ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/cvsroot/libeio/eio.c

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.66 by root, Sun Jun 5 19:58:37 2011 UTC vs.
Revision 1.97 by root, Sun Jul 24 03:32:54 2011 UTC

…		…
40	#ifndef _WIN32	40	#ifndef _WIN32
41	# include "config.h"	41	# include "config.h"
42	#endif	42	#endif
43		43
44	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
45		46
46	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
47	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
48	#endif	49	#endif
49	#include "xthread.h"	50	#include "xthread.h"
…		…
53	#include <stdlib.h>	54	#include <stdlib.h>
54	#include <string.h>	55	#include <string.h>
55	#include <errno.h>	56	#include <errno.h>
56	#include <sys/types.h>	57	#include <sys/types.h>
57	#include <sys/stat.h>	58	#include <sys/stat.h>
58	#include <sys/statvfs.h>
59	#include <limits.h>	59	#include <limits.h>
60	#include <fcntl.h>	60	#include <fcntl.h>
61	#include <assert.h>	61	#include <assert.h>
62		62
63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */	63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64	/* intptr_t only comes form stdint.h, says idiot openbsd coder */	64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65	#if HAVE_STDINT_H	65	#if HAVE_STDINT_H
66	# include <stdint.h>	66	# include <stdint.h>
67	#endif	67	#endif
68		68
		69	#ifndef ECANCELED
		70	# define ECANCELED EDOM
		71	#endif
		72	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
		77	# define ENOTSOCK WSAENOTSOCK
		78	#endif
		79
		80	static void eio_destroy (eio_req *req);
		81
69	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
70	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	83	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
71	#endif	84	#endif
72		85
73	#ifndef EIO_DESTROY	86	#ifndef EIO_DESTROY
…		…
76		89
77	#ifndef EIO_FEED	90	#ifndef EIO_FEED
78	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	91	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
79	#endif	92	#endif
80		93
		94	#ifndef EIO_FD_TO_WIN32_HANDLE
		95	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		96	#endif
		97	#ifndef EIO_WIN32_HANDLE_TO_FD
		98	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		99	#endif
		100
		101	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		102
		103	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		104
81	#ifdef _WIN32	105	#ifdef _WIN32
82		106
83	/*doh*/	107	#undef PAGESIZE
		108	#define PAGESIZE 4096 /* GetSystemInfo? */
		109
		110	#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
84	#else	163	#else
85		164
86	# include <sys/time.h>	165	#include <sys/time.h>
87	# include <sys/select.h>	166	#include <sys/select.h>
		167	#include <sys/statvfs.h>
88	# 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
89	# include <utime.h>	203	# include <utime.h>
90	# include <signal.h>
91	# include <dirent.h>
92
93	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94	# include <sys/mman.h>
95	#endif
96
97	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99	# define _DIRENT_HAVE_D_TYPE /* sigh */
100	# define D_INO(de) (de)->d_fileno
101	# define D_NAMLEN(de) (de)->d_namlen
102	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103	# define D_INO(de) (de)->d_ino
104	# endif
105
106	#ifdef _D_EXACT_NAMLEN
107	# undef D_NAMLEN
108	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109	#endif
110
111	# ifdef _DIRENT_HAVE_D_TYPE
112	# define D_TYPE(de) (de)->d_type
113	# endif
114
115	# ifndef EIO_STRUCT_DIRENT
116	# define EIO_STRUCT_DIRENT struct dirent
117	# endif
118
119	#endif	204	#endif
120		205
121	#if HAVE_SENDFILE	206	#if HAVE_SENDFILE
122	# if __linux	207	# if __linux
123	# include <sys/sendfile.h>	208	# include <sys/sendfile.h>
…		…
131	# else	216	# else
132	# error sendfile support requested but not available	217	# error sendfile support requested but not available
133	# endif	218	# endif
134	#endif	219	#endif
135		220
		221	#if HAVE_SYS_SYSCALL_H
		222	# include <sys/syscall.h>
		223	#endif
		224
136	#ifndef D_TYPE	225	#ifndef D_TYPE
137	# define D_TYPE(de) 0	226	# define D_TYPE(de) 0
138	#endif	227	#endif
139	#ifndef D_INO	228	#ifndef D_INO
140	# define D_INO(de) 0	229	# define D_INO(de) 0
141	#endif	230	#endif
142	#ifndef D_NAMLEN	231	#ifndef D_NAMLEN
143	# define D_NAMLEN(de) strlen ((de)->d_name)	232	# define D_NAMLEN(entp) strlen (D_NAME (entp))
144	#endif	233	#endif
145		234
146	/* used for struct dirent, AIX doesn't provide it */	235	/* used for struct dirent, AIX doesn't provide it */
147	#ifndef NAME_MAX	236	#ifndef NAME_MAX
148	# define NAME_MAX 4096	237	# define NAME_MAX 4096
…		…
155		244
156	/* buffer size for various temporary buffers */	245	/* buffer size for various temporary buffers */
157	#define EIO_BUFSIZE 65536	246	#define EIO_BUFSIZE 65536
158		247
159	#define dBUF \	248	#define dBUF \
160	char *eio_buf; \
161	ETP_WORKER_LOCK (self); \
162	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	249	char *eio_buf = malloc (EIO_BUFSIZE); \
163	ETP_WORKER_UNLOCK (self); \
164	errno = ENOMEM; \	250	errno = ENOMEM; \
165	if (!eio_buf) \	251	if (!eio_buf) \
166	return -1;	252	return -1
		253
		254	#define FUBd \
		255	free (eio_buf)
167		256
168	#define EIO_TICKS ((1000000 + 1023) >> 10)	257	#define EIO_TICKS ((1000000 + 1023) >> 10)
169
170	/*****************************************************************************/
171
172	#if __GNUC__ >= 3
173	# define expect(expr,value) __builtin_expect ((expr),(value))
174	#else
175	# define expect(expr,value) (expr)
176	#endif
177
178	#define expect_false(expr) expect ((expr) != 0, 0)
179	#define expect_true(expr) expect ((expr) != 0, 1)
180
181	/*****************************************************************************/
182		258
183	#define ETP_PRI_MIN EIO_PRI_MIN	259	#define ETP_PRI_MIN EIO_PRI_MIN
184	#define ETP_PRI_MAX EIO_PRI_MAX	260	#define ETP_PRI_MAX EIO_PRI_MAX
185		261
186	struct etp_worker;	262	struct etp_worker;
…		…
190	static int eio_finish (eio_req *req);	266	static int eio_finish (eio_req *req);
191	#define ETP_FINISH(req) eio_finish (req)	267	#define ETP_FINISH(req) eio_finish (req)
192	static void eio_execute (struct etp_worker *self, eio_req *req);	268	static void eio_execute (struct etp_worker *self, eio_req *req);
193	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	269	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194		270
195	#define ETP_WORKER_CLEAR(req) \
196	if (wrk->dbuf) \
197	{ \
198	free (wrk->dbuf); \
199	wrk->dbuf = 0; \
200	} \
201	\
202	if (wrk->dirp) \
203	{ \
204	closedir (wrk->dirp); \
205	wrk->dirp = 0; \
206	}
207
208	#define ETP_WORKER_COMMON \
209	void *dbuf; \
210	DIR *dirp;
211
212	/*****************************************************************************/	271	/*****************************************************************************/
213		272
214	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	273	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215		274
216	/* calculate time difference in ~1/EIO_TICKS of a second */	275	/* calculate time difference in ~1/EIO_TICKS of a second */
		276	ecb_inline int
217	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	277	tvdiff (struct timeval *tv1, struct timeval *tv2)
218	{	278	{
219	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	279	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	280	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
221	}	281	}
222		282
…		…
243	/*	303	/*
244	* make our pread/pwrite emulation safe against themselves, but not against	304	* make our pread/pwrite emulation safe against themselves, but not against
245	* normal read/write by using a mutex. slows down execution a lot,	305	* normal read/write by using a mutex. slows down execution a lot,
246	* but that's your problem, not mine.	306	* but that's your problem, not mine.
