ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.17 by root, Tue Jun 3 05:12:51 2008 UTC vs.
Revision 1.101 by root, Wed Aug 3 15:25:38 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	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
		64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
		65	#if HAVE_STDINT_H
		66	# include <stdint.h>
		67	#endif
		68
		69	#ifndef ECANCELED
		70	# define ECANCELED EDOM
		71	#endif
		72	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
		77	# define ENOTSOCK WSAENOTSOCK
		78	#endif
		79
		80	static void eio_destroy (eio_req *req);
		81
54	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
55	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	83	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
56	#endif	84	#endif
57		85
58	#ifndef EIO_DESTROY	86	#ifndef EIO_DESTROY
…		…
61		89
62	#ifndef EIO_FEED	90	#ifndef EIO_FEED
63	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	91	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
64	#endif	92	#endif
65		93
		94	#ifndef EIO_FD_TO_WIN32_HANDLE
		95	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		96	#endif
		97	#ifndef EIO_WIN32_HANDLE_TO_FD
		98	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		99	#endif
		100
		101	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		102
		103	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		104
66	#ifdef _WIN32	105	#ifdef _WIN32
67		106
68	/*doh*/	107	#undef PAGESIZE
		108	#define PAGESIZE 4096 /* GetSystemInfo? */
		109
		110	/* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
		111
		112	#ifdef EIO_STRUCT_STATI64
		113	/* look at perl's non-sloppy stat */
		114	#define stat(path,buf) _stati64 (path,buf)
		115	#define fstat(fd,buf) _fstati64 (fd,buf)
		116	#endif
		117	#define lstat(path,buf) stat (path,buf)
		118	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		119	#define mkdir(path,mode) _mkdir (path)
		120	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		121
		122	#define chmod(path,mode) _chmod (path, mode)
		123	#define dup(fd) _dup (fd)
		124	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		125
		126	#define fchmod(fd,mode) EIO_ENOSYS ()
		127	#define chown(path,uid,gid) EIO_ENOSYS ()
		128	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		129	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		130	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		131	#define mknod(path,mode,dev) EIO_ENOSYS ()
		132	#define sync() EIO_ENOSYS ()
		133	#define readlink(path,buf,s) EIO_ENOSYS ()
		134	#define statvfs(path,buf) EIO_ENOSYS ()
		135	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		136
		137	/* rename() uses MoveFile, which fails to overwrite */
		138	#define rename(old,neu) eio__rename (old, neu)
		139
		140	static int
		141	eio__rename (const char *old, const char *neu)
		142	{
		143	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		144	return 0;
		145
		146	/* should steal _dosmaperr */
		147	switch (GetLastError ())
		148	{
		149	case ERROR_FILE_NOT_FOUND:
		150	case ERROR_PATH_NOT_FOUND:
		151	case ERROR_INVALID_DRIVE:
		152	case ERROR_NO_MORE_FILES:
		153	case ERROR_BAD_NETPATH:
		154	case ERROR_BAD_NET_NAME:
		155	case ERROR_BAD_PATHNAME:
		156	case ERROR_FILENAME_EXCED_RANGE:
		157	errno = ENOENT;
		158	break;
		159
		160	default:
		161	errno = EACCES;
		162	break;
		163	}
		164
		165	return -1;
		166	}
		167
		168	/* we could even stat and see if it exists */
		169	static int
		170	symlink (const char *old, const char *neu)
		171	{
		172	#if WINVER >= 0x0600
		173	if (CreateSymbolicLink (neu, old, 1))
		174	return 0;
		175
		176	if (CreateSymbolicLink (neu, old, 0))
		177	return 0;
		178	#endif
		179
		180	return EIO_ERRNO (ENOENT, -1);
		181	}
		182
		183	/* POSIX API only */
		184	#define CreateHardLink(neu,old,flags) 0
		185	#define CreateSymbolicLink(neu,old,flags) 0
		186
		187	struct statvfs
		188	{
		189	int dummy;
		190	};
		191
		192	#define DT_DIR EIO_DT_DIR
		193	#define DT_REG EIO_DT_REG
		194	#define D_NAME(entp) entp.cFileName
		195	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
69		196
70	#else	197	#else
71		198
72	# include "config.h"
73	# include <sys/time.h>	199	#include <sys/time.h>
74	# include <sys/select.h>	200	#include <sys/select.h>
		201	#include <sys/statvfs.h>
75	# include <unistd.h>	202	#include <unistd.h>
		203	#include <signal.h>
		204	#include <dirent.h>
		205
		206	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		207	#include <sys/mman.h>
		208	#endif
		209
		210	#define D_NAME(entp) entp->d_name
		211
		212	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		213	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		214	#define _DIRENT_HAVE_D_TYPE /* sigh */
		215	#define D_INO(de) (de)->d_fileno
		216	#define D_NAMLEN(de) (de)->d_namlen
		217	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		218	#define D_INO(de) (de)->d_ino
		219	#endif
		220
		221	#ifdef _D_EXACT_NAMLEN
		222	#undef D_NAMLEN
		223	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		224	#endif
		225
		226	#ifdef _DIRENT_HAVE_D_TYPE
		227	#define D_TYPE(de) (de)->d_type
		228	#endif
		229
		230	#ifndef EIO_STRUCT_DIRENT
		231	#define EIO_STRUCT_DIRENT struct dirent
		232	#endif
		233
		234	#endif
		235
		236	#if HAVE_UTIMES
76	# include <utime.h>	237	# include <utime.h>
77	# include <signal.h>
78	# include <dirent.h>
79
80	# ifndef EIO_STRUCT_DIRENT
81	# define EIO_STRUCT_DIRENT struct dirent
82	# endif	238	#endif
83		239
		240	#if HAVE_SYS_SYSCALL_H
		241	# include <sys/syscall.h>
		242	#endif
		243
		244	#if HAVE_SYS_PRCTL_H
		245	# include <sys/prctl.h>
84	#endif	246	#endif
85		247
86	#if HAVE_SENDFILE	248	#if HAVE_SENDFILE
87	# if __linux	249	# if __linux
88	# include <sys/sendfile.h>	250	# include <sys/sendfile.h>
89	# elif __freebsd	251	# elif __FreeBSD__ || defined __APPLE__
90	# include <sys/socket.h>	252	# include <sys/socket.h>
91	# include <sys/uio.h>	253	# include <sys/uio.h>
92	# elif __hpux	254	# elif __hpux
93	# include <sys/socket.h>	255	# include <sys/socket.h>
94	# elif __solaris /* not yet */	256	# elif __solaris
95	# include <sys/sendfile.h>	257	# include <sys/sendfile.h>
96	# else	258	# else
97	# error sendfile support requested but not available	259	# error sendfile support requested but not available
98	# endif	260	# endif
99	#endif	261	#endif
100		262
101	/* number of seconds after which an idle threads exit */	263	#ifndef D_TYPE
102	#define IDLE_TIMEOUT 10	264	# define D_TYPE(de) 0
		265	#endif
		266	#ifndef D_INO
		267	# define D_INO(de) 0
		268	#endif
		269	#ifndef D_NAMLEN
		270	# define D_NAMLEN(entp) strlen (D_NAME (entp))
		271	#endif
103		272
104	/* used for struct dirent, AIX doesn't provide it */	273	/* used for struct dirent, AIX doesn't provide it */
105	#ifndef NAME_MAX	274	#ifndef NAME_MAX
106	# define NAME_MAX 4096	275	# define NAME_MAX 4096
107	#endif	276	#endif
108		277
		278	/* used for readlink etc. */
		279	#ifndef PATH_MAX
		280	# define PATH_MAX 4096
		281	#endif
		282
109	/* buffer size for various temporary buffers */	283	/* buffer size for various temporary buffers */
110	#define EIO_BUFSIZE 65536	284	#define EIO_BUFSIZE 65536
111		285
112	#define dBUF \	286	#define dBUF \
113	char *eio_buf; \
114	ETP_WORKER_LOCK (self); \
115	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	287	char *eio_buf = malloc (EIO_BUFSIZE); \
116	ETP_WORKER_UNLOCK (self); \
117	errno = ENOMEM; \	288	errno = ENOMEM; \
118	if (!eio_buf) \	289	if (!eio_buf) \
119	return -1;	290	return -1
		291
		292	#define FUBd \
		293	free (eio_buf)
120		294
121	#define EIO_TICKS ((1000000 + 1023) >> 10)	295	#define EIO_TICKS ((1000000 + 1023) >> 10)
122
123	/*****************************************************************************/
124		296
125	#define ETP_PRI_MIN EIO_PRI_MIN	297	#define ETP_PRI_MIN EIO_PRI_MIN
126	#define ETP_PRI_MAX EIO_PRI_MAX	298	#define ETP_PRI_MAX EIO_PRI_MAX
127		299
128	struct etp_worker;	300	struct etp_worker;
…		…
132	static int eio_finish (eio_req *req);	304	static int eio_finish (eio_req *req);
133	#define ETP_FINISH(req) eio_finish (req)	305	#define ETP_FINISH(req) eio_finish (req)
134	static void eio_execute (struct etp_worker *self, eio_req *req);	306	static void eio_execute (struct etp_worker *self, eio_req *req);
135	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	307	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
136		308
137	#define ETP_WORKER_CLEAR(req) \
138	if (wrk->dbuf) \
139	{ \
140	free (wrk->dbuf); \
141	wrk->dbuf = 0; \
142	} \
143	\
144	if (wrk->dirp) \
145	{ \
146	closedir (wrk->dirp); \
147	wrk->dirp = 0; \
148	}
149	#define ETP_WORKER_COMMON \
150	void *dbuf; \
151	DIR *dirp;
152
153	/*****************************************************************************/	309	/*****************************************************************************/
154		310
155	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	311	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
156		312
157	/* calculcate time difference in ~1/EIO_TICKS of a second */	313	/* calculate time difference in ~1/EIO_TICKS of a second */
		314	ecb_inline int
158	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	315	tvdiff (struct timeval *tv1, struct timeval *tv2)
159	{	316	{
160	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	317	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
161	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	318	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
162	}	319	}
163		320
…		…
167	static void (*done_poll_cb) (void);	324	static void (*done_poll_cb) (void);
168		325
169	static unsigned int max_poll_time; /* reslock */	326	static unsigned int max_poll_time; /* reslock */
170	static unsigned int max_poll_reqs; /* reslock */	327	static unsigned int max_poll_reqs; /* reslock */
171		328
172	static volatile unsigned int nreqs; /* reqlock */	329	static unsigned int nreqs; /* reqlock */
173	static volatile unsigned int nready; /* reqlock */	330	static unsigned int nready; /* reqlock */
174	static volatile unsigned int npending; /* reqlock */	331	static unsigned int npending; /* reqlock */
175	static volatile unsigned int max_idle = 4;	332	static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		333	static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
176		334
177	static mutex_t wrklock = X_MUTEX_INIT;	335	static xmutex_t wrklock;
178	static mutex_t reslock = X_MUTEX_INIT;	336	static xmutex_t reslock;
179	static mutex_t reqlock = X_MUTEX_INIT;	337	static xmutex_t reqlock;
180	static cond_t reqwait = X_COND_INIT;	338	static xcond_t reqwait;
		339
		340	#if !HAVE_PREADWRITE
		341	/*
		342	* make our pread/pwrite emulation safe against themselves, but not against
		343	* normal read/write by using a mutex. slows down execution a lot,
		344	* but that's your problem, not mine.
