ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.68 by root, Sun Jun 5 22:44:30 2011 UTC vs.
Revision 1.120 by root, Tue Apr 24 18:47:50 2012 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>	4	* Copyright (c) 2007,2008,2009,2010,2011,2012 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	*
…		…
40	#ifndef _WIN32	40	#ifndef _WIN32
41	# include "config.h"	41	# include "config.h"
42	#endif	42	#endif
43		43
44	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
45		46
46	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
47	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define X_STACKSIZE EIO_STACKSIZE
48	#endif	49	#endif
49	#include "xthread.h"	50	#include "xthread.h"
50		51
51	#include <errno.h>	52	#include <errno.h>
52	#include <stddef.h>	53	#include <stddef.h>
53	#include <stdlib.h>	54	#include <stdlib.h>
54	#include <string.h>	55	#include <string.h>
55	#include <errno.h>	56	#include <errno.h>
56	#include <sys/types.h>	57	#include <sys/types.h>
57	#include <sys/stat.h>	58	#include <sys/stat.h>
58	#include <sys/statvfs.h>
59	#include <limits.h>	59	#include <limits.h>
60	#include <fcntl.h>	60	#include <fcntl.h>
61	#include <assert.h>	61	#include <assert.h>
62		62
63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */	63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64	/* intptr_t only comes form stdint.h, says idiot openbsd coder */	64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65	#if HAVE_STDINT_H	65	#if HAVE_STDINT_H
66	# include <stdint.h>	66	# include <stdint.h>
67	#endif	67	#endif
68		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
69	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
70	# 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
71	#endif	84	#endif
72		85
73	#ifndef EIO_DESTROY	86	#ifndef EIO_DESTROY
…		…
76		89
77	#ifndef EIO_FEED	90	#ifndef EIO_FEED
78	# 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)
79	#endif	92	#endif
80		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
81	#ifdef _WIN32	105	#ifdef _WIN32
82		106
83	/*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)
		196
84	#else	197	#else
85		198
86	# include <sys/time.h>	199	#include <sys/time.h>
87	# include <sys/select.h>	200	#include <sys/select.h>
		201	#include <sys/statvfs.h>
88	# 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__ || __NetBSD__ || __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
89	# include <utime.h>	237	# include <utime.h>
90	# include <signal.h>
91	# include <dirent.h>
92
93	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94	# include <sys/mman.h>
95	#endif
96
97	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99	# define _DIRENT_HAVE_D_TYPE /* sigh */
100	# define D_INO(de) (de)->d_fileno
101	# define D_NAMLEN(de) (de)->d_namlen
102	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103	# define D_INO(de) (de)->d_ino
104	# endif	238	#endif
105		239
106	#ifdef _D_EXACT_NAMLEN	240	#if HAVE_SYS_SYSCALL_H
107	# undef D_NAMLEN	241	# include <sys/syscall.h>
108	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109	#endif
110
111	# ifdef _DIRENT_HAVE_D_TYPE
112	# define D_TYPE(de) (de)->d_type
113	# endif	242	#endif
114		243
115	# ifndef EIO_STRUCT_DIRENT	244	#if HAVE_SYS_PRCTL_H
116	# define EIO_STRUCT_DIRENT struct dirent	245	# include <sys/prctl.h>
117	# endif
118
119	#endif	246	#endif
120		247
121	#if HAVE_SENDFILE	248	#if HAVE_SENDFILE
122	# if __linux	249	# if __linux
123	# include <sys/sendfile.h>	250	# include <sys/sendfile.h>
…		…
138	#endif	265	#endif
139	#ifndef D_INO	266	#ifndef D_INO
140	# define D_INO(de) 0	267	# define D_INO(de) 0
141	#endif	268	#endif
142	#ifndef D_NAMLEN	269	#ifndef D_NAMLEN
143	# define D_NAMLEN(de) strlen ((de)->d_name)	270	# define D_NAMLEN(entp) strlen (D_NAME (entp))
144	#endif	271	#endif
145		272
146	/* used for struct dirent, AIX doesn't provide it */	273	/* used for struct dirent, AIX doesn't provide it */
147	#ifndef NAME_MAX	274	#ifndef NAME_MAX
148	# define NAME_MAX 4096	275	# define NAME_MAX 4096
…		…
155		282
156	/* buffer size for various temporary buffers */	283	/* buffer size for various temporary buffers */
157	#define EIO_BUFSIZE 65536	284	#define EIO_BUFSIZE 65536
158		285
159	#define dBUF \	286	#define dBUF \
160	char *eio_buf; \
161	ETP_WORKER_LOCK (self); \
162	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	287	char *eio_buf = malloc (EIO_BUFSIZE); \
163	ETP_WORKER_UNLOCK (self); \
164	errno = ENOMEM; \	288	errno = ENOMEM; \
165	if (!eio_buf) \	289	if (!eio_buf) \
166	return -1;	290	return -1
		291
		292	#define FUBd \
		293	free (eio_buf)
167		294
168	#define EIO_TICKS ((1000000 + 1023) >> 10)	295	#define EIO_TICKS ((1000000 + 1023) >> 10)
169		296
170	/*****************************************************************************/	297	/*****************************************************************************/
171		298
172	#if __GNUC__ >= 3	299	struct tmpbuf
173	# define expect(expr,value) __builtin_expect ((expr),(value))	300	{
		301	void *ptr;
		302	int len;
		303	};
		304
		305	static void *
		306	tmpbuf_get (struct tmpbuf *buf, int len)
		307	{
		308	if (buf->len < len)
		309	{
		310	free (buf->ptr);
		311	buf->ptr = malloc (buf->len = len);
		312	}
		313
		314	return buf->ptr;
		315	}
		316
		317	struct tmpbuf;
		318
		319	#if _POSIX_VERSION >= 200809L
		320	#define HAVE_AT 1
		321	#define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
		322	#ifndef O_SEARCH
		323	#define O_SEARCH O_RDONLY
		324	#endif
174	#else	325	#else
175	# define expect(expr,value) (expr)	326	#define HAVE_AT 0
		327	static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
176	#endif	328	#endif
177		329
178	#define expect_false(expr) expect ((expr) != 0, 0)	330	struct eio_pwd
179	#define expect_true(expr) expect ((expr) != 0, 1)	331	{
		332	#if HAVE_AT
		333	int fd;
		334	#endif
		335	int len;
		336	char str[1]; /* actually, a 0-terminated canonical path */
		337	};
180		338
181	/*****************************************************************************/	339	/*****************************************************************************/
182		340
183	#define ETP_PRI_MIN EIO_PRI_MIN	341	#define ETP_PRI_MIN EIO_PRI_MIN
184	#define ETP_PRI_MAX EIO_PRI_MAX	342	#define ETP_PRI_MAX EIO_PRI_MAX
…		…
190	static int eio_finish (eio_req *req);	348	static int eio_finish (eio_req *req);
191	#define ETP_FINISH(req) eio_finish (req)	349	#define ETP_FINISH(req) eio_finish (req)
192	static void eio_execute (struct etp_worker *self, eio_req *req);	350	static void eio_execute (struct etp_worker *self, eio_req *req);
193	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	351	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194		352
195	#define ETP_WORKER_CLEAR(req) \
196	if (wrk->dbuf) \
197	{ \
198	free (wrk->dbuf); \
199	wrk->dbuf = 0; \
200	} \
201	\
202	if (wrk->dirp) \
203	{ \
204	closedir (wrk->dirp); \
205	wrk->dirp = 0; \
206	}
207
208	#define ETP_WORKER_COMMON \
209	void *dbuf; \
210	DIR *dirp;
211
212	/*****************************************************************************/	353	/*****************************************************************************/
213		354
214	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	355	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215		356
216	/* calculate time difference in ~1/EIO_TICKS of a second */	357	/* calculate time difference in ~1/EIO_TICKS of a second */
		358	ecb_inline int
217	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	359	tvdiff (struct timeval *tv1, struct timeval *tv2)
218	{	360	{
219	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	361	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	362	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
221	}	363	}
222		364
…		…
226	static void (*done_poll_cb) (void);	368	static void (*done_poll_cb) (void);
227		369
228	static unsigned int max_poll_time; /* reslock */	370	static unsigned int max_poll_time; /* reslock */
229	static unsigned int max_poll_reqs; /* reslock */	371	static unsigned int max_poll_reqs; /* reslock */
230		372
231	static volatile unsigned int nreqs; /* reqlock */	373	static unsigned int nreqs; /* reqlock */
232	static volatile unsigned int nready; /* reqlock */	374	static unsigned int nready; /* reqlock */
233	static volatile unsigned int npending; /* reqlock */	375	static unsigned int npending; /* reqlock */
234	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */	376	static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
235	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */	377	static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
236		378
237	static xmutex_t wrklock;	379	static xmutex_t wrklock;
238	static xmutex_t reslock;	380	static xmutex_t reslock;
239	static xmutex_t reqlock;	381	static xmutex_t reqlock;
240	static xcond_t reqwait;	382	static xcond_t reqwait;
…		…
243	/*	385	/*
244	* make our pread/pwrite emulation safe against themselves, but not against	386	* make our pread/pwrite emulation safe against themselves, but not against
245	* normal read/write by using a mutex. slows down execution a lot,	387	* normal read/write by using a mutex. slows down execution a lot,
246	* but that's your problem, not mine.	388	* but that's your problem, not mine.
