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 1.95 by root, Tue Jul 19 04:56:43 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>
89	# include <utime.h>	169	#include <utime.h>
90	# include <signal.h>	170	#include <signal.h>
91	# include <dirent.h>	171	#include <dirent.h>
92		172
93	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES	173	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94	# include <sys/mman.h>	174	#include <sys/mman.h>
95	#endif	175	#endif
96		176
		177	#define D_NAME(entp) entp->d_name
		178
97	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */	179	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__	180	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99	# define _DIRENT_HAVE_D_TYPE /* sigh */	181	#define _DIRENT_HAVE_D_TYPE /* sigh */
100	# define D_INO(de) (de)->d_fileno	182	#define D_INO(de) (de)->d_fileno
101	# define D_NAMLEN(de) (de)->d_namlen	183	#define D_NAMLEN(de) (de)->d_namlen
102	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600	184	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103	# define D_INO(de) (de)->d_ino	185	#define D_INO(de) (de)->d_ino
104	# endif	186	#endif
105		187
106	#ifdef _D_EXACT_NAMLEN	188	#ifdef _D_EXACT_NAMLEN
107	# undef D_NAMLEN	189	#undef D_NAMLEN
108	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)	190	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109	#endif	191	#endif
110		192
111	# ifdef _DIRENT_HAVE_D_TYPE	193	#ifdef _DIRENT_HAVE_D_TYPE
112	# define D_TYPE(de) (de)->d_type	194	#define D_TYPE(de) (de)->d_type
113	# endif	195	#endif
114		196
115	# ifndef EIO_STRUCT_DIRENT	197	#ifndef EIO_STRUCT_DIRENT
116	# define EIO_STRUCT_DIRENT struct dirent	198	#define EIO_STRUCT_DIRENT struct dirent
117	# endif	199	#endif
118		200
119	#endif	201	#endif
120		202
121	#if HAVE_SENDFILE	203	#if HAVE_SENDFILE
122	# if __linux	204	# if __linux
…		…
138	#endif	220	#endif
139	#ifndef D_INO	221	#ifndef D_INO
140	# define D_INO(de) 0	222	# define D_INO(de) 0
141	#endif	223	#endif
142	#ifndef D_NAMLEN	224	#ifndef D_NAMLEN
143	# define D_NAMLEN(de) strlen ((de)->d_name)	225	# define D_NAMLEN(entp) strlen (D_NAME (entp))
144	#endif	226	#endif
145		227
146	/* used for struct dirent, AIX doesn't provide it */	228	/* used for struct dirent, AIX doesn't provide it */
147	#ifndef NAME_MAX	229	#ifndef NAME_MAX
148	# define NAME_MAX 4096	230	# define NAME_MAX 4096
…		…
155		237
156	/* buffer size for various temporary buffers */	238	/* buffer size for various temporary buffers */
157	#define EIO_BUFSIZE 65536	239	#define EIO_BUFSIZE 65536
158		240
159	#define dBUF \	241	#define dBUF \
160	char *eio_buf; \
161	ETP_WORKER_LOCK (self); \
162	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	242	char *eio_buf = malloc (EIO_BUFSIZE); \
163	ETP_WORKER_UNLOCK (self); \
164	errno = ENOMEM; \	243	errno = ENOMEM; \
165	if (!eio_buf) \	244	if (!eio_buf) \
166	return -1;	245	return -1
		246
		247	#define FUBd \
		248	free (eio_buf)
167		249
168	#define EIO_TICKS ((1000000 + 1023) >> 10)	250	#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		251
183	#define ETP_PRI_MIN EIO_PRI_MIN	252	#define ETP_PRI_MIN EIO_PRI_MIN
184	#define ETP_PRI_MAX EIO_PRI_MAX	253	#define ETP_PRI_MAX EIO_PRI_MAX
185		254
186	struct etp_worker;	255	struct etp_worker;
…		…
190	static int eio_finish (eio_req *req);	259	static int eio_finish (eio_req *req);
191	#define ETP_FINISH(req) eio_finish (req)	260	#define ETP_FINISH(req) eio_finish (req)
192	static void eio_execute (struct etp_worker *self, eio_req *req);	261	static void eio_execute (struct etp_worker *self, eio_req *req);
193	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	262	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194		263
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	/*****************************************************************************/	264	/*****************************************************************************/
213		265
214	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	266	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215		267
216	/* calculate time difference in ~1/EIO_TICKS of a second */	268	/* calculate time difference in ~1/EIO_TICKS of a second */
		269	ecb_inline int
217	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	270	tvdiff (struct timeval *tv1, struct timeval *tv2)
218	{	271	{
219	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	272	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	273	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
221	}	274	}
222		275
…		…
243	/*	296	/*
244	* make our pread/pwrite emulation safe against themselves, but not against	297	* make our pread/pwrite emulation safe against themselves, but not against
245	* normal read/write by using a mutex. slows down execution a lot,	298	* normal read/write by using a mutex. slows down execution a lot,
246	* but that's your problem, not mine.	299	* but that's your problem, not mine.
247	*/	300	*/
248	static xmutex_t preadwritelock = X_MUTEX_INIT;	301	static xmutex_t preadwritelock;
249	#endif	302	#endif
250		303
251	typedef struct etp_worker	304	typedef struct etp_worker
252	{	305	{
253	/* locked by wrklock */	306	/* locked by wrklock */
…		…
256	xthread_t tid;	309	xthread_t tid;
257		310
258	/* locked by reslock, reqlock or wrklock */	311	/* locked by reslock, reqlock or wrklock */
259	ETP_REQ *req; /* currently processed request */	312	ETP_REQ *req; /* currently processed request */
260		313
		314	#ifdef ETP_WORKER_COMMON
261	ETP_WORKER_COMMON	315	ETP_WORKER_COMMON
		316	#endif
262	} etp_worker;	317	} etp_worker;
263		318
264	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	319	static etp_worker wrk_first; /* NOT etp */
265		320
266	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	321	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	322	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268		323
269	/* worker threads management */	324	/* worker threads management */
270		325
		326	static void ecb_cold
271	static void etp_worker_clear (etp_worker *wrk)	327	etp_worker_clear (etp_worker *wrk)
272	{	328	{
273	ETP_WORKER_CLEAR (wrk);
274	}	329	}
275		330
		331	static void ecb_cold
276	static void etp_worker_free (etp_worker *wrk)	332	etp_worker_free (etp_worker *wrk)
277	{	333	{
278	wrk->next->prev = wrk->prev;	334	wrk->next->prev = wrk->prev;
279	wrk->prev->next = wrk->next;	335	wrk->prev->next = wrk->next;
280		336
281	free (wrk);	337	free (wrk);
282	}	338	}
283		339
284	static unsigned int etp_nreqs (void)	340	static unsigned int
		341	etp_nreqs (void)
285	{	342	{
286	int retval;	343	int retval;
287	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	344	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288	retval = nreqs;	345	retval = nreqs;
289	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	346	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290	return retval;	347	return retval;
291	}	348	}
292		349
293	static unsigned int etp_nready (void)	350	static unsigned int
		351	etp_nready (void)
294	{	352	{
295	unsigned int retval;	353	unsigned int retval;
296		354
297	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	355	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298	retval = nready;	356	retval = nready;
