ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/cvsroot/libeio/eio.c

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.66 by root, Sun Jun 5 19:58:37 2011 UTC vs.
Revision 1.121 by root, Sat Jun 2 20:13:26 2012 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>	4	* Copyright (c) 2007,2008,2009,2010,2011,2012 Marc Alexander Lehmann <libeio@schmorp.de>
5	* All rights reserved.	5	* All rights reserved.
6	*	6	*
7	* Redistribution and use in source and binary forms, with or without modifica-	7	* Redistribution and use in source and binary forms, with or without modifica-
8	* tion, are permitted provided that the following conditions are met:	8	* tion, are permitted provided that the following conditions are met:
9	*	9	*
…		…
40	#ifndef _WIN32	40	#ifndef _WIN32
41	# include "config.h"	41	# include "config.h"
42	#endif	42	#endif
43		43
44	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
45		46
46	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
47	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define X_STACKSIZE EIO_STACKSIZE
48	#endif	49	#endif
49	#include "xthread.h"	50	#include "xthread.h"
50		51
51	#include <errno.h>	52	#include <errno.h>
52	#include <stddef.h>	53	#include <stddef.h>
53	#include <stdlib.h>	54	#include <stdlib.h>
54	#include <string.h>	55	#include <string.h>
55	#include <errno.h>	56	#include <errno.h>
56	#include <sys/types.h>	57	#include <sys/types.h>
57	#include <sys/stat.h>	58	#include <sys/stat.h>
58	#include <sys/statvfs.h>
59	#include <limits.h>	59	#include <limits.h>
60	#include <fcntl.h>	60	#include <fcntl.h>
61	#include <assert.h>	61	#include <assert.h>
62		62
63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */	63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64	/* intptr_t only comes form stdint.h, says idiot openbsd coder */	64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65	#if HAVE_STDINT_H	65	#if HAVE_STDINT_H
66	# include <stdint.h>	66	# include <stdint.h>
67	#endif	67	#endif
68		68
		69	#ifndef ECANCELED
		70	# define ECANCELED EDOM
		71	#endif
		72	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
		77	# define ENOTSOCK WSAENOTSOCK
		78	#endif
		79
		80	static void eio_destroy (eio_req *req);
		81
69	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
70	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	83	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
71	#endif	84	#endif
72		85
73	#ifndef EIO_DESTROY	86	#ifndef EIO_DESTROY
…		…
76		89
77	#ifndef EIO_FEED	90	#ifndef EIO_FEED
78	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	91	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
79	#endif	92	#endif
80		93
		94	#ifndef EIO_FD_TO_WIN32_HANDLE
		95	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		96	#endif
		97	#ifndef EIO_WIN32_HANDLE_TO_FD
		98	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		99	#endif
		100
		101	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		102
		103	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		104
81	#ifdef _WIN32	105	#ifdef _WIN32
82		106
83	/*doh*/	107	#undef PAGESIZE
		108	#define PAGESIZE 4096 /* GetSystemInfo? */
		109
		110	/* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
		111
		112	#ifdef EIO_STRUCT_STATI64
		113	/* look at perl's non-sloppy stat */
		114	#define stat(path,buf) _stati64 (path,buf)
		115	#define fstat(fd,buf) _fstati64 (fd,buf)
		116	#endif
		117	#define lstat(path,buf) stat (path,buf)
		118	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		119	#define mkdir(path,mode) _mkdir (path)
		120	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		121
		122	#define chmod(path,mode) _chmod (path, mode)
		123	#define dup(fd) _dup (fd)
		124	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		125
		126	#define fchmod(fd,mode) EIO_ENOSYS ()
		127	#define chown(path,uid,gid) EIO_ENOSYS ()
		128	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		129	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		130	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		131	#define mknod(path,mode,dev) EIO_ENOSYS ()
		132	#define sync() EIO_ENOSYS ()
		133	#define readlink(path,buf,s) EIO_ENOSYS ()
		134	#define statvfs(path,buf) EIO_ENOSYS ()
		135	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		136
		137	/* rename() uses MoveFile, which fails to overwrite */
		138	#define rename(old,neu) eio__rename (old, neu)
		139
		140	static int
		141	eio__rename (const char *old, const char *neu)
		142	{
		143	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		144	return 0;
		145
		146	/* should steal _dosmaperr */
		147	switch (GetLastError ())
		148	{
		149	case ERROR_FILE_NOT_FOUND:
		150	case ERROR_PATH_NOT_FOUND:
		151	case ERROR_INVALID_DRIVE:
		152	case ERROR_NO_MORE_FILES:
		153	case ERROR_BAD_NETPATH:
		154	case ERROR_BAD_NET_NAME:
		155	case ERROR_BAD_PATHNAME:
		156	case ERROR_FILENAME_EXCED_RANGE:
		157	errno = ENOENT;
		158	break;
		159
		160	default:
		161	errno = EACCES;
		162	break;
		163	}
		164
		165	return -1;
		166	}
		167
		168	/* we could even stat and see if it exists */
		169	static int
		170	symlink (const char *old, const char *neu)
		171	{
		172	#if WINVER >= 0x0600
		173	if (CreateSymbolicLink (neu, old, 1))
		174	return 0;
		175
		176	if (CreateSymbolicLink (neu, old, 0))
		177	return 0;
		178	#endif
		179
		180	return EIO_ERRNO (ENOENT, -1);
		181	}
		182
		183	/* POSIX API only */
		184	#define CreateHardLink(neu,old,flags) 0
		185	#define CreateSymbolicLink(neu,old,flags) 0
		186
		187	struct statvfs
		188	{
		189	int dummy;
		190	};
		191
		192	#define DT_DIR EIO_DT_DIR
		193	#define DT_REG EIO_DT_REG
		194	#define D_NAME(entp) entp.cFileName
		195	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
		196
84	#else	197	#else
85		198
86	# include <sys/time.h>	199	#include <sys/time.h>
87	# include <sys/select.h>	200	#include <sys/select.h>
		201	#include <sys/statvfs.h>
88	# include <unistd.h>	202	#include <unistd.h>
		203	#include <signal.h>
		204	#include <dirent.h>
		205
		206	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		207	#include <sys/mman.h>
		208	#endif
		209
		210	#define D_NAME(entp) entp->d_name
		211
		212	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		213	#if __FreeBSD__ || __NetBSD__ || __OpenBSD__
		214	#define _DIRENT_HAVE_D_TYPE /* sigh */
		215	#define D_INO(de) (de)->d_fileno
		216	#define D_NAMLEN(de) (de)->d_namlen
		217	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		218	#define D_INO(de) (de)->d_ino
		219	#endif
		220
		221	#ifdef _D_EXACT_NAMLEN
		222	#undef D_NAMLEN
		223	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		224	#endif
		225
		226	#ifdef _DIRENT_HAVE_D_TYPE
		227	#define D_TYPE(de) (de)->d_type
		228	#endif
		229
		230	#ifndef EIO_STRUCT_DIRENT
		231	#define EIO_STRUCT_DIRENT struct dirent
		232	#endif
		233
		234	#endif
		235
		236	#if HAVE_UTIMES
89	# include <utime.h>	237	# include <utime.h>
90	# include <signal.h>
91	# include <dirent.h>
92
93	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94	# include <sys/mman.h>
95	#endif
96
97	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99	# define _DIRENT_HAVE_D_TYPE /* sigh */
100	# define D_INO(de) (de)->d_fileno
101	# define D_NAMLEN(de) (de)->d_namlen
102	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103	# define D_INO(de) (de)->d_ino
104	# endif	238	#endif
105		239
106	#ifdef _D_EXACT_NAMLEN	240	#if HAVE_SYS_SYSCALL_H
107	# undef D_NAMLEN	241	# include <sys/syscall.h>
108	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109	#endif
110
111	# ifdef _DIRENT_HAVE_D_TYPE
112	# define D_TYPE(de) (de)->d_type
113	# endif	242	#endif
114		243
115	# ifndef EIO_STRUCT_DIRENT	244	#if HAVE_SYS_PRCTL_H
116	# define EIO_STRUCT_DIRENT struct dirent	245	# include <sys/prctl.h>
117	# endif
118
119	#endif	246	#endif
120		247
121	#if HAVE_SENDFILE	248	#if HAVE_SENDFILE
122	# if __linux	249	# if __linux
123	# include <sys/sendfile.h>	250	# include <sys/sendfile.h>
…		…
138	#endif	265	#endif
139	#ifndef D_INO	266	#ifndef D_INO
140	# define D_INO(de) 0	267	# define D_INO(de) 0
141	#endif	268	#endif
142	#ifndef D_NAMLEN	269	#ifndef D_NAMLEN
143	# define D_NAMLEN(de) strlen ((de)->d_name)	270	# define D_NAMLEN(entp) strlen (D_NAME (entp))
144	#endif	271	#endif
145		272
146	/* used for struct dirent, AIX doesn't provide it */	273	/* used for struct dirent, AIX doesn't provide it */
147	#ifndef NAME_MAX	274	#ifndef NAME_MAX
148	# define NAME_MAX 4096	275	# define NAME_MAX 4096
…		…
155		282
156	/* buffer size for various temporary buffers */	283	/* buffer size for various temporary buffers */
157	#define EIO_BUFSIZE 65536	284	#define EIO_BUFSIZE 65536
158		285
159	#define dBUF \	286	#define dBUF \
160	char *eio_buf; \
161	ETP_WORKER_LOCK (self); \
162	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	287	char *eio_buf = malloc (EIO_BUFSIZE); \
163	ETP_WORKER_UNLOCK (self); \
164	errno = ENOMEM; \	288	errno = ENOMEM; \
165	if (!eio_buf) \	289	if (!eio_buf) \
166	return -1;	290	return -1
		291
		292	#define FUBd \
		293	free (eio_buf)
167		294
168	#define EIO_TICKS ((1000000 + 1023) >> 10)	295	#define EIO_TICKS ((1000000 + 1023) >> 10)
169		296
170	/*****************************************************************************/	297	/*****************************************************************************/
171		298
172	#if __GNUC__ >= 3	299	struct tmpbuf
173	# define expect(expr,value) __builtin_expect ((expr),(value))	300	{
		301	void *ptr;
		302	int len;
		303	};
		304
		305	static void *
		306	tmpbuf_get (struct tmpbuf *buf, int len)
		307	{
		308	if (buf->len < len)
		309	{
		310	free (buf->ptr);
		311	buf->ptr = malloc (buf->len = len);
		312	}
		313
		314	return buf->ptr;
		315	}
		316
		317	struct tmpbuf;
		318
		319	#if _POSIX_VERSION >= 200809L
		320	#define HAVE_AT 1
		321	#define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
		322	#ifndef O_SEARCH
		323	#define O_SEARCH O_RDONLY
		324	#endif
174	#else	325	#else
175	# define expect(expr,value) (expr)	326	#define HAVE_AT 0
		327	static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
176	#endif	328	#endif
177		329
178	#define expect_false(expr) expect ((expr) != 0, 0)	330	struct eio_pwd
179	#define expect_true(expr) expect ((expr) != 0, 1)	331	{
		332	#if HAVE_AT
		333	int fd;
		334	#endif
		335	int len;
		336	char str[1]; /* actually, a 0-terminated canonical path */
		337	};
180		338
181	/*****************************************************************************/	339	/*****************************************************************************/
182		340
183	#define ETP_PRI_MIN EIO_PRI_MIN	341	#define ETP_PRI_MIN EIO_PRI_MIN
184	#define ETP_PRI_MAX EIO_PRI_MAX	342	#define ETP_PRI_MAX EIO_PRI_MAX
…		…
190	static int eio_finish (eio_req *req);	348	static int eio_finish (eio_req *req);
191	#define ETP_FINISH(req) eio_finish (req)	349	#define ETP_FINISH(req) eio_finish (req)
192	static void eio_execute (struct etp_worker *self, eio_req *req);	350	static void eio_execute (struct etp_worker *self, eio_req *req);
193	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	351	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194		352
195	#define ETP_WORKER_CLEAR(req) \
196	if (wrk->dbuf) \
197	{ \
198	free (wrk->dbuf); \
199	wrk->dbuf = 0; \
200	} \
201	\
202	if (wrk->dirp) \
203	{ \
204	closedir (wrk->dirp); \
205	wrk->dirp = 0; \
206	}
207
208	#define ETP_WORKER_COMMON \
209	void *dbuf; \
210	DIR *dirp;
211
212	/*****************************************************************************/	353	/*****************************************************************************/
213		354
214	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	355	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215		356
216	/* calculate time difference in ~1/EIO_TICKS of a second */	357	/* calculate time difference in ~1/EIO_TICKS of a second */
		358	ecb_inline int
217	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	359	tvdiff (struct timeval *tv1, struct timeval *tv2)
218	{	360	{
219	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	361	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	362	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
221	}	363	}
222		364
…		…
226	static void (*done_poll_cb) (void);	368	static void (*done_poll_cb) (void);
227		369
228	static unsigned int max_poll_time; /* reslock */	370	static unsigned int max_poll_time; /* reslock */
229	static unsigned int max_poll_reqs; /* reslock */	371	static unsigned int max_poll_reqs; /* reslock */
230		372
231	static volatile unsigned int nreqs; /* reqlock */	373	static unsigned int nreqs; /* reqlock */
232	static volatile unsigned int nready; /* reqlock */	374	static unsigned int nready; /* reqlock */
233	static volatile unsigned int npending; /* reqlock */	375	static unsigned int npending; /* reqlock */
234	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */	376	static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
235	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */	377	static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
236		378
237	static xmutex_t wrklock;	379	static xmutex_t wrklock;
238	static xmutex_t reslock;	380	static xmutex_t reslock;
239	static xmutex_t reqlock;	381	static xmutex_t reqlock;
240	static xcond_t reqwait;	382	static xcond_t reqwait;
241		383
242	#if !HAVE_PREADWRITE
243	/*
244	* make our pread/pwrite emulation safe against themselves, but not against
245	* normal read/write by using a mutex. slows down execution a lot,
246	* but that's your problem, not mine.