247	*/	389	*/
248	static xmutex_t preadwritelock = X_MUTEX_INIT;	390	static xmutex_t preadwritelock;
249	#endif	391	#endif
250		392
251	typedef struct etp_worker	393	typedef struct etp_worker
252	{	394	{
		395	struct tmpbuf tmpbuf;
		396
253	/* locked by wrklock */	397	/* locked by wrklock */
254	struct etp_worker *prev, *next;	398	struct etp_worker *prev, *next;
255		399
256	xthread_t tid;	400	xthread_t tid;
257		401
258	/* locked by reslock, reqlock or wrklock */	402	#ifdef ETP_WORKER_COMMON
259	ETP_REQ *req; /* currently processed request */
260
261	ETP_WORKER_COMMON	403	ETP_WORKER_COMMON
		404	#endif
262	} etp_worker;	405	} etp_worker;
263		406
264	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	407	static etp_worker wrk_first; /* NOT etp */
265		408
266	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	409	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	410	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268		411
269	/* worker threads management */	412	/* worker threads management */
270		413
		414	static void
271	static void etp_worker_clear (etp_worker *wrk)	415	etp_worker_clear (etp_worker *wrk)
272	{	416	{
273	ETP_WORKER_CLEAR (wrk);
274	}	417	}
275		418
		419	static void ecb_cold
276	static void etp_worker_free (etp_worker *wrk)	420	etp_worker_free (etp_worker *wrk)
277	{	421	{
		422	free (wrk->tmpbuf.ptr);
		423
278	wrk->next->prev = wrk->prev;	424	wrk->next->prev = wrk->prev;
279	wrk->prev->next = wrk->next;	425	wrk->prev->next = wrk->next;
280		426
281	free (wrk);	427	free (wrk);
282	}	428	}
283		429
284	static unsigned int etp_nreqs (void)	430	static unsigned int
		431	etp_nreqs (void)
285	{	432	{
286	int retval;	433	int retval;
287	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	434	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288	retval = nreqs;	435	retval = nreqs;
289	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	436	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290	return retval;	437	return retval;
291	}	438	}
292		439
293	static unsigned int etp_nready (void)	440	static unsigned int
		441	etp_nready (void)
294	{	442	{
295	unsigned int retval;	443	unsigned int retval;
296		444
297	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	445	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298	retval = nready;	446	retval = nready;
299	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	447	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300		448
301	return retval;	449	return retval;
302	}	450	}
303		451
304	static unsigned int etp_npending (void)	452	static unsigned int
		453	etp_npending (void)
305	{	454	{
306	unsigned int retval;	455	unsigned int retval;
307		456
308	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	457	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309	retval = npending;	458	retval = npending;
310	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	459	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311		460
312	return retval;	461	return retval;
313	}	462	}
314		463
315	static unsigned int etp_nthreads (void)	464	static unsigned int
		465	etp_nthreads (void)
316	{	466	{
317	unsigned int retval;	467	unsigned int retval;
318		468
319	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	469	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320	retval = started;	470	retval = started;
…		…
334	} etp_reqq;	484	} etp_reqq;
335		485
336	static etp_reqq req_queue;	486	static etp_reqq req_queue;
337	static etp_reqq res_queue;	487	static etp_reqq res_queue;
338		488
		489	static void ecb_noinline ecb_cold
		490	reqq_init (etp_reqq *q)
		491	{
		492	int pri;
		493
		494	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		495	q->qs[pri] = q->qe[pri] = 0;
		496
		497	q->size = 0;
		498	}
		499
		500	static int ecb_noinline
339	static int reqq_push (etp_reqq *q, ETP_REQ *req)	501	reqq_push (etp_reqq *q, ETP_REQ *req)
340	{	502	{
341	int pri = req->pri;	503	int pri = req->pri;
342	req->next = 0;	504	req->next = 0;
343		505
344	if (q->qe[pri])	506	if (q->qe[pri])
…		…
350	q->qe[pri] = q->qs[pri] = req;	512	q->qe[pri] = q->qs[pri] = req;
351		513
352	return q->size++;	514	return q->size++;
353	}	515	}
354		516
		517	static ETP_REQ * ecb_noinline
355	static ETP_REQ *reqq_shift (etp_reqq *q)	518	reqq_shift (etp_reqq *q)
356	{	519	{
357	int pri;	520	int pri;
358		521
359	if (!q->size)	522	if (!q->size)
360	return 0;	523	return 0;
…		…
375	}	538	}
376		539
377	abort ();	540	abort ();
378	}	541	}
379		542
380	static void etp_thread_init (void)	543	static int ecb_cold
		544	etp_init (void (*want_poll)(void), void (*done_poll)(void))
381	{	545	{
382	X_MUTEX_CREATE (wrklock);	546	X_MUTEX_CREATE (wrklock);
383	X_MUTEX_CREATE (reslock);	547	X_MUTEX_CREATE (reslock);
384	X_MUTEX_CREATE (reqlock);	548	X_MUTEX_CREATE (reqlock);
385	X_COND_CREATE (reqwait);	549	X_COND_CREATE (reqwait);
386	}
387		550
388	static void etp_atfork_prepare (void)	551	reqq_init (&req_queue);
389	{	552	reqq_init (&res_queue);
390	X_LOCK (wrklock);
391	X_LOCK (reqlock);
392	X_LOCK (reslock);
393	#if !HAVE_PREADWRITE
394	X_LOCK (preadwritelock);
395	#endif
396	}
397		553
398	static void etp_atfork_parent (void)	554	wrk_first.next =
399	{	555	wrk_first.prev = &wrk_first;
400	#if !HAVE_PREADWRITE
401	X_UNLOCK (preadwritelock);
402	#endif
403	X_UNLOCK (reslock);
404	X_UNLOCK (reqlock);
405	X_UNLOCK (wrklock);
406	}
407
408	static void etp_atfork_child (void)
409	{
410	ETP_REQ *prv;
411
412	while ((prv = reqq_shift (&req_queue)))
413	ETP_DESTROY (prv);
414
415	while ((prv = reqq_shift (&res_queue)))
416	ETP_DESTROY (prv);
417
418	while (wrk_first.next != &wrk_first)
419	{
420	etp_worker *wrk = wrk_first.next;
421
422	if (wrk->req)
423	ETP_DESTROY (wrk->req);
424
425	etp_worker_clear (wrk);
426	etp_worker_free (wrk);
427	}
428		556
429	started = 0;	557	started = 0;
430	idle = 0;	558	idle = 0;
431	nreqs = 0;	559	nreqs = 0;
432	nready = 0;	560	nready = 0;
433	npending = 0;	561	npending = 0;
434		562
435	etp_thread_init ();
436	}
437
438	static void
439	etp_once_init (void)
440	{
441	etp_thread_init ();
442	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
443	}
444
445	static int
446	etp_init (void (*want_poll)(void), void (*done_poll)(void))
447	{
448	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
449
450	pthread_once (&doinit, etp_once_init);
451
452	want_poll_cb = want_poll;	563	want_poll_cb = want_poll;
453	done_poll_cb = done_poll;	564	done_poll_cb = done_poll;
454		565
455	return 0;	566	return 0;
456	}	567	}
457		568
458	X_THREAD_PROC (etp_proc);	569	X_THREAD_PROC (etp_proc);
459		570
		571	static void ecb_cold
460	static void etp_start_thread (void)	572	etp_start_thread (void)
461	{	573	{
462	etp_worker *wrk = calloc (1, sizeof (etp_worker));	574	etp_worker *wrk = calloc (1, sizeof (etp_worker));
463		575
464	/*TODO*/	576	/*TODO*/
465	assert (("unable to allocate worker thread data", wrk));	577	assert (("unable to allocate worker thread data", wrk));
…		…
478	free (wrk);	590	free (wrk);
479		591
480	X_UNLOCK (wrklock);	592	X_UNLOCK (wrklock);
481	}	593	}
482		594
		595	static void
483	static void etp_maybe_start_thread (void)	596	etp_maybe_start_thread (void)
484	{	597	{
485	if (expect_true (etp_nthreads () >= wanted))	598	if (ecb_expect_true (etp_nthreads () >= wanted))
486	return;	599	return;
487		600
488	/* todo: maybe use idle here, but might be less exact */	601	/* todo: maybe use idle here, but might be less exact */
489	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	602	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
490	return;	603	return;
491		604
492	etp_start_thread ();	605	etp_start_thread ();
493	}	606	}
494		607
		608	static void ecb_cold
495	static void etp_end_thread (void)	609	etp_end_thread (void)
496	{	610	{
497	eio_req *req = calloc (1, sizeof (eio_req));	611	eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
498		612
499	req->type = -1;	613	req->type = -1;
500	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	614	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
501		615
502	X_LOCK (reqlock);	616	X_LOCK (reqlock);
…		…
507	X_LOCK (wrklock);	621	X_LOCK (wrklock);
508	--started;	622	--started;
509	X_UNLOCK (wrklock);	623	X_UNLOCK (wrklock);
510	}	624	}
511		625
512	static int etp_poll (void)	626	static int
		627	etp_poll (void)
513	{	628	{
514	unsigned int maxreqs;	629	unsigned int maxreqs;
515	unsigned int maxtime;	630	unsigned int maxtime;
516	struct timeval tv_start, tv_now;	631	struct timeval tv_start, tv_now;
517		632
…		…
547		662
548	X_LOCK (reqlock);	663	X_LOCK (reqlock);
549	--nreqs;	664	--nreqs;
550	X_UNLOCK (reqlock);	665	X_UNLOCK (reqlock);
551		666
552	if (expect_false (req->type == EIO_GROUP && req->size))	667	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
553	{	668	{
554	req->int1 = 1; /* mark request as delayed */	669	req->int1 = 1; /* mark request as delayed */
555	continue;	670	continue;
556	}	671	}
557	else	672	else
558	{	673	{
559	int res = ETP_FINISH (req);	674	int res = ETP_FINISH (req);
560	if (expect_false (res))	675	if (ecb_expect_false (res))
561	return res;	676	return res;
562	}	677	}
563		678
564	if (expect_false (maxreqs && !--maxreqs))	679	if (ecb_expect_false (maxreqs && !--maxreqs))
565	break;	680	break;
566		681
567	if (maxtime)	682	if (maxtime)
568	{	683	{
569	gettimeofday (&tv_now, 0);	684	gettimeofday (&tv_now, 0);
…		…
575		690
576	errno = EAGAIN;	691	errno = EAGAIN;
577	return -1;	692	return -1;
578	}	693	}
579		694
		695	static void
580	static void etp_cancel (ETP_REQ *req)	696	etp_cancel (ETP_REQ *req)
581	{	697	{
582	X_LOCK (wrklock);	698	req->cancelled = 1;
583	req->flags |= EIO_FLAG_CANCELLED;
584	X_UNLOCK (wrklock);
585		699
586	eio_grp_cancel (req);	700	eio_grp_cancel (req);
587	}	701	}
588		702
		703	static void
589	static void etp_submit (ETP_REQ *req)	704	etp_submit (ETP_REQ *req)
590	{	705	{
591	req->pri -= ETP_PRI_MIN;	706	req->pri -= ETP_PRI_MIN;
592		707
593	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	708	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
594	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	709	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595		710
596	if (expect_false (req->type == EIO_GROUP))	711	if (ecb_expect_false (req->type == EIO_GROUP))
597	{	712	{
598	/* I hope this is worth it :/ */	713	/* I hope this is worth it :/ */
599	X_LOCK (reqlock);	714	X_LOCK (reqlock);
600	++nreqs;	715	++nreqs;
601	X_UNLOCK (reqlock);	716	X_UNLOCK (reqlock);
…		…
620		735
621	etp_maybe_start_thread ();	736	etp_maybe_start_thread ();
622	}	737	}
623	}	738	}
624		739
		740	static void ecb_cold
625	static void etp_set_max_poll_time (double nseconds)	741	etp_set_max_poll_time (double nseconds)
626	{	742	{
627	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	743	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628	max_poll_time = nseconds * EIO_TICKS;	744	max_poll_time = nseconds * EIO_TICKS;
629	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	745	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630	}	746	}
631		747
		748	static void ecb_cold
632	static void etp_set_max_poll_reqs (unsigned int maxreqs)	749	etp_set_max_poll_reqs (unsigned int maxreqs)
633	{	750	{
634	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	751	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635	max_poll_reqs = maxreqs;	752	max_poll_reqs = maxreqs;
