ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/cvsroot/libeio/eio.c

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.38 by root, Fri Jun 12 20:01:42 2009 UTC vs.
Revision 1.96 by root, Tue Jul 19 05:20:36 2011 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008,2009 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"
…		…
53	#include <sys/stat.h>	58	#include <sys/stat.h>
54	#include <limits.h>	59	#include <limits.h>
55	#include <fcntl.h>	60	#include <fcntl.h>
56	#include <assert.h>	61	#include <assert.h>
57		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
58	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
59	# 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
60	#endif	84	#endif
61		85
62	#ifndef EIO_DESTROY	86	#ifndef EIO_DESTROY
…		…
65		89
66	#ifndef EIO_FEED	90	#ifndef EIO_FEED
67	# 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)
68	#endif	92	#endif
69		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
70	#ifdef _WIN32	105	#ifdef _WIN32
71		106
72	/*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	/* we could even stat and see if it exists */
		135	static int
		136	symlink (const char *old, const char *neu)
		137	{
		138	#if WINVER >= 0x0600
		139	if (CreateSymbolicLink (neu, old, 1))
		140	return 0;
		141
		142	if (CreateSymbolicLink (neu, old, 0))
		143	return 0;
		144	#endif
		145
		146	return EIO_ERRNO (ENOENT, -1);
		147	}
		148
		149	/* POSIX API only */
		150	#define CreateHardLink(neu,old,flags) 0
		151	#define CreateSymbolicLink(neu,old,flags) 0
		152
		153	struct statvfs
		154	{
		155	int dummy;
		156	};
		157
		158	#define DT_DIR EIO_DT_DIR
		159	#define DT_REG EIO_DT_REG
		160	#define D_NAME(entp) entp.cFileName
		161	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
		162
73	#else	163	#else
74		164
75	# include "config.h"
76	# include <sys/time.h>	165	#include <sys/time.h>
77	# include <sys/select.h>	166	#include <sys/select.h>
78	# include <sys/mman.h>	167	#include <sys/statvfs.h>
79	# include <unistd.h>	168	#include <unistd.h>
		169	#include <signal.h>
		170	#include <dirent.h>
		171
		172	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		173	#include <sys/mman.h>
		174	#endif
		175
		176	#define D_NAME(entp) entp->d_name
		177
		178	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		179	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		180	#define _DIRENT_HAVE_D_TYPE /* sigh */
		181	#define D_INO(de) (de)->d_fileno
		182	#define D_NAMLEN(de) (de)->d_namlen
		183	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		184	#define D_INO(de) (de)->d_ino
		185	#endif
		186
		187	#ifdef _D_EXACT_NAMLEN
		188	#undef D_NAMLEN
		189	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		190	#endif
		191
		192	#ifdef _DIRENT_HAVE_D_TYPE
		193	#define D_TYPE(de) (de)->d_type
		194	#endif
		195
		196	#ifndef EIO_STRUCT_DIRENT
		197	#define EIO_STRUCT_DIRENT struct dirent
		198	#endif
		199
		200	#endif
		201
		202	#if HAVE_UTIMES
80	# include <utime.h>	203	# include <utime.h>
81	# include <signal.h>
82	# include <dirent.h>
83
84	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85	# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
86	# define _DIRENT_HAVE_D_TYPE /* sigh */
87	# define D_INO(de) (de)->d_fileno
88	# define D_NAMLEN(de) (de)->d_namlen
89	# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
90	# define D_INO(de) (de)->d_ino
91	# endif
92
93	#ifdef _D_EXACT_NAMLEN
94	# undef D_NAMLEN
95	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96	#endif
97
98	# ifdef _DIRENT_HAVE_D_TYPE
99	# define D_TYPE(de) (de)->d_type
100	# endif
101
102	# ifndef EIO_STRUCT_DIRENT
103	# define EIO_STRUCT_DIRENT struct dirent
104	# endif
105
106	#endif	204	#endif
107		205
108	#if HAVE_SENDFILE	206	#if HAVE_SENDFILE
109	# if __linux	207	# if __linux
110	# include <sys/sendfile.h>	208	# include <sys/sendfile.h>
111	# elif __freebsd	209	# elif __FreeBSD__ || defined __APPLE__
112	# include <sys/socket.h>	210	# include <sys/socket.h>
113	# include <sys/uio.h>	211	# include <sys/uio.h>
114	# elif __hpux	212	# elif __hpux
115	# include <sys/socket.h>	213	# include <sys/socket.h>
116	# elif __solaris /* not yet */	214	# elif __solaris
117	# include <sys/sendfile.h>	215	# include <sys/sendfile.h>
118	# else	216	# else
119	# error sendfile support requested but not available	217	# error sendfile support requested but not available
120	# endif	218	# endif
121	#endif	219	#endif
…		…
125	#endif	223	#endif
126	#ifndef D_INO	224	#ifndef D_INO
127	# define D_INO(de) 0	225	# define D_INO(de) 0
128	#endif	226	#endif
129	#ifndef D_NAMLEN	227	#ifndef D_NAMLEN
130	# define D_NAMLEN(de) strlen ((de)->d_name)	228	# define D_NAMLEN(entp) strlen (D_NAME (entp))
131	#endif	229	#endif
132
133	/* number of seconds after which an idle threads exit */
134	#define IDLE_TIMEOUT 10
135		230
136	/* used for struct dirent, AIX doesn't provide it */	231	/* used for struct dirent, AIX doesn't provide it */
137	#ifndef NAME_MAX	232	#ifndef NAME_MAX
138	# define NAME_MAX 4096	233	# define NAME_MAX 4096
139	#endif	234	#endif
140		235
		236	/* used for readlink etc. */
		237	#ifndef PATH_MAX
		238	# define PATH_MAX 4096
		239	#endif
		240
141	/* buffer size for various temporary buffers */	241	/* buffer size for various temporary buffers */
142	#define EIO_BUFSIZE 65536	242	#define EIO_BUFSIZE 65536
143		243
144	#define dBUF \	244	#define dBUF \
145	char *eio_buf; \
146	ETP_WORKER_LOCK (self); \
147	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	245	char *eio_buf = malloc (EIO_BUFSIZE); \
148	ETP_WORKER_UNLOCK (self); \
149	errno = ENOMEM; \	246	errno = ENOMEM; \
150	if (!eio_buf) \	247	if (!eio_buf) \
151	return -1;	248	return -1
		249
		250	#define FUBd \
		251	free (eio_buf)
152		252
153	#define EIO_TICKS ((1000000 + 1023) >> 10)	253	#define EIO_TICKS ((1000000 + 1023) >> 10)
154
155	/*****************************************************************************/
156
157	#if __GNUC__ >= 3
158	# define expect(expr,value) __builtin_expect ((expr),(value))
159	#else
160	# define expect(expr,value) (expr)
161	#endif
162
163	#define expect_false(expr) expect ((expr) != 0, 0)
164	#define expect_true(expr) expect ((expr) != 0, 1)
165
166	/*****************************************************************************/
167		254
168	#define ETP_PRI_MIN EIO_PRI_MIN	255	#define ETP_PRI_MIN EIO_PRI_MIN
169	#define ETP_PRI_MAX EIO_PRI_MAX	256	#define ETP_PRI_MAX EIO_PRI_MAX
170		257
171	struct etp_worker;	258	struct etp_worker;
…		…
175	static int eio_finish (eio_req *req);	262	static int eio_finish (eio_req *req);
176	#define ETP_FINISH(req) eio_finish (req)	263	#define ETP_FINISH(req) eio_finish (req)
177	static void eio_execute (struct etp_worker *self, eio_req *req);	264	static void eio_execute (struct etp_worker *self, eio_req *req);
178	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	265	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
179		266
180	#define ETP_WORKER_CLEAR(req) \
181	if (wrk->dbuf) \
182	{ \
183	free (wrk->dbuf); \
184	wrk->dbuf = 0; \
185	} \
186	\
187	if (wrk->dirp) \
188	{ \
189	closedir (wrk->dirp); \
190	wrk->dirp = 0; \
191	}
192
193	#define ETP_WORKER_COMMON \
194	void *dbuf; \
195	DIR *dirp;
196
197	/*****************************************************************************/	267	/*****************************************************************************/
198		268
199	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	269	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200		270
201	/* calculcate time difference in ~1/EIO_TICKS of a second */	271	/* calculate time difference in ~1/EIO_TICKS of a second */
		272	ecb_inline int
202	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	273	tvdiff (struct timeval *tv1, struct timeval *tv2)
203	{	274	{
204	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	275	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	276	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
206	}	277	}
207		278
…		…
214	static unsigned int max_poll_reqs; /* reslock */	285	static unsigned int max_poll_reqs; /* reslock */
215		286
216	static volatile unsigned int nreqs; /* reqlock */	287	static volatile unsigned int nreqs; /* reqlock */
217	static volatile unsigned int nready; /* reqlock */	288	static volatile unsigned int nready; /* reqlock */
218	static volatile unsigned int npending; /* reqlock */	289	static volatile unsigned int npending; /* reqlock */
219	static volatile unsigned int max_idle = 4;	290	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		291	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
220		292
221	static mutex_t wrklock = X_MUTEX_INIT;	293	static xmutex_t wrklock;
222	static mutex_t reslock = X_MUTEX_INIT;	294	static xmutex_t reslock;
223	static mutex_t reqlock = X_MUTEX_INIT;	295	static xmutex_t reqlock;
224	static cond_t reqwait = X_COND_INIT;	296	static xcond_t reqwait;
225		297
226	#if !HAVE_PREADWRITE	298	#if !HAVE_PREADWRITE
227	/*	299	/*
228	* make our pread/pwrite emulation safe against themselves, but not against	300	* make our pread/pwrite emulation safe against themselves, but not against
229	* normal read/write by using a mutex. slows down execution a lot,	301	* normal read/write by using a mutex. slows down execution a lot,
230	* but that's your problem, not mine.	302	* but that's your problem, not mine.