247	*/
248	static xmutex_t preadwritelock = X_MUTEX_INIT;
249	#endif
250
251	typedef struct etp_worker	384	typedef struct etp_worker
252	{	385	{
		386	struct tmpbuf tmpbuf;
		387
253	/* locked by wrklock */	388	/* locked by wrklock */
254	struct etp_worker *prev, *next;	389	struct etp_worker *prev, *next;
255		390
256	xthread_t tid;	391	xthread_t tid;
257		392
258	/* locked by reslock, reqlock or wrklock */	393	#ifdef ETP_WORKER_COMMON
259	ETP_REQ *req; /* currently processed request */
260
261	ETP_WORKER_COMMON	394	ETP_WORKER_COMMON
		395	#endif
262	} etp_worker;	396	} etp_worker;
263		397
264	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	398	static etp_worker wrk_first; /* NOT etp */
265		399
266	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	400	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	401	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268		402
269	/* worker threads management */	403	/* worker threads management */
270		404
		405	static void
271	static void etp_worker_clear (etp_worker *wrk)	406	etp_worker_clear (etp_worker *wrk)
272	{	407	{
273	ETP_WORKER_CLEAR (wrk);
274	}	408	}
275		409
		410	static void ecb_cold
276	static void etp_worker_free (etp_worker *wrk)	411	etp_worker_free (etp_worker *wrk)
277	{	412	{
		413	free (wrk->tmpbuf.ptr);
		414
278	wrk->next->prev = wrk->prev;	415	wrk->next->prev = wrk->prev;
279	wrk->prev->next = wrk->next;	416	wrk->prev->next = wrk->next;
280		417
281	free (wrk);	418	free (wrk);
282	}	419	}
283		420
284	static unsigned int etp_nreqs (void)	421	static unsigned int
		422	etp_nreqs (void)
285	{	423	{
286	int retval;	424	int retval;
287	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	425	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288	retval = nreqs;	426	retval = nreqs;
289	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	427	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290	return retval;	428	return retval;
291	}	429	}
292		430
293	static unsigned int etp_nready (void)	431	static unsigned int
		432	etp_nready (void)
294	{	433	{
295	unsigned int retval;	434	unsigned int retval;
296		435
297	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	436	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298	retval = nready;	437	retval = nready;
299	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	438	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300		439
301	return retval;	440	return retval;
302	}	441	}
303		442
304	static unsigned int etp_npending (void)	443	static unsigned int
		444	etp_npending (void)
305	{	445	{
306	unsigned int retval;	446	unsigned int retval;
307		447
308	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	448	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309	retval = npending;	449	retval = npending;
310	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	450	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311		451
312	return retval;	452	return retval;
313	}	453	}
314		454
315	static unsigned int etp_nthreads (void)	455	static unsigned int
		456	etp_nthreads (void)
316	{	457	{
317	unsigned int retval;	458	unsigned int retval;
318		459
319	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	460	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320	retval = started;	461	retval = started;
…		…
334	} etp_reqq;	475	} etp_reqq;
335		476
336	static etp_reqq req_queue;	477	static etp_reqq req_queue;
337	static etp_reqq res_queue;	478	static etp_reqq res_queue;
338		479
		480	static void ecb_noinline ecb_cold
		481	reqq_init (etp_reqq *q)
		482	{
		483	int pri;
		484
		485	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		486	q->qs[pri] = q->qe[pri] = 0;
		487
		488	q->size = 0;
		489	}
		490
		491	static int ecb_noinline
339	static int reqq_push (etp_reqq *q, ETP_REQ *req)	492	reqq_push (etp_reqq *q, ETP_REQ *req)
340	{	493	{
341	int pri = req->pri;	494	int pri = req->pri;
342	req->next = 0;	495	req->next = 0;
343		496
344	if (q->qe[pri])	497	if (q->qe[pri])
…		…
350	q->qe[pri] = q->qs[pri] = req;	503	q->qe[pri] = q->qs[pri] = req;
351		504
352	return q->size++;	505	return q->size++;
353	}	506	}
354		507
		508	static ETP_REQ * ecb_noinline
355	static ETP_REQ *reqq_shift (etp_reqq *q)	509	reqq_shift (etp_reqq *q)
356	{	510	{
357	int pri;	511	int pri;
358		512
359	if (!q->size)	513	if (!q->size)
360	return 0;	514	return 0;
…		…
375	}	529	}
376		530
377	abort ();	531	abort ();
378	}	532	}
379		533
380	static void etp_thread_init (void)	534	static int ecb_cold
		535	etp_init (void (*want_poll)(void), void (*done_poll)(void))
381	{	536	{
382	X_MUTEX_CREATE (wrklock);	537	X_MUTEX_CREATE (wrklock);
383	X_MUTEX_CREATE (reslock);	538	X_MUTEX_CREATE (reslock);
384	X_MUTEX_CREATE (reqlock);	539	X_MUTEX_CREATE (reqlock);
385	X_COND_CREATE (reqwait);	540	X_COND_CREATE (reqwait);
386	}
387		541
388	static void etp_atfork_prepare (void)	542	reqq_init (&req_queue);
389	{	543	reqq_init (&res_queue);
390	X_LOCK (wrklock);
391	X_LOCK (reqlock);
392	X_LOCK (reslock);
393	#if !HAVE_PREADWRITE
394	X_LOCK (preadwritelock);
395	#endif
396	}
397		544
398	static void etp_atfork_parent (void)	545	wrk_first.next =
399	{	546	wrk_first.prev = &wrk_first;
400	#if !HAVE_PREADWRITE
401	X_UNLOCK (preadwritelock);
402	#endif
403	X_UNLOCK (reslock);
404	X_UNLOCK (reqlock);
405	X_UNLOCK (wrklock);
406	}
407
408	static void etp_atfork_child (void)
409	{
410	ETP_REQ *prv;
411
412	while ((prv = reqq_shift (&req_queue)))
413	ETP_DESTROY (prv);
414
415	while ((prv = reqq_shift (&res_queue)))
416	ETP_DESTROY (prv);
417
418	while (wrk_first.next != &wrk_first)
419	{
420	etp_worker *wrk = wrk_first.next;
421
422	if (wrk->req)
423	ETP_DESTROY (wrk->req);
424
425	etp_worker_clear (wrk);
426	etp_worker_free (wrk);
427	}
428		547
429	started = 0;	548	started = 0;
430	idle = 0;	549	idle = 0;
431	nreqs = 0;	550	nreqs = 0;
432	nready = 0;	551	nready = 0;
433	npending = 0;	552	npending = 0;
434		553
435	etp_thread_init ();
436	}
437
438	static void
439	etp_once_init (void)
440	{
441	etp_thread_init ();
442	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
443	}
444
445	static int
446	etp_init (void (*want_poll)(void), void (*done_poll)(void))
447	{
448	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
449
450	pthread_once (&doinit, etp_once_init);
451
452	want_poll_cb = want_poll;	554	want_poll_cb = want_poll;
453	done_poll_cb = done_poll;	555	done_poll_cb = done_poll;
454		556
455	return 0;	557	return 0;
456	}	558	}
457		559
458	X_THREAD_PROC (etp_proc);	560	X_THREAD_PROC (etp_proc);
459		561
		562	static void ecb_cold
460	static void etp_start_thread (void)	563	etp_start_thread (void)
461	{	564	{
462	etp_worker *wrk = calloc (1, sizeof (etp_worker));	565	etp_worker *wrk = calloc (1, sizeof (etp_worker));
463		566
464	/*TODO*/	567	/*TODO*/
465	assert (("unable to allocate worker thread data", wrk));	568	assert (("unable to allocate worker thread data", wrk));
…		…
478	free (wrk);	581	free (wrk);
479		582
480	X_UNLOCK (wrklock);	583	X_UNLOCK (wrklock);
481	}	584	}
482		585
		586	static void
483	static void etp_maybe_start_thread (void)	587	etp_maybe_start_thread (void)
484	{	588	{
485	if (expect_true (etp_nthreads () >= wanted))	589	if (ecb_expect_true (etp_nthreads () >= wanted))
486	return;	590	return;
487		591
488	/* todo: maybe use idle here, but might be less exact */	592	/* todo: maybe use idle here, but might be less exact */
489	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	593	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
490	return;	594	return;
491		595
492	etp_start_thread ();	596	etp_start_thread ();
493	}	597	}
494		598
		599	static void ecb_cold
495	static void etp_end_thread (void)	600	etp_end_thread (void)
496	{	601	{
497	eio_req *req = calloc (1, sizeof (eio_req));	602	eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
498		603
499	req->type = -1;	604	req->type = -1;
500	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	605	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
501		606
502	X_LOCK (reqlock);	607	X_LOCK (reqlock);
…		…
507	X_LOCK (wrklock);	612	X_LOCK (wrklock);
508	--started;	613	--started;
509	X_UNLOCK (wrklock);	614	X_UNLOCK (wrklock);
510	}	615	}
511		616
512	static int etp_poll (void)	617	static int
		618	etp_poll (void)
513	{	619	{
514	unsigned int maxreqs;	620	unsigned int maxreqs;
515	unsigned int maxtime;	621	unsigned int maxtime;
516	struct timeval tv_start, tv_now;	622	struct timeval tv_start, tv_now;
517		623
…		…
547		653
548	X_LOCK (reqlock);	654	X_LOCK (reqlock);
549	--nreqs;	655	--nreqs;
550	X_UNLOCK (reqlock);	656	X_UNLOCK (reqlock);
551		657
552	if (expect_false (req->type == EIO_GROUP && req->size))	658	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
553	{	659	{
554	req->int1 = 1; /* mark request as delayed */	660	req->int1 = 1; /* mark request as delayed */
555	continue;	661	continue;
556	}	662	}
557	else	663	else
558	{	664	{
559	int res = ETP_FINISH (req);	665	int res = ETP_FINISH (req);
560	if (expect_false (res))	666	if (ecb_expect_false (res))
561	return res;	667	return res;
562	}	668	}
563		669
564	if (expect_false (maxreqs && !--maxreqs))	670	if (ecb_expect_false (maxreqs && !--maxreqs))
565	break;	671	break;
566		672
567	if (maxtime)	673	if (maxtime)
568	{	674	{
569	gettimeofday (&tv_now, 0);	675	gettimeofday (&tv_now, 0);
…		…
575		681
576	errno = EAGAIN;	682	errno = EAGAIN;
577	return -1;	683	return -1;
578	}	684	}
579		685
		686	static void
580	static void etp_cancel (ETP_REQ *req)	687	etp_cancel (ETP_REQ *req)
581	{	688	{
582	X_LOCK (wrklock);	689	req->cancelled = 1;
583	req->flags |= EIO_FLAG_CANCELLED;
584	X_UNLOCK (wrklock);
585		690
586	eio_grp_cancel (req);	691	eio_grp_cancel (req);
587	}	692	}
588		693
		694	static void
589	static void etp_submit (ETP_REQ *req)	695	etp_submit (ETP_REQ *req)
590	{	696	{
591	req->pri -= ETP_PRI_MIN;	697	req->pri -= ETP_PRI_MIN;
592		698
593	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	699	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
594	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	700	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595		701
596	if (expect_false (req->type == EIO_GROUP))	702	if (ecb_expect_false (req->type == EIO_GROUP))
597	{	703	{
598	/* I hope this is worth it :/ */	704	/* I hope this is worth it :/ */
599	X_LOCK (reqlock);	705	X_LOCK (reqlock);
600	++nreqs;	706	++nreqs;
601	X_UNLOCK (reqlock);	707	X_UNLOCK (reqlock);
…		…
620		726
621	etp_maybe_start_thread ();	727	etp_maybe_start_thread ();
622	}	728	}
623	}	729	}
624		730
		731	static void ecb_cold
625	static void etp_set_max_poll_time (double nseconds)	732	etp_set_max_poll_time (double nseconds)
626	{	733	{
627	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	734	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628	max_poll_time = nseconds * EIO_TICKS;	735	max_poll_time = nseconds * EIO_TICKS;
629	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	736	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630	}	737	}
631		738
		739	static void ecb_cold
632	static void etp_set_max_poll_reqs (unsigned int maxreqs)	740	etp_set_max_poll_reqs (unsigned int maxreqs)
633	{	741	{