636	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	753	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637	}	754	}
638		755
		756	static void ecb_cold
639	static void etp_set_max_idle (unsigned int nthreads)	757	etp_set_max_idle (unsigned int nthreads)
640	{	758	{
641	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	759	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642	max_idle = nthreads;	760	max_idle = nthreads;
643	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	761	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644	}	762	}
645		763
		764	static void ecb_cold
646	static void etp_set_idle_timeout (unsigned int seconds)	765	etp_set_idle_timeout (unsigned int seconds)
647	{	766	{
648	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	767	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649	idle_timeout = seconds;	768	idle_timeout = seconds;
650	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	769	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651	}	770	}
652		771
		772	static void ecb_cold
653	static void etp_set_min_parallel (unsigned int nthreads)	773	etp_set_min_parallel (unsigned int nthreads)
654	{	774	{
655	if (wanted < nthreads)	775	if (wanted < nthreads)
656	wanted = nthreads;	776	wanted = nthreads;
657	}	777	}
658		778
		779	static void ecb_cold
659	static void etp_set_max_parallel (unsigned int nthreads)	780	etp_set_max_parallel (unsigned int nthreads)
660	{	781	{
661	if (wanted > nthreads)	782	if (wanted > nthreads)
662	wanted = nthreads;	783	wanted = nthreads;
663		784
664	while (started > wanted)	785	while (started > wanted)
665	etp_end_thread ();	786	etp_end_thread ();
666	}	787	}
667		788
668	/*****************************************************************************/	789	/*****************************************************************************/
669		790
		791	static void
670	static void grp_try_feed (eio_req *grp)	792	grp_try_feed (eio_req *grp)
671	{	793	{
672	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	794	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673	{	795	{
674	grp->flags &= ~EIO_FLAG_GROUPADD;	796	grp->flags &= ~EIO_FLAG_GROUPADD;
675		797
…		…
682	break;	804	break;
683	}	805	}
684	}	806	}
685	}	807	}
686		808
		809	static int
687	static int grp_dec (eio_req *grp)	810	grp_dec (eio_req *grp)
688	{	811	{
689	--grp->size;	812	--grp->size;
690		813
691	/* call feeder, if applicable */	814	/* call feeder, if applicable */
692	grp_try_feed (grp);	815	grp_try_feed (grp);
…		…
696	return eio_finish (grp);	819	return eio_finish (grp);
697	else	820	else
698	return 0;	821	return 0;
699	}	822	}
700		823
		824	static void
701	void eio_destroy (eio_req *req)	825	eio_destroy (eio_req *req)
702	{	826	{
703	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	827	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	828	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705		829
706	EIO_DESTROY (req);	830	EIO_DESTROY (req);
707	}	831	}
708		832
		833	static int
709	static int eio_finish (eio_req *req)	834	eio_finish (eio_req *req)
710	{	835	{
711	int res = EIO_FINISH (req);	836	int res = EIO_FINISH (req);
712		837
713	if (req->grp)	838	if (req->grp)
714	{	839	{
…		…
722	if (grp->grp_first == req)	847	if (grp->grp_first == req)
723	grp->grp_first = req->grp_next;	848	grp->grp_first = req->grp_next;
724		849
725	res2 = grp_dec (grp);	850	res2 = grp_dec (grp);
726		851
727	if (!res && res2)	852	if (!res)
728	res = res2;	853	res = res2;
729	}	854	}
730		855
731	eio_destroy (req);	856	eio_destroy (req);
732		857
733	return res;	858	return res;
734	}	859	}
735		860
		861	void
736	void eio_grp_cancel (eio_req *grp)	862	eio_grp_cancel (eio_req *grp)
737	{	863	{
738	for (grp = grp->grp_first; grp; grp = grp->grp_next)	864	for (grp = grp->grp_first; grp; grp = grp->grp_next)
739	eio_cancel (grp);	865	eio_cancel (grp);
740	}	866	}
741		867
		868	void
742	void eio_cancel (eio_req *req)	869	eio_cancel (eio_req *req)
743	{	870	{
744	etp_cancel (req);	871	etp_cancel (req);
745	}	872	}
746		873
		874	void
747	void eio_submit (eio_req *req)	875	eio_submit (eio_req *req)
748	{	876	{
749	etp_submit (req);	877	etp_submit (req);
750	}	878	}
751		879
752	unsigned int eio_nreqs (void)	880	unsigned int
		881	eio_nreqs (void)
753	{	882	{
754	return etp_nreqs ();	883	return etp_nreqs ();
755	}	884	}
756		885
757	unsigned int eio_nready (void)	886	unsigned int
		887	eio_nready (void)
758	{	888	{
759	return etp_nready ();	889	return etp_nready ();
760	}	890	}
761		891
762	unsigned int eio_npending (void)	892	unsigned int
		893	eio_npending (void)
763	{	894	{
764	return etp_npending ();	895	return etp_npending ();
765	}	896	}
766		897
767	unsigned int eio_nthreads (void)	898	unsigned int ecb_cold
		899	eio_nthreads (void)
768	{	900	{
769	return etp_nthreads ();	901	return etp_nthreads ();
770	}	902	}
771		903
		904	void ecb_cold
772	void eio_set_max_poll_time (double nseconds)	905	eio_set_max_poll_time (double nseconds)
773	{	906	{
774	etp_set_max_poll_time (nseconds);	907	etp_set_max_poll_time (nseconds);
775	}	908	}
776		909
		910	void ecb_cold
777	void eio_set_max_poll_reqs (unsigned int maxreqs)	911	eio_set_max_poll_reqs (unsigned int maxreqs)
778	{	912	{
779	etp_set_max_poll_reqs (maxreqs);	913	etp_set_max_poll_reqs (maxreqs);
780	}	914	}
781		915
		916	void ecb_cold
782	void eio_set_max_idle (unsigned int nthreads)	917	eio_set_max_idle (unsigned int nthreads)
783	{	918	{
784	etp_set_max_idle (nthreads);	919	etp_set_max_idle (nthreads);
785	}	920	}
786		921
		922	void ecb_cold
787	void eio_set_idle_timeout (unsigned int seconds)	923	eio_set_idle_timeout (unsigned int seconds)
788	{	924	{
789	etp_set_idle_timeout (seconds);	925	etp_set_idle_timeout (seconds);
790	}	926	}
791		927
		928	void ecb_cold
792	void eio_set_min_parallel (unsigned int nthreads)	929	eio_set_min_parallel (unsigned int nthreads)
793	{	930	{
794	etp_set_min_parallel (nthreads);	931	etp_set_min_parallel (nthreads);
795	}	932	}
796		933
		934	void ecb_cold
797	void eio_set_max_parallel (unsigned int nthreads)	935	eio_set_max_parallel (unsigned int nthreads)
798	{	936	{
799	etp_set_max_parallel (nthreads);	937	etp_set_max_parallel (nthreads);
800	}	938	}
801		939
802	int eio_poll (void)	940	int eio_poll (void)
…		…
811	# undef pread	949	# undef pread
812	# undef pwrite	950	# undef pwrite
813	# define pread eio__pread	951	# define pread eio__pread
814	# define pwrite eio__pwrite	952	# define pwrite eio__pwrite
815		953
816	static ssize_t	954	static eio_ssize_t
817	eio__pread (int fd, void *buf, size_t count, off_t offset)	955	eio__pread (int fd, void *buf, size_t count, off_t offset)
818	{	956	{
819	ssize_t res;	957	eio_ssize_t res;
820	off_t ooffset;	958	off_t ooffset;
821		959
822	X_LOCK (preadwritelock);	960	X_LOCK (preadwritelock);
823	ooffset = lseek (fd, 0, SEEK_CUR);	961	ooffset = lseek (fd, 0, SEEK_CUR);
824	lseek (fd, offset, SEEK_SET);	962	lseek (fd, offset, SEEK_SET);
…		…
827	X_UNLOCK (preadwritelock);	965	X_UNLOCK (preadwritelock);
828		966
829	return res;	967	return res;
830	}	968	}
831		969
832	static ssize_t	970	static eio_ssize_t
833	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	971	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834	{	972	{
835	ssize_t res;	973	eio_ssize_t res;
836	off_t ooffset;	974	off_t ooffset;
837		975
838	X_LOCK (preadwritelock);	976	X_LOCK (preadwritelock);
839	ooffset = lseek (fd, 0, SEEK_CUR);	977	ooffset = lseek (fd, 0, SEEK_CUR);
840	lseek (fd, offset, SEEK_SET);	978	lseek (fd, offset, SEEK_SET);
…		…
872	#ifndef HAVE_FUTIMES	1010	#ifndef HAVE_FUTIMES
873		1011
874	# undef futimes	1012	# undef futimes
875	# define futimes(fd,times) eio__futimes (fd, times)	1013	# define futimes(fd,times) eio__futimes (fd, times)
876		1014
		1015	static int
877	static int eio__futimes (int fd, const struct timeval tv[2])	1016	eio__futimes (int fd, const struct timeval tv[2])
878	{	1017	{
879	errno = ENOSYS;	1018	errno = ENOSYS;
880	return -1;	1019	return -1;
881	}	1020	}
882		1021
…		…
885	#if !HAVE_FDATASYNC	1024	#if !HAVE_FDATASYNC
886	# undef fdatasync	1025	# undef fdatasync
887	# define fdatasync(fd) fsync (fd)	1026	# define fdatasync(fd) fsync (fd)
888	#endif	1027	#endif
889		1028
		1029	static int
		1030	eio__syncfs (int fd)
		1031	{
		1032	int res;
		1033
		1034	#if HAVE_SYS_SYNCFS
		1035	res = (int)syscall (__NR_syncfs, (int)(fd));
		1036	#else
		1037	res = EIO_ENOSYS ();
		1038	#endif
		1039
		1040	if (res < 0 && errno == ENOSYS && fd >= 0)
		1041	sync ();
		1042
		1043	return res;
		1044	}
		1045
890	/* sync_file_range always needs emulation */	1046	/* sync_file_range always needs emulation */
891	int	1047	static int
892	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	1048	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893	{	1049	{
894	#if HAVE_SYNC_FILE_RANGE	1050	#if HAVE_SYNC_FILE_RANGE
895	int res;	1051	int res;
896		1052
…		…
913	/* even though we could play tricks with the flags, it's better to always	1069	/* even though we could play tricks with the flags, it's better to always
914	* call fdatasync, as that matches the expectation of its users best */	1070	* call fdatasync, as that matches the expectation of its users best */
915	return fdatasync (fd);	1071	return fdatasync (fd);
916	}	1072	}
917		1073
		1074	static int
		1075	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		1076	{
		1077	#if HAVE_LINUX_FALLOCATE
		1078	return fallocate (fd, mode, offset, len);
		1079	#else
		1080	return EIO_ENOSYS ();
		1081	#endif
		1082	}
		1083
918	#if !HAVE_READAHEAD	1084	#if !HAVE_READAHEAD
919	# undef readahead	1085	# undef readahead
920	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1086	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921		1087
922	static ssize_t	1088	static eio_ssize_t
923	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1089	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
924	{	1090	{
925	size_t todo = count;	1091	size_t todo = count;
926	dBUF;	1092	dBUF;
927		1093
…		…
932	pread (fd, eio_buf, len, offset);	1098	pread (fd, eio_buf, len, offset);
933	offset += len;	1099	offset += len;
934	todo -= len;	1100	todo -= len;
935	}	1101	}
936		1102
937	errno = 0;	1103	FUBd;
		1104
		1105	/* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
		1106	/* but not for e.g. EIO or eof, so we also never fail */
938	return count;	1107	return 0;
939	}	1108	}
940		1109
941	#endif	1110	#endif
942		1111
943	/* sendfile always needs emulation */	1112	/* sendfile always needs emulation */
944	static ssize_t	1113	static eio_ssize_t
945	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1114	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
946	{	1115	{
		1116	eio_ssize_t written = 0;
947	ssize_t res;	1117	eio_ssize_t res;
948		1118
949	if (!count)	1119	if (!count)
950	return 0;	1120	return 0;
951		1121
		1122	for (;;)
		1123	{
		1124	#ifdef __APPLE__
		1125	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1126	#endif
952	#if HAVE_SENDFILE	1127	#if HAVE_SENDFILE
953	# if __linux	1128	# if __linux
		1129	off_t soffset = offset;
954	res = sendfile (ofd, ifd, &offset, count);	1130	res = sendfile (ofd, ifd, &soffset, count);
955		1131
956	# elif __FreeBSD__	1132	# elif __FreeBSD__
957	/*	1133	/*
958	* Of course, the freebsd sendfile is a dire hack with no thoughts	1134	* Of course, the freebsd sendfile is a dire hack with no thoughts
959	* wasted on making it similar to other I/O functions.	1135	* wasted on making it similar to other I/O functions.