247	*/	307	*/
248	static xmutex_t preadwritelock = X_MUTEX_INIT;	308	static xmutex_t preadwritelock;
249	#endif	309	#endif
250		310
251	typedef struct etp_worker	311	typedef struct etp_worker
252	{	312	{
253	/* locked by wrklock */	313	/* locked by wrklock */
…		…
256	xthread_t tid;	316	xthread_t tid;
257		317
258	/* locked by reslock, reqlock or wrklock */	318	/* locked by reslock, reqlock or wrklock */
259	ETP_REQ *req; /* currently processed request */	319	ETP_REQ *req; /* currently processed request */
260		320
		321	#ifdef ETP_WORKER_COMMON
261	ETP_WORKER_COMMON	322	ETP_WORKER_COMMON
		323	#endif
262	} etp_worker;	324	} etp_worker;
263		325
264	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	326	static etp_worker wrk_first; /* NOT etp */
265		327
266	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	328	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	329	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268		330
269	/* worker threads management */	331	/* worker threads management */
270		332
		333	static void ecb_cold
271	static void etp_worker_clear (etp_worker *wrk)	334	etp_worker_clear (etp_worker *wrk)
272	{	335	{
273	ETP_WORKER_CLEAR (wrk);
274	}	336	}
275		337
		338	static void ecb_cold
276	static void etp_worker_free (etp_worker *wrk)	339	etp_worker_free (etp_worker *wrk)
277	{	340	{
278	wrk->next->prev = wrk->prev;	341	wrk->next->prev = wrk->prev;
279	wrk->prev->next = wrk->next;	342	wrk->prev->next = wrk->next;
280		343
281	free (wrk);	344	free (wrk);
282	}	345	}
283		346
284	static unsigned int etp_nreqs (void)	347	static unsigned int
		348	etp_nreqs (void)
285	{	349	{
286	int retval;	350	int retval;
287	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	351	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288	retval = nreqs;	352	retval = nreqs;
289	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	353	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290	return retval;	354	return retval;
291	}	355	}
292		356
293	static unsigned int etp_nready (void)	357	static unsigned int
		358	etp_nready (void)
294	{	359	{
295	unsigned int retval;	360	unsigned int retval;
296		361
297	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	362	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298	retval = nready;	363	retval = nready;
299	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	364	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300		365
301	return retval;	366	return retval;
302	}	367	}
303		368
304	static unsigned int etp_npending (void)	369	static unsigned int
		370	etp_npending (void)
305	{	371	{
306	unsigned int retval;	372	unsigned int retval;
307		373
308	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	374	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309	retval = npending;	375	retval = npending;
310	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	376	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311		377
312	return retval;	378	return retval;
313	}	379	}
314		380
315	static unsigned int etp_nthreads (void)	381	static unsigned int
		382	etp_nthreads (void)
316	{	383	{
317	unsigned int retval;	384	unsigned int retval;
318		385
319	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	386	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320	retval = started;	387	retval = started;
…		…
334	} etp_reqq;	401	} etp_reqq;
335		402
336	static etp_reqq req_queue;	403	static etp_reqq req_queue;
337	static etp_reqq res_queue;	404	static etp_reqq res_queue;
338		405
		406	static void ecb_noinline ecb_cold
		407	reqq_init (etp_reqq *q)
		408	{
		409	int pri;
		410
		411	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		412	q->qs[pri] = q->qe[pri] = 0;
		413
		414	q->size = 0;
		415	}
		416
		417	static int ecb_noinline
339	static int reqq_push (etp_reqq *q, ETP_REQ *req)	418	reqq_push (etp_reqq *q, ETP_REQ *req)
340	{	419	{
341	int pri = req->pri;	420	int pri = req->pri;
342	req->next = 0;	421	req->next = 0;
343		422
344	if (q->qe[pri])	423	if (q->qe[pri])
…		…
350	q->qe[pri] = q->qs[pri] = req;	429	q->qe[pri] = q->qs[pri] = req;
351		430
352	return q->size++;	431	return q->size++;
353	}	432	}
354		433
		434	static ETP_REQ * ecb_noinline
355	static ETP_REQ *reqq_shift (etp_reqq *q)	435	reqq_shift (etp_reqq *q)
356	{	436	{
357	int pri;	437	int pri;
358		438
359	if (!q->size)	439	if (!q->size)
360	return 0;	440	return 0;
…		…
375	}	455	}
376		456
377	abort ();	457	abort ();
378	}	458	}
379		459
380	static void etp_thread_init (void)	460	static int ecb_cold
		461	etp_init (void (*want_poll)(void), void (*done_poll)(void))
381	{	462	{
382	X_MUTEX_CREATE (wrklock);	463	X_MUTEX_CREATE (wrklock);
383	X_MUTEX_CREATE (reslock);	464	X_MUTEX_CREATE (reslock);
384	X_MUTEX_CREATE (reqlock);	465	X_MUTEX_CREATE (reqlock);
385	X_COND_CREATE (reqwait);	466	X_COND_CREATE (reqwait);
386	}
387		467
388	static void etp_atfork_prepare (void)	468	reqq_init (&req_queue);
389	{	469	reqq_init (&res_queue);
390	X_LOCK (wrklock);
391	X_LOCK (reqlock);
392	X_LOCK (reslock);
393	#if !HAVE_PREADWRITE
394	X_LOCK (preadwritelock);
395	#endif
396	}
397		470
398	static void etp_atfork_parent (void)	471	wrk_first.next =
399	{	472	wrk_first.prev = &wrk_first;
400	#if !HAVE_PREADWRITE
401	X_UNLOCK (preadwritelock);
402	#endif
403	X_UNLOCK (reslock);
404	X_UNLOCK (reqlock);
405	X_UNLOCK (wrklock);
406	}
407
408	static void etp_atfork_child (void)
409	{
410	ETP_REQ *prv;
411
412	while ((prv = reqq_shift (&req_queue)))
413	ETP_DESTROY (prv);
414
415	while ((prv = reqq_shift (&res_queue)))
416	ETP_DESTROY (prv);
417
418	while (wrk_first.next != &wrk_first)
419	{
420	etp_worker *wrk = wrk_first.next;
421
422	if (wrk->req)
423	ETP_DESTROY (wrk->req);
424
425	etp_worker_clear (wrk);
426	etp_worker_free (wrk);
427	}
428		473
429	started = 0;	474	started = 0;
430	idle = 0;	475	idle = 0;
431	nreqs = 0;	476	nreqs = 0;
432	nready = 0;	477	nready = 0;
433	npending = 0;	478	npending = 0;
434		479
435	etp_thread_init ();
436	}
437
438	static void
439	etp_once_init (void)
440	{
441	etp_thread_init ();
442	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
443	}
444
445	static int
446	etp_init (void (*want_poll)(void), void (*done_poll)(void))
447	{
448	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
449
450	pthread_once (&doinit, etp_once_init);
451
452	want_poll_cb = want_poll;	480	want_poll_cb = want_poll;
453	done_poll_cb = done_poll;	481	done_poll_cb = done_poll;
454		482
455	return 0;	483	return 0;
456	}	484	}
457		485
458	X_THREAD_PROC (etp_proc);	486	X_THREAD_PROC (etp_proc);
459		487
		488	static void ecb_cold
460	static void etp_start_thread (void)	489	etp_start_thread (void)
461	{	490	{
462	etp_worker *wrk = calloc (1, sizeof (etp_worker));	491	etp_worker *wrk = calloc (1, sizeof (etp_worker));
463		492
464	/*TODO*/	493	/*TODO*/
465	assert (("unable to allocate worker thread data", wrk));	494	assert (("unable to allocate worker thread data", wrk));
…		…
478	free (wrk);	507	free (wrk);
479		508
480	X_UNLOCK (wrklock);	509	X_UNLOCK (wrklock);
481	}	510	}
482		511
		512	static void
483	static void etp_maybe_start_thread (void)	513	etp_maybe_start_thread (void)
484	{	514	{
485	if (expect_true (etp_nthreads () >= wanted))	515	if (ecb_expect_true (etp_nthreads () >= wanted))
486	return;	516	return;
487		517
488	/* todo: maybe use idle here, but might be less exact */	518	/* todo: maybe use idle here, but might be less exact */
489	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	519	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
490	return;	520	return;