231	*/	303	*/
232	static mutex_t preadwritelock = X_MUTEX_INIT;	304	static xmutex_t preadwritelock;
233	#endif	305	#endif
234		306
235	typedef struct etp_worker	307	typedef struct etp_worker
236	{	308	{
237	/* locked by wrklock */	309	/* locked by wrklock */
238	struct etp_worker *prev, *next;	310	struct etp_worker *prev, *next;
239		311
240	thread_t tid;	312	xthread_t tid;
241		313
242	/* locked by reslock, reqlock or wrklock */	314	/* locked by reslock, reqlock or wrklock */
243	ETP_REQ *req; /* currently processed request */	315	ETP_REQ *req; /* currently processed request */
244		316
		317	#ifdef ETP_WORKER_COMMON
245	ETP_WORKER_COMMON	318	ETP_WORKER_COMMON
		319	#endif
246	} etp_worker;	320	} etp_worker;
247		321
248	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	322	static etp_worker wrk_first; /* NOT etp */
249		323
250	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	324	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
251	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	325	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
252		326
253	/* worker threads management */	327	/* worker threads management */
254		328
		329	static void ecb_cold
255	static void etp_worker_clear (etp_worker *wrk)	330	etp_worker_clear (etp_worker *wrk)
256	{	331	{
257	ETP_WORKER_CLEAR (wrk);
258	}	332	}
259		333
		334	static void ecb_cold
260	static void etp_worker_free (etp_worker *wrk)	335	etp_worker_free (etp_worker *wrk)
261	{	336	{
262	wrk->next->prev = wrk->prev;	337	wrk->next->prev = wrk->prev;
263	wrk->prev->next = wrk->next;	338	wrk->prev->next = wrk->next;
264		339
265	free (wrk);	340	free (wrk);
266	}	341	}
267		342
268	static unsigned int etp_nreqs (void)	343	static unsigned int
		344	etp_nreqs (void)
269	{	345	{
270	int retval;	346	int retval;
271	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	347	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
272	retval = nreqs;	348	retval = nreqs;
273	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	349	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
274	return retval;	350	return retval;
275	}	351	}
276		352
277	static unsigned int etp_nready (void)	353	static unsigned int
		354	etp_nready (void)
278	{	355	{
279	unsigned int retval;	356	unsigned int retval;
280		357
281	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	358	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282	retval = nready;	359	retval = nready;
283	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	360	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284		361
285	return retval;	362	return retval;
286	}	363	}
287		364
288	static unsigned int etp_npending (void)	365	static unsigned int
		366	etp_npending (void)
289	{	367	{
290	unsigned int retval;	368	unsigned int retval;
291		369
292	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	370	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
293	retval = npending;	371	retval = npending;
294	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	372	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
295		373
296	return retval;	374	return retval;
297	}	375	}
298		376
299	static unsigned int etp_nthreads (void)	377	static unsigned int
		378	etp_nthreads (void)
300	{	379	{
301	unsigned int retval;	380	unsigned int retval;
302		381
303	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	382	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
304	retval = started;	383	retval = started;
…		…
318	} etp_reqq;	397	} etp_reqq;
319		398
320	static etp_reqq req_queue;	399	static etp_reqq req_queue;
321	static etp_reqq res_queue;	400	static etp_reqq res_queue;
322		401
		402	static void ecb_noinline ecb_cold
		403	reqq_init (etp_reqq *q)
		404	{
		405	int pri;
		406
		407	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		408	q->qs[pri] = q->qe[pri] = 0;
		409
		410	q->size = 0;
		411	}
		412
		413	static int ecb_noinline
323	static int reqq_push (etp_reqq *q, ETP_REQ *req)	414	reqq_push (etp_reqq *q, ETP_REQ *req)
324	{	415	{
325	int pri = req->pri;	416	int pri = req->pri;
326	req->next = 0;	417	req->next = 0;
327		418
328	if (q->qe[pri])	419	if (q->qe[pri])
…		…
334	q->qe[pri] = q->qs[pri] = req;	425	q->qe[pri] = q->qs[pri] = req;
335		426
336	return q->size++;	427	return q->size++;
337	}	428	}
338		429
		430	static ETP_REQ * ecb_noinline
339	static ETP_REQ *reqq_shift (etp_reqq *q)	431	reqq_shift (etp_reqq *q)
340	{	432	{
341	int pri;	433	int pri;
342		434
343	if (!q->size)	435	if (!q->size)
344	return 0;	436	return 0;
…		…
359	}	451	}
360		452
361	abort ();	453	abort ();
362	}	454	}
363		455
364	static void etp_atfork_prepare (void)	456	static int ecb_cold
		457	etp_init (void (*want_poll)(void), void (*done_poll)(void))
365	{	458	{
366	X_LOCK (wrklock);	459	X_MUTEX_CREATE (wrklock);
367	X_LOCK (reqlock);	460	X_MUTEX_CREATE (reslock);
368	X_LOCK (reslock);	461	X_MUTEX_CREATE (reqlock);
369	#if !HAVE_PREADWRITE	462	X_COND_CREATE (reqwait);
370	X_LOCK (preadwritelock);
371	#endif
372	}
373		463
374	static void etp_atfork_parent (void)	464	reqq_init (&req_queue);
375	{	465	reqq_init (&res_queue);
376	#if !HAVE_PREADWRITE
377	X_UNLOCK (preadwritelock);
378	#endif
379	X_UNLOCK (reslock);
380	X_UNLOCK (reqlock);
381	X_UNLOCK (wrklock);
382	}
383		466
384	static void etp_atfork_child (void)	467	wrk_first.next =
385	{	468	wrk_first.prev = &wrk_first;
386	ETP_REQ *prv;
387
388	while ((prv = reqq_shift (&req_queue)))
389	ETP_DESTROY (prv);
390
391	while ((prv = reqq_shift (&res_queue)))
392	ETP_DESTROY (prv);
393
394	while (wrk_first.next != &wrk_first)
395	{
396	etp_worker *wrk = wrk_first.next;
397
398	if (wrk->req)
399	ETP_DESTROY (wrk->req);
400
401	etp_worker_clear (wrk);
402	etp_worker_free (wrk);
403	}
404		469
405	started = 0;	470	started = 0;
406	idle = 0;	471	idle = 0;
407	nreqs = 0;	472	nreqs = 0;
408	nready = 0;	473	nready = 0;
409	npending = 0;	474	npending = 0;
410		475
411	etp_atfork_parent ();
412	}
413
414	static void
415	etp_once_init (void)
416	{
417	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
418	}
419
420	static int
421	etp_init (void (*want_poll)(void), void (*done_poll)(void))
422	{
423	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
424
425	pthread_once (&doinit, etp_once_init);
426
427	want_poll_cb = want_poll;	476	want_poll_cb = want_poll;
428	done_poll_cb = done_poll;	477	done_poll_cb = done_poll;
429		478
430	return 0;	479	return 0;
431	}	480	}
432		481
433	X_THREAD_PROC (etp_proc);	482	X_THREAD_PROC (etp_proc);
434		483
		484	static void ecb_cold
435	static void etp_start_thread (void)	485	etp_start_thread (void)
436	{	486	{
437	etp_worker *wrk = calloc (1, sizeof (etp_worker));	487	etp_worker *wrk = calloc (1, sizeof (etp_worker));
438		488
439	/*TODO*/	489	/*TODO*/
440	assert (("unable to allocate worker thread data", wrk));	490	assert (("unable to allocate worker thread data", wrk));
…		…
453	free (wrk);	503	free (wrk);
454		504
455	X_UNLOCK (wrklock);	505	X_UNLOCK (wrklock);
456	}	506	}
457		507
		508	static void
458	static void etp_maybe_start_thread (void)	509	etp_maybe_start_thread (void)
459	{	510	{
460	if (expect_true (etp_nthreads () >= wanted))	511	if (ecb_expect_true (etp_nthreads () >= wanted))
461	return;	512	return;
462		513
463	/* todo: maybe use idle here, but might be less exact */	514	/* todo: maybe use idle here, but might be less exact */
464	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	515	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
465	return;	516	return;
466		517
467	etp_start_thread ();	518	etp_start_thread ();
468	}	519	}
469		520
		521	static void ecb_cold
470	static void etp_end_thread (void)	522	etp_end_thread (void)
471	{	523	{
472	eio_req *req = calloc (1, sizeof (eio_req));	524	eio_req *req = calloc (1, sizeof (eio_req));
473		525
474	req->type = -1;	526	req->type = -1;
475	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	527	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
482	X_LOCK (wrklock);	534	X_LOCK (wrklock);
483	--started;	535	--started;
484	X_UNLOCK (wrklock);	536	X_UNLOCK (wrklock);
485	}	537	}
486		538
487	static int etp_poll (void)	539	static int
		540	etp_poll (void)
488	{	541	{
489	unsigned int maxreqs;	542	unsigned int maxreqs;
490	unsigned int maxtime;	543	unsigned int maxtime;
491	struct timeval tv_start, tv_now;	544	struct timeval tv_start, tv_now;
492		545
…		…
522		575
523	X_LOCK (reqlock);	576	X_LOCK (reqlock);
524	--nreqs;	577	--nreqs;
525	X_UNLOCK (reqlock);	578	X_UNLOCK (reqlock);
526		579
527	if (expect_false (req->type == EIO_GROUP && req->size))	580	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
528	{	581	{
529	req->int1 = 1; /* mark request as delayed */	582	req->int1 = 1; /* mark request as delayed */
530	continue;	583	continue;
531	}	584	}
532	else	585	else
533	{	586	{
534	int res = ETP_FINISH (req);	587	int res = ETP_FINISH (req);
535	if (expect_false (res))	588	if (ecb_expect_false (res))
536	return res;	589	return res;
537	}	590	}
538		591
539	if (expect_false (maxreqs && !--maxreqs))	592	if (ecb_expect_false (maxreqs && !--maxreqs))
540	break;	593	break;
541		594
542	if (maxtime)	595	if (maxtime)
543	{	596	{
544	gettimeofday (&tv_now, 0);	597	gettimeofday (&tv_now, 0);
…		…
550		603
551	errno = EAGAIN;	604	errno = EAGAIN;
552	return -1;	605	return -1;
553	}	606	}
554		607
		608	static void
555	static void etp_cancel (ETP_REQ *req)	609	etp_cancel (ETP_REQ *req)
556	{	610	{
557	X_LOCK (wrklock);	611	req->cancelled = 1;
558	req->flags |= EIO_FLAG_CANCELLED;
559	X_UNLOCK (wrklock);
560		612