960	*/	1136	*/
961	{
962	off_t sbytes;	1137	off_t sbytes;
963	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1138	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
964		1139
965	#if 0 /* according to the manpage, this is correct, but broken behaviour */	1140	#if 0 /* according to the manpage, this is correct, but broken behaviour */
966	/* freebsd' sendfile will return 0 on success */	1141	/* freebsd' sendfile will return 0 on success */
967	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */	1142	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
968	/* not on e.g. EIO or EPIPE - sounds broken */	1143	/* not on e.g. EIO or EPIPE - sounds broken */
969	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)	1144	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
970	res = sbytes;	1145	res = sbytes;
971	#endif	1146	#endif
972		1147
973	/* according to source inspection, this is correct, and useful behaviour */	1148	/* according to source inspection, this is correct, and useful behaviour */
974	if (sbytes)	1149	if (sbytes)
975	res = sbytes;	1150	res = sbytes;
976	}
977		1151
978	# elif defined (__APPLE__)	1152	# elif defined __APPLE__
979
980	{
981	off_t sbytes = count;	1153	off_t sbytes = count;
982	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	1154	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983		1155
984	/* according to the manpage, sbytes is always valid */	1156	/* according to the manpage, sbytes is always valid */
985	if (sbytes)	1157	if (sbytes)
986	res = sbytes;	1158	res = sbytes;
987	}
988		1159
989	# elif __hpux	1160	# elif __hpux
990	res = sendfile (ofd, ifd, offset, count, 0, 0);	1161	res = sendfile (ofd, ifd, offset, count, 0, 0);
991		1162
992	# elif __solaris	1163	# elif __solaris
993	{
994	struct sendfilevec vec;	1164	struct sendfilevec vec;
995	size_t sbytes;	1165	size_t sbytes;
996		1166
997	vec.sfv_fd = ifd;	1167	vec.sfv_fd = ifd;
998	vec.sfv_flag = 0;	1168	vec.sfv_flag = 0;
999	vec.sfv_off = offset;	1169	vec.sfv_off = offset;
1000	vec.sfv_len = count;	1170	vec.sfv_len = count;
1001		1171
1002	res = sendfilev (ofd, &vec, 1, &sbytes);	1172	res = sendfilev (ofd, &vec, 1, &sbytes);
1003		1173
1004	if (res < 0 && sbytes)	1174	if (res < 0 && sbytes)
1005	res = sbytes;	1175	res = sbytes;
1006	}
1007		1176
1008	# endif	1177	# endif
1009		1178
1010	#elif defined (_WIN32)	1179	#elif defined (_WIN32) && 0
1011
1012	/* does not work, just for documentation of what would need to be done */	1180	/* does not work, just for documentation of what would need to be done */
1013	{	1181	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1182	/* libeio guarantees that the file offset does not change, and windows */
		1183	/* has no way to get an independent handle to the same file description */
1014	HANDLE h = TO_SOCKET (ifd);	1184	HANDLE h = TO_SOCKET (ifd);
1015	SetFilePointer (h, offset, 0, FILE_BEGIN);	1185	SetFilePointer (h, offset, 0, FILE_BEGIN);
1016	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);	1186	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1017	}
1018		1187
1019	#else	1188	#else
1020	res = -1;	1189	res = EIO_ENOSYS ();
1021	errno = ENOSYS;
1022	#endif	1190	#endif
1023		1191
		1192	/* we assume sendfile can copy at least 128mb in one go */
		1193	if (res <= 128 * 1024 * 1024)
		1194	{
		1195	if (res > 0)
		1196	written += res;
		1197
		1198	if (written)
		1199	return written;
		1200
		1201	break;
		1202	}
		1203	else
		1204	{
		1205	/* if we requested more, then probably the kernel was lazy */
		1206	written += res;
		1207	offset += res;
		1208	count -= res;
		1209
		1210	if (!count)
		1211	return written;
		1212	}
		1213	}
		1214
1024	if (res < 0	1215	if (res < 0
1025	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1216	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1026	/* BSDs */	1217	/* BSDs */
1027	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1218	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1028	|| errno == ENOTSUP	1219	|| errno == ENOTSUP
1029	#endif	1220	#endif
		1221	#ifdef EOPNOTSUPP /* windows */
1030	|| errno == EOPNOTSUPP /* BSDs */	1222	|| errno == EOPNOTSUPP /* BSDs */
		1223	#endif
1031	#if __solaris	1224	#if __solaris
1032	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1225	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1033	#endif	1226	#endif
1034	)	1227	)
1035	)	1228	)
…		…
1039		1232
1040	res = 0;	1233	res = 0;
1041		1234
1042	while (count)	1235	while (count)
1043	{	1236	{
1044	ssize_t cnt;	1237	eio_ssize_t cnt;
1045		1238
1046	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1239	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1047		1240
1048	if (cnt <= 0)	1241	if (cnt <= 0)
1049	{	1242	{
…		…
1061		1254
1062	offset += cnt;	1255	offset += cnt;
1063	res += cnt;	1256	res += cnt;
1064	count -= cnt;	1257	count -= cnt;
1065	}	1258	}
		1259
		1260	FUBd;
1066	}	1261	}
1067		1262
1068	return res;	1263	return res;
		1264	}
		1265
		1266	#ifdef PAGESIZE
		1267	# define eio_pagesize() PAGESIZE
		1268	#else
		1269	static intptr_t
		1270	eio_pagesize (void)
		1271	{
		1272	static intptr_t page;
		1273
		1274	if (!page)
		1275	page = sysconf (_SC_PAGESIZE);
		1276
		1277	return page;
		1278	}
		1279	#endif
		1280
		1281	static void
		1282	eio_page_align (void **addr, size_t *length)
		1283	{
		1284	intptr_t mask = eio_pagesize () - 1;
		1285
		1286	/* round down addr */
		1287	intptr_t adj = mask & (intptr_t)*addr;
		1288
		1289	*addr = (void *)((intptr_t)*addr - adj);
		1290	*length += adj;
		1291
		1292	/* round up length */
		1293	*length = (*length + mask) & ~mask;
		1294	}
		1295
		1296	#if !_POSIX_MEMLOCK
		1297	# define eio__mlockall(a) EIO_ENOSYS ()
		1298	#else
		1299
		1300	static int
		1301	eio__mlockall (int flags)
		1302	{
		1303	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1304	extern int mallopt (int, int);
		1305	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1306	#endif
		1307
		1308	if (EIO_MCL_CURRENT != MCL_CURRENT
		1309	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1310	{
		1311	flags = 0
		1312	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1313	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1314	}
		1315
		1316	return mlockall (flags);
		1317	}
		1318	#endif
		1319
		1320	#if !_POSIX_MEMLOCK_RANGE
		1321	# define eio__mlock(a,b) EIO_ENOSYS ()
		1322	#else
		1323
		1324	static int
		1325	eio__mlock (void *addr, size_t length)
		1326	{
		1327	eio_page_align (&addr, &length);
		1328
		1329	return mlock (addr, length);
		1330	}
		1331
		1332	#endif
		1333
		1334	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1335	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1336	#else
		1337
		1338	static int
		1339	eio__msync (void *mem, size_t len, int flags)
		1340	{
		1341	eio_page_align (&mem, &len);
		1342
		1343	if (EIO_MS_ASYNC != MS_SYNC
		1344	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1345	|| EIO_MS_SYNC != MS_SYNC)
		1346	{
		1347	flags = 0
		1348	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1349	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1350	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1351	}
		1352
		1353	return msync (mem, len, flags);
		1354	}
		1355
		1356	#endif
		1357
		1358	static int
		1359	eio__mtouch (eio_req *req)
		1360	{
		1361	void *mem = req->ptr2;
		1362	size_t len = req->size;
		1363	int flags = req->int1;
		1364
		1365	eio_page_align (&mem, &len);
		1366
		1367	{
		1368	intptr_t addr = (intptr_t)mem;
		1369	intptr_t end = addr + len;
		1370	intptr_t page = eio_pagesize ();
		1371
		1372	if (addr < end)
		1373	if (flags & EIO_MT_MODIFY) /* modify */
		1374	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1375	else
		1376	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1377	}
		1378
		1379	return 0;
		1380	}
		1381
		1382	/*****************************************************************************/
		1383	/* requests implemented outside eio_execute, because they are so large */
		1384
		1385	static void
		1386	eio__lseek (eio_req *req)
		1387	{
		1388	/* this usually gets optimised away completely, or your compiler sucks, */
		1389	/* or the whence constants really are not 0, 1, 2 */
		1390	int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
		1391	: req->int2 == EIO_SEEK_CUR ? SEEK_CUR
		1392	: req->int2 == EIO_SEEK_END ? SEEK_END
		1393	: req->int2;
		1394
		1395	req->offs = lseek (req->int1, req->offs, whence);
		1396	req->result = req->offs == (off_t)-1 ? -1 : 0;
		1397	}
		1398
		1399	/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
		1400	static int
		1401	eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1402	{
		1403	const char *rel = path;
		1404	char *res;
		1405	char *tmp1, *tmp2;
		1406	#if SYMLOOP_MAX > 32
		1407	int symlinks = SYMLOOP_MAX;
		1408	#else
		1409	int symlinks = 32;
		1410	#endif
		1411
		1412	errno = EINVAL;
		1413	if (!rel)
		1414	return -1;
		1415
		1416	errno = ENOENT;
		1417	if (!*rel)
		1418	return -1;
		1419
		1420	res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
		1421	tmp1 = res + PATH_MAX;
		1422	tmp2 = tmp1 + PATH_MAX;
		1423
		1424	#if 0 /* disabled, the musl way to do things is just too racy */
		1425	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1426	/* on linux we may be able to ask the kernel */
		1427	{
		1428	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1429
		1430	if (fd >= 0)
		1431	{
		1432	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1433	req->result = readlink (tmp1, res, PATH_MAX);
		1434	close (fd);
		1435
		1436	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1437
		1438	if (req->result > 0)
		1439	goto done;
		1440	}
		1441	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1442	return;
		1443	}
		1444	#endif
		1445	#endif
		1446
		1447	if (*rel != '/')
		1448	{
		1449	int len;
		1450
		1451	errno = ENOENT;
		1452	if (wd == EIO_INVALID_WD)
		1453	return -1;
		1454
		1455	if (wd == EIO_CWD)
		1456	{
		1457	if (!getcwd (res, PATH_MAX))
		1458	return -1;
		1459
		1460	len = strlen (res);
		1461	}
		1462	else
		1463	memcpy (res, wd->str, len = wd->len);
		1464
		1465	if (res [1]) /* only use if not / */
		1466	res += len;
		1467	}
		1468
		1469	while (*rel)
		1470	{
		1471	eio_ssize_t len, linklen;
		1472	const char *beg = rel;
		1473
		1474	while (*rel && *rel != '/')
		1475	++rel;
		1476
		1477	len = rel - beg;
		1478
		1479	if (!len) /* skip slashes */
		1480	{
		1481	++rel;
		1482	continue;
		1483	}
		1484
		1485	if (beg [0] == '.')
