ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.53 by root, Sat Jan 9 10:03:09 2010 UTC vs.
Revision 1.110 by root, Tue Sep 27 12:36:19 2011 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008,2009,2010 Marc Alexander Lehmann <libeio@schmorp.de>	4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5	* All rights reserved.	5	* All rights reserved.
6	*	6	*
7	* Redistribution and use in source and binary forms, with or without modifica-	7	* Redistribution and use in source and binary forms, with or without modifica-
8	* tion, are permitted provided that the following conditions are met:	8	* tion, are permitted provided that the following conditions are met:
9	*	9	*
…		…
35	* and other provisions required by the GPL. If you do not delete the	35	* and other provisions required by the GPL. If you do not delete the
36	* provisions above, a recipient may use your version of this file under	36	* provisions above, a recipient may use your version of this file under
37	* either the BSD or the GPL.	37	* either the BSD or the GPL.
38	*/	38	*/
39		39
		40	#ifndef _WIN32
		41	# include "config.h"
		42	#endif
		43
40	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
41		46
42	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
43	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define X_STACKSIZE EIO_STACKSIZE
44	#endif	49	#endif
45	#include "xthread.h"	50	#include "xthread.h"
46		51
47	#include <errno.h>	52	#include <errno.h>
48	#include <stddef.h>	53	#include <stddef.h>
49	#include <stdlib.h>	54	#include <stdlib.h>
50	#include <string.h>	55	#include <string.h>
51	#include <errno.h>	56	#include <errno.h>
52	#include <sys/types.h>	57	#include <sys/types.h>
53	#include <sys/stat.h>	58	#include <sys/stat.h>
54	#include <sys/statvfs.h>
55	#include <limits.h>	59	#include <limits.h>
56	#include <fcntl.h>	60	#include <fcntl.h>
57	#include <assert.h>	61	#include <assert.h>
58		62
		63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
		64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
		65	#if HAVE_STDINT_H
		66	# include <stdint.h>
		67	#endif
		68
		69	#ifndef ECANCELED
		70	# define ECANCELED EDOM
		71	#endif
		72	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
		77	# define ENOTSOCK WSAENOTSOCK
		78	#endif
		79
		80	static void eio_destroy (eio_req *req);
		81
59	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
60	# 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
61	#endif	84	#endif
62		85
63	#ifndef EIO_DESTROY	86	#ifndef EIO_DESTROY
…		…
66		89
67	#ifndef EIO_FEED	90	#ifndef EIO_FEED
68	# 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)
69	#endif	92	#endif
70		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
71	#ifdef _WIN32	105	#ifdef _WIN32
72		106
73	/*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
74	#else	197	#else
75		198
76	# include "config.h"
77	# include <sys/time.h>	199	#include <sys/time.h>
78	# include <sys/select.h>	200	#include <sys/select.h>
79	# include <sys/mman.h>	201	#include <sys/statvfs.h>
80	# include <unistd.h>	202	#include <unistd.h>
		203	#include <signal.h>
		204	#include <dirent.h>
		205
		206	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		207	#include <sys/mman.h>
		208	#endif
		209
		210	#define D_NAME(entp) entp->d_name
		211
		212	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		213	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		214	#define _DIRENT_HAVE_D_TYPE /* sigh */
		215	#define D_INO(de) (de)->d_fileno
		216	#define D_NAMLEN(de) (de)->d_namlen
		217	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		218	#define D_INO(de) (de)->d_ino
		219	#endif
		220
		221	#ifdef _D_EXACT_NAMLEN
		222	#undef D_NAMLEN
		223	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		224	#endif
		225
		226	#ifdef _DIRENT_HAVE_D_TYPE
		227	#define D_TYPE(de) (de)->d_type
		228	#endif
		229
		230	#ifndef EIO_STRUCT_DIRENT
		231	#define EIO_STRUCT_DIRENT struct dirent
		232	#endif
		233
		234	#endif
		235
		236	#if HAVE_UTIMES
81	# include <utime.h>	237	# include <utime.h>
82	# include <signal.h>
83	# include <dirent.h>
84
85	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
86	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
87	# define _DIRENT_HAVE_D_TYPE /* sigh */
88	# define D_INO(de) (de)->d_fileno
89	# define D_NAMLEN(de) (de)->d_namlen
90	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
91	# define D_INO(de) (de)->d_ino
92	# endif	238	#endif
93		239
94	#ifdef _D_EXACT_NAMLEN	240	#if HAVE_SYS_SYSCALL_H
95	# undef D_NAMLEN	241	# include <sys/syscall.h>
96	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
97	#endif
98
99	# ifdef _DIRENT_HAVE_D_TYPE
100	# define D_TYPE(de) (de)->d_type
101	# endif	242	#endif
102		243
103	# ifndef EIO_STRUCT_DIRENT	244	#if HAVE_SYS_PRCTL_H
104	# define EIO_STRUCT_DIRENT struct dirent	245	# include <sys/prctl.h>
105	# endif
106
107	#endif	246	#endif
108		247
109	#if HAVE_SENDFILE	248	#if HAVE_SENDFILE
110	# if __linux	249	# if __linux
111	# include <sys/sendfile.h>	250	# include <sys/sendfile.h>
…		…
126	#endif	265	#endif
127	#ifndef D_INO	266	#ifndef D_INO
128	# define D_INO(de) 0	267	# define D_INO(de) 0
129	#endif	268	#endif
130	#ifndef D_NAMLEN	269	#ifndef D_NAMLEN
131	# define D_NAMLEN(de) strlen ((de)->d_name)	270	# define D_NAMLEN(entp) strlen (D_NAME (entp))
132	#endif	271	#endif
133
134	/* number of seconds after which an idle threads exit */
135	#define IDLE_TIMEOUT 10
136		272
137	/* used for struct dirent, AIX doesn't provide it */	273	/* used for struct dirent, AIX doesn't provide it */
138	#ifndef NAME_MAX	274	#ifndef NAME_MAX
139	# define NAME_MAX 4096	275	# define NAME_MAX 4096
140	#endif	276	#endif
141		277
		278	/* used for readlink etc. */
		279	#ifndef PATH_MAX
		280	# define PATH_MAX 4096
		281	#endif
		282
142	/* buffer size for various temporary buffers */	283	/* buffer size for various temporary buffers */
143	#define EIO_BUFSIZE 65536	284	#define EIO_BUFSIZE 65536
144		285
145	#define dBUF \	286	#define dBUF \
146	char *eio_buf; \
147	ETP_WORKER_LOCK (self); \
148	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	287	char *eio_buf = malloc (EIO_BUFSIZE); \
149	ETP_WORKER_UNLOCK (self); \
150	errno = ENOMEM; \	288	errno = ENOMEM; \
151	if (!eio_buf) \	289	if (!eio_buf) \
152	return -1;	290	return -1
		291
		292	#define FUBd \
		293	free (eio_buf)
153		294
154	#define EIO_TICKS ((1000000 + 1023) >> 10)	295	#define EIO_TICKS ((1000000 + 1023) >> 10)
155		296
156	/*****************************************************************************/	297	/*****************************************************************************/
157		298
158	#if __GNUC__ >= 3	299	struct tmpbuf
159	# 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
160	#else	325	#else
161	# 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);
162	#endif	328	#endif
163		329
164	#define expect_false(expr) expect ((expr) != 0, 0)	330	struct eio_pwd
165	#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	};
166		338
167	/*****************************************************************************/	339	/*****************************************************************************/
168		340
169	#define ETP_PRI_MIN EIO_PRI_MIN	341	#define ETP_PRI_MIN EIO_PRI_MIN
170	#define ETP_PRI_MAX EIO_PRI_MAX	342	#define ETP_PRI_MAX EIO_PRI_MAX
…		…
176	static int eio_finish (eio_req *req);	348	static int eio_finish (eio_req *req);
177	#define ETP_FINISH(req) eio_finish (req)	349	#define ETP_FINISH(req) eio_finish (req)
178	static void eio_execute (struct etp_worker *self, eio_req *req);	350	static void eio_execute (struct etp_worker *self, eio_req *req);
179	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	351	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
180		352
181	#define ETP_WORKER_CLEAR(req) \
182	if (wrk->dbuf) \
183	{ \
184	free (wrk->dbuf); \
185	wrk->dbuf = 0; \
186	} \
187	\
188	if (wrk->dirp) \
189	{ \
190	closedir (wrk->dirp); \
191	wrk->dirp = 0; \
192	}
193
194	#define ETP_WORKER_COMMON \
195	void *dbuf; \
196	DIR *dirp;
197
198	/*****************************************************************************/	353	/*****************************************************************************/
199		354
200	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	355	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
201		356
202	/* 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
203	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	359	tvdiff (struct timeval *tv1, struct timeval *tv2)
204	{	360	{
205	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	361	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
206	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	362	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
207	}	363	}
208		364
…		…
212	static void (*done_poll_cb) (void);	368	static void (*done_poll_cb) (void);
213		369
214	static unsigned int max_poll_time; /* reslock */	370	static unsigned int max_poll_time; /* reslock */
215	static unsigned int max_poll_reqs; /* reslock */	371	static unsigned int max_poll_reqs; /* reslock */
216		372
217	static volatile unsigned int nreqs; /* reqlock */	373	static unsigned int nreqs; /* reqlock */
218	static volatile unsigned int nready; /* reqlock */	374	static unsigned int nready; /* reqlock */
219	static volatile unsigned int npending; /* reqlock */	375	static unsigned int npending; /* reqlock */
220	static volatile unsigned int max_idle = 4;	376	static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		377	static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
221		378
222	static mutex_t wrklock = X_MUTEX_INIT;	379	static xmutex_t wrklock;
223	static mutex_t reslock = X_MUTEX_INIT;	380	static xmutex_t reslock;
224	static mutex_t reqlock = X_MUTEX_INIT;	381	static xmutex_t reqlock;
225	static cond_t reqwait = X_COND_INIT;	382	static xcond_t reqwait;
226		383
227	#if !HAVE_PREADWRITE	384	#if !HAVE_PREADWRITE
228	/*	385	/*
229	* make our pread/pwrite emulation safe against themselves, but not against	386	* make our pread/pwrite emulation safe against themselves, but not against
230	* 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,
231	* but that's your problem, not mine.	388	* but that's your problem, not mine.
