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 1.96 by root, Tue Jul 19 05:20:36 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>
…		…
138	#endif	223	#endif
139	#ifndef D_INO	224	#ifndef D_INO
140	# define D_INO(de) 0	225	# define D_INO(de) 0
141	#endif	226	#endif
142	#ifndef D_NAMLEN	227	#ifndef D_NAMLEN
143	# define D_NAMLEN(de) strlen ((de)->d_name)	228	# define D_NAMLEN(entp) strlen (D_NAME (entp))
144	#endif	229	#endif
145		230
146	/* used for struct dirent, AIX doesn't provide it */	231	/* used for struct dirent, AIX doesn't provide it */
147	#ifndef NAME_MAX	232	#ifndef NAME_MAX
148	# define NAME_MAX 4096	233	# define NAME_MAX 4096
…		…
155		240
156	/* buffer size for various temporary buffers */	241	/* buffer size for various temporary buffers */
157	#define EIO_BUFSIZE 65536	242	#define EIO_BUFSIZE 65536
158		243
159	#define dBUF \	244	#define dBUF \
160	char *eio_buf; \
161	ETP_WORKER_LOCK (self); \
162	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	245	char *eio_buf = malloc (EIO_BUFSIZE); \
163	ETP_WORKER_UNLOCK (self); \
164	errno = ENOMEM; \	246	errno = ENOMEM; \
165	if (!eio_buf) \	247	if (!eio_buf) \
166	return -1;	248	return -1
		249
		250	#define FUBd \
		251	free (eio_buf)
167		252
168	#define EIO_TICKS ((1000000 + 1023) >> 10)	253	#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		254
183	#define ETP_PRI_MIN EIO_PRI_MIN	255	#define ETP_PRI_MIN EIO_PRI_MIN
184	#define ETP_PRI_MAX EIO_PRI_MAX	256	#define ETP_PRI_MAX EIO_PRI_MAX
185		257
186	struct etp_worker;	258	struct etp_worker;
…		…
190	static int eio_finish (eio_req *req);	262	static int eio_finish (eio_req *req);
191	#define ETP_FINISH(req) eio_finish (req)	263	#define ETP_FINISH(req) eio_finish (req)
192	static void eio_execute (struct etp_worker *self, eio_req *req);	264	static void eio_execute (struct etp_worker *self, eio_req *req);
193	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	265	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194		266
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	/*****************************************************************************/	267	/*****************************************************************************/
213		268
214	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	269	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215		270
216	/* calculate time difference in ~1/EIO_TICKS of a second */	271	/* calculate time difference in ~1/EIO_TICKS of a second */
		272	ecb_inline int
217	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	273	tvdiff (struct timeval *tv1, struct timeval *tv2)
218	{	274	{
219	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	275	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	276	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
221	}	277	}
222		278
…		…
243	/*	299	/*
244	* make our pread/pwrite emulation safe against themselves, but not against	300	* make our pread/pwrite emulation safe against themselves, but not against
245	* normal read/write by using a mutex. slows down execution a lot,	301	* normal read/write by using a mutex. slows down execution a lot,
246	* but that's your problem, not mine.	302	* but that's your problem, not mine.