491		521
492	etp_start_thread ();	522	etp_start_thread ();
493	}	523	}
494		524
		525	static void ecb_cold
495	static void etp_end_thread (void)	526	etp_end_thread (void)
496	{	527	{
497	eio_req *req = calloc (1, sizeof (eio_req));	528	eio_req *req = calloc (1, sizeof (eio_req));
498		529
499	req->type = -1;	530	req->type = -1;
500	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	531	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
507	X_LOCK (wrklock);	538	X_LOCK (wrklock);
508	--started;	539	--started;
509	X_UNLOCK (wrklock);	540	X_UNLOCK (wrklock);
510	}	541	}
511		542
512	static int etp_poll (void)	543	static int
		544	etp_poll (void)
513	{	545	{
514	unsigned int maxreqs;	546	unsigned int maxreqs;
515	unsigned int maxtime;	547	unsigned int maxtime;
516	struct timeval tv_start, tv_now;	548	struct timeval tv_start, tv_now;
517		549
…		…
547		579
548	X_LOCK (reqlock);	580	X_LOCK (reqlock);
549	--nreqs;	581	--nreqs;
550	X_UNLOCK (reqlock);	582	X_UNLOCK (reqlock);
551		583
552	if (expect_false (req->type == EIO_GROUP && req->size))	584	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
553	{	585	{
554	req->int1 = 1; /* mark request as delayed */	586	req->int1 = 1; /* mark request as delayed */
555	continue;	587	continue;
556	}	588	}
557	else	589	else
558	{	590	{
559	int res = ETP_FINISH (req);	591	int res = ETP_FINISH (req);
560	if (expect_false (res))	592	if (ecb_expect_false (res))
561	return res;	593	return res;
562	}	594	}
563		595
564	if (expect_false (maxreqs && !--maxreqs))	596	if (ecb_expect_false (maxreqs && !--maxreqs))
565	break;	597	break;
566		598
567	if (maxtime)	599	if (maxtime)
568	{	600	{
569	gettimeofday (&tv_now, 0);	601	gettimeofday (&tv_now, 0);
…		…
575		607
576	errno = EAGAIN;	608	errno = EAGAIN;
577	return -1;	609	return -1;
578	}	610	}
579		611
		612	static void
580	static void etp_cancel (ETP_REQ *req)	613	etp_cancel (ETP_REQ *req)
581	{	614	{
582	X_LOCK (wrklock);	615	req->cancelled = 1;
583	req->flags |= EIO_FLAG_CANCELLED;
584	X_UNLOCK (wrklock);
585		616
586	eio_grp_cancel (req);	617	eio_grp_cancel (req);
587	}	618	}
588		619
		620	static void
589	static void etp_submit (ETP_REQ *req)	621	etp_submit (ETP_REQ *req)
590	{	622	{
591	req->pri -= ETP_PRI_MIN;	623	req->pri -= ETP_PRI_MIN;
592		624
593	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	625	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
594	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	626	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595		627
596	if (expect_false (req->type == EIO_GROUP))	628	if (ecb_expect_false (req->type == EIO_GROUP))
597	{	629	{
598	/* I hope this is worth it :/ */	630	/* I hope this is worth it :/ */
599	X_LOCK (reqlock);	631	X_LOCK (reqlock);
600	++nreqs;	632	++nreqs;
601	X_UNLOCK (reqlock);	633	X_UNLOCK (reqlock);
…		…
620		652
621	etp_maybe_start_thread ();	653	etp_maybe_start_thread ();
622	}	654	}
623	}	655	}
624		656
		657	static void ecb_cold
625	static void etp_set_max_poll_time (double nseconds)	658	etp_set_max_poll_time (double nseconds)
626	{	659	{
627	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	660	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628	max_poll_time = nseconds * EIO_TICKS;	661	max_poll_time = nseconds * EIO_TICKS;
629	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	662	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630	}	663	}
631		664
		665	static void ecb_cold
632	static void etp_set_max_poll_reqs (unsigned int maxreqs)	666	etp_set_max_poll_reqs (unsigned int maxreqs)
633	{	667	{
634	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	668	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635	max_poll_reqs = maxreqs;	669	max_poll_reqs = maxreqs;
636	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	670	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637	}	671	}
638		672
		673	static void ecb_cold
639	static void etp_set_max_idle (unsigned int nthreads)	674	etp_set_max_idle (unsigned int nthreads)
640	{	675	{
641	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	676	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642	max_idle = nthreads;	677	max_idle = nthreads;
643	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	678	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644	}	679	}
645		680
		681	static void ecb_cold
646	static void etp_set_idle_timeout (unsigned int seconds)	682	etp_set_idle_timeout (unsigned int seconds)
647	{	683	{
648	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	684	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649	idle_timeout = seconds;	685	idle_timeout = seconds;
650	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	686	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651	}	687	}
652		688
		689	static void ecb_cold
653	static void etp_set_min_parallel (unsigned int nthreads)	690	etp_set_min_parallel (unsigned int nthreads)
654	{	691	{
655	if (wanted < nthreads)	692	if (wanted < nthreads)
656	wanted = nthreads;	693	wanted = nthreads;
657	}	694	}
658		695
		696	static void ecb_cold
659	static void etp_set_max_parallel (unsigned int nthreads)	697	etp_set_max_parallel (unsigned int nthreads)
660	{	698	{
661	if (wanted > nthreads)	699	if (wanted > nthreads)
662	wanted = nthreads;	700	wanted = nthreads;
663		701
664	while (started > wanted)	702	while (started > wanted)
665	etp_end_thread ();	703	etp_end_thread ();
666	}	704	}
667		705
668	/*****************************************************************************/	706	/*****************************************************************************/
669		707
		708	static void
670	static void grp_try_feed (eio_req *grp)	709	grp_try_feed (eio_req *grp)
671	{	710	{
672	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	711	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673	{	712	{
674	grp->flags &= ~EIO_FLAG_GROUPADD;	713	grp->flags &= ~EIO_FLAG_GROUPADD;
675		714
…		…
682	break;	721	break;
683	}	722	}
684	}	723	}
685	}	724	}
686		725
		726	static int
687	static int grp_dec (eio_req *grp)	727	grp_dec (eio_req *grp)
688	{	728	{
689	--grp->size;	729	--grp->size;
690		730
691	/* call feeder, if applicable */	731	/* call feeder, if applicable */
692	grp_try_feed (grp);	732	grp_try_feed (grp);
…		…
696	return eio_finish (grp);	736	return eio_finish (grp);
697	else	737	else
698	return 0;	738	return 0;
699	}	739	}
700		740
		741	static void
701	void eio_destroy (eio_req *req)	742	eio_destroy (eio_req *req)
702	{	743	{
703	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	744	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	745	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705		746
706	EIO_DESTROY (req);	747	EIO_DESTROY (req);
707	}	748	}
708		749
		750	static int
709	static int eio_finish (eio_req *req)	751	eio_finish (eio_req *req)
710	{	752	{
711	int res = EIO_FINISH (req);	753	int res = EIO_FINISH (req);
712		754
713	if (req->grp)	755	if (req->grp)
714	{	756	{
…		…
722	if (grp->grp_first == req)	764	if (grp->grp_first == req)
723	grp->grp_first = req->grp_next;	765	grp->grp_first = req->grp_next;
724		766
725	res2 = grp_dec (grp);	767	res2 = grp_dec (grp);
726		768
727	if (!res && res2)	769	if (!res)
728	res = res2;	770	res = res2;
729	}	771	}
730		772
731	eio_destroy (req);	773	eio_destroy (req);
732		774
733	return res;	775	return res;
734	}	776	}
735		777
		778	void
736	void eio_grp_cancel (eio_req *grp)	779	eio_grp_cancel (eio_req *grp)
737	{	780	{
738	for (grp = grp->grp_first; grp; grp = grp->grp_next)	781	for (grp = grp->grp_first; grp; grp = grp->grp_next)