634	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	742	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635	max_poll_reqs = maxreqs;	743	max_poll_reqs = maxreqs;
636	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	744	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637	}	745	}
638		746
		747	static void ecb_cold
639	static void etp_set_max_idle (unsigned int nthreads)	748	etp_set_max_idle (unsigned int nthreads)
640	{	749	{
641	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	750	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642	max_idle = nthreads;	751	max_idle = nthreads;
643	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	752	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644	}	753	}
645		754
		755	static void ecb_cold
646	static void etp_set_idle_timeout (unsigned int seconds)	756	etp_set_idle_timeout (unsigned int seconds)
647	{	757	{
648	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	758	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649	idle_timeout = seconds;	759	idle_timeout = seconds;
650	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	760	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651	}	761	}
652		762
		763	static void ecb_cold
653	static void etp_set_min_parallel (unsigned int nthreads)	764	etp_set_min_parallel (unsigned int nthreads)
654	{	765	{
655	if (wanted < nthreads)	766	if (wanted < nthreads)
656	wanted = nthreads;	767	wanted = nthreads;
657	}	768	}
658		769
		770	static void ecb_cold
659	static void etp_set_max_parallel (unsigned int nthreads)	771	etp_set_max_parallel (unsigned int nthreads)
660	{	772	{
661	if (wanted > nthreads)	773	if (wanted > nthreads)
662	wanted = nthreads;	774	wanted = nthreads;
663		775
664	while (started > wanted)	776	while (started > wanted)
665	etp_end_thread ();	777	etp_end_thread ();
666	}	778	}
667		779
668	/*****************************************************************************/	780	/*****************************************************************************/
669		781
		782	static void
670	static void grp_try_feed (eio_req *grp)	783	grp_try_feed (eio_req *grp)
671	{	784	{
672	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	785	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673	{	786	{
674	grp->flags &= ~EIO_FLAG_GROUPADD;	787	grp->flags &= ~EIO_FLAG_GROUPADD;
675		788
…		…
682	break;	795	break;
683	}	796	}
684	}	797	}
685	}	798	}
686		799
		800	static int
687	static int grp_dec (eio_req *grp)	801	grp_dec (eio_req *grp)
688	{	802	{
689	--grp->size;	803	--grp->size;
690		804
691	/* call feeder, if applicable */	805	/* call feeder, if applicable */
692	grp_try_feed (grp);	806	grp_try_feed (grp);
…		…
696	return eio_finish (grp);	810	return eio_finish (grp);
697	else	811	else
698	return 0;	812	return 0;
699	}	813	}
700		814
		815	static void
701	void eio_destroy (eio_req *req)	816	eio_destroy (eio_req *req)
702	{	817	{
703	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	818	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	819	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705		820
706	EIO_DESTROY (req);	821	EIO_DESTROY (req);
707	}	822	}
708		823
		824	static int
709	static int eio_finish (eio_req *req)	825	eio_finish (eio_req *req)
710	{	826	{
711	int res = EIO_FINISH (req);	827	int res = EIO_FINISH (req);
712		828
713	if (req->grp)	829	if (req->grp)
714	{	830	{
…		…
722	if (grp->grp_first == req)	838	if (grp->grp_first == req)
723	grp->grp_first = req->grp_next;	839	grp->grp_first = req->grp_next;
724		840
725	res2 = grp_dec (grp);	841	res2 = grp_dec (grp);
726		842
727	if (!res && res2)	843	if (!res)
728	res = res2;	844	res = res2;
729	}	845	}
730		846
731	eio_destroy (req);	847	eio_destroy (req);
732		848
733	return res;	849	return res;
734	}	850	}
735		851
		852	void
736	void eio_grp_cancel (eio_req *grp)	853	eio_grp_cancel (eio_req *grp)
737	{	854	{
738	for (grp = grp->grp_first; grp; grp = grp->grp_next)	855	for (grp = grp->grp_first; grp; grp = grp->grp_next)
739	eio_cancel (grp);	856	eio_cancel (grp);
740	}	857	}
741		858
		859	void
742	void eio_cancel (eio_req *req)	860	eio_cancel (eio_req *req)
743	{	861	{
744	etp_cancel (req);	862	etp_cancel (req);
745	}	863	}
746		864
		865	void
747	void eio_submit (eio_req *req)	866	eio_submit (eio_req *req)
748	{	867	{
749	etp_submit (req);	868	etp_submit (req);
750	}	869	}
751		870
752	unsigned int eio_nreqs (void)	871	unsigned int
		872	eio_nreqs (void)
753	{	873	{
754	return etp_nreqs ();	874	return etp_nreqs ();
755	}	875	}
756		876
757	unsigned int eio_nready (void)	877	unsigned int
		878	eio_nready (void)
758	{	879	{
759	return etp_nready ();	880	return etp_nready ();
760	}	881	}
761		882
762	unsigned int eio_npending (void)	883	unsigned int
		884	eio_npending (void)
763	{	885	{
764	return etp_npending ();	886	return etp_npending ();
765	}	887	}
766		888
767	unsigned int eio_nthreads (void)	889	unsigned int ecb_cold
		890	eio_nthreads (void)
768	{	891	{
769	return etp_nthreads ();	892	return etp_nthreads ();
770	}	893	}
771		894
		895	void ecb_cold
772	void eio_set_max_poll_time (double nseconds)	896	eio_set_max_poll_time (double nseconds)
773	{	897	{
774	etp_set_max_poll_time (nseconds);	898	etp_set_max_poll_time (nseconds);
775	}	899	}
776		900
		901	void ecb_cold
777	void eio_set_max_poll_reqs (unsigned int maxreqs)	902	eio_set_max_poll_reqs (unsigned int maxreqs)
778	{	903	{
779	etp_set_max_poll_reqs (maxreqs);	904	etp_set_max_poll_reqs (maxreqs);
780	}	905	}
781		906
		907	void ecb_cold
782	void eio_set_max_idle (unsigned int nthreads)	908	eio_set_max_idle (unsigned int nthreads)
783	{	909	{
784	etp_set_max_idle (nthreads);	910	etp_set_max_idle (nthreads);
785	}	911	}
786		912
		913	void ecb_cold
787	void eio_set_idle_timeout (unsigned int seconds)	914	eio_set_idle_timeout (unsigned int seconds)
788	{	915	{
789	etp_set_idle_timeout (seconds);	916	etp_set_idle_timeout (seconds);
790	}	917	}
791		918
		919	void ecb_cold
792	void eio_set_min_parallel (unsigned int nthreads)	920	eio_set_min_parallel (unsigned int nthreads)
793	{	921	{
794	etp_set_min_parallel (nthreads);	922	etp_set_min_parallel (nthreads);
795	}	923	}
796		924
		925	void ecb_cold
797	void eio_set_max_parallel (unsigned int nthreads)	926	eio_set_max_parallel (unsigned int nthreads)
798	{	927	{
799	etp_set_max_parallel (nthreads);	928	etp_set_max_parallel (nthreads);
800	}	929	}
801		930
802	int eio_poll (void)	931	int eio_poll (void)
…		…
804	return etp_poll ();	933	return etp_poll ();
805	}	934	}
806		935
807	/*****************************************************************************/	936	/*****************************************************************************/
808	/* work around various missing functions */	937	/* work around various missing functions */
809
810	#if !HAVE_PREADWRITE
811	# undef pread
812	# undef pwrite
813	# define pread eio__pread
814	# define pwrite eio__pwrite
815
816	static ssize_t
817	eio__pread (int fd, void *buf, size_t count, off_t offset)
818	{
819	ssize_t res;
820	off_t ooffset;
821
822	X_LOCK (preadwritelock);
823	ooffset = lseek (fd, 0, SEEK_CUR);
824	lseek (fd, offset, SEEK_SET);
825	res = read (fd, buf, count);
826	lseek (fd, ooffset, SEEK_SET);
827	X_UNLOCK (preadwritelock);
828
829	return res;
830	}
831
832	static ssize_t
833	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834	{
835	ssize_t res;
836	off_t ooffset;
837
838	X_LOCK (preadwritelock);
839	ooffset = lseek (fd, 0, SEEK_CUR);
840	lseek (fd, offset, SEEK_SET);
841	res = write (fd, buf, count);
842	lseek (fd, ooffset, SEEK_SET);
843	X_UNLOCK (preadwritelock);
844
845	return res;
846	}
847	#endif
848		938
849	#ifndef HAVE_UTIMES	939	#ifndef HAVE_UTIMES
850		940
851	# undef utimes	941	# undef utimes
852	# define utimes(path,times) eio__utimes (path, times)	942	# define utimes(path,times) eio__utimes (path, times)
…		…
872	#ifndef HAVE_FUTIMES	962	#ifndef HAVE_FUTIMES
873		963
874	# undef futimes	964	# undef futimes
875	# define futimes(fd,times) eio__futimes (fd, times)	965	# define futimes(fd,times) eio__futimes (fd, times)
876		966
		967	static int
877	static int eio__futimes (int fd, const struct timeval tv[2])	968	eio__futimes (int fd, const struct timeval tv[2])
878	{	969	{
879	errno = ENOSYS;	970	errno = ENOSYS;
880	return -1;	971	return -1;
881	}	972	}
882		973
…		…
885	#if !HAVE_FDATASYNC	976	#if !HAVE_FDATASYNC
886	# undef fdatasync	977	# undef fdatasync
887	# define fdatasync(fd) fsync (fd)	978	# define fdatasync(fd) fsync (fd)
888	#endif	979	#endif
889		980
		981	static int
		982	eio__syncfs (int fd)
		983	{
		984	int res;
		985
		986	#if HAVE_SYS_SYNCFS
		987	res = (int)syscall (__NR_syncfs, (int)(fd));
		988	#else
		989	res = EIO_ENOSYS ();
		990	#endif
		991
		992	if (res < 0 && errno == ENOSYS && fd >= 0)
		993	sync ();
		994
		995	return res;
		996	}
		997
890	/* sync_file_range always needs emulation */	998	/* sync_file_range always needs emulation */
891	int	999	static int
892	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	1000	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893	{	1001	{
894	#if HAVE_SYNC_FILE_RANGE	1002	#if HAVE_SYNC_FILE_RANGE
895	int res;	1003	int res;
896		1004
…		…
913	/* even though we could play tricks with the flags, it's better to always	1021	/* even though we could play tricks with the flags, it's better to always
914	* call fdatasync, as that matches the expectation of its users best */	1022	* call fdatasync, as that matches the expectation of its users best */
915	return fdatasync (fd);	1023	return fdatasync (fd);
916	}	1024	}
917		1025
		1026	static int
		1027	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		1028	{
		1029	#if HAVE_LINUX_FALLOCATE
		1030	return fallocate (fd, mode, offset, len);
		1031	#else
		1032	return EIO_ENOSYS ();
		1033	#endif
		1034	}
		1035
918	#if !HAVE_READAHEAD	1036	#if !HAVE_READAHEAD
919	# undef readahead	1037	# undef readahead
920	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1038	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921		1039
922	static ssize_t	1040	static eio_ssize_t
923	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1041	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
924	{	1042	{
925	size_t todo = count;	1043	size_t todo = count;
926	dBUF;	1044	dBUF;
927		1045
…		…
932	pread (fd, eio_buf, len, offset);	1050	pread (fd, eio_buf, len, offset);
933	offset += len;	1051	offset += len;
934	todo -= len;	1052	todo -= len;
935	}	1053	}
936		1054
937	errno = 0;	1055	FUBd;
		1056
		1057	/* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
		1058	/* but not for e.g. EIO or eof, so we also never fail */
938	return count;	1059	return 0;
939	}	1060	}
940		1061
941	#endif	1062	#endif
942		1063
943	/* sendfile always needs emulation */	1064	/* sendfile always needs emulation */
944	static ssize_t	1065	static eio_ssize_t
945	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1066	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
946	{	1067	{
		1068	eio_ssize_t written = 0;
947	ssize_t res;	1069	eio_ssize_t res;
948		1070
949	if (!count)	1071	if (!count)
950	return 0;	1072	return 0;
951		1073
		1074	for (;;)