561	eio_grp_cancel (req);	613	eio_grp_cancel (req);
562	}	614	}
563		615
		616	static void
564	static void etp_submit (ETP_REQ *req)	617	etp_submit (ETP_REQ *req)
565	{	618	{
566	req->pri -= ETP_PRI_MIN;	619	req->pri -= ETP_PRI_MIN;
567		620
568	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	621	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
569	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	622	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
570		623
571	if (expect_false (req->type == EIO_GROUP))	624	if (ecb_expect_false (req->type == EIO_GROUP))
572	{	625	{
573	/* I hope this is worth it :/ */	626	/* I hope this is worth it :/ */
574	X_LOCK (reqlock);	627	X_LOCK (reqlock);
575	++nreqs;	628	++nreqs;
576	X_UNLOCK (reqlock);	629	X_UNLOCK (reqlock);
…		…
595		648
596	etp_maybe_start_thread ();	649	etp_maybe_start_thread ();
597	}	650	}
598	}	651	}
599		652
		653	static void ecb_cold
600	static void etp_set_max_poll_time (double nseconds)	654	etp_set_max_poll_time (double nseconds)
601	{	655	{
602	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	656	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603	max_poll_time = nseconds;	657	max_poll_time = nseconds * EIO_TICKS;
604	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	658	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605	}	659	}
606		660
		661	static void ecb_cold
607	static void etp_set_max_poll_reqs (unsigned int maxreqs)	662	etp_set_max_poll_reqs (unsigned int maxreqs)
608	{	663	{
609	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	664	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
610	max_poll_reqs = maxreqs;	665	max_poll_reqs = maxreqs;
611	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	666	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
612	}	667	}
613		668
		669	static void ecb_cold
614	static void etp_set_max_idle (unsigned int nthreads)	670	etp_set_max_idle (unsigned int nthreads)
615	{	671	{
616	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	672	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617	max_idle = nthreads <= 0 ? 1 : nthreads;	673	max_idle = nthreads;
618	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	674	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619	}	675	}
620		676
		677	static void ecb_cold
		678	etp_set_idle_timeout (unsigned int seconds)
		679	{
		680	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		681	idle_timeout = seconds;
		682	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		683	}
		684
		685	static void ecb_cold
621	static void etp_set_min_parallel (unsigned int nthreads)	686	etp_set_min_parallel (unsigned int nthreads)
622	{	687	{
623	if (wanted < nthreads)	688	if (wanted < nthreads)
624	wanted = nthreads;	689	wanted = nthreads;
625	}	690	}
626		691
		692	static void ecb_cold
627	static void etp_set_max_parallel (unsigned int nthreads)	693	etp_set_max_parallel (unsigned int nthreads)
628	{	694	{
629	if (wanted > nthreads)	695	if (wanted > nthreads)
630	wanted = nthreads;	696	wanted = nthreads;
631		697
632	while (started > wanted)	698	while (started > wanted)
633	etp_end_thread ();	699	etp_end_thread ();
634	}	700	}
635		701
636	/*****************************************************************************/	702	/*****************************************************************************/
637		703
		704	static void
638	static void grp_try_feed (eio_req *grp)	705	grp_try_feed (eio_req *grp)
639	{	706	{
640	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	707	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641	{	708	{
642	grp->flags &= ~EIO_FLAG_GROUPADD;	709	grp->flags &= ~EIO_FLAG_GROUPADD;
643		710
…		…
650	break;	717	break;
651	}	718	}
652	}	719	}
653	}	720	}
654		721
		722	static int
655	static int grp_dec (eio_req *grp)	723	grp_dec (eio_req *grp)
656	{	724	{
657	--grp->size;	725	--grp->size;
658		726
659	/* call feeder, if applicable */	727	/* call feeder, if applicable */
660	grp_try_feed (grp);	728	grp_try_feed (grp);
…		…
664	return eio_finish (grp);	732	return eio_finish (grp);
665	else	733	else
666	return 0;	734	return 0;
667	}	735	}
668		736
		737	static void
669	void eio_destroy (eio_req *req)	738	eio_destroy (eio_req *req)
670	{	739	{
671	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	740	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	741	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673		742
674	EIO_DESTROY (req);	743	EIO_DESTROY (req);
675	}	744	}
676		745
		746	static int
677	static int eio_finish (eio_req *req)	747	eio_finish (eio_req *req)
678	{	748	{
679	int res = EIO_FINISH (req);	749	int res = EIO_FINISH (req);
680		750
681	if (req->grp)	751	if (req->grp)
682	{	752	{
…		…
690	if (grp->grp_first == req)	760	if (grp->grp_first == req)
691	grp->grp_first = req->grp_next;	761	grp->grp_first = req->grp_next;
692		762
693	res2 = grp_dec (grp);	763	res2 = grp_dec (grp);
694		764
695	if (!res && res2)	765	if (!res)
696	res = res2;	766	res = res2;
697	}	767	}
698		768
699	eio_destroy (req);	769	eio_destroy (req);
700		770
701	return res;	771	return res;
702	}	772	}
703		773
		774	void
704	void eio_grp_cancel (eio_req *grp)	775	eio_grp_cancel (eio_req *grp)
705	{	776	{
706	for (grp = grp->grp_first; grp; grp = grp->grp_next)	777	for (grp = grp->grp_first; grp; grp = grp->grp_next)
707	eio_cancel (grp);	778	eio_cancel (grp);
708	}	779	}
709		780
		781	void
710	void eio_cancel (eio_req *req)	782	eio_cancel (eio_req *req)
711	{	783	{
712	etp_cancel (req);	784	etp_cancel (req);
713	}	785	}
714		786
		787	void
715	void eio_submit (eio_req *req)	788	eio_submit (eio_req *req)
716	{	789	{
717	etp_submit (req);	790	etp_submit (req);
718	}	791	}
719		792
720	unsigned int eio_nreqs (void)	793	unsigned int
		794	eio_nreqs (void)
721	{	795	{
722	return etp_nreqs ();	796	return etp_nreqs ();
723	}	797	}
724		798
725	unsigned int eio_nready (void)	799	unsigned int
		800	eio_nready (void)
726	{	801	{
727	return etp_nready ();	802	return etp_nready ();
728	}	803	}
729		804
730	unsigned int eio_npending (void)	805	unsigned int
		806	eio_npending (void)
731	{	807	{
732	return etp_npending ();	808	return etp_npending ();
733	}	809	}
734		810
735	unsigned int eio_nthreads (void)	811	unsigned int ecb_cold
		812	eio_nthreads (void)
736	{	813	{
737	return etp_nthreads ();	814	return etp_nthreads ();
738	}	815	}
739		816
		817	void ecb_cold
740	void eio_set_max_poll_time (double nseconds)	818	eio_set_max_poll_time (double nseconds)
741	{	819	{
742	etp_set_max_poll_time (nseconds);	820	etp_set_max_poll_time (nseconds);
743	}	821	}
744		822
		823	void ecb_cold
745	void eio_set_max_poll_reqs (unsigned int maxreqs)	824	eio_set_max_poll_reqs (unsigned int maxreqs)
746	{	825	{
747	etp_set_max_poll_reqs (maxreqs);	826	etp_set_max_poll_reqs (maxreqs);
748	}	827	}
749		828
		829	void ecb_cold
750	void eio_set_max_idle (unsigned int nthreads)	830	eio_set_max_idle (unsigned int nthreads)
751	{	831	{
752	etp_set_max_idle (nthreads);	832	etp_set_max_idle (nthreads);
753	}	833	}
754		834
		835	void ecb_cold
		836	eio_set_idle_timeout (unsigned int seconds)
		837	{
		838	etp_set_idle_timeout (seconds);
		839	}
		840
		841	void ecb_cold
755	void eio_set_min_parallel (unsigned int nthreads)	842	eio_set_min_parallel (unsigned int nthreads)
756	{	843	{
757	etp_set_min_parallel (nthreads);	844	etp_set_min_parallel (nthreads);
758	}	845	}
759		846
		847	void ecb_cold
760	void eio_set_max_parallel (unsigned int nthreads)	848	eio_set_max_parallel (unsigned int nthreads)
761	{	849	{
762	etp_set_max_parallel (nthreads);	850	etp_set_max_parallel (nthreads);
763	}	851	}
764		852
765	int eio_poll (void)	853	int eio_poll (void)
…		…
774	# undef pread	862	# undef pread
775	# undef pwrite	863	# undef pwrite
776	# define pread eio__pread	864	# define pread eio__pread
777	# define pwrite eio__pwrite	865	# define pwrite eio__pwrite
778		866
779	static ssize_t	867	static eio_ssize_t
780	eio__pread (int fd, void *buf, size_t count, off_t offset)	868	eio__pread (int fd, void *buf, size_t count, off_t offset)
781	{	869	{
782	ssize_t res;	870	eio_ssize_t res;
783	off_t ooffset;	871	off_t ooffset;
784		872
785	X_LOCK (preadwritelock);	873	X_LOCK (preadwritelock);
786	ooffset = lseek (fd, 0, SEEK_CUR);	874	ooffset = lseek (fd, 0, SEEK_CUR);
787	lseek (fd, offset, SEEK_SET);	875	lseek (fd, offset, SEEK_SET);
…		…
790	X_UNLOCK (preadwritelock);	878	X_UNLOCK (preadwritelock);
791		879
792	return res;	880	return res;
793	}	881	}
794		882
795	static ssize_t	883	static eio_ssize_t
796	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	884	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
797	{	885	{
798	ssize_t res;	886	eio_ssize_t res;
799	off_t ooffset;	887	off_t ooffset;
800		888
801	X_LOCK (preadwritelock);	889	X_LOCK (preadwritelock);
802	ooffset = lseek (fd, 0, SEEK_CUR);	890	ooffset = lseek (fd, 0, SEEK_CUR);
803	lseek (fd, offset, SEEK_SET);	891	lseek (fd, offset, SEEK_SET);
…		…
807		895
808	return res;	896	return res;
809	}	897	}
810	#endif	898	#endif
811		899
812	#ifndef HAVE_FUTIMES	900	#ifndef HAVE_UTIMES
813		901
814	# undef utimes	902	# undef utimes
815	# undef futimes
816	# define utimes(path,times) eio__utimes (path, times)	903	# define utimes(path,times) eio__utimes (path, times)
817	# define futimes(fd,times) eio__futimes (fd, times)
818		904
819	static int	905	static int
820	eio__utimes (const char *filename, const struct timeval times[2])	906	eio__utimes (const char *filename, const struct timeval times[2])
821	{	907	{
822	if (times)	908	if (times)
…		…
830	}	916	}
831	else	917	else