739	eio_cancel (grp);	782	eio_cancel (grp);
740	}	783	}
741		784
		785	void
742	void eio_cancel (eio_req *req)	786	eio_cancel (eio_req *req)
743	{	787	{
744	etp_cancel (req);	788	etp_cancel (req);
745	}	789	}
746		790
		791	void
747	void eio_submit (eio_req *req)	792	eio_submit (eio_req *req)
748	{	793	{
749	etp_submit (req);	794	etp_submit (req);
750	}	795	}
751		796
752	unsigned int eio_nreqs (void)	797	unsigned int
		798	eio_nreqs (void)
753	{	799	{
754	return etp_nreqs ();	800	return etp_nreqs ();
755	}	801	}
756		802
757	unsigned int eio_nready (void)	803	unsigned int
		804	eio_nready (void)
758	{	805	{
759	return etp_nready ();	806	return etp_nready ();
760	}	807	}
761		808
762	unsigned int eio_npending (void)	809	unsigned int
		810	eio_npending (void)
763	{	811	{
764	return etp_npending ();	812	return etp_npending ();
765	}	813	}
766		814
767	unsigned int eio_nthreads (void)	815	unsigned int ecb_cold
		816	eio_nthreads (void)
768	{	817	{
769	return etp_nthreads ();	818	return etp_nthreads ();
770	}	819	}
771		820
		821	void ecb_cold
772	void eio_set_max_poll_time (double nseconds)	822	eio_set_max_poll_time (double nseconds)
773	{	823	{
774	etp_set_max_poll_time (nseconds);	824	etp_set_max_poll_time (nseconds);
775	}	825	}
776		826
		827	void ecb_cold
777	void eio_set_max_poll_reqs (unsigned int maxreqs)	828	eio_set_max_poll_reqs (unsigned int maxreqs)
778	{	829	{
779	etp_set_max_poll_reqs (maxreqs);	830	etp_set_max_poll_reqs (maxreqs);
780	}	831	}
781		832
		833	void ecb_cold
782	void eio_set_max_idle (unsigned int nthreads)	834	eio_set_max_idle (unsigned int nthreads)
783	{	835	{
784	etp_set_max_idle (nthreads);	836	etp_set_max_idle (nthreads);
785	}	837	}
786		838
		839	void ecb_cold
787	void eio_set_idle_timeout (unsigned int seconds)	840	eio_set_idle_timeout (unsigned int seconds)
788	{	841	{
789	etp_set_idle_timeout (seconds);	842	etp_set_idle_timeout (seconds);
790	}	843	}
791		844
		845	void ecb_cold
792	void eio_set_min_parallel (unsigned int nthreads)	846	eio_set_min_parallel (unsigned int nthreads)
793	{	847	{
794	etp_set_min_parallel (nthreads);	848	etp_set_min_parallel (nthreads);
795	}	849	}
796		850
		851	void ecb_cold
797	void eio_set_max_parallel (unsigned int nthreads)	852	eio_set_max_parallel (unsigned int nthreads)
798	{	853	{
799	etp_set_max_parallel (nthreads);	854	etp_set_max_parallel (nthreads);
800	}	855	}
801		856
802	int eio_poll (void)	857	int eio_poll (void)
…		…
811	# undef pread	866	# undef pread
812	# undef pwrite	867	# undef pwrite
813	# define pread eio__pread	868	# define pread eio__pread
814	# define pwrite eio__pwrite	869	# define pwrite eio__pwrite
815		870
816	static ssize_t	871	static eio_ssize_t
817	eio__pread (int fd, void *buf, size_t count, off_t offset)	872	eio__pread (int fd, void *buf, size_t count, off_t offset)
818	{	873	{
819	ssize_t res;	874	eio_ssize_t res;
820	off_t ooffset;	875	off_t ooffset;
821		876
822	X_LOCK (preadwritelock);	877	X_LOCK (preadwritelock);
823	ooffset = lseek (fd, 0, SEEK_CUR);	878	ooffset = lseek (fd, 0, SEEK_CUR);
824	lseek (fd, offset, SEEK_SET);	879	lseek (fd, offset, SEEK_SET);
…		…
827	X_UNLOCK (preadwritelock);	882	X_UNLOCK (preadwritelock);
828		883
829	return res;	884	return res;
830	}	885	}
831		886
832	static ssize_t	887	static eio_ssize_t
833	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	888	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834	{	889	{
835	ssize_t res;	890	eio_ssize_t res;
836	off_t ooffset;	891	off_t ooffset;
837		892
838	X_LOCK (preadwritelock);	893	X_LOCK (preadwritelock);
839	ooffset = lseek (fd, 0, SEEK_CUR);	894	ooffset = lseek (fd, 0, SEEK_CUR);
840	lseek (fd, offset, SEEK_SET);	895	lseek (fd, offset, SEEK_SET);
…		…
872	#ifndef HAVE_FUTIMES	927	#ifndef HAVE_FUTIMES
873		928
874	# undef futimes	929	# undef futimes
875	# define futimes(fd,times) eio__futimes (fd, times)	930	# define futimes(fd,times) eio__futimes (fd, times)
876		931
		932	static int
877	static int eio__futimes (int fd, const struct timeval tv[2])	933	eio__futimes (int fd, const struct timeval tv[2])
878	{	934	{
879	errno = ENOSYS;	935	errno = ENOSYS;
880	return -1;	936	return -1;
881	}	937	}
882		938
…		…
885	#if !HAVE_FDATASYNC	941	#if !HAVE_FDATASYNC
886	# undef fdatasync	942	# undef fdatasync
887	# define fdatasync(fd) fsync (fd)	943	# define fdatasync(fd) fsync (fd)
888	#endif	944	#endif
889		945
		946	static int
		947	eio__syncfs (int fd)
		948	{
		949	int res;
		950
		951	#if HAVE_SYS_SYNCFS
		952	res = (int)syscall (__NR_syncfs, (int)(fd));
		953	#else
		954	res = -1;
		955	errno = ENOSYS;
		956	#endif
		957
		958	if (res < 0 && errno == ENOSYS && fd >= 0)
		959	sync ();
		960
		961	return res;
		962	}
		963
890	/* sync_file_range always needs emulation */	964	/* sync_file_range always needs emulation */
891	int	965	static int
892	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	966	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893	{	967	{
894	#if HAVE_SYNC_FILE_RANGE	968	#if HAVE_SYNC_FILE_RANGE
895	int res;	969	int res;
896		970
…		…
913	/* even though we could play tricks with the flags, it's better to always	987	/* even though we could play tricks with the flags, it's better to always
914	* call fdatasync, as that matches the expectation of its users best */	988	* call fdatasync, as that matches the expectation of its users best */
915	return fdatasync (fd);	989	return fdatasync (fd);
916	}	990	}
917		991
		992	static int
		993	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		994	{
		995	#if HAVE_FALLOCATE
		996	return fallocate (fd, mode, offset, len);
		997	#else
		998	errno = ENOSYS;
		999	return -1;
		1000	#endif
		1001	}
		1002
918	#if !HAVE_READAHEAD	1003	#if !HAVE_READAHEAD
919	# undef readahead	1004	# undef readahead
920	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1005	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921		1006
922	static ssize_t	1007	static eio_ssize_t
923	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1008	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
924	{	1009	{
925	size_t todo = count;	1010	size_t todo = count;
926	dBUF;	1011	dBUF;
927		1012
…		…
932	pread (fd, eio_buf, len, offset);	1017	pread (fd, eio_buf, len, offset);
933	offset += len;	1018	offset += len;
934	todo -= len;	1019	todo -= len;
935	}	1020	}
936		1021
		1022	FUBd;
		1023
937	errno = 0;	1024	errno = 0;
938	return count;	1025	return count;
939	}	1026	}
940		1027
941	#endif	1028	#endif
942		1029
943	/* sendfile always needs emulation */	1030	/* sendfile always needs emulation */
944	static ssize_t	1031	static eio_ssize_t
945	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1032	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
946	{	1033	{
		1034	eio_ssize_t written = 0;
947	ssize_t res;	1035	eio_ssize_t res;
948		1036
949	if (!count)	1037	if (!count)
950	return 0;	1038	return 0;
951		1039
		1040	for (;;)
		1041	{
		1042	#ifdef __APPLE__
		1043	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1044	#endif
952	#if HAVE_SENDFILE	1045	#if HAVE_SENDFILE
953	# if __linux	1046	# if __linux
		1047	off_t soffset = offset;
954	res = sendfile (ofd, ifd, &offset, count);	1048	res = sendfile (ofd, ifd, &soffset, count);
955		1049
956	# elif __FreeBSD__	1050	# elif __FreeBSD__
957	/*	1051	/*
958	* Of course, the freebsd sendfile is a dire hack with no thoughts	1052	* Of course, the freebsd sendfile is a dire hack with no thoughts
959	* wasted on making it similar to other I/O functions.	1053	* wasted on making it similar to other I/O functions.