299	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	357	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300		358
301	return retval;	359	return retval;
302	}	360	}
303		361
304	static unsigned int etp_npending (void)	362	static unsigned int
		363	etp_npending (void)
305	{	364	{
306	unsigned int retval;	365	unsigned int retval;
307		366
308	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	367	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309	retval = npending;	368	retval = npending;
310	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	369	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311		370
312	return retval;	371	return retval;
313	}	372	}
314		373
315	static unsigned int etp_nthreads (void)	374	static unsigned int
		375	etp_nthreads (void)
316	{	376	{
317	unsigned int retval;	377	unsigned int retval;
318		378
319	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	379	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320	retval = started;	380	retval = started;
…		…
334	} etp_reqq;	394	} etp_reqq;
335		395
336	static etp_reqq req_queue;	396	static etp_reqq req_queue;
337	static etp_reqq res_queue;	397	static etp_reqq res_queue;
338		398
		399	static void ecb_noinline ecb_cold
		400	reqq_init (etp_reqq *q)
		401	{
		402	int pri;
		403
		404	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		405	q->qs[pri] = q->qe[pri] = 0;
		406
		407	q->size = 0;
		408	}
		409
		410	static int ecb_noinline
339	static int reqq_push (etp_reqq *q, ETP_REQ *req)	411	reqq_push (etp_reqq *q, ETP_REQ *req)
340	{	412	{
341	int pri = req->pri;	413	int pri = req->pri;
342	req->next = 0;	414	req->next = 0;
343		415
344	if (q->qe[pri])	416	if (q->qe[pri])
…		…
350	q->qe[pri] = q->qs[pri] = req;	422	q->qe[pri] = q->qs[pri] = req;
351		423
352	return q->size++;	424	return q->size++;
353	}	425	}
354		426
		427	static ETP_REQ * ecb_noinline
355	static ETP_REQ *reqq_shift (etp_reqq *q)	428	reqq_shift (etp_reqq *q)
356	{	429	{
357	int pri;	430	int pri;
358		431
359	if (!q->size)	432	if (!q->size)
360	return 0;	433	return 0;
…		…
375	}	448	}
376		449
377	abort ();	450	abort ();
378	}	451	}
379		452
380	static void etp_thread_init (void)	453	static int ecb_cold
		454	etp_init (void (*want_poll)(void), void (*done_poll)(void))
381	{	455	{
382	X_MUTEX_CREATE (wrklock);	456	X_MUTEX_CREATE (wrklock);
383	X_MUTEX_CREATE (reslock);	457	X_MUTEX_CREATE (reslock);
384	X_MUTEX_CREATE (reqlock);	458	X_MUTEX_CREATE (reqlock);
385	X_COND_CREATE (reqwait);	459	X_COND_CREATE (reqwait);
386	}
387		460
388	static void etp_atfork_prepare (void)	461	reqq_init (&req_queue);
389	{	462	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		463
398	static void etp_atfork_parent (void)	464	wrk_first.next =
399	{	465	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		466
429	started = 0;	467	started = 0;
430	idle = 0;	468	idle = 0;
431	nreqs = 0;	469	nreqs = 0;
432	nready = 0;	470	nready = 0;
433	npending = 0;	471	npending = 0;
434		472
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;	473	want_poll_cb = want_poll;
453	done_poll_cb = done_poll;	474	done_poll_cb = done_poll;
454		475
455	return 0;	476	return 0;
456	}	477	}
457		478
458	X_THREAD_PROC (etp_proc);	479	X_THREAD_PROC (etp_proc);
459		480
		481	static void ecb_cold
460	static void etp_start_thread (void)	482	etp_start_thread (void)
461	{	483	{
462	etp_worker *wrk = calloc (1, sizeof (etp_worker));	484	etp_worker *wrk = calloc (1, sizeof (etp_worker));
463		485
464	/*TODO*/	486	/*TODO*/
465	assert (("unable to allocate worker thread data", wrk));	487	assert (("unable to allocate worker thread data", wrk));
…		…
478	free (wrk);	500	free (wrk);
479		501
480	X_UNLOCK (wrklock);	502	X_UNLOCK (wrklock);
481	}	503	}
482		504
		505	static void
483	static void etp_maybe_start_thread (void)	506	etp_maybe_start_thread (void)
484	{	507	{
485	if (expect_true (etp_nthreads () >= wanted))	508	if (ecb_expect_true (etp_nthreads () >= wanted))
486	return;	509	return;
487		510
488	/* todo: maybe use idle here, but might be less exact */	511	/* todo: maybe use idle here, but might be less exact */
489	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	512	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
490	return;	513	return;
491		514
492	etp_start_thread ();	515	etp_start_thread ();
493	}	516	}
494		517
		518	static void ecb_cold
495	static void etp_end_thread (void)	519	etp_end_thread (void)
496	{	520	{
497	eio_req *req = calloc (1, sizeof (eio_req));	521	eio_req *req = calloc (1, sizeof (eio_req));
498		522
499	req->type = -1;	523	req->type = -1;
500	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	524	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
507	X_LOCK (wrklock);	531	X_LOCK (wrklock);
508	--started;	532	--started;
509	X_UNLOCK (wrklock);	533	X_UNLOCK (wrklock);
510	}	534	}
511		535
512	static int etp_poll (void)	536	static int
		537	etp_poll (void)
513	{	538	{
514	unsigned int maxreqs;	539	unsigned int maxreqs;
515	unsigned int maxtime;	540	unsigned int maxtime;
516	struct timeval tv_start, tv_now;	541	struct timeval tv_start, tv_now;
517		542
…		…
547		572
548	X_LOCK (reqlock);	573	X_LOCK (reqlock);
549	--nreqs;	574	--nreqs;
550	X_UNLOCK (reqlock);	575	X_UNLOCK (reqlock);
551		576
552	if (expect_false (req->type == EIO_GROUP && req->size))	577	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
553	{	578	{
554	req->int1 = 1; /* mark request as delayed */	579	req->int1 = 1; /* mark request as delayed */
555	continue;	580	continue;
556	}	581	}
557	else	582	else
558	{	583	{
559	int res = ETP_FINISH (req);	584	int res = ETP_FINISH (req);
560	if (expect_false (res))	585	if (ecb_expect_false (res))
561	return res;	586	return res;
562	}	587	}
563		588
564	if (expect_false (maxreqs && !--maxreqs))	589	if (ecb_expect_false (maxreqs && !--maxreqs))
565	break;	590	break;
566		591
567	if (maxtime)	592	if (maxtime)
568	{	593	{
569	gettimeofday (&tv_now, 0);	594	gettimeofday (&tv_now, 0);
…		…
575		600
576	errno = EAGAIN;	601	errno = EAGAIN;
577	return -1;	602	return -1;
578	}	603	}
579		604
		605	static void
580	static void etp_cancel (ETP_REQ *req)	606	etp_cancel (ETP_REQ *req)
581	{	607	{
582	X_LOCK (wrklock);	608	req->cancelled = 1;
583	req->flags |= EIO_FLAG_CANCELLED;
584	X_UNLOCK (wrklock);
585		609
586	eio_grp_cancel (req);	610	eio_grp_cancel (req);
587	}	611	}
588		612
		613	static void
589	static void etp_submit (ETP_REQ *req)	614	etp_submit (ETP_REQ *req)
590	{	615	{
591	req->pri -= ETP_PRI_MIN;	616	req->pri -= ETP_PRI_MIN;
592		617
593	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	618	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;	619	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595		620
596	if (expect_false (req->type == EIO_GROUP))	621	if (ecb_expect_false (req->type == EIO_GROUP))
597	{	622	{
598	/* I hope this is worth it :/ */	623	/* I hope this is worth it :/ */
599	X_LOCK (reqlock);	624	X_LOCK (reqlock);
600	++nreqs;	625	++nreqs;
601	X_UNLOCK (reqlock);	626	X_UNLOCK (reqlock);
…		…
620		645
621	etp_maybe_start_thread ();	646	etp_maybe_start_thread ();
622	}	647	}
623	}	648	}
624		649