832	return utime (filename, 0);	918	return utime (filename, 0);
833	}	919	}
834		920
		921	#endif
		922
		923	#ifndef HAVE_FUTIMES
		924
		925	# undef futimes
		926	# define futimes(fd,times) eio__futimes (fd, times)
		927
		928	static int
835	static int eio__futimes (int fd, const struct timeval tv[2])	929	eio__futimes (int fd, const struct timeval tv[2])
836	{	930	{
837	errno = ENOSYS;	931	errno = ENOSYS;
838	return -1;	932	return -1;
839	}	933	}
840		934
…		…
844	# undef fdatasync	938	# undef fdatasync
845	# define fdatasync(fd) fsync (fd)	939	# define fdatasync(fd) fsync (fd)
846	#endif	940	#endif
847		941
848	/* sync_file_range always needs emulation */	942	/* sync_file_range always needs emulation */
849	int	943	static int
850	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	944	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
851	{	945	{
852	#if HAVE_SYNC_FILE_RANGE	946	#if HAVE_SYNC_FILE_RANGE
853	int res;	947	int res;
854		948
…		…
867	if (!res || errno != ENOSYS)	961	if (!res || errno != ENOSYS)
868	return res;	962	return res;
869	#endif	963	#endif
870		964
871	/* even though we could play tricks with the flags, it's better to always	965	/* even though we could play tricks with the flags, it's better to always
872	* call fdatasync, as thta matches the expectation of it's users best */	966	* call fdatasync, as that matches the expectation of its users best */
873	return fdatasync (fd);	967	return fdatasync (fd);
		968	}
		969
		970	static int
		971	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		972	{
		973	#if HAVE_FALLOCATE
		974	return fallocate (fd, mode, offset, len);
		975	#else
		976	errno = ENOSYS;
		977	return -1;
		978	#endif
874	}	979	}
875		980
876	#if !HAVE_READAHEAD	981	#if !HAVE_READAHEAD
877	# undef readahead	982	# undef readahead
878	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	983	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
879		984
880	static ssize_t	985	static eio_ssize_t
881	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	986	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
882	{	987	{
883	size_t todo = count;	988	size_t todo = count;
884	dBUF;	989	dBUF;
885		990
…		…
890	pread (fd, eio_buf, len, offset);	995	pread (fd, eio_buf, len, offset);
891	offset += len;	996	offset += len;
892	todo -= len;	997	todo -= len;
893	}	998	}
894		999
		1000	FUBd;
		1001
895	errno = 0;	1002	errno = 0;
896	return count;	1003	return count;
897	}	1004	}
898		1005
899	#endif	1006	#endif
900		1007
901	/* sendfile always needs emulation */	1008	/* sendfile always needs emulation */
902	static ssize_t	1009	static eio_ssize_t
903	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1010	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
904	{	1011	{
		1012	eio_ssize_t written = 0;
905	ssize_t res;	1013	eio_ssize_t res;
906		1014
907	if (!count)	1015	if (!count)
908	return 0;	1016	return 0;
909		1017
		1018	for (;;)
		1019	{
		1020	#ifdef __APPLE__
		1021	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1022	#endif
910	#if HAVE_SENDFILE	1023	#if HAVE_SENDFILE
911	# if __linux	1024	# if __linux
		1025	off_t soffset = offset;
912	res = sendfile (ofd, ifd, &offset, count);	1026	res = sendfile (ofd, ifd, &soffset, count);
913		1027
914	# elif __freebsd	1028	# elif __FreeBSD__
915	/*	1029	/*
916	* Of course, the freebsd sendfile is a dire hack with no thoughts	1030	* Of course, the freebsd sendfile is a dire hack with no thoughts
917	* wasted on making it similar to other I/O functions.	1031	* wasted on making it similar to other I/O functions.
918	*/	1032	*/
919	{
920	off_t sbytes;	1033	off_t sbytes;
921	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1034	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922		1035
923	if (res < 0 && sbytes)	1036	#if 0 /* according to the manpage, this is correct, but broken behaviour */
924	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1037	/* freebsd' sendfile will return 0 on success */
		1038	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1039	/* not on e.g. EIO or EPIPE - sounds broken */
		1040	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
925	res = sbytes;	1041	res = sbytes;
926	}	1042	#endif
		1043
		1044	/* according to source inspection, this is correct, and useful behaviour */
		1045	if (sbytes)
		1046	res = sbytes;
		1047
		1048	# elif defined (__APPLE__)
		1049	off_t sbytes = count;
		1050	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1051
		1052	/* according to the manpage, sbytes is always valid */
		1053	if (sbytes)
		1054	res = sbytes;
927		1055
928	# elif __hpux	1056	# elif __hpux
929	res = sendfile (ofd, ifd, offset, count, 0, 0);	1057	res = sendfile (ofd, ifd, offset, count, 0, 0);
930		1058
931	# elif __solaris	1059	# elif __solaris
932	{
933	struct sendfilevec vec;	1060	struct sendfilevec vec;
934	size_t sbytes;	1061	size_t sbytes;
935		1062
936	vec.sfv_fd = ifd;	1063	vec.sfv_fd = ifd;
937	vec.sfv_flag = 0;	1064	vec.sfv_flag = 0;
938	vec.sfv_off = offset;	1065	vec.sfv_off = offset;
939	vec.sfv_len = count;	1066	vec.sfv_len = count;
940		1067
941	res = sendfilev (ofd, &vec, 1, &sbytes);	1068	res = sendfilev (ofd, &vec, 1, &sbytes);
942		1069
943	if (res < 0 && sbytes)	1070	if (res < 0 && sbytes)
944	res = sbytes;	1071	res = sbytes;
945	}
946		1072
947	# endif	1073	# endif
		1074
		1075	#elif defined (_WIN32) && 0
		1076	/* does not work, just for documentation of what would need to be done */
		1077	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1078	/* libeio guarantees that the file offset does not change, and windows */
		1079	/* has no way to get an independent handle to the same file description */
		1080	HANDLE h = TO_SOCKET (ifd);
		1081	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1082	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1083
948	#else	1084	#else
949	res = -1;	1085	res = -1;
950	errno = ENOSYS;	1086	errno = ENOSYS;
951	#endif	1087	#endif
952		1088
		1089	/* we assume sendfile can copy at least 128mb in one go */
		1090	if (res <= 128 * 1024 * 1024)
		1091	{
		1092	if (res > 0)
		1093	written += res;
		1094
		1095	if (written)
		1096	return written;
		1097
		1098	break;
		1099	}
		1100	else
		1101	{
		1102	/* if we requested more, then probably the kernel was lazy */
		1103	written += res;
		1104	offset += res;
		1105	count -= res;
		1106
		1107	if (!count)
		1108	return written;
		1109	}
		1110	}
		1111
953	if (res < 0	1112	if (res < 0
954	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1113	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1114	/* BSDs */
		1115	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1116	|| errno == ENOTSUP
		1117	#endif
		1118	#ifdef EOPNOTSUPP /* windows */
		1119	|| errno == EOPNOTSUPP /* BSDs */
		1120	#endif
955	#if __solaris	1121	#if __solaris
956	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1122	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
957	#endif	1123	#endif
958	)	1124	)
959	)	1125	)
…		…
963		1129
964	res = 0;	1130	res = 0;
965		1131
966	while (count)	1132	while (count)
967	{	1133	{
968	ssize_t cnt;	1134	eio_ssize_t cnt;
969		1135
970	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1136	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
971		1137
972	if (cnt <= 0)	1138	if (cnt <= 0)
973	{	1139	{
…		…
985		1151
986	offset += cnt;	1152	offset += cnt;
987	res += cnt;	1153	res += cnt;
988	count -= cnt;	1154	count -= cnt;
989	}	1155	}
		1156
		1157	FUBd;
990	}	1158	}
991		1159
992	return res;	1160	return res;
		1161	}
		1162
		1163	#ifdef PAGESIZE
		1164	# define eio_pagesize() PAGESIZE
		1165	#else
		1166	static intptr_t
		1167	eio_pagesize (void)
		1168	{
		1169	static intptr_t page;
		1170
		1171	if (!page)
		1172	page = sysconf (_SC_PAGESIZE);
		1173
		1174	return page;
		1175	}
		1176	#endif
		1177
		1178	static void
		1179	eio_page_align (void **addr, size_t *length)
		1180	{
		1181	intptr_t mask = eio_pagesize () - 1;
		1182
		1183	/* round down addr */
		1184	intptr_t adj = mask & (intptr_t)*addr;
		1185
		1186	*addr = (void *)((intptr_t)*addr - adj);
		1187	*length += adj;
		1188
		1189	/* round up length */
		1190	*length = (*length + mask) & ~mask;
		1191	}
		1192
		1193	#if !_POSIX_MEMLOCK
		1194	# define eio__mlockall(a) EIO_ENOSYS ()
		1195	#else
		1196
		1197	static int
		1198	eio__mlockall (int flags)
		1199	{
		1200	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1201	extern int mallopt (int, int);
		1202	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1203	#endif
		1204
		1205	if (EIO_MCL_CURRENT != MCL_CURRENT
		1206	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1207	{
		1208	flags = 0
		1209	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1210	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1211	}
		1212
		1213	return mlockall (flags);
		1214	}
		1215	#endif
		1216
		1217	#if !_POSIX_MEMLOCK_RANGE
		1218	# define eio__mlock(a,b) EIO_ENOSYS ()
		1219	#else
		1220
		1221	static int
		1222	eio__mlock (void *addr, size_t length)
		1223	{
		1224	eio_page_align (&addr, &length);
		1225
		1226	return mlock (addr, length);
		1227	}
		1228
		1229	#endif
		1230
		1231	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1232	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1233	#else
		1234
		1235	static int
		1236	eio__msync (void *mem, size_t len, int flags)
		1237	{
		1238	eio_page_align (&mem, &len);
		1239
		1240	if (EIO_MS_ASYNC != MS_SYNC
		1241	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1242	|| EIO_MS_SYNC != MS_SYNC)
		1243	{
		1244	flags = 0
