ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.60 by root, Fri Jan 7 21:45:42 2011 UTC vs.
Revision 1.99 by root, Tue Jul 26 11:07:08 2011 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008,2009,2010 Marc Alexander Lehmann <libeio@schmorp.de>	4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5	* All rights reserved.	5	* All rights reserved.
6	*	6	*
7	* Redistribution and use in source and binary forms, with or without modifica-	7	* Redistribution and use in source and binary forms, with or without modifica-
8	* tion, are permitted provided that the following conditions are met:	8	* tion, are permitted provided that the following conditions are met:
9	*	9	*
…		…
35	* and other provisions required by the GPL. If you do not delete the	35	* and other provisions required by the GPL. If you do not delete the
36	* provisions above, a recipient may use your version of this file under	36	* provisions above, a recipient may use your version of this file under
37	* either the BSD or the GPL.	37	* either the BSD or the GPL.
38	*/	38	*/
39		39
		40	#ifndef _WIN32
		41	# include "config.h"
		42	#endif
		43
40	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
41		46
42	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
43	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
44	#endif	49	#endif
45	#include "xthread.h"	50	#include "xthread.h"
…		…
49	#include <stdlib.h>	54	#include <stdlib.h>
50	#include <string.h>	55	#include <string.h>
51	#include <errno.h>	56	#include <errno.h>
52	#include <sys/types.h>	57	#include <sys/types.h>
53	#include <sys/stat.h>	58	#include <sys/stat.h>
54	#include <sys/statvfs.h>
55	#include <limits.h>	59	#include <limits.h>
56	#include <fcntl.h>	60	#include <fcntl.h>
57	#include <assert.h>	61	#include <assert.h>
58		62
		63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
		64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
		65	#if HAVE_STDINT_H
		66	# include <stdint.h>
		67	#endif
		68
		69	#ifndef ECANCELED
		70	# define ECANCELED EDOM
		71	#endif
		72	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
		77	# define ENOTSOCK WSAENOTSOCK
		78	#endif
		79
		80	static void eio_destroy (eio_req *req);
		81
59	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
60	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	83	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
61	#endif	84	#endif
62		85
63	#ifndef EIO_DESTROY	86	#ifndef EIO_DESTROY
…		…
66		89
67	#ifndef EIO_FEED	90	#ifndef EIO_FEED
68	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	91	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
69	#endif	92	#endif
70		93
		94	#ifndef EIO_FD_TO_WIN32_HANDLE
		95	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		96	#endif
		97	#ifndef EIO_WIN32_HANDLE_TO_FD
		98	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		99	#endif
		100
		101	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		102
		103	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		104
71	#ifdef _WIN32	105	#ifdef _WIN32
72		106
73	/*doh*/	107	#undef PAGESIZE
		108	#define PAGESIZE 4096 /* GetSystemInfo? */
		109
		110	#ifdef EIO_STRUCT_STATI64
		111	#define stat(path,buf) _stati64 (path,buf)
		112	#define fstat(fd,buf) _fstati64 (fd,buf)
		113	#endif
		114	#define lstat(path,buf) stat (path,buf)
		115	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		116	#define mkdir(path,mode) _mkdir (path)
		117	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		118
		119	#define chmod(path,mode) _chmod (path, mode)
		120	#define dup(fd) _dup (fd)
		121	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		122
		123	#define fchmod(fd,mode) EIO_ENOSYS ()
		124	#define chown(path,uid,gid) EIO_ENOSYS ()
		125	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		126	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		127	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		128	#define mknod(path,mode,dev) EIO_ENOSYS ()
		129	#define sync() EIO_ENOSYS ()
		130	#define readlink(path,buf,s) EIO_ENOSYS ()
		131	#define statvfs(path,buf) EIO_ENOSYS ()
		132	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		133
		134	/* rename() uses MoveFile, which fails to overwrite */
		135	#define rename(old,neu) eio__rename (old, neu)
		136
		137	static int
		138	eio__rename (const char *old, const char *neu)
		139	{
		140	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		141	return 0;
		142
		143	/* should steal _dosmaperr */
		144	switch (GetLastError ())
		145	{
		146	case ERROR_FILE_NOT_FOUND:
		147	case ERROR_PATH_NOT_FOUND:
		148	case ERROR_INVALID_DRIVE:
		149	case ERROR_NO_MORE_FILES:
		150	case ERROR_BAD_NETPATH:
		151	case ERROR_BAD_NET_NAME:
		152	case ERROR_BAD_PATHNAME:
		153	case ERROR_FILENAME_EXCED_RANGE:
		154	errno = ENOENT;
		155	break;
		156
		157	default:
		158	errno = EACCES;
		159	break;
		160	}
		161
		162	return -1;
		163	}
		164
		165	/* we could even stat and see if it exists */
		166	static int
		167	symlink (const char *old, const char *neu)
		168	{
		169	#if WINVER >= 0x0600
		170	if (CreateSymbolicLink (neu, old, 1))
		171	return 0;
		172
		173	if (CreateSymbolicLink (neu, old, 0))
		174	return 0;
		175	#endif
		176
		177	return EIO_ERRNO (ENOENT, -1);
		178	}
		179
		180	/* POSIX API only */
		181	#define CreateHardLink(neu,old,flags) 0
		182	#define CreateSymbolicLink(neu,old,flags) 0
		183
		184	struct statvfs
		185	{
		186	int dummy;
		187	};
		188
		189	#define DT_DIR EIO_DT_DIR
		190	#define DT_REG EIO_DT_REG
		191	#define D_NAME(entp) entp.cFileName
		192	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
		193
74	#else	194	#else
75		195
76	# include "config.h"
77	# include <sys/time.h>	196	#include <sys/time.h>
78	# include <sys/select.h>	197	#include <sys/select.h>
		198	#include <sys/statvfs.h>
79	# include <unistd.h>	199	#include <unistd.h>
		200	#include <signal.h>
		201	#include <dirent.h>
		202
		203	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		204	#include <sys/mman.h>
		205	#endif
		206
		207	#define D_NAME(entp) entp->d_name
		208
		209	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		210	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		211	#define _DIRENT_HAVE_D_TYPE /* sigh */
		212	#define D_INO(de) (de)->d_fileno
		213	#define D_NAMLEN(de) (de)->d_namlen
		214	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		215	#define D_INO(de) (de)->d_ino
		216	#endif
		217
		218	#ifdef _D_EXACT_NAMLEN
		219	#undef D_NAMLEN
		220	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		221	#endif
		222
		223	#ifdef _DIRENT_HAVE_D_TYPE
		224	#define D_TYPE(de) (de)->d_type
		225	#endif
		226
		227	#ifndef EIO_STRUCT_DIRENT
		228	#define EIO_STRUCT_DIRENT struct dirent
		229	#endif
		230
		231	#endif
		232
		233	#if HAVE_UTIMES
80	# include <utime.h>	234	# include <utime.h>
81	# include <signal.h>
82	# include <dirent.h>
83
84	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
85	# include <sys/mman.h>
86	#endif
87
88	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
89	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
90	# define _DIRENT_HAVE_D_TYPE /* sigh */
91	# define D_INO(de) (de)->d_fileno
92	# define D_NAMLEN(de) (de)->d_namlen
93	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
94	# define D_INO(de) (de)->d_ino
95	# endif	235	#endif
96		236
97	#ifdef _D_EXACT_NAMLEN	237	#if HAVE_SYS_SYSCALL_H
98	# undef D_NAMLEN	238	# include <sys/syscall.h>
99	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
100	#endif
101
102	# ifdef _DIRENT_HAVE_D_TYPE
103	# define D_TYPE(de) (de)->d_type
104	# endif	239	#endif
105		240
106	# ifndef EIO_STRUCT_DIRENT	241	#if HAVE_SYS_PRCTL_H
107	# define EIO_STRUCT_DIRENT struct dirent	242	# include <sys/prctl.h>
108	# endif
109
110	#endif	243	#endif
111		244
112	#if HAVE_SENDFILE	245	#if HAVE_SENDFILE
113	# if __linux	246	# if __linux
114	# include <sys/sendfile.h>	247	# include <sys/sendfile.h>
…		…
129	#endif	262	#endif
130	#ifndef D_INO	263	#ifndef D_INO
131	# define D_INO(de) 0	264	# define D_INO(de) 0
132	#endif	265	#endif
133	#ifndef D_NAMLEN	266	#ifndef D_NAMLEN
134	# define D_NAMLEN(de) strlen ((de)->d_name)	267	# define D_NAMLEN(entp) strlen (D_NAME (entp))
135	#endif	268	#endif
136
137	/* number of seconds after which an idle threads exit */
138	#define IDLE_TIMEOUT 10
139		269
140	/* used for struct dirent, AIX doesn't provide it */	270	/* used for struct dirent, AIX doesn't provide it */
141	#ifndef NAME_MAX	271	#ifndef NAME_MAX
142	# define NAME_MAX 4096	272	# define NAME_MAX 4096
143	#endif	273	#endif
…		…
149		279
150	/* buffer size for various temporary buffers */	280	/* buffer size for various temporary buffers */
151	#define EIO_BUFSIZE 65536	281	#define EIO_BUFSIZE 65536
152		282
153	#define dBUF \	283	#define dBUF \