		1075	{
		1076	#ifdef __APPLE__
		1077	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1078	#endif
952	#if HAVE_SENDFILE	1079	#if HAVE_SENDFILE
953	# if __linux	1080	# if __linux
		1081	off_t soffset = offset;
954	res = sendfile (ofd, ifd, &offset, count);	1082	res = sendfile (ofd, ifd, &soffset, count);
955		1083
956	# elif __FreeBSD__	1084	# elif __FreeBSD__
957	/*	1085	/*
958	* Of course, the freebsd sendfile is a dire hack with no thoughts	1086	* Of course, the freebsd sendfile is a dire hack with no thoughts
959	* wasted on making it similar to other I/O functions.	1087	* wasted on making it similar to other I/O functions.
960	*/	1088	*/
961	{
962	off_t sbytes;	1089	off_t sbytes;
963	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1090	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
964		1091
965	#if 0 /* according to the manpage, this is correct, but broken behaviour */	1092	#if 0 /* according to the manpage, this is correct, but broken behaviour */
966	/* freebsd' sendfile will return 0 on success */	1093	/* freebsd' sendfile will return 0 on success */
967	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */	1094	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
968	/* not on e.g. EIO or EPIPE - sounds broken */	1095	/* not on e.g. EIO or EPIPE - sounds broken */
969	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)	1096	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
970	res = sbytes;	1097	res = sbytes;
971	#endif	1098	#endif
972		1099
973	/* according to source inspection, this is correct, and useful behaviour */	1100	/* according to source inspection, this is correct, and useful behaviour */
974	if (sbytes)	1101	if (sbytes)
975	res = sbytes;	1102	res = sbytes;
976	}
977		1103
978	# elif defined (__APPLE__)	1104	# elif defined __APPLE__
979
980	{
981	off_t sbytes = count;	1105	off_t sbytes = count;
982	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	1106	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983		1107
984	/* according to the manpage, sbytes is always valid */	1108	/* according to the manpage, sbytes is always valid */
985	if (sbytes)	1109	if (sbytes)
986	res = sbytes;	1110	res = sbytes;
987	}
988		1111
989	# elif __hpux	1112	# elif __hpux
990	res = sendfile (ofd, ifd, offset, count, 0, 0);	1113	res = sendfile (ofd, ifd, offset, count, 0, 0);
991		1114
992	# elif __solaris	1115	# elif __solaris
993	{
994	struct sendfilevec vec;	1116	struct sendfilevec vec;
995	size_t sbytes;	1117	size_t sbytes;
996		1118
997	vec.sfv_fd = ifd;	1119	vec.sfv_fd = ifd;
998	vec.sfv_flag = 0;	1120	vec.sfv_flag = 0;
999	vec.sfv_off = offset;	1121	vec.sfv_off = offset;
1000	vec.sfv_len = count;	1122	vec.sfv_len = count;
1001		1123
1002	res = sendfilev (ofd, &vec, 1, &sbytes);	1124	res = sendfilev (ofd, &vec, 1, &sbytes);
1003		1125
1004	if (res < 0 && sbytes)	1126	if (res < 0 && sbytes)
1005	res = sbytes;	1127	res = sbytes;
1006	}
1007		1128
1008	# endif	1129	# endif
1009		1130
1010	#elif defined (_WIN32)	1131	#elif defined (_WIN32) && 0
1011
1012	/* does not work, just for documentation of what would need to be done */	1132	/* does not work, just for documentation of what would need to be done */
1013	{	1133	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1134	/* libeio guarantees that the file offset does not change, and windows */
		1135	/* has no way to get an independent handle to the same file description */
1014	HANDLE h = TO_SOCKET (ifd);	1136	HANDLE h = TO_SOCKET (ifd);
1015	SetFilePointer (h, offset, 0, FILE_BEGIN);	1137	SetFilePointer (h, offset, 0, FILE_BEGIN);
1016	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);	1138	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1017	}
1018		1139
1019	#else	1140	#else
1020	res = -1;	1141	res = EIO_ENOSYS ();
1021	errno = ENOSYS;
1022	#endif	1142	#endif
1023		1143
		1144	/* we assume sendfile can copy at least 128mb in one go */
		1145	if (res <= 128 * 1024 * 1024)
		1146	{
		1147	if (res > 0)
		1148	written += res;
		1149
		1150	if (written)
		1151	return written;
		1152
		1153	break;
		1154	}
		1155	else
		1156	{
		1157	/* if we requested more, then probably the kernel was lazy */
		1158	written += res;
		1159	offset += res;
		1160	count -= res;
		1161
		1162	if (!count)
		1163	return written;
		1164	}
		1165	}
		1166
1024	if (res < 0	1167	if (res < 0
1025	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1168	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1026	/* BSDs */	1169	/* BSDs */
1027	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1170	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1028	|| errno == ENOTSUP	1171	|| errno == ENOTSUP
1029	#endif	1172	#endif
		1173	#ifdef EOPNOTSUPP /* windows */
1030	|| errno == EOPNOTSUPP /* BSDs */	1174	|| errno == EOPNOTSUPP /* BSDs */
		1175	#endif
1031	#if __solaris	1176	#if __solaris
1032	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1177	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1033	#endif	1178	#endif
1034	)	1179	)
1035	)	1180	)
…		…
1039		1184
1040	res = 0;	1185	res = 0;
1041		1186
1042	while (count)	1187	while (count)
1043	{	1188	{
1044	ssize_t cnt;	1189	eio_ssize_t cnt;
1045		1190
1046	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1191	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1047		1192
1048	if (cnt <= 0)	1193	if (cnt <= 0)
1049	{	1194	{
…		…
1061		1206
1062	offset += cnt;	1207	offset += cnt;
1063	res += cnt;	1208	res += cnt;
1064	count -= cnt;	1209	count -= cnt;
1065	}	1210	}
		1211
		1212	FUBd;
1066	}	1213	}
1067		1214
1068	return res;	1215	return res;
		1216	}
		1217
		1218	#ifdef PAGESIZE
		1219	# define eio_pagesize() PAGESIZE
		1220	#else
		1221	static intptr_t
		1222	eio_pagesize (void)
		1223	{
		1224	static intptr_t page;
		1225
		1226	if (!page)
		1227	page = sysconf (_SC_PAGESIZE);
		1228
		1229	return page;
		1230	}
		1231	#endif
		1232
		1233	static void
		1234	eio_page_align (void **addr, size_t *length)
		1235	{
		1236	intptr_t mask = eio_pagesize () - 1;
		1237
		1238	/* round down addr */
		1239	intptr_t adj = mask & (intptr_t)*addr;
		1240
		1241	*addr = (void *)((intptr_t)*addr - adj);
		1242	*length += adj;
		1243
		1244	/* round up length */
		1245	*length = (*length + mask) & ~mask;
		1246	}
		1247
		1248	#if !_POSIX_MEMLOCK
		1249	# define eio__mlockall(a) EIO_ENOSYS ()
		1250	#else
		1251
		1252	static int
		1253	eio__mlockall (int flags)
		1254	{
		1255	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1256	extern int mallopt (int, int);
		1257	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1258	#endif
		1259
		1260	if (EIO_MCL_CURRENT != MCL_CURRENT
		1261	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1262	{
		1263	flags = 0
		1264	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1265	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1266	}
		1267
		1268	return mlockall (flags);
		1269	}
		1270	#endif
		1271
		1272	#if !_POSIX_MEMLOCK_RANGE
		1273	# define eio__mlock(a,b) EIO_ENOSYS ()
		1274	#else
		1275
		1276	static int
		1277	eio__mlock (void *addr, size_t length)
		1278	{
		1279	eio_page_align (&addr, &length);
		1280
		1281	return mlock (addr, length);
		1282	}
		1283
		1284	#endif
		1285
		1286	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1287	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1288	#else
		1289
		1290	static int
		1291	eio__msync (void *mem, size_t len, int flags)
		1292	{
		1293	eio_page_align (&mem, &len);
		1294
		1295	if (EIO_MS_ASYNC != MS_SYNC
		1296	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1297	|| EIO_MS_SYNC != MS_SYNC)
		1298	{
		1299	flags = 0
		1300	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1301	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1302	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1303	}
		1304
		1305	return msync (mem, len, flags);
		1306	}
		1307
		1308	#endif
		1309
		1310	static int
		1311	eio__mtouch (eio_req *req)
		1312	{
		1313	void *mem = req->ptr2;
		1314	size_t len = req->size;
		1315	int flags = req->int1;
		1316
		1317	eio_page_align (&mem, &len);
		1318
		1319	{
		1320	intptr_t addr = (intptr_t)mem;
		1321	intptr_t end = addr + len;
		1322	intptr_t page = eio_pagesize ();
		1323
		1324	if (addr < end)
		1325	if (flags & EIO_MT_MODIFY) /* modify */
		1326	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1327	else
		1328	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1329	}
		1330
		1331	return 0;
		1332	}
		1333
		1334	/*****************************************************************************/
		1335	/* requests implemented outside eio_execute, because they are so large */
		1336
		1337	static void
		1338	eio__lseek (eio_req *req)
		1339	{
		1340	/* this usually gets optimised away completely, or your compiler sucks, */
		1341	/* or the whence constants really are not 0, 1, 2 */
		1342	int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
		1343	: req->int2 == EIO_SEEK_CUR ? SEEK_CUR
		1344	: req->int2 == EIO_SEEK_END ? SEEK_END
		1345	: req->int2;
		1346
		1347	req->offs = lseek (req->int1, req->offs, whence);
		1348	req->result = req->offs == (off_t)-1 ? -1 : 0;
		1349	}
		1350
		1351	/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
		1352	static int
		1353	eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1354	{
		1355	const char *rel = path;
		1356	char *res;
		1357	char *tmp1, *tmp2;
		1358	#if SYMLOOP_MAX > 32
		1359	int symlinks = SYMLOOP_MAX;
		1360	#else
		1361	int symlinks = 32;
		1362	#endif
		1363
		1364	errno = EINVAL;
		1365	if (!rel)
		1366	return -1;
		1367
		1368	errno = ENOENT;
		1369	if (!*rel)
		1370	return -1;
		1371
		1372	res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
		1373	tmp1 = res + PATH_MAX;
		1374	tmp2 = tmp1 + PATH_MAX;
		1375
		1376	#if 0 /* disabled, the musl way to do things is just too racy */
		1377	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1378	/* on linux we may be able to ask the kernel */
		1379	{
		1380	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1381
		1382	if (fd >= 0)
		1383	{
		1384	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1385	req->result = readlink (tmp1, res, PATH_MAX);
		1386	close (fd);
		1387
		1388	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1389
		1390	if (req->result > 0)
		1391	goto done;
		1392	}
		1393	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1394	return;
		1395	}
		1396	#endif
		1397	#endif
		1398
		1399	if (*rel != '/')
		1400	{
		1401	int len;
		1402
		1403	errno = ENOENT;
		1404	if (wd == EIO_INVALID_WD)
		1405	return -1;
		1406
		1407	if (wd == EIO_CWD)
		1408	{
		1409	if (!getcwd (res, PATH_MAX))
		1410	return -1;
		1411
		1412	len = strlen (res);
		1413	}
		1414	else
		1415	memcpy (res, wd->str, len = wd->len);
		1416
		1417	if (res [1]) /* only use if not / */
		1418	res += len;
		1419	}
		1420
		1421	while (*rel)
		1422	{
		1423	eio_ssize_t len, linklen;
		1424	const char *beg = rel;
		1425
		1426	while (*rel && *rel != '/')
		1427	++rel;
		1428
		1429	len = rel - beg;
		1430
		1431	if (!len) /* skip slashes */
		1432	{
		1433	++rel;
		1434	continue;
		1435	}
		1436
		1437	if (beg [0] == '.')