960	*/	1054	*/
961	{
962	off_t sbytes;	1055	off_t sbytes;
963	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1056	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
964		1057
965	#if 0 /* according to the manpage, this is correct, but broken behaviour */	1058	#if 0 /* according to the manpage, this is correct, but broken behaviour */
966	/* freebsd' sendfile will return 0 on success */	1059	/* freebsd' sendfile will return 0 on success */
967	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */	1060	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
968	/* not on e.g. EIO or EPIPE - sounds broken */	1061	/* not on e.g. EIO or EPIPE - sounds broken */
969	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)	1062	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
970	res = sbytes;	1063	res = sbytes;
971	#endif	1064	#endif
972		1065
973	/* according to source inspection, this is correct, and useful behaviour */	1066	/* according to source inspection, this is correct, and useful behaviour */
974	if (sbytes)	1067	if (sbytes)
975	res = sbytes;	1068	res = sbytes;
976	}
977		1069
978	# elif defined (__APPLE__)	1070	# elif defined (__APPLE__)
979
980	{
981	off_t sbytes = count;	1071	off_t sbytes = count;
982	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	1072	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983		1073
984	/* according to the manpage, sbytes is always valid */	1074	/* according to the manpage, sbytes is always valid */
985	if (sbytes)	1075	if (sbytes)
986	res = sbytes;	1076	res = sbytes;
987	}
988		1077
989	# elif __hpux	1078	# elif __hpux
990	res = sendfile (ofd, ifd, offset, count, 0, 0);	1079	res = sendfile (ofd, ifd, offset, count, 0, 0);
991		1080
992	# elif __solaris	1081	# elif __solaris
993	{
994	struct sendfilevec vec;	1082	struct sendfilevec vec;
995	size_t sbytes;	1083	size_t sbytes;
996		1084
997	vec.sfv_fd = ifd;	1085	vec.sfv_fd = ifd;
998	vec.sfv_flag = 0;	1086	vec.sfv_flag = 0;
999	vec.sfv_off = offset;	1087	vec.sfv_off = offset;
1000	vec.sfv_len = count;	1088	vec.sfv_len = count;
1001		1089
1002	res = sendfilev (ofd, &vec, 1, &sbytes);	1090	res = sendfilev (ofd, &vec, 1, &sbytes);
1003		1091
1004	if (res < 0 && sbytes)	1092	if (res < 0 && sbytes)
1005	res = sbytes;	1093	res = sbytes;
1006	}
1007		1094
1008	# endif	1095	# endif
1009		1096
1010	#elif defined (_WIN32)	1097	#elif defined (_WIN32) && 0
1011
1012	/* does not work, just for documentation of what would need to be done */	1098	/* does not work, just for documentation of what would need to be done */
1013	{	1099	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1100	/* libeio guarantees that the file offset does not change, and windows */
		1101	/* has no way to get an independent handle to the same file description */
1014	HANDLE h = TO_SOCKET (ifd);	1102	HANDLE h = TO_SOCKET (ifd);
1015	SetFilePointer (h, offset, 0, FILE_BEGIN);	1103	SetFilePointer (h, offset, 0, FILE_BEGIN);
1016	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);	1104	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1017	}
1018		1105
1019	#else	1106	#else
1020	res = -1;	1107	res = -1;
1021	errno = ENOSYS;	1108	errno = ENOSYS;
1022	#endif	1109	#endif
1023		1110
		1111	/* we assume sendfile can copy at least 128mb in one go */
		1112	if (res <= 128 * 1024 * 1024)
		1113	{
		1114	if (res > 0)
		1115	written += res;
		1116
		1117	if (written)
		1118	return written;
		1119
		1120	break;
		1121	}
		1122	else
		1123	{
		1124	/* if we requested more, then probably the kernel was lazy */
		1125	written += res;
		1126	offset += res;
		1127	count -= res;
		1128
		1129	if (!count)
		1130	return written;
		1131	}
		1132	}
		1133
1024	if (res < 0	1134	if (res < 0
1025	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1135	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1026	/* BSDs */	1136	/* BSDs */
1027	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1137	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1028	|| errno == ENOTSUP	1138	|| errno == ENOTSUP
1029	#endif	1139	#endif
		1140	#ifdef EOPNOTSUPP /* windows */
1030	|| errno == EOPNOTSUPP /* BSDs */	1141	|| errno == EOPNOTSUPP /* BSDs */
		1142	#endif
1031	#if __solaris	1143	#if __solaris
1032	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1144	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1033	#endif	1145	#endif
1034	)	1146	)
1035	)	1147	)
…		…
1039		1151
1040	res = 0;	1152	res = 0;
1041		1153
1042	while (count)	1154	while (count)
1043	{	1155	{
1044	ssize_t cnt;	1156	eio_ssize_t cnt;
1045		1157
1046	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1158	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1047		1159
1048	if (cnt <= 0)	1160	if (cnt <= 0)
1049	{	1161	{
…		…
1061		1173
1062	offset += cnt;	1174	offset += cnt;
1063	res += cnt;	1175	res += cnt;
1064	count -= cnt;	1176	count -= cnt;
1065	}	1177	}
		1178
		1179	FUBd;
1066	}	1180	}
1067		1181
1068	return res;	1182	return res;
		1183	}
		1184
		1185	#ifdef PAGESIZE
		1186	# define eio_pagesize() PAGESIZE
		1187	#else
		1188	static intptr_t
		1189	eio_pagesize (void)
		1190	{
		1191	static intptr_t page;
		1192
		1193	if (!page)
		1194	page = sysconf (_SC_PAGESIZE);
		1195
		1196	return page;
		1197	}
		1198	#endif
		1199
		1200	static void
		1201	eio_page_align (void **addr, size_t *length)
		1202	{
		1203	intptr_t mask = eio_pagesize () - 1;
		1204
		1205	/* round down addr */
		1206	intptr_t adj = mask & (intptr_t)*addr;
		1207
		1208	*addr = (void *)((intptr_t)*addr - adj);
		1209	*length += adj;
		1210
		1211	/* round up length */
		1212	*length = (*length + mask) & ~mask;
		1213	}
		1214
		1215	#if !_POSIX_MEMLOCK
		1216	# define eio__mlockall(a) EIO_ENOSYS ()
		1217	#else
		1218
		1219	static int
		1220	eio__mlockall (int flags)
		1221	{
		1222	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1223	extern int mallopt (int, int);
		1224	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1225	#endif
		1226
		1227	if (EIO_MCL_CURRENT != MCL_CURRENT
		1228	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1229	{
		1230	flags = 0
		1231	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1232	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1233	}
		1234
		1235	return mlockall (flags);
		1236	}
		1237	#endif
		1238
		1239	#if !_POSIX_MEMLOCK_RANGE
		1240	# define eio__mlock(a,b) EIO_ENOSYS ()
		1241	#else
		1242
		1243	static int
		1244	eio__mlock (void *addr, size_t length)
		1245	{
		1246	eio_page_align (&addr, &length);
		1247
		1248	return mlock (addr, length);
		1249	}
		1250
		1251	#endif
		1252
		1253	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1254	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1255	#else
		1256
		1257	static int
		1258	eio__msync (void *mem, size_t len, int flags)
		1259	{
		1260	eio_page_align (&mem, &len);
		1261
		1262	if (EIO_MS_ASYNC != MS_SYNC
		1263	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1264	|| EIO_MS_SYNC != MS_SYNC)
		1265	{
		1266	flags = 0
		1267	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1268	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1269	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1270	}
		1271
		1272	return msync (mem, len, flags);
		1273	}
		1274
		1275	#endif
		1276
		1277	static int
		1278	eio__mtouch (eio_req *req)
		1279	{
		1280	void *mem = req->ptr2;
		1281	size_t len = req->size;
		1282	int flags = req->int1;
		1283
		1284	eio_page_align (&mem, &len);
		1285
		1286	{
		1287	intptr_t addr = (intptr_t)mem;
		1288	intptr_t end = addr + len;
		1289	intptr_t page = eio_pagesize ();
		1290
		1291	if (addr < end)
		1292	if (flags & EIO_MT_MODIFY) /* modify */
		1293	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1294	else
		1295	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1296	}
		1297
		1298	return 0;
		1299	}
		1300
		1301	/*****************************************************************************/
		1302	/* requests implemented outside eio_execute, because they are so large */
		1303
		1304	static void
		1305	eio__realpath (eio_req *req, etp_worker *self)
		1306	{
		1307	char *rel = req->ptr1;
		1308	char *res;
		1309	char *tmp1, *tmp2;
		1310	#if SYMLOOP_MAX > 32
		1311	int symlinks = SYMLOOP_MAX;
		1312	#else
		1313	int symlinks = 32;
		1314	#endif
		1315
		1316	req->result = -1;
		1317
		1318	errno = EINVAL;
		1319	if (!rel)
		1320	return;
		1321
		1322	errno = ENOENT;
		1323	if (!*rel)
		1324	return;
		1325
		1326	if (!req->ptr2)
		1327	{
		1328	X_LOCK (wrklock);
		1329	req->flags |= EIO_FLAG_PTR2_FREE;
		1330	X_UNLOCK (wrklock);
		1331	req->ptr2 = malloc (PATH_MAX * 3);
		1332
		1333	errno = ENOMEM;
		1334	if (!req->ptr2)
		1335	return;
		1336	}
		1337
		1338	res = req->ptr2;
		1339	tmp1 = res + PATH_MAX;
		1340	tmp2 = tmp1 + PATH_MAX;
		1341
		1342	#if 0 /* disabled, the musl way to do things is just too racy */
		1343	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1344	/* on linux we may be able to ask the kernel */
		1345	{
		1346	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1347
		1348	if (fd >= 0)
		1349	{
		1350	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1351	req->result = readlink (tmp1, res, PATH_MAX);
		1352	close (fd);
		1353
		1354	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1355
		1356	if (req->result > 0)
		1357	goto done;
		1358	}
		1359	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1360	return;
		1361	}
		1362	#endif
		1363	#endif
		1364
		1365	if (*rel != '/')
		1366	{
		1367	if (!getcwd (res, PATH_MAX))
		1368	return;
		1369
		1370	if (res [1]) /* only use if not / */
		1371	res += strlen (res);
		1372	}
		1373
		1374	while (*rel)
		1375	{
		1376	eio_ssize_t len, linklen;
		1377	char *beg = rel;
		1378
		1379	while (*rel && *rel != '/')
		1380	++rel;
		1381
		1382	len = rel - beg;
		1383
		1384	if (!len) /* skip slashes */
		1385	{
		1386	++rel;
		1387	continue;
		1388	}
		1389
		1390	if (beg [0] == '.')