154	char *eio_buf; \
155	ETP_WORKER_LOCK (self); \
156	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	284	char *eio_buf = malloc (EIO_BUFSIZE); \
157	ETP_WORKER_UNLOCK (self); \
158	errno = ENOMEM; \	285	errno = ENOMEM; \
159	if (!eio_buf) \	286	if (!eio_buf) \
160	return -1;	287	return -1
		288
		289	#define FUBd \
		290	free (eio_buf)
161		291
162	#define EIO_TICKS ((1000000 + 1023) >> 10)	292	#define EIO_TICKS ((1000000 + 1023) >> 10)
163
164	/*****************************************************************************/
165
166	#if __GNUC__ >= 3
167	# define expect(expr,value) __builtin_expect ((expr),(value))
168	#else
169	# define expect(expr,value) (expr)
170	#endif
171
172	#define expect_false(expr) expect ((expr) != 0, 0)
173	#define expect_true(expr) expect ((expr) != 0, 1)
174
175	/*****************************************************************************/
176		293
177	#define ETP_PRI_MIN EIO_PRI_MIN	294	#define ETP_PRI_MIN EIO_PRI_MIN
178	#define ETP_PRI_MAX EIO_PRI_MAX	295	#define ETP_PRI_MAX EIO_PRI_MAX
179		296
180	struct etp_worker;	297	struct etp_worker;
…		…
184	static int eio_finish (eio_req *req);	301	static int eio_finish (eio_req *req);
185	#define ETP_FINISH(req) eio_finish (req)	302	#define ETP_FINISH(req) eio_finish (req)
186	static void eio_execute (struct etp_worker *self, eio_req *req);	303	static void eio_execute (struct etp_worker *self, eio_req *req);
187	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	304	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
188		305
189	#define ETP_WORKER_CLEAR(req) \
190	if (wrk->dbuf) \
191	{ \
192	free (wrk->dbuf); \
193	wrk->dbuf = 0; \
194	} \
195	\
196	if (wrk->dirp) \
197	{ \
198	closedir (wrk->dirp); \
199	wrk->dirp = 0; \
200	}
201
202	#define ETP_WORKER_COMMON \
203	void *dbuf; \
204	DIR *dirp;
205
206	/*****************************************************************************/	306	/*****************************************************************************/
207		307
208	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	308	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209		309
210	/* calculate time difference in ~1/EIO_TICKS of a second */	310	/* calculate time difference in ~1/EIO_TICKS of a second */
		311	ecb_inline int
211	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	312	tvdiff (struct timeval *tv1, struct timeval *tv2)
212	{	313	{
213	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	314	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
214	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	315	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
215	}	316	}
216		317
…		…
223	static unsigned int max_poll_reqs; /* reslock */	324	static unsigned int max_poll_reqs; /* reslock */
224		325
225	static volatile unsigned int nreqs; /* reqlock */	326	static volatile unsigned int nreqs; /* reqlock */
226	static volatile unsigned int nready; /* reqlock */	327	static volatile unsigned int nready; /* reqlock */
227	static volatile unsigned int npending; /* reqlock */	328	static volatile unsigned int npending; /* reqlock */
228	static volatile unsigned int max_idle = 4;	329	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		330	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
229		331
230	static xmutex_t wrklock = X_MUTEX_INIT;	332	static xmutex_t wrklock;
231	static xmutex_t reslock = X_MUTEX_INIT;	333	static xmutex_t reslock;
232	static xmutex_t reqlock = X_MUTEX_INIT;	334	static xmutex_t reqlock;
233	static xcond_t reqwait = X_COND_INIT;	335	static xcond_t reqwait;
234		336
235	#if !HAVE_PREADWRITE	337	#if !HAVE_PREADWRITE
236	/*	338	/*
237	* make our pread/pwrite emulation safe against themselves, but not against	339	* make our pread/pwrite emulation safe against themselves, but not against
238	* normal read/write by using a mutex. slows down execution a lot,	340	* normal read/write by using a mutex. slows down execution a lot,
239	* but that's your problem, not mine.	341	* but that's your problem, not mine.
240	*/	342	*/
241	static xmutex_t preadwritelock = X_MUTEX_INIT;	343	static xmutex_t preadwritelock;
242	#endif	344	#endif
243		345
244	typedef struct etp_worker	346	typedef struct etp_worker
245	{	347	{
246	/* locked by wrklock */	348	/* locked by wrklock */
…		…
249	xthread_t tid;	351	xthread_t tid;
250		352
251	/* locked by reslock, reqlock or wrklock */	353	/* locked by reslock, reqlock or wrklock */
252	ETP_REQ *req; /* currently processed request */	354	ETP_REQ *req; /* currently processed request */
253		355
		356	#ifdef ETP_WORKER_COMMON
254	ETP_WORKER_COMMON	357	ETP_WORKER_COMMON
		358	#endif
255	} etp_worker;	359	} etp_worker;
256		360
257	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	361	static etp_worker wrk_first; /* NOT etp */
258		362
259	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	363	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
260	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	364	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
261		365
262	/* worker threads management */	366	/* worker threads management */
263		367
		368	static void ecb_cold
264	static void etp_worker_clear (etp_worker *wrk)	369	etp_worker_clear (etp_worker *wrk)
265	{	370	{
266	ETP_WORKER_CLEAR (wrk);
267	}	371	}
268		372
		373	static void ecb_cold
269	static void etp_worker_free (etp_worker *wrk)	374	etp_worker_free (etp_worker *wrk)
270	{	375	{
271	wrk->next->prev = wrk->prev;	376	wrk->next->prev = wrk->prev;
272	wrk->prev->next = wrk->next;	377	wrk->prev->next = wrk->next;
273		378
274	free (wrk);	379	free (wrk);
275	}	380	}
276		381
277	static unsigned int etp_nreqs (void)	382	static unsigned int
		383	etp_nreqs (void)
278	{	384	{
279	int retval;	385	int retval;
280	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	386	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281	retval = nreqs;	387	retval = nreqs;
282	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	388	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283	return retval;	389	return retval;
284	}	390	}
285		391
286	static unsigned int etp_nready (void)	392	static unsigned int
		393	etp_nready (void)
287	{	394	{
288	unsigned int retval;	395	unsigned int retval;
289		396
290	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	397	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
291	retval = nready;	398	retval = nready;
292	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	399	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
293		400
294	return retval;	401	return retval;
295	}	402	}
296		403
297	static unsigned int etp_npending (void)	404	static unsigned int
		405	etp_npending (void)
298	{	406	{
299	unsigned int retval;	407	unsigned int retval;
300		408
301	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	409	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
302	retval = npending;	410	retval = npending;
303	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	411	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
304		412
305	return retval;	413	return retval;
306	}	414	}
307		415
308	static unsigned int etp_nthreads (void)	416	static unsigned int
		417	etp_nthreads (void)
309	{	418	{
310	unsigned int retval;	419	unsigned int retval;
311		420
312	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	421	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
313	retval = started;	422	retval = started;
…		…
327	} etp_reqq;	436	} etp_reqq;
328		437
329	static etp_reqq req_queue;	438	static etp_reqq req_queue;
330	static etp_reqq res_queue;	439	static etp_reqq res_queue;
331		440
		441	static void ecb_noinline ecb_cold
		442	reqq_init (etp_reqq *q)
		443	{
		444	int pri;
		445
		446	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		447	q->qs[pri] = q->qe[pri] = 0;
		448
		449	q->size = 0;
		450	}
		451
		452	static int ecb_noinline
332	static int reqq_push (etp_reqq *q, ETP_REQ *req)	453	reqq_push (etp_reqq *q, ETP_REQ *req)
333	{	454	{
334	int pri = req->pri;	455	int pri = req->pri;
335	req->next = 0;	456	req->next = 0;
336		457
337	if (q->qe[pri])	458	if (q->qe[pri])
…		…
343	q->qe[pri] = q->qs[pri] = req;	464	q->qe[pri] = q->qs[pri] = req;
344		465
345	return q->size++;	466	return q->size++;
346	}	467	}
347		468
		469	static ETP_REQ * ecb_noinline
348	static ETP_REQ *reqq_shift (etp_reqq *q)	470	reqq_shift (etp_reqq *q)
349	{	471	{
350	int pri;	472	int pri;
351		473
352	if (!q->size)	474	if (!q->size)
353	return 0;	475	return 0;
…		…
368	}	490	}
369		491
370	abort ();	492	abort ();
371	}	493	}
372		494
373	static void etp_atfork_prepare (void)	495	static int ecb_cold
		496	etp_init (void (*want_poll)(void), void (*done_poll)(void))
374	{	497	{
375	X_LOCK (wrklock);	498	X_MUTEX_CREATE (wrklock);