		1438	{
		1439	if (len == 1)
		1440	continue; /* . - nop */
		1441
		1442	if (beg [1] == '.' && len == 2)
		1443	{
		1444	/* .. - back up one component, if possible */
		1445
		1446	while (res != tmpbuf->ptr)
		1447	if (*--res == '/')
		1448	break;
		1449
		1450	continue;
		1451	}
		1452	}
		1453
		1454	errno = ENAMETOOLONG;
		1455	if (res + 1 + len + 1 >= tmp1)
		1456	return -1;
		1457
		1458	/* copy one component */
		1459	*res = '/';
		1460	memcpy (res + 1, beg, len);
		1461
		1462	/* zero-terminate, for readlink */
		1463	res [len + 1] = 0;
		1464
		1465	/* now check if it's a symlink */
		1466	linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
		1467
		1468	if (linklen < 0)
		1469	{
		1470	if (errno != EINVAL)
		1471	return -1;
		1472
		1473	/* it's a normal directory. hopefully */
		1474	res += len + 1;
		1475	}
		1476	else
		1477	{
		1478	/* yay, it was a symlink - build new path in tmp2 */
		1479	int rellen = strlen (rel);
		1480
		1481	errno = ENAMETOOLONG;
		1482	if (linklen + 1 + rellen >= PATH_MAX)
		1483	return -1;
		1484
		1485	errno = ELOOP;
		1486	if (!--symlinks)
		1487	return -1;
		1488
		1489	if (*tmp1 == '/')
		1490	res = tmpbuf->ptr; /* symlink resolves to an absolute path */
		1491
		1492	/* we need to be careful, as rel might point into tmp2 already */
		1493	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1494	tmp2 [linklen] = '/';
		1495	memcpy (tmp2, tmp1, linklen);
		1496
		1497	rel = tmp2;
		1498	}
		1499	}
		1500
		1501	/* special case for the lone root path */
		1502	if (res == tmpbuf->ptr)
		1503	*res++ = '/';
		1504
		1505	return res - (char *)tmpbuf->ptr;
1069	}	1506	}
1070		1507
1071	static signed char	1508	static signed char
1072	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1509	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1073	{	1510	{
…		…
1081		1518
1082	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1519	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1083	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1520	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1084		1521
1085	static void	1522	static void
1086	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1523	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1087	{	1524	{
1088	unsigned char bits [9 + sizeof (ino_t) * 8];	1525	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1089	unsigned char *bit = bits;	1526	unsigned char *bit = bits;
1090		1527
1091	assert (CHAR_BIT == 8);	1528	assert (CHAR_BIT == 8);
1092	assert (sizeof (eio_dirent) * 8 < 256);	1529	assert (sizeof (eio_dirent) * 8 < 256);
1093	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1530	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1095		1532
1096	if (size <= EIO_SORT_FAST)	1533	if (size <= EIO_SORT_FAST)
1097	return;	1534	return;
1098		1535
1099	/* first prepare an array of bits to test in our radix sort */	1536	/* first prepare an array of bits to test in our radix sort */
1100	/* try to take endianness into account, as well as differences in ino_t sizes */	1537	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1101	/* inode_bits must contain all inodes ORed together */	1538	/* inode_bits must contain all inodes ORed together */
1102	/* which is used to skip bits that are 0 everywhere, which is very common */	1539	/* which is used to skip bits that are 0 everywhere, which is very common */
1103	{	1540	{
1104	ino_t endianness;	1541	eio_ino_t endianness;
1105	int i, j;	1542	int i, j;
1106		1543
1107	/* we store the byte offset of byte n into byte n of "endianness" */	1544	/* we store the byte offset of byte n into byte n of "endianness" */
1108	for (i = 0; i < sizeof (ino_t); ++i)	1545	for (i = 0; i < sizeof (eio_ino_t); ++i)
1109	((unsigned char *)&endianness)[i] = i;	1546	((unsigned char *)&endianness)[i] = i;
1110		1547
1111	*bit++ = 0;	1548	*bit++ = 0;
1112		1549
1113	for (i = 0; i < sizeof (ino_t); ++i)	1550	for (i = 0; i < sizeof (eio_ino_t); ++i)
1114	{	1551	{
1115	/* shifting off the byte offsets out of "endianness" */	1552	/* shifting off the byte offsets out of "endianness" */
1116	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1553	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1117	endianness >>= 8;	1554	endianness >>= 8;
1118		1555
1119	for (j = 0; j < 8; ++j)	1556	for (j = 0; j < 8; ++j)
1120	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1557	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1121	*bit++ = offs + j;	1558	*bit++ = offs + j;
1122	}	1559	}
1123		1560
1124	for (j = 0; j < 8; ++j)	1561	for (j = 0; j < 8; ++j)
1125	if (score_bits & (1 << j))	1562	if (score_bits & (1 << j))
1126	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1563	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1127	}	1564	}
1128		1565
1129	/* now actually do the sorting (a variant of MSD radix sort) */	1566	/* now actually do the sorting (a variant of MSD radix sort) */
1130	{	1567	{
1131	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1568	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1132	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1569	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1133	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1570	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1134	int stk_idx = 0;	1571	int stk_idx = 0;
1135		1572
1136	base_stk [stk_idx] = dents;	1573	base_stk [stk_idx] = dents;
1137	end_stk [stk_idx] = dents + size;	1574	end_stk [stk_idx] = dents + size;
1138	bit_stk [stk_idx] = bit - 1;	1575	bit_stk [stk_idx] = bit - 1;
…		…
1217	}	1654	}
1218	}	1655	}
1219	}	1656	}
1220		1657
1221	static void	1658	static void
1222	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1659	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1223	{	1660	{
1224	if (size <= 1)	1661	if (size <= 1)
1225	return; /* our insertion sort relies on size > 0 */	1662	return; /* our insertion sort relies on size > 0 */
1226		1663
1227	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1664	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1235		1672
1236	/* read a full directory */	1673	/* read a full directory */
1237	static void	1674	static void
1238	eio__scandir (eio_req *req, etp_worker *self)	1675	eio__scandir (eio_req *req, etp_worker *self)
1239	{	1676	{
1240	DIR *dirp;
1241	EIO_STRUCT_DIRENT *entp;
1242	char *name, *names;	1677	char *name, *names;
1243	int namesalloc = 4096;	1678	int namesalloc = 4096 - sizeof (void *) * 4;
1244	int namesoffs = 0;	1679	int namesoffs = 0;
1245	int flags = req->int1;	1680	int flags = req->int1;
1246	eio_dirent *dents = 0;	1681	eio_dirent *dents = 0;
1247	int dentalloc = 128;	1682	int dentalloc = 128;
1248	int dentoffs = 0;	1683	int dentoffs = 0;
1249	ino_t inode_bits = 0;	1684	eio_ino_t inode_bits = 0;
		1685	#ifdef _WIN32
		1686	HANDLE dirp;
		1687	WIN32_FIND_DATA entp;
		1688	#else
		1689	DIR *dirp;
		1690	EIO_STRUCT_DIRENT *entp;
		1691	#endif
1250		1692
1251	req->result = -1;	1693	req->result = -1;
1252		1694
1253	if (!(flags & EIO_READDIR_DENTS))	1695	if (!(flags & EIO_READDIR_DENTS))
1254	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1696	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1255		1697
1256	X_LOCK (wrklock);	1698	#ifdef _WIN32
1257	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1699	{
		1700	int len = strlen ((const char *)req->ptr1);
		1701	char *path = malloc (MAX_PATH);
		1702	const char *fmt;
		1703	const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		1704
		1705	if (!len)
		1706	fmt = "./*";
		1707	else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
		1708	fmt = "%s*";
		1709	else
		1710	fmt = "%s/*";
		1711
		1712	_snprintf (path, MAX_PATH, fmt, reqpath);
		1713	dirp = FindFirstFile (path, &entp);
		1714	free (path);
		1715
		1716	if (dirp == INVALID_HANDLE_VALUE)
		1717	{
		1718	/* should steal _dosmaperr */
		1719	switch (GetLastError ())
		1720	{
		1721	case ERROR_FILE_NOT_FOUND:
		1722	req->result = 0;
		1723	break;
		1724
		1725	case ERROR_INVALID_NAME:
		1726	case ERROR_PATH_NOT_FOUND:
		1727	case ERROR_NO_MORE_FILES:
		1728	errno = ENOENT;
		1729	break;
		1730
		1731	case ERROR_NOT_ENOUGH_MEMORY:
		1732	errno = ENOMEM;
		1733	break;
		1734
		1735	default:
		1736	errno = EINVAL;
		1737	break;
		1738	}
		1739
		1740	return;
		1741	}
		1742	}
		1743	#else
		1744	#if HAVE_AT
		1745	if (req->wd)
		1746	{
		1747	int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1748
		1749	if (fd < 0)
		1750	return;
		1751
		1752	dirp = fdopendir (fd);
		1753
		1754	if (!dirp)
		1755	close (fd);
		1756	}
		1757	else
1258	self->dirp = dirp = opendir (req->ptr1);	1758	dirp = opendir (req->ptr1);
		1759	#else
		1760	dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
		1761	#endif
		1762
		1763	if (!dirp)
		1764	return;
		1765	#endif
		1766
		1767	if (req->flags & EIO_FLAG_PTR1_FREE)
		1768	free (req->ptr1);
1259		1769
1260	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1770	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1261	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1771	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1262	req->ptr2 = names = malloc (namesalloc);	1772	req->ptr2 = names = malloc (namesalloc);
1263	X_UNLOCK (wrklock);
1264		1773
1265	if (dirp && names && (!flags || dents))	1774	if (!names || (flags && !dents))
		1775	return;
		1776
1266	for (;;)	1777	for (;;)
1267	{	1778	{
		1779	int done;
		1780
		1781	#ifdef _WIN32
		1782	done = !dirp;
		1783	#else
1268	errno = 0;	1784	errno = 0;
1269	entp = readdir (dirp);	1785	entp = readdir (dirp);
		1786	done = !entp;
		1787	#endif
1270		1788
1271	if (!entp)	1789	if (done)
1272	{	1790	{
		1791	#ifndef _WIN32
		1792	int old_errno = errno;
		1793	closedir (dirp);
		1794	errno = old_errno;
		1795
1273	if (errno)	1796	if (errno)
1274	break;	1797	break;
		1798	#endif
1275		1799
1276	/* sort etc. */	1800	/* sort etc. */
1277	req->int1 = flags;	1801	req->int1 = flags;
1278	req->result = dentoffs;	1802	req->result = dentoffs;
1279		1803
1280	if (flags & EIO_READDIR_STAT_ORDER)	1804	if (flags & EIO_READDIR_STAT_ORDER)
1281	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);	1805	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1282	else if (flags & EIO_READDIR_DIRS_FIRST)	1806	else if (flags & EIO_READDIR_DIRS_FIRST)
1283	if (flags & EIO_READDIR_FOUND_UNKNOWN)	1807	if (flags & EIO_READDIR_FOUND_UNKNOWN)
1284	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */	1808	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1285	else	1809	else
		1810	{