		345	*/
		346	static xmutex_t preadwritelock;
		347	#endif
181		348
182	typedef struct etp_worker	349	typedef struct etp_worker
183	{	350	{
184	/* locked by wrklock */	351	/* locked by wrklock */
185	struct etp_worker *prev, *next;	352	struct etp_worker *prev, *next;
186		353
187	thread_t tid;	354	xthread_t tid;
188		355
189	/* locked by reslock, reqlock or wrklock */	356	/* locked by reslock, reqlock or wrklock */
190	ETP_REQ *req; /* currently processed request */	357	ETP_REQ *req; /* currently processed request */
191		358
		359	#ifdef ETP_WORKER_COMMON
192	ETP_WORKER_COMMON	360	ETP_WORKER_COMMON
		361	#endif
193	} etp_worker;	362	} etp_worker;
194		363
195	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	364	static etp_worker wrk_first; /* NOT etp */
196		365
197	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	366	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
198	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	367	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
199		368
200	/* worker threads management */	369	/* worker threads management */
201		370
		371	static void ecb_cold
202	static void etp_worker_clear (etp_worker *wrk)	372	etp_worker_clear (etp_worker *wrk)
203	{	373	{
204	ETP_WORKER_CLEAR (wrk);
205	}	374	}
206		375
		376	static void ecb_cold
207	static void etp_worker_free (etp_worker *wrk)	377	etp_worker_free (etp_worker *wrk)
208	{	378	{
209	wrk->next->prev = wrk->prev;	379	wrk->next->prev = wrk->prev;
210	wrk->prev->next = wrk->next;	380	wrk->prev->next = wrk->next;
211		381
212	free (wrk);	382	free (wrk);
213	}	383	}
214		384
215	static unsigned int etp_nreqs (void)	385	static unsigned int
		386	etp_nreqs (void)
216	{	387	{
217	int retval;	388	int retval;
218	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	389	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
219	retval = nreqs;	390	retval = nreqs;
220	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	391	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
221	return retval;	392	return retval;
222	}	393	}
223		394
224	static unsigned int etp_nready (void)	395	static unsigned int
		396	etp_nready (void)
225	{	397	{
226	unsigned int retval;	398	unsigned int retval;
227		399
228	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	400	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
229	retval = nready;	401	retval = nready;
230	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	402	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
231		403
232	return retval;	404	return retval;
233	}	405	}
234		406
235	static unsigned int etp_npending (void)	407	static unsigned int
		408	etp_npending (void)
236	{	409	{
237	unsigned int retval;	410	unsigned int retval;
238		411
239	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	412	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
240	retval = npending;	413	retval = npending;
241	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	414	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
242		415
243	return retval;	416	return retval;
244	}	417	}
245		418
246	static unsigned int etp_nthreads (void)	419	static unsigned int
		420	etp_nthreads (void)
247	{	421	{
248	unsigned int retval;	422	unsigned int retval;
249		423
250	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	424	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
251	retval = started;	425	retval = started;
…		…
265	} etp_reqq;	439	} etp_reqq;
266		440
267	static etp_reqq req_queue;	441	static etp_reqq req_queue;
268	static etp_reqq res_queue;	442	static etp_reqq res_queue;
269		443
		444	static void ecb_noinline ecb_cold
		445	reqq_init (etp_reqq *q)
		446	{
		447	int pri;
		448
		449	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		450	q->qs[pri] = q->qe[pri] = 0;
		451
		452	q->size = 0;
		453	}
		454
		455	static int ecb_noinline
270	static int reqq_push (etp_reqq *q, ETP_REQ *req)	456	reqq_push (etp_reqq *q, ETP_REQ *req)
271	{	457	{
272	int pri = req->pri;	458	int pri = req->pri;
273	req->next = 0;	459	req->next = 0;
274		460
275	if (q->qe[pri])	461	if (q->qe[pri])
…		…
281	q->qe[pri] = q->qs[pri] = req;	467	q->qe[pri] = q->qs[pri] = req;
282		468
283	return q->size++;	469	return q->size++;
284	}	470	}
285		471
		472	static ETP_REQ * ecb_noinline
286	static ETP_REQ *reqq_shift (etp_reqq *q)	473	reqq_shift (etp_reqq *q)
287	{	474	{
288	int pri;	475	int pri;
289		476
290	if (!q->size)	477	if (!q->size)
291	return 0;	478	return 0;
…		…
306	}	493	}
307		494
308	abort ();	495	abort ();
309	}	496	}
310		497
311	static void etp_atfork_prepare (void)	498	static int ecb_cold
		499	etp_init (void (*want_poll)(void), void (*done_poll)(void))
312	{	500	{
313	X_LOCK (wrklock);	501	X_MUTEX_CREATE (wrklock);
314	X_LOCK (reqlock);	502	X_MUTEX_CREATE (reslock);
315	X_LOCK (reslock);	503	X_MUTEX_CREATE (reqlock);
316	#if !HAVE_PREADWRITE	504	X_COND_CREATE (reqwait);
317	X_LOCK (preadwritelock);
318	#endif
319	}
320		505
321	static void etp_atfork_parent (void)	506	reqq_init (&req_queue);
322	{	507	reqq_init (&res_queue);
323	#if !HAVE_PREADWRITE
324	X_UNLOCK (preadwritelock);
325	#endif
326	X_UNLOCK (reslock);
327	X_UNLOCK (reqlock);
328	X_UNLOCK (wrklock);
329	}
330		508
331	static void etp_atfork_child (void)	509	wrk_first.next =
332	{	510	wrk_first.prev = &wrk_first;
333	ETP_REQ *prv;
334
335	while (prv = reqq_shift (&req_queue))
336	ETP_DESTROY (prv);
337
338	while (prv = reqq_shift (&res_queue))
339	ETP_DESTROY (prv);
340
341	while (wrk_first.next != &wrk_first)
342	{
343	etp_worker *wrk = wrk_first.next;
344
345	if (wrk->req)
346	ETP_DESTROY (wrk->req);
347
348	etp_worker_clear (wrk);
349	etp_worker_free (wrk);
350	}
351		511
352	started = 0;	512	started = 0;
353	idle = 0;	513	idle = 0;
354	nreqs = 0;	514	nreqs = 0;
355	nready = 0;	515	nready = 0;
356	npending = 0;	516	npending = 0;
357		517
358	etp_atfork_parent ();
359	}
360
361	static void
362	etp_once_init (void)
363	{
364	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
365	}
366
367	static int
368	etp_init (void (*want_poll)(void), void (*done_poll)(void))
369	{
370	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
371
372	pthread_once (&doinit, etp_once_init);
373
374	want_poll_cb = want_poll;	518	want_poll_cb = want_poll;
375	done_poll_cb = done_poll;	519	done_poll_cb = done_poll;
		520
		521	return 0;
376	}	522	}
377		523
378	X_THREAD_PROC (etp_proc);	524	X_THREAD_PROC (etp_proc);
379		525
		526	static void ecb_cold
380	static void etp_start_thread (void)	527	etp_start_thread (void)
381	{	528	{
382	etp_worker *wrk = calloc (1, sizeof (etp_worker));	529	etp_worker *wrk = calloc (1, sizeof (etp_worker));
383		530
384	/*TODO*/	531	/*TODO*/
385	assert (("unable to allocate worker thread data", wrk));	532	assert (("unable to allocate worker thread data", wrk));
…		…
398	free (wrk);	545	free (wrk);
399		546
400	X_UNLOCK (wrklock);	547	X_UNLOCK (wrklock);
401	}	548	}
402		549
		550	static void
403	static void etp_maybe_start_thread (void)	551	etp_maybe_start_thread (void)
404	{	552	{
405	if (etp_nthreads () >= wanted)	553	if (ecb_expect_true (etp_nthreads () >= wanted))
406	return;	554	return;
407		555
408	/* todo: maybe use idle here, but might be less exact */	556	/* todo: maybe use idle here, but might be less exact */
409	if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())	557	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
410	return;	558	return;
411		559
412	etp_start_thread ();	560	etp_start_thread ();
413	}	561	}
414		562
		563	static void ecb_cold
415	static void etp_end_thread (void)	564	etp_end_thread (void)
416	{	565	{
417	eio_req *req = calloc (1, sizeof (eio_req));	566	eio_req *req = calloc (1, sizeof (eio_req));
418		567
419	req->type = -1;	568	req->type = -1;
420	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	569	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
427	X_LOCK (wrklock);	576	X_LOCK (wrklock);
428	--started;	577	--started;
429	X_UNLOCK (wrklock);	578	X_UNLOCK (wrklock);
430	}	579	}
431		580
432	static int etp_poll (void)	581	static int
		582	etp_poll (void)
433	{	583	{
434	unsigned int maxreqs;	584	unsigned int maxreqs;