232	*/	389	*/
233	static mutex_t preadwritelock = X_MUTEX_INIT;	390	static xmutex_t preadwritelock;
234	#endif	391	#endif
235		392
236	typedef struct etp_worker	393	typedef struct etp_worker
237	{	394	{
		395	struct tmpbuf tmpbuf;
		396
238	/* locked by wrklock */	397	/* locked by wrklock */
239	struct etp_worker *prev, *next;	398	struct etp_worker *prev, *next;
240		399
241	thread_t tid;	400	xthread_t tid;
242		401
243	/* locked by reslock, reqlock or wrklock */	402	#ifdef ETP_WORKER_COMMON
244	ETP_REQ *req; /* currently processed request */
245
246	ETP_WORKER_COMMON	403	ETP_WORKER_COMMON
		404	#endif
247	} etp_worker;	405	} etp_worker;
248		406
249	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	407	static etp_worker wrk_first; /* NOT etp */
250		408
251	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	409	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
252	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	410	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
253		411
254	/* worker threads management */	412	/* worker threads management */
255		413
		414	static void
256	static void etp_worker_clear (etp_worker *wrk)	415	etp_worker_clear (etp_worker *wrk)
257	{	416	{
258	ETP_WORKER_CLEAR (wrk);
259	}	417	}
260		418
		419	static void ecb_cold
261	static void etp_worker_free (etp_worker *wrk)	420	etp_worker_free (etp_worker *wrk)
262	{	421	{
		422	free (wrk->tmpbuf.ptr);
		423
263	wrk->next->prev = wrk->prev;	424	wrk->next->prev = wrk->prev;
264	wrk->prev->next = wrk->next;	425	wrk->prev->next = wrk->next;
265		426
266	free (wrk);	427	free (wrk);
267	}	428	}
268		429
269	static unsigned int etp_nreqs (void)	430	static unsigned int
		431	etp_nreqs (void)
270	{	432	{
271	int retval;	433	int retval;
272	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	434	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
273	retval = nreqs;	435	retval = nreqs;
274	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	436	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
275	return retval;	437	return retval;
276	}	438	}
277		439
278	static unsigned int etp_nready (void)	440	static unsigned int
		441	etp_nready (void)
279	{	442	{
280	unsigned int retval;	443	unsigned int retval;
281		444
282	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	445	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
283	retval = nready;	446	retval = nready;
284	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	447	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
285		448
286	return retval;	449	return retval;
287	}	450	}
288		451
289	static unsigned int etp_npending (void)	452	static unsigned int
		453	etp_npending (void)
290	{	454	{
291	unsigned int retval;	455	unsigned int retval;
292		456
293	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	457	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
294	retval = npending;	458	retval = npending;
295	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	459	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
296		460
297	return retval;	461	return retval;
298	}	462	}
299		463
300	static unsigned int etp_nthreads (void)	464	static unsigned int
		465	etp_nthreads (void)
301	{	466	{
302	unsigned int retval;	467	unsigned int retval;
303		468
304	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	469	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
305	retval = started;	470	retval = started;
…		…
319	} etp_reqq;	484	} etp_reqq;
320		485
321	static etp_reqq req_queue;	486	static etp_reqq req_queue;
322	static etp_reqq res_queue;	487	static etp_reqq res_queue;
323		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
324	static int reqq_push (etp_reqq *q, ETP_REQ *req)	501	reqq_push (etp_reqq *q, ETP_REQ *req)
325	{	502	{
326	int pri = req->pri;	503	int pri = req->pri;
327	req->next = 0;	504	req->next = 0;
328		505
329	if (q->qe[pri])	506	if (q->qe[pri])
…		…
335	q->qe[pri] = q->qs[pri] = req;	512	q->qe[pri] = q->qs[pri] = req;
336		513
337	return q->size++;	514	return q->size++;
338	}	515	}
339		516
		517	static ETP_REQ * ecb_noinline
340	static ETP_REQ *reqq_shift (etp_reqq *q)	518	reqq_shift (etp_reqq *q)
341	{	519	{
342	int pri;	520	int pri;
343		521
344	if (!q->size)	522	if (!q->size)
345	return 0;	523	return 0;
…		…
360	}	538	}
361		539
362	abort ();	540	abort ();
363	}	541	}
364		542
365	static void etp_atfork_prepare (void)	543	static int ecb_cold
		544	etp_init (void (*want_poll)(void), void (*done_poll)(void))
366	{	545	{
367	X_LOCK (wrklock);	546	X_MUTEX_CREATE (wrklock);
368	X_LOCK (reqlock);	547	X_MUTEX_CREATE (reslock);
369	X_LOCK (reslock);	548	X_MUTEX_CREATE (reqlock);
370	#if !HAVE_PREADWRITE	549	X_COND_CREATE (reqwait);
371	X_LOCK (preadwritelock);
372	#endif
373	}
374		550
375	static void etp_atfork_parent (void)	551	reqq_init (&req_queue);
376	{	552	reqq_init (&res_queue);
377	#if !HAVE_PREADWRITE
378	X_UNLOCK (preadwritelock);
379	#endif
380	X_UNLOCK (reslock);
381	X_UNLOCK (reqlock);
382	X_UNLOCK (wrklock);
383	}
384		553
385	static void etp_atfork_child (void)	554	wrk_first.next =
386	{	555	wrk_first.prev = &wrk_first;
387	ETP_REQ *prv;
388
389	while ((prv = reqq_shift (&req_queue)))
390	ETP_DESTROY (prv);
391
392	while ((prv = reqq_shift (&res_queue)))
393	ETP_DESTROY (prv);
394
395	while (wrk_first.next != &wrk_first)
396	{
397	etp_worker *wrk = wrk_first.next;
398
399	if (wrk->req)
400	ETP_DESTROY (wrk->req);
401
402	etp_worker_clear (wrk);
403	etp_worker_free (wrk);
404	}
405		556
406	started = 0;	557	started = 0;
407	idle = 0;	558	idle = 0;
408	nreqs = 0;	559	nreqs = 0;
409	nready = 0;	560	nready = 0;
410	npending = 0;	561	npending = 0;
411		562
412	etp_atfork_parent ();
413	}
414
415	static void
416	etp_once_init (void)
417	{
418	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
419	}
420
421	static int
422	etp_init (void (*want_poll)(void), void (*done_poll)(void))
423	{
424	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
425
426	pthread_once (&doinit, etp_once_init);
427
428	want_poll_cb = want_poll;	563	want_poll_cb = want_poll;
429	done_poll_cb = done_poll;	564	done_poll_cb = done_poll;
430		565
431	return 0;	566	return 0;
432	}	567	}
433		568
434	X_THREAD_PROC (etp_proc);	569	X_THREAD_PROC (etp_proc);
435		570
		571	static void ecb_cold
436	static void etp_start_thread (void)	572	etp_start_thread (void)
437	{	573	{
438	etp_worker *wrk = calloc (1, sizeof (etp_worker));	574	etp_worker *wrk = calloc (1, sizeof (etp_worker));
439		575
440	/*TODO*/	576	/*TODO*/
441	assert (("unable to allocate worker thread data", wrk));	577	assert (("unable to allocate worker thread data", wrk));
…		…
454	free (wrk);	590	free (wrk);
455		591
456	X_UNLOCK (wrklock);	592	X_UNLOCK (wrklock);
457	}	593	}
458		594
		595	static void
459	static void etp_maybe_start_thread (void)	596	etp_maybe_start_thread (void)
460	{	597	{
461	if (expect_true (etp_nthreads () >= wanted))	598	if (ecb_expect_true (etp_nthreads () >= wanted))
462	return;	599	return;
463		600
464	/* todo: maybe use idle here, but might be less exact */	601	/* todo: maybe use idle here, but might be less exact */
465	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 ()))
466	return;	603	return;
467		604
468	etp_start_thread ();	605	etp_start_thread ();
469	}	606	}
470		607
		608	static void ecb_cold
471	static void etp_end_thread (void)	609	etp_end_thread (void)
472	{	610	{
473	eio_req *req = calloc (1, sizeof (eio_req));	611	eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
474		612
475	req->type = -1;	613	req->type = -1;
476	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	614	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
477		615
478	X_LOCK (reqlock);	616	X_LOCK (reqlock);
…		…
483	X_LOCK (wrklock);	621	X_LOCK (wrklock);
484	--started;	622	--started;
485	X_UNLOCK (wrklock);	623	X_UNLOCK (wrklock);
486	}	624	}
487		625
488	static int etp_poll (void)	626	static int
		627	etp_poll (void)
489	{	628	{
490	unsigned int maxreqs;	629	unsigned int maxreqs;
491	unsigned int maxtime;	630	unsigned int maxtime;
492	struct timeval tv_start, tv_now;	631	struct timeval tv_start, tv_now;
493		632
…		…
523		662
524	X_LOCK (reqlock);	663	X_LOCK (reqlock);
525	--nreqs;	664	--nreqs;
526	X_UNLOCK (reqlock);	665	X_UNLOCK (reqlock);
527		666
528	if (expect_false (req->type == EIO_GROUP && req->size))	667	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
529	{	668	{
530	req->int1 = 1; /* mark request as delayed */	669	req->int1 = 1; /* mark request as delayed */
531	continue;	670	continue;
532	}	671	}
533	else	672	else
534	{	673	{
535	int res = ETP_FINISH (req);	674	int res = ETP_FINISH (req);
536	if (expect_false (res))	675	if (ecb_expect_false (res))
537	return res;	676	return res;
538	}	677	}
539		678
540	if (expect_false (maxreqs && !--maxreqs))	679	if (ecb_expect_false (maxreqs && !--maxreqs))
541	break;	680	break;
542		681
543	if (maxtime)	682	if (maxtime)
544	{	683	{
545	gettimeofday (&tv_now, 0);	684	gettimeofday (&tv_now, 0);
…		…
551		690
552	errno = EAGAIN;	691	errno = EAGAIN;
553	return -1;	692	return -1;
554	}	693	}
555		694
		695	static void
556	static void etp_cancel (ETP_REQ *req)	696	etp_cancel (ETP_REQ *req)
557	{	697	{
558	X_LOCK (wrklock);	698	req->cancelled = 1;
559	req->flags |= EIO_FLAG_CANCELLED;
560	X_UNLOCK (wrklock);
561		699
562	eio_grp_cancel (req);	700	eio_grp_cancel (req);
563	}	701	}
564		702
		703	static void
565	static void etp_submit (ETP_REQ *req)	704	etp_submit (ETP_REQ *req)
566	{	705	{
567	req->pri -= ETP_PRI_MIN;	706	req->pri -= ETP_PRI_MIN;
568		707
569	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;
570	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;
571		710
572	if (expect_false (req->type == EIO_GROUP))	711	if (ecb_expect_false (req->type == EIO_GROUP))
573	{	712	{
574	/* I hope this is worth it :/ */	713	/* I hope this is worth it :/ */
575	X_LOCK (reqlock);	714	X_LOCK (reqlock);
576	++nreqs;	715	++nreqs;
577	X_UNLOCK (reqlock);	716	X_UNLOCK (reqlock);
…		…
596		735
597	etp_maybe_start_thread ();	736	etp_maybe_start_thread ();
598	}	737	}
599	}	738	}
600		739
		740	static void ecb_cold
601	static void etp_set_max_poll_time (double nseconds)	741	etp_set_max_poll_time (double nseconds)
602	{	742	{
603	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	743	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
604	max_poll_time = nseconds * EIO_TICKS;	744	max_poll_time = nseconds * EIO_TICKS;
605	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	745	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
606	}	746	}
607		747
		748	static void ecb_cold
608	static void etp_set_max_poll_reqs (unsigned int maxreqs)	749	etp_set_max_poll_reqs (unsigned int maxreqs)
609	{	750	{
610	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	751	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
611	max_poll_reqs = maxreqs;	752	max_poll_reqs = maxreqs;
612	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	753	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
613	}	754	}
614		755
		756	static void ecb_cold
615	static void etp_set_max_idle (unsigned int nthreads)	757	etp_set_max_idle (unsigned int nthreads)
616	{	758	{
617	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	759	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