		1486	{
		1487	if (len == 1)
		1488	continue; /* . - nop */
		1489
		1490	if (beg [1] == '.' && len == 2)
		1491	{
		1492	/* .. - back up one component, if possible */
		1493
		1494	while (res != tmpbuf->ptr)
		1495	if (*--res == '/')
		1496	break;
		1497
		1498	continue;
		1499	}
		1500	}
		1501
		1502	errno = ENAMETOOLONG;
		1503	if (res + 1 + len + 1 >= tmp1)
		1504	return -1;
		1505
		1506	/* copy one component */
		1507	*res = '/';
		1508	memcpy (res + 1, beg, len);
		1509
		1510	/* zero-terminate, for readlink */
		1511	res [len + 1] = 0;
		1512
		1513	/* now check if it's a symlink */
		1514	linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
		1515
		1516	if (linklen < 0)
		1517	{
		1518	if (errno != EINVAL)
		1519	return -1;
		1520
		1521	/* it's a normal directory. hopefully */
		1522	res += len + 1;
		1523	}
		1524	else
		1525	{
		1526	/* yay, it was a symlink - build new path in tmp2 */
		1527	int rellen = strlen (rel);
		1528
		1529	errno = ENAMETOOLONG;
		1530	if (linklen + 1 + rellen >= PATH_MAX)
		1531	return -1;
		1532
		1533	errno = ELOOP;
		1534	if (!--symlinks)
		1535	return -1;
		1536
		1537	if (*tmp1 == '/')
		1538	res = tmpbuf->ptr; /* symlink resolves to an absolute path */
		1539
		1540	/* we need to be careful, as rel might point into tmp2 already */
		1541	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1542	tmp2 [linklen] = '/';
		1543	memcpy (tmp2, tmp1, linklen);
		1544
		1545	rel = tmp2;
		1546	}
		1547	}
		1548
		1549	/* special case for the lone root path */
		1550	if (res == tmpbuf->ptr)
		1551	*res++ = '/';
		1552
		1553	return res - (char *)tmpbuf->ptr;
1069	}	1554	}
1070		1555
1071	static signed char	1556	static signed char
1072	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1557	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1073	{	1558	{
…		…
1081		1566
1082	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1567	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1083	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1568	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1084		1569
1085	static void	1570	static void
1086	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1571	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1087	{	1572	{
1088	unsigned char bits [9 + sizeof (ino_t) * 8];	1573	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1089	unsigned char *bit = bits;	1574	unsigned char *bit = bits;
1090		1575
1091	assert (CHAR_BIT == 8);	1576	assert (CHAR_BIT == 8);
1092	assert (sizeof (eio_dirent) * 8 < 256);	1577	assert (sizeof (eio_dirent) * 8 < 256);
1093	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1578	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1095		1580
1096	if (size <= EIO_SORT_FAST)	1581	if (size <= EIO_SORT_FAST)
1097	return;	1582	return;
1098		1583
1099	/* first prepare an array of bits to test in our radix sort */	1584	/* first prepare an array of bits to test in our radix sort */
1100	/* try to take endianness into account, as well as differences in ino_t sizes */	1585	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1101	/* inode_bits must contain all inodes ORed together */	1586	/* inode_bits must contain all inodes ORed together */
1102	/* which is used to skip bits that are 0 everywhere, which is very common */	1587	/* which is used to skip bits that are 0 everywhere, which is very common */
1103	{	1588	{
1104	ino_t endianness;	1589	eio_ino_t endianness;
1105	int i, j;	1590	int i, j;
1106		1591
1107	/* we store the byte offset of byte n into byte n of "endianness" */	1592	/* we store the byte offset of byte n into byte n of "endianness" */
1108	for (i = 0; i < sizeof (ino_t); ++i)	1593	for (i = 0; i < sizeof (eio_ino_t); ++i)
1109	((unsigned char *)&endianness)[i] = i;	1594	((unsigned char *)&endianness)[i] = i;
1110		1595
1111	*bit++ = 0;	1596	*bit++ = 0;
1112		1597
1113	for (i = 0; i < sizeof (ino_t); ++i)	1598	for (i = 0; i < sizeof (eio_ino_t); ++i)
1114	{	1599	{
1115	/* shifting off the byte offsets out of "endianness" */	1600	/* shifting off the byte offsets out of "endianness" */
1116	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1601	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1117	endianness >>= 8;	1602	endianness >>= 8;
1118		1603
1119	for (j = 0; j < 8; ++j)	1604	for (j = 0; j < 8; ++j)
1120	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1605	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1121	*bit++ = offs + j;	1606	*bit++ = offs + j;
1122	}	1607	}
1123		1608
1124	for (j = 0; j < 8; ++j)	1609	for (j = 0; j < 8; ++j)
1125	if (score_bits & (1 << j))	1610	if (score_bits & (1 << j))
1126	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1611	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1127	}	1612	}
1128		1613
1129	/* now actually do the sorting (a variant of MSD radix sort) */	1614	/* now actually do the sorting (a variant of MSD radix sort) */
1130	{	1615	{
1131	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1616	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1132	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1617	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1133	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1618	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1134	int stk_idx = 0;	1619	int stk_idx = 0;
1135		1620
1136	base_stk [stk_idx] = dents;	1621	base_stk [stk_idx] = dents;
1137	end_stk [stk_idx] = dents + size;	1622	end_stk [stk_idx] = dents + size;
1138	bit_stk [stk_idx] = bit - 1;	1623	bit_stk [stk_idx] = bit - 1;
…		…
1217	}	1702	}
1218	}	1703	}
1219	}	1704	}
1220		1705
1221	static void	1706	static void
1222	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1707	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1223	{	1708	{
1224	if (size <= 1)	1709	if (size <= 1)
1225	return; /* our insertion sort relies on size > 0 */	1710	return; /* our insertion sort relies on size > 0 */
1226		1711
1227	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1712	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1235		1720
1236	/* read a full directory */	1721	/* read a full directory */
1237	static void	1722	static void
1238	eio__scandir (eio_req *req, etp_worker *self)	1723	eio__scandir (eio_req *req, etp_worker *self)
1239	{	1724	{
1240	DIR *dirp;
1241	EIO_STRUCT_DIRENT *entp;
1242	char *name, *names;	1725	char *name, *names;
1243	int namesalloc = 4096;	1726	int namesalloc = 4096 - sizeof (void *) * 4;
1244	int namesoffs = 0;	1727	int namesoffs = 0;
1245	int flags = req->int1;	1728	int flags = req->int1;
1246	eio_dirent *dents = 0;	1729	eio_dirent *dents = 0;
1247	int dentalloc = 128;	1730	int dentalloc = 128;
1248	int dentoffs = 0;	1731	int dentoffs = 0;
1249	ino_t inode_bits = 0;	1732	eio_ino_t inode_bits = 0;
		1733	#ifdef _WIN32
		1734	HANDLE dirp;
		1735	WIN32_FIND_DATA entp;
		1736	#else
		1737	DIR *dirp;
		1738	EIO_STRUCT_DIRENT *entp;
		1739	#endif
1250		1740
1251	req->result = -1;	1741	req->result = -1;
1252		1742
1253	if (!(flags & EIO_READDIR_DENTS))	1743	if (!(flags & EIO_READDIR_DENTS))
1254	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1744	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1255		1745
1256	X_LOCK (wrklock);	1746	#ifdef _WIN32
1257	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1747	{
		1748	int len = strlen ((const char *)req->ptr1);
		1749	char *path = malloc (MAX_PATH);
		1750	const char *fmt;
		1751	const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		1752
		1753	if (!len)
		1754	fmt = "./*";
		1755	else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
		1756	fmt = "%s*";
		1757	else
		1758	fmt = "%s/*";
		1759
		1760	_snprintf (path, MAX_PATH, fmt, reqpath);
		1761	dirp = FindFirstFile (path, &entp);
		1762	free (path);
		1763
		1764	if (dirp == INVALID_HANDLE_VALUE)
		1765	{
		1766	/* should steal _dosmaperr */
		1767	switch (GetLastError ())
		1768	{
		1769	case ERROR_FILE_NOT_FOUND:
		1770	req->result = 0;
		1771	break;
		1772
		1773	case ERROR_INVALID_NAME:
		1774	case ERROR_PATH_NOT_FOUND:
		1775	case ERROR_NO_MORE_FILES:
		1776	errno = ENOENT;
		1777	break;
		1778
		1779	case ERROR_NOT_ENOUGH_MEMORY:
		1780	errno = ENOMEM;
		1781	break;
		1782
		1783	default:
		1784	errno = EINVAL;
		1785	break;
		1786	}
		1787
		1788	return;
		1789	}
		1790	}
		1791	#else
		1792	#if HAVE_AT
		1793	if (req->wd)
		1794	{
		1795	int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1796
		1797	if (fd < 0)
		1798	return;
		1799
		1800	dirp = fdopendir (fd);
		1801
		1802	if (!dirp)
		1803	close (fd);
		1804	}
		1805	else
1258	self->dirp = dirp = opendir (req->ptr1);	1806	dirp = opendir (req->ptr1);
		1807	#else
		1808	dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
		1809	#endif
		1810
		1811	if (!dirp)
		1812	return;
		1813	#endif
		1814
		1815	if (req->flags & EIO_FLAG_PTR1_FREE)
		1816	free (req->ptr1);
1259		1817
1260	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1818	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1261	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1819	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1262	req->ptr2 = names = malloc (namesalloc);	1820	req->ptr2 = names = malloc (namesalloc);
1263	X_UNLOCK (wrklock);
1264		1821
1265	if (dirp && names && (!flags || dents))	1822	if (!names || (flags && !dents))
		1823	return;
		1824
1266	for (;;)	1825	for (;;)
1267	{	1826	{
		1827	int done;
		1828
		1829	#ifdef _WIN32
		1830	done = !dirp;
		1831	#else
1268	errno = 0;	1832	errno = 0;
1269	entp = readdir (dirp);	1833	entp = readdir (dirp);
		1834	done = !entp;
		1835	#endif
1270		1836
1271	if (!entp)	1837	if (done)
1272	{	1838	{
		1839	#ifndef _WIN32
		1840	int old_errno = errno;
		1841	closedir (dirp);
		1842	errno = old_errno;
		1843
1273	if (errno)	1844	if (errno)
1274	break;	1845	break;
		1846	#endif
1275		1847
1276	/* sort etc. */	1848	/* sort etc. */
1277	req->int1 = flags;	1849	req->int1 = flags;
1278	req->result = dentoffs;	1850	req->result = dentoffs;
1279		1851
1280	if (flags & EIO_READDIR_STAT_ORDER)	1852	if (flags & EIO_READDIR_STAT_ORDER)
1281	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);	1853	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1282	else if (flags & EIO_READDIR_DIRS_FIRST)	1854	else if (flags & EIO_READDIR_DIRS_FIRST)
1283	if (flags & EIO_READDIR_FOUND_UNKNOWN)	1855	if (flags & EIO_READDIR_FOUND_UNKNOWN)
1284	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */	1856	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1285	else	1857	else
		1858	{