		650	static void ecb_cold
625	static void etp_set_max_poll_time (double nseconds)	651	etp_set_max_poll_time (double nseconds)
626	{	652	{
627	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	653	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628	max_poll_time = nseconds * EIO_TICKS;	654	max_poll_time = nseconds * EIO_TICKS;
629	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	655	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630	}	656	}
631		657
		658	static void ecb_cold
632	static void etp_set_max_poll_reqs (unsigned int maxreqs)	659	etp_set_max_poll_reqs (unsigned int maxreqs)
633	{	660	{
634	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	661	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635	max_poll_reqs = maxreqs;	662	max_poll_reqs = maxreqs;
636	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	663	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637	}	664	}
638		665
		666	static void ecb_cold
639	static void etp_set_max_idle (unsigned int nthreads)	667	etp_set_max_idle (unsigned int nthreads)
640	{	668	{
641	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	669	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642	max_idle = nthreads;	670	max_idle = nthreads;
643	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	671	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644	}	672	}
645		673
		674	static void ecb_cold
646	static void etp_set_idle_timeout (unsigned int seconds)	675	etp_set_idle_timeout (unsigned int seconds)
647	{	676	{
648	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	677	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649	idle_timeout = seconds;	678	idle_timeout = seconds;
650	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	679	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651	}	680	}
652		681
		682	static void ecb_cold
653	static void etp_set_min_parallel (unsigned int nthreads)	683	etp_set_min_parallel (unsigned int nthreads)
654	{	684	{
655	if (wanted < nthreads)	685	if (wanted < nthreads)
656	wanted = nthreads;	686	wanted = nthreads;
657	}	687	}
658		688
		689	static void ecb_cold
659	static void etp_set_max_parallel (unsigned int nthreads)	690	etp_set_max_parallel (unsigned int nthreads)
660	{	691	{
661	if (wanted > nthreads)	692	if (wanted > nthreads)
662	wanted = nthreads;	693	wanted = nthreads;
663		694
664	while (started > wanted)	695	while (started > wanted)
665	etp_end_thread ();	696	etp_end_thread ();
666	}	697	}
667		698
668	/*****************************************************************************/	699	/*****************************************************************************/
669		700
		701	static void
670	static void grp_try_feed (eio_req *grp)	702	grp_try_feed (eio_req *grp)
671	{	703	{
672	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	704	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673	{	705	{
674	grp->flags &= ~EIO_FLAG_GROUPADD;	706	grp->flags &= ~EIO_FLAG_GROUPADD;
675		707
…		…
682	break;	714	break;
683	}	715	}
684	}	716	}
685	}	717	}
686		718
		719	static int
687	static int grp_dec (eio_req *grp)	720	grp_dec (eio_req *grp)
688	{	721	{
689	--grp->size;	722	--grp->size;
690		723
691	/* call feeder, if applicable */	724	/* call feeder, if applicable */
692	grp_try_feed (grp);	725	grp_try_feed (grp);
…		…
696	return eio_finish (grp);	729	return eio_finish (grp);
697	else	730	else
698	return 0;	731	return 0;
699	}	732	}
700		733
		734	static void
701	void eio_destroy (eio_req *req)	735	eio_destroy (eio_req *req)
702	{	736	{
703	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	737	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	738	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705		739
706	EIO_DESTROY (req);	740	EIO_DESTROY (req);
707	}	741	}
708		742
		743	static int
709	static int eio_finish (eio_req *req)	744	eio_finish (eio_req *req)
710	{	745	{
711	int res = EIO_FINISH (req);	746	int res = EIO_FINISH (req);
712		747
713	if (req->grp)	748	if (req->grp)
714	{	749	{
…		…
722	if (grp->grp_first == req)	757	if (grp->grp_first == req)
723	grp->grp_first = req->grp_next;	758	grp->grp_first = req->grp_next;
724		759
725	res2 = grp_dec (grp);	760	res2 = grp_dec (grp);
726		761
727	if (!res && res2)	762	if (!res)
728	res = res2;	763	res = res2;
729	}	764	}
730		765
731	eio_destroy (req);	766	eio_destroy (req);
732		767
733	return res;	768	return res;
734	}	769	}
735		770
		771	void
736	void eio_grp_cancel (eio_req *grp)	772	eio_grp_cancel (eio_req *grp)
737	{	773	{
738	for (grp = grp->grp_first; grp; grp = grp->grp_next)	774	for (grp = grp->grp_first; grp; grp = grp->grp_next)
739	eio_cancel (grp);	775	eio_cancel (grp);
740	}	776	}
741		777
		778	void
742	void eio_cancel (eio_req *req)	779	eio_cancel (eio_req *req)
743	{	780	{
744	etp_cancel (req);	781	etp_cancel (req);
745	}	782	}
746		783
		784	void
747	void eio_submit (eio_req *req)	785	eio_submit (eio_req *req)
748	{	786	{
749	etp_submit (req);	787	etp_submit (req);
750	}	788	}
751		789
752	unsigned int eio_nreqs (void)	790	unsigned int
		791	eio_nreqs (void)
753	{	792	{
754	return etp_nreqs ();	793	return etp_nreqs ();
755	}	794	}
756		795
757	unsigned int eio_nready (void)	796	unsigned int
		797	eio_nready (void)
758	{	798	{
759	return etp_nready ();	799	return etp_nready ();
760	}	800	}
761		801
762	unsigned int eio_npending (void)	802	unsigned int
		803	eio_npending (void)
763	{	804	{
764	return etp_npending ();	805	return etp_npending ();
765	}	806	}
766		807
767	unsigned int eio_nthreads (void)	808	unsigned int ecb_cold
		809	eio_nthreads (void)
768	{	810	{
769	return etp_nthreads ();	811	return etp_nthreads ();
770	}	812	}
771		813
		814	void ecb_cold
772	void eio_set_max_poll_time (double nseconds)	815	eio_set_max_poll_time (double nseconds)
773	{	816	{
774	etp_set_max_poll_time (nseconds);	817	etp_set_max_poll_time (nseconds);
775	}	818	}
776		819
		820	void ecb_cold
777	void eio_set_max_poll_reqs (unsigned int maxreqs)	821	eio_set_max_poll_reqs (unsigned int maxreqs)
778	{	822	{
779	etp_set_max_poll_reqs (maxreqs);	823	etp_set_max_poll_reqs (maxreqs);
780	}	824	}
781		825
		826	void ecb_cold
782	void eio_set_max_idle (unsigned int nthreads)	827	eio_set_max_idle (unsigned int nthreads)
783	{	828	{
784	etp_set_max_idle (nthreads);	829	etp_set_max_idle (nthreads);
785	}	830	}
786		831
		832	void ecb_cold
787	void eio_set_idle_timeout (unsigned int seconds)	833	eio_set_idle_timeout (unsigned int seconds)
788	{	834	{
789	etp_set_idle_timeout (seconds);	835	etp_set_idle_timeout (seconds);
790	}	836	}
791		837
		838	void ecb_cold
792	void eio_set_min_parallel (unsigned int nthreads)	839	eio_set_min_parallel (unsigned int nthreads)
793	{	840	{
794	etp_set_min_parallel (nthreads);	841	etp_set_min_parallel (nthreads);
795	}	842	}
796		843
		844	void ecb_cold
797	void eio_set_max_parallel (unsigned int nthreads)	845	eio_set_max_parallel (unsigned int nthreads)
798	{	846	{
799	etp_set_max_parallel (nthreads);	847	etp_set_max_parallel (nthreads);
800	}	848	}
801		849
802	int eio_poll (void)	850	int eio_poll (void)
…		…
811	# undef pread	859	# undef pread
812	# undef pwrite	860	# undef pwrite
813	# define pread eio__pread	861	# define pread eio__pread