435	unsigned int maxtime;	585	unsigned int maxtime;
436	struct timeval tv_start, tv_now;	586	struct timeval tv_start, tv_now;
437		587
…		…
467		617
468	X_LOCK (reqlock);	618	X_LOCK (reqlock);
469	--nreqs;	619	--nreqs;
470	X_UNLOCK (reqlock);	620	X_UNLOCK (reqlock);
471		621
472	if (req->type == EIO_GROUP && req->size)	622	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
473	{	623	{
474	req->int1 = 1; /* mark request as delayed */	624	req->int1 = 1; /* mark request as delayed */
475	continue;	625	continue;
476	}	626	}
477	else	627	else
478	{	628	{
479	int res = ETP_FINISH (req);	629	int res = ETP_FINISH (req);
480	if (res)	630	if (ecb_expect_false (res))
481	return res;	631	return res;
482	}	632	}
483		633
484	if (maxreqs && !--maxreqs)	634	if (ecb_expect_false (maxreqs && !--maxreqs))
485	break;	635	break;
486		636
487	if (maxtime)	637	if (maxtime)
488	{	638	{
489	gettimeofday (&tv_now, 0);	639	gettimeofday (&tv_now, 0);
…		…
495		645
496	errno = EAGAIN;	646	errno = EAGAIN;
497	return -1;	647	return -1;
498	}	648	}
499		649
		650	static void
500	static void etp_cancel (ETP_REQ *req)	651	etp_cancel (ETP_REQ *req)
501	{	652	{
502	X_LOCK (wrklock);	653	req->cancelled = 1;
503	req->flags |= EIO_FLAG_CANCELLED;
504	X_UNLOCK (wrklock);
505		654
506	eio_grp_cancel (req);	655	eio_grp_cancel (req);
507	}	656	}
508		657
		658	static void
509	static void etp_submit (ETP_REQ *req)	659	etp_submit (ETP_REQ *req)
510	{	660	{
511	req->pri -= ETP_PRI_MIN;	661	req->pri -= ETP_PRI_MIN;
512		662
513	if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	663	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
514	if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	664	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
515		665
		666	if (ecb_expect_false (req->type == EIO_GROUP))
		667	{
		668	/* I hope this is worth it :/ */
516	X_LOCK (reqlock);	669	X_LOCK (reqlock);
517	++nreqs;	670	++nreqs;
		671	X_UNLOCK (reqlock);
		672
		673	X_LOCK (reslock);
		674
		675	++npending;
		676
		677	if (!reqq_push (&res_queue, req) && want_poll_cb)
		678	want_poll_cb ();
		679
		680	X_UNLOCK (reslock);
		681	}
		682	else
		683	{
		684	X_LOCK (reqlock);
		685	++nreqs;
518	++nready;	686	++nready;
519	reqq_push (&req_queue, req);	687	reqq_push (&req_queue, req);
520	X_COND_SIGNAL (reqwait);	688	X_COND_SIGNAL (reqwait);
521	X_UNLOCK (reqlock);	689	X_UNLOCK (reqlock);
522		690
523	etp_maybe_start_thread ();	691	etp_maybe_start_thread ();
		692	}
524	}	693	}
525		694
		695	static void ecb_cold
526	static void etp_set_max_poll_time (double nseconds)	696	etp_set_max_poll_time (double nseconds)
527	{	697	{
528	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	698	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
529	max_poll_time = nseconds;	699	max_poll_time = nseconds * EIO_TICKS;
530	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	700	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
531	}	701	}
532		702
		703	static void ecb_cold
533	static void etp_set_max_poll_reqs (unsigned int maxreqs)	704	etp_set_max_poll_reqs (unsigned int maxreqs)
534	{	705	{
535	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	706	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
536	max_poll_reqs = maxreqs;	707	max_poll_reqs = maxreqs;
537	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	708	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
538	}	709	}
539		710
		711	static void ecb_cold
540	static void etp_set_max_idle (unsigned int nthreads)	712	etp_set_max_idle (unsigned int nthreads)
541	{	713	{
542	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	714	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
543	max_idle = nthreads <= 0 ? 1 : nthreads;	715	max_idle = nthreads;
544	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	716	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
545	}	717	}
546		718
		719	static void ecb_cold
		720	etp_set_idle_timeout (unsigned int seconds)
		721	{
		722	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		723	idle_timeout = seconds;
		724	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		725	}
		726
		727	static void ecb_cold
547	static void etp_set_min_parallel (unsigned int nthreads)	728	etp_set_min_parallel (unsigned int nthreads)
548	{	729	{
549	if (wanted < nthreads)	730	if (wanted < nthreads)
550	wanted = nthreads;	731	wanted = nthreads;
551	}	732	}
552		733
		734	static void ecb_cold
553	static void etp_set_max_parallel (unsigned int nthreads)	735	etp_set_max_parallel (unsigned int nthreads)
554	{	736	{
555	if (wanted > nthreads)	737	if (wanted > nthreads)
556	wanted = nthreads;	738	wanted = nthreads;
557		739
558	while (started > wanted)	740	while (started > wanted)
559	etp_end_thread ();	741	etp_end_thread ();
560	}	742	}
561		743
562	/*****************************************************************************/	744	/*****************************************************************************/
563		745
		746	static void
564	static void grp_try_feed (eio_req *grp)	747	grp_try_feed (eio_req *grp)
565	{	748	{
566	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	749	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
567	{	750	{
568	int old_len = grp->size;	751	grp->flags &= ~EIO_FLAG_GROUPADD;
569		752
570	EIO_FEED (grp);	753	EIO_FEED (grp);
571		754
572	/* stop if no progress has been made */	755	/* stop if no progress has been made */
573	if (old_len == grp->size)	756	if (!(grp->flags & EIO_FLAG_GROUPADD))
574	{	757	{
575	grp->feed = 0;	758	grp->feed = 0;
576	break;	759	break;
577	}	760	}
578	}	761	}
579	}	762	}
580		763
		764	static int
581	static int grp_dec (eio_req *grp)	765	grp_dec (eio_req *grp)
582	{	766	{
583	--grp->size;	767	--grp->size;
584		768
585	/* call feeder, if applicable */	769	/* call feeder, if applicable */
586	grp_try_feed (grp);	770	grp_try_feed (grp);
…		…
590	return eio_finish (grp);	774	return eio_finish (grp);
591	else	775	else
592	return 0;	776	return 0;
593	}	777	}
594		778
		779	static void
595	void eio_destroy (eio_req *req)	780	eio_destroy (eio_req *req)
596	{	781	{
597	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	782	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
598	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	783	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
599		784
600	EIO_DESTROY (req);	785	EIO_DESTROY (req);
601	}	786	}
602		787
		788	static int
603	static int eio_finish (eio_req *req)	789	eio_finish (eio_req *req)
604	{	790	{
605	int res = EIO_FINISH (req);	791	int res = EIO_FINISH (req);
606		792
607	if (req->grp)	793	if (req->grp)
608	{	794	{
…		…
616	if (grp->grp_first == req)	802	if (grp->grp_first == req)
617	grp->grp_first = req->grp_next;	803	grp->grp_first = req->grp_next;
618		804
619	res2 = grp_dec (grp);	805	res2 = grp_dec (grp);
620		806
621	if (!res && res2)	807	if (!res)
622	res = res2;	808	res = res2;
623	}	809	}
624		810
625	eio_destroy (req);	811	eio_destroy (req);
626		812
627	return res;	813	return res;
628	}	814	}
629		815
		816	void
630	void eio_grp_cancel (eio_req *grp)	817	eio_grp_cancel (eio_req *grp)
631	{	818	{
632	for (grp = grp->grp_first; grp; grp = grp->grp_next)	819	for (grp = grp->grp_first; grp; grp = grp->grp_next)
633	eio_cancel (grp);	820	eio_cancel (grp);
634	}	821	}
635		822
		823	void
636	void eio_cancel (eio_req *req)	824	eio_cancel (eio_req *req)
637	{	825	{
638	etp_cancel (req);	826	etp_cancel (req);
639	}	827	}
640		828
		829	void
641	void eio_submit (eio_req *req)	830	eio_submit (eio_req *req)
642	{	831	{
643	etp_submit (req);	832	etp_submit (req);
644	}	833	}
645		834
646	unsigned int eio_nreqs (void)	835	unsigned int
		836	eio_nreqs (void)
647	{	837	{
648	return etp_nreqs ();	838	return etp_nreqs ();
649	}	839	}
650		840
651	unsigned int eio_nready (void)	841	unsigned int
		842	eio_nready (void)
652	{	843	{
653	return etp_nready ();	844	return etp_nready ();
654	}	845	}
655		846
656	unsigned int eio_npending (void)	847	unsigned int
		848	eio_npending (void)
657	{	849	{
658	return etp_npending ();	850	return etp_npending ();
659	}	851	}
660		852