618	max_idle = nthreads <= 0 ? 1 : nthreads;	760	max_idle = nthreads;
619	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	761	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
620	}	762	}
621		763
		764	static void ecb_cold
		765	etp_set_idle_timeout (unsigned int seconds)
		766	{
		767	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		768	idle_timeout = seconds;
		769	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		770	}
		771
		772	static void ecb_cold
622	static void etp_set_min_parallel (unsigned int nthreads)	773	etp_set_min_parallel (unsigned int nthreads)
623	{	774	{
624	if (wanted < nthreads)	775	if (wanted < nthreads)
625	wanted = nthreads;	776	wanted = nthreads;
626	}	777	}
627		778
		779	static void ecb_cold
628	static void etp_set_max_parallel (unsigned int nthreads)	780	etp_set_max_parallel (unsigned int nthreads)
629	{	781	{
630	if (wanted > nthreads)	782	if (wanted > nthreads)
631	wanted = nthreads;	783	wanted = nthreads;
632		784
633	while (started > wanted)	785	while (started > wanted)
634	etp_end_thread ();	786	etp_end_thread ();
635	}	787	}
636		788
637	/*****************************************************************************/	789	/*****************************************************************************/
638		790
		791	static void
639	static void grp_try_feed (eio_req *grp)	792	grp_try_feed (eio_req *grp)
640	{	793	{
641	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	794	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
642	{	795	{
643	grp->flags &= ~EIO_FLAG_GROUPADD;	796	grp->flags &= ~EIO_FLAG_GROUPADD;
644		797
…		…
651	break;	804	break;
652	}	805	}
653	}	806	}
654	}	807	}
655		808
		809	static int
656	static int grp_dec (eio_req *grp)	810	grp_dec (eio_req *grp)
657	{	811	{
658	--grp->size;	812	--grp->size;
659		813
660	/* call feeder, if applicable */	814	/* call feeder, if applicable */
661	grp_try_feed (grp);	815	grp_try_feed (grp);
…		…
665	return eio_finish (grp);	819	return eio_finish (grp);
666	else	820	else
667	return 0;	821	return 0;
668	}	822	}
669		823
		824	static void
670	void eio_destroy (eio_req *req)	825	eio_destroy (eio_req *req)
671	{	826	{
672	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	827	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
673	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	828	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
674		829
675	EIO_DESTROY (req);	830	EIO_DESTROY (req);
676	}	831	}
677		832
		833	static int
678	static int eio_finish (eio_req *req)	834	eio_finish (eio_req *req)
679	{	835	{
680	int res = EIO_FINISH (req);	836	int res = EIO_FINISH (req);
681		837
682	if (req->grp)	838	if (req->grp)
683	{	839	{
…		…
691	if (grp->grp_first == req)	847	if (grp->grp_first == req)
692	grp->grp_first = req->grp_next;	848	grp->grp_first = req->grp_next;
693		849
694	res2 = grp_dec (grp);	850	res2 = grp_dec (grp);
695		851
696	if (!res && res2)	852	if (!res)
697	res = res2;	853	res = res2;
698	}	854	}
699		855
700	eio_destroy (req);	856	eio_destroy (req);
701		857
702	return res;	858	return res;
703	}	859	}
704		860
		861	void
705	void eio_grp_cancel (eio_req *grp)	862	eio_grp_cancel (eio_req *grp)
706	{	863	{
707	for (grp = grp->grp_first; grp; grp = grp->grp_next)	864	for (grp = grp->grp_first; grp; grp = grp->grp_next)
708	eio_cancel (grp);	865	eio_cancel (grp);
709	}	866	}
710		867
		868	void
711	void eio_cancel (eio_req *req)	869	eio_cancel (eio_req *req)
712	{	870	{
713	etp_cancel (req);	871	etp_cancel (req);
714	}	872	}
715		873
		874	void
716	void eio_submit (eio_req *req)	875	eio_submit (eio_req *req)
717	{	876	{
718	etp_submit (req);	877	etp_submit (req);
719	}	878	}
720		879
721	unsigned int eio_nreqs (void)	880	unsigned int
		881	eio_nreqs (void)
722	{	882	{
723	return etp_nreqs ();	883	return etp_nreqs ();
724	}	884	}
725		885
726	unsigned int eio_nready (void)	886	unsigned int
		887	eio_nready (void)
727	{	888	{
728	return etp_nready ();	889	return etp_nready ();
729	}	890	}
730		891
731	unsigned int eio_npending (void)	892	unsigned int
		893	eio_npending (void)
732	{	894	{
733	return etp_npending ();	895	return etp_npending ();
734	}	896	}
735		897
736	unsigned int eio_nthreads (void)	898	unsigned int ecb_cold
		899	eio_nthreads (void)
737	{	900	{
738	return etp_nthreads ();	901	return etp_nthreads ();
739	}	902	}
740		903
		904	void ecb_cold
741	void eio_set_max_poll_time (double nseconds)	905	eio_set_max_poll_time (double nseconds)
742	{	906	{
743	etp_set_max_poll_time (nseconds);	907	etp_set_max_poll_time (nseconds);
744	}	908	}
745		909
		910	void ecb_cold
746	void eio_set_max_poll_reqs (unsigned int maxreqs)	911	eio_set_max_poll_reqs (unsigned int maxreqs)
747	{	912	{
748	etp_set_max_poll_reqs (maxreqs);	913	etp_set_max_poll_reqs (maxreqs);
749	}	914	}
750		915
		916	void ecb_cold
751	void eio_set_max_idle (unsigned int nthreads)	917	eio_set_max_idle (unsigned int nthreads)
752	{	918	{
753	etp_set_max_idle (nthreads);	919	etp_set_max_idle (nthreads);
754	}	920	}
755		921
		922	void ecb_cold
		923	eio_set_idle_timeout (unsigned int seconds)
		924	{
		925	etp_set_idle_timeout (seconds);
		926	}
		927
		928	void ecb_cold
756	void eio_set_min_parallel (unsigned int nthreads)	929	eio_set_min_parallel (unsigned int nthreads)
757	{	930	{
758	etp_set_min_parallel (nthreads);	931	etp_set_min_parallel (nthreads);
759	}	932	}
760		933
		934	void ecb_cold
761	void eio_set_max_parallel (unsigned int nthreads)	935	eio_set_max_parallel (unsigned int nthreads)
762	{	936	{
763	etp_set_max_parallel (nthreads);	937	etp_set_max_parallel (nthreads);
764	}	938	}
765		939
766	int eio_poll (void)	940	int eio_poll (void)
…		…
775	# undef pread	949	# undef pread
776	# undef pwrite	950	# undef pwrite
777	# define pread eio__pread	951	# define pread eio__pread
778	# define pwrite eio__pwrite	952	# define pwrite eio__pwrite
779		953
780	static ssize_t	954	static eio_ssize_t
781	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)
782	{	956	{
783	ssize_t res;	957	eio_ssize_t res;
784	off_t ooffset;	958	off_t ooffset;
785		959
786	X_LOCK (preadwritelock);	960	X_LOCK (preadwritelock);
787	ooffset = lseek (fd, 0, SEEK_CUR);	961	ooffset = lseek (fd, 0, SEEK_CUR);
788	lseek (fd, offset, SEEK_SET);	962	lseek (fd, offset, SEEK_SET);
…		…
791	X_UNLOCK (preadwritelock);	965	X_UNLOCK (preadwritelock);
792		966
793	return res;	967	return res;
794	}	968	}
795		969
796	static ssize_t	970	static eio_ssize_t
797	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)
798	{	972	{
799	ssize_t res;	973	eio_ssize_t res;
800	off_t ooffset;	974	off_t ooffset;
801		975
802	X_LOCK (preadwritelock);	976	X_LOCK (preadwritelock);
803	ooffset = lseek (fd, 0, SEEK_CUR);	977	ooffset = lseek (fd, 0, SEEK_CUR);
804	lseek (fd, offset, SEEK_SET);	978	lseek (fd, offset, SEEK_SET);
…		…
808		982
809	return res;	983	return res;
810	}	984	}
811	#endif	985	#endif
812		986
813	#ifndef HAVE_FUTIMES	987	#ifndef HAVE_UTIMES
814		988
815	# undef utimes	989	# undef utimes
816	# undef futimes
817	# define utimes(path,times) eio__utimes (path, times)	990	# define utimes(path,times) eio__utimes (path, times)
818	# define futimes(fd,times) eio__futimes (fd, times)
819		991
820	static int	992	static int
821	eio__utimes (const char *filename, const struct timeval times[2])	993	eio__utimes (const char *filename, const struct timeval times[2])
822	{	994	{
823	if (times)	995	if (times)
…		…
831	}	1003	}
832	else	1004	else
833	return utime (filename, 0);	1005	return utime (filename, 0);
834	}	1006	}
835		1007
		1008	#endif
		1009
		1010	#ifndef HAVE_FUTIMES
		1011
		1012	# undef futimes
		1013	# define futimes(fd,times) eio__futimes (fd, times)
		1014
		1015	static int
836	static int eio__futimes (int fd, const struct timeval tv[2])	1016	eio__futimes (int fd, const struct timeval tv[2])
837	{	1017	{
838	errno = ENOSYS;	1018	errno = ENOSYS;
839	return -1;	1019	return -1;
840	}	1020	}
841		1021
…		…
844	#if !HAVE_FDATASYNC	1024	#if !HAVE_FDATASYNC
845	# undef fdatasync	1025	# undef fdatasync
846	# define fdatasync(fd) fsync (fd)	1026	# define fdatasync(fd) fsync (fd)
847	#endif	1027	#endif
848		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 = -1;
		1038	errno = ENOSYS;
		1039	#endif
		1040
		1041	if (res < 0 && errno == ENOSYS && fd >= 0)
		1042	sync ();
		1043
		1044	return res;
		1045	}
		1046
849	/* sync_file_range always needs emulation */	1047	/* sync_file_range always needs emulation */
850	int	1048	static int
851	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	1049	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
852	{	1050	{
853	#if HAVE_SYNC_FILE_RANGE	1051	#if HAVE_SYNC_FILE_RANGE
854	int res;	1052	int res;
855		1053
…		…
872	/* even though we could play tricks with the flags, it's better to always	1070	/* even though we could play tricks with the flags, it's better to always
873	* call fdatasync, as that matches the expectation of its users best */	1071	* call fdatasync, as that matches the expectation of its users best */
874	return fdatasync (fd);	1072	return fdatasync (fd);
875	}	1073	}
876		1074
		1075	static int
		1076	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		1077	{
		1078	#if HAVE_FALLOCATE
		1079	return fallocate (fd, mode, offset, len);
		1080	#else
		1081	errno = ENOSYS;
		1082	return -1;
		1083	#endif
		1084	}
		1085
877	#if !HAVE_READAHEAD	1086	#if !HAVE_READAHEAD
878	# undef readahead	1087	# undef readahead
879	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1088	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
880		1089
881	static ssize_t	1090	static eio_ssize_t
882	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1091	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
883	{	1092	{
884	size_t todo = count;	1093	size_t todo = count;
885	dBUF;	1094	dBUF;
886		1095
…		…
891	pread (fd, eio_buf, len, offset);	1100	pread (fd, eio_buf, len, offset);
892	offset += len;	1101	offset += len;
893	todo -= len;	1102	todo -= len;
894	}	1103	}
895		1104
		1105	FUBd;
		1106
896	errno = 0;	1107	errno = 0;
897	return count;	1108	return count;
898	}	1109	}
899		1110
900	#endif	1111	#endif
901		1112
902	/* sendfile always needs emulation */	1113	/* sendfile always needs emulation */
903	static ssize_t	1114	static eio_ssize_t
904	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1115	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
905	{	1116	{
		1117	eio_ssize_t written = 0;
906	ssize_t res;	1118	eio_ssize_t res;
907		1119
908	if (!count)	1120	if (!count)
909	return 0;	1121	return 0;
910		1122
		1123	for (;;)
		1124	{
		1125	#ifdef __APPLE__
		1126	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1127	#endif
911	#if HAVE_SENDFILE	1128	#if HAVE_SENDFILE
912	# if __linux	1129	# if __linux
		1130	off_t soffset = offset;
913	res = sendfile (ofd, ifd, &offset, count);	1131	res = sendfile (ofd, ifd, &soffset, count);
914		1132
915	# elif __FreeBSD__	1133	# elif __FreeBSD__
916	/*	1134	/*
917	* Of course, the freebsd sendfile is a dire hack with no thoughts	1135	* Of course, the freebsd sendfile is a dire hack with no thoughts
918	* wasted on making it similar to other I/O functions.	1136	* wasted on making it similar to other I/O functions.