		1811	/* in this case, all is known, and we just put dirs first and sort them */
		1812	eio_dirent *oth = dents + dentoffs;
		1813	eio_dirent *dir = dents;
		1814
		1815	/* now partition dirs to the front, and non-dirs to the back */
		1816	/* by walking from both sides and swapping if necessary */
		1817	while (oth > dir)
		1818	{
		1819	if (dir->type == EIO_DT_DIR)
		1820	++dir;
		1821	else if ((--oth)->type == EIO_DT_DIR)
		1822	{
		1823	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1824
		1825	++dir;
		1826	}
		1827	}
		1828
		1829	/* now sort the dirs only (dirs all have the same score) */
		1830	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1831	}
		1832
		1833	break;
		1834	}
		1835
		1836	/* now add the entry to our list(s) */
		1837	name = D_NAME (entp);
		1838
		1839	/* skip . and .. entries */
		1840	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
		1841	{
		1842	int len = D_NAMLEN (entp) + 1;
		1843
		1844	while (ecb_expect_false (namesoffs + len > namesalloc))
		1845	{
		1846	namesalloc *= 2;
		1847	req->ptr2 = names = realloc (names, namesalloc);
		1848
		1849	if (!names)
		1850	break;
		1851	}
		1852
		1853	memcpy (names + namesoffs, name, len);
		1854
		1855	if (dents)
		1856	{
		1857	struct eio_dirent *ent;
		1858
		1859	if (ecb_expect_false (dentoffs == dentalloc))
1286	{	1860	{
1287	/* in this case, all is known, and we just put dirs first and sort them */	1861	dentalloc *= 2;
		1862	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1863
		1864	if (!dents)
		1865	break;
		1866	}
		1867
1288	eio_dirent *oth = dents + dentoffs;	1868	ent = dents + dentoffs;
1289	eio_dirent *dir = dents;
1290		1869
1291	/* now partition dirs to the front, and non-dirs to the back */	1870	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1292	/* by walking from both sides and swapping if necessary */	1871	ent->namelen = len - 1;
1293	while (oth > dir)	1872	ent->inode = D_INO (entp);
		1873
		1874	inode_bits |= ent->inode;
		1875
		1876	switch (D_TYPE (entp))
		1877	{
		1878	default:
		1879	ent->type = EIO_DT_UNKNOWN;
		1880	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1881	break;
		1882
		1883	#ifdef DT_FIFO
		1884	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1885	#endif
		1886	#ifdef DT_CHR
		1887	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1888	#endif
		1889	#ifdef DT_MPC
		1890	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1891	#endif
		1892	#ifdef DT_DIR
		1893	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1894	#endif
		1895	#ifdef DT_NAM
		1896	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1897	#endif
		1898	#ifdef DT_BLK
		1899	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1900	#endif
		1901	#ifdef DT_MPB
		1902	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1903	#endif
		1904	#ifdef DT_REG
		1905	case DT_REG: ent->type = EIO_DT_REG; break;
		1906	#endif
		1907	#ifdef DT_NWK
		1908	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1909	#endif
		1910	#ifdef DT_CMP
		1911	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1912	#endif
		1913	#ifdef DT_LNK
		1914	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1915	#endif
		1916	#ifdef DT_SOCK
		1917	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1918	#endif
		1919	#ifdef DT_DOOR
		1920	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1921	#endif
		1922	#ifdef DT_WHT
		1923	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1924	#endif
		1925	}
		1926
		1927	ent->score = 7;
		1928
		1929	if (flags & EIO_READDIR_DIRS_FIRST)
		1930	{
		1931	if (ent->type == EIO_DT_UNKNOWN)
1294	{	1932	{
1295	if (dir->type == EIO_DT_DIR)	1933	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1296	++dir;	1934	ent->score = 1;
1297	else if ((--oth)->type == EIO_DT_DIR)	1935	else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1298	{	1936	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1299	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1300
1301	++dir;
1302	}
1303	}	1937	}
1304		1938	else if (ent->type == EIO_DT_DIR)
1305	/* now sort the dirs only (dirs all have the same score) */	1939	ent->score = 0;
1306	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1307	}	1940	}
1308
1309	break;
1310	}
1311
1312	/* now add the entry to our list(s) */
1313	name = entp->d_name;
1314
1315	/* skip . and .. entries */
1316	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1317	{
1318	int len = D_NAMLEN (entp) + 1;
1319
1320	while (expect_false (namesoffs + len > namesalloc))
1321	{
1322	namesalloc *= 2;
1323	X_LOCK (wrklock);
1324	req->ptr2 = names = realloc (names, namesalloc);
1325	X_UNLOCK (wrklock);
1326
1327	if (!names)
1328	break;
1329	}	1941	}
1330		1942
1331	memcpy (names + namesoffs, name, len);
1332
1333	if (dents)
1334	{
1335	struct eio_dirent *ent;
1336
1337	if (expect_false (dentoffs == dentalloc))
1338	{
1339	dentalloc *= 2;
1340	X_LOCK (wrklock);
1341	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1342	X_UNLOCK (wrklock);
1343
1344	if (!dents)
1345	break;
1346	}
1347
1348	ent = dents + dentoffs;
1349
1350	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1351	ent->namelen = len - 1;
1352	ent->inode = D_INO (entp);
1353
1354	inode_bits |= ent->inode;
1355
1356	switch (D_TYPE (entp))
1357	{
1358	default:
1359	ent->type = EIO_DT_UNKNOWN;
1360	flags |= EIO_READDIR_FOUND_UNKNOWN;
1361	break;
1362
1363	#ifdef DT_FIFO
1364	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1365	#endif
1366	#ifdef DT_CHR
1367	case DT_CHR: ent->type = EIO_DT_CHR; break;
1368	#endif
1369	#ifdef DT_MPC
1370	case DT_MPC: ent->type = EIO_DT_MPC; break;
1371	#endif
1372	#ifdef DT_DIR
1373	case DT_DIR: ent->type = EIO_DT_DIR; break;
1374	#endif
1375	#ifdef DT_NAM
1376	case DT_NAM: ent->type = EIO_DT_NAM; break;
1377	#endif
1378	#ifdef DT_BLK
1379	case DT_BLK: ent->type = EIO_DT_BLK; break;
1380	#endif
1381	#ifdef DT_MPB
1382	case DT_MPB: ent->type = EIO_DT_MPB; break;
1383	#endif
1384	#ifdef DT_REG
1385	case DT_REG: ent->type = EIO_DT_REG; break;
1386	#endif
1387	#ifdef DT_NWK
1388	case DT_NWK: ent->type = EIO_DT_NWK; break;
1389	#endif
1390	#ifdef DT_CMP
1391	case DT_CMP: ent->type = EIO_DT_CMP; break;
1392	#endif
1393	#ifdef DT_LNK
1394	case DT_LNK: ent->type = EIO_DT_LNK; break;
1395	#endif
1396	#ifdef DT_SOCK
1397	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1398	#endif
1399	#ifdef DT_DOOR
1400	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1401	#endif
1402	#ifdef DT_WHT
1403	case DT_WHT: ent->type = EIO_DT_WHT; break;
1404	#endif
1405	}
1406
1407	ent->score = 7;
1408
1409	if (flags & EIO_READDIR_DIRS_FIRST)
1410	{
1411	if (ent->type == EIO_DT_UNKNOWN)
1412	{
1413	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1414	ent->score = 1;
1415	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1416	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1417	}
1418	else if (ent->type == EIO_DT_DIR)
1419	ent->score = 0;
1420	}
1421	}
1422
1423	namesoffs += len;	1943	namesoffs += len;
1424	++dentoffs;	1944	++dentoffs;
1425	}	1945	}
1426		1946
1427	if (EIO_CANCELLED (req))	1947	if (EIO_CANCELLED (req))
1428	{	1948	{
1429	errno = ECANCELED;	1949	errno = ECANCELED;
1430	break;	1950	break;
1431	}	1951	}
		1952
		1953	#ifdef _WIN32
		1954	if (!FindNextFile (dirp, &entp))
		1955	{
		1956	FindClose (dirp);
		1957	dirp = 0;
1432	}	1958	}
		1959	#endif
		1960	}
1433	}	1961	}
1434		1962
1435	#ifdef PAGESIZE	1963	/*****************************************************************************/
1436	# define eio_pagesize() PAGESIZE	1964	/* working directory stuff */
1437	#else	1965	/* various deficiencies in the posix 2008 api force us to */
1438	static intptr_t	1966	/* keep the absolute path in string form at all times */
1439	eio_pagesize (void)	1967	/* fuck yeah. */
1440	{
1441	static intptr_t page;
1442		1968
1443	if (!page)	1969	#if !HAVE_AT
1444	page = sysconf (_SC_PAGESIZE);
1445		1970
		1971	/* a bit like realpath, but usually faster because it doesn'T have to return */
		1972	/* an absolute or canonical path */
		1973	static const char *
		1974	wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1975	{
		1976	if (!wd || *path == '/')
		1977	return path;
		1978
		1979	if (path [0] == '.' && !path [1])
		1980	return wd->str;
		1981
		1982	{
		1983	int l1 = wd->len;
		1984	int l2 = strlen (path);
		1985
		1986	char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
		1987
		1988	memcpy (res, wd->str, l1);
		1989	res [l1] = '/';
		1990	memcpy (res + l1 + 1, path, l2 + 1);
		1991
		1992	return res;
		1993	}
		1994	}
		1995
		1996	#endif
		1997
		1998	static eio_wd
		1999	eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2000	{
		2001	int fd;
		2002	eio_wd res;
		2003	int len = eio__realpath (tmpbuf, wd, path);
		2004
		2005	if (len < 0)
		2006	return EIO_INVALID_WD;
		2007
		2008	#if HAVE_AT
		2009	fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		2010
		2011	if (fd < 0)
		2012	return EIO_INVALID_WD;
		2013	#endif
		2014
		2015	res = malloc (sizeof (*res) + len); /* one extra 0-byte */
		2016
		2017	#if HAVE_AT
		2018	res->fd = fd;
		2019	#endif
		2020
		2021	res->len = len;
		2022	memcpy (res->str, tmpbuf->ptr, len);
		2023	res->str [len] = 0;
		2024
1446	return page;	2025	return res;
1447	}	2026	}
1448	#endif
1449		2027
1450	static void	2028	eio_wd
1451	eio_page_align (void **addr, size_t *length)	2029	eio_wd_open_sync (eio_wd wd, const char *path)
1452	{	2030	{
1453	intptr_t mask = eio_pagesize () - 1;	2031	struct tmpbuf tmpbuf = { 0 };
		2032	wd = eio__wd_open_sync (&tmpbuf, wd, path);
		2033	free (tmpbuf.ptr);
1454		2034
1455	/* round down addr */	2035	return wd;
1456	intptr_t adj = mask & (intptr_t)*addr;
1457
1458	*addr = (void *)((intptr_t)*addr - adj);
1459	*length += adj;
1460
1461	/* round up length */
1462	*length = (*length + mask) & ~mask;
1463	}	2036	}
1464		2037
1465	#if !_POSIX_MEMLOCK	2038	void
1466	# define eio__mlockall(a) ((errno = ENOSYS), -1)	2039	eio_wd_close_sync (eio_wd wd)
1467	#else	2040	{
		2041	if (wd != EIO_INVALID_WD && wd != EIO_CWD)
		2042	{
		2043	#if HAVE_AT
		2044	close (wd->fd);
		2045	#endif
		2046	free (wd);
		2047	}
		2048	}
		2049
		2050	#if HAVE_AT
		2051
		2052	/* they forgot these */
1468		2053
1469	static int	2054	static int
1470	eio__mlockall (int flags)	2055	eio__truncateat (int dirfd, const char *path, off_t length)
1471	{	2056	{
1472	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7	2057	int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1473	extern int mallopt (int, int);	2058	int res;
1474	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1475	#endif