247	*/	303	*/
248	static xmutex_t preadwritelock = X_MUTEX_INIT;	304	static xmutex_t preadwritelock;
249	#endif	305	#endif
250		306
251	typedef struct etp_worker	307	typedef struct etp_worker
252	{	308	{
253	/* locked by wrklock */	309	/* locked by wrklock */
…		…
256	xthread_t tid;	312	xthread_t tid;
257		313
258	/* locked by reslock, reqlock or wrklock */	314	/* locked by reslock, reqlock or wrklock */
259	ETP_REQ *req; /* currently processed request */	315	ETP_REQ *req; /* currently processed request */
260		316
		317	#ifdef ETP_WORKER_COMMON
261	ETP_WORKER_COMMON	318	ETP_WORKER_COMMON
		319	#endif
262	} etp_worker;	320	} etp_worker;
263		321
264	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	322	static etp_worker wrk_first; /* NOT etp */
265		323
266	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	324	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	325	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268		326
269	/* worker threads management */	327	/* worker threads management */
270		328
		329	static void ecb_cold
271	static void etp_worker_clear (etp_worker *wrk)	330	etp_worker_clear (etp_worker *wrk)
272	{	331	{
273	ETP_WORKER_CLEAR (wrk);
274	}	332	}
275		333
		334	static void ecb_cold
276	static void etp_worker_free (etp_worker *wrk)	335	etp_worker_free (etp_worker *wrk)
277	{	336	{
278	wrk->next->prev = wrk->prev;	337	wrk->next->prev = wrk->prev;
279	wrk->prev->next = wrk->next;	338	wrk->prev->next = wrk->next;
280		339
281	free (wrk);	340	free (wrk);
282	}	341	}
283		342
284	static unsigned int etp_nreqs (void)	343	static unsigned int
		344	etp_nreqs (void)
285	{	345	{
286	int retval;	346	int retval;
287	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	347	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288	retval = nreqs;	348	retval = nreqs;
289	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	349	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290	return retval;	350	return retval;
291	}	351	}
292		352
293	static unsigned int etp_nready (void)	353	static unsigned int
		354	etp_nready (void)
294	{	355	{
295	unsigned int retval;	356	unsigned int retval;
296		357
297	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	358	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298	retval = nready;	359	retval = nready;
299	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	360	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300		361
301	return retval;	362	return retval;
302	}	363	}
303		364
304	static unsigned int etp_npending (void)	365	static unsigned int
		366	etp_npending (void)
305	{	367	{
306	unsigned int retval;	368	unsigned int retval;
307		369
308	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	370	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309	retval = npending;	371	retval = npending;
310	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	372	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311		373
312	return retval;	374	return retval;
313	}	375	}
314		376
315	static unsigned int etp_nthreads (void)	377	static unsigned int
		378	etp_nthreads (void)
316	{	379	{
317	unsigned int retval;	380	unsigned int retval;
318		381
319	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	382	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320	retval = started;	383	retval = started;
…		…
334	} etp_reqq;	397	} etp_reqq;
335		398
336	static etp_reqq req_queue;	399	static etp_reqq req_queue;
337	static etp_reqq res_queue;	400	static etp_reqq res_queue;
338		401
		402	static void ecb_noinline ecb_cold
		403	reqq_init (etp_reqq *q)
		404	{
		405	int pri;
		406
		407	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		408	q->qs[pri] = q->qe[pri] = 0;
		409
		410	q->size = 0;
		411	}
		412
		413	static int ecb_noinline
339	static int reqq_push (etp_reqq *q, ETP_REQ *req)	414	reqq_push (etp_reqq *q, ETP_REQ *req)
340	{	415	{
341	int pri = req->pri;	416	int pri = req->pri;
342	req->next = 0;	417	req->next = 0;
343		418
344	if (q->qe[pri])	419	if (q->qe[pri])
…		…
350	q->qe[pri] = q->qs[pri] = req;	425	q->qe[pri] = q->qs[pri] = req;
351		426
352	return q->size++;	427	return q->size++;
353	}	428	}
354		429
		430	static ETP_REQ * ecb_noinline