661	unsigned int eio_nthreads (void)	853	unsigned int ecb_cold
		854	eio_nthreads (void)
662	{	855	{
663	return etp_nthreads ();	856	return etp_nthreads ();
664	}	857	}
665		858
		859	void ecb_cold
666	void eio_set_max_poll_time (double nseconds)	860	eio_set_max_poll_time (double nseconds)
667	{	861	{
668	etp_set_max_poll_time (nseconds);	862	etp_set_max_poll_time (nseconds);
669	}	863	}
670		864
		865	void ecb_cold
671	void eio_set_max_poll_reqs (unsigned int maxreqs)	866	eio_set_max_poll_reqs (unsigned int maxreqs)
672	{	867	{
673	etp_set_max_poll_reqs (maxreqs);	868	etp_set_max_poll_reqs (maxreqs);
674	}	869	}
675		870
		871	void ecb_cold
676	void eio_set_max_idle (unsigned int nthreads)	872	eio_set_max_idle (unsigned int nthreads)
677	{	873	{
678	etp_set_max_idle (nthreads);	874	etp_set_max_idle (nthreads);
679	}	875	}
680		876
		877	void ecb_cold
		878	eio_set_idle_timeout (unsigned int seconds)
		879	{
		880	etp_set_idle_timeout (seconds);
		881	}
		882
		883	void ecb_cold
681	void eio_set_min_parallel (unsigned int nthreads)	884	eio_set_min_parallel (unsigned int nthreads)
682	{	885	{
683	etp_set_min_parallel (nthreads);	886	etp_set_min_parallel (nthreads);
684	}	887	}
685		888
		889	void ecb_cold
686	void eio_set_max_parallel (unsigned int nthreads)	890	eio_set_max_parallel (unsigned int nthreads)
687	{	891	{
688	etp_set_max_parallel (nthreads);	892	etp_set_max_parallel (nthreads);
689	}	893	}
690		894
691	int eio_poll (void)	895	int eio_poll (void)
…		…
695		899
696	/*****************************************************************************/	900	/*****************************************************************************/
697	/* work around various missing functions */	901	/* work around various missing functions */
698		902
699	#if !HAVE_PREADWRITE	903	#if !HAVE_PREADWRITE
		904	# undef pread
		905	# undef pwrite
700	# define pread eio__pread	906	# define pread eio__pread
701	# define pwrite eio__pwrite	907	# define pwrite eio__pwrite
702		908
703	/*
704	* make our pread/pwrite safe against themselves, but not against
705	* normal read/write by using a mutex. slows down execution a lot,
706	* but that's your problem, not mine.
707	*/
708	static mutex_t preadwritelock = X_MUTEX_INIT;
709
710	static ssize_t	909	static eio_ssize_t
711	eio__pread (int fd, void *buf, size_t count, off_t offset)	910	eio__pread (int fd, void *buf, size_t count, off_t offset)
712	{	911	{
713	ssize_t res;	912	eio_ssize_t res;
714	off_t ooffset;	913	off_t ooffset;
715		914
716	X_LOCK (preadwritelock);	915	X_LOCK (preadwritelock);
717	ooffset = lseek (fd, 0, SEEK_CUR);	916	ooffset = lseek (fd, 0, SEEK_CUR);
718	lseek (fd, offset, SEEK_SET);	917	lseek (fd, offset, SEEK_SET);
…		…
721	X_UNLOCK (preadwritelock);	920	X_UNLOCK (preadwritelock);
722		921
723	return res;	922	return res;
724	}	923	}
725		924
726	static ssize_t	925	static eio_ssize_t
727	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	926	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
728	{	927	{
729	ssize_t res;	928	eio_ssize_t res;
730	off_t ooffset;	929	off_t ooffset;
731		930
732	X_LOCK (preadwritelock);	931	X_LOCK (preadwritelock);
733	ooffset = lseek (fd, 0, SEEK_CUR);	932	ooffset = lseek (fd, 0, SEEK_CUR);
734	lseek (fd, offset, SEEK_SET);	933	lseek (fd, offset, SEEK_SET);
735	res = write (fd, buf, count);	934	res = write (fd, buf, count);
736	lseek (fd, offset, SEEK_SET);	935	lseek (fd, ooffset, SEEK_SET);
737	X_UNLOCK (preadwritelock);	936	X_UNLOCK (preadwritelock);
738		937
739	return res;	938	return res;
740	}	939	}
741	#endif	940	#endif
742		941
743	#ifndef HAVE_FUTIMES	942	#ifndef HAVE_UTIMES
744		943
		944	# undef utimes
745	# define utimes(path,times) eio__utimes (path, times)	945	# define utimes(path,times) eio__utimes (path, times)
746	# define futimes(fd,times) eio__futimes (fd, times)
747		946
748	static int	947	static int
749	eio__utimes (const char *filename, const struct timeval times[2])	948	eio__utimes (const char *filename, const struct timeval times[2])
750	{	949	{
751	if (times)	950	if (times)
…		…
759	}	958	}
760	else	959	else
761	return utime (filename, 0);	960	return utime (filename, 0);
762	}	961	}
763		962
		963	#endif
		964
		965	#ifndef HAVE_FUTIMES
		966
		967	# undef futimes
		968	# define futimes(fd,times) eio__futimes (fd, times)
		969
		970	static int
764	static int eio__futimes (int fd, const struct timeval tv[2])	971	eio__futimes (int fd, const struct timeval tv[2])
765	{	972	{
766	errno = ENOSYS;	973	errno = ENOSYS;
767	return -1;	974	return -1;
768	}	975	}
769		976
770	#endif	977	#endif
771		978
772	#if !HAVE_FDATASYNC	979	#if !HAVE_FDATASYNC
		980	# undef fdatasync
773	# define fdatasync fsync	981	# define fdatasync(fd) fsync (fd)
774	#endif	982	#endif
		983
		984	static int
		985	eio__syncfs (int fd)
		986	{
		987	int res;
		988
		989	#if HAVE_SYS_SYNCFS
		990	res = (int)syscall (__NR_syncfs, (int)(fd));
		991	#else
		992	res = -1;
		993	errno = ENOSYS;
		994	#endif
		995
		996	if (res < 0 && errno == ENOSYS && fd >= 0)
		997	sync ();
		998
		999	return res;
		1000	}
		1001
		1002	/* sync_file_range always needs emulation */
		1003	static int
		1004	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		1005	{
		1006	#if HAVE_SYNC_FILE_RANGE
		1007	int res;
		1008
		1009	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		1010	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		1011	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		1012	{
		1013	flags = 0
		1014	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		1015	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		1016	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		1017	}
		1018
		1019	res = sync_file_range (fd, offset, nbytes, flags);
		1020
		1021	if (!res || errno != ENOSYS)
		1022	return res;
		1023	#endif
		1024
		1025	/* even though we could play tricks with the flags, it's better to always
		1026	* call fdatasync, as that matches the expectation of its users best */
		1027	return fdatasync (fd);
		1028	}
		1029
		1030	static int
		1031	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		1032	{
		1033	#if HAVE_FALLOCATE
		1034	return fallocate (fd, mode, offset, len);
		1035	#else
		1036	errno = ENOSYS;
		1037	return -1;
		1038	#endif
		1039	}
775		1040
776	#if !HAVE_READAHEAD	1041	#if !HAVE_READAHEAD
		1042	# undef readahead
777	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1043	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
778		1044
779	static ssize_t	1045	static eio_ssize_t
780	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1046	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
781	{	1047	{
782	size_t todo = count;	1048	size_t todo = count;
783	dBUF;	1049	dBUF;
784		1050
…		…
789	pread (fd, eio_buf, len, offset);	1055	pread (fd, eio_buf, len, offset);
790	offset += len;	1056	offset += len;
791	todo -= len;	1057	todo -= len;
792	}	1058	}
793		1059
		1060	FUBd;
		1061
794	errno = 0;	1062	errno = 0;
795	return count;	1063	return count;
796	}	1064	}
797		1065
798	#endif	1066	#endif
799		1067
800	/* sendfile always needs emulation */	1068	/* sendfile always needs emulation */
801	static ssize_t	1069	static eio_ssize_t
802	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1070	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
803	{	1071	{
		1072	eio_ssize_t written = 0;
804	ssize_t res;	1073	eio_ssize_t res;
805		1074
806	if (!count)	1075	if (!count)
807	return 0;	1076	return 0;
808		1077
		1078	for (;;)
		1079	{
		1080	#ifdef __APPLE__
		1081	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1082	#endif
809	#if HAVE_SENDFILE	1083	#if HAVE_SENDFILE
810	# if __linux	1084	# if __linux
		1085	off_t soffset = offset;
811	res = sendfile (ofd, ifd, &offset, count);	1086	res = sendfile (ofd, ifd, &soffset, count);
812		1087
813	# elif __freebsd	1088	# elif __FreeBSD__
814	/*	1089	/*
815	* Of course, the freebsd sendfile is a dire hack with no thoughts	1090	* Of course, the freebsd sendfile is a dire hack with no thoughts
816	* wasted on making it similar to other I/O functions.	1091	* wasted on making it similar to other I/O functions.