		1245	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1246	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1247	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1248	}
		1249
		1250	return msync (mem, len, flags);
		1251	}
		1252
		1253	#endif
		1254
		1255	static int
		1256	eio__mtouch (eio_req *req)
		1257	{
		1258	void *mem = req->ptr2;
		1259	size_t len = req->size;
		1260	int flags = req->int1;
		1261
		1262	eio_page_align (&mem, &len);
		1263
		1264	{
		1265	intptr_t addr = (intptr_t)mem;
		1266	intptr_t end = addr + len;
		1267	intptr_t page = eio_pagesize ();
		1268
		1269	if (addr < end)
		1270	if (flags & EIO_MT_MODIFY) /* modify */
		1271	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1272	else
		1273	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1274	}
		1275
		1276	return 0;
		1277	}
		1278
		1279	/*****************************************************************************/
		1280	/* requests implemented outside eio_execute, because they are so large */
		1281
		1282	static void
		1283	eio__realpath (eio_req *req, etp_worker *self)
		1284	{
		1285	char *rel = req->ptr1;
		1286	char *res;
		1287	char *tmp1, *tmp2;
		1288	#if SYMLOOP_MAX > 32
		1289	int symlinks = SYMLOOP_MAX;
		1290	#else
		1291	int symlinks = 32;
		1292	#endif
		1293
		1294	req->result = -1;
		1295
		1296	errno = EINVAL;
		1297	if (!rel)
		1298	return;
		1299
		1300	errno = ENOENT;
		1301	if (!*rel)
		1302	return;
		1303
		1304	if (!req->ptr2)
		1305	{
		1306	X_LOCK (wrklock);
		1307	req->flags |= EIO_FLAG_PTR2_FREE;
		1308	X_UNLOCK (wrklock);
		1309	req->ptr2 = malloc (PATH_MAX * 3);
		1310
		1311	errno = ENOMEM;
		1312	if (!req->ptr2)
		1313	return;
		1314	}
		1315
		1316	res = req->ptr2;
		1317	tmp1 = res + PATH_MAX;
		1318	tmp2 = tmp1 + PATH_MAX;
		1319
		1320	#if 0 /* disabled, the musl way to do things is just too racy */
		1321	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1322	/* on linux we may be able to ask the kernel */
		1323	{
		1324	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1325
		1326	if (fd >= 0)
		1327	{
		1328	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1329	req->result = readlink (tmp1, res, PATH_MAX);
		1330	close (fd);
		1331
		1332	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1333
		1334	if (req->result > 0)
		1335	goto done;
		1336	}
		1337	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1338	return;
		1339	}
		1340	#endif
		1341	#endif
		1342
		1343	if (*rel != '/')
		1344	{
		1345	if (!getcwd (res, PATH_MAX))
		1346	return;
		1347
		1348	if (res [1]) /* only use if not / */
		1349	res += strlen (res);
		1350	}
		1351
		1352	while (*rel)
		1353	{
		1354	eio_ssize_t len, linklen;
		1355	char *beg = rel;
		1356
		1357	while (*rel && *rel != '/')
		1358	++rel;
		1359
		1360	len = rel - beg;
		1361
		1362	if (!len) /* skip slashes */
		1363	{
		1364	++rel;
		1365	continue;
		1366	}
		1367
		1368	if (beg [0] == '.')
		1369	{
		1370	if (len == 1)
		1371	continue; /* . - nop */
		1372
		1373	if (beg [1] == '.' && len == 2)
		1374	{
		1375	/* .. - back up one component, if possible */
		1376
		1377	while (res != req->ptr2)
		1378	if (*--res == '/')
		1379	break;
		1380
		1381	continue;
		1382	}
		1383	}
		1384
		1385	errno = ENAMETOOLONG;
		1386	if (res + 1 + len + 1 >= tmp1)
		1387	return;
		1388
		1389	/* copy one component */
		1390	*res = '/';
		1391	memcpy (res + 1, beg, len);
		1392
		1393	/* zero-terminate, for readlink */
		1394	res [len + 1] = 0;
		1395
		1396	/* now check if it's a symlink */
		1397	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1398
		1399	if (linklen < 0)
		1400	{
		1401	if (errno != EINVAL)
		1402	return;
		1403
		1404	/* it's a normal directory. hopefully */
		1405	res += len + 1;
		1406	}
		1407	else
		1408	{
		1409	/* yay, it was a symlink - build new path in tmp2 */
		1410	int rellen = strlen (rel);
		1411
		1412	errno = ENAMETOOLONG;
		1413	if (linklen + 1 + rellen >= PATH_MAX)
		1414	return;
		1415
		1416	errno = ELOOP;
		1417	if (!--symlinks)
		1418	return;
		1419
		1420	if (*tmp1 == '/')
		1421	res = req->ptr2; /* symlink resolves to an absolute path */
		1422
		1423	/* we need to be careful, as rel might point into tmp2 already */
		1424	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1425	tmp2 [linklen] = '/';
		1426	memcpy (tmp2, tmp1, linklen);
		1427
		1428	rel = tmp2;
		1429	}
		1430	}
		1431
		1432	/* special case for the lone root path */
		1433	if (res == req->ptr2)
		1434	*res++ = '/';
		1435
		1436	req->result = res - (char *)req->ptr2;
		1437
		1438	done:
		1439	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
993	}	1440	}
994		1441
995	static signed char	1442	static signed char
996	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1443	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997	{	1444	{
998	return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */	1445	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
999	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;	1446	: a->inode < b->inode ? -1
		1447	: a->inode > b->inode ? 1
		1448	: 0;
1000	}	1449	}
1001		1450
		1451	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1452
1002	#define EIO_QSORT_CUTOFF 20 /* quite high, but performs well on many filesystems */	1453	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1454	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1003		1455
1004	static void	1456	static void
1005	eio_dent_sort (eio_dirent *dents, int size)	1457	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1006	{	1458	{
		1459	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
		1460	unsigned char *bit = bits;
		1461
		1462	assert (CHAR_BIT == 8);
		1463	assert (sizeof (eio_dirent) * 8 < 256);
		1464	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1465	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1466
		1467	if (size <= EIO_SORT_FAST)
		1468	return;
		1469
		1470	/* first prepare an array of bits to test in our radix sort */
		1471	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
		1472	/* inode_bits must contain all inodes ORed together */
		1473	/* which is used to skip bits that are 0 everywhere, which is very common */
		1474	{
		1475	eio_ino_t endianness;
1007	int i, j;	1476	int i, j;
1008		1477
1009	if (size > EIO_QSORT_CUTOFF * 3) /* skip quicksort for small directories */	1478	/* we store the byte offset of byte n into byte n of "endianness" */
		1479	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1480	((unsigned char *)&endianness)[i] = i;
		1481
		1482	*bit++ = 0;
		1483
		1484	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1485	{
		1486	/* shifting off the byte offsets out of "endianness" */
		1487	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1488	endianness >>= 8;
		1489
		1490	for (j = 0; j < 8; ++j)
		1491	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
		1492	*bit++ = offs + j;
		1493	}
		1494
		1495	for (j = 0; j < 8; ++j)
		1496	if (score_bits & (1 << j))
		1497	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1498	}
		1499
		1500	/* now actually do the sorting (a variant of MSD radix sort) */
		1501	{
		1502	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
		1503	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
		1504	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
		1505	int stk_idx = 0;
		1506
		1507	base_stk [stk_idx] = dents;
		1508	end_stk [stk_idx] = dents + size;
		1509	bit_stk [stk_idx] = bit - 1;
		1510
		1511	do
		1512	{
		1513	base = base_stk [stk_idx];
		1514	end = end_stk [stk_idx];
		1515	bit = bit_stk [stk_idx];
		1516
		1517	for (;;)
		1518	{
		1519	unsigned char O = *bit >> 3;
		1520	unsigned char M = 1 << (*bit & 7);
		1521
		1522	eio_dirent *a = base;
		1523	eio_dirent *b = end;
		1524
		1525	if (b - a < EIO_SORT_CUTOFF)
		1526	break;
		1527
		1528	/* now bit-partition the array on the bit */
		1529	/* this ugly asymmetric loop seems to perform much better than typical */
		1530	/* partition algos found in the literature */
		1531	do
		1532	if (!(((unsigned char *)a)[O] & M))
		1533	++a;
		1534	else if (!(((unsigned char *)--b)[O] & M))
		1535	{
		1536	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1537	++a;
		1538	}
		1539	while (b > a);
		1540
		1541	/* next bit, or stop, if no bits left in this path */
		1542	if (!*--bit)
		1543	break;
		1544
		1545	base_stk [stk_idx] = a;
		1546	end_stk [stk_idx] = end;
		1547	bit_stk [stk_idx] = bit;
		1548	++stk_idx;
		1549
		1550	end = a;
		1551	}
		1552	}
		1553	while (stk_idx--);
		1554	}
		1555	}
		1556
		1557	static void
		1558	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1559	{
		1560	/* first move the smallest element to the front, to act as a sentinel */
		1561	{
		1562	int i;
		1563	eio_dirent *min = dents;
		1564
		1565	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1566	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1567	if (EIO_DENT_CMP (dents [i], <, *min))
		1568	min = &dents [i];
		1569
		1570	/* swap elements 0 and j (minimum) */
1010	{	1571	{
1011	/* first, use quicksort */	1572	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1012	/* should be good for 2**31 entries */
1013	struct rng { int l, r; } rng [32];
1014
1015	i = 0;
1016	rng[0].l = 0;
1017	rng[0].r = size;
1018
1019	while (expect_true (i >= 0))
1020	{
1021	int L = rng [i].l;
1022	int R = rng [i].r - 1;
1023
1024	if (expect_false (L + EIO_QSORT_CUTOFF < R))
1025	{
1026	eio_dirent piv = dents [L];
1027
1028	while (L < R)
1029	{
1030	while (eio_dent_cmp (&dents [R], &piv) >= 0 && L < R)