		1391	{
		1392	if (len == 1)
		1393	continue; /* . - nop */
		1394
		1395	if (beg [1] == '.' && len == 2)
		1396	{
		1397	/* .. - back up one component, if possible */
		1398
		1399	while (res != req->ptr2)
		1400	if (*--res == '/')
		1401	break;
		1402
		1403	continue;
		1404	}
		1405	}
		1406
		1407	errno = ENAMETOOLONG;
		1408	if (res + 1 + len + 1 >= tmp1)
		1409	return;
		1410
		1411	/* copy one component */
		1412	*res = '/';
		1413	memcpy (res + 1, beg, len);
		1414
		1415	/* zero-terminate, for readlink */
		1416	res [len + 1] = 0;
		1417
		1418	/* now check if it's a symlink */
		1419	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1420
		1421	if (linklen < 0)
		1422	{
		1423	if (errno != EINVAL)
		1424	return;
		1425
		1426	/* it's a normal directory. hopefully */
		1427	res += len + 1;
		1428	}
		1429	else
		1430	{
		1431	/* yay, it was a symlink - build new path in tmp2 */
		1432	int rellen = strlen (rel);
		1433
		1434	errno = ENAMETOOLONG;
		1435	if (linklen + 1 + rellen >= PATH_MAX)
		1436	return;
		1437
		1438	errno = ELOOP;
		1439	if (!--symlinks)
		1440	return;
		1441
		1442	if (*tmp1 == '/')
		1443	res = req->ptr2; /* symlink resolves to an absolute path */
		1444
		1445	/* we need to be careful, as rel might point into tmp2 already */
		1446	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1447	tmp2 [linklen] = '/';
		1448	memcpy (tmp2, tmp1, linklen);
		1449
		1450	rel = tmp2;
		1451	}
		1452	}
		1453
		1454	/* special case for the lone root path */
		1455	if (res == req->ptr2)
		1456	*res++ = '/';
		1457
		1458	req->result = res - (char *)req->ptr2;
		1459
		1460	done:
		1461	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1069	}	1462	}
1070		1463
1071	static signed char	1464	static signed char
1072	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1465	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1073	{	1466	{
…		…
1081		1474
1082	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1475	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1083	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1476	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1084		1477
1085	static void	1478	static void
1086	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1479	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1087	{	1480	{
1088	unsigned char bits [9 + sizeof (ino_t) * 8];	1481	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1089	unsigned char *bit = bits;	1482	unsigned char *bit = bits;
1090		1483
1091	assert (CHAR_BIT == 8);	1484	assert (CHAR_BIT == 8);
1092	assert (sizeof (eio_dirent) * 8 < 256);	1485	assert (sizeof (eio_dirent) * 8 < 256);
1093	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1486	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1095		1488
1096	if (size <= EIO_SORT_FAST)	1489	if (size <= EIO_SORT_FAST)
1097	return;	1490	return;
1098		1491
1099	/* first prepare an array of bits to test in our radix sort */	1492	/* first prepare an array of bits to test in our radix sort */
1100	/* try to take endianness into account, as well as differences in ino_t sizes */	1493	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1101	/* inode_bits must contain all inodes ORed together */	1494	/* inode_bits must contain all inodes ORed together */
1102	/* which is used to skip bits that are 0 everywhere, which is very common */	1495	/* which is used to skip bits that are 0 everywhere, which is very common */
1103	{	1496	{
1104	ino_t endianness;	1497	eio_ino_t endianness;
1105	int i, j;	1498	int i, j;
1106		1499
1107	/* we store the byte offset of byte n into byte n of "endianness" */	1500	/* we store the byte offset of byte n into byte n of "endianness" */
1108	for (i = 0; i < sizeof (ino_t); ++i)	1501	for (i = 0; i < sizeof (eio_ino_t); ++i)
1109	((unsigned char *)&endianness)[i] = i;	1502	((unsigned char *)&endianness)[i] = i;
1110		1503
1111	*bit++ = 0;	1504	*bit++ = 0;
1112		1505
1113	for (i = 0; i < sizeof (ino_t); ++i)	1506	for (i = 0; i < sizeof (eio_ino_t); ++i)
1114	{	1507	{
1115	/* shifting off the byte offsets out of "endianness" */	1508	/* shifting off the byte offsets out of "endianness" */
1116	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1509	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1117	endianness >>= 8;	1510	endianness >>= 8;
1118		1511
1119	for (j = 0; j < 8; ++j)	1512	for (j = 0; j < 8; ++j)
1120	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1513	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1121	*bit++ = offs + j;	1514	*bit++ = offs + j;
1122	}	1515	}
1123		1516
1124	for (j = 0; j < 8; ++j)	1517	for (j = 0; j < 8; ++j)
1125	if (score_bits & (1 << j))	1518	if (score_bits & (1 << j))
1126	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1519	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1127	}	1520	}
1128		1521
1129	/* now actually do the sorting (a variant of MSD radix sort) */	1522	/* now actually do the sorting (a variant of MSD radix sort) */
1130	{	1523	{
1131	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1524	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1132	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1525	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1133	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1526	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1134	int stk_idx = 0;	1527	int stk_idx = 0;
1135		1528
1136	base_stk [stk_idx] = dents;	1529	base_stk [stk_idx] = dents;
1137	end_stk [stk_idx] = dents + size;	1530	end_stk [stk_idx] = dents + size;
1138	bit_stk [stk_idx] = bit - 1;	1531	bit_stk [stk_idx] = bit - 1;
…		…
1217	}	1610	}
1218	}	1611	}
1219	}	1612	}
1220		1613
1221	static void	1614	static void
1222	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1615	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1223	{	1616	{
1224	if (size <= 1)	1617	if (size <= 1)
1225	return; /* our insertion sort relies on size > 0 */	1618	return; /* our insertion sort relies on size > 0 */
1226		1619
1227	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1620	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1235		1628
1236	/* read a full directory */	1629	/* read a full directory */
1237	static void	1630	static void
1238	eio__scandir (eio_req *req, etp_worker *self)	1631	eio__scandir (eio_req *req, etp_worker *self)
1239	{	1632	{
1240	DIR *dirp;
1241	EIO_STRUCT_DIRENT *entp;
1242	char *name, *names;	1633	char *name, *names;
1243	int namesalloc = 4096;	1634	int namesalloc = 4096 - sizeof (void *) * 4;
1244	int namesoffs = 0;	1635	int namesoffs = 0;
1245	int flags = req->int1;	1636	int flags = req->int1;
1246	eio_dirent *dents = 0;	1637	eio_dirent *dents = 0;
1247	int dentalloc = 128;	1638	int dentalloc = 128;
1248	int dentoffs = 0;	1639	int dentoffs = 0;
1249	ino_t inode_bits = 0;	1640	eio_ino_t inode_bits = 0;
		1641	#ifdef _WIN32
		1642	HANDLE dirp;
		1643	WIN32_FIND_DATA entp;
		1644	#else
		1645	DIR *dirp;
		1646	EIO_STRUCT_DIRENT *entp;
		1647	#endif
1250		1648
1251	req->result = -1;	1649	req->result = -1;
1252		1650
1253	if (!(flags & EIO_READDIR_DENTS))	1651	if (!(flags & EIO_READDIR_DENTS))
1254	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1652	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1255		1653
1256	X_LOCK (wrklock);	1654	#ifdef _WIN32
1257	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1655	{
		1656	int len = strlen ((const char *)req->ptr1);
		1657	char *path = malloc (MAX_PATH);
		1658	const char *fmt;
		1659
		1660	if (!len)
		1661	fmt = "./*";
		1662	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1663	fmt = "%s*";
		1664	else
		1665	fmt = "%s/*";
		1666
		1667	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1668	dirp = FindFirstFile (path, &entp);
		1669	free (path);
		1670
		1671	if (dirp == INVALID_HANDLE_VALUE)
		1672	{
		1673	dirp = 0;
		1674
		1675	switch (GetLastError ())
		1676	{
		1677	case ERROR_FILE_NOT_FOUND:
		1678	req->result = 0;
		1679	break;
		1680
		1681	case ERROR_INVALID_NAME:
		1682	case ERROR_PATH_NOT_FOUND:
		1683	case ERROR_NO_MORE_FILES:
		1684	errno = ENOENT;
		1685	break;
		1686
		1687	case ERROR_NOT_ENOUGH_MEMORY:
		1688	errno = ENOMEM;
		1689	break;
		1690
		1691	default:
		1692	errno = EINVAL;
		1693	break;
		1694	}
		1695	}
		1696	}
		1697	#else
1258	self->dirp = dirp = opendir (req->ptr1);	1698	dirp = opendir (req->ptr1);
		1699	#endif
		1700
		1701	if (req->flags & EIO_FLAG_PTR1_FREE)
		1702	free (req->ptr1);
1259		1703
1260	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1704	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1261	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1705	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1262	req->ptr2 = names = malloc (namesalloc);	1706	req->ptr2 = names = malloc (namesalloc);
1263	X_UNLOCK (wrklock);
1264		1707
1265	if (dirp && names && (!flags || dents))	1708	if (dirp && names && (!flags || dents))
1266	for (;;)	1709	for (;;)
1267	{	1710	{
		1711	int done;
		1712
		1713	#ifdef _WIN32
		1714	done = !dirp;
		1715	#else
1268	errno = 0;	1716	errno = 0;
1269	entp = readdir (dirp);	1717	entp = readdir (dirp);
		1718	done = !entp;
		1719	#endif
1270		1720
1271	if (!entp)	1721	if (done)
1272	{	1722	{
		1723	#ifndef _WIN32
		1724	int old_errno = errno;
		1725	closedir (dirp);
		1726	errno = old_errno;
		1727