919	*/	1137	*/
920	{
921	off_t sbytes;	1138	off_t sbytes;
922	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1139	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
923		1140
924	#if 0 /* according to the manpage, this is correct, but broken behaviour */	1141	#if 0 /* according to the manpage, this is correct, but broken behaviour */
925	/* freebsd' sendfile will return 0 on success */	1142	/* freebsd' sendfile will return 0 on success */
926	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */	1143	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
927	/* not on e.g. EIO or EPIPE - sounds broken */	1144	/* not on e.g. EIO or EPIPE - sounds broken */
928	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)	1145	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
929	res = sbytes;	1146	res = sbytes;
930	#endif	1147	#endif
931		1148
932	/* according to source inspection, this is correct, and useful behaviour */	1149	/* according to source inspection, this is correct, and useful behaviour */
933	if (sbytes)	1150	if (sbytes)
934	res = sbytes;	1151	res = sbytes;
935	}
936		1152
937	# elif defined (__APPLE__)	1153	# elif defined (__APPLE__)
938
939	{
940	off_t sbytes = count;	1154	off_t sbytes = count;
941	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	1155	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
942		1156
943	/* according to the manpage, sbytes is always valid */	1157	/* according to the manpage, sbytes is always valid */
944	if (sbytes)	1158	if (sbytes)
945	res = sbytes;	1159	res = sbytes;
946	}
947		1160
948	# elif __hpux	1161	# elif __hpux
949	res = sendfile (ofd, ifd, offset, count, 0, 0);	1162	res = sendfile (ofd, ifd, offset, count, 0, 0);
950		1163
951	# elif __solaris	1164	# elif __solaris
952	{
953	struct sendfilevec vec;	1165	struct sendfilevec vec;
954	size_t sbytes;	1166	size_t sbytes;
955		1167
956	vec.sfv_fd = ifd;	1168	vec.sfv_fd = ifd;
957	vec.sfv_flag = 0;	1169	vec.sfv_flag = 0;
958	vec.sfv_off = offset;	1170	vec.sfv_off = offset;
959	vec.sfv_len = count;	1171	vec.sfv_len = count;
960		1172
961	res = sendfilev (ofd, &vec, 1, &sbytes);	1173	res = sendfilev (ofd, &vec, 1, &sbytes);
962		1174
963	if (res < 0 && sbytes)	1175	if (res < 0 && sbytes)
964	res = sbytes;	1176	res = sbytes;
965	}
966		1177
967	# endif	1178	# endif
968		1179
969	#elif defined (_WIN32)	1180	#elif defined (_WIN32) && 0
970
971	/* does not work, just for documentation of what would need to be done */	1181	/* does not work, just for documentation of what would need to be done */
972	{	1182	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1183	/* libeio guarantees that the file offset does not change, and windows */
		1184	/* has no way to get an independent handle to the same file description */
973	HANDLE h = TO_SOCKET (ifd);	1185	HANDLE h = TO_SOCKET (ifd);
974	SetFilePointer (h, offset, 0, FILE_BEGIN);	1186	SetFilePointer (h, offset, 0, FILE_BEGIN);
975	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);	1187	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
976	}
977		1188
978	#else	1189	#else
979	res = -1;	1190	res = -1;
980	errno = ENOSYS;	1191	errno = ENOSYS;
981	#endif	1192	#endif
982		1193
		1194	/* we assume sendfile can copy at least 128mb in one go */
		1195	if (res <= 128 * 1024 * 1024)
		1196	{
		1197	if (res > 0)
		1198	written += res;
		1199
		1200	if (written)
		1201	return written;
		1202
		1203	break;
		1204	}
		1205	else
		1206	{
		1207	/* if we requested more, then probably the kernel was lazy */
		1208	written += res;
		1209	offset += res;
		1210	count -= res;
		1211
		1212	if (!count)
		1213	return written;
		1214	}
		1215	}
		1216
983	if (res < 0	1217	if (res < 0
984	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1218	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
985	/* BSDs */	1219	/* BSDs */
986	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1220	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
987	|| errno == ENOTSUP	1221	|| errno == ENOTSUP
988	#endif	1222	#endif
		1223	#ifdef EOPNOTSUPP /* windows */
989	|| errno == EOPNOTSUPP /* BSDs */	1224	|| errno == EOPNOTSUPP /* BSDs */
		1225	#endif
990	#if __solaris	1226	#if __solaris
991	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1227	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
992	#endif	1228	#endif
993	)	1229	)
994	)	1230	)
…		…
998		1234
999	res = 0;	1235	res = 0;
1000		1236
1001	while (count)	1237	while (count)
1002	{	1238	{
1003	ssize_t cnt;	1239	eio_ssize_t cnt;
1004		1240
1005	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1241	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1006		1242
1007	if (cnt <= 0)	1243	if (cnt <= 0)
1008	{	1244	{
…		…
1020		1256
1021	offset += cnt;	1257	offset += cnt;
1022	res += cnt;	1258	res += cnt;
1023	count -= cnt;	1259	count -= cnt;
1024	}	1260	}
		1261
		1262	FUBd;
1025	}	1263	}
1026		1264
1027	return res;	1265	return res;
		1266	}
		1267
		1268	#ifdef PAGESIZE
		1269	# define eio_pagesize() PAGESIZE
		1270	#else
		1271	static intptr_t
		1272	eio_pagesize (void)
		1273	{
		1274	static intptr_t page;
		1275
		1276	if (!page)
		1277	page = sysconf (_SC_PAGESIZE);
		1278
		1279	return page;
		1280	}
		1281	#endif
		1282
		1283	static void
		1284	eio_page_align (void **addr, size_t *length)
		1285	{
		1286	intptr_t mask = eio_pagesize () - 1;
		1287
		1288	/* round down addr */
		1289	intptr_t adj = mask & (intptr_t)*addr;
		1290
		1291	*addr = (void *)((intptr_t)*addr - adj);
		1292	*length += adj;
		1293
		1294	/* round up length */
		1295	*length = (*length + mask) & ~mask;
		1296	}
		1297
		1298	#if !_POSIX_MEMLOCK
		1299	# define eio__mlockall(a) EIO_ENOSYS ()
		1300	#else
		1301
		1302	static int
		1303	eio__mlockall (int flags)
		1304	{
		1305	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1306	extern int mallopt (int, int);
		1307	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1308	#endif
		1309
		1310	if (EIO_MCL_CURRENT != MCL_CURRENT
		1311	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1312	{
		1313	flags = 0
		1314	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1315	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1316	}
		1317
		1318	return mlockall (flags);
		1319	}
		1320	#endif
		1321
		1322	#if !_POSIX_MEMLOCK_RANGE
		1323	# define eio__mlock(a,b) EIO_ENOSYS ()
		1324	#else
		1325
		1326	static int
		1327	eio__mlock (void *addr, size_t length)
		1328	{
		1329	eio_page_align (&addr, &length);
		1330
		1331	return mlock (addr, length);
		1332	}
		1333
		1334	#endif
		1335
		1336	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1337	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1338	#else
		1339
		1340	static int
		1341	eio__msync (void *mem, size_t len, int flags)
		1342	{
		1343	eio_page_align (&mem, &len);
		1344
		1345	if (EIO_MS_ASYNC != MS_SYNC
		1346	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1347	|| EIO_MS_SYNC != MS_SYNC)
		1348	{
		1349	flags = 0
		1350	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1351	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1352	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1353	}
		1354
		1355	return msync (mem, len, flags);
		1356	}
		1357
		1358	#endif
		1359
		1360	static int
		1361	eio__mtouch (eio_req *req)
		1362	{
		1363	void *mem = req->ptr2;
		1364	size_t len = req->size;
		1365	int flags = req->int1;
		1366
		1367	eio_page_align (&mem, &len);
		1368
		1369	{
		1370	intptr_t addr = (intptr_t)mem;
		1371	intptr_t end = addr + len;
		1372	intptr_t page = eio_pagesize ();
		1373
		1374	if (addr < end)
		1375	if (flags & EIO_MT_MODIFY) /* modify */
		1376	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1377	else
		1378	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1379	}
		1380
		1381	return 0;
		1382	}
		1383
		1384	/*****************************************************************************/
		1385	/* requests implemented outside eio_execute, because they are so large */
		1386
		1387	/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
		1388	static int
		1389	eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1390	{
		1391	const char *rel = path;
		1392	char *res;
		1393	char *tmp1, *tmp2;
		1394	#if SYMLOOP_MAX > 32
		1395	int symlinks = SYMLOOP_MAX;
		1396	#else
		1397	int symlinks = 32;
		1398	#endif
		1399
		1400	errno = EINVAL;
		1401	if (!rel)
		1402	return -1;
		1403
		1404	errno = ENOENT;
		1405	if (!*rel)
		1406	return -1;
		1407
		1408	res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
		1409	tmp1 = res + PATH_MAX;
		1410	tmp2 = tmp1 + PATH_MAX;
		1411
		1412	#if 0 /* disabled, the musl way to do things is just too racy */
		1413	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1414	/* on linux we may be able to ask the kernel */
		1415	{
		1416	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1417
		1418	if (fd >= 0)
		1419	{
		1420	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1421	req->result = readlink (tmp1, res, PATH_MAX);
		1422	close (fd);
		1423
		1424	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1425
		1426	if (req->result > 0)
		1427	goto done;
		1428	}
		1429	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1430	return;
		1431	}
		1432	#endif
		1433	#endif
		1434
		1435	if (*rel != '/')
		1436	{
		1437	int len;
		1438
		1439	errno = ENOENT;
		1440	if (wd == EIO_INVALID_WD)
		1441	return -1;
		1442
		1443	if (wd == EIO_CWD)
		1444	{
		1445	if (!getcwd (res, PATH_MAX))
		1446	return -1;
		1447
		1448	len = strlen (res);
		1449	}
		1450	else
		1451	memcpy (res, wd->str, len = wd->len);
		1452
		1453	if (res [1]) /* only use if not / */
		1454	res += len;
		1455	}
		1456
		1457	while (*rel)
		1458	{
		1459	eio_ssize_t len, linklen;
		1460	const char *beg = rel;
		1461
		1462	while (*rel && *rel != '/')
		1463	++rel;
		1464
		1465	len = rel - beg;
		1466
		1467	if (!len) /* skip slashes */
		1468	{
		1469	++rel;
		1470	continue;
		1471	}
		1472
		1473	if (beg [0] == '.')