376	X_LOCK (reqlock);	499	X_MUTEX_CREATE (reslock);
377	X_LOCK (reslock);	500	X_MUTEX_CREATE (reqlock);
378	#if !HAVE_PREADWRITE	501	X_COND_CREATE (reqwait);
379	X_LOCK (preadwritelock);
380	#endif
381	}
382		502
383	static void etp_atfork_parent (void)	503	reqq_init (&req_queue);
384	{	504	reqq_init (&res_queue);
385	#if !HAVE_PREADWRITE
386	X_UNLOCK (preadwritelock);
387	#endif
388	X_UNLOCK (reslock);
389	X_UNLOCK (reqlock);
390	X_UNLOCK (wrklock);
391	}
392		505
393	static void etp_atfork_child (void)	506	wrk_first.next =
394	{	507	wrk_first.prev = &wrk_first;
395	ETP_REQ *prv;
396
397	while ((prv = reqq_shift (&req_queue)))
398	ETP_DESTROY (prv);
399
400	while ((prv = reqq_shift (&res_queue)))
401	ETP_DESTROY (prv);
402
403	while (wrk_first.next != &wrk_first)
404	{
405	etp_worker *wrk = wrk_first.next;
406
407	if (wrk->req)
408	ETP_DESTROY (wrk->req);
409
410	etp_worker_clear (wrk);
411	etp_worker_free (wrk);
412	}
413		508
414	started = 0;	509	started = 0;
415	idle = 0;	510	idle = 0;
416	nreqs = 0;	511	nreqs = 0;
417	nready = 0;	512	nready = 0;
418	npending = 0;	513	npending = 0;
419		514
420	etp_atfork_parent ();
421	}
422
423	static void
424	etp_once_init (void)
425	{
426	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
427	}
428
429	static int
430	etp_init (void (*want_poll)(void), void (*done_poll)(void))
431	{
432	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
433
434	pthread_once (&doinit, etp_once_init);
435
436	want_poll_cb = want_poll;	515	want_poll_cb = want_poll;
437	done_poll_cb = done_poll;	516	done_poll_cb = done_poll;
438		517
439	return 0;	518	return 0;
440	}	519	}
441		520
442	X_THREAD_PROC (etp_proc);	521	X_THREAD_PROC (etp_proc);
443		522
		523	static void ecb_cold
444	static void etp_start_thread (void)	524	etp_start_thread (void)
445	{	525	{
446	etp_worker *wrk = calloc (1, sizeof (etp_worker));	526	etp_worker *wrk = calloc (1, sizeof (etp_worker));
447		527
448	/*TODO*/	528	/*TODO*/
449	assert (("unable to allocate worker thread data", wrk));	529	assert (("unable to allocate worker thread data", wrk));
…		…
462	free (wrk);	542	free (wrk);
463		543
464	X_UNLOCK (wrklock);	544	X_UNLOCK (wrklock);
465	}	545	}
466		546
		547	static void
467	static void etp_maybe_start_thread (void)	548	etp_maybe_start_thread (void)
468	{	549	{
469	if (expect_true (etp_nthreads () >= wanted))	550	if (ecb_expect_true (etp_nthreads () >= wanted))
470	return;	551	return;
471		552
472	/* todo: maybe use idle here, but might be less exact */	553	/* todo: maybe use idle here, but might be less exact */
473	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	554	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
474	return;	555	return;
475		556
476	etp_start_thread ();	557	etp_start_thread ();
477	}	558	}
478		559
		560	static void ecb_cold
479	static void etp_end_thread (void)	561	etp_end_thread (void)
480	{	562	{
481	eio_req *req = calloc (1, sizeof (eio_req));	563	eio_req *req = calloc (1, sizeof (eio_req));
482		564
483	req->type = -1;	565	req->type = -1;
484	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	566	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
491	X_LOCK (wrklock);	573	X_LOCK (wrklock);
492	--started;	574	--started;
493	X_UNLOCK (wrklock);	575	X_UNLOCK (wrklock);
494	}	576	}
495		577
496	static int etp_poll (void)	578	static int
		579	etp_poll (void)
497	{	580	{
498	unsigned int maxreqs;	581	unsigned int maxreqs;
499	unsigned int maxtime;	582	unsigned int maxtime;
500	struct timeval tv_start, tv_now;	583	struct timeval tv_start, tv_now;
501		584
…		…
531		614
532	X_LOCK (reqlock);	615	X_LOCK (reqlock);
533	--nreqs;	616	--nreqs;
534	X_UNLOCK (reqlock);	617	X_UNLOCK (reqlock);
535		618
536	if (expect_false (req->type == EIO_GROUP && req->size))	619	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
537	{	620	{
538	req->int1 = 1; /* mark request as delayed */	621	req->int1 = 1; /* mark request as delayed */
539	continue;	622	continue;
540	}	623	}
541	else	624	else
542	{	625	{
543	int res = ETP_FINISH (req);	626	int res = ETP_FINISH (req);
544	if (expect_false (res))	627	if (ecb_expect_false (res))
545	return res;	628	return res;
546	}	629	}
547		630
548	if (expect_false (maxreqs && !--maxreqs))	631	if (ecb_expect_false (maxreqs && !--maxreqs))
549	break;	632	break;
550		633
551	if (maxtime)	634	if (maxtime)
552	{	635	{
553	gettimeofday (&tv_now, 0);	636	gettimeofday (&tv_now, 0);
…		…
559		642
560	errno = EAGAIN;	643	errno = EAGAIN;
561	return -1;	644	return -1;
562	}	645	}
563		646
		647	static void
564	static void etp_cancel (ETP_REQ *req)	648	etp_cancel (ETP_REQ *req)
565	{	649	{
566	X_LOCK (wrklock);	650	req->cancelled = 1;
567	req->flags |= EIO_FLAG_CANCELLED;
568	X_UNLOCK (wrklock);
569		651
570	eio_grp_cancel (req);	652	eio_grp_cancel (req);
571	}	653	}
572		654
		655	static void
573	static void etp_submit (ETP_REQ *req)	656	etp_submit (ETP_REQ *req)
574	{	657	{
575	req->pri -= ETP_PRI_MIN;	658	req->pri -= ETP_PRI_MIN;
576		659
577	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	660	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
578	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	661	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
579		662
580	if (expect_false (req->type == EIO_GROUP))	663	if (ecb_expect_false (req->type == EIO_GROUP))
581	{	664	{
582	/* I hope this is worth it :/ */	665	/* I hope this is worth it :/ */
583	X_LOCK (reqlock);	666	X_LOCK (reqlock);
584	++nreqs;	667	++nreqs;
585	X_UNLOCK (reqlock);	668	X_UNLOCK (reqlock);
…		…
604		687
605	etp_maybe_start_thread ();	688	etp_maybe_start_thread ();
606	}	689	}
607	}	690	}
608		691
		692	static void ecb_cold
609	static void etp_set_max_poll_time (double nseconds)	693	etp_set_max_poll_time (double nseconds)
610	{	694	{
611	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	695	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
612	max_poll_time = nseconds * EIO_TICKS;	696	max_poll_time = nseconds * EIO_TICKS;
613	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	697	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
614	}	698	}
615		699
		700	static void ecb_cold
616	static void etp_set_max_poll_reqs (unsigned int maxreqs)	701	etp_set_max_poll_reqs (unsigned int maxreqs)
617	{	702	{
618	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	703	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
619	max_poll_reqs = maxreqs;	704	max_poll_reqs = maxreqs;
620	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	705	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
621	}	706	}
622		707
		708	static void ecb_cold
623	static void etp_set_max_idle (unsigned int nthreads)	709	etp_set_max_idle (unsigned int nthreads)
624	{	710	{
625	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	711	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
626	max_idle = nthreads <= 0 ? 1 : nthreads;	712	max_idle = nthreads;
627	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	713	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
628	}	714	}
629		715
		716	static void ecb_cold
		717	etp_set_idle_timeout (unsigned int seconds)
		718	{
		719	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		720	idle_timeout = seconds;
		721	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		722	}
		723
		724	static void ecb_cold
630	static void etp_set_min_parallel (unsigned int nthreads)	725	etp_set_min_parallel (unsigned int nthreads)
631	{	726	{
632	if (wanted < nthreads)	727	if (wanted < nthreads)
633	wanted = nthreads;	728	wanted = nthreads;
634	}	729	}
635		730
		731	static void ecb_cold
636	static void etp_set_max_parallel (unsigned int nthreads)	732	etp_set_max_parallel (unsigned int nthreads)
637	{	733	{
638	if (wanted > nthreads)	734	if (wanted > nthreads)
639	wanted = nthreads;	735	wanted = nthreads;
640		736
641	while (started > wanted)	737	while (started > wanted)
642	etp_end_thread ();	738	etp_end_thread ();
643	}	739	}
644		740
645	/*****************************************************************************/	741	/*****************************************************************************/
646		742
		743	static void
647	static void grp_try_feed (eio_req *grp)	744	grp_try_feed (eio_req *grp)
648	{	745	{
649	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	746	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
650	{	747	{
651	grp->flags &= ~EIO_FLAG_GROUPADD;	748	grp->flags &= ~EIO_FLAG_GROUPADD;
652		749
…		…
659	break;	756	break;
660	}	757	}
661	}	758	}
662	}	759	}
663		760
		761	static int
664	static int grp_dec (eio_req *grp)	762	grp_dec (eio_req *grp)
665	{	763	{
666	--grp->size;	764	--grp->size;
667		765
668	/* call feeder, if applicable */	766	/* call feeder, if applicable */