1273	if (errno)	1728	if (errno)
1274	break;	1729	break;
		1730	#endif
1275		1731
1276	/* sort etc. */	1732	/* sort etc. */
1277	req->int1 = flags;	1733	req->int1 = flags;
1278	req->result = dentoffs;	1734	req->result = dentoffs;
1279		1735
…		…
1308		1764
1309	break;	1765	break;
1310	}	1766	}
1311		1767
1312	/* now add the entry to our list(s) */	1768	/* now add the entry to our list(s) */
1313	name = entp->d_name;	1769	name = D_NAME (entp);
1314		1770
1315	/* skip . and .. entries */	1771	/* skip . and .. entries */
1316	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1772	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1317	{	1773	{
1318	int len = D_NAMLEN (entp) + 1;	1774	int len = D_NAMLEN (entp) + 1;
1319		1775
1320	while (expect_false (namesoffs + len > namesalloc))	1776	while (ecb_expect_false (namesoffs + len > namesalloc))
1321	{	1777	{
1322	namesalloc *= 2;	1778	namesalloc *= 2;
1323	X_LOCK (wrklock);
1324	req->ptr2 = names = realloc (names, namesalloc);	1779	req->ptr2 = names = realloc (names, namesalloc);
1325	X_UNLOCK (wrklock);
1326		1780
1327	if (!names)	1781	if (!names)
1328	break;	1782	break;
1329	}	1783	}
1330		1784
…		…
1332		1786
1333	if (dents)	1787	if (dents)
1334	{	1788	{
1335	struct eio_dirent *ent;	1789	struct eio_dirent *ent;
1336		1790
1337	if (expect_false (dentoffs == dentalloc))	1791	if (ecb_expect_false (dentoffs == dentalloc))
1338	{	1792	{
1339	dentalloc *= 2;	1793	dentalloc *= 2;
1340	X_LOCK (wrklock);
1341	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1794	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1342	X_UNLOCK (wrklock);
1343		1795
1344	if (!dents)	1796	if (!dents)
1345	break;	1797	break;
1346	}	1798	}
1347		1799
…		…
1427	if (EIO_CANCELLED (req))	1879	if (EIO_CANCELLED (req))
1428	{	1880	{
1429	errno = ECANCELED;	1881	errno = ECANCELED;
1430	break;	1882	break;
1431	}	1883	}
		1884
		1885	#ifdef _WIN32
		1886	if (!FindNextFile (dirp, &entp))
		1887	{
		1888	FindClose (dirp);
		1889	dirp = 0;
		1890	}
		1891	#endif
1432	}	1892	}
1433	}
1434
1435	#ifdef PAGESIZE
1436	# define eio_pagesize() PAGESIZE
1437	#else
1438	static intptr_t
1439	eio_pagesize (void)
1440	{
1441	static intptr_t page;
1442
1443	if (!page)
1444	page = sysconf (_SC_PAGESIZE);
1445
1446	return page;
1447	}
1448	#endif
1449
1450	static void
1451	eio_page_align (void **addr, size_t *length)
1452	{
1453	intptr_t mask = eio_pagesize () - 1;
1454
1455	/* round down addr */
1456	intptr_t adj = mask & (intptr_t)*addr;
1457
1458	*addr = (void *)((intptr_t)*addr - adj);
1459	*length += adj;
1460
1461	/* round up length */
1462	*length = (*length + mask) & ~mask;
1463	}
1464
1465	#if !_POSIX_MEMLOCK
1466	# define eio__mlockall(a) ((errno = ENOSYS), -1)
1467	#else
1468
1469	static int
1470	eio__mlockall (int flags)
1471	{
1472	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1473	extern int mallopt (int, int);
1474	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1475	#endif
1476
1477	if (EIO_MCL_CURRENT != MCL_CURRENT
1478	|| EIO_MCL_FUTURE != MCL_FUTURE)
1479	{
1480	flags = 0
1481	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1482	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1483	}
1484
1485	return mlockall (flags);
1486	}
1487	#endif
1488
1489	#if !_POSIX_MEMLOCK_RANGE
1490	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1491	#else
1492
1493	static int
1494	eio__mlock (void *addr, size_t length)
1495	{
1496	eio_page_align (&addr, &length);
1497
1498	return mlock (addr, length);
1499	}
1500
1501	#endif
1502
1503	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1504	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1505	#else
1506
1507	int
1508	eio__msync (void *mem, size_t len, int flags)
1509	{
1510	eio_page_align (&mem, &len);
1511
1512	if (EIO_MS_ASYNC != MS_SYNC
1513	|| EIO_MS_INVALIDATE != MS_INVALIDATE
1514	|| EIO_MS_SYNC != MS_SYNC)
1515	{
1516	flags = 0
1517	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1518	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1519	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1520	}
1521
1522	return msync (mem, len, flags);
1523	}
1524
1525	#endif
1526
1527	int
1528	eio__mtouch (void *mem, size_t len, int flags)
1529	{
1530	eio_page_align (&mem, &len);
1531
1532	{
1533	intptr_t addr = (intptr_t)mem;
1534	intptr_t end = addr + len;
1535	intptr_t page = eio_pagesize ();
1536
1537	if (addr < end)
1538	if (flags & EIO_MT_MODIFY) /* modify */
1539	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1540	else
1541	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1542	}
1543
1544	return 0;
1545	}	1893	}
1546		1894
1547	/*****************************************************************************/	1895	/*****************************************************************************/
1548		1896
1549	#define ALLOC(len) \	1897	#define ALLOC(len) \
…		…
1565	{	1913	{
1566	ETP_REQ *req;	1914	ETP_REQ *req;
1567	struct timespec ts;	1915	struct timespec ts;
1568	etp_worker *self = (etp_worker *)thr_arg;	1916	etp_worker *self = (etp_worker *)thr_arg;
1569		1917
1570	/* try to distribute timeouts somewhat randomly */	1918	/* try to distribute timeouts somewhat evenly */
1571	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1919	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1572		1920
1573	for (;;)	1921	for (;;)
1574	{	1922	{
		1923	ts.tv_sec = 0;
		1924
1575	X_LOCK (reqlock);	1925	X_LOCK (reqlock);
1576		1926
1577	for (;;)	1927	for (;;)
1578	{	1928	{
1579	self->req = req = reqq_shift (&req_queue);	1929	self->req = req = reqq_shift (&req_queue);
1580		1930
1581	if (req)	1931	if (req)
1582	break;	1932	break;
1583		1933
		1934	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1935	{
		1936	X_UNLOCK (reqlock);
		1937	X_LOCK (wrklock);
		1938	--started;
		1939	X_UNLOCK (wrklock);
		1940	goto quit;
		1941	}
		1942
1584	++idle;	1943	++idle;
1585		1944
1586	ts.tv_sec = time (0) + idle_timeout;	1945	if (idle <= max_idle)
1587	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1946	/* we are allowed to idle, so do so without any timeout */
		1947	X_COND_WAIT (reqwait, reqlock);
		1948	else
1588	{	1949	{
1589	if (idle > max_idle)	1950	/* initialise timeout once */
1590	{	1951	if (!ts.tv_sec)
1591	--idle;	1952	ts.tv_sec = time (0) + idle_timeout;
1592	X_UNLOCK (reqlock);
1593	X_LOCK (wrklock);
1594	--started;
1595	X_UNLOCK (wrklock);
1596	goto quit;
1597	}
1598		1953
1599	/* we are allowed to idle, so do so without any timeout */
1600	X_COND_WAIT (reqwait, reqlock);	1954	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1955	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1601	}	1956	}
1602		1957
1603	--idle;	1958	--idle;
1604	}	1959	}
1605		1960
…		…
1608	X_UNLOCK (reqlock);	1963	X_UNLOCK (reqlock);
1609		1964
1610	if (req->type < 0)	1965	if (req->type < 0)
1611	goto quit;	1966	goto quit;
1612		1967
1613	if (!EIO_CANCELLED (req))
1614	ETP_EXECUTE (self, req);	1968	ETP_EXECUTE (self, req);
1615		1969
1616	X_LOCK (reslock);	1970	X_LOCK (reslock);
1617		1971
1618	++npending;	1972	++npending;
1619		1973
…		…
1634	return 0;	1988	return 0;
1635	}	1989	}
1636		1990
1637	/*****************************************************************************/	1991	/*****************************************************************************/
1638		1992
		1993	int ecb_cold
1639	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1994	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1640	{	1995	{
		1996	#if !HAVE_PREADWRITE
		1997	X_MUTEX_CREATE (preadwritelock);
		1998	#endif
		1999
1641	return etp_init (want_poll, done_poll);	2000	return etp_init (want_poll, done_poll);
1642	}	2001	}
1643		2002
		2003	ecb_inline void
1644	static void eio_api_destroy (eio_req *req)	2004	eio_api_destroy (eio_req *req)
1645	{	2005	{
1646	free (req);	2006	free (req);
1647	}	2007	}
1648		2008
1649	#define REQ(rtype) \	2009	#define REQ(rtype) \
…		…
1668	{ \	2028	{ \
1669	eio_api_destroy (req); \	2029	eio_api_destroy (req); \
1670	return 0; \	2030	return 0; \
1671	}	2031	}
1672		2032
		2033	static void
1673	static void eio_execute (etp_worker *self, eio_req *req)	2034	eio_execute (etp_worker *self, eio_req *req)
1674	{	2035	{
		2036	if (ecb_expect_false (EIO_CANCELLED (req)))
		2037	{
		2038	req->result = -1;
		2039	req->errorno = ECANCELED;
		2040	return;
		2041	}
		2042
1675	switch (req->type)	2043	switch (req->type)
1676	{	2044	{
1677	case EIO_READ: ALLOC (req->size);	2045	case EIO_READ: ALLOC (req->size);
1678	req->result = req->offs >= 0	2046	req->result = req->offs >= 0
1679	? pread (req->int1, req->ptr2, req->size, req->offs)	2047	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1681	case EIO_WRITE: req->result = req->offs >= 0	2049	case EIO_WRITE: req->result = req->offs >= 0
1682	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2050	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1683	: write (req->int1, req->ptr2, req->size); break;	2051	: write (req->int1, req->ptr2, req->size); break;
1684		2052
1685	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2053	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1686	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2054	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1687		2055