817	*/	1092	*/
818	{
819	off_t sbytes;	1093	off_t sbytes;
820	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1094	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
821		1095
822	if (res < 0 && sbytes)	1096	#if 0 /* according to the manpage, this is correct, but broken behaviour */
823	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1097	/* freebsd' sendfile will return 0 on success */
		1098	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1099	/* not on e.g. EIO or EPIPE - sounds broken */
		1100	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
824	res = sbytes;	1101	res = sbytes;
825	}	1102	#endif
		1103
		1104	/* according to source inspection, this is correct, and useful behaviour */
		1105	if (sbytes)
		1106	res = sbytes;
		1107
		1108	# elif defined (__APPLE__)
		1109	off_t sbytes = count;
		1110	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1111
		1112	/* according to the manpage, sbytes is always valid */
		1113	if (sbytes)
		1114	res = sbytes;
826		1115
827	# elif __hpux	1116	# elif __hpux
828	res = sendfile (ofd, ifd, offset, count, 0, 0);	1117	res = sendfile (ofd, ifd, offset, count, 0, 0);
829		1118
830	# elif __solaris	1119	# elif __solaris
831	{
832	struct sendfilevec vec;	1120	struct sendfilevec vec;
833	size_t sbytes;	1121	size_t sbytes;
834		1122
835	vec.sfv_fd = ifd;	1123	vec.sfv_fd = ifd;
836	vec.sfv_flag = 0;	1124	vec.sfv_flag = 0;
837	vec.sfv_off = offset;	1125	vec.sfv_off = offset;
838	vec.sfv_len = count;	1126	vec.sfv_len = count;
839		1127
840	res = sendfilev (ofd, &vec, 1, &sbytes);	1128	res = sendfilev (ofd, &vec, 1, &sbytes);
841		1129
842	if (res < 0 && sbytes)	1130	if (res < 0 && sbytes)
843	res = sbytes;	1131	res = sbytes;
844	}
845		1132
846	# endif	1133	# endif
		1134
		1135	#elif defined (_WIN32) && 0
		1136	/* does not work, just for documentation of what would need to be done */
		1137	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1138	/* libeio guarantees that the file offset does not change, and windows */
		1139	/* has no way to get an independent handle to the same file description */
		1140	HANDLE h = TO_SOCKET (ifd);
		1141	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1142	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1143
847	#else	1144	#else
848	res = -1;	1145	res = -1;
849	errno = ENOSYS;	1146	errno = ENOSYS;
850	#endif	1147	#endif
851		1148
		1149	/* we assume sendfile can copy at least 128mb in one go */
		1150	if (res <= 128 * 1024 * 1024)
		1151	{
		1152	if (res > 0)
		1153	written += res;
		1154
		1155	if (written)
		1156	return written;
		1157
		1158	break;
		1159	}
		1160	else
		1161	{
		1162	/* if we requested more, then probably the kernel was lazy */
		1163	written += res;
		1164	offset += res;
		1165	count -= res;
		1166
		1167	if (!count)
		1168	return written;
		1169	}
		1170	}
		1171
852	if (res < 0	1172	if (res < 0
853	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1173	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1174	/* BSDs */
		1175	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1176	|| errno == ENOTSUP
		1177	#endif
		1178	#ifdef EOPNOTSUPP /* windows */
		1179	|| errno == EOPNOTSUPP /* BSDs */
		1180	#endif
854	#if __solaris	1181	#if __solaris
855	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1182	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
856	#endif	1183	#endif
857	)	1184	)
858	)	1185	)
…		…
862		1189
863	res = 0;	1190	res = 0;
864		1191
865	while (count)	1192	while (count)
866	{	1193	{
867	ssize_t cnt;	1194	eio_ssize_t cnt;
868		1195
869	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1196	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
870		1197
871	if (cnt <= 0)	1198	if (cnt <= 0)
872	{	1199	{
…		…
884		1211
885	offset += cnt;	1212	offset += cnt;
886	res += cnt;	1213	res += cnt;
887	count -= cnt;	1214	count -= cnt;
888	}	1215	}
		1216
		1217	FUBd;
889	}	1218	}
890		1219
891	return res;	1220	return res;
		1221	}
		1222
		1223	#ifdef PAGESIZE
		1224	# define eio_pagesize() PAGESIZE
		1225	#else
		1226	static intptr_t
		1227	eio_pagesize (void)
		1228	{
		1229	static intptr_t page;
		1230
		1231	if (!page)
		1232	page = sysconf (_SC_PAGESIZE);
		1233
		1234	return page;
		1235	}
		1236	#endif
		1237
		1238	static void
		1239	eio_page_align (void **addr, size_t *length)
		1240	{
		1241	intptr_t mask = eio_pagesize () - 1;
		1242
		1243	/* round down addr */
		1244	intptr_t adj = mask & (intptr_t)*addr;
		1245
		1246	*addr = (void *)((intptr_t)*addr - adj);
		1247	*length += adj;
		1248
		1249	/* round up length */
		1250	*length = (*length + mask) & ~mask;
		1251	}
		1252
		1253	#if !_POSIX_MEMLOCK
		1254	# define eio__mlockall(a) EIO_ENOSYS ()
		1255	#else
		1256
		1257	static int
		1258	eio__mlockall (int flags)
		1259	{
		1260	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1261	extern int mallopt (int, int);
		1262	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1263	#endif
		1264
		1265	if (EIO_MCL_CURRENT != MCL_CURRENT
		1266	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1267	{
		1268	flags = 0
		1269	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1270	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1271	}
		1272
		1273	return mlockall (flags);
		1274	}
		1275	#endif
		1276
		1277	#if !_POSIX_MEMLOCK_RANGE
		1278	# define eio__mlock(a,b) EIO_ENOSYS ()
		1279	#else
		1280
		1281	static int
		1282	eio__mlock (void *addr, size_t length)
		1283	{
		1284	eio_page_align (&addr, &length);
		1285
		1286	return mlock (addr, length);
		1287	}
		1288
		1289	#endif
		1290
		1291	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1292	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1293	#else
		1294
		1295	static int
		1296	eio__msync (void *mem, size_t len, int flags)
		1297	{
		1298	eio_page_align (&mem, &len);
		1299
		1300	if (EIO_MS_ASYNC != MS_SYNC
		1301	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1302	|| EIO_MS_SYNC != MS_SYNC)
		1303	{
		1304	flags = 0
		1305	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1306	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1307	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1308	}
		1309
		1310	return msync (mem, len, flags);
		1311	}
		1312
		1313	#endif
		1314
		1315	static int
		1316	eio__mtouch (eio_req *req)
		1317	{
		1318	void *mem = req->ptr2;
		1319	size_t len = req->size;
		1320	int flags = req->int1;
		1321
		1322	eio_page_align (&mem, &len);
		1323
		1324	{
		1325	intptr_t addr = (intptr_t)mem;
		1326	intptr_t end = addr + len;
		1327	intptr_t page = eio_pagesize ();
		1328
		1329	if (addr < end)
		1330	if (flags & EIO_MT_MODIFY) /* modify */
		1331	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1332	else
		1333	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1334	}
		1335
		1336	return 0;
		1337	}
		1338
		1339	/*****************************************************************************/
		1340	/* requests implemented outside eio_execute, because they are so large */
		1341
		1342	static void
		1343	eio__realpath (eio_req *req, etp_worker *self)
		1344	{
		1345	char *rel = req->ptr1;
		1346	char *res;
		1347	char *tmp1, *tmp2;
		1348	#if SYMLOOP_MAX > 32
		1349	int symlinks = SYMLOOP_MAX;
		1350	#else
		1351	int symlinks = 32;
		1352	#endif
		1353
		1354	req->result = -1;
		1355
		1356	errno = EINVAL;
		1357	if (!rel)
		1358	return;
		1359
		1360	errno = ENOENT;
		1361	if (!*rel)
		1362	return;
		1363
		1364	if (!req->ptr2)
		1365	{
		1366	X_LOCK (wrklock);
		1367	req->flags |= EIO_FLAG_PTR2_FREE;
		1368	X_UNLOCK (wrklock);
		1369	req->ptr2 = malloc (PATH_MAX * 3);
		1370
		1371	errno = ENOMEM;
		1372	if (!req->ptr2)
		1373	return;
		1374	}
		1375
		1376	res = req->ptr2;
		1377	tmp1 = res + PATH_MAX;
		1378	tmp2 = tmp1 + PATH_MAX;
		1379
		1380	#if 0 /* disabled, the musl way to do things is just too racy */
		1381	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1382	/* on linux we may be able to ask the kernel */
		1383	{
		1384	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1385
		1386	if (fd >= 0)
		1387	{
		1388	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1389	req->result = readlink (tmp1, res, PATH_MAX);
		1390	close (fd);
		1391
		1392	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1393
		1394	if (req->result > 0)
		1395	goto done;
		1396	}
		1397	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1398	return;
		1399	}
		1400	#endif
		1401	#endif
		1402
		1403	if (*rel != '/')
		1404	{
		1405	if (!getcwd (res, PATH_MAX))
		1406	return;
		1407
		1408	if (res [1]) /* only use if not / */
		1409	res += strlen (res);
		1410	}
		1411
		1412	while (*rel)
		1413	{
		1414	eio_ssize_t len, linklen;
		1415	char *beg = rel;
		1416
		1417	while (*rel && *rel != '/')
		1418	++rel;
		1419
		1420	len = rel - beg;
		1421
		1422	if (!len) /* skip slashes */
		1423	{
		1424	++rel;
		1425	continue;
		1426	}
		1427
		1428	if (beg [0] == '.')