669	grp_try_feed (grp);	767	grp_try_feed (grp);
…		…
673	return eio_finish (grp);	771	return eio_finish (grp);
674	else	772	else
675	return 0;	773	return 0;
676	}	774	}
677		775
		776	static void
678	void eio_destroy (eio_req *req)	777	eio_destroy (eio_req *req)
679	{	778	{
680	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	779	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
681	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	780	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
682		781
683	EIO_DESTROY (req);	782	EIO_DESTROY (req);
684	}	783	}
685		784
		785	static int
686	static int eio_finish (eio_req *req)	786	eio_finish (eio_req *req)
687	{	787	{
688	int res = EIO_FINISH (req);	788	int res = EIO_FINISH (req);
689		789
690	if (req->grp)	790	if (req->grp)
691	{	791	{
…		…
699	if (grp->grp_first == req)	799	if (grp->grp_first == req)
700	grp->grp_first = req->grp_next;	800	grp->grp_first = req->grp_next;
701		801
702	res2 = grp_dec (grp);	802	res2 = grp_dec (grp);
703		803
704	if (!res && res2)	804	if (!res)
705	res = res2;	805	res = res2;
706	}	806	}
707		807
708	eio_destroy (req);	808	eio_destroy (req);
709		809
710	return res;	810	return res;
711	}	811	}
712		812
		813	void
713	void eio_grp_cancel (eio_req *grp)	814	eio_grp_cancel (eio_req *grp)
714	{	815	{
715	for (grp = grp->grp_first; grp; grp = grp->grp_next)	816	for (grp = grp->grp_first; grp; grp = grp->grp_next)
716	eio_cancel (grp);	817	eio_cancel (grp);
717	}	818	}
718		819
		820	void
719	void eio_cancel (eio_req *req)	821	eio_cancel (eio_req *req)
720	{	822	{
721	etp_cancel (req);	823	etp_cancel (req);
722	}	824	}
723		825
		826	void
724	void eio_submit (eio_req *req)	827	eio_submit (eio_req *req)
725	{	828	{
726	etp_submit (req);	829	etp_submit (req);
727	}	830	}
728		831
729	unsigned int eio_nreqs (void)	832	unsigned int
		833	eio_nreqs (void)
730	{	834	{
731	return etp_nreqs ();	835	return etp_nreqs ();
732	}	836	}
733		837
734	unsigned int eio_nready (void)	838	unsigned int
		839	eio_nready (void)
735	{	840	{
736	return etp_nready ();	841	return etp_nready ();
737	}	842	}
738		843
739	unsigned int eio_npending (void)	844	unsigned int
		845	eio_npending (void)
740	{	846	{
741	return etp_npending ();	847	return etp_npending ();
742	}	848	}
743		849
744	unsigned int eio_nthreads (void)	850	unsigned int ecb_cold
		851	eio_nthreads (void)
745	{	852	{
746	return etp_nthreads ();	853	return etp_nthreads ();
747	}	854	}
748		855
		856	void ecb_cold
749	void eio_set_max_poll_time (double nseconds)	857	eio_set_max_poll_time (double nseconds)
750	{	858	{
751	etp_set_max_poll_time (nseconds);	859	etp_set_max_poll_time (nseconds);
752	}	860	}
753		861
		862	void ecb_cold
754	void eio_set_max_poll_reqs (unsigned int maxreqs)	863	eio_set_max_poll_reqs (unsigned int maxreqs)
755	{	864	{
756	etp_set_max_poll_reqs (maxreqs);	865	etp_set_max_poll_reqs (maxreqs);
757	}	866	}
758		867
		868	void ecb_cold
759	void eio_set_max_idle (unsigned int nthreads)	869	eio_set_max_idle (unsigned int nthreads)
760	{	870	{
761	etp_set_max_idle (nthreads);	871	etp_set_max_idle (nthreads);
762	}	872	}
763		873
		874	void ecb_cold
		875	eio_set_idle_timeout (unsigned int seconds)
		876	{
		877	etp_set_idle_timeout (seconds);
		878	}
		879
		880	void ecb_cold
764	void eio_set_min_parallel (unsigned int nthreads)	881	eio_set_min_parallel (unsigned int nthreads)
765	{	882	{
766	etp_set_min_parallel (nthreads);	883	etp_set_min_parallel (nthreads);
767	}	884	}
768		885
		886	void ecb_cold
769	void eio_set_max_parallel (unsigned int nthreads)	887	eio_set_max_parallel (unsigned int nthreads)
770	{	888	{
771	etp_set_max_parallel (nthreads);	889	etp_set_max_parallel (nthreads);
772	}	890	}
773		891
774	int eio_poll (void)	892	int eio_poll (void)
…		…
783	# undef pread	901	# undef pread
784	# undef pwrite	902	# undef pwrite
785	# define pread eio__pread	903	# define pread eio__pread
786	# define pwrite eio__pwrite	904	# define pwrite eio__pwrite
787		905
788	static ssize_t	906	static eio_ssize_t
789	eio__pread (int fd, void *buf, size_t count, off_t offset)	907	eio__pread (int fd, void *buf, size_t count, off_t offset)
790	{	908	{
791	ssize_t res;	909	eio_ssize_t res;
792	off_t ooffset;	910	off_t ooffset;
793		911
794	X_LOCK (preadwritelock);	912	X_LOCK (preadwritelock);
795	ooffset = lseek (fd, 0, SEEK_CUR);	913	ooffset = lseek (fd, 0, SEEK_CUR);
796	lseek (fd, offset, SEEK_SET);	914	lseek (fd, offset, SEEK_SET);
…		…
799	X_UNLOCK (preadwritelock);	917	X_UNLOCK (preadwritelock);
800		918
801	return res;	919	return res;
802	}	920	}
803		921
804	static ssize_t	922	static eio_ssize_t
805	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	923	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
806	{	924	{
807	ssize_t res;	925	eio_ssize_t res;
808	off_t ooffset;	926	off_t ooffset;
809		927
810	X_LOCK (preadwritelock);	928	X_LOCK (preadwritelock);
811	ooffset = lseek (fd, 0, SEEK_CUR);	929	ooffset = lseek (fd, 0, SEEK_CUR);
812	lseek (fd, offset, SEEK_SET);	930	lseek (fd, offset, SEEK_SET);
…		…
844	#ifndef HAVE_FUTIMES	962	#ifndef HAVE_FUTIMES
845		963
846	# undef futimes	964	# undef futimes
847	# define futimes(fd,times) eio__futimes (fd, times)	965	# define futimes(fd,times) eio__futimes (fd, times)
848		966
		967	static int
849	static int eio__futimes (int fd, const struct timeval tv[2])	968	eio__futimes (int fd, const struct timeval tv[2])
850	{	969	{
851	errno = ENOSYS;	970	errno = ENOSYS;
852	return -1;	971	return -1;
853	}	972	}
854		973
…		…
857	#if !HAVE_FDATASYNC	976	#if !HAVE_FDATASYNC
858	# undef fdatasync	977	# undef fdatasync
859	# define fdatasync(fd) fsync (fd)	978	# define fdatasync(fd) fsync (fd)
860	#endif	979	#endif
861		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 = -1;
		990	errno = ENOSYS;
		991	#endif
		992
		993	if (res < 0 && errno == ENOSYS && fd >= 0)
		994	sync ();
		995
		996	return res;
		997	}
		998
862	/* sync_file_range always needs emulation */	999	/* sync_file_range always needs emulation */
863	int	1000	static int
864	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	1001	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
865	{	1002	{
866	#if HAVE_SYNC_FILE_RANGE	1003	#if HAVE_SYNC_FILE_RANGE
867	int res;	1004	int res;
868		1005
…		…
885	/* even though we could play tricks with the flags, it's better to always	1022	/* even though we could play tricks with the flags, it's better to always
886	* call fdatasync, as that matches the expectation of its users best */	1023	* call fdatasync, as that matches the expectation of its users best */
887	return fdatasync (fd);	1024	return fdatasync (fd);
888	}	1025	}
889		1026
		1027	static int
		1028	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		1029	{
		1030	#if HAVE_FALLOCATE
		1031	return fallocate (fd, mode, offset, len);
		1032	#else
		1033	errno = ENOSYS;
		1034	return -1;
		1035	#endif
		1036	}
		1037
890	#if !HAVE_READAHEAD	1038	#if !HAVE_READAHEAD
891	# undef readahead	1039	# undef readahead
892	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1040	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
893		1041
894	static ssize_t	1042	static eio_ssize_t
895	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1043	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
896	{	1044	{
897	size_t todo = count;	1045	size_t todo = count;
898	dBUF;	1046	dBUF;
899		1047
…		…
904	pread (fd, eio_buf, len, offset);	1052	pread (fd, eio_buf, len, offset);
905	offset += len;	1053	offset += len;
906	todo -= len;	1054	todo -= len;
907	}	1055	}
908		1056
		1057	FUBd;
		1058
909	errno = 0;	1059	errno = 0;
910	return count;	1060	return count;
911	}	1061	}
912		1062
913	#endif	1063	#endif
914		1064
915	/* sendfile always needs emulation */	1065	/* sendfile always needs emulation */
916	static ssize_t	1066	static eio_ssize_t
917	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1067	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
918	{	1068	{
		1069	eio_ssize_t written = 0;
919	ssize_t res;	1070	eio_ssize_t res;
920		1071
921	if (!count)	1072	if (!count)
922	return 0;	1073	return 0;
923		1074
		1075	for (;;)
		1076	{
		1077	#ifdef __APPLE__
		1078	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1079	#endif
924	#if HAVE_SENDFILE	1080	#if HAVE_SENDFILE
925	# if __linux	1081	# if __linux
		1082	off_t soffset = offset;
926	res = sendfile (ofd, ifd, &offset, count);	1083	res = sendfile (ofd, ifd, &soffset, count);
927		1084
928	# elif __FreeBSD__	1085	# elif __FreeBSD__
929	/*	1086	/*
930	* Of course, the freebsd sendfile is a dire hack with no thoughts	1087	* Of course, the freebsd sendfile is a dire hack with no thoughts
931	* wasted on making it similar to other I/O functions.	1088	* wasted on making it similar to other I/O functions.