		1859	/* in this case, all is known, and we just put dirs first and sort them */
		1860	eio_dirent *oth = dents + dentoffs;
		1861	eio_dirent *dir = dents;
		1862
		1863	/* now partition dirs to the front, and non-dirs to the back */
		1864	/* by walking from both sides and swapping if necessary */
		1865	while (oth > dir)
		1866	{
		1867	if (dir->type == EIO_DT_DIR)
		1868	++dir;
		1869	else if ((--oth)->type == EIO_DT_DIR)
		1870	{
		1871	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1872
		1873	++dir;
		1874	}
		1875	}
		1876
		1877	/* now sort the dirs only (dirs all have the same score) */
		1878	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1879	}
		1880
		1881	break;
		1882	}
		1883
		1884	/* now add the entry to our list(s) */
		1885	name = D_NAME (entp);
		1886
		1887	/* skip . and .. entries */
		1888	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
		1889	{
		1890	int len = D_NAMLEN (entp) + 1;
		1891
		1892	while (ecb_expect_false (namesoffs + len > namesalloc))
		1893	{
		1894	namesalloc *= 2;
		1895	req->ptr2 = names = realloc (names, namesalloc);
		1896
		1897	if (!names)
		1898	break;
		1899	}
		1900
		1901	memcpy (names + namesoffs, name, len);
		1902
		1903	if (dents)
		1904	{
		1905	struct eio_dirent *ent;
		1906
		1907	if (ecb_expect_false (dentoffs == dentalloc))
1286	{	1908	{
1287	/* in this case, all is known, and we just put dirs first and sort them */	1909	dentalloc *= 2;
		1910	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1911
		1912	if (!dents)
		1913	break;
		1914	}
		1915
1288	eio_dirent *oth = dents + dentoffs;	1916	ent = dents + dentoffs;
1289	eio_dirent *dir = dents;
1290		1917
1291	/* now partition dirs to the front, and non-dirs to the back */	1918	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1292	/* by walking from both sides and swapping if necessary */	1919	ent->namelen = len - 1;
1293	while (oth > dir)	1920	ent->inode = D_INO (entp);
		1921
		1922	inode_bits |= ent->inode;
		1923
		1924	switch (D_TYPE (entp))
		1925	{
		1926	default:
		1927	ent->type = EIO_DT_UNKNOWN;
		1928	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1929	break;
		1930
		1931	#ifdef DT_FIFO
		1932	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1933	#endif
		1934	#ifdef DT_CHR
		1935	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1936	#endif
		1937	#ifdef DT_MPC
		1938	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1939	#endif
		1940	#ifdef DT_DIR
		1941	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1942	#endif
		1943	#ifdef DT_NAM
		1944	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1945	#endif
		1946	#ifdef DT_BLK
		1947	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1948	#endif
		1949	#ifdef DT_MPB
		1950	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1951	#endif
		1952	#ifdef DT_REG
		1953	case DT_REG: ent->type = EIO_DT_REG; break;
		1954	#endif
		1955	#ifdef DT_NWK
		1956	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1957	#endif
		1958	#ifdef DT_CMP
		1959	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1960	#endif
		1961	#ifdef DT_LNK
		1962	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1963	#endif
		1964	#ifdef DT_SOCK
		1965	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1966	#endif
		1967	#ifdef DT_DOOR
		1968	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1969	#endif
		1970	#ifdef DT_WHT
		1971	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1972	#endif
		1973	}
		1974
		1975	ent->score = 7;
		1976
		1977	if (flags & EIO_READDIR_DIRS_FIRST)
		1978	{
		1979	if (ent->type == EIO_DT_UNKNOWN)
1294	{	1980	{
1295	if (dir->type == EIO_DT_DIR)	1981	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1296	++dir;	1982	ent->score = 1;
1297	else if ((--oth)->type == EIO_DT_DIR)	1983	else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1298	{	1984	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1299	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1300
1301	++dir;
1302	}
1303	}	1985	}
1304		1986	else if (ent->type == EIO_DT_DIR)
1305	/* now sort the dirs only (dirs all have the same score) */	1987	ent->score = 0;
1306	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1307	}	1988	}
1308
1309	break;
1310	}
1311
1312	/* now add the entry to our list(s) */
1313	name = entp->d_name;
1314
1315	/* skip . and .. entries */
1316	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1317	{
1318	int len = D_NAMLEN (entp) + 1;
1319
1320	while (expect_false (namesoffs + len > namesalloc))
1321	{
1322	namesalloc *= 2;
1323	X_LOCK (wrklock);
1324	req->ptr2 = names = realloc (names, namesalloc);
1325	X_UNLOCK (wrklock);
1326
1327	if (!names)
1328	break;
1329	}	1989	}
1330		1990
1331	memcpy (names + namesoffs, name, len);
1332
1333	if (dents)
1334	{
1335	struct eio_dirent *ent;
1336
1337	if (expect_false (dentoffs == dentalloc))
1338	{
1339	dentalloc *= 2;
1340	X_LOCK (wrklock);
1341	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1342	X_UNLOCK (wrklock);
1343
1344	if (!dents)
1345	break;
1346	}
1347
1348	ent = dents + dentoffs;
1349
1350	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1351	ent->namelen = len - 1;
1352	ent->inode = D_INO (entp);
1353
1354	inode_bits |= ent->inode;
1355
1356	switch (D_TYPE (entp))
1357	{
1358	default:
1359	ent->type = EIO_DT_UNKNOWN;
1360	flags |= EIO_READDIR_FOUND_UNKNOWN;
1361	break;
1362
1363	#ifdef DT_FIFO
1364	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1365	#endif
1366	#ifdef DT_CHR
1367	case DT_CHR: ent->type = EIO_DT_CHR; break;
1368	#endif
1369	#ifdef DT_MPC
1370	case DT_MPC: ent->type = EIO_DT_MPC; break;
1371	#endif
1372	#ifdef DT_DIR
1373	case DT_DIR: ent->type = EIO_DT_DIR; break;
1374	#endif
1375	#ifdef DT_NAM
1376	case DT_NAM: ent->type = EIO_DT_NAM; break;
1377	#endif
1378	#ifdef DT_BLK
1379	case DT_BLK: ent->type = EIO_DT_BLK; break;
1380	#endif
1381	#ifdef DT_MPB
1382	case DT_MPB: ent->type = EIO_DT_MPB; break;
1383	#endif
1384	#ifdef DT_REG
1385	case DT_REG: ent->type = EIO_DT_REG; break;
1386	#endif
1387	#ifdef DT_NWK
1388	case DT_NWK: ent->type = EIO_DT_NWK; break;
1389	#endif
1390	#ifdef DT_CMP
1391	case DT_CMP: ent->type = EIO_DT_CMP; break;
1392	#endif
1393	#ifdef DT_LNK
1394	case DT_LNK: ent->type = EIO_DT_LNK; break;
1395	#endif
1396	#ifdef DT_SOCK
1397	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1398	#endif
1399	#ifdef DT_DOOR
1400	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1401	#endif
1402	#ifdef DT_WHT
1403	case DT_WHT: ent->type = EIO_DT_WHT; break;
1404	#endif
1405	}
1406
1407	ent->score = 7;
1408
1409	if (flags & EIO_READDIR_DIRS_FIRST)
1410	{
1411	if (ent->type == EIO_DT_UNKNOWN)
1412	{
1413	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1414	ent->score = 1;
1415	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1416	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1417	}
1418	else if (ent->type == EIO_DT_DIR)
1419	ent->score = 0;
1420	}
1421	}
1422
1423	namesoffs += len;	1991	namesoffs += len;
1424	++dentoffs;	1992	++dentoffs;
1425	}	1993	}
1426		1994
1427	if (EIO_CANCELLED (req))	1995	if (EIO_CANCELLED (req))
1428	{	1996	{
1429	errno = ECANCELED;	1997	errno = ECANCELED;
1430	break;	1998	break;
1431	}	1999	}
		2000
		2001	#ifdef _WIN32
		2002	if (!FindNextFile (dirp, &entp))
		2003	{
		2004	FindClose (dirp);
		2005	dirp = 0;
1432	}	2006	}
		2007	#endif
		2008	}
1433	}	2009	}
1434		2010
1435	#ifdef PAGESIZE	2011	/*****************************************************************************/
1436	# define eio_pagesize() PAGESIZE	2012	/* working directory stuff */
1437	#else	2013	/* various deficiencies in the posix 2008 api force us to */
1438	static intptr_t	2014	/* keep the absolute path in string form at all times */
1439	eio_pagesize (void)	2015	/* fuck yeah. */
1440	{
1441	static intptr_t page;
1442		2016
1443	if (!page)	2017	#if !HAVE_AT
1444	page = sysconf (_SC_PAGESIZE);
1445		2018
		2019	/* a bit like realpath, but usually faster because it doesn'T have to return */
		2020	/* an absolute or canonical path */
		2021	static const char *
		2022	wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2023	{
		2024	if (!wd || *path == '/')
		2025	return path;
		2026
		2027	if (path [0] == '.' && !path [1])
		2028	return wd->str;
		2029
		2030	{
		2031	int l1 = wd->len;
		2032	int l2 = strlen (path);
		2033
		2034	char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
		2035
		2036	memcpy (res, wd->str, l1);
		2037	res [l1] = '/';
		2038	memcpy (res + l1 + 1, path, l2 + 1);
		2039
		2040	return res;
		2041	}
		2042	}
		2043
		2044	#endif
		2045
		2046	static eio_wd
		2047	eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2048	{
		2049	int fd;
		2050	eio_wd res;
		2051	int len = eio__realpath (tmpbuf, wd, path);
		2052
		2053	if (len < 0)
		2054	return EIO_INVALID_WD;
		2055
		2056	#if HAVE_AT
		2057	fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		2058
		2059	if (fd < 0)
		2060	return EIO_INVALID_WD;
		2061	#endif
		2062
		2063	res = malloc (sizeof (*res) + len); /* one extra 0-byte */
		2064
		2065	#if HAVE_AT
		2066	res->fd = fd;
		2067	#endif
		2068
		2069	res->len = len;
		2070	memcpy (res->str, tmpbuf->ptr, len);
		2071	res->str [len] = 0;
		2072
1446	return page;	2073	return res;
1447	}	2074	}
1448	#endif
1449		2075
1450	static void	2076	eio_wd
1451	eio_page_align (void **addr, size_t *length)	2077	eio_wd_open_sync (eio_wd wd, const char *path)
1452	{	2078	{
1453	intptr_t mask = eio_pagesize () - 1;	2079	struct tmpbuf tmpbuf = { 0 };
		2080	wd = eio__wd_open_sync (&tmpbuf, wd, path);
		2081	free (tmpbuf.ptr);
1454		2082
1455	/* round down addr */	2083	return wd;
1456	intptr_t adj = mask & (intptr_t)*addr;
1457
1458	*addr = (void *)((intptr_t)*addr - adj);
1459	*length += adj;
1460
1461	/* round up length */
1462	*length = (*length + mask) & ~mask;
1463	}	2084	}
1464		2085
1465	#if !_POSIX_MEMLOCK	2086	void
1466	# define eio__mlockall(a) ((errno = ENOSYS), -1)	2087	eio_wd_close_sync (eio_wd wd)
1467	#else	2088	{
		2089	if (wd != EIO_INVALID_WD && wd != EIO_CWD)
		2090	{
		2091	#if HAVE_AT
		2092	close (wd->fd);
		2093	#endif
		2094	free (wd);
		2095	}
		2096	}
		2097
		2098	#if HAVE_AT
		2099
		2100	/* they forgot these */
1468		2101
1469	static int	2102	static int
1470	eio__mlockall (int flags)	2103	eio__truncateat (int dirfd, const char *path, off_t length)
1471	{	2104	{
1472	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7	2105	int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1473	extern int mallopt (int, int);	2106	int res;