		1429	{
		1430	if (len == 1)
		1431	continue; /* . - nop */
		1432
		1433	if (beg [1] == '.' && len == 2)
		1434	{
		1435	/* .. - back up one component, if possible */
		1436
		1437	while (res != req->ptr2)
		1438	if (*--res == '/')
		1439	break;
		1440
		1441	continue;
		1442	}
		1443	}
		1444
		1445	errno = ENAMETOOLONG;
		1446	if (res + 1 + len + 1 >= tmp1)
		1447	return;
		1448
		1449	/* copy one component */
		1450	*res = '/';
		1451	memcpy (res + 1, beg, len);
		1452
		1453	/* zero-terminate, for readlink */
		1454	res [len + 1] = 0;
		1455
		1456	/* now check if it's a symlink */
		1457	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1458
		1459	if (linklen < 0)
		1460	{
		1461	if (errno != EINVAL)
		1462	return;
		1463
		1464	/* it's a normal directory. hopefully */
		1465	res += len + 1;
		1466	}
		1467	else
		1468	{
		1469	/* yay, it was a symlink - build new path in tmp2 */
		1470	int rellen = strlen (rel);
		1471
		1472	errno = ENAMETOOLONG;
		1473	if (linklen + 1 + rellen >= PATH_MAX)
		1474	return;
		1475
		1476	errno = ELOOP;
		1477	if (!--symlinks)
		1478	return;
		1479
		1480	if (*tmp1 == '/')
		1481	res = req->ptr2; /* symlink resolves to an absolute path */
		1482
		1483	/* we need to be careful, as rel might point into tmp2 already */
		1484	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1485	tmp2 [linklen] = '/';
		1486	memcpy (tmp2, tmp1, linklen);
		1487
		1488	rel = tmp2;
		1489	}
		1490	}
		1491
		1492	/* special case for the lone root path */
		1493	if (res == req->ptr2)
		1494	*res++ = '/';
		1495
		1496	req->result = res - (char *)req->ptr2;
		1497
		1498	done:
		1499	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1500	}
		1501
		1502	static signed char
		1503	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1504	{
		1505	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1506	: a->inode < b->inode ? -1
		1507	: a->inode > b->inode ? 1
		1508	: 0;
		1509	}
		1510
		1511	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1512
		1513	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1514	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1515
		1516	static void
		1517	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1518	{
		1519	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
		1520	unsigned char *bit = bits;
		1521
		1522	assert (CHAR_BIT == 8);
		1523	assert (sizeof (eio_dirent) * 8 < 256);
		1524	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1525	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1526
		1527	if (size <= EIO_SORT_FAST)
		1528	return;
		1529
		1530	/* first prepare an array of bits to test in our radix sort */
		1531	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
		1532	/* inode_bits must contain all inodes ORed together */
		1533	/* which is used to skip bits that are 0 everywhere, which is very common */
		1534	{
		1535	eio_ino_t endianness;
		1536	int i, j;
		1537
		1538	/* we store the byte offset of byte n into byte n of "endianness" */
		1539	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1540	((unsigned char *)&endianness)[i] = i;
		1541
		1542	*bit++ = 0;
		1543
		1544	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1545	{
		1546	/* shifting off the byte offsets out of "endianness" */
		1547	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1548	endianness >>= 8;
		1549
		1550	for (j = 0; j < 8; ++j)
		1551	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
		1552	*bit++ = offs + j;
		1553	}
		1554
		1555	for (j = 0; j < 8; ++j)
		1556	if (score_bits & (1 << j))
		1557	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1558	}
		1559
		1560	/* now actually do the sorting (a variant of MSD radix sort) */
		1561	{
		1562	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
		1563	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
		1564	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
		1565	int stk_idx = 0;
		1566
		1567	base_stk [stk_idx] = dents;
		1568	end_stk [stk_idx] = dents + size;
		1569	bit_stk [stk_idx] = bit - 1;
		1570
		1571	do
		1572	{
		1573	base = base_stk [stk_idx];
		1574	end = end_stk [stk_idx];
		1575	bit = bit_stk [stk_idx];
		1576
		1577	for (;;)
		1578	{
		1579	unsigned char O = *bit >> 3;
		1580	unsigned char M = 1 << (*bit & 7);
		1581
		1582	eio_dirent *a = base;
		1583	eio_dirent *b = end;
		1584
		1585	if (b - a < EIO_SORT_CUTOFF)
		1586	break;
		1587
		1588	/* now bit-partition the array on the bit */
		1589	/* this ugly asymmetric loop seems to perform much better than typical */
		1590	/* partition algos found in the literature */
		1591	do
		1592	if (!(((unsigned char *)a)[O] & M))
		1593	++a;
		1594	else if (!(((unsigned char *)--b)[O] & M))
		1595	{
		1596	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1597	++a;
		1598	}
		1599	while (b > a);
		1600
		1601	/* next bit, or stop, if no bits left in this path */
		1602	if (!*--bit)
		1603	break;
		1604
		1605	base_stk [stk_idx] = a;
		1606	end_stk [stk_idx] = end;
		1607	bit_stk [stk_idx] = bit;
		1608	++stk_idx;
		1609
		1610	end = a;
		1611	}
		1612	}
		1613	while (stk_idx--);
		1614	}
		1615	}
		1616
		1617	static void
		1618	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1619	{
		1620	/* first move the smallest element to the front, to act as a sentinel */
		1621	{
		1622	int i;
		1623	eio_dirent *min = dents;
		1624
		1625	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1626	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1627	if (EIO_DENT_CMP (dents [i], <, *min))
		1628	min = &dents [i];
		1629
		1630	/* swap elements 0 and j (minimum) */
		1631	{
		1632	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1633	}
		1634	}
		1635
		1636	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1637	{
		1638	eio_dirent *i, *j;
		1639
		1640	for (i = dents + 1; i < dents + size; ++i)
		1641	{
		1642	eio_dirent value = *i;
		1643
		1644	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1645	j [1] = j [0];
		1646
		1647	j [1] = value;
		1648	}
		1649	}
		1650	}
		1651
		1652	static void
		1653	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1654	{
		1655	if (size <= 1)
		1656	return; /* our insertion sort relies on size > 0 */
		1657
		1658	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1659	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1660	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1661
		1662	/* use an insertion sort at the end, or for small arrays, */
		1663	/* as insertion sort is more efficient for small partitions */
		1664	eio_dent_insertion_sort (dents, size);
892	}	1665	}
893		1666
894	/* read a full directory */	1667	/* read a full directory */
895	static void	1668	static void
896	eio__scandir (eio_req *req, etp_worker *self)	1669	eio__scandir (eio_req *req, etp_worker *self)
897	{	1670	{
		1671	char *name, *names;
		1672	int namesalloc = 4096 - sizeof (void *) * 4;
		1673	int namesoffs = 0;
		1674	int flags = req->int1;
		1675	eio_dirent *dents = 0;
		1676	int dentalloc = 128;
		1677	int dentoffs = 0;
		1678	eio_ino_t inode_bits = 0;
		1679	#ifdef _WIN32
		1680	HANDLE dirp;
		1681	WIN32_FIND_DATA entp;
		1682	#else
898	DIR *dirp;	1683	DIR *dirp;
899	EIO_STRUCT_DIRENT *entp;	1684	EIO_STRUCT_DIRENT *entp;
900	char *name, *names;	1685	#endif
901	int memlen = 4096;
902	int memofs = 0;
903	int res = 0;
904		1686
905	X_LOCK (wrklock);	1687	req->result = -1;
		1688
		1689	if (!(flags & EIO_READDIR_DENTS))
		1690	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
		1691
		1692	#ifdef _WIN32
		1693	{
		1694	int len = strlen ((const char *)req->ptr1);
		1695	char *path = malloc (MAX_PATH);
		1696	const char *fmt;
		1697
		1698	if (!len)
		1699	fmt = "./*";
		1700	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1701	fmt = "%s*";
		1702	else
		1703	fmt = "%s/*";
		1704
		1705	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1706	dirp = FindFirstFile (path, &entp);
		1707	free (path);
		1708
		1709	if (dirp == INVALID_HANDLE_VALUE)
		1710	{
		1711	dirp = 0;
		1712
		1713	/* should steal _dosmaperr */
		1714	switch (GetLastError ())
		1715	{
		1716	case ERROR_FILE_NOT_FOUND:
		1717	req->result = 0;
		1718	break;
		1719
		1720	case ERROR_INVALID_NAME:
		1721	case ERROR_PATH_NOT_FOUND:
		1722	case ERROR_NO_MORE_FILES:
		1723	errno = ENOENT;
		1724	break;
		1725
		1726	case ERROR_NOT_ENOUGH_MEMORY:
		1727	errno = ENOMEM;
		1728	break;
		1729
		1730	default:
		1731	errno = EINVAL;
		1732	break;
		1733	}
		1734	}
		1735	}
		1736	#else
906	self->dirp = dirp = opendir (req->ptr1);	1737	dirp = opendir (req->ptr1);
		1738	#endif
		1739
907	req->flags |= EIO_FLAG_PTR2_FREE;	1740	if (req->flags & EIO_FLAG_PTR1_FREE)
		1741	free (req->ptr1);
		1742
		1743	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1744	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
908	req->ptr2 = names = malloc (memlen);	1745	req->ptr2 = names = malloc (namesalloc);
909	X_UNLOCK (wrklock);
910		1746
911	if (dirp && names)	1747	if (dirp && names && (!flags || dents))
912	for (;;)	1748	for (;;)
913	{	1749	{
		1750	int done;
		1751
		1752	#ifdef _WIN32
		1753	done = !dirp;
		1754	#else
914	errno = 0;	1755	errno = 0;
915	entp = readdir (dirp);	1756	entp = readdir (dirp);
		1757	done = !entp;
		1758	#endif
916		1759
917	if (!entp)	1760	if (done)
		1761	{
		1762	#ifndef _WIN32
		1763	int old_errno = errno;
		1764	closedir (dirp);
		1765	errno = old_errno;
		1766
		1767	if (errno)
		1768	break;
		1769	#endif
		1770
		1771	/* sort etc. */
		1772	req->int1 = flags;
		1773	req->result = dentoffs;
		1774
		1775	if (flags & EIO_READDIR_STAT_ORDER)
		1776	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1777	else if (flags & EIO_READDIR_DIRS_FIRST)
		1778	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1779	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1780	else
		1781	{
		1782	/* in this case, all is known, and we just put dirs first and sort them */
		1783	eio_dirent *oth = dents + dentoffs;
		1784	eio_dirent *dir = dents;
		1785
		1786	/* now partition dirs to the front, and non-dirs to the back */
		1787	/* by walking from both sides and swapping if necessary */
		1788	while (oth > dir)
		1789	{
		1790	if (dir->type == EIO_DT_DIR)
		1791	++dir;
		1792	else if ((--oth)->type == EIO_DT_DIR)
		1793	{
		1794	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1795
		1796	++dir;
		1797	}
		1798	}
		1799
		1800	/* now sort the dirs only (dirs all have the same score) */
		1801	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1802	}
		1803
918	break;	1804	break;
		1805	}
919		1806
920	name = entp->d_name;	1807	/* now add the entry to our list(s) */
		1808	name = D_NAME (entp);
921		1809
		1810	/* skip . and .. entries */