		1474	{
		1475	if (len == 1)
		1476	continue; /* . - nop */
		1477
		1478	if (beg [1] == '.' && len == 2)
		1479	{
		1480	/* .. - back up one component, if possible */
		1481
		1482	while (res != tmpbuf->ptr)
		1483	if (*--res == '/')
		1484	break;
		1485
		1486	continue;
		1487	}
		1488	}
		1489
		1490	errno = ENAMETOOLONG;
		1491	if (res + 1 + len + 1 >= tmp1)
		1492	return;
		1493
		1494	/* copy one component */
		1495	*res = '/';
		1496	memcpy (res + 1, beg, len);
		1497
		1498	/* zero-terminate, for readlink */
		1499	res [len + 1] = 0;
		1500
		1501	/* now check if it's a symlink */
		1502	linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
		1503
		1504	if (linklen < 0)
		1505	{
		1506	if (errno != EINVAL)
		1507	return -1;
		1508
		1509	/* it's a normal directory. hopefully */
		1510	res += len + 1;
		1511	}
		1512	else
		1513	{
		1514	/* yay, it was a symlink - build new path in tmp2 */
		1515	int rellen = strlen (rel);
		1516
		1517	errno = ENAMETOOLONG;
		1518	if (linklen + 1 + rellen >= PATH_MAX)
		1519	return -1;
		1520
		1521	errno = ELOOP;
		1522	if (!--symlinks)
		1523	return -1;
		1524
		1525	if (*tmp1 == '/')
		1526	res = tmpbuf->ptr; /* symlink resolves to an absolute path */
		1527
		1528	/* we need to be careful, as rel might point into tmp2 already */
		1529	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1530	tmp2 [linklen] = '/';
		1531	memcpy (tmp2, tmp1, linklen);
		1532
		1533	rel = tmp2;
		1534	}
		1535	}
		1536
		1537	/* special case for the lone root path */
		1538	if (res == tmpbuf->ptr)
		1539	*res++ = '/';
		1540
		1541	return res - (char *)tmpbuf->ptr;
1028	}	1542	}
1029		1543
1030	static signed char	1544	static signed char
1031	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1545	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1032	{	1546	{
1033	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */	1547	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1034	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;	1548	: a->inode < b->inode ? -1
		1549	: a->inode > b->inode ? 1
		1550	: 0;
1035	}	1551	}
1036		1552
1037	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0	1553	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1038		1554
1039	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1555	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1040	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1556	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1041		1557
1042	static void	1558	static void
1043	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1559	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1044	{	1560	{
1045	unsigned char bits [9 + sizeof (ino_t) * 8];	1561	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1046	unsigned char *bit = bits;	1562	unsigned char *bit = bits;
1047		1563
1048	assert (CHAR_BIT == 8);	1564	assert (CHAR_BIT == 8);
1049	assert (sizeof (eio_dirent) * 8 < 256);	1565	assert (sizeof (eio_dirent) * 8 < 256);
1050	assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */	1566	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1051	assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */	1567	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1052		1568
1053	if (size <= EIO_SORT_FAST)	1569	if (size <= EIO_SORT_FAST)
1054	return;	1570	return;
1055		1571
1056	/* first prepare an array of bits to test in our radix sort */	1572	/* first prepare an array of bits to test in our radix sort */
1057	/* try to take endianness into account, as well as differences in ino_t sizes */	1573	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1058	/* inode_bits must contain all inodes ORed together */	1574	/* inode_bits must contain all inodes ORed together */
1059	/* which is used to skip bits that are 0 everywhere, which is very common */	1575	/* which is used to skip bits that are 0 everywhere, which is very common */
1060	{	1576	{
1061	ino_t endianness;	1577	eio_ino_t endianness;
1062	int i, j;	1578	int i, j;
1063		1579
1064	/* we store the byte offset of byte n into byte n of "endianness" */	1580	/* we store the byte offset of byte n into byte n of "endianness" */
1065	for (i = 0; i < sizeof (ino_t); ++i)	1581	for (i = 0; i < sizeof (eio_ino_t); ++i)
1066	((unsigned char *)&endianness)[i] = i;	1582	((unsigned char *)&endianness)[i] = i;
1067		1583
1068	*bit++ = 0;	1584	*bit++ = 0;
1069		1585
1070	for (i = 0; i < sizeof (ino_t); ++i)	1586	for (i = 0; i < sizeof (eio_ino_t); ++i)
1071	{	1587	{
1072	/* shifting off the byte offsets out of "endianness" */	1588	/* shifting off the byte offsets out of "endianness" */
1073	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1589	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1074	endianness >>= 8;	1590	endianness >>= 8;
1075		1591
1076	for (j = 0; j < 8; ++j)	1592	for (j = 0; j < 8; ++j)
1077	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1593	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1078	*bit++ = offs + j;	1594	*bit++ = offs + j;
1079	}	1595	}
1080		1596
1081	for (j = 0; j < 8; ++j)	1597	for (j = 0; j < 8; ++j)
1082	if (score_bits & (1 << j))	1598	if (score_bits & (1 << j))
1083	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1599	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1084	}	1600	}
1085		1601
1086	/* now actually do the sorting (a variant of MSD radix sort) */	1602	/* now actually do the sorting (a variant of MSD radix sort) */
1087	{	1603	{
1088	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1604	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1089	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1605	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1090	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1606	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1091	int stk_idx = 0;	1607	int stk_idx = 0;
1092		1608
1093	base_stk [stk_idx] = dents;	1609	base_stk [stk_idx] = dents;
1094	end_stk [stk_idx] = dents + size;	1610	end_stk [stk_idx] = dents + size;
1095	bit_stk [stk_idx] = bit - 1;	1611	bit_stk [stk_idx] = bit - 1;
…		…
1174	}	1690	}
1175	}	1691	}
1176	}	1692	}
1177		1693
1178	static void	1694	static void
1179	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1695	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1180	{	1696	{
1181	if (size <= 1)	1697	if (size <= 1)
1182	return; /* our insertion sort relies on size > 0 */	1698	return; /* our insertion sort relies on size > 0 */
1183		1699
1184	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1700	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1192		1708
1193	/* read a full directory */	1709	/* read a full directory */
1194	static void	1710	static void
1195	eio__scandir (eio_req *req, etp_worker *self)	1711	eio__scandir (eio_req *req, etp_worker *self)
1196	{	1712	{
1197	DIR *dirp;
1198	EIO_STRUCT_DIRENT *entp;
1199	char *name, *names;	1713	char *name, *names;
1200	int namesalloc = 4096;	1714	int namesalloc = 4096 - sizeof (void *) * 4;
1201	int namesoffs = 0;	1715	int namesoffs = 0;
1202	int flags = req->int1;	1716	int flags = req->int1;
1203	eio_dirent *dents = 0;	1717	eio_dirent *dents = 0;
1204	int dentalloc = 128;	1718	int dentalloc = 128;
1205	int dentoffs = 0;	1719	int dentoffs = 0;
1206	ino_t inode_bits = 0;	1720	eio_ino_t inode_bits = 0;
		1721	#ifdef _WIN32
		1722	HANDLE dirp;
		1723	WIN32_FIND_DATA entp;
		1724	#else
		1725	DIR *dirp;
		1726	EIO_STRUCT_DIRENT *entp;
		1727	#endif
1207		1728
1208	req->result = -1;	1729	req->result = -1;
1209		1730
1210	if (!(flags & EIO_READDIR_DENTS))	1731	if (!(flags & EIO_READDIR_DENTS))
1211	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1732	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1212		1733
1213	X_LOCK (wrklock);	1734	#ifdef _WIN32
1214	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1735	{
		1736	int len = strlen ((const char *)req->ptr1);
		1737	char *path = malloc (MAX_PATH);
		1738	const char *fmt;
		1739	const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		1740
		1741	if (!len)
		1742	fmt = "./*";
		1743	else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
		1744	fmt = "%s*";
		1745	else
		1746	fmt = "%s/*";
		1747
		1748	_snprintf (path, MAX_PATH, fmt, reqpath);
		1749	dirp = FindFirstFile (path, &entp);
		1750	free (path);
		1751
		1752	if (dirp == INVALID_HANDLE_VALUE)
		1753	{
		1754	/* should steal _dosmaperr */
		1755	switch (GetLastError ())
		1756	{
		1757	case ERROR_FILE_NOT_FOUND:
		1758	req->result = 0;
		1759	break;
		1760
		1761	case ERROR_INVALID_NAME:
		1762	case ERROR_PATH_NOT_FOUND:
		1763	case ERROR_NO_MORE_FILES:
		1764	errno = ENOENT;
		1765	break;
		1766
		1767	case ERROR_NOT_ENOUGH_MEMORY:
		1768	errno = ENOMEM;
		1769	break;
		1770
		1771	default:
		1772	errno = EINVAL;
		1773	break;
		1774	}
		1775
		1776	return;
		1777	}
		1778	}
		1779	#else
		1780	#if HAVE_AT
		1781	if (req->wd)
		1782	{
		1783	int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1784
		1785	if (fd < 0)
		1786	return;
		1787
		1788	dirp = fdopendir (fd);
		1789
		1790	if (!dirp)
		1791	close (fd);
		1792	}
		1793	else
1215	self->dirp = dirp = opendir (req->ptr1);	1794	dirp = opendir (req->ptr1);
		1795	#else
		1796	dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
		1797	#endif
		1798
		1799	if (!dirp)
		1800	return;
		1801	#endif
		1802
		1803	if (req->flags & EIO_FLAG_PTR1_FREE)
		1804	free (req->ptr1);
		1805
1216	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1806	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1217	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1807	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1218	req->ptr2 = names = malloc (namesalloc);	1808	req->ptr2 = names = malloc (namesalloc);
1219	X_UNLOCK (wrklock);
1220		1809
1221	if (dirp && names && (!flags || dents))	1810	if (!names || (flags && !dents))
		1811	return;
		1812
1222	for (;;)	1813	for (;;)
1223	{	1814	{
		1815	int done;
		1816
		1817	#ifdef _WIN32
		1818	done = !dirp;
		1819	#else
1224	errno = 0;	1820	errno = 0;
1225	entp = readdir (dirp);	1821	entp = readdir (dirp);
		1822	done = !entp;
		1823	#endif
1226		1824
1227	if (!entp)	1825	if (done)
1228	{	1826	{
		1827	#ifndef _WIN32
		1828	int old_errno = errno;
		1829	closedir (dirp);
		1830	errno = old_errno;
		1831
1229	if (errno)	1832	if (errno)
1230	break;	1833	break;
		1834	#endif
1231		1835
1232	/* sort etc. */	1836	/* sort etc. */
1233	req->int1 = flags;	1837	req->int1 = flags;
1234	req->result = dentoffs;	1838	req->result = dentoffs;
1235		1839
1236	if (flags & EIO_READDIR_STAT_ORDER)	1840	if (flags & EIO_READDIR_STAT_ORDER)
1237	eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */	1841	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1238	else if (flags & EIO_READDIR_DIRS_FIRST)	1842	else if (flags & EIO_READDIR_DIRS_FIRST)