814	# define pwrite eio__pwrite	862	# define pwrite eio__pwrite
815		863
816	static ssize_t	864	static eio_ssize_t
817	eio__pread (int fd, void *buf, size_t count, off_t offset)	865	eio__pread (int fd, void *buf, size_t count, off_t offset)
818	{	866	{
819	ssize_t res;	867	eio_ssize_t res;
820	off_t ooffset;	868	off_t ooffset;
821		869
822	X_LOCK (preadwritelock);	870	X_LOCK (preadwritelock);
823	ooffset = lseek (fd, 0, SEEK_CUR);	871	ooffset = lseek (fd, 0, SEEK_CUR);
824	lseek (fd, offset, SEEK_SET);	872	lseek (fd, offset, SEEK_SET);
…		…
827	X_UNLOCK (preadwritelock);	875	X_UNLOCK (preadwritelock);
828		876
829	return res;	877	return res;
830	}	878	}
831		879
832	static ssize_t	880	static eio_ssize_t
833	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	881	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834	{	882	{
835	ssize_t res;	883	eio_ssize_t res;
836	off_t ooffset;	884	off_t ooffset;
837		885
838	X_LOCK (preadwritelock);	886	X_LOCK (preadwritelock);
839	ooffset = lseek (fd, 0, SEEK_CUR);	887	ooffset = lseek (fd, 0, SEEK_CUR);
840	lseek (fd, offset, SEEK_SET);	888	lseek (fd, offset, SEEK_SET);
…		…
872	#ifndef HAVE_FUTIMES	920	#ifndef HAVE_FUTIMES
873		921
874	# undef futimes	922	# undef futimes
875	# define futimes(fd,times) eio__futimes (fd, times)	923	# define futimes(fd,times) eio__futimes (fd, times)
876		924
		925	static int
877	static int eio__futimes (int fd, const struct timeval tv[2])	926	eio__futimes (int fd, const struct timeval tv[2])
878	{	927	{
879	errno = ENOSYS;	928	errno = ENOSYS;
880	return -1;	929	return -1;
881	}	930	}
882		931
…		…
886	# undef fdatasync	935	# undef fdatasync
887	# define fdatasync(fd) fsync (fd)	936	# define fdatasync(fd) fsync (fd)
888	#endif	937	#endif
889		938
890	/* sync_file_range always needs emulation */	939	/* sync_file_range always needs emulation */
891	int	940	static int
892	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	941	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893	{	942	{
894	#if HAVE_SYNC_FILE_RANGE	943	#if HAVE_SYNC_FILE_RANGE
895	int res;	944	int res;
896		945
…		…
913	/* even though we could play tricks with the flags, it's better to always	962	/* 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 */	963	* call fdatasync, as that matches the expectation of its users best */
915	return fdatasync (fd);	964	return fdatasync (fd);
916	}	965	}
917		966
		967	static int
		968	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		969	{
		970	#if HAVE_FALLOCATE
		971	return fallocate (fd, mode, offset, len);
		972	#else
		973	errno = ENOSYS;
		974	return -1;
		975	#endif
		976	}
		977
918	#if !HAVE_READAHEAD	978	#if !HAVE_READAHEAD
919	# undef readahead	979	# undef readahead
920	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	980	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921		981
922	static ssize_t	982	static eio_ssize_t
923	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	983	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
924	{	984	{
925	size_t todo = count;	985	size_t todo = count;
926	dBUF;	986	dBUF;
927		987
…		…
932	pread (fd, eio_buf, len, offset);	992	pread (fd, eio_buf, len, offset);
933	offset += len;	993	offset += len;
934	todo -= len;	994	todo -= len;
935	}	995	}
936		996
		997	FUBd;
		998
937	errno = 0;	999	errno = 0;
938	return count;	1000	return count;
939	}	1001	}
940		1002
941	#endif	1003	#endif
942		1004
943	/* sendfile always needs emulation */	1005	/* sendfile always needs emulation */
944	static ssize_t	1006	static eio_ssize_t
945	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1007	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
946	{	1008	{
		1009	eio_ssize_t written = 0;
947	ssize_t res;	1010	eio_ssize_t res;
948		1011
949	if (!count)	1012	if (!count)
950	return 0;	1013	return 0;
951		1014
		1015	for (;;)
		1016	{
		1017	#ifdef __APPLE__
		1018	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1019	#endif
952	#if HAVE_SENDFILE	1020	#if HAVE_SENDFILE
953	# if __linux	1021	# if __linux
		1022	off_t soffset = offset;
954	res = sendfile (ofd, ifd, &offset, count);	1023	res = sendfile (ofd, ifd, &soffset, count);
955		1024
956	# elif __FreeBSD__	1025	# elif __FreeBSD__
957	/*	1026	/*
958	* Of course, the freebsd sendfile is a dire hack with no thoughts	1027	* Of course, the freebsd sendfile is a dire hack with no thoughts
959	* wasted on making it similar to other I/O functions.	1028	* wasted on making it similar to other I/O functions.
960	*/	1029	*/
961	{
962	off_t sbytes;	1030	off_t sbytes;
963	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1031	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
964		1032
965	#if 0 /* according to the manpage, this is correct, but broken behaviour */	1033	#if 0 /* according to the manpage, this is correct, but broken behaviour */
966	/* freebsd' sendfile will return 0 on success */	1034	/* freebsd' sendfile will return 0 on success */
967	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */	1035	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
968	/* not on e.g. EIO or EPIPE - sounds broken */	1036	/* not on e.g. EIO or EPIPE - sounds broken */
969	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)	1037	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
970	res = sbytes;	1038	res = sbytes;
971	#endif	1039	#endif
972		1040
973	/* according to source inspection, this is correct, and useful behaviour */	1041	/* according to source inspection, this is correct, and useful behaviour */
974	if (sbytes)	1042	if (sbytes)
975	res = sbytes;	1043	res = sbytes;
976	}
977		1044
978	# elif defined (__APPLE__)	1045	# elif defined (__APPLE__)
979
980	{
981	off_t sbytes = count;	1046	off_t sbytes = count;
982	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	1047	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983		1048
984	/* according to the manpage, sbytes is always valid */	1049	/* according to the manpage, sbytes is always valid */
985	if (sbytes)	1050	if (sbytes)
986	res = sbytes;	1051	res = sbytes;
987	}
988		1052
989	# elif __hpux	1053	# elif __hpux
990	res = sendfile (ofd, ifd, offset, count, 0, 0);	1054	res = sendfile (ofd, ifd, offset, count, 0, 0);
991		1055
992	# elif __solaris	1056	# elif __solaris
993	{
994	struct sendfilevec vec;	1057	struct sendfilevec vec;
995	size_t sbytes;	1058	size_t sbytes;
996		1059
997	vec.sfv_fd = ifd;	1060	vec.sfv_fd = ifd;
998	vec.sfv_flag = 0;	1061	vec.sfv_flag = 0;
999	vec.sfv_off = offset;	1062	vec.sfv_off = offset;
1000	vec.sfv_len = count;	1063	vec.sfv_len = count;
1001		1064
1002	res = sendfilev (ofd, &vec, 1, &sbytes);	1065	res = sendfilev (ofd, &vec, 1, &sbytes);
1003		1066
1004	if (res < 0 && sbytes)	1067	if (res < 0 && sbytes)
1005	res = sbytes;	1068	res = sbytes;
1006	}
1007		1069
1008	# endif	1070	# endif
1009		1071
1010	#elif defined (_WIN32)	1072	#elif defined (_WIN32) && 0
1011
1012	/* does not work, just for documentation of what would need to be done */	1073	/* does not work, just for documentation of what would need to be done */
1013	{	1074	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1075	/* libeio guarantees that the file offset does not change, and windows */