932	*/	1089	*/
933	{
934	off_t sbytes;	1090	off_t sbytes;
935	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1091	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
936		1092
937	#if 0 /* according to the manpage, this is correct, but broken behaviour */	1093	#if 0 /* according to the manpage, this is correct, but broken behaviour */
938	/* freebsd' sendfile will return 0 on success */	1094	/* freebsd' sendfile will return 0 on success */
939	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */	1095	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
940	/* not on e.g. EIO or EPIPE - sounds broken */	1096	/* not on e.g. EIO or EPIPE - sounds broken */
941	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)	1097	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
942	res = sbytes;	1098	res = sbytes;
943	#endif	1099	#endif
944		1100
945	/* according to source inspection, this is correct, and useful behaviour */	1101	/* according to source inspection, this is correct, and useful behaviour */
946	if (sbytes)	1102	if (sbytes)
947	res = sbytes;	1103	res = sbytes;
948	}
949		1104
950	# elif defined (__APPLE__)	1105	# elif defined (__APPLE__)
951
952	{
953	off_t sbytes = count;	1106	off_t sbytes = count;
954	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	1107	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
955		1108
956	/* according to the manpage, sbytes is always valid */	1109	/* according to the manpage, sbytes is always valid */
957	if (sbytes)	1110	if (sbytes)
958	res = sbytes;	1111	res = sbytes;
959	}
960		1112
961	# elif __hpux	1113	# elif __hpux
962	res = sendfile (ofd, ifd, offset, count, 0, 0);	1114	res = sendfile (ofd, ifd, offset, count, 0, 0);
963		1115
964	# elif __solaris	1116	# elif __solaris
965	{
966	struct sendfilevec vec;	1117	struct sendfilevec vec;
967	size_t sbytes;	1118	size_t sbytes;
968		1119
969	vec.sfv_fd = ifd;	1120	vec.sfv_fd = ifd;
970	vec.sfv_flag = 0;	1121	vec.sfv_flag = 0;
971	vec.sfv_off = offset;	1122	vec.sfv_off = offset;
972	vec.sfv_len = count;	1123	vec.sfv_len = count;
973		1124
974	res = sendfilev (ofd, &vec, 1, &sbytes);	1125	res = sendfilev (ofd, &vec, 1, &sbytes);
975		1126
976	if (res < 0 && sbytes)	1127	if (res < 0 && sbytes)
977	res = sbytes;	1128	res = sbytes;
978	}
979		1129
980	# endif	1130	# endif
981		1131
982	#elif defined (_WIN32)	1132	#elif defined (_WIN32) && 0
983
984	/* does not work, just for documentation of what would need to be done */	1133	/* does not work, just for documentation of what would need to be done */
985	{	1134	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1135	/* libeio guarantees that the file offset does not change, and windows */
		1136	/* has no way to get an independent handle to the same file description */
986	HANDLE h = TO_SOCKET (ifd);	1137	HANDLE h = TO_SOCKET (ifd);
987	SetFilePointer (h, offset, 0, FILE_BEGIN);	1138	SetFilePointer (h, offset, 0, FILE_BEGIN);
988	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);	1139	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
989	}
990		1140
991	#else	1141	#else
992	res = -1;	1142	res = -1;
993	errno = ENOSYS;	1143	errno = ENOSYS;
994	#endif	1144	#endif
995		1145
		1146	/* we assume sendfile can copy at least 128mb in one go */
		1147	if (res <= 128 * 1024 * 1024)
		1148	{
		1149	if (res > 0)
		1150	written += res;
		1151
		1152	if (written)
		1153	return written;
		1154
		1155	break;
		1156	}
		1157	else
		1158	{
		1159	/* if we requested more, then probably the kernel was lazy */
		1160	written += res;
		1161	offset += res;
		1162	count -= res;
		1163
		1164	if (!count)
		1165	return written;
		1166	}
		1167	}
		1168
996	if (res < 0	1169	if (res < 0
997	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1170	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
998	/* BSDs */	1171	/* BSDs */
999	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1172	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1000	|| errno == ENOTSUP	1173	|| errno == ENOTSUP
1001	#endif	1174	#endif
		1175	#ifdef EOPNOTSUPP /* windows */
1002	|| errno == EOPNOTSUPP /* BSDs */	1176	|| errno == EOPNOTSUPP /* BSDs */
		1177	#endif
1003	#if __solaris	1178	#if __solaris
1004	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1179	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1005	#endif	1180	#endif
1006	)	1181	)
1007	)	1182	)
…		…
1011		1186
1012	res = 0;	1187	res = 0;
1013		1188
1014	while (count)	1189	while (count)
1015	{	1190	{
1016	ssize_t cnt;	1191	eio_ssize_t cnt;
1017		1192
1018	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1193	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1019		1194
1020	if (cnt <= 0)	1195	if (cnt <= 0)
1021	{	1196	{
…		…
1033		1208
1034	offset += cnt;	1209	offset += cnt;
1035	res += cnt;	1210	res += cnt;
1036	count -= cnt;	1211	count -= cnt;
1037	}	1212	}
		1213
		1214	FUBd;
1038	}	1215	}
1039		1216
1040	return res;	1217	return res;
		1218	}
		1219
		1220	#ifdef PAGESIZE
		1221	# define eio_pagesize() PAGESIZE
		1222	#else
		1223	static intptr_t
		1224	eio_pagesize (void)
		1225	{
		1226	static intptr_t page;
		1227
		1228	if (!page)
		1229	page = sysconf (_SC_PAGESIZE);
		1230
		1231	return page;
		1232	}
		1233	#endif
		1234
		1235	static void
		1236	eio_page_align (void **addr, size_t *length)
		1237	{
		1238	intptr_t mask = eio_pagesize () - 1;
		1239
		1240	/* round down addr */
		1241	intptr_t adj = mask & (intptr_t)*addr;
		1242
		1243	*addr = (void *)((intptr_t)*addr - adj);
		1244	*length += adj;
		1245
		1246	/* round up length */
		1247	*length = (*length + mask) & ~mask;
		1248	}
		1249
		1250	#if !_POSIX_MEMLOCK
		1251	# define eio__mlockall(a) EIO_ENOSYS ()
		1252	#else
		1253
		1254	static int
		1255	eio__mlockall (int flags)
		1256	{
		1257	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1258	extern int mallopt (int, int);
		1259	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1260	#endif
		1261
		1262	if (EIO_MCL_CURRENT != MCL_CURRENT
		1263	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1264	{
		1265	flags = 0
		1266	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1267	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1268	}
		1269
		1270	return mlockall (flags);
		1271	}
		1272	#endif
		1273
		1274	#if !_POSIX_MEMLOCK_RANGE
		1275	# define eio__mlock(a,b) EIO_ENOSYS ()
		1276	#else
		1277
		1278	static int
		1279	eio__mlock (void *addr, size_t length)
		1280	{
		1281	eio_page_align (&addr, &length);
		1282
		1283	return mlock (addr, length);
		1284	}
		1285
		1286	#endif
		1287
		1288	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1289	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1290	#else
		1291
		1292	static int
		1293	eio__msync (void *mem, size_t len, int flags)
		1294	{
		1295	eio_page_align (&mem, &len);
		1296
		1297	if (EIO_MS_ASYNC != MS_SYNC
		1298	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1299	|| EIO_MS_SYNC != MS_SYNC)
		1300	{
		1301	flags = 0
		1302	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1303	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1304	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1305	}
		1306
		1307	return msync (mem, len, flags);
		1308	}
		1309
		1310	#endif
		1311
		1312	static int
		1313	eio__mtouch (eio_req *req)
		1314	{
		1315	void *mem = req->ptr2;
		1316	size_t len = req->size;
		1317	int flags = req->int1;
		1318
		1319	eio_page_align (&mem, &len);
		1320
		1321	{
		1322	intptr_t addr = (intptr_t)mem;
		1323	intptr_t end = addr + len;
		1324	intptr_t page = eio_pagesize ();
		1325
		1326	if (addr < end)
		1327	if (flags & EIO_MT_MODIFY) /* modify */
		1328	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1329	else
		1330	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1331	}
		1332
		1333	return 0;
		1334	}
		1335
		1336	/*****************************************************************************/
		1337	/* requests implemented outside eio_execute, because they are so large */
		1338
		1339	static void
		1340	eio__realpath (eio_req *req, etp_worker *self)
		1341	{
		1342	char *rel = req->ptr1;
		1343	char *res;
		1344	char *tmp1, *tmp2;
		1345	#if SYMLOOP_MAX > 32
		1346	int symlinks = SYMLOOP_MAX;
		1347	#else
		1348	int symlinks = 32;
		1349	#endif
		1350
		1351	req->result = -1;
		1352
		1353	errno = EINVAL;
		1354	if (!rel)
		1355	return;
		1356
		1357	errno = ENOENT;
		1358	if (!*rel)
		1359	return;
		1360
		1361	if (!req->ptr2)
		1362	{
		1363	X_LOCK (wrklock);
		1364	req->flags |= EIO_FLAG_PTR2_FREE;
		1365	X_UNLOCK (wrklock);
		1366	req->ptr2 = malloc (PATH_MAX * 3);
		1367
		1368	errno = ENOMEM;
		1369	if (!req->ptr2)
		1370	return;
		1371	}
		1372
		1373	res = req->ptr2;
		1374	tmp1 = res + PATH_MAX;
		1375	tmp2 = tmp1 + PATH_MAX;
		1376
		1377	#if 0 /* disabled, the musl way to do things is just too racy */
		1378	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1379	/* on linux we may be able to ask the kernel */
		1380	{
		1381	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1382
		1383	if (fd >= 0)
		1384	{
		1385	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1386	req->result = readlink (tmp1, res, PATH_MAX);
		1387	close (fd);
		1388
		1389	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1390
		1391	if (req->result > 0)
		1392	goto done;
		1393	}
		1394	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1395	return;
		1396	}
		1397	#endif
		1398	#endif
		1399
		1400	if (*rel != '/')
		1401	{
		1402	if (!getcwd (res, PATH_MAX))
		1403	return;
		1404
		1405	if (res [1]) /* only use if not / */
		1406	res += strlen (res);
		1407	}
		1408
		1409	while (*rel)
		1410	{
		1411	eio_ssize_t len, linklen;
		1412	char *beg = rel;
		1413
		1414	while (*rel && *rel != '/')
		1415	++rel;
		1416
		1417	len = rel - beg;
		1418
		1419	if (!len) /* skip slashes */
		1420	{
		1421	++rel;
		1422	continue;
		1423	}
		1424
		1425	if (beg [0] == '.')