1239	if (flags & EIO_READDIR_FOUND_UNKNOWN)	1843	if (flags & EIO_READDIR_FOUND_UNKNOWN)
1240	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */	1844	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1241	else	1845	else
		1846	{
		1847	/* in this case, all is known, and we just put dirs first and sort them */
		1848	eio_dirent *oth = dents + dentoffs;
		1849	eio_dirent *dir = dents;
		1850
		1851	/* now partition dirs to the front, and non-dirs to the back */
		1852	/* by walking from both sides and swapping if necessary */
		1853	while (oth > dir)
		1854	{
		1855	if (dir->type == EIO_DT_DIR)
		1856	++dir;
		1857	else if ((--oth)->type == EIO_DT_DIR)
		1858	{
		1859	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1860
		1861	++dir;
		1862	}
		1863	}
		1864
		1865	/* now sort the dirs only (dirs all have the same score) */
		1866	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1867	}
		1868
		1869	break;
		1870	}
		1871
		1872	/* now add the entry to our list(s) */
		1873	name = D_NAME (entp);
		1874
		1875	/* skip . and .. entries */
		1876	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
		1877	{
		1878	int len = D_NAMLEN (entp) + 1;
		1879
		1880	while (ecb_expect_false (namesoffs + len > namesalloc))
		1881	{
		1882	namesalloc *= 2;
		1883	req->ptr2 = names = realloc (names, namesalloc);
		1884
		1885	if (!names)
		1886	break;
		1887	}
		1888
		1889	memcpy (names + namesoffs, name, len);
		1890
		1891	if (dents)
		1892	{
		1893	struct eio_dirent *ent;
		1894
		1895	if (ecb_expect_false (dentoffs == dentalloc))
1242	{	1896	{
1243	/* in this case, all is known, and we just put dirs first and sort them */	1897	dentalloc *= 2;
		1898	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1899
		1900	if (!dents)
		1901	break;
		1902	}
		1903
1244	eio_dirent *oth = dents + dentoffs;	1904	ent = dents + dentoffs;
1245	eio_dirent *dir = dents;
1246		1905
1247	/* now partition dirs to the front, and non-dirs to the back */	1906	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1248	/* by walking from both sides and swapping if necessary */	1907	ent->namelen = len - 1;
1249	/* also clear score, so it doesn't influence sorting */	1908	ent->inode = D_INO (entp);
1250	while (oth > dir)	1909
		1910	inode_bits |= ent->inode;
		1911
		1912	switch (D_TYPE (entp))
		1913	{
		1914	default:
		1915	ent->type = EIO_DT_UNKNOWN;
		1916	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1917	break;
		1918
		1919	#ifdef DT_FIFO
		1920	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1921	#endif
		1922	#ifdef DT_CHR
		1923	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1924	#endif
		1925	#ifdef DT_MPC
		1926	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1927	#endif
		1928	#ifdef DT_DIR
		1929	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1930	#endif
		1931	#ifdef DT_NAM
		1932	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1933	#endif
		1934	#ifdef DT_BLK
		1935	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1936	#endif
		1937	#ifdef DT_MPB
		1938	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1939	#endif
		1940	#ifdef DT_REG
		1941	case DT_REG: ent->type = EIO_DT_REG; break;
		1942	#endif
		1943	#ifdef DT_NWK
		1944	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1945	#endif
		1946	#ifdef DT_CMP
		1947	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1948	#endif
		1949	#ifdef DT_LNK
		1950	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1951	#endif
		1952	#ifdef DT_SOCK
		1953	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1954	#endif
		1955	#ifdef DT_DOOR
		1956	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1957	#endif
		1958	#ifdef DT_WHT
		1959	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1960	#endif
		1961	}
		1962
		1963	ent->score = 7;
		1964
		1965	if (flags & EIO_READDIR_DIRS_FIRST)
		1966	{
		1967	if (ent->type == EIO_DT_UNKNOWN)
1251	{	1968	{
1252	if (dir->type == EIO_DT_DIR)	1969	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1253	++dir;	1970	ent->score = 1;
1254	else if ((--oth)->type == EIO_DT_DIR)	1971	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1255	{	1972	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1256	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1257
1258	++dir;
1259	}
1260	}	1973	}
1261		1974	else if (ent->type == EIO_DT_DIR)
1262	/* now sort the dirs only */	1975	ent->score = 0;
1263	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1264	}	1976	}
1265
1266	break;
1267	}
1268
1269	/* now add the entry to our list(s) */
1270	name = entp->d_name;
1271
1272	/* skip . and .. entries */
1273	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1274	{
1275	int len = D_NAMLEN (entp) + 1;
1276
1277	while (expect_false (namesoffs + len > namesalloc))
1278	{
1279	namesalloc *= 2;
1280	X_LOCK (wrklock);
1281	req->ptr2 = names = realloc (names, namesalloc);
1282	X_UNLOCK (wrklock);
1283
1284	if (!names)
1285	break;
1286	}	1977	}
1287		1978
1288	memcpy (names + namesoffs, name, len);
1289
1290	if (dents)
1291	{
1292	struct eio_dirent *ent;
1293
1294	if (expect_false (dentoffs == dentalloc))
1295	{
1296	dentalloc *= 2;
1297	X_LOCK (wrklock);
1298	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1299	X_UNLOCK (wrklock);
1300
1301	if (!dents)
1302	break;
1303	}
1304
1305	ent = dents + dentoffs;
1306
1307	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1308	ent->namelen = len - 1;
1309	ent->inode = D_INO (entp);
1310
1311	inode_bits |= ent->inode;
1312
1313	switch (D_TYPE (entp))
1314	{
1315	default:
1316	ent->type = EIO_DT_UNKNOWN;
1317	flags |= EIO_READDIR_FOUND_UNKNOWN;
1318	break;
1319
1320	#ifdef DT_FIFO
1321	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1322	#endif
1323	#ifdef DT_CHR
1324	case DT_CHR: ent->type = EIO_DT_CHR; break;
1325	#endif
1326	#ifdef DT_MPC
1327	case DT_MPC: ent->type = EIO_DT_MPC; break;
1328	#endif
1329	#ifdef DT_DIR
1330	case DT_DIR: ent->type = EIO_DT_DIR; break;
1331	#endif
1332	#ifdef DT_NAM
1333	case DT_NAM: ent->type = EIO_DT_NAM; break;
1334	#endif
1335	#ifdef DT_BLK
1336	case DT_BLK: ent->type = EIO_DT_BLK; break;
1337	#endif
1338	#ifdef DT_MPB
1339	case DT_MPB: ent->type = EIO_DT_MPB; break;
1340	#endif
1341	#ifdef DT_REG
1342	case DT_REG: ent->type = EIO_DT_REG; break;
1343	#endif
1344	#ifdef DT_NWK
1345	case DT_NWK: ent->type = EIO_DT_NWK; break;
1346	#endif
1347	#ifdef DT_CMP
1348	case DT_CMP: ent->type = EIO_DT_CMP; break;
1349	#endif
1350	#ifdef DT_LNK
1351	case DT_LNK: ent->type = EIO_DT_LNK; break;
1352	#endif
1353	#ifdef DT_SOCK
1354	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1355	#endif
1356	#ifdef DT_DOOR
1357	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1358	#endif
1359	#ifdef DT_WHT
1360	case DT_WHT: ent->type = EIO_DT_WHT; break;
1361	#endif
1362	}
1363
1364	ent->score = 7;
1365
1366	if (flags & EIO_READDIR_DIRS_FIRST)
1367	{
1368	if (ent->type == EIO_DT_UNKNOWN)
1369	{
1370	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1371	ent->score = 1;
1372	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1373	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1374	}
1375	else if (ent->type == EIO_DT_DIR)
1376	ent->score = 0;
1377	}
1378	}
1379
1380	namesoffs += len;	1979	namesoffs += len;
1381	++dentoffs;	1980	++dentoffs;
1382	}	1981	}
1383		1982
1384	if (EIO_CANCELLED (req))	1983	if (EIO_CANCELLED (req))
1385	{	1984	{
1386	errno = ECANCELED;	1985	errno = ECANCELED;
1387	break;	1986	break;
1388	}	1987	}
		1988
		1989	#ifdef _WIN32
		1990	if (!FindNextFile (dirp, &entp))
		1991	{
		1992	FindClose (dirp);
		1993	dirp = 0;
1389	}	1994	}
		1995	#endif
		1996	}
1390	}	1997	}
1391		1998
1392	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)	1999	/*****************************************************************************/
1393	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)	2000	/* working directory stuff */
1394	#else	2001	/* various deficiencies in the posix 2008 api force us to */
		2002	/* keep the absolute path in string form at all times */
		2003	/* fuck yeah. */
1395		2004
1396	int	2005	#if !HAVE_AT
1397	eio__msync (void *mem, size_t len, int flags)	2006
		2007	/* a bit like realpath, but usually faster because it doesn'T have to return */
		2008	/* an absolute or canonical path */
		2009	static const char *
		2010	wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1398	{	2011	{
1399	if (EIO_MS_ASYNC != MS_SYNC	2012	if (!wd || *path == '/')
1400	|| EIO_MS_INVALIDATE != MS_INVALIDATE	2013	return path;
1401	|| EIO_MS_SYNC != MS_SYNC)	2014
		2015	if (path [0] == '.' && !path [1])
		2016	return wd->str;
		2017
		2018	{
		2019	int l1 = wd->len;
		2020	int l2 = strlen (path);
		2021
		2022	char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
		2023
		2024	memcpy (res, wd->str, l1);
		2025	res [l1] = '/';
		2026	memcpy (res + l1 + 1, path, l2 + 1);
		2027
		2028	return res;
		2029	}
		2030	}
		2031
		2032	#endif
		2033
		2034	static eio_wd
		2035	eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2036	{
		2037	int fd;
		2038	eio_wd res;
		2039	int len = eio__realpath (tmpbuf, wd, path);
		2040
		2041	if (len < 0)
		2042	return EIO_INVALID_WD;
		2043
		2044	#if HAVE_AT
		2045	fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		2046
		2047	if (fd < 0)
		2048	return EIO_INVALID_WD;
		2049	#endif
		2050
		2051	res = malloc (sizeof (*res) + len); /* one extra 0-byte */
		2052
		2053	#if HAVE_AT
		2054	res->fd = fd;
		2055	#endif
		2056
		2057	res->len = len;
		2058	memcpy (res->str, tmpbuf->ptr, len);
		2059	res->str [len] = 0;
		2060
		2061	return res;
		2062	}
		2063
		2064	eio_wd
		2065	eio_wd_open_sync (eio_wd wd, const char *path)
		2066	{
		2067	struct tmpbuf tmpbuf = { 0 };
		2068	wd = eio__wd_open_sync (&tmpbuf, wd, path);
		2069	free (tmpbuf.ptr);
		2070
		2071	return wd;
		2072	}
		2073
		2074	void
		2075	eio_wd_close_sync (eio_wd wd)
		2076	{
		2077	if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1402	{	2078	{
1403	flags = 0	2079	#if HAVE_AT
1404	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)	2080	close (wd->fd);
1405	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)	2081	#endif
1406	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);	2082	free (wd);
1407	}	2083	}
1408
1409	return msync (mem, len, flags);
1410	}	2084	}
1411		2085
1412	#endif	2086	#if HAVE_AT
1413		2087
1414	int	2088	/* they forgot these */
1415	eio__mtouch (void *mem, size_t len, int flags)
1416	{
1417	intptr_t addr = (intptr_t)mem;
1418	intptr_t end = addr + len;
1419	#ifdef PAGESIZE
1420	const intptr_t page = PAGESIZE;
1421	#else
1422	static intptr_t page;