1688	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2056	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1689	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2057	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1690	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2058	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1691	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2059	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
…		…
1713	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2081	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1714	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2082	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1715	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2083	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1716	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;	2084	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1717		2085
		2086	case EIO_REALPATH: eio__realpath (req, self); break;
		2087
1718	case EIO_READLINK: ALLOC (PATH_MAX);	2088	case EIO_READLINK: ALLOC (PATH_MAX);
1719	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;	2089	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1720		2090
1721	case EIO_SYNC: req->result = 0; sync (); break;	2091	case EIO_SYNC: req->result = 0; sync (); break;
1722	case EIO_FSYNC: req->result = fsync (req->int1); break;	2092	case EIO_FSYNC: req->result = fsync (req->int1); break;
1723	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2093	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2094	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2095	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1724	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2096	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1725	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	2097	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1726	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2098	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1727	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2099	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1728	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	2100	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1729		2101
1730	case EIO_READDIR: eio__scandir (req, self); break;	2102	case EIO_READDIR: eio__scandir (req, self); break;
1731		2103
1732	case EIO_BUSY:	2104	case EIO_BUSY:
1733	#ifdef _WIN32	2105	#ifdef _WIN32
…		…
1774	case EIO_NOP:	2146	case EIO_NOP:
1775	req->result = 0;	2147	req->result = 0;
1776	break;	2148	break;
1777		2149
1778	case EIO_CUSTOM:	2150	case EIO_CUSTOM:
1779	((void (*)(eio_req *))req->feed) (req);	2151	req->feed (req);
1780	break;	2152	break;
1781		2153
1782	default:	2154	default:
1783	errno = ENOSYS;	2155	errno = ENOSYS;
1784	req->result = -1;	2156	req->result = -1;
…		…
1813	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2185	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1814	{	2186	{
1815	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2187	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1816	}	2188	}
1817		2189
		2190	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		2191	{
		2192	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2193	}
		2194
		2195	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2196	{
		2197	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2198	}
		2199
		2200	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2201	{
		2202	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2203	}
		2204
1818	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2205	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1819	{	2206	{
1820	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2207	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1821	}	2208	}
1822		2209
…		…
1828	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)	2215	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1829	{	2216	{
1830	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;	2217	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1831	}	2218	}
1832		2219
1833	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2220	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1834	{	2221	{
1835	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2222	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1836	}
1837
1838	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1839	{
1840	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1841	}	2223	}
1842		2224
1843	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2225	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1844	{	2226	{
1845	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2227	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
1883	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2265	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1884	{	2266	{
1885	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2267	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1886	}	2268	}
1887		2269
1888	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2270	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1889	{	2271	{
1890	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2272	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1891	}	2273	}
1892		2274
1893	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2275	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1913	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2295	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1914	{	2296	{
1915	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2297	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1916	}	2298	}
1917		2299
1918	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2300	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1919	{	2301	{
1920	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2302	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1921	}	2303	}
1922		2304
1923	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2305	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1937	}	2319	}
1938		2320
1939	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2321	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1940	{	2322	{
1941	return eio__1path (EIO_READLINK, path, pri, cb, data);	2323	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2324	}
		2325
		2326	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2327	{
		2328	return eio__1path (EIO_REALPATH, path, pri, cb, data);
1942	}	2329	}
1943		2330
1944	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2331	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1945	{	2332	{
1946	return eio__1path (EIO_STAT, path, pri, cb, data);	2333	return eio__1path (EIO_STAT, path, pri, cb, data);
…		…
2005	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2392	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2006	{	2393	{
2007	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2394	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2008	}	2395	}
2009		2396
2010	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2397	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2011	{	2398	{
2012	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2399	REQ (EIO_CUSTOM); req->feed = execute; SEND;
2013	}	2400	}
2014		2401
2015	#endif	2402	#endif
2016		2403
2017	eio_req *eio_grp (eio_cb cb, void *data)	2404	eio_req *eio_grp (eio_cb cb, void *data)
…		…
2026	#undef SEND	2413	#undef SEND
2027		2414
2028	/*****************************************************************************/	2415	/*****************************************************************************/
2029	/* grp functions */	2416	/* grp functions */
2030		2417
		2418	void
2031	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2419	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2032	{	2420	{
2033	grp->int2 = limit;	2421	grp->int2 = limit;
2034	grp->feed = feed;	2422	grp->feed = feed;
2035		2423
2036	grp_try_feed (grp);	2424	grp_try_feed (grp);
2037	}	2425	}
2038		2426
		2427	void
2039	void eio_grp_limit (eio_req *grp, int limit)	2428	eio_grp_limit (eio_req *grp, int limit)
2040	{	2429	{
2041	grp->int2 = limit;	2430	grp->int2 = limit;
2042		2431
2043	grp_try_feed (grp);	2432	grp_try_feed (grp);
2044	}	2433	}
2045		2434
		2435	void
2046	void eio_grp_add (eio_req *grp, eio_req *req)	2436	eio_grp_add (eio_req *grp, eio_req *req)
2047	{	2437	{
2048	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2438	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2049		2439
2050	grp->flags |= EIO_FLAG_GROUPADD;	2440	grp->flags |= EIO_FLAG_GROUPADD;
2051		2441
…		…
2062	}	2452	}
2063		2453
2064	/*****************************************************************************/	2454	/*****************************************************************************/
2065	/* misc garbage */	2455	/* misc garbage */
2066		2456
		2457	eio_ssize_t
2067	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2458	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2068	{	2459	{
2069	etp_worker wrk;
2070	ssize_t ret;
2071
2072	wrk.dbuf = 0;
2073
2074	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2460	return eio__sendfile (ofd, ifd, offset, count);
2075
2076	if (wrk.dbuf)
2077	free (wrk.dbuf);
2078
2079	return ret;
2080	}	2461	}
2081		2462

Diff Legend

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