922	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1811	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
923	{	1812	{
924	int len = strlen (name) + 1;	1813	int len = D_NAMLEN (entp) + 1;
925		1814
926	res++;	1815	while (ecb_expect_false (namesoffs + len > namesalloc))
927
928	while (memofs + len > memlen)
929	{	1816	{
930	memlen *= 2;	1817	namesalloc *= 2;
931	X_LOCK (wrklock);
932	req->ptr2 = names = realloc (names, memlen);	1818	req->ptr2 = names = realloc (names, namesalloc);
933	X_UNLOCK (wrklock);
934		1819
935	if (!names)	1820	if (!names)
936	break;	1821	break;
937	}	1822	}
938		1823
939	memcpy (names + memofs, name, len);	1824	memcpy (names + namesoffs, name, len);
		1825
		1826	if (dents)
		1827	{
		1828	struct eio_dirent *ent;
		1829
		1830	if (ecb_expect_false (dentoffs == dentalloc))
		1831	{
		1832	dentalloc *= 2;
		1833	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1834
		1835	if (!dents)
		1836	break;
		1837	}
		1838
		1839	ent = dents + dentoffs;
		1840
		1841	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1842	ent->namelen = len - 1;
		1843	ent->inode = D_INO (entp);
		1844
		1845	inode_bits |= ent->inode;
		1846
		1847	switch (D_TYPE (entp))
		1848	{
		1849	default:
		1850	ent->type = EIO_DT_UNKNOWN;
		1851	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1852	break;
		1853
		1854	#ifdef DT_FIFO
		1855	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1856	#endif
		1857	#ifdef DT_CHR
		1858	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1859	#endif
		1860	#ifdef DT_MPC
		1861	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1862	#endif
		1863	#ifdef DT_DIR
		1864	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1865	#endif
		1866	#ifdef DT_NAM
		1867	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1868	#endif
		1869	#ifdef DT_BLK
		1870	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1871	#endif
		1872	#ifdef DT_MPB
		1873	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1874	#endif
		1875	#ifdef DT_REG
		1876	case DT_REG: ent->type = EIO_DT_REG; break;
		1877	#endif
		1878	#ifdef DT_NWK
		1879	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1880	#endif
		1881	#ifdef DT_CMP
		1882	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1883	#endif
		1884	#ifdef DT_LNK
		1885	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1886	#endif
		1887	#ifdef DT_SOCK
		1888	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1889	#endif
		1890	#ifdef DT_DOOR
		1891	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1892	#endif
		1893	#ifdef DT_WHT
		1894	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1895	#endif
		1896	}
		1897
		1898	ent->score = 7;
		1899
		1900	if (flags & EIO_READDIR_DIRS_FIRST)
		1901	{
		1902	if (ent->type == EIO_DT_UNKNOWN)
		1903	{
		1904	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1905	ent->score = 1;
		1906	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1907	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1908	}
		1909	else if (ent->type == EIO_DT_DIR)
		1910	ent->score = 0;
		1911	}
		1912	}
		1913
940	memofs += len;	1914	namesoffs += len;
		1915	++dentoffs;
941	}	1916	}
		1917
		1918	if (EIO_CANCELLED (req))
		1919	{
		1920	errno = ECANCELED;
		1921	break;
		1922	}
		1923
		1924	#ifdef _WIN32
		1925	if (!FindNextFile (dirp, &entp))
		1926	{
		1927	FindClose (dirp);
		1928	dirp = 0;
		1929	}
		1930	#endif
942	}	1931	}
943
944	if (errno)
945	res = -1;
946
947	req->result = res;
948	}	1932	}
949		1933
950	/*****************************************************************************/	1934	/*****************************************************************************/
951		1935
952	#define ALLOC(len) \	1936	#define ALLOC(len) \
…		…
968	{	1952	{
969	ETP_REQ *req;	1953	ETP_REQ *req;
970	struct timespec ts;	1954	struct timespec ts;
971	etp_worker *self = (etp_worker *)thr_arg;	1955	etp_worker *self = (etp_worker *)thr_arg;
972		1956
		1957	#if HAVE_PRCTL_SET_NAME
		1958	prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
		1959	#endif
		1960
973	/* try to distribute timeouts somewhat randomly */	1961	/* try to distribute timeouts somewhat evenly */
974	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1962	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
975		1963
976	for (;;)	1964	for (;;)
977	{	1965	{
		1966	ts.tv_sec = 0;
		1967
978	X_LOCK (reqlock);	1968	X_LOCK (reqlock);
979		1969
980	for (;;)	1970	for (;;)
981	{	1971	{
982	self->req = req = reqq_shift (&req_queue);	1972	self->req = req = reqq_shift (&req_queue);
983		1973
984	if (req)	1974	if (req)
985	break;	1975	break;
986		1976
		1977	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1978	{
		1979	X_UNLOCK (reqlock);
		1980	X_LOCK (wrklock);
		1981	--started;
		1982	X_UNLOCK (wrklock);
		1983	goto quit;
		1984	}
		1985
987	++idle;	1986	++idle;
988		1987
989	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1988	if (idle <= max_idle)
990	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1989	/* we are allowed to idle, so do so without any timeout */
		1990	X_COND_WAIT (reqwait, reqlock);
		1991	else
991	{	1992	{
992	if (idle > max_idle)	1993	/* initialise timeout once */
993	{	1994	if (!ts.tv_sec)
994	--idle;	1995	ts.tv_sec = time (0) + idle_timeout;
995	X_UNLOCK (reqlock);
996	X_LOCK (wrklock);
997	--started;
998	X_UNLOCK (wrklock);
999	goto quit;
1000	}
1001		1996
1002	/* we are allowed to idle, so do so without any timeout */
1003	X_COND_WAIT (reqwait, reqlock);	1997	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1998	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1004	}	1999	}
1005		2000
1006	--idle;	2001	--idle;
1007	}	2002	}
1008		2003
…		…
1011	X_UNLOCK (reqlock);	2006	X_UNLOCK (reqlock);
1012		2007
1013	if (req->type < 0)	2008	if (req->type < 0)
1014	goto quit;	2009	goto quit;
1015		2010
1016	if (!EIO_CANCELLED (req))
1017	ETP_EXECUTE (self, req);	2011	ETP_EXECUTE (self, req);
1018		2012
1019	X_LOCK (reslock);	2013	X_LOCK (reslock);
1020		2014
1021	++npending;	2015	++npending;
1022		2016
…		…
1037	return 0;	2031	return 0;
1038	}	2032	}
1039		2033
1040	/*****************************************************************************/	2034	/*****************************************************************************/
1041		2035
		2036	int ecb_cold
1042	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	2037	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1043	{	2038	{
		2039	#if !HAVE_PREADWRITE
		2040	X_MUTEX_CREATE (preadwritelock);
		2041	#endif
		2042
1044	etp_init (want_poll, done_poll);	2043	return etp_init (want_poll, done_poll);
1045	}	2044	}
1046		2045
		2046	ecb_inline void
1047	static void eio_api_destroy (eio_req *req)	2047	eio_api_destroy (eio_req *req)
1048	{	2048	{
1049	free (req);	2049	free (req);
1050	}	2050	}
1051		2051
1052	#define REQ(rtype) \	2052	#define REQ(rtype) \
…		…
1071	{ \	2071	{ \
1072	eio_api_destroy (req); \	2072	eio_api_destroy (req); \
1073	return 0; \	2073	return 0; \
1074	}	2074	}
1075		2075
		2076	static void
1076	static void eio_execute (etp_worker *self, eio_req *req)	2077	eio_execute (etp_worker *self, eio_req *req)
1077	{	2078	{
1078	errno = 0;	2079	if (ecb_expect_false (EIO_CANCELLED (req)))
		2080	{
		2081	req->result = -1;
		2082	req->errorno = ECANCELED;
		2083	return;
		2084	}
1079		2085
1080	switch (req->type)	2086	switch (req->type)
1081	{	2087	{
1082	case EIO_READ: ALLOC (req->size);	2088	case EIO_READ: ALLOC (req->size);
1083	req->result = req->offs >= 0	2089	req->result = req->offs >= 0
…		…
1086	case EIO_WRITE: req->result = req->offs >= 0	2092	case EIO_WRITE: req->result = req->offs >= 0
1087	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2093	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1088	: write (req->int1, req->ptr2, req->size); break;	2094	: write (req->int1, req->ptr2, req->size); break;
1089		2095
1090	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2096	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1091	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2097	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1092		2098
1093	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2099	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1094	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2100	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1095	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2101	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1096	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2102	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1097	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2103	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1098	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2104	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2105
		2106	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2107	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2108	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2109	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1099		2110
1100	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	2111	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1101	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	2112	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1102	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	2113	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1103	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	2114	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
…		…
1113	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2124	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1114	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2125	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1115	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2126	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1116	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;	2127	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1117		2128
		2129	case EIO_REALPATH: eio__realpath (req, self); break;
		2130
1118	case EIO_READLINK: ALLOC (NAME_MAX);	2131	case EIO_READLINK: ALLOC (PATH_MAX);
1119	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	2132	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1120		2133
1121	case EIO_SYNC: req->result = 0; sync (); break;	2134	case EIO_SYNC: req->result = 0; sync (); break;
1122	case EIO_FSYNC: req->result = fsync (req->int1); break;	2135	case EIO_FSYNC: req->result = fsync (req->int1); break;