1423		2089
1424	if (!page)	2090	static int
1425	page = sysconf (_SC_PAGESIZE);	2091	eio__truncateat (int dirfd, const char *path, off_t length)
1426	#endif	2092	{
		2093	int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
		2094	int res;
1427		2095
1428	/* round down to start of page, although this is probably useless */	2096	if (fd < 0)
1429	addr &= ~(page - 1); /* assume page size is always a power of two */	2097	return fd;
1430		2098
1431	if (addr < end)	2099	res = ftruncate (fd, length);
1432	if (flags & EIO_MT_MODIFY) /* modify */	2100	close (fd);
1433	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1434	else
1435	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1436
1437	return 0;	2101	return res;
1438	}	2102	}
		2103
		2104	static int
		2105	eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
		2106	{
		2107	int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
		2108	int res;
		2109
		2110	if (fd < 0)
		2111	return fd;
		2112
		2113	res = fstatvfs (fd, buf);
		2114	close (fd);
		2115	return res;
		2116
		2117	}
		2118
		2119	#endif
1439		2120
1440	/*****************************************************************************/	2121	/*****************************************************************************/
1441		2122
1442	#define ALLOC(len) \	2123	#define ALLOC(len) \
1443	if (!req->ptr2) \	2124	if (!req->ptr2) \
…		…
1458	{	2139	{
1459	ETP_REQ *req;	2140	ETP_REQ *req;
1460	struct timespec ts;	2141	struct timespec ts;
1461	etp_worker *self = (etp_worker *)thr_arg;	2142	etp_worker *self = (etp_worker *)thr_arg;
1462		2143
		2144	#if HAVE_PRCTL_SET_NAME
		2145	prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
		2146	#endif
		2147
1463	/* try to distribute timeouts somewhat randomly */	2148	/* try to distribute timeouts somewhat evenly */
1464	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	2149	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1465		2150
1466	for (;;)	2151	for (;;)
1467	{	2152	{
		2153	ts.tv_sec = 0;
		2154
1468	X_LOCK (reqlock);	2155	X_LOCK (reqlock);
1469		2156
1470	for (;;)	2157	for (;;)
1471	{	2158	{
1472	self->req = req = reqq_shift (&req_queue);	2159	req = reqq_shift (&req_queue);
1473		2160
1474	if (req)	2161	if (req)
1475	break;	2162	break;
1476		2163
		2164	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		2165	{
		2166	X_UNLOCK (reqlock);
		2167	X_LOCK (wrklock);
		2168	--started;
		2169	X_UNLOCK (wrklock);
		2170	goto quit;
		2171	}
		2172
1477	++idle;	2173	++idle;
1478		2174
1479	ts.tv_sec = time (0) + IDLE_TIMEOUT;	2175	if (idle <= max_idle)
1480	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	2176	/* we are allowed to idle, so do so without any timeout */
		2177	X_COND_WAIT (reqwait, reqlock);
		2178	else
1481	{	2179	{
1482	if (idle > max_idle)	2180	/* initialise timeout once */
1483	{	2181	if (!ts.tv_sec)
1484	--idle;	2182	ts.tv_sec = time (0) + idle_timeout;
1485	X_UNLOCK (reqlock);
1486	X_LOCK (wrklock);
1487	--started;
1488	X_UNLOCK (wrklock);
1489	goto quit;
1490	}
1491		2183
1492	/* we are allowed to idle, so do so without any timeout */
1493	X_COND_WAIT (reqwait, reqlock);	2184	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		2185	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1494	}	2186	}
1495		2187
1496	--idle;	2188	--idle;
1497	}	2189	}
1498		2190
…		…
1501	X_UNLOCK (reqlock);	2193	X_UNLOCK (reqlock);
1502		2194
1503	if (req->type < 0)	2195	if (req->type < 0)
1504	goto quit;	2196	goto quit;
1505		2197
1506	if (!EIO_CANCELLED (req))
1507	ETP_EXECUTE (self, req);	2198	ETP_EXECUTE (self, req);
1508		2199
1509	X_LOCK (reslock);	2200	X_LOCK (reslock);
1510		2201
1511	++npending;	2202	++npending;
1512		2203
1513	if (!reqq_push (&res_queue, req) && want_poll_cb)	2204	if (!reqq_push (&res_queue, req) && want_poll_cb)
1514	want_poll_cb ();	2205	want_poll_cb ();
1515		2206
1516	self->req = 0;
1517	etp_worker_clear (self);	2207	etp_worker_clear (self);
1518		2208
1519	X_UNLOCK (reslock);	2209	X_UNLOCK (reslock);
1520	}	2210	}
1521		2211
1522	quit:	2212	quit:
		2213	free (req);
		2214
1523	X_LOCK (wrklock);	2215	X_LOCK (wrklock);
1524	etp_worker_free (self);	2216	etp_worker_free (self);
1525	X_UNLOCK (wrklock);	2217	X_UNLOCK (wrklock);
1526		2218
1527	return 0;	2219	return 0;
1528	}	2220	}
1529		2221
1530	/*****************************************************************************/	2222	/*****************************************************************************/
1531		2223
		2224	int ecb_cold
1532	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	2225	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1533	{	2226	{
		2227	#if !HAVE_PREADWRITE
		2228	X_MUTEX_CREATE (preadwritelock);
		2229	#endif
		2230
1534	return etp_init (want_poll, done_poll);	2231	return etp_init (want_poll, done_poll);
1535	}	2232	}
1536		2233
		2234	ecb_inline void
1537	static void eio_api_destroy (eio_req *req)	2235	eio_api_destroy (eio_req *req)
1538	{	2236	{
1539	free (req);	2237	free (req);
1540	}	2238	}
1541		2239
1542	#define REQ(rtype) \	2240	#define REQ(rtype) \
…		…
1561	{ \	2259	{ \
1562	eio_api_destroy (req); \	2260	eio_api_destroy (req); \
1563	return 0; \	2261	return 0; \
1564	}	2262	}
1565		2263
		2264	static void
1566	static void eio_execute (etp_worker *self, eio_req *req)	2265	eio_execute (etp_worker *self, eio_req *req)
1567	{	2266	{
1568	errno = 0;	2267	#if HAVE_AT
		2268	int dirfd;
		2269	#else
		2270	const char *path;
		2271	#endif
		2272
		2273	if (ecb_expect_false (EIO_CANCELLED (req)))
		2274	{
		2275	req->result = -1;
		2276	req->errorno = ECANCELED;
		2277	return;
		2278	}
		2279
		2280	if (ecb_expect_false (req->wd == EIO_INVALID_WD))
		2281	{
		2282	req->result = -1;
		2283	req->errorno = ENOENT;
		2284	return;
		2285	}
		2286
		2287	if (req->type >= EIO_OPEN)
		2288	{
		2289	#if HAVE_AT
		2290	dirfd = WD2FD (req->wd);
		2291	#else
		2292	path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		2293	#endif
		2294	}
1569		2295
1570	switch (req->type)	2296	switch (req->type)
1571	{	2297	{
		2298	case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
		2299	req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
		2300	break;
		2301	case EIO_WD_CLOSE: req->result = 0;
		2302	eio_wd_close_sync (req->wd); break;
		2303
1572	case EIO_READ: ALLOC (req->size);	2304	case EIO_READ: ALLOC (req->size);
1573	req->result = req->offs >= 0	2305	req->result = req->offs >= 0
1574	? pread (req->int1, req->ptr2, req->size, req->offs)	2306	? pread (req->int1, req->ptr2, req->size, req->offs)
1575	: read (req->int1, req->ptr2, req->size); break;	2307	: read (req->int1, req->ptr2, req->size); break;
1576	case EIO_WRITE: req->result = req->offs >= 0	2308	case EIO_WRITE: req->result = req->offs >= 0
1577	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2309	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1578	: write (req->int1, req->ptr2, req->size); break;	2310	: write (req->int1, req->ptr2, req->size); break;
1579		2311
1580	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2312	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1581	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2313	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
		2314
		2315	#if HAVE_AT
1582		2316
1583	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2317	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1584	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2318	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1585	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2319	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1586	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2320	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1587	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2321	case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1588	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2322	case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
		2323	case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
		2324	case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1589		2325
		2326	case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
		2327	case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
		2328	case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
		2329	case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
		2330	case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
		2331	case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
		2332	case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2333	case EIO_READLINK: ALLOC (PATH_MAX);
		2334	req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1590	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));	2335	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1591	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;	2336	req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1592	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1593	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1594
1595	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1596	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1597	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1598	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1599	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1600	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1601
1602	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1603	case EIO_CLOSE: req->result = close (req->int1); break;
1604	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1605	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1606	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1607	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1608	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1609	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1610	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1611	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1612
1613	case EIO_READLINK: ALLOC (NAME_MAX);
1614	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1615
1616	case EIO_SYNC: req->result = 0; sync (); break;
1617	case EIO_FSYNC: req->result = fsync (req->int1); break;
1618	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1619	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1620	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1621	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1622
1623	case EIO_READDIR: eio__scandir (req, self); break;
1624
1625	case EIO_BUSY:	2337	case EIO_UTIME:
1626	#ifdef _WIN32	2338	case EIO_FUTIME:
1627	Sleep (req->nv1 * 1000.);	2339	{
		2340	struct timespec ts[2];
		2341	struct timespec *times;
		2342
		2343	if (req->nv1 != -1. || req->nv2 != -1.)