1031	--R;
1032
1033	if (L < R)
1034	dents [L++] = dents [R];
1035
1036	while (eio_dent_cmp (&dents [L], &piv) <= 0 && L < R)
1037	++L;
1038
1039	if (L < R)
1040	dents [R--] = dents [L];
1041	}
1042
1043	dents [L] = piv;
1044
1045	++i;
1046	rng [i].l = L + 1;
1047	rng [i].r = rng [i - 1].r;
1048	rng [i - 1].r = L;
1049
1050	if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l)
1051	{
1052	struct rng t;
1053
1054	t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t;
1055	}
1056	}
1057	else
1058	--i;
1059	}
1060	}	1573	}
		1574	}
1061		1575
1062	/* use a simple insertion sort at the end */	1576	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1577	{
		1578	eio_dirent *i, *j;
		1579
1063	for (i = 1; i < size; ++i)	1580	for (i = dents + 1; i < dents + size; ++i)
1064	{	1581	{
1065	eio_dirent value = dents [i];	1582	eio_dirent value = *i;
1066		1583
1067	for (j = i - 1; j >= 0 && eio_dent_cmp (&dents [j], &value) > 0; --j)	1584	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1068	dents [j + 1] = dents [j];	1585	j [1] = j [0];
1069		1586
1070	dents [j + 1] = value;	1587	j [1] = value;
1071	}	1588	}
		1589	}
		1590	}
		1591
		1592	static void
		1593	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1594	{
		1595	if (size <= 1)
		1596	return; /* our insertion sort relies on size > 0 */
		1597
		1598	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1599	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1600	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1601
		1602	/* use an insertion sort at the end, or for small arrays, */
		1603	/* as insertion sort is more efficient for small partitions */
		1604	eio_dent_insertion_sort (dents, size);
1072	}	1605	}
1073		1606
1074	/* read a full directory */	1607	/* read a full directory */
1075	static void	1608	static void
1076	eio__scandir (eio_req *req, etp_worker *self)	1609	eio__scandir (eio_req *req, etp_worker *self)
1077	{	1610	{
1078	DIR *dirp;
1079	EIO_STRUCT_DIRENT *entp;
1080	char *name, *names;	1611	char *name, *names;
1081	int namesalloc = 4096;	1612	int namesalloc = 4096 - sizeof (void *) * 4;
1082	int namesoffs = 0;	1613	int namesoffs = 0;
1083	int flags = req->int1;	1614	int flags = req->int1;
1084	eio_dirent *dents = 0;	1615	eio_dirent *dents = 0;
1085	int dentalloc = 128;	1616	int dentalloc = 128;
1086	int dentoffs = 0;	1617	int dentoffs = 0;
		1618	eio_ino_t inode_bits = 0;
		1619	#ifdef _WIN32
		1620	HANDLE dirp;
		1621	WIN32_FIND_DATA entp;
		1622	#else
		1623	DIR *dirp;
		1624	EIO_STRUCT_DIRENT *entp;
		1625	#endif
1087		1626
1088	req->result = -1;	1627	req->result = -1;
1089		1628
1090	if (!(flags & EIO_READDIR_DENTS))	1629	if (!(flags & EIO_READDIR_DENTS))
1091	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1630	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1092		1631
1093	X_LOCK (wrklock);	1632	#ifdef _WIN32
1094	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1633	{
		1634	int len = strlen ((const char *)req->ptr1);
		1635	char *path = malloc (MAX_PATH);
		1636	const char *fmt;
		1637
		1638	if (!len)
		1639	fmt = "./*";
		1640	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1641	fmt = "%s*";
		1642	else
		1643	fmt = "%s/*";
		1644
		1645	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1646	dirp = FindFirstFile (path, &entp);
		1647	free (path);
		1648
		1649	if (dirp == INVALID_HANDLE_VALUE)
		1650	{
		1651	dirp = 0;
		1652
		1653	switch (GetLastError ())
		1654	{
		1655	case ERROR_FILE_NOT_FOUND:
		1656	req->result = 0;
		1657	break;
		1658
		1659	case ERROR_INVALID_NAME:
		1660	case ERROR_PATH_NOT_FOUND:
		1661	case ERROR_NO_MORE_FILES:
		1662	errno = ENOENT;
		1663	break;
		1664
		1665	case ERROR_NOT_ENOUGH_MEMORY:
		1666	errno = ENOMEM;
		1667	break;
		1668
		1669	default:
		1670	errno = EINVAL;
		1671	break;
		1672	}
		1673	}
		1674	}
		1675	#else
1095	self->dirp = dirp = opendir (req->ptr1);	1676	dirp = opendir (req->ptr1);
		1677	#endif
		1678
		1679	if (req->flags & EIO_FLAG_PTR1_FREE)
		1680	free (req->ptr1);
		1681
1096	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1682	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1097	req->ptr1 = names = malloc (namesalloc);
1098	req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1683	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1099	X_UNLOCK (wrklock);	1684	req->ptr2 = names = malloc (namesalloc);
1100		1685
1101	if (dirp && names && (!flags || dents))	1686	if (dirp && names && (!flags || dents))
1102	for (;;)	1687	for (;;)
1103	{	1688	{
		1689	int done;
		1690
		1691	#ifdef _WIN32
		1692	done = !dirp;
		1693	#else
1104	errno = 0;	1694	errno = 0;
1105	entp = readdir (dirp);	1695	entp = readdir (dirp);
		1696	done = !entp;
		1697	#endif
1106		1698
1107	if (!entp)	1699	if (done)
1108	{	1700	{
		1701	#ifndef _WIN32
		1702	int old_errno = errno;
		1703	closedir (dirp);
		1704	errno = old_errno;
		1705
1109	if (errno)	1706	if (errno)
1110	break;	1707	break;
		1708	#endif
1111		1709
1112	/* sort etc. */	1710	/* sort etc. */
1113	req->int1 = flags;	1711	req->int1 = flags;
1114	req->result = dentoffs;	1712	req->result = dentoffs;
1115		1713
1116	if (dents)
1117	{
1118	eio_dirent *ent = dents + dentoffs;
1119
1120	while (ent > dents)
1121	(--ent)->name = names + (size_t)ent->name;
1122	}
1123
1124	if (flags & EIO_READDIR_STAT_ORDER	1714	if (flags & EIO_READDIR_STAT_ORDER)
1125	|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))	1715	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1126	eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */
1127	else if (flags & EIO_READDIR_DIRS_FIRST)	1716	else if (flags & EIO_READDIR_DIRS_FIRST)
		1717	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1718	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1719	else
1128	{	1720	{
1129	/* in this case, all is known, and we just put dirs first and sort them */	1721	/* in this case, all is known, and we just put dirs first and sort them */
1130	eio_dirent *ent = dents + dentoffs;	1722	eio_dirent *oth = dents + dentoffs;
1131	eio_dirent *dir = dents;	1723	eio_dirent *dir = dents;
1132		1724
1133	/* now move dirs to the front, and non-dirs to the back */	1725	/* now partition dirs to the front, and non-dirs to the back */
1134	/* by walking from both sides and swapping if necessary */	1726	/* by walking from both sides and swapping if necessary */
1135	while (ent > dir)	1727	while (oth > dir)
1136	{	1728	{
1137	if (dir->type == DT_DIR)	1729	if (dir->type == EIO_DT_DIR)
1138	++dir;	1730	++dir;
1139	else
1140	{
1141	--ent;
1142
1143	if (ent->type == DT_DIR)	1731	else if ((--oth)->type == EIO_DT_DIR)
1144	{	1732	{
1145	eio_dirent tmp = *dir;	1733	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1146	*dir = *ent;
1147	*ent = tmp;
1148		1734
1149	++dir;	1735	++dir;
1150	}	1736	}
1151	}	1737	}
		1738
		1739	/* now sort the dirs only (dirs all have the same score) */
		1740	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1152	}	1741	}
1153
1154	/* now sort the dirs only */
1155	eio_dent_sort (dents, dir - dents);
1156	}
1157
1158	/* only provide the names array unless DENTS is specified */
1159	if (!(flags & EIO_READDIR_DENTS))
1160	{
1161	X_LOCK (wrklock);
1162	assert (!dents);
1163	req->ptr1 = 0;
1164	req->ptr2 = names;
1165	X_UNLOCK (wrklock);
1166	}
1167		1742
1168	break;	1743	break;
1169	}	1744	}
1170		1745
1171	/* now add the entry to our list(s) */	1746	/* now add the entry to our list(s) */
1172	name = entp->d_name;	1747	name = D_NAME (entp);
1173		1748
1174	/* skip . and .. entries */	1749	/* skip . and .. entries */
1175	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1750	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1176	{	1751	{
1177	int len = D_NAMLEN (entp) + 1;	1752	int len = D_NAMLEN (entp) + 1;
1178		1753
1179	while (expect_false (namesoffs + len > namesalloc))	1754	while (ecb_expect_false (namesoffs + len > namesalloc))
1180	{	1755	{
1181	namesalloc *= 2;	1756	namesalloc *= 2;
1182	X_LOCK (wrklock);
1183	req->ptr1 = names = realloc (names, namesalloc);	1757	req->ptr2 = names = realloc (names, namesalloc);
1184	X_UNLOCK (wrklock);
1185		1758
1186	if (!names)	1759	if (!names)
1187	break;	1760	break;
1188	}	1761	}
1189		1762
…		…
1191		1764
1192	if (dents)	1765	if (dents)
1193	{	1766	{
1194	struct eio_dirent *ent;	1767	struct eio_dirent *ent;
1195		1768
1196	if (expect_false (dentoffs == dentalloc))	1769	if (ecb_expect_false (dentoffs == dentalloc))
1197	{	1770	{
1198	dentalloc *= 2;	1771	dentalloc *= 2;
1199	X_LOCK (wrklock);
1200	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1772	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1201	X_UNLOCK (wrklock);
1202		1773
1203	if (!dents)	1774	if (!dents)
1204	break;	1775	break;
1205	}	1776	}
1206		1777
1207	ent = dents + dentoffs;	1778	ent = dents + dentoffs;
1208		1779
1209	ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */	1780	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1210	ent->namelen = len - 1;	1781	ent->namelen = len - 1;
1211	ent->inode = D_INO (entp);	1782	ent->inode = D_INO (entp);
		1783
		1784	inode_bits |= ent->inode;
1212		1785
1213	switch (D_TYPE (entp))	1786	switch (D_TYPE (entp))
1214	{	1787	{
1215	default:	1788	default:
1216	ent->type = EIO_DT_UNKNOWN;	1789	ent->type = EIO_DT_UNKNOWN;
…		…
1259	#ifdef DT_WHT	1832	#ifdef DT_WHT
1260	case DT_WHT: ent->type = EIO_DT_WHT; break;	1833	case DT_WHT: ent->type = EIO_DT_WHT; break;
1261	#endif	1834	#endif
1262	}	1835	}
1263		1836
1264	ent->score = 0;	1837	ent->score = 7;
1265		1838