1476		2059
1477	if (EIO_MCL_CURRENT != MCL_CURRENT	2060	if (fd < 0)
1478	|| EIO_MCL_FUTURE != MCL_FUTURE)	2061	return fd;
1479	{
1480	flags = 0
1481	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1482	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1483	}
1484		2062
1485	return mlockall (flags);	2063	res = ftruncate (fd, length);
		2064	close (fd);
		2065	return res;
1486	}	2066	}
1487	#endif
1488
1489	#if !_POSIX_MEMLOCK_RANGE
1490	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1491	#else
1492		2067
1493	static int	2068	static int
1494	eio__mlock (void *addr, size_t length)	2069	eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1495	{	2070	{
1496	eio_page_align (&addr, &length);	2071	int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
		2072	int res;
1497		2073
1498	return mlock (addr, length);	2074	if (fd < 0)
1499	}	2075	return fd;
1500		2076
1501	#endif	2077	res = fstatvfs (fd, buf);
1502		2078	close (fd);
1503	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1504	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1505	#else
1506
1507	int
1508	eio__msync (void *mem, size_t len, int flags)
1509	{
1510	eio_page_align (&mem, &len);
1511
1512	if (EIO_MS_ASYNC != MS_SYNC
1513	|| EIO_MS_INVALIDATE != MS_INVALIDATE
1514	|| EIO_MS_SYNC != MS_SYNC)
1515	{
1516	flags = 0
1517	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1518	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1519	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1520	}
1521
1522	return msync (mem, len, flags);
1523	}
1524
1525	#endif
1526
1527	int
1528	eio__mtouch (void *mem, size_t len, int flags)
1529	{
1530	eio_page_align (&mem, &len);
1531
1532	{
1533	intptr_t addr = (intptr_t)mem;
1534	intptr_t end = addr + len;
1535	intptr_t page = eio_pagesize ();
1536
1537	if (addr < end)
1538	if (flags & EIO_MT_MODIFY) /* modify */
1539	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1540	else
1541	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1542	}
1543
1544	return 0;	2079	return res;
		2080
1545	}	2081	}
		2082
		2083	#endif
1546		2084
1547	/*****************************************************************************/	2085	/*****************************************************************************/
1548		2086
1549	#define ALLOC(len) \	2087	#define ALLOC(len) \
1550	if (!req->ptr2) \	2088	if (!req->ptr2) \
…		…
1559	req->result = -1; \	2097	req->result = -1; \
1560	break; \	2098	break; \
1561	} \	2099	} \
1562	}	2100	}
1563		2101
		2102	static void ecb_noinline ecb_cold
		2103	etp_proc_init (void)
		2104	{
		2105	#if HAVE_PRCTL_SET_NAME
		2106	/* provide a more sensible "thread name" */
		2107	char name[16 + 1];
		2108	const int namelen = sizeof (name) - 1;
		2109	int len;
		2110
		2111	prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
		2112	name [namelen] = 0;
		2113	len = strlen (name);
		2114	strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
		2115	prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
		2116	#endif
		2117	}
		2118
1564	X_THREAD_PROC (etp_proc)	2119	X_THREAD_PROC (etp_proc)
1565	{	2120	{
1566	ETP_REQ *req;	2121	ETP_REQ *req;
1567	struct timespec ts;	2122	struct timespec ts;
1568	etp_worker *self = (etp_worker *)thr_arg;	2123	etp_worker *self = (etp_worker *)thr_arg;
1569		2124
		2125	etp_proc_init ();
		2126
1570	/* try to distribute timeouts somewhat randomly */	2127	/* try to distribute timeouts somewhat evenly */
1571	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	2128	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1572		2129
1573	for (;;)	2130	for (;;)
1574	{	2131	{
		2132	ts.tv_sec = 0;
		2133
1575	X_LOCK (reqlock);	2134	X_LOCK (reqlock);
1576		2135
1577	for (;;)	2136	for (;;)
1578	{	2137	{
1579	self->req = req = reqq_shift (&req_queue);	2138	req = reqq_shift (&req_queue);
1580		2139
1581	if (req)	2140	if (req)
1582	break;	2141	break;
1583		2142
		2143	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		2144	{
		2145	X_UNLOCK (reqlock);
		2146	X_LOCK (wrklock);
		2147	--started;
		2148	X_UNLOCK (wrklock);
		2149	goto quit;
		2150	}
		2151
1584	++idle;	2152	++idle;
1585		2153
1586	ts.tv_sec = time (0) + idle_timeout;	2154	if (idle <= max_idle)
1587	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	2155	/* we are allowed to idle, so do so without any timeout */
		2156	X_COND_WAIT (reqwait, reqlock);
		2157	else
1588	{	2158	{
1589	if (idle > max_idle)	2159	/* initialise timeout once */
1590	{	2160	if (!ts.tv_sec)
1591	--idle;	2161	ts.tv_sec = time (0) + idle_timeout;
1592	X_UNLOCK (reqlock);
1593	X_LOCK (wrklock);
1594	--started;
1595	X_UNLOCK (wrklock);
1596	goto quit;
1597	}
1598		2162
1599	/* we are allowed to idle, so do so without any timeout */
1600	X_COND_WAIT (reqwait, reqlock);	2163	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		2164	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1601	}	2165	}
1602		2166
1603	--idle;	2167	--idle;
1604	}	2168	}
1605		2169
…		…
1608	X_UNLOCK (reqlock);	2172	X_UNLOCK (reqlock);
1609		2173
1610	if (req->type < 0)	2174	if (req->type < 0)
1611	goto quit;	2175	goto quit;
1612		2176
1613	if (!EIO_CANCELLED (req))
1614	ETP_EXECUTE (self, req);	2177	ETP_EXECUTE (self, req);
1615		2178
1616	X_LOCK (reslock);	2179	X_LOCK (reslock);
1617		2180
1618	++npending;	2181	++npending;
1619		2182
1620	if (!reqq_push (&res_queue, req) && want_poll_cb)	2183	if (!reqq_push (&res_queue, req) && want_poll_cb)
1621	want_poll_cb ();	2184	want_poll_cb ();
1622		2185
1623	self->req = 0;
1624	etp_worker_clear (self);	2186	etp_worker_clear (self);
1625		2187
1626	X_UNLOCK (reslock);	2188	X_UNLOCK (reslock);
1627	}	2189	}
1628		2190
1629	quit:	2191	quit:
		2192	free (req);
		2193
1630	X_LOCK (wrklock);	2194	X_LOCK (wrklock);
1631	etp_worker_free (self);	2195	etp_worker_free (self);
1632	X_UNLOCK (wrklock);	2196	X_UNLOCK (wrklock);
1633
1634	return 0;
1635	}	2197	}
1636		2198
1637	/*****************************************************************************/	2199	/*****************************************************************************/
1638		2200
		2201	int ecb_cold
1639	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	2202	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1640	{	2203	{
1641	return etp_init (want_poll, done_poll);	2204	return etp_init (want_poll, done_poll);
1642	}	2205	}
1643		2206
		2207	ecb_inline void
1644	static void eio_api_destroy (eio_req *req)	2208	eio_api_destroy (eio_req *req)
1645	{	2209	{
1646	free (req);	2210	free (req);
1647	}	2211	}
1648		2212
1649	#define REQ(rtype) \	2213	#define REQ(rtype) \
1650	eio_req *req; \	2214	eio_req *req; \
1651	\	2215	\
1652	req = (eio_req *)calloc (1, sizeof *req); \	2216	req = (eio_req *)calloc (1, sizeof *req); \
1653	if (!req) \	2217	if (!req) \
1654	return 0; \	2218	return 0; \
…		…
1668	{ \	2232	{ \
1669	eio_api_destroy (req); \	2233	eio_api_destroy (req); \
1670	return 0; \	2234	return 0; \
1671	}	2235	}
1672		2236
		2237	static void
1673	static void eio_execute (etp_worker *self, eio_req *req)	2238	eio_execute (etp_worker *self, eio_req *req)
1674	{	2239	{
		2240	#if HAVE_AT
		2241	int dirfd;
		2242	#else
		2243	const char *path;
		2244	#endif
		2245
		2246	if (ecb_expect_false (EIO_CANCELLED (req)))
		2247	{
		2248	req->result = -1;
		2249	req->errorno = ECANCELED;
		2250	return;
		2251	}
		2252
		2253	if (ecb_expect_false (req->wd == EIO_INVALID_WD))
		2254	{
		2255	req->result = -1;
		2256	req->errorno = ENOENT;
		2257	return;
		2258	}
		2259
		2260	if (req->type >= EIO_OPEN)
		2261	{
		2262	#if HAVE_AT
		2263	dirfd = WD2FD (req->wd);
		2264	#else
		2265	path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		2266	#endif
		2267	}
		2268
1675	switch (req->type)	2269	switch (req->type)
1676	{	2270	{
		2271	case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
		2272	req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
		2273	break;
		2274	case EIO_WD_CLOSE: req->result = 0;
		2275	eio_wd_close_sync (req->wd); break;
		2276
		2277	case EIO_SEEK: eio__lseek (req); break;
1677	case EIO_READ: ALLOC (req->size);	2278	case EIO_READ: ALLOC (req->size);
1678	req->result = req->offs >= 0	2279	req->result = req->offs >= 0
1679	? pread (req->int1, req->ptr2, req->size, req->offs)	2280	? pread (req->int1, req->ptr2, req->size, req->offs)
1680	: read (req->int1, req->ptr2, req->size); break;	2281	: read (req->int1, req->ptr2, req->size); break;
1681	case EIO_WRITE: req->result = req->offs >= 0	2282	case EIO_WRITE: req->result = req->offs >= 0
1682	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2283	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1683	: write (req->int1, req->ptr2, req->size); break;	2284	: write (req->int1, req->ptr2, req->size); break;
1684		2285
1685	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2286	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1686	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2287	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
		2288
		2289	#if HAVE_AT
1687		2290
1688	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2291	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1689	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2292	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1690	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2293	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2294	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
		2295	case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
		2296	case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
		2297	case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
		2298	case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
		2299
		2300	case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
		2301	case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
		2302	case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
		2303	case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
		2304	case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
		2305	case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
		2306	case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2307	case EIO_READLINK: ALLOC (PATH_MAX);
		2308	req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
		2309	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2310	req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2311	case EIO_UTIME:
		2312	case EIO_FUTIME:
		2313	{
		2314	struct timespec ts[2];
		2315	struct timespec *times;
		2316
		2317	if (req->nv1 != -1. || req->nv2 != -1.)