1123	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2136	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2137	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2138	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2139	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2140	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		2141	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2142	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		2143	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1124		2144
1125	case EIO_READDIR: eio__scandir (req, self); break;	2145	case EIO_READDIR: eio__scandir (req, self); break;
1126		2146
1127	case EIO_BUSY:	2147	case EIO_BUSY:
1128	#ifdef _WIN32	2148	#ifdef _WIN32
1129	Sleep (req->nv1 * 1000.);	2149	Sleep (req->nv1 * 1e3);
1130	#else	2150	#else
1131	{	2151	{
1132	struct timeval tv;	2152	struct timeval tv;
1133		2153
1134	tv.tv_sec = req->nv1;	2154	tv.tv_sec = req->nv1;
1135	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	2155	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1136		2156
1137	req->result = select (0, 0, 0, 0, &tv);	2157	req->result = select (0, 0, 0, 0, &tv);
1138	}	2158	}
1139	#endif	2159	#endif
1140	break;	2160	break;
…		…
1155	times = tv;	2175	times = tv;
1156	}	2176	}
1157	else	2177	else
1158	times = 0;	2178	times = 0;
1159		2179
1160
1161	req->result = req->type == EIO_FUTIME	2180	req->result = req->type == EIO_FUTIME
1162	? futimes (req->int1, times)	2181	? futimes (req->int1, times)
1163	: utimes (req->ptr1, times);	2182	: utimes (req->ptr1, times);
1164	}	2183	}
		2184	break;
1165		2185
1166	case EIO_GROUP:	2186	case EIO_GROUP:
		2187	abort (); /* handled in eio_request */
		2188
1167	case EIO_NOP:	2189	case EIO_NOP:
1168	req->result = 0;	2190	req->result = 0;
1169	break;	2191	break;
1170		2192
1171	case EIO_CUSTOM:	2193	case EIO_CUSTOM:
1172	req->feed (req);	2194	req->feed (req);
1173	break;	2195	break;
1174		2196
1175	default:	2197	default:
		2198	errno = ENOSYS;
1176	req->result = -1;	2199	req->result = -1;
1177	break;	2200	break;
1178	}	2201	}
1179		2202
1180	req->errorno = errno;	2203	req->errorno = errno;
…		…
1200	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	2223	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1201	{	2224	{
1202	REQ (EIO_FSYNC); req->int1 = fd; SEND;	2225	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1203	}	2226	}
1204		2227
		2228	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2229	{
		2230	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2231	}
		2232
1205	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2233	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1206	{	2234	{
1207	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2235	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1208	}	2236	}
1209		2237
		2238	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2239	{
		2240	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2241	}
		2242
		2243	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2244	{
		2245	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2246	}
		2247
		2248	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2249	{
		2250	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2251	}
		2252
		2253	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2254	{
		2255	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2256	}
		2257
		2258	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2259	{
		2260	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2261	}
		2262
		2263	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2264	{
		2265	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2266	}
		2267
1210	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2268	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1211	{	2269	{
1212	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2270	REQ (EIO_CLOSE); req->int1 = fd; SEND;
1213	}	2271	}
1214		2272
…		…
1230	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2288	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1231	{	2289	{
1232	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2290	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1233	}	2291	}
1234		2292
		2293	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2294	{
		2295	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2296	}
		2297
1235	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2298	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1236	{	2299	{
1237	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2300	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1238	}	2301	}
1239		2302
…		…
1245	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2308	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1246	{	2309	{
1247	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2310	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1248	}	2311	}
1249		2312
1250	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2313	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1251	{	2314	{
1252	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2315	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1253	}	2316	}
1254		2317
1255	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2318	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1275	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2338	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1276	{	2339	{
1277	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2340	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1278	}	2341	}
1279		2342
1280	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2343	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1281	{	2344	{
1282	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2345	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1283	}	2346	}
1284		2347
1285	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2348	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1301	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2364	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1302	{	2365	{
1303	return eio__1path (EIO_READLINK, path, pri, cb, data);	2366	return eio__1path (EIO_READLINK, path, pri, cb, data);
1304	}	2367	}
1305		2368
		2369	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2370	{
		2371	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2372	}
		2373
1306	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2374	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1307	{	2375	{
1308	return eio__1path (EIO_STAT, path, pri, cb, data);	2376	return eio__1path (EIO_STAT, path, pri, cb, data);
1309	}	2377	}
1310		2378
1311	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2379	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1312	{	2380	{
1313	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2381	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1314	}	2382	}
1315		2383
		2384	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2385	{
		2386	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2387	}
		2388
1316	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2389	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1317	{	2390	{
1318	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2391	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1319	}	2392	}
1320		2393
1321	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	2394	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1322	{	2395	{
1323	return eio__1path (EIO_RMDIR, path, pri, cb, data);	2396	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1324	}	2397	}
1325		2398
1326	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	2399	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1327	{	2400	{
1328	return eio__1path (EIO_READDIR, path, pri, cb, data);	2401	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1329	}	2402	}
1330		2403
1331	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2404	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1332	{	2405	{
1333	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	2406	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1334	}	2407	}
1335		2408
1336	static eio_req *	2409	static eio_req *
1337	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2410	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1338	{	2411	{
…		…
1362	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2435	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1363	{	2436	{
1364	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2437	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1365	}	2438	}
1366		2439
1367	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2440	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1368	{	2441	{
1369	REQ (EIO_CUSTOM); req->feed = execute; SEND;	2442	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1370	}	2443	}
1371		2444
1372	#endif	2445	#endif
…		…
1383	#undef SEND	2456	#undef SEND
1384		2457
1385	/*****************************************************************************/	2458	/*****************************************************************************/
1386	/* grp functions */	2459	/* grp functions */
1387		2460
		2461	void
1388	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2462	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1389	{	2463	{
1390	grp->int2 = limit;	2464	grp->int2 = limit;
1391	grp->feed = feed;	2465	grp->feed = feed;
1392		2466
1393	grp_try_feed (grp);	2467	grp_try_feed (grp);
1394	}	2468	}
1395		2469
		2470	void
1396	void eio_grp_limit (eio_req *grp, int limit)	2471	eio_grp_limit (eio_req *grp, int limit)
1397	{	2472	{
1398	grp->int2 = limit;	2473	grp->int2 = limit;
1399		2474
1400	grp_try_feed (grp);	2475	grp_try_feed (grp);
1401	}	2476	}
1402		2477
		2478	void
1403	void eio_grp_add (eio_req *grp, eio_req *req)	2479	eio_grp_add (eio_req *grp, eio_req *req)
1404	{	2480	{
1405	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2481	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
		2482
		2483	grp->flags |= EIO_FLAG_GROUPADD;
1406		2484
1407	++grp->size;	2485	++grp->size;
1408	req->grp = grp;	2486	req->grp = grp;
1409		2487
1410	req->grp_prev = 0;	2488	req->grp_prev = 0;
…		…
1417	}	2495	}
1418		2496
1419	/*****************************************************************************/	2497	/*****************************************************************************/
1420	/* misc garbage */	2498	/* misc garbage */
1421		2499
		2500	eio_ssize_t
1422	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2501	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1423	{	2502	{
1424	etp_worker wrk;
1425
1426	wrk.dbuf = 0;
1427
1428	eio__sendfile (ofd, ifd, offset, count, &wrk);	2503	return eio__sendfile (ofd, ifd, offset, count);
1429
1430	if (wrk.dbuf)
1431	free (wrk.dbuf);
1432	}	2504	}
1433		2505

Diff Legend

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