv1 * 1e3);
1132	#else	2027	#else
1133	{	2028	{
1134	struct timeval tv;	2029	struct timeval tv;
1135		2030
1136	tv.tv_sec = req->nv1;	2031	tv.tv_sec = req->nv1;
1137	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	2032	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1138		2033
1139	req->result = select (0, 0, 0, 0, &tv);	2034	req->result = select (0, 0, 0, 0, &tv);
1140	}	2035	}
1141	#endif	2036	#endif
1142	break;	2037	break;
…		…
1157	times = tv;	2052	times = tv;
1158	}	2053	}
1159	else	2054	else
1160	times = 0;	2055	times = 0;
1161		2056
1162
1163	req->result = req->type == EIO_FUTIME	2057	req->result = req->type == EIO_FUTIME
1164	? futimes (req->int1, times)	2058	? futimes (req->int1, times)
1165	: utimes (req->ptr1, times);	2059	: utimes (req->ptr1, times);
1166	}	2060	}
		2061	break;
1167		2062
1168	case EIO_GROUP:	2063	case EIO_GROUP:
		2064	abort (); /* handled in eio_request */
		2065
1169	case EIO_NOP:	2066	case EIO_NOP:
1170	req->result = 0;	2067	req->result = 0;
1171	break;	2068	break;
1172		2069
1173	case EIO_CUSTOM:	2070	case EIO_CUSTOM:
1174	req->feed (req);	2071	req->feed (req);
1175	break;	2072	break;
1176		2073
1177	default:	2074	default:
		2075	errno = ENOSYS;
1178	req->result = -1;	2076	req->result = -1;
1179	break;	2077	break;
1180	}	2078	}
1181		2079
1182	req->errorno = errno;	2080	req->errorno = errno;
…		…
1202	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	2100	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1203	{	2101	{
1204	REQ (EIO_FSYNC); req->int1 = fd; SEND;	2102	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1205	}	2103	}
1206		2104
		2105	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2106	{
		2107	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2108	}
		2109
		2110	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2111	{
		2112	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2113	}
		2114
		2115	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2116	{
		2117	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2118	}
		2119
		2120	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2121	{
		2122	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2123	}
		2124
		2125	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2126	{
		2127	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2128	}
		2129
		2130	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2131	{
		2132	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2133	}
		2134
1207	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2135	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1208	{	2136	{
1209	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2137	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1210	}	2138	}
1211		2139
…		…
1230	}	2158	}
1231		2159
1232	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2160	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1233	{	2161	{
1234	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2162	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		2163	}
		2164
		2165	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2166	{
		2167	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1235	}	2168	}
1236		2169
1237	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2170	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1238	{	2171	{
1239	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2172	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
…		…
1303	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2236	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1304	{	2237	{
1305	return eio__1path (EIO_READLINK, path, pri, cb, data);	2238	return eio__1path (EIO_READLINK, path, pri, cb, data);
1306	}	2239	}
1307		2240
		2241	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2242	{
		2243	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2244	}
		2245
1308	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2246	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1309	{	2247	{
1310	return eio__1path (EIO_STAT, path, pri, cb, data);	2248	return eio__1path (EIO_STAT, path, pri, cb, data);
1311	}	2249	}
1312		2250
1313	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2251	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1314	{	2252	{
1315	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2253	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1316	}	2254	}
1317		2255
		2256	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2257	{
		2258	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2259	}
		2260
1318	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2261	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1319	{	2262	{
1320	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2263	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1321	}	2264	}
1322		2265
1323	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	2266	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1324	{	2267	{
1325	return eio__1path (EIO_RMDIR, path, pri, cb, data);	2268	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1326	}	2269	}
1327		2270
1328	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	2271	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1329	{	2272	{
1330	return eio__1path (EIO_READDIR, path, pri, cb, data);	2273	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1331	}	2274	}
1332		2275
1333	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2276	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1334	{	2277	{
1335	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2278	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1336	}	2279	}
1337		2280
1338	static eio_req *	2281	static eio_req *
1339	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2282	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1340	{	2283	{
…		…
1364	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2307	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1365	{	2308	{
1366	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2309	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1367	}	2310	}
1368		2311
1369	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2312	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1370	{	2313	{
1371	REQ (EIO_CUSTOM); req->feed = execute; SEND;	2314	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1372	}	2315	}
1373		2316
1374	#endif	2317	#endif
…		…
1385	#undef SEND	2328	#undef SEND
1386		2329
1387	/*****************************************************************************/	2330	/*****************************************************************************/
1388	/* grp functions */	2331	/* grp functions */
1389		2332
		2333	void
1390	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2334	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1391	{	2335	{
1392	grp->int2 = limit;	2336	grp->int2 = limit;
1393	grp->feed = feed;	2337	grp->feed = feed;
1394		2338
1395	grp_try_feed (grp);	2339	grp_try_feed (grp);
1396	}	2340	}
1397		2341
		2342	void
1398	void eio_grp_limit (eio_req *grp, int limit)	2343	eio_grp_limit (eio_req *grp, int limit)
1399	{	2344	{
1400	grp->int2 = limit;	2345	grp->int2 = limit;
1401		2346
1402	grp_try_feed (grp);	2347	grp_try_feed (grp);
1403	}	2348	}
1404		2349
		2350	void
1405	void eio_grp_add (eio_req *grp, eio_req *req)	2351	eio_grp_add (eio_req *grp, eio_req *req)
1406	{	2352	{
1407	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2353	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
		2354
		2355	grp->flags |= EIO_FLAG_GROUPADD;
1408		2356
1409	++grp->size;	2357	++grp->size;
1410	req->grp = grp;	2358	req->grp = grp;
1411		2359
1412	req->grp_prev = 0;	2360	req->grp_prev = 0;
…		…
1419	}	2367	}
1420		2368
1421	/*****************************************************************************/	2369	/*****************************************************************************/
1422	/* misc garbage */	2370	/* misc garbage */
1423		2371
		2372	ssize_t
1424	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2373	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1425	{	2374	{
1426	etp_worker wrk;	2375	etp_worker wrk;
		2376	ssize_t ret;
1427		2377
1428	wrk.dbuf = 0;	2378	wrk.dbuf = 0;
1429		2379
1430	eio__sendfile (ofd, ifd, offset, count, &wrk);	2380	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1431		2381
1432	if (wrk.dbuf)	2382	if (wrk.dbuf)
1433	free (wrk.dbuf);	2383	free (wrk.dbuf);
1434	}
1435		2384
		2385	return ret;
		2386	}
		2387

Diff Legend

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