1266	if (flags & EIO_READDIR_DIRS_FIRST)	1839	if (flags & EIO_READDIR_DIRS_FIRST)
1267	{	1840	{
1268	if (ent->type == EIO_DT_UNKNOWN)	1841	if (ent->type == EIO_DT_UNKNOWN)
1269	{	1842	{
1270	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */	1843	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1271	ent->score = 98;	1844	ent->score = 1;
1272	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */	1845	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1273	ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */	1846	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1274	}	1847	}
1275	else if (ent->type == EIO_DT_DIR)	1848	else if (ent->type == EIO_DT_DIR)
1276	ent->score = 100;	1849	ent->score = 0;
1277	}	1850	}
1278	}	1851	}
1279		1852
1280	namesoffs += len;	1853	namesoffs += len;
1281	++dentoffs;	1854	++dentoffs;
…		…
1284	if (EIO_CANCELLED (req))	1857	if (EIO_CANCELLED (req))
1285	{	1858	{
1286	errno = ECANCELED;	1859	errno = ECANCELED;
1287	break;	1860	break;
1288	}	1861	}
		1862
		1863	#ifdef _WIN32
		1864	if (!FindNextFile (dirp, &entp))
		1865	{
		1866	FindClose (dirp);
		1867	dirp = 0;
		1868	}
		1869	#endif
1289	}	1870	}
1290	}
1291
1292	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1293	# undef msync
1294	# define msync(a,b,c) ((errno = ENOSYS), -1)
1295	#endif
1296
1297	int
1298	eio__mtouch (void *mem, size_t len, int flags)
1299	{
1300	intptr_t addr = (intptr_t)mem;
1301	intptr_t end = addr + len;
1302	#ifdef PAGESIZE
1303	const intptr_t page = PAGESIZE;
1304	#else
1305	static intptr_t page;
1306
1307	if (!page)
1308	page = sysconf (_SC_PAGESIZE);
1309	#endif
1310
1311	addr &= ~(page - 1); /* assume page size is always a power of two */
1312
1313	if (addr < end)
1314	if (flags) /* modify */
1315	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1316	else
1317	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1318
1319	return 0;
1320	}	1871	}
1321		1872
1322	/*****************************************************************************/	1873	/*****************************************************************************/
1323		1874
1324	#define ALLOC(len) \	1875	#define ALLOC(len) \
…		…
1340	{	1891	{
1341	ETP_REQ *req;	1892	ETP_REQ *req;
1342	struct timespec ts;	1893	struct timespec ts;
1343	etp_worker *self = (etp_worker *)thr_arg;	1894	etp_worker *self = (etp_worker *)thr_arg;
1344		1895
1345	/* try to distribute timeouts somewhat randomly */	1896	/* try to distribute timeouts somewhat evenly */
1346	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1897	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1347		1898
1348	for (;;)	1899	for (;;)
1349	{	1900	{
		1901	ts.tv_sec = 0;
		1902
1350	X_LOCK (reqlock);	1903	X_LOCK (reqlock);
1351		1904
1352	for (;;)	1905	for (;;)
1353	{	1906	{
1354	self->req = req = reqq_shift (&req_queue);	1907	self->req = req = reqq_shift (&req_queue);
1355		1908
1356	if (req)	1909	if (req)
1357	break;	1910	break;
1358		1911
		1912	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1913	{
		1914	X_UNLOCK (reqlock);
		1915	X_LOCK (wrklock);
		1916	--started;
		1917	X_UNLOCK (wrklock);
		1918	goto quit;
		1919	}
		1920
1359	++idle;	1921	++idle;
1360		1922
1361	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1923	if (idle <= max_idle)
1362	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1924	/* we are allowed to idle, so do so without any timeout */
		1925	X_COND_WAIT (reqwait, reqlock);
		1926	else
1363	{	1927	{
1364	if (idle > max_idle)	1928	/* initialise timeout once */
1365	{	1929	if (!ts.tv_sec)
1366	--idle;	1930	ts.tv_sec = time (0) + idle_timeout;
1367	X_UNLOCK (reqlock);
1368	X_LOCK (wrklock);
1369	--started;
1370	X_UNLOCK (wrklock);
1371	goto quit;
1372	}
1373		1931
1374	/* we are allowed to idle, so do so without any timeout */
1375	X_COND_WAIT (reqwait, reqlock);	1932	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1933	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1376	}	1934	}
1377		1935
1378	--idle;	1936	--idle;
1379	}	1937	}
1380		1938
…		…
1383	X_UNLOCK (reqlock);	1941	X_UNLOCK (reqlock);
1384		1942
1385	if (req->type < 0)	1943	if (req->type < 0)
1386	goto quit;	1944	goto quit;
1387		1945
1388	if (!EIO_CANCELLED (req))
1389	ETP_EXECUTE (self, req);	1946	ETP_EXECUTE (self, req);
1390		1947
1391	X_LOCK (reslock);	1948	X_LOCK (reslock);
1392		1949
1393	++npending;	1950	++npending;
1394		1951
…		…
1409	return 0;	1966	return 0;
1410	}	1967	}
1411		1968
1412	/*****************************************************************************/	1969	/*****************************************************************************/
1413		1970
		1971	int ecb_cold
1414	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1972	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1415	{	1973	{
		1974	#if !HAVE_PREADWRITE
		1975	X_MUTEX_CREATE (preadwritelock);
		1976	#endif
		1977
1416	return etp_init (want_poll, done_poll);	1978	return etp_init (want_poll, done_poll);
1417	}	1979	}
1418		1980
		1981	ecb_inline void
1419	static void eio_api_destroy (eio_req *req)	1982	eio_api_destroy (eio_req *req)
1420	{	1983	{
1421	free (req);	1984	free (req);
1422	}	1985	}
1423		1986
1424	#define REQ(rtype) \	1987	#define REQ(rtype) \
…		…
1443	{ \	2006	{ \
1444	eio_api_destroy (req); \	2007	eio_api_destroy (req); \
1445	return 0; \	2008	return 0; \
1446	}	2009	}
1447		2010
		2011	static void
1448	static void eio_execute (etp_worker *self, eio_req *req)	2012	eio_execute (etp_worker *self, eio_req *req)
1449	{	2013	{
1450	errno = 0;	2014	if (ecb_expect_false (EIO_CANCELLED (req)))
		2015	{
		2016	req->result = -1;
		2017	req->errorno = ECANCELED;
		2018	return;
		2019	}
1451		2020
1452	switch (req->type)	2021	switch (req->type)
1453	{	2022	{
1454	case EIO_READ: ALLOC (req->size);	2023	case EIO_READ: ALLOC (req->size);
1455	req->result = req->offs >= 0	2024	req->result = req->offs >= 0
…		…
1458	case EIO_WRITE: req->result = req->offs >= 0	2027	case EIO_WRITE: req->result = req->offs >= 0
1459	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2028	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1460	: write (req->int1, req->ptr2, req->size); break;	2029	: write (req->int1, req->ptr2, req->size); break;
1461		2030
1462	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2031	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1463	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2032	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1464		2033
1465	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2034	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1466	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2035	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1467	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2036	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1468	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2037	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1469	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2038	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1470	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2039	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2040
		2041	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2042	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2043	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2044	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1471		2045
1472	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	2046	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1473	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	2047	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1474	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	2048	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1475	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	2049	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
…		…
1483	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	2057	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1484	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	2058	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1485	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2059	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1486	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2060	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1487	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2061	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1488	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;	2062	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1489		2063
		2064	case EIO_REALPATH: eio__realpath (req, self); break;
		2065
1490	case EIO_READLINK: ALLOC (NAME_MAX);	2066	case EIO_READLINK: ALLOC (PATH_MAX);
1491	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	2067	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1492		2068
1493	case EIO_SYNC: req->result = 0; sync (); break;	2069	case EIO_SYNC: req->result = 0; sync (); break;
1494	case EIO_FSYNC: req->result = fsync (req->int1); break;	2070	case EIO_FSYNC: req->result = fsync (req->int1); break;
1495	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2071	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1496	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;	2072	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2073	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1497	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	2074	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2075	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1498	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	2076	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2077	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1499		2078