		1426	{
		1427	if (len == 1)
		1428	continue; /* . - nop */
		1429
		1430	if (beg [1] == '.' && len == 2)
		1431	{
		1432	/* .. - back up one component, if possible */
		1433
		1434	while (res != req->ptr2)
		1435	if (*--res == '/')
		1436	break;
		1437
		1438	continue;
		1439	}
		1440	}
		1441
		1442	errno = ENAMETOOLONG;
		1443	if (res + 1 + len + 1 >= tmp1)
		1444	return;
		1445
		1446	/* copy one component */
		1447	*res = '/';
		1448	memcpy (res + 1, beg, len);
		1449
		1450	/* zero-terminate, for readlink */
		1451	res [len + 1] = 0;
		1452
		1453	/* now check if it's a symlink */
		1454	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1455
		1456	if (linklen < 0)
		1457	{
		1458	if (errno != EINVAL)
		1459	return;
		1460
		1461	/* it's a normal directory. hopefully */
		1462	res += len + 1;
		1463	}
		1464	else
		1465	{
		1466	/* yay, it was a symlink - build new path in tmp2 */
		1467	int rellen = strlen (rel);
		1468
		1469	errno = ENAMETOOLONG;
		1470	if (linklen + 1 + rellen >= PATH_MAX)
		1471	return;
		1472
		1473	errno = ELOOP;
		1474	if (!--symlinks)
		1475	return;
		1476
		1477	if (*tmp1 == '/')
		1478	res = req->ptr2; /* symlink resolves to an absolute path */
		1479
		1480	/* we need to be careful, as rel might point into tmp2 already */
		1481	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1482	tmp2 [linklen] = '/';
		1483	memcpy (tmp2, tmp1, linklen);
		1484
		1485	rel = tmp2;
		1486	}
		1487	}
		1488
		1489	/* special case for the lone root path */
		1490	if (res == req->ptr2)
		1491	*res++ = '/';
		1492
		1493	req->result = res - (char *)req->ptr2;
		1494
		1495	done:
		1496	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1041	}	1497	}
1042		1498
1043	static signed char	1499	static signed char
1044	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1500	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1045	{	1501	{
1046	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */	1502	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1047	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;	1503	: a->inode < b->inode ? -1
		1504	: a->inode > b->inode ? 1
		1505	: 0;
1048	}	1506	}
1049		1507
1050	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0	1508	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1051		1509
1052	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1510	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1053	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1511	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1054		1512
1055	static void	1513	static void
1056	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1514	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1057	{	1515	{
1058	unsigned char bits [9 + sizeof (ino_t) * 8];	1516	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1059	unsigned char *bit = bits;	1517	unsigned char *bit = bits;
1060		1518
1061	assert (CHAR_BIT == 8);	1519	assert (CHAR_BIT == 8);
1062	assert (sizeof (eio_dirent) * 8 < 256);	1520	assert (sizeof (eio_dirent) * 8 < 256);
1063	assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */	1521	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1064	assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */	1522	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1065		1523
1066	if (size <= EIO_SORT_FAST)	1524	if (size <= EIO_SORT_FAST)
1067	return;	1525	return;
1068		1526
1069	/* first prepare an array of bits to test in our radix sort */	1527	/* first prepare an array of bits to test in our radix sort */
1070	/* try to take endianness into account, as well as differences in ino_t sizes */	1528	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1071	/* inode_bits must contain all inodes ORed together */	1529	/* inode_bits must contain all inodes ORed together */
1072	/* which is used to skip bits that are 0 everywhere, which is very common */	1530	/* which is used to skip bits that are 0 everywhere, which is very common */
1073	{	1531	{
1074	ino_t endianness;	1532	eio_ino_t endianness;
1075	int i, j;	1533	int i, j;
1076		1534
1077	/* we store the byte offset of byte n into byte n of "endianness" */	1535	/* we store the byte offset of byte n into byte n of "endianness" */
1078	for (i = 0; i < sizeof (ino_t); ++i)	1536	for (i = 0; i < sizeof (eio_ino_t); ++i)
1079	((unsigned char *)&endianness)[i] = i;	1537	((unsigned char *)&endianness)[i] = i;
1080		1538
1081	*bit++ = 0;	1539	*bit++ = 0;
1082		1540
1083	for (i = 0; i < sizeof (ino_t); ++i)	1541	for (i = 0; i < sizeof (eio_ino_t); ++i)
1084	{	1542	{
1085	/* shifting off the byte offsets out of "endianness" */	1543	/* shifting off the byte offsets out of "endianness" */
1086	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1544	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1087	endianness >>= 8;	1545	endianness >>= 8;
1088		1546
1089	for (j = 0; j < 8; ++j)	1547	for (j = 0; j < 8; ++j)
1090	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1548	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1091	*bit++ = offs + j;	1549	*bit++ = offs + j;
1092	}	1550	}
1093		1551
1094	for (j = 0; j < 8; ++j)	1552	for (j = 0; j < 8; ++j)
1095	if (score_bits & (1 << j))	1553	if (score_bits & (1 << j))
1096	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1554	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1097	}	1555	}
1098		1556
1099	/* now actually do the sorting (a variant of MSD radix sort) */	1557	/* now actually do the sorting (a variant of MSD radix sort) */
1100	{	1558	{
1101	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1559	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1102	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1560	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1103	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1561	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1104	int stk_idx = 0;	1562	int stk_idx = 0;
1105		1563
1106	base_stk [stk_idx] = dents;	1564	base_stk [stk_idx] = dents;
1107	end_stk [stk_idx] = dents + size;	1565	end_stk [stk_idx] = dents + size;
1108	bit_stk [stk_idx] = bit - 1;	1566	bit_stk [stk_idx] = bit - 1;
…		…
1187	}	1645	}
1188	}	1646	}
1189	}	1647	}
1190		1648
1191	static void	1649	static void
1192	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1650	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1193	{	1651	{
1194	if (size <= 1)	1652	if (size <= 1)
1195	return; /* our insertion sort relies on size > 0 */	1653	return; /* our insertion sort relies on size > 0 */
1196		1654
1197	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1655	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1205		1663
1206	/* read a full directory */	1664	/* read a full directory */
1207	static void	1665	static void
1208	eio__scandir (eio_req *req, etp_worker *self)	1666	eio__scandir (eio_req *req, etp_worker *self)
1209	{	1667	{
1210	DIR *dirp;
1211	EIO_STRUCT_DIRENT *entp;
1212	char *name, *names;	1668	char *name, *names;
1213	int namesalloc = 4096;	1669	int namesalloc = 4096 - sizeof (void *) * 4;
1214	int namesoffs = 0;	1670	int namesoffs = 0;
1215	int flags = req->int1;	1671	int flags = req->int1;
1216	eio_dirent *dents = 0;	1672	eio_dirent *dents = 0;
1217	int dentalloc = 128;	1673	int dentalloc = 128;
1218	int dentoffs = 0;	1674	int dentoffs = 0;
1219	ino_t inode_bits = 0;	1675	eio_ino_t inode_bits = 0;
		1676	#ifdef _WIN32
		1677	HANDLE dirp;
		1678	WIN32_FIND_DATA entp;
		1679	#else
		1680	DIR *dirp;
		1681	EIO_STRUCT_DIRENT *entp;
		1682	#endif
1220		1683
1221	req->result = -1;	1684	req->result = -1;
1222		1685
1223	if (!(flags & EIO_READDIR_DENTS))	1686	if (!(flags & EIO_READDIR_DENTS))
1224	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1687	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1225		1688
1226	X_LOCK (wrklock);	1689	#ifdef _WIN32
1227	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1690	{
		1691	int len = strlen ((const char *)req->ptr1);
		1692	char *path = malloc (MAX_PATH);
		1693	const char *fmt;
		1694
		1695	if (!len)
		1696	fmt = "./*";
		1697	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1698	fmt = "%s*";
		1699	else
		1700	fmt = "%s/*";
		1701
		1702	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1703	dirp = FindFirstFile (path, &entp);
		1704	free (path);
		1705
		1706	if (dirp == INVALID_HANDLE_VALUE)
		1707	{
		1708	dirp = 0;
		1709
		1710	/* should steal _dosmaperr */
		1711	switch (GetLastError ())
		1712	{
		1713	case ERROR_FILE_NOT_FOUND:
		1714	req->result = 0;
		1715	break;
		1716
		1717	case ERROR_INVALID_NAME:
		1718	case ERROR_PATH_NOT_FOUND:
		1719	case ERROR_NO_MORE_FILES:
		1720	errno = ENOENT;
		1721	break;
		1722
		1723	case ERROR_NOT_ENOUGH_MEMORY:
		1724	errno = ENOMEM;
		1725	break;
		1726
		1727	default:
		1728	errno = EINVAL;
		1729	break;
		1730	}
		1731	}
		1732	}
		1733	#else
1228	self->dirp = dirp = opendir (req->ptr1);	1734	dirp = opendir (req->ptr1);
		1735	#endif
		1736
		1737	if (req->flags & EIO_FLAG_PTR1_FREE)
		1738	free (req->ptr1);
		1739
1229	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1740	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1230	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1741	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1231	req->ptr2 = names = malloc (namesalloc);	1742	req->ptr2 = names = malloc (namesalloc);
1232	X_UNLOCK (wrklock);
1233		1743
1234	if (dirp && names && (!flags || dents))	1744	if (dirp && names && (!flags || dents))
1235	for (;;)	1745	for (;;)
1236	{	1746	{
		1747	int done;
		1748
		1749	#ifdef _WIN32
		1750	done = !dirp;
		1751	#else
1237	errno = 0;	1752	errno = 0;
1238	entp = readdir (dirp);	1753	entp = readdir (dirp);
		1754	done = !entp;
		1755	#endif
1239		1756
1240	if (!entp)	1757	if (done)
1241	{	1758	{
		1759	#ifndef _WIN32
		1760	int old_errno = errno;
		1761	closedir (dirp);
		1762	errno = old_errno;
		1763
1242	if (errno)	1764	if (errno)
1243	break;	1765	break;
		1766	#endif
1244		1767
1245	/* sort etc. */	1768	/* sort etc. */
1246	req->int1 = flags;	1769	req->int1 = flags;
1247	req->result = dentoffs;	1770	req->result = dentoffs;
1248		1771
1249	if (flags & EIO_READDIR_STAT_ORDER)	1772	if (flags & EIO_READDIR_STAT_ORDER)
1250	eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */	1773	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1251	else if (flags & EIO_READDIR_DIRS_FIRST)	1774	else if (flags & EIO_READDIR_DIRS_FIRST)