		1076	/* has no way to get an independent handle to the same file description */
1014	HANDLE h = TO_SOCKET (ifd);	1077	HANDLE h = TO_SOCKET (ifd);
1015	SetFilePointer (h, offset, 0, FILE_BEGIN);	1078	SetFilePointer (h, offset, 0, FILE_BEGIN);
1016	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);	1079	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1017	}
1018		1080
1019	#else	1081	#else
1020	res = -1;	1082	res = -1;
1021	errno = ENOSYS;	1083	errno = ENOSYS;
1022	#endif	1084	#endif
1023		1085
		1086	/* we assume sendfile can copy at least 128mb in one go */
		1087	if (res <= 128 * 1024 * 1024)
		1088	{
		1089	if (res > 0)
		1090	written += res;
		1091
		1092	if (written)
		1093	return written;
		1094
		1095	break;
		1096	}
		1097	else
		1098	{
		1099	/* if we requested more, then probably the kernel was lazy */
		1100	written += res;
		1101	offset += res;
		1102	count -= res;
		1103
		1104	if (!count)
		1105	return written;
		1106	}
		1107	}
		1108
1024	if (res < 0	1109	if (res < 0
1025	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1110	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1026	/* BSDs */	1111	/* BSDs */
1027	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1112	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1028	|| errno == ENOTSUP	1113	|| errno == ENOTSUP
1029	#endif	1114	#endif
		1115	#ifdef EOPNOTSUPP /* windows */
1030	|| errno == EOPNOTSUPP /* BSDs */	1116	|| errno == EOPNOTSUPP /* BSDs */
		1117	#endif
1031	#if __solaris	1118	#if __solaris
1032	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1119	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1033	#endif	1120	#endif
1034	)	1121	)
1035	)	1122	)
…		…
1039		1126
1040	res = 0;	1127	res = 0;
1041		1128
1042	while (count)	1129	while (count)
1043	{	1130	{
1044	ssize_t cnt;	1131	eio_ssize_t cnt;
1045		1132
1046	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1133	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1047		1134
1048	if (cnt <= 0)	1135	if (cnt <= 0)
1049	{	1136	{
…		…
1061		1148
1062	offset += cnt;	1149	offset += cnt;
1063	res += cnt;	1150	res += cnt;
1064	count -= cnt;	1151	count -= cnt;
1065	}	1152	}
		1153
		1154	FUBd;
1066	}	1155	}
1067		1156
1068	return res;	1157	return res;
		1158	}
		1159
		1160	#ifdef PAGESIZE
		1161	# define eio_pagesize() PAGESIZE
		1162	#else
		1163	static intptr_t
		1164	eio_pagesize (void)
		1165	{
		1166	static intptr_t page;
		1167
		1168	if (!page)
		1169	page = sysconf (_SC_PAGESIZE);
		1170
		1171	return page;
		1172	}
		1173	#endif
		1174
		1175	static void
		1176	eio_page_align (void **addr, size_t *length)
		1177	{
		1178	intptr_t mask = eio_pagesize () - 1;
		1179
		1180	/* round down addr */
		1181	intptr_t adj = mask & (intptr_t)*addr;
		1182
		1183	*addr = (void *)((intptr_t)*addr - adj);
		1184	*length += adj;
		1185
		1186	/* round up length */
		1187	*length = (*length + mask) & ~mask;
		1188	}
		1189
		1190	#if !_POSIX_MEMLOCK
		1191	# define eio__mlockall(a) EIO_ENOSYS ()
		1192	#else
		1193
		1194	static int
		1195	eio__mlockall (int flags)
		1196	{
		1197	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1198	extern int mallopt (int, int);
		1199	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1200	#endif
		1201
		1202	if (EIO_MCL_CURRENT != MCL_CURRENT
		1203	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1204	{
		1205	flags = 0
		1206	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1207	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1208	}
		1209
		1210	return mlockall (flags);
		1211	}
		1212	#endif
		1213
		1214	#if !_POSIX_MEMLOCK_RANGE
		1215	# define eio__mlock(a,b) EIO_ENOSYS ()
		1216	#else
		1217
		1218	static int
		1219	eio__mlock (void *addr, size_t length)
		1220	{
		1221	eio_page_align (&addr, &length);
		1222
		1223	return mlock (addr, length);
		1224	}
		1225
		1226	#endif
		1227
		1228	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1229	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1230	#else
		1231
		1232	static int
		1233	eio__msync (void *mem, size_t len, int flags)
		1234	{
		1235	eio_page_align (&mem, &len);
		1236
		1237	if (EIO_MS_ASYNC != MS_SYNC
		1238	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1239	|| EIO_MS_SYNC != MS_SYNC)
		1240	{
		1241	flags = 0
		1242	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1243	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1244	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1245	}
		1246
		1247	return msync (mem, len, flags);
		1248	}
		1249
		1250	#endif
		1251
		1252	static int
		1253	eio__mtouch (eio_req *req)
		1254	{
		1255	void *mem = req->ptr2;
		1256	size_t len = req->size;
		1257	int flags = req->int1;
		1258
		1259	eio_page_align (&mem, &len);
		1260
		1261	{
		1262	intptr_t addr = (intptr_t)mem;
		1263	intptr_t end = addr + len;
		1264	intptr_t page = eio_pagesize ();
		1265
		1266	if (addr < end)
		1267	if (flags & EIO_MT_MODIFY) /* modify */
		1268	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1269	else
		1270	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1271	}
		1272
		1273	return 0;
		1274	}
		1275
		1276	/*****************************************************************************/
		1277	/* requests implemented outside eio_execute, because they are so large */
		1278
		1279	static void
		1280	eio__realpath (eio_req *req, etp_worker *self)
		1281	{
		1282	char *rel = req->ptr1;
		1283	char *res;
		1284	char *tmp1, *tmp2;
		1285	#if SYMLOOP_MAX > 32
		1286	int symlinks = SYMLOOP_MAX;
		1287	#else
		1288	int symlinks = 32;
		1289	#endif
		1290
		1291	req->result = -1;
		1292
		1293	errno = EINVAL;
		1294	if (!rel)
		1295	return;
		1296
		1297	errno = ENOENT;
		1298	if (!*rel)
		1299	return;
		1300
		1301	if (!req->ptr2)
		1302	{
		1303	X_LOCK (wrklock);
		1304	req->flags |= EIO_FLAG_PTR2_FREE;
		1305	X_UNLOCK (wrklock);
		1306	req->ptr2 = malloc (PATH_MAX * 3);
		1307
		1308	errno = ENOMEM;
		1309	if (!req->ptr2)
		1310	return;
		1311	}
		1312
		1313	res = req->ptr2;
		1314	tmp1 = res + PATH_MAX;
		1315	tmp2 = tmp1 + PATH_MAX;
		1316
		1317	#if 0 /* disabled, the musl way to do things is just too racy */
		1318	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1319	/* on linux we may be able to ask the kernel */
		1320	{
		1321	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1322
		1323	if (fd >= 0)
		1324	{
		1325	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1326	req->result = readlink (tmp1, res, PATH_MAX);
		1327	close (fd);
		1328
		1329	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1330
		1331	if (req->result > 0)
		1332	goto done;
		1333	}
		1334	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1335	return;
		1336	}
		1337	#endif
		1338	#endif
		1339
		1340	if (*rel != '/')
		1341	{
		1342	if (!getcwd (res, PATH_MAX))
		1343	return;
		1344
		1345	if (res [1]) /* only use if not / */
		1346	res += strlen (res);
		1347	}
		1348
		1349	while (*rel)
		1350	{
		1351	eio_ssize_t len, linklen;
		1352	char *beg = rel;
		1353
		1354	while (*rel && *rel != '/')
		1355	++rel;
		1356
		1357	len = rel - beg;
		1358
		1359	if (!len) /* skip slashes */
		1360	{
		1361	++rel;
		1362	continue;
		1363	}
		1364
		1365	if (beg [0] == '.')