1500	case EIO_READDIR: eio__scandir (req, self); break;	2079	case EIO_READDIR: eio__scandir (req, self); break;
1501		2080
1502	case EIO_BUSY:	2081	case EIO_BUSY:
1503	#ifdef _WIN32	2082	#ifdef _WIN32
1504	Sleep (req->nv1 * 1000.);	2083	Sleep (req->nv1 * 1e3);
1505	#else	2084	#else
1506	{	2085	{
1507	struct timeval tv;	2086	struct timeval tv;
1508		2087
1509	tv.tv_sec = req->nv1;	2088	tv.tv_sec = req->nv1;
1510	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	2089	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1511		2090
1512	req->result = select (0, 0, 0, 0, &tv);	2091	req->result = select (0, 0, 0, 0, &tv);
1513	}	2092	}
1514	#endif	2093	#endif
1515	break;	2094	break;
…		…
1530	times = tv;	2109	times = tv;
1531	}	2110	}
1532	else	2111	else
1533	times = 0;	2112	times = 0;
1534		2113
1535
1536	req->result = req->type == EIO_FUTIME	2114	req->result = req->type == EIO_FUTIME
1537	? futimes (req->int1, times)	2115	? futimes (req->int1, times)
1538	: utimes (req->ptr1, times);	2116	: utimes (req->ptr1, times);
1539	}	2117	}
1540	break;	2118	break;
…		…
1545	case EIO_NOP:	2123	case EIO_NOP:
1546	req->result = 0;	2124	req->result = 0;
1547	break;	2125	break;
1548		2126
1549	case EIO_CUSTOM:	2127	case EIO_CUSTOM:
1550	((void (*)(eio_req *))req->feed) (req);	2128	req->feed (req);
1551	break;	2129	break;
1552		2130
1553	default:	2131	default:
		2132	errno = ENOSYS;
1554	req->result = -1;	2133	req->result = -1;
1555	break;	2134	break;
1556	}	2135	}
1557		2136
1558	req->errorno = errno;	2137	req->errorno = errno;
…		…
1588	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2167	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1589	{	2168	{
1590	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2169	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1591	}	2170	}
1592		2171
		2172	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2173	{
		2174	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2175	}
		2176
		2177	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2178	{
		2179	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2180	}
		2181
1593	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2182	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1594	{	2183	{
1595	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2184	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1596	}	2185	}
1597		2186
		2187	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2188	{
		2189	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2190	}
		2191
1598	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2192	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1599	{	2193	{
1600	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2194	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1601	}	2195	}
1602		2196
…		…
1623	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2217	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1624	{	2218	{
1625	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2219	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1626	}	2220	}
1627		2221
		2222	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2223	{
		2224	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2225	}
		2226
1628	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2227	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1629	{	2228	{
1630	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2229	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1631	}	2230	}
1632		2231
…		…
1638	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2237	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1639	{	2238	{
1640	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2239	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1641	}	2240	}
1642		2241
1643	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2242	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1644	{	2243	{
1645	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2244	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1646	}	2245	}
1647		2246
1648	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2247	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1668	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2267	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1669	{	2268	{
1670	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2269	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1671	}	2270	}
1672		2271
1673	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2272	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1674	{	2273	{
1675	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2274	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1676	}	2275	}
1677		2276
1678	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2277	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1694	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2293	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1695	{	2294	{
1696	return eio__1path (EIO_READLINK, path, pri, cb, data);	2295	return eio__1path (EIO_READLINK, path, pri, cb, data);
1697	}	2296	}
1698		2297
		2298	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2299	{
		2300	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2301	}
		2302
1699	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2303	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1700	{	2304	{
1701	return eio__1path (EIO_STAT, path, pri, cb, data);	2305	return eio__1path (EIO_STAT, path, pri, cb, data);
1702	}	2306	}
1703		2307
1704	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2308	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1705	{	2309	{
1706	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2310	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1707	}	2311	}
1708		2312
		2313	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2314	{
		2315	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2316	}
		2317
1709	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2318	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1710	{	2319	{
1711	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2320	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1712	}	2321	}
1713		2322
…		…
1721	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2330	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1722	}	2331	}
1723		2332
1724	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2333	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1725	{	2334	{
1726	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2335	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1727	}	2336	}
1728		2337
1729	static eio_req *	2338	static eio_req *
1730	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2339	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1731	{	2340	{
…		…
1755	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2364	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1756	{	2365	{
1757	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2366	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1758	}	2367	}
1759		2368
1760	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2369	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1761	{	2370	{
1762	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2371	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1763	}	2372	}
1764		2373
1765	#endif	2374	#endif
1766		2375
1767	eio_req *eio_grp (eio_cb cb, void *data)	2376	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1776	#undef SEND	2385	#undef SEND
1777		2386
1778	/*****************************************************************************/	2387	/*****************************************************************************/
1779	/* grp functions */	2388	/* grp functions */
1780		2389
		2390	void
1781	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2391	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1782	{	2392	{
1783	grp->int2 = limit;	2393	grp->int2 = limit;
1784	grp->feed = feed;	2394	grp->feed = feed;
1785		2395
1786	grp_try_feed (grp);	2396	grp_try_feed (grp);
1787	}	2397	}
1788		2398
		2399	void
1789	void eio_grp_limit (eio_req *grp, int limit)	2400	eio_grp_limit (eio_req *grp, int limit)
1790	{	2401	{
1791	grp->int2 = limit;	2402	grp->int2 = limit;
1792		2403
1793	grp_try_feed (grp);	2404	grp_try_feed (grp);
1794	}	2405	}
1795		2406
		2407	void
1796	void eio_grp_add (eio_req *grp, eio_req *req)	2408	eio_grp_add (eio_req *grp, eio_req *req)
1797	{	2409	{
1798	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2410	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1799		2411
1800	grp->flags |= EIO_FLAG_GROUPADD;	2412	grp->flags |= EIO_FLAG_GROUPADD;
1801		2413
…		…
1812	}	2424	}
1813		2425
1814	/*****************************************************************************/	2426	/*****************************************************************************/
1815	/* misc garbage */	2427	/* misc garbage */
1816		2428
		2429	eio_ssize_t
1817	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2430	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1818	{	2431	{
1819	etp_worker wrk;
1820	ssize_t ret;
1821
1822	wrk.dbuf = 0;
1823
1824	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2432	return eio__sendfile (ofd, ifd, offset, count);
1825
1826	if (wrk.dbuf)
1827	free (wrk.dbuf);
1828
1829	return ret;
1830	}	2433	}
1831		2434

Diff Legend

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