1474	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1475	#endif
1476		2107
1477	if (EIO_MCL_CURRENT != MCL_CURRENT	2108	if (fd < 0)
1478	|| EIO_MCL_FUTURE != MCL_FUTURE)	2109	return fd;
1479	{
1480	flags = 0
1481	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1482	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1483	}
1484		2110
1485	return mlockall (flags);	2111	res = ftruncate (fd, length);
		2112	close (fd);
		2113	return res;
1486	}	2114	}
1487	#endif
1488
1489	#if !_POSIX_MEMLOCK_RANGE
1490	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1491	#else
1492		2115
1493	static int	2116	static int
1494	eio__mlock (void *addr, size_t length)	2117	eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1495	{	2118	{
1496	eio_page_align (&addr, &length);	2119	int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
		2120	int res;
1497		2121
1498	return mlock (addr, length);	2122	if (fd < 0)
1499	}	2123	return fd;
1500		2124
1501	#endif	2125	res = fstatvfs (fd, buf);
1502		2126	close (fd);
1503	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1504	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1505	#else
1506
1507	int
1508	eio__msync (void *mem, size_t len, int flags)
1509	{
1510	eio_page_align (&mem, &len);
1511
1512	if (EIO_MS_ASYNC != MS_SYNC
1513	|| EIO_MS_INVALIDATE != MS_INVALIDATE
1514	|| EIO_MS_SYNC != MS_SYNC)
1515	{
1516	flags = 0
1517	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1518	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1519	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1520	}
1521
1522	return msync (mem, len, flags);
1523	}
1524
1525	#endif
1526
1527	int
1528	eio__mtouch (eio_req *req)
1529	{
1530	void *mem = req->ptr2;
1531	size_t len = req->size;
1532	int flags = req->int1;
1533
1534	eio_page_align (&mem, &len);
1535
1536	{
1537	intptr_t addr = (intptr_t)mem;
1538	intptr_t end = addr + len;
1539	intptr_t page = eio_pagesize ();
1540
1541	if (addr < end)
1542	if (flags & EIO_MT_MODIFY) /* modify */
1543	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1544	else
1545	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1546	}
1547
1548	return 0;	2127	return res;
		2128
1549	}	2129	}
		2130
		2131	#endif
1550		2132
1551	/*****************************************************************************/	2133	/*****************************************************************************/
1552		2134
1553	#define ALLOC(len) \	2135	#define ALLOC(len) \
1554	if (!req->ptr2) \	2136	if (!req->ptr2) \
…		…
1563	req->result = -1; \	2145	req->result = -1; \
1564	break; \	2146	break; \
1565	} \	2147	} \
1566	}	2148	}
1567		2149
		2150	static void ecb_noinline ecb_cold
		2151	etp_proc_init (void)
		2152	{
		2153	#if HAVE_PRCTL_SET_NAME
		2154	/* provide a more sensible "thread name" */
		2155	char name[16 + 1];
		2156	const int namelen = sizeof (name) - 1;
		2157	int len;
		2158
		2159	prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
		2160	name [namelen] = 0;
		2161	len = strlen (name);
		2162	strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
		2163	prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
		2164	#endif
		2165	}
		2166
1568	X_THREAD_PROC (etp_proc)	2167	X_THREAD_PROC (etp_proc)
1569	{	2168	{
1570	ETP_REQ *req;	2169	ETP_REQ *req;
1571	struct timespec ts;	2170	struct timespec ts;
1572	etp_worker *self = (etp_worker *)thr_arg;	2171	etp_worker *self = (etp_worker *)thr_arg;
1573		2172
		2173	etp_proc_init ();
		2174
1574	/* try to distribute timeouts somewhat randomly */	2175	/* try to distribute timeouts somewhat evenly */
1575	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	2176	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1576		2177
1577	for (;;)	2178	for (;;)
1578	{	2179	{
		2180	ts.tv_sec = 0;
		2181
1579	X_LOCK (reqlock);	2182	X_LOCK (reqlock);
1580		2183
1581	for (;;)	2184	for (;;)
1582	{	2185	{
1583	self->req = req = reqq_shift (&req_queue);	2186	req = reqq_shift (&req_queue);
1584		2187
1585	if (req)	2188	if (req)
1586	break;	2189	break;
1587		2190
		2191	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		2192	{
		2193	X_UNLOCK (reqlock);
		2194	X_LOCK (wrklock);
		2195	--started;
		2196	X_UNLOCK (wrklock);
		2197	goto quit;
		2198	}
		2199
1588	++idle;	2200	++idle;
1589		2201
1590	ts.tv_sec = time (0) + idle_timeout;	2202	if (idle <= max_idle)
1591	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	2203	/* we are allowed to idle, so do so without any timeout */
		2204	X_COND_WAIT (reqwait, reqlock);
		2205	else
1592	{	2206	{
1593	if (idle > max_idle)	2207	/* initialise timeout once */
1594	{	2208	if (!ts.tv_sec)
1595	--idle;	2209	ts.tv_sec = time (0) + idle_timeout;
1596	X_UNLOCK (reqlock);
1597	X_LOCK (wrklock);
1598	--started;
1599	X_UNLOCK (wrklock);
1600	goto quit;
1601	}
1602		2210
1603	/* we are allowed to idle, so do so without any timeout */
1604	X_COND_WAIT (reqwait, reqlock);	2211	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		2212	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1605	}	2213	}
1606		2214
1607	--idle;	2215	--idle;
1608	}	2216	}
1609		2217
…		…
1612	X_UNLOCK (reqlock);	2220	X_UNLOCK (reqlock);
1613		2221
1614	if (req->type < 0)	2222	if (req->type < 0)
1615	goto quit;	2223	goto quit;
1616		2224
1617	if (!EIO_CANCELLED (req))
1618	ETP_EXECUTE (self, req);	2225	ETP_EXECUTE (self, req);
1619		2226
1620	X_LOCK (reslock);	2227	X_LOCK (reslock);
1621		2228
1622	++npending;	2229	++npending;
1623		2230
1624	if (!reqq_push (&res_queue, req) && want_poll_cb)	2231	if (!reqq_push (&res_queue, req) && want_poll_cb)
1625	want_poll_cb ();	2232	want_poll_cb ();
1626		2233
1627	self->req = 0;
1628	etp_worker_clear (self);	2234	etp_worker_clear (self);
1629		2235
1630	X_UNLOCK (reslock);	2236	X_UNLOCK (reslock);
1631	}	2237	}
1632		2238
1633	quit:	2239	quit:
		2240	free (req);
		2241
1634	X_LOCK (wrklock);	2242	X_LOCK (wrklock);
1635	etp_worker_free (self);	2243	etp_worker_free (self);
1636	X_UNLOCK (wrklock);	2244	X_UNLOCK (wrklock);
1637
1638	return 0;
1639	}	2245	}
1640		2246
1641	/*****************************************************************************/	2247	/*****************************************************************************/
1642		2248
		2249	int ecb_cold
1643	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	2250	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1644	{	2251	{
		2252	#if !HAVE_PREADWRITE
		2253	X_MUTEX_CREATE (preadwritelock);
		2254	#endif
		2255
1645	return etp_init (want_poll, done_poll);	2256	return etp_init (want_poll, done_poll);
1646	}	2257	}
1647		2258
		2259	ecb_inline void
1648	static void eio_api_destroy (eio_req *req)	2260	eio_api_destroy (eio_req *req)
1649	{	2261	{
1650	free (req);	2262	free (req);
1651	}	2263	}
1652		2264
1653	#define REQ(rtype) \	2265	#define REQ(rtype) \
…		…
1672	{ \	2284	{ \
1673	eio_api_destroy (req); \	2285	eio_api_destroy (req); \
1674	return 0; \	2286	return 0; \
1675	}	2287	}
1676		2288
		2289	static void
1677	static void eio_execute (etp_worker *self, eio_req *req)	2290	eio_execute (etp_worker *self, eio_req *req)
1678	{	2291	{
		2292	#if HAVE_AT
		2293	int dirfd;
		2294	#else
		2295	const char *path;
		2296	#endif
		2297
		2298	if (ecb_expect_false (EIO_CANCELLED (req)))
		2299	{
		2300	req->result = -1;
		2301	req->errorno = ECANCELED;
		2302	return;
		2303	}
		2304
		2305	if (ecb_expect_false (req->wd == EIO_INVALID_WD))
		2306	{
		2307	req->result = -1;
		2308	req->errorno = ENOENT;
		2309	return;
		2310	}
		2311
		2312	if (req->type >= EIO_OPEN)
		2313	{
		2314	#if HAVE_AT
		2315	dirfd = WD2FD (req->wd);
		2316	#else
		2317	path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		2318	#endif
		2319	}
		2320
1679	switch (req->type)	2321	switch (req->type)
1680	{	2322	{
		2323	case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
		2324	req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
		2325	break;
		2326	case EIO_WD_CLOSE: req->result = 0;
		2327	eio_wd_close_sync (req->wd); break;
		2328
		2329	case EIO_SEEK: eio__lseek (req); break;
1681	case EIO_READ: ALLOC (req->size);	2330	case EIO_READ: ALLOC (req->size);
1682	req->result = req->offs >= 0	2331	req->result = req->offs >= 0
1683	? pread (req->int1, req->ptr2, req->size, req->offs)	2332	? pread (req->int1, req->ptr2, req->size, req->offs)
1684	: read (req->int1, req->ptr2, req->size); break;	2333	: read (req->int1, req->ptr2, req->size); break;
1685	case EIO_WRITE: req->result = req->offs >= 0	2334	case EIO_WRITE: req->result = req->offs >= 0
1686	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2335	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1687	: write (req->int1, req->ptr2, req->size); break;	2336	: write (req->int1, req->ptr2, req->size); break;
1688		2337
1689	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2338	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1690	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2339	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
		2340
		2341	#if HAVE_AT
1691		2342
1692	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2343	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1693	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2344	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1694	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2345	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2346	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
		2347	case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
		2348	case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
		2349	case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
		2350	case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
		2351
		2352	case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
		2353	case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
		2354	case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
		2355	case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
		2356	case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
		2357	case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
		2358	case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2359	case EIO_READLINK: ALLOC (PATH_MAX);
		2360	req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
		2361	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2362	req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2363	case EIO_UTIME:
		2364	case EIO_FUTIME:
		2365	{
		2366	struct timespec ts[2];
		2367	struct timespec *times;
		2368
		2369	if (req->nv1 != -1. || req->nv2 != -1.)