		1366	{
		1367	if (len == 1)
		1368	continue; /* . - nop */
		1369
		1370	if (beg [1] == '.' && len == 2)
		1371	{
		1372	/* .. - back up one component, if possible */
		1373
		1374	while (res != req->ptr2)
		1375	if (*--res == '/')
		1376	break;
		1377
		1378	continue;
		1379	}
		1380	}
		1381
		1382	errno = ENAMETOOLONG;
		1383	if (res + 1 + len + 1 >= tmp1)
		1384	return;
		1385
		1386	/* copy one component */
		1387	*res = '/';
		1388	memcpy (res + 1, beg, len);
		1389
		1390	/* zero-terminate, for readlink */
		1391	res [len + 1] = 0;
		1392
		1393	/* now check if it's a symlink */
		1394	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1395
		1396	if (linklen < 0)
		1397	{
		1398	if (errno != EINVAL)
		1399	return;
		1400
		1401	/* it's a normal directory. hopefully */
		1402	res += len + 1;
		1403	}
		1404	else
		1405	{
		1406	/* yay, it was a symlink - build new path in tmp2 */
		1407	int rellen = strlen (rel);
		1408
		1409	errno = ENAMETOOLONG;
		1410	if (linklen + 1 + rellen >= PATH_MAX)
		1411	return;
		1412
		1413	errno = ELOOP;
		1414	if (!--symlinks)
		1415	return;
		1416
		1417	if (*tmp1 == '/')
		1418	res = req->ptr2; /* symlink resolves to an absolute path */
		1419
		1420	/* we need to be careful, as rel might point into tmp2 already */
		1421	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1422	tmp2 [linklen] = '/';
		1423	memcpy (tmp2, tmp1, linklen);
		1424
		1425	rel = tmp2;
		1426	}
		1427	}
		1428
		1429	/* special case for the lone root path */
		1430	if (res == req->ptr2)
		1431	*res++ = '/';
		1432
		1433	req->result = res - (char *)req->ptr2;
		1434
		1435	done:
		1436	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1069	}	1437	}
1070		1438
1071	static signed char	1439	static signed char
1072	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1440	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1073	{	1441	{
…		…
1081		1449
1082	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1450	#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 */	1451	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1084		1452
1085	static void	1453	static void
1086	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1454	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1087	{	1455	{
1088	unsigned char bits [9 + sizeof (ino_t) * 8];	1456	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1089	unsigned char *bit = bits;	1457	unsigned char *bit = bits;
1090		1458
1091	assert (CHAR_BIT == 8);	1459	assert (CHAR_BIT == 8);
1092	assert (sizeof (eio_dirent) * 8 < 256);	1460	assert (sizeof (eio_dirent) * 8 < 256);
1093	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1461	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1095		1463
1096	if (size <= EIO_SORT_FAST)	1464	if (size <= EIO_SORT_FAST)
1097	return;	1465	return;
1098		1466
1099	/* first prepare an array of bits to test in our radix sort */	1467	/* 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 */	1468	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1101	/* inode_bits must contain all inodes ORed together */	1469	/* inode_bits must contain all inodes ORed together */
1102	/* which is used to skip bits that are 0 everywhere, which is very common */	1470	/* which is used to skip bits that are 0 everywhere, which is very common */
1103	{	1471	{
1104	ino_t endianness;	1472	eio_ino_t endianness;
1105	int i, j;	1473	int i, j;
1106		1474
1107	/* we store the byte offset of byte n into byte n of "endianness" */	1475	/* we store the byte offset of byte n into byte n of "endianness" */
1108	for (i = 0; i < sizeof (ino_t); ++i)	1476	for (i = 0; i < sizeof (eio_ino_t); ++i)
1109	((unsigned char *)&endianness)[i] = i;	1477	((unsigned char *)&endianness)[i] = i;
1110		1478
1111	*bit++ = 0;	1479	*bit++ = 0;
1112		1480
1113	for (i = 0; i < sizeof (ino_t); ++i)	1481	for (i = 0; i < sizeof (eio_ino_t); ++i)
1114	{	1482	{
1115	/* shifting off the byte offsets out of "endianness" */	1483	/* shifting off the byte offsets out of "endianness" */
1116	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1484	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1117	endianness >>= 8;	1485	endianness >>= 8;
1118		1486
1119	for (j = 0; j < 8; ++j)	1487	for (j = 0; j < 8; ++j)
1120	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1488	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1121	*bit++ = offs + j;	1489	*bit++ = offs + j;
1122	}	1490	}
1123		1491
1124	for (j = 0; j < 8; ++j)	1492	for (j = 0; j < 8; ++j)
1125	if (score_bits & (1 << j))	1493	if (score_bits & (1 << j))
1126	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1494	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1127	}	1495	}
1128		1496
1129	/* now actually do the sorting (a variant of MSD radix sort) */	1497	/* now actually do the sorting (a variant of MSD radix sort) */
1130	{	1498	{
1131	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1499	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1132	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1500	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1133	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1501	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1134	int stk_idx = 0;	1502	int stk_idx = 0;
1135		1503
1136	base_stk [stk_idx] = dents;	1504	base_stk [stk_idx] = dents;
1137	end_stk [stk_idx] = dents + size;	1505	end_stk [stk_idx] = dents + size;
1138	bit_stk [stk_idx] = bit - 1;	1506	bit_stk [stk_idx] = bit - 1;
…		…
1217	}	1585	}
1218	}	1586	}
1219	}	1587	}
1220		1588
1221	static void	1589	static void
1222	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1590	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1223	{	1591	{
1224	if (size <= 1)	1592	if (size <= 1)
1225	return; /* our insertion sort relies on size > 0 */	1593	return; /* our insertion sort relies on size > 0 */
1226		1594
1227	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1595	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1235		1603
1236	/* read a full directory */	1604	/* read a full directory */
1237	static void	1605	static void
1238	eio__scandir (eio_req *req, etp_worker *self)	1606	eio__scandir (eio_req *req, etp_worker *self)
1239	{	1607	{
1240	DIR *dirp;
1241	EIO_STRUCT_DIRENT *entp;
1242	char *name, *names;	1608	char *name, *names;
1243	int namesalloc = 4096;	1609	int namesalloc = 4096 - sizeof (void *) * 4;
1244	int namesoffs = 0;	1610	int namesoffs = 0;
1245	int flags = req->int1;	1611	int flags = req->int1;
1246	eio_dirent *dents = 0;	1612	eio_dirent *dents = 0;
1247	int dentalloc = 128;	1613	int dentalloc = 128;
1248	int dentoffs = 0;	1614	int dentoffs = 0;
1249	ino_t inode_bits = 0;	1615	eio_ino_t inode_bits = 0;
		1616	#ifdef _WIN32
		1617	HANDLE dirp;
		1618	WIN32_FIND_DATA entp;
		1619	#else
		1620	DIR *dirp;
		1621	EIO_STRUCT_DIRENT *entp;
		1622	#endif
1250		1623
1251	req->result = -1;	1624	req->result = -1;
1252		1625
1253	if (!(flags & EIO_READDIR_DENTS))	1626	if (!(flags & EIO_READDIR_DENTS))
1254	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1627	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1255		1628
1256	X_LOCK (wrklock);	1629	#ifdef _WIN32
1257	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1630	{
		1631	int len = strlen ((const char *)req->ptr1);
		1632	char *path = malloc (MAX_PATH);
		1633	const char *fmt;
		1634
		1635	if (!len)
		1636	fmt = "./*";
		1637	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1638	fmt = "%s*";
		1639	else
		1640	fmt = "%s/*";
		1641
		1642	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1643	dirp = FindFirstFile (path, &entp);
		1644	free (path);
		1645
		1646	if (dirp == INVALID_HANDLE_VALUE)
		1647	{
		1648	dirp = 0;
		1649
		1650	switch (GetLastError ())
		1651	{
		1652	case ERROR_FILE_NOT_FOUND:
		1653	req->result = 0;
		1654	break;
		1655
		1656	case ERROR_INVALID_NAME:
		1657	case ERROR_PATH_NOT_FOUND:
		1658	case ERROR_NO_MORE_FILES:
		1659	errno = ENOENT;
		1660	break;
		1661
		1662	case ERROR_NOT_ENOUGH_MEMORY:
		1663	errno = ENOMEM;
		1664	break;
		1665
		1666	default:
		1667	errno = EINVAL;
		1668	break;
		1669	}
		1670	}
		1671	}
		1672	#else
1258	self->dirp = dirp = opendir (req->ptr1);	1673	dirp = opendir (req->ptr1);
		1674	#endif
		1675
		1676	if (req->flags & EIO_FLAG_PTR1_FREE)
		1677	free (req->ptr1);
1259		1678
1260	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1679	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1261	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1680	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1262	req->ptr2 = names = malloc (namesalloc);	1681	req->ptr2 = names = malloc (namesalloc);
1263	X_UNLOCK (wrklock);
1264		1682
1265	if (dirp && names && (!flags || dents))	1683	if (dirp && names && (!flags || dents))
1266	for (;;)	1684	for (;;)
1267	{	1685	{
		1686	int done;
		1687
		1688	#ifdef _WIN32
		1689	done = !dirp;
		1690	#else
1268	errno = 0;	1691	errno = 0;
1269	entp = readdir (dirp);	1692	entp = readdir (dirp);
		1693	done = !entp;
		1694	#endif
1270		1695
1271	if (!entp)	1696	if (done)
1272	{	1697	{
		1698	#ifndef _WIN32
		1699	int old_errno = errno;
		1700	closedir (dirp);
		1701	errno = old_errno;
		1702
1273	if (errno)	1703	if (errno)
1274	break;	1704	break;
		1705	#endif
1275		1706
1276	/* sort etc. */	1707	/* sort etc. */
1277	req->int1 = flags;	1708	req->int1 = flags;
1278	req->result = dentoffs;	1709	req->result = dentoffs;
1279		1710
…		…
1308		1739
1309	break;	1740	break;
1310	}	1741	}
1311		1742
1312	/* now add the entry to our list(s) */	1743	/* now add the entry to our list(s) */
1313	name = entp->d_name;	1744	name = D_NAME (entp);
1314		1745
1315	/* skip . and .. entries */	1746	/* skip . and .. entries */
1316	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1747	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1317	{	1748	{
1318	int len = D_NAMLEN (entp) + 1;	1749	int len = D_NAMLEN (entp) + 1;
1319		1750
1320	while (expect_false (namesoffs + len > namesalloc))	1751	while (ecb_expect_false (namesoffs + len > namesalloc))
1321	{	1752	{
1322	namesalloc *= 2;	1753	namesalloc *= 2;
1323	X_LOCK (wrklock);
1324	req->ptr2 = names = realloc (names, namesalloc);	1754	req->ptr2 = names = realloc (names, namesalloc);
1325	X_UNLOCK (wrklock);
1326		1755
1327	if (!names)	1756	if (!names)
1328	break;	1757	break;
1329	}	1758	}
1330		1759
…		…
1332		1761
1333	if (dents)	1762	if (dents)
1334	{	1763	{
1335	struct eio_dirent *ent;	1764	struct eio_dirent *ent;
1336		1765
1337	if (expect_false (dentoffs == dentalloc))	1766	if (ecb_expect_false (dentoffs == dentalloc))
1338	{	1767	{
1339	dentalloc *= 2;	1768	dentalloc *= 2;
1340	X_LOCK (wrklock);
1341	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1769	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1342	X_UNLOCK (wrklock);
1343		1770
1344	if (!dents)	1771	if (!dents)
1345	break;	1772	break;
1346	}	1773	}
1347		1774
…		…
1427	if (EIO_CANCELLED (req))	1854	if (EIO_CANCELLED (req))
1428	{	1855	{
1429	errno = ECANCELED;	1856	errno = ECANCELED;
1430	break;	1857	break;
1431	}	1858	}
		1859
		1860	#ifdef _WIN32
		1861	if (!FindNextFile (dirp, &entp))
		1862	{
		1863	FindClose (dirp);
		1864	dirp = 0;
		1865	}
		1866	#endif
1432	}	1867	}
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	}	1868	}
1546		1869
1547	/*****************************************************************************/	1870	/*****************************************************************************/
1548		1871
1549	#define ALLOC(len) \	1872	#define ALLOC(len) \
…		…
1565	{	1888	{
1566	ETP_REQ *req;	1889	ETP_REQ *req;
1567	struct timespec ts;	1890	struct timespec ts;
1568	etp_worker *self = (etp_worker *)thr_arg;	1891	etp_worker *self = (etp_worker *)thr_arg;
1569		1892
1570	/* try to distribute timeouts somewhat randomly */	1893	/* try to distribute timeouts somewhat evenly */
1571	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1894	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1572		1895
1573	for (;;)	1896	for (;;)
1574	{	1897	{
		1898	ts.tv_sec = 0;
		1899
1575	X_LOCK (reqlock);	1900	X_LOCK (reqlock);
1576		1901
1577	for (;;)	1902	for (;;)
1578	{	1903	{
1579	self->req = req = reqq_shift (&req_queue);	1904	self->req = req = reqq_shift (&req_queue);
1580		1905
1581	if (req)	1906	if (req)
1582	break;	1907	break;
1583		1908
		1909	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1910	{
		1911	X_UNLOCK (reqlock);
		1912	X_LOCK (wrklock);
		1913	--started;
		1914	X_UNLOCK (wrklock);
		1915	goto quit;
		1916	}
		1917
1584	++idle;	1918	++idle;
1585		1919
1586	ts.tv_sec = time (0) + idle_timeout;	1920	if (idle <= max_idle)
1587	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1921	/* we are allowed to idle, so do so without any timeout */
		1922	X_COND_WAIT (reqwait, reqlock);
		1923	else
1588	{	1924	{
1589	if (idle > max_idle)	1925	/* initialise timeout once */
1590	{	1926	if (!ts.tv_sec)
1591	--idle;	1927	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		1928
1599	/* we are allowed to idle, so do so without any timeout */
1600	X_COND_WAIT (reqwait, reqlock);	1929	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1930	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1601	}	1931	}
1602		1932
1603	--idle;	1933	--idle;
1604	}	1934	}
1605		1935
…		…
1608	X_UNLOCK (reqlock);	1938	X_UNLOCK (reqlock);
1609		1939
1610	if (req->type < 0)	1940	if (req->type < 0)
1611	goto quit;	1941	goto quit;
1612		1942
1613	if (!EIO_CANCELLED (req))
1614	ETP_EXECUTE (self, req);	1943	ETP_EXECUTE (self, req);
1615		1944
1616	X_LOCK (reslock);	1945	X_LOCK (reslock);
1617		1946
1618	++npending;	1947	++npending;
1619		1948
…		…
1634	return 0;	1963	return 0;
1635	}	1964	}
1636		1965
1637	/*****************************************************************************/	1966	/*****************************************************************************/
1638		1967
		1968	int ecb_cold
1639	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1969	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1640	{	1970	{
		1971	#if !HAVE_PREADWRITE
		1972	X_MUTEX_CREATE (preadwritelock);
		1973	#endif
		1974
1641	return etp_init (want_poll, done_poll);	1975	return etp_init (want_poll, done_poll);
1642	}	1976	}
1643		1977
		1978	ecb_inline void
1644	static void eio_api_destroy (eio_req *req)	1979	eio_api_destroy (eio_req *req)
1645	{	1980	{
1646	free (req);	1981	free (req);
1647	}	1982	}
1648		1983
1649	#define REQ(rtype) \	1984	#define REQ(rtype) \
…		…
1668	{ \	2003	{ \
1669	eio_api_destroy (req); \	2004	eio_api_destroy (req); \
1670	return 0; \	2005	return 0; \
1671	}	2006	}
1672		2007
		2008	static void
1673	static void eio_execute (etp_worker *self, eio_req *req)	2009	eio_execute (etp_worker *self, eio_req *req)
1674	{	2010	{
		2011	if (ecb_expect_false (EIO_CANCELLED (req)))
		2012	{
		2013	req->result = -1;
		2014	req->errorno = ECANCELED;
		2015	return;
		2016	}
		2017
1675	switch (req->type)	2018	switch (req->type)
1676	{	2019	{
1677	case EIO_READ: ALLOC (req->size);	2020	case EIO_READ: ALLOC (req->size);
1678	req->result = req->offs >= 0	2021	req->result = req->offs >= 0
1679	? pread (req->int1, req->ptr2, req->size, req->offs)	2022	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1681	case EIO_WRITE: req->result = req->offs >= 0	2024	case EIO_WRITE: req->result = req->offs >= 0
1682	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2025	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1683	: write (req->int1, req->ptr2, req->size); break;	2026	: write (req->int1, req->ptr2, req->size); break;
1684		2027
1685	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2028	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;	2029	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1687		2030