		2344	{
		2345	ts[0].tv_sec = req->nv1;
		2346	ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
		2347	ts[1].tv_sec = req->nv2;
		2348	ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
		2349
		2350	times = ts;
		2351	}
		2352	else
		2353	times = 0;
		2354
		2355	req->result = req->type == EIO_FUTIME
		2356	? futimens (req->int1, times)
		2357	: utimensat (dirfd, req->ptr1, times, 0);
		2358	}
		2359	break;
		2360
1628	#else	2361	#else
1629	{
1630	struct timeval tv;
1631		2362
1632	tv.tv_sec = req->nv1;	2363	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1633	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	2364	req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2365	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2366	req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2367	case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
		2368	case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
		2369	case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
		2370	case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1634		2371
1635	req->result = select (0, 0, 0, 0, &tv);	2372	case EIO_UNLINK: req->result = unlink (path ); break;
1636	}	2373	case EIO_RMDIR: req->result = rmdir (path ); break;
1637	#endif	2374	case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1638	break;	2375	case EIO_RENAME: req->result = rename (path , req->ptr2); break;
		2376	case EIO_LINK: req->result = link (path , req->ptr2); break;
		2377	case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
		2378	case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
		2379	case EIO_READLINK: ALLOC (PATH_MAX);
		2380	req->result = readlink (path, req->ptr2, PATH_MAX); break;
		2381	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2382	req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1639		2383
1640	case EIO_UTIME:	2384	case EIO_UTIME:
1641	case EIO_FUTIME:	2385	case EIO_FUTIME:
1642	{	2386	{
1643	struct timeval tv[2];	2387	struct timeval tv[2];
1644	struct timeval *times;	2388	struct timeval *times;
1645		2389
1646	if (req->nv1 != -1. || req->nv2 != -1.)	2390	if (req->nv1 != -1. || req->nv2 != -1.)
1647	{	2391	{
1648	tv[0].tv_sec = req->nv1;	2392	tv[0].tv_sec = req->nv1;
1649	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;	2393	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1650	tv[1].tv_sec = req->nv2;	2394	tv[1].tv_sec = req->nv2;
1651	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;	2395	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1652		2396
1653	times = tv;	2397	times = tv;
1654	}	2398	}
1655	else	2399	else
1656	times = 0;	2400	times = 0;
1657
1658		2401
1659	req->result = req->type == EIO_FUTIME	2402	req->result = req->type == EIO_FUTIME
1660	? futimes (req->int1, times)	2403	? futimes (req->int1, times)
1661	: utimes (req->ptr1, times);	2404	: utimes (req->ptr1, times);
1662	}	2405	}
1663	break;	2406	break;
1664		2407
		2408	#endif
		2409
		2410	case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
		2411	{
		2412	ALLOC (req->result);
		2413	memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
		2414	}
		2415	break;
		2416
		2417	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2418	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2419
		2420	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2421	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2422
		2423	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		2424	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		2425	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		2426
		2427	case EIO_CLOSE: req->result = close (req->int1); break;
		2428	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		2429	case EIO_SYNC: req->result = 0; sync (); break;
		2430	case EIO_FSYNC: req->result = fsync (req->int1); break;
		2431	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2432	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2433	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2434	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2435	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		2436	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2437	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		2438	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
		2439
		2440	case EIO_READDIR: eio__scandir (req, self); break;
		2441
		2442	case EIO_BUSY:
		2443	#ifdef _WIN32
		2444	Sleep (req->nv1 * 1e3);
		2445	#else
		2446	{
		2447	struct timeval tv;
		2448
		2449	tv.tv_sec = req->nv1;
		2450	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
		2451
		2452	req->result = select (0, 0, 0, 0, &tv);
		2453	}
		2454	#endif
		2455	break;
		2456
1665	case EIO_GROUP:	2457	case EIO_GROUP:
1666	abort (); /* handled in eio_request */	2458	abort (); /* handled in eio_request */
1667		2459
1668	case EIO_NOP:	2460	case EIO_NOP:
1669	req->result = 0;	2461	req->result = 0;
1670	break;	2462	break;
1671		2463
1672	case EIO_CUSTOM:	2464	case EIO_CUSTOM:
1673	((void (*)(eio_req *))req->feed) (req);	2465	req->feed (req);
1674	break;	2466	break;
1675		2467
1676	default:	2468	default:
		2469	errno = ENOSYS;
1677	req->result = -1;	2470	req->result = -1;
1678	break;	2471	break;
1679	}	2472	}
1680		2473
1681	req->errorno = errno;	2474	req->errorno = errno;
1682	}	2475	}
1683		2476
1684	#ifndef EIO_NO_WRAPPERS	2477	#ifndef EIO_NO_WRAPPERS
1685		2478
		2479	eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
		2480	{
		2481	REQ (EIO_WD_OPEN); PATH; SEND;
		2482	}
		2483
		2484	eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
		2485	{
		2486	REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
		2487	}
		2488
1686	eio_req *eio_nop (int pri, eio_cb cb, void *data)	2489	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1687	{	2490	{
1688	REQ (EIO_NOP); SEND;	2491	REQ (EIO_NOP); SEND;
1689	}	2492	}
1690		2493
…		…
1706	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2509	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1707	{	2510	{
1708	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2511	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1709	}	2512	}
1710		2513
		2514	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		2515	{
		2516	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2517	}
		2518
		2519	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2520	{
		2521	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2522	}
		2523
		2524	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2525	{
		2526	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2527	}
		2528
1711	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2529	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1712	{	2530	{
1713	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2531	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1714	}	2532	}
1715		2533
1716	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2534	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1717	{	2535	{
1718	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2536	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1719	}	2537	}
1720		2538
1721	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2539	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1722	{	2540	{
1723	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2541	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2542	}
		2543
		2544	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2545	{
		2546	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1724	}	2547	}
1725		2548
1726	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2549	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1727	{	2550	{
1728	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2551	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
1766	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2589	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1767	{	2590	{
1768	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2591	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1769	}	2592	}
1770		2593
1771	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2594	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1772	{	2595	{
1773	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2596	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1774	}	2597	}
1775		2598
1776	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2599	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1796	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2619	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1797	{	2620	{
1798	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2621	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1799	}	2622	}
1800		2623
1801	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2624	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1802	{	2625	{
1803	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2626	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1804	}	2627	}
1805		2628
1806	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2629	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1822	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2645	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1823	{	2646	{
1824	return eio__1path (EIO_READLINK, path, pri, cb, data);	2647	return eio__1path (EIO_READLINK, path, pri, cb, data);
1825	}	2648	}
1826		2649
		2650	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2651	{
		2652	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2653	}
		2654
1827	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2655	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1828	{	2656	{
1829	return eio__1path (EIO_STAT, path, pri, cb, data);	2657	return eio__1path (EIO_STAT, path, pri, cb, data);
1830	}	2658	}
1831		2659
…		…
1854	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2682	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1855	}	2683	}
1856		2684
1857	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2685	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1858	{	2686	{
1859	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2687	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1860	}	2688	}
1861		2689
1862	static eio_req *	2690	static eio_req *
1863	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2691	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1864	{	2692	{
…		…
1888	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2716	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1889	{	2717	{
1890	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2718	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1891	}	2719	}
1892		2720
1893	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2721	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1894	{	2722	{
1895	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2723	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1896	}	2724	}
1897		2725
1898	#endif	2726	#endif
1899		2727
1900	eio_req *eio_grp (eio_cb cb, void *data)	2728	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1909	#undef SEND	2737	#undef SEND
1910		2738
1911	/*****************************************************************************/	2739	/*****************************************************************************/
1912	/* grp functions */	2740	/* grp functions */
1913		2741
		2742	void
1914	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2743	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1915	{	2744	{
1916	grp->int2 = limit;	2745	grp->int2 = limit;
1917	grp->feed = feed;	2746	grp->feed = feed;
1918		2747
1919	grp_try_feed (grp);	2748	grp_try_feed (grp);
1920	}	2749	}
1921		2750
		2751	void
1922	void eio_grp_limit (eio_req *grp, int limit)	2752	eio_grp_limit (eio_req *grp, int limit)
1923	{	2753	{
1924	grp->int2 = limit;	2754	grp->int2 = limit;
1925		2755
1926	grp_try_feed (grp);	2756	grp_try_feed (grp);
1927	}	2757	}
1928		2758
		2759	void
1929	void eio_grp_add (eio_req *grp, eio_req *req)	2760	eio_grp_add (eio_req *grp, eio_req *req)
1930	{	2761	{
1931	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2762	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1932		2763
1933	grp->flags |= EIO_FLAG_GROUPADD;	2764	grp->flags |= EIO_FLAG_GROUPADD;
1934		2765
…		…
1945	}	2776	}
1946		2777
1947	/*****************************************************************************/	2778	/*****************************************************************************/
1948	/* misc garbage */	2779	/* misc garbage */
1949		2780
		2781	eio_ssize_t
1950	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2782	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1951	{	2783	{
1952	etp_worker wrk;
1953	ssize_t ret;
1954
1955	wrk.dbuf = 0;
1956
1957	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2784	return eio__sendfile (ofd, ifd, offset, count);
1958
1959	if (wrk.dbuf)
1960	free (wrk.dbuf);
1961
1962	return ret;
1963	}	2785	}
1964		2786

Diff Legend

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