		2318	{
		2319	ts[0].tv_sec = req->nv1;
		2320	ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
		2321	ts[1].tv_sec = req->nv2;
		2322	ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
		2323
		2324	times = ts;
		2325	}
		2326	else
		2327	times = 0;
		2328
		2329	req->result = req->type == EIO_FUTIME
		2330	? futimens (req->int1, times)
		2331	: utimensat (dirfd, req->ptr1, times, 0);
		2332	}
		2333	break;
		2334
		2335	#else
		2336
		2337	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2338	req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2339	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1691	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2340	req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2341	case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
		2342	case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
		2343	case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
		2344	case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
		2345
		2346	case EIO_UNLINK: req->result = unlink (path ); break;
		2347	case EIO_RMDIR: req->result = rmdir (path ); break;
		2348	case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
		2349	case EIO_RENAME: req->result = rename (path , req->ptr2); break;
		2350	case EIO_LINK: req->result = link (path , req->ptr2); break;
		2351	case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
		2352	case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
		2353	case EIO_READLINK: ALLOC (PATH_MAX);
		2354	req->result = readlink (path, req->ptr2, PATH_MAX); break;
		2355	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2356	req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2357
		2358	case EIO_UTIME:
		2359	case EIO_FUTIME:
		2360	{
		2361	struct timeval tv[2];
		2362	struct timeval *times;
		2363
		2364	if (req->nv1 != -1. || req->nv2 != -1.)
		2365	{
		2366	tv[0].tv_sec = req->nv1;
		2367	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
		2368	tv[1].tv_sec = req->nv2;
		2369	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
		2370
		2371	times = tv;
		2372	}
		2373	else
		2374	times = 0;
		2375
		2376	req->result = req->type == EIO_FUTIME
		2377	? futimes (req->int1, times)
		2378	: utimes (req->ptr1, times);
		2379	}
		2380	break;
		2381
		2382	#endif
		2383
		2384	case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
		2385	{
		2386	ALLOC (req->result);
		2387	memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
		2388	}
		2389	break;
		2390
1692	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2391	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1693	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2392	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1694		2393
1695	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1696	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1697	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));	2394	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1698	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;	2395	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1699		2396
1700	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1701	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	2397	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1702	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1703	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	2398	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1704	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1705	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;	2399	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1706		2400
1707	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1708	case EIO_CLOSE: req->result = close (req->int1); break;	2401	case EIO_CLOSE: req->result = close (req->int1); break;
1709	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;	2402	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1710	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1711	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1712	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1713	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1714	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1715	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1716	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1717
1718	case EIO_READLINK: ALLOC (PATH_MAX);
1719	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1720
1721	case EIO_SYNC: req->result = 0; sync (); break;	2403	case EIO_SYNC: req->result = 0; sync (); break;
1722	case EIO_FSYNC: req->result = fsync (req->int1); break;	2404	case EIO_FSYNC: req->result = fsync (req->int1); break;
1723	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2405	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2406	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2407	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1724	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2408	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1725	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	2409	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1726	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2410	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1727	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2411	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1728	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	2412	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1729		2413
1730	case EIO_READDIR: eio__scandir (req, self); break;	2414	case EIO_READDIR: eio__scandir (req, self); break;
1731		2415
1732	case EIO_BUSY:	2416	case EIO_BUSY:
1733	#ifdef _WIN32	2417	#ifdef _WIN32
…		…
1742	req->result = select (0, 0, 0, 0, &tv);	2426	req->result = select (0, 0, 0, 0, &tv);
1743	}	2427	}
1744	#endif	2428	#endif
1745	break;	2429	break;
1746		2430
1747	case EIO_UTIME:
1748	case EIO_FUTIME:
1749	{
1750	struct timeval tv[2];
1751	struct timeval *times;
1752
1753	if (req->nv1 != -1. || req->nv2 != -1.)
1754	{
1755	tv[0].tv_sec = req->nv1;
1756	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1757	tv[1].tv_sec = req->nv2;
1758	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1759
1760	times = tv;
1761	}
1762	else
1763	times = 0;
1764
1765	req->result = req->type == EIO_FUTIME
1766	? futimes (req->int1, times)
1767	: utimes (req->ptr1, times);
1768	}
1769	break;
1770
1771	case EIO_GROUP:	2431	case EIO_GROUP:
1772	abort (); /* handled in eio_request */	2432	abort (); /* handled in eio_request */
1773		2433
1774	case EIO_NOP:	2434	case EIO_NOP:
1775	req->result = 0;	2435	req->result = 0;
1776	break;	2436	break;
1777		2437
1778	case EIO_CUSTOM:	2438	case EIO_CUSTOM:
1779	((void (*)(eio_req *))req->feed) (req);	2439	req->feed (req);
1780	break;	2440	break;
1781		2441
1782	default:	2442	default:
1783	errno = ENOSYS;
1784	req->result = -1;	2443	req->result = EIO_ENOSYS ();
1785	break;	2444	break;
1786	}	2445	}
1787		2446
1788	req->errorno = errno;	2447	req->errorno = errno;
1789	}	2448	}
1790		2449
1791	#ifndef EIO_NO_WRAPPERS	2450	#ifndef EIO_NO_WRAPPERS
1792		2451
		2452	eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
		2453	{
		2454	REQ (EIO_WD_OPEN); PATH; SEND;
		2455	}
		2456
		2457	eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
		2458	{
		2459	REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
		2460	}
		2461
1793	eio_req *eio_nop (int pri, eio_cb cb, void *data)	2462	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1794	{	2463	{
1795	REQ (EIO_NOP); SEND;	2464	REQ (EIO_NOP); SEND;
1796	}	2465	}
1797		2466
…		…
1813	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2482	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1814	{	2483	{
1815	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2484	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1816	}	2485	}
1817		2486
		2487	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		2488	{
		2489	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2490	}
		2491
		2492	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2493	{
		2494	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2495	}
		2496
		2497	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2498	{
		2499	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2500	}
		2501
1818	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2502	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1819	{	2503	{
1820	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2504	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1821	}	2505	}
1822		2506
…		…
1828	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)	2512	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1829	{	2513	{
1830	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;	2514	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1831	}	2515	}
1832		2516
1833	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2517	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1834	{	2518	{
1835	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2519	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1836	}
1837
1838	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1839	{
1840	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1841	}	2520	}
1842		2521
1843	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2522	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1844	{	2523	{
1845	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2524	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
1848	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)	2527	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1849	{	2528	{
1850	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;	2529	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1851	}	2530	}
1852		2531
		2532	eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
		2533	{
		2534	REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
		2535	}
		2536
1853	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)	2537	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1854	{	2538	{
1855	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;	2539	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1856	}	2540	}
1857		2541
…		…
1883	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2567	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1884	{	2568	{
1885	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2569	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1886	}	2570	}
1887		2571
1888	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2572	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1889	{	2573	{
1890	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2574	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1891	}	2575	}
1892		2576
1893	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2577	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1913	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2597	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1914	{	2598	{
1915	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2599	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1916	}	2600	}
1917		2601
1918	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2602	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1919	{	2603	{
1920	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2604	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1921	}	2605	}
1922		2606
1923	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2607	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1937	}	2621	}
1938		2622
1939	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2623	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1940	{	2624	{
1941	return eio__1path (EIO_READLINK, path, pri, cb, data);	2625	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2626	}
		2627
		2628	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2629	{
		2630	return eio__1path (EIO_REALPATH, path, pri, cb, data);
1942	}	2631	}
1943		2632
1944	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2633	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1945	{	2634	{
1946	return eio__1path (EIO_STAT, path, pri, cb, data);	2635	return eio__1path (EIO_STAT, path, pri, cb, data);
…		…
2005	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2694	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2006	{	2695	{
2007	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2696	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2008	}	2697	}
2009		2698
2010	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2699	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2011	{	2700	{
2012	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2701	REQ (EIO_CUSTOM); req->feed = execute; SEND;
2013	}	2702	}
2014		2703
2015	#endif	2704	#endif
2016		2705
2017	eio_req *eio_grp (eio_cb cb, void *data)	2706	eio_req *eio_grp (eio_cb cb, void *data)
…		…
2026	#undef SEND	2715	#undef SEND
2027		2716
2028	/*****************************************************************************/	2717	/*****************************************************************************/
2029	/* grp functions */	2718	/* grp functions */
2030		2719
		2720	void
2031	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2721	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2032	{	2722	{
2033	grp->int2 = limit;	2723	grp->int2 = limit;
2034	grp->feed = feed;	2724	grp->feed = feed;
2035		2725
2036	grp_try_feed (grp);	2726	grp_try_feed (grp);
2037	}	2727	}
2038		2728
		2729	void
2039	void eio_grp_limit (eio_req *grp, int limit)	2730	eio_grp_limit (eio_req *grp, int limit)
2040	{	2731	{
2041	grp->int2 = limit;	2732	grp->int2 = limit;
2042		2733
2043	grp_try_feed (grp);	2734	grp_try_feed (grp);
2044	}	2735	}
2045		2736
		2737	void
2046	void eio_grp_add (eio_req *grp, eio_req *req)	2738	eio_grp_add (eio_req *grp, eio_req *req)
2047	{	2739	{
2048	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2740	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2049		2741
2050	grp->flags |= EIO_FLAG_GROUPADD;	2742	grp->flags |= EIO_FLAG_GROUPADD;
2051		2743
…		…
2062	}	2754	}
2063		2755
2064	/*****************************************************************************/	2756	/*****************************************************************************/
2065	/* misc garbage */	2757	/* misc garbage */
2066		2758
		2759	eio_ssize_t
2067	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2760	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2068	{	2761	{
2069	etp_worker wrk;
2070	ssize_t ret;
2071
2072	wrk.dbuf = 0;
2073
2074	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2762	return eio__sendfile (ofd, ifd, offset, count);
2075
2076	if (wrk.dbuf)
2077	free (wrk.dbuf);
2078
2079	return ret;
2080	}	2763	}
2081		2764

Diff Legend

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