		2370	{
		2371	ts[0].tv_sec = req->nv1;
		2372	ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
		2373	ts[1].tv_sec = req->nv2;
		2374	ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
		2375
		2376	times = ts;
		2377	}
		2378	else
		2379	times = 0;
		2380
		2381	req->result = req->type == EIO_FUTIME
		2382	? futimens (req->int1, times)
		2383	: utimensat (dirfd, req->ptr1, times, 0);
		2384	}
		2385	break;
		2386
		2387	#else
		2388
		2389	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2390	req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2391	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1695	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2392	req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2393	case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
		2394	case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
		2395	case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
		2396	case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
		2397
		2398	case EIO_UNLINK: req->result = unlink (path ); break;
		2399	case EIO_RMDIR: req->result = rmdir (path ); break;
		2400	case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
		2401	case EIO_RENAME: req->result = rename (path , req->ptr2); break;
		2402	case EIO_LINK: req->result = link (path , req->ptr2); break;
		2403	case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
		2404	case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
		2405	case EIO_READLINK: ALLOC (PATH_MAX);
		2406	req->result = readlink (path, req->ptr2, PATH_MAX); break;
		2407	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2408	req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2409
		2410	case EIO_UTIME:
		2411	case EIO_FUTIME:
		2412	{
		2413	struct timeval tv[2];
		2414	struct timeval *times;
		2415
		2416	if (req->nv1 != -1. || req->nv2 != -1.)
		2417	{
		2418	tv[0].tv_sec = req->nv1;
		2419	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
		2420	tv[1].tv_sec = req->nv2;
		2421	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
		2422
		2423	times = tv;
		2424	}
		2425	else
		2426	times = 0;
		2427
		2428	req->result = req->type == EIO_FUTIME
		2429	? futimes (req->int1, times)
		2430	: utimes (req->ptr1, times);
		2431	}
		2432	break;
		2433
		2434	#endif
		2435
		2436	case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
		2437	{
		2438	ALLOC (req->result);
		2439	memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
		2440	}
		2441	break;
		2442
1696	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2443	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1697	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2444	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1698		2445
1699	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1700	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1701	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));	2446	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1702	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;	2447	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1703		2448
1704	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1705	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	2449	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1706	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1707	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	2450	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1708	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1709	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;	2451	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1710		2452
1711	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1712	case EIO_CLOSE: req->result = close (req->int1); break;	2453	case EIO_CLOSE: req->result = close (req->int1); break;
1713	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;	2454	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1714	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1715	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1716	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1717	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1718	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1719	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1720	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1721
1722	case EIO_READLINK: ALLOC (PATH_MAX);
1723	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1724
1725	case EIO_SYNC: req->result = 0; sync (); break;	2455	case EIO_SYNC: req->result = 0; sync (); break;
1726	case EIO_FSYNC: req->result = fsync (req->int1); break;	2456	case EIO_FSYNC: req->result = fsync (req->int1); break;
1727	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2457	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2458	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2459	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1728	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2460	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1729	case EIO_MTOUCH: req->result = eio__mtouch (req); break;	2461	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1730	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2462	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1731	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2463	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1732	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	2464	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1733		2465
1734	case EIO_READDIR: eio__scandir (req, self); break;	2466	case EIO_READDIR: eio__scandir (req, self); break;
1735		2467
1736	case EIO_BUSY:	2468	case EIO_BUSY:
1737	#ifdef _WIN32	2469	#ifdef _WIN32
…		…
1746	req->result = select (0, 0, 0, 0, &tv);	2478	req->result = select (0, 0, 0, 0, &tv);
1747	}	2479	}
1748	#endif	2480	#endif
1749	break;	2481	break;
1750		2482
1751	case EIO_UTIME:
1752	case EIO_FUTIME:
1753	{
1754	struct timeval tv[2];
1755	struct timeval *times;
1756
1757	if (req->nv1 != -1. || req->nv2 != -1.)
1758	{
1759	tv[0].tv_sec = req->nv1;
1760	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1761	tv[1].tv_sec = req->nv2;
1762	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1763
1764	times = tv;
1765	}
1766	else
1767	times = 0;
1768
1769	req->result = req->type == EIO_FUTIME
1770	? futimes (req->int1, times)
1771	: utimes (req->ptr1, times);
1772	}
1773	break;
1774
1775	case EIO_GROUP:	2483	case EIO_GROUP:
1776	abort (); /* handled in eio_request */	2484	abort (); /* handled in eio_request */
1777		2485
1778	case EIO_NOP:	2486	case EIO_NOP:
1779	req->result = 0;	2487	req->result = 0;
…		…
1782	case EIO_CUSTOM:	2490	case EIO_CUSTOM:
1783	req->feed (req);	2491	req->feed (req);
1784	break;	2492	break;
1785		2493
1786	default:	2494	default:
1787	errno = ENOSYS;
1788	req->result = -1;	2495	req->result = EIO_ENOSYS ();
1789	break;	2496	break;
1790	}	2497	}
1791		2498
1792	req->errorno = errno;	2499	req->errorno = errno;
1793	}	2500	}
1794		2501
1795	#ifndef EIO_NO_WRAPPERS	2502	#ifndef EIO_NO_WRAPPERS
1796		2503
		2504	eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
		2505	{
		2506	REQ (EIO_WD_OPEN); PATH; SEND;
		2507	}
		2508
		2509	eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
		2510	{
		2511	REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
		2512	}
		2513
1797	eio_req *eio_nop (int pri, eio_cb cb, void *data)	2514	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1798	{	2515	{
1799	REQ (EIO_NOP); SEND;	2516	REQ (EIO_NOP); SEND;
1800	}	2517	}
1801		2518
…		…
1817	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2534	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1818	{	2535	{
1819	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2536	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1820	}	2537	}
1821		2538
		2539	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		2540	{
		2541	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2542	}
		2543
		2544	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2545	{
		2546	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2547	}
		2548
		2549	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2550	{
		2551	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2552	}
		2553
1822	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2554	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1823	{	2555	{
1824	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2556	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1825	}	2557	}
1826		2558
…		…
1832	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)	2564	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1833	{	2565	{
1834	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;	2566	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1835	}	2567	}
1836		2568
1837	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2569	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1838	{	2570	{
1839	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2571	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1840	}
1841
1842	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1843	{
1844	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1845	}	2572	}
1846		2573
1847	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2574	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1848	{	2575	{
1849	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2576	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
1852	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)	2579	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1853	{	2580	{
1854	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;	2581	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1855	}	2582	}
1856		2583
		2584	eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
		2585	{
		2586	REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
		2587	}
		2588
1857	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)	2589	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1858	{	2590	{
1859	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;	2591	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1860	}	2592	}
1861		2593
…		…
1887	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2619	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1888	{	2620	{
1889	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2621	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1890	}	2622	}
1891		2623
1892	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2624	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1893	{	2625	{
1894	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2626	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1895	}	2627	}
1896		2628
1897	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2629	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1917	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2649	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1918	{	2650	{
1919	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2651	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1920	}	2652	}
1921		2653
1922	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2654	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1923	{	2655	{
1924	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2656	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1925	}	2657	}
1926		2658
1927	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2659	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1941	}	2673	}
1942		2674
1943	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2675	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1944	{	2676	{
1945	return eio__1path (EIO_READLINK, path, pri, cb, data);	2677	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2678	}
		2679
		2680	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2681	{
		2682	return eio__1path (EIO_REALPATH, path, pri, cb, data);
1946	}	2683	}
1947		2684
1948	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2685	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1949	{	2686	{
1950	return eio__1path (EIO_STAT, path, pri, cb, data);	2687	return eio__1path (EIO_STAT, path, pri, cb, data);
…		…
2009	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2746	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2010	{	2747	{
2011	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2748	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2012	}	2749	}
2013		2750
2014	eio_req *eio_custom (void (*)(eio_req *) execute, int pri, eio_cb cb, void *data);	2751	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2015	{	2752	{
2016	REQ (EIO_CUSTOM); req->feed = execute; SEND;	2753	REQ (EIO_CUSTOM); req->feed = execute; SEND;
2017	}	2754	}
2018		2755
2019	#endif	2756	#endif
…		…
2030	#undef SEND	2767	#undef SEND
2031		2768
2032	/*****************************************************************************/	2769	/*****************************************************************************/
2033	/* grp functions */	2770	/* grp functions */
2034		2771
		2772	void
2035	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2773	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2036	{	2774	{
2037	grp->int2 = limit;	2775	grp->int2 = limit;
2038	grp->feed = feed;	2776	grp->feed = feed;
2039		2777
2040	grp_try_feed (grp);	2778	grp_try_feed (grp);
2041	}	2779	}
2042		2780
		2781	void
2043	void eio_grp_limit (eio_req *grp, int limit)	2782	eio_grp_limit (eio_req *grp, int limit)
2044	{	2783	{
2045	grp->int2 = limit;	2784	grp->int2 = limit;
2046		2785
2047	grp_try_feed (grp);	2786	grp_try_feed (grp);
2048	}	2787	}
2049		2788
		2789	void
2050	void eio_grp_add (eio_req *grp, eio_req *req)	2790	eio_grp_add (eio_req *grp, eio_req *req)
2051	{	2791	{
2052	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2792	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2053		2793
2054	grp->flags |= EIO_FLAG_GROUPADD;	2794	grp->flags |= EIO_FLAG_GROUPADD;
2055		2795
…		…
2066	}	2806	}
2067		2807
2068	/*****************************************************************************/	2808	/*****************************************************************************/
2069	/* misc garbage */	2809	/* misc garbage */
2070		2810
		2811	eio_ssize_t
2071	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2812	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2072	{	2813	{
2073	etp_worker wrk;
2074	ssize_t ret;
2075
2076	wrk.dbuf = 0;
2077
2078	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2814	return eio__sendfile (ofd, ifd, offset, count);
2079
2080	if (wrk.dbuf)
2081	free (wrk.dbuf);
2082
2083	return ret;
2084	}	2815	}
2085		2816

Diff Legend

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