1252	if (flags & EIO_READDIR_FOUND_UNKNOWN)	1775	if (flags & EIO_READDIR_FOUND_UNKNOWN)
1253	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */	1776	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1254	else	1777	else
1255	{	1778	{
…		…
1257	eio_dirent *oth = dents + dentoffs;	1780	eio_dirent *oth = dents + dentoffs;
1258	eio_dirent *dir = dents;	1781	eio_dirent *dir = dents;
1259		1782
1260	/* now partition dirs to the front, and non-dirs to the back */	1783	/* now partition dirs to the front, and non-dirs to the back */
1261	/* by walking from both sides and swapping if necessary */	1784	/* by walking from both sides and swapping if necessary */
1262	/* also clear score, so it doesn't influence sorting */
1263	while (oth > dir)	1785	while (oth > dir)
1264	{	1786	{
1265	if (dir->type == EIO_DT_DIR)	1787	if (dir->type == EIO_DT_DIR)
1266	++dir;	1788	++dir;
1267	else if ((--oth)->type == EIO_DT_DIR)	1789	else if ((--oth)->type == EIO_DT_DIR)
…		…
1270		1792
1271	++dir;	1793	++dir;
1272	}	1794	}
1273	}	1795	}
1274		1796
1275	/* now sort the dirs only */	1797	/* now sort the dirs only (dirs all have the same score) */
1276	eio_dent_sort (dents, dir - dents, 0, inode_bits);	1798	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1277	}	1799	}
1278		1800
1279	break;	1801	break;
1280	}	1802	}
1281		1803
1282	/* now add the entry to our list(s) */	1804	/* now add the entry to our list(s) */
1283	name = entp->d_name;	1805	name = D_NAME (entp);
1284		1806
1285	/* skip . and .. entries */	1807	/* skip . and .. entries */
1286	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1808	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1287	{	1809	{
1288	int len = D_NAMLEN (entp) + 1;	1810	int len = D_NAMLEN (entp) + 1;
1289		1811
1290	while (expect_false (namesoffs + len > namesalloc))	1812	while (ecb_expect_false (namesoffs + len > namesalloc))
1291	{	1813	{
1292	namesalloc *= 2;	1814	namesalloc *= 2;
1293	X_LOCK (wrklock);
1294	req->ptr2 = names = realloc (names, namesalloc);	1815	req->ptr2 = names = realloc (names, namesalloc);
1295	X_UNLOCK (wrklock);
1296		1816
1297	if (!names)	1817	if (!names)
1298	break;	1818	break;
1299	}	1819	}
1300		1820
…		…
1302		1822
1303	if (dents)	1823	if (dents)
1304	{	1824	{
1305	struct eio_dirent *ent;	1825	struct eio_dirent *ent;
1306		1826
1307	if (expect_false (dentoffs == dentalloc))	1827	if (ecb_expect_false (dentoffs == dentalloc))
1308	{	1828	{
1309	dentalloc *= 2;	1829	dentalloc *= 2;
1310	X_LOCK (wrklock);
1311	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1830	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1312	X_UNLOCK (wrklock);
1313		1831
1314	if (!dents)	1832	if (!dents)
1315	break;	1833	break;
1316	}	1834	}
1317		1835
…		…
1397	if (EIO_CANCELLED (req))	1915	if (EIO_CANCELLED (req))
1398	{	1916	{
1399	errno = ECANCELED;	1917	errno = ECANCELED;
1400	break;	1918	break;
1401	}	1919	}
		1920
		1921	#ifdef _WIN32
		1922	if (!FindNextFile (dirp, &entp))
		1923	{
		1924	FindClose (dirp);
		1925	dirp = 0;
		1926	}
		1927	#endif
1402	}	1928	}
1403	}
1404
1405	#ifdef PAGESIZE
1406	# define eio_pagesize() PAGESIZE
1407	#else
1408	static intptr_t
1409	eio_pagesize (void)
1410	{
1411	static intptr_t page;
1412
1413	if (!page)
1414	page = sysconf (_SC_PAGESIZE);
1415
1416	return page;
1417	}
1418	#endif
1419
1420	static void
1421	eio_page_align (void **addr, size_t *length)
1422	{
1423	intptr_t mask = eio_pagesize () - 1;
1424
1425	/* round down addr */
1426	intptr_t adj = mask & (intptr_t)*addr;
1427
1428	*addr = (void *)((intptr_t)*addr - adj);
1429	*length += adj;
1430
1431	/* round up length */
1432	*length = (*length + mask) & ~mask;
1433	}
1434
1435	#if !_POSIX_MEMLOCK
1436	# define eio__mlockall(a) ((errno = ENOSYS), -1)
1437	#else
1438
1439	static int
1440	eio__mlockall (int flags)
1441	{
1442	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1443	extern int mallopt (int, int);
1444	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1445	#endif
1446
1447	if (EIO_MCL_CURRENT != MCL_CURRENT
1448	|| EIO_MCL_FUTURE != MCL_FUTURE)
1449	{
1450	flags = 0
1451	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1452	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1453	}
1454
1455	return mlockall (flags);
1456	}
1457	#endif
1458
1459	#if !_POSIX_MEMLOCK_RANGE
1460	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1461	#else
1462
1463	static int
1464	eio__mlock (void *addr, size_t length)
1465	{
1466	eio_page_align (&addr, &length);
1467
1468	return mlock (addr, length);
1469	}
1470
1471	#endif
1472
1473	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1474	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1475	#else
1476
1477	int
1478	eio__msync (void *mem, size_t len, int flags)
1479	{
1480	eio_page_align (&mem, &len);
1481
1482	if (EIO_MS_ASYNC != MS_SYNC
1483	|| EIO_MS_INVALIDATE != MS_INVALIDATE
1484	|| EIO_MS_SYNC != MS_SYNC)
1485	{
1486	flags = 0
1487	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1488	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1489	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1490	}
1491
1492	return msync (mem, len, flags);
1493	}
1494
1495	#endif
1496
1497	int
1498	eio__mtouch (void *mem, size_t len, int flags)
1499	{
1500	eio_page_align (&mem, &len);
1501
1502	{
1503	intptr_t addr = (intptr_t)mem;
1504	intptr_t end = addr + len;
1505	intptr_t page = eio_pagesize ();
1506
1507	if (addr < end)
1508	if (flags & EIO_MT_MODIFY) /* modify */
1509	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1510	else
1511	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1512	}
1513
1514	return 0;
1515	}	1929	}
1516		1930
1517	/*****************************************************************************/	1931	/*****************************************************************************/
1518		1932
1519	#define ALLOC(len) \	1933	#define ALLOC(len) \
…		…
1535	{	1949	{
1536	ETP_REQ *req;	1950	ETP_REQ *req;
1537	struct timespec ts;	1951	struct timespec ts;
1538	etp_worker *self = (etp_worker *)thr_arg;	1952	etp_worker *self = (etp_worker *)thr_arg;
1539		1953
		1954	#if HAVE_PRCTL_SET_NAME
		1955	prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
		1956	#endif
		1957
1540	/* try to distribute timeouts somewhat randomly */	1958	/* try to distribute timeouts somewhat evenly */
1541	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1959	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1542		1960
1543	for (;;)	1961	for (;;)
1544	{	1962	{
		1963	ts.tv_sec = 0;
		1964
1545	X_LOCK (reqlock);	1965	X_LOCK (reqlock);
1546		1966
1547	for (;;)	1967	for (;;)
1548	{	1968	{
1549	self->req = req = reqq_shift (&req_queue);	1969	self->req = req = reqq_shift (&req_queue);
1550		1970
1551	if (req)	1971	if (req)
1552	break;	1972	break;
1553		1973
		1974	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1975	{
		1976	X_UNLOCK (reqlock);
		1977	X_LOCK (wrklock);
		1978	--started;
		1979	X_UNLOCK (wrklock);
		1980	goto quit;
		1981	}
		1982
1554	++idle;	1983	++idle;
1555		1984
1556	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1985	if (idle <= max_idle)
1557	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1986	/* we are allowed to idle, so do so without any timeout */
		1987	X_COND_WAIT (reqwait, reqlock);
		1988	else
1558	{	1989	{
1559	if (idle > max_idle)	1990	/* initialise timeout once */
1560	{	1991	if (!ts.tv_sec)
1561	--idle;	1992	ts.tv_sec = time (0) + idle_timeout;
1562	X_UNLOCK (reqlock);
1563	X_LOCK (wrklock);
1564	--started;
1565	X_UNLOCK (wrklock);
1566	goto quit;
1567	}
1568		1993
1569	/* we are allowed to idle, so do so without any timeout */
1570	X_COND_WAIT (reqwait, reqlock);	1994	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1995	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1571	}	1996	}
1572		1997
1573	--idle;	1998	--idle;
1574	}	1999	}
1575		2000
…		…
1578	X_UNLOCK (reqlock);	2003	X_UNLOCK (reqlock);
1579		2004
1580	if (req->type < 0)	2005	if (req->type < 0)
1581	goto quit;	2006	goto quit;
1582		2007
1583	if (!EIO_CANCELLED (req))
1584	ETP_EXECUTE (self, req);	2008	ETP_EXECUTE (self, req);
1585		2009
1586	X_LOCK (reslock);	2010	X_LOCK (reslock);
1587		2011
1588	++npending;	2012	++npending;
1589		2013
…		…
1604	return 0;	2028	return 0;
1605	}	2029	}
1606		2030
1607	/*****************************************************************************/	2031	/*****************************************************************************/
1608		2032
		2033	int ecb_cold
1609	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	2034	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1610	{	2035	{
		2036	#if !HAVE_PREADWRITE
		2037	X_MUTEX_CREATE (preadwritelock);
		2038	#endif
		2039
1611	return etp_init (want_poll, done_poll);	2040	return etp_init (want_poll, done_poll);
1612	}	2041	}
1613		2042
		2043	ecb_inline void
1614	static void eio_api_destroy (eio_req *req)	2044	eio_api_destroy (eio_req *req)
1615	{	2045	{
1616	free (req);	2046	free (req);
1617	}	2047	}
1618		2048
1619	#define REQ(rtype) \	2049	#define REQ(rtype) \
…		…
1638	{ \	2068	{ \
1639	eio_api_destroy (req); \	2069	eio_api_destroy (req); \
1640	return 0; \	2070	return 0; \
1641	}	2071	}
1642		2072
		2073	static void
1643	static void eio_execute (etp_worker *self, eio_req *req)	2074	eio_execute (etp_worker *self, eio_req *req)
1644	{	2075	{
		2076	if (ecb_expect_false (EIO_CANCELLED (req)))
		2077	{
		2078	req->result = -1;
		2079	req->errorno = ECANCELED;
		2080	return;
		2081	}
		2082
1645	switch (req->type)	2083	switch (req->type)
1646	{	2084	{
1647	case EIO_READ: ALLOC (req->size);	2085	case EIO_READ: ALLOC (req->size);
1648	req->result = req->offs >= 0	2086	req->result = req->offs >= 0
1649	? pread (req->int1, req->ptr2, req->size, req->offs)	2087	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1651	case EIO_WRITE: req->result = req->offs >= 0	2089	case EIO_WRITE: req->result = req->offs >= 0
1652	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2090	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1653	: write (req->int1, req->ptr2, req->size); break;	2091	: write (req->int1, req->ptr2, req->size); break;
1654		2092
1655	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2093	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1656	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2094	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1657		2095