1688	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2031	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1689	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2032	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1690	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2033	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1691	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2034	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
…		…
1713	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2056	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1714	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2057	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1715	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2058	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;	2059	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1717		2060
		2061	case EIO_REALPATH: eio__realpath (req, self); break;
		2062
1718	case EIO_READLINK: ALLOC (PATH_MAX);	2063	case EIO_READLINK: ALLOC (PATH_MAX);
1719	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;	2064	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1720		2065
1721	case EIO_SYNC: req->result = 0; sync (); break;	2066	case EIO_SYNC: req->result = 0; sync (); break;
1722	case EIO_FSYNC: req->result = fsync (req->int1); break;	2067	case EIO_FSYNC: req->result = fsync (req->int1); break;
1723	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2068	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1724	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2069	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;	2070	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1726	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2071	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1727	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2072	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;	2073	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2074	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1729		2075
1730	case EIO_READDIR: eio__scandir (req, self); break;	2076	case EIO_READDIR: eio__scandir (req, self); break;
1731		2077
1732	case EIO_BUSY:	2078	case EIO_BUSY:
1733	#ifdef _WIN32	2079	#ifdef _WIN32
…		…
1774	case EIO_NOP:	2120	case EIO_NOP:
1775	req->result = 0;	2121	req->result = 0;
1776	break;	2122	break;
1777		2123
1778	case EIO_CUSTOM:	2124	case EIO_CUSTOM:
1779	((void (*)(eio_req *))req->feed) (req);	2125	req->feed (req);
1780	break;	2126	break;
1781		2127
1782	default:	2128	default:
1783	errno = ENOSYS;	2129	errno = ENOSYS;
1784	req->result = -1;	2130	req->result = -1;
…		…
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)	2179	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1834	{	2180	{
1835	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2181	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1836	}	2182	}
1837		2183
		2184	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2185	{
		2186	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2187	}
		2188
1838	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2189	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1839	{	2190	{
1840	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2191	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1841	}	2192	}
1842		2193
…		…
1883	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2234	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1884	{	2235	{
1885	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2236	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1886	}	2237	}
1887		2238
1888	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2239	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1889	{	2240	{
1890	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2241	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1891	}	2242	}
1892		2243
1893	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2244	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)	2264	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1914	{	2265	{
1915	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2266	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1916	}	2267	}
1917		2268
1918	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2269	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1919	{	2270	{
1920	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2271	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1921	}	2272	}
1922		2273
1923	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2274	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1937	}	2288	}
1938		2289
1939	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2290	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1940	{	2291	{
1941	return eio__1path (EIO_READLINK, path, pri, cb, data);	2292	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2293	}
		2294
		2295	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2296	{
		2297	return eio__1path (EIO_REALPATH, path, pri, cb, data);
1942	}	2298	}
1943		2299
1944	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2300	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1945	{	2301	{
1946	return eio__1path (EIO_STAT, path, pri, cb, data);	2302	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)	2361	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2006	{	2362	{
2007	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2363	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2008	}	2364	}
2009		2365
2010	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2366	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2011	{	2367	{
2012	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2368	REQ (EIO_CUSTOM); req->feed = execute; SEND;
2013	}	2369	}
2014		2370
2015	#endif	2371	#endif
2016		2372
2017	eio_req *eio_grp (eio_cb cb, void *data)	2373	eio_req *eio_grp (eio_cb cb, void *data)
…		…
2026	#undef SEND	2382	#undef SEND
2027		2383
2028	/*****************************************************************************/	2384	/*****************************************************************************/
2029	/* grp functions */	2385	/* grp functions */
2030		2386
		2387	void
2031	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2388	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2032	{	2389	{
2033	grp->int2 = limit;	2390	grp->int2 = limit;
2034	grp->feed = feed;	2391	grp->feed = feed;
2035		2392
2036	grp_try_feed (grp);	2393	grp_try_feed (grp);
2037	}	2394	}
2038		2395
		2396	void
2039	void eio_grp_limit (eio_req *grp, int limit)	2397	eio_grp_limit (eio_req *grp, int limit)
2040	{	2398	{
2041	grp->int2 = limit;	2399	grp->int2 = limit;
2042		2400
2043	grp_try_feed (grp);	2401	grp_try_feed (grp);
2044	}	2402	}
2045		2403
		2404	void
2046	void eio_grp_add (eio_req *grp, eio_req *req)	2405	eio_grp_add (eio_req *grp, eio_req *req)
2047	{	2406	{
2048	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2407	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2049		2408
2050	grp->flags |= EIO_FLAG_GROUPADD;	2409	grp->flags |= EIO_FLAG_GROUPADD;
2051		2410
…		…
2062	}	2421	}
2063		2422
2064	/*****************************************************************************/	2423	/*****************************************************************************/
2065	/* misc garbage */	2424	/* misc garbage */
2066		2425
		2426	eio_ssize_t
2067	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2427	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2068	{	2428	{
2069	etp_worker wrk;
2070	ssize_t ret;
2071
2072	wrk.dbuf = 0;
2073
2074	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2429	return eio__sendfile (ofd, ifd, offset, count);
2075
2076	if (wrk.dbuf)
2077	free (wrk.dbuf);
2078
2079	return ret;
2080	}	2430	}
2081		2431

Diff Legend

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