1658	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2096	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1659	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2097	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1660	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2098	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1661	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2099	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
…		…
1681	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	2119	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1682	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	2120	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1683	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2121	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1684	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2122	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1685	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2123	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1686	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;	2124	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2125
		2126	case EIO_REALPATH: eio__realpath (req, self); break;
1687		2127
1688	case EIO_READLINK: ALLOC (PATH_MAX);	2128	case EIO_READLINK: ALLOC (PATH_MAX);
1689	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;	2129	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1690		2130
1691	case EIO_SYNC: req->result = 0; sync (); break;	2131	case EIO_SYNC: req->result = 0; sync (); break;
1692	case EIO_FSYNC: req->result = fsync (req->int1); break;	2132	case EIO_FSYNC: req->result = fsync (req->int1); break;
1693	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2133	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2134	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2135	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1694	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2136	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1695	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	2137	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1696	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2138	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1697	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2139	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1698	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	2140	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1699		2141
1700	case EIO_READDIR: eio__scandir (req, self); break;	2142	case EIO_READDIR: eio__scandir (req, self); break;
1701		2143
1702	case EIO_BUSY:	2144	case EIO_BUSY:
1703	#ifdef _WIN32	2145	#ifdef _WIN32
…		…
1744	case EIO_NOP:	2186	case EIO_NOP:
1745	req->result = 0;	2187	req->result = 0;
1746	break;	2188	break;
1747		2189
1748	case EIO_CUSTOM:	2190	case EIO_CUSTOM:
1749	((void (*)(eio_req *))req->feed) (req);	2191	req->feed (req);
1750	break;	2192	break;
1751		2193
1752	default:	2194	default:
1753	errno = ENOSYS;	2195	errno = ENOSYS;
1754	req->result = -1;	2196	req->result = -1;
…		…
1783	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2225	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1784	{	2226	{
1785	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2227	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1786	}	2228	}
1787		2229
		2230	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		2231	{
		2232	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2233	}
		2234
		2235	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2236	{
		2237	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2238	}
		2239
		2240	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2241	{
		2242	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2243	}
		2244
1788	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2245	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1789	{	2246	{
1790	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2247	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1791	}	2248	}
1792		2249
…		…
1798	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)	2255	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1799	{	2256	{
1800	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;	2257	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1801	}	2258	}
1802		2259
1803	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2260	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1804	{	2261	{
1805	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2262	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1806	}
1807
1808	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1809	{
1810	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1811	}	2263	}
1812		2264
1813	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2265	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1814	{	2266	{
1815	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2267	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
1853	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2305	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1854	{	2306	{
1855	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2307	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1856	}	2308	}
1857		2309
1858	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2310	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1859	{	2311	{
1860	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2312	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1861	}	2313	}
1862		2314
1863	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2315	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1883	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2335	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1884	{	2336	{
1885	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2337	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1886	}	2338	}
1887		2339
1888	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2340	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1889	{	2341	{
1890	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2342	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1891	}	2343	}
1892		2344
1893	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2345	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1909	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2361	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1910	{	2362	{
1911	return eio__1path (EIO_READLINK, path, pri, cb, data);	2363	return eio__1path (EIO_READLINK, path, pri, cb, data);
1912	}	2364	}
1913		2365
		2366	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2367	{
		2368	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2369	}
		2370
1914	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2371	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1915	{	2372	{
1916	return eio__1path (EIO_STAT, path, pri, cb, data);	2373	return eio__1path (EIO_STAT, path, pri, cb, data);
1917	}	2374	}
1918		2375
…		…
1941	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2398	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1942	}	2399	}
1943		2400
1944	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2401	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1945	{	2402	{
1946	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2403	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1947	}	2404	}
1948		2405
1949	static eio_req *	2406	static eio_req *
1950	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2407	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1951	{	2408	{
…		…
1975	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2432	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1976	{	2433	{
1977	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2434	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1978	}	2435	}
1979		2436
1980	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2437	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1981	{	2438	{
1982	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2439	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1983	}	2440	}
1984		2441
1985	#endif	2442	#endif
1986		2443
1987	eio_req *eio_grp (eio_cb cb, void *data)	2444	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1996	#undef SEND	2453	#undef SEND
1997		2454
1998	/*****************************************************************************/	2455	/*****************************************************************************/
1999	/* grp functions */	2456	/* grp functions */
2000		2457
		2458	void
2001	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2459	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2002	{	2460	{
2003	grp->int2 = limit;	2461	grp->int2 = limit;
2004	grp->feed = feed;	2462	grp->feed = feed;
2005		2463
2006	grp_try_feed (grp);	2464	grp_try_feed (grp);
2007	}	2465	}
2008		2466
		2467	void
2009	void eio_grp_limit (eio_req *grp, int limit)	2468	eio_grp_limit (eio_req *grp, int limit)
2010	{	2469	{
2011	grp->int2 = limit;	2470	grp->int2 = limit;
2012		2471
2013	grp_try_feed (grp);	2472	grp_try_feed (grp);
2014	}	2473	}
2015		2474
		2475	void
2016	void eio_grp_add (eio_req *grp, eio_req *req)	2476	eio_grp_add (eio_req *grp, eio_req *req)
2017	{	2477	{
2018	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2478	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2019		2479
2020	grp->flags |= EIO_FLAG_GROUPADD;	2480	grp->flags |= EIO_FLAG_GROUPADD;
2021		2481
…		…
2032	}	2492	}
2033		2493
2034	/*****************************************************************************/	2494	/*****************************************************************************/
2035	/* misc garbage */	2495	/* misc garbage */
2036		2496
		2497	eio_ssize_t
2037	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2498	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2038	{	2499	{
2039	etp_worker wrk;
2040	ssize_t ret;
2041
2042	wrk.dbuf = 0;
2043
2044	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2500	return eio__sendfile (ofd, ifd, offset, count);
2045
2046	if (wrk.dbuf)
2047	free (wrk.dbuf);
2048
2049	return ret;
2050	}	2501	}
2051		2502

Diff Legend

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