355	static ETP_REQ *reqq_shift (etp_reqq *q)	431	reqq_shift (etp_reqq *q)
356	{	432	{
357	int pri;	433	int pri;
358		434
359	if (!q->size)	435	if (!q->size)
360	return 0;	436	return 0;
…		…
375	}	451	}
376		452
377	abort ();	453	abort ();
378	}	454	}
379		455
380	static void etp_thread_init (void)	456	static int ecb_cold
		457	etp_init (void (*want_poll)(void), void (*done_poll)(void))
381	{	458	{
382	X_MUTEX_CREATE (wrklock);	459	X_MUTEX_CREATE (wrklock);
383	X_MUTEX_CREATE (reslock);	460	X_MUTEX_CREATE (reslock);
384	X_MUTEX_CREATE (reqlock);	461	X_MUTEX_CREATE (reqlock);
385	X_COND_CREATE (reqwait);	462	X_COND_CREATE (reqwait);
386	}
387		463
388	static void etp_atfork_prepare (void)	464	reqq_init (&req_queue);
389	{	465	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		466
398	static void etp_atfork_parent (void)	467	wrk_first.next =
399	{	468	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		469
429	started = 0;	470	started = 0;
430	idle = 0;	471	idle = 0;
431	nreqs = 0;	472	nreqs = 0;
432	nready = 0;	473	nready = 0;
433	npending = 0;	474	npending = 0;
434		475
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;	476	want_poll_cb = want_poll;
453	done_poll_cb = done_poll;	477	done_poll_cb = done_poll;
454		478
455	return 0;	479	return 0;
456	}	480	}
457		481
458	X_THREAD_PROC (etp_proc);	482	X_THREAD_PROC (etp_proc);
459		483
		484	static void ecb_cold
460	static void etp_start_thread (void)	485	etp_start_thread (void)
461	{	486	{
462	etp_worker *wrk = calloc (1, sizeof (etp_worker));	487	etp_worker *wrk = calloc (1, sizeof (etp_worker));
463		488
464	/*TODO*/	489	/*TODO*/
465	assert (("unable to allocate worker thread data", wrk));	490	assert (("unable to allocate worker thread data", wrk));
…		…
478	free (wrk);	503	free (wrk);
479		504
480	X_UNLOCK (wrklock);	505	X_UNLOCK (wrklock);
481	}	506	}
482		507
		508	static void
483	static void etp_maybe_start_thread (void)	509	etp_maybe_start_thread (void)
484	{	510	{
485	if (expect_true (etp_nthreads () >= wanted))	511	if (ecb_expect_true (etp_nthreads () >= wanted))
486	return;	512	return;
487		513
488	/* todo: maybe use idle here, but might be less exact */	514	/* todo: maybe use idle here, but might be less exact */
489	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	515	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
490	return;	516	return;
491		517
492	etp_start_thread ();	518	etp_start_thread ();
493	}	519	}
494		520
		521	static void ecb_cold
495	static void etp_end_thread (void)	522	etp_end_thread (void)
496	{	523	{
497	eio_req *req = calloc (1, sizeof (eio_req));	524	eio_req *req = calloc (1, sizeof (eio_req));
498		525
499	req->type = -1;	526	req->type = -1;
500	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	527	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
507	X_LOCK (wrklock);	534	X_LOCK (wrklock);
508	--started;	535	--started;
509	X_UNLOCK (wrklock);	536	X_UNLOCK (wrklock);
510	}	537	}
511		538
512	static int etp_poll (void)	539	static int
		540	etp_poll (void)
513	{	541	{
514	unsigned int maxreqs;	542	unsigned int maxreqs;
515	unsigned int maxtime;	543	unsigned int maxtime;
516	struct timeval tv_start, tv_now;	544	struct timeval tv_start, tv_now;
517		545
…		…
547		575
548	X_LOCK (reqlock);	576	X_LOCK (reqlock);
549	--nreqs;	577	--nreqs;
550	X_UNLOCK (reqlock);	578	X_UNLOCK (reqlock);
551		579
552	if (expect_false (req->type == EIO_GROUP && req->size))	580	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
553	{	581	{
554	req->int1 = 1; /* mark request as delayed */	582	req->int1 = 1; /* mark request as delayed */
555	continue;	583	continue;
556	}	584	}
557	else	585	else
558	{	586	{
559	int res = ETP_FINISH (req);	587	int res = ETP_FINISH (req);
560	if (expect_false (res))	588	if (ecb_expect_false (res))
561	return res;	589	return res;
562	}	590	}
563		591
564	if (expect_false (maxreqs && !--maxreqs))	592	if (ecb_expect_false (maxreqs && !--maxreqs))
565	break;	593	break;
566		594
567	if (maxtime)	595	if (maxtime)
568	{	596	{
569	gettimeofday (&tv_now, 0);	597	gettimeofday (&tv_now, 0);
…		…
575		603
576	errno = EAGAIN;	604	errno = EAGAIN;
577	return -1;	605	return -1;
578	}	606	}
579		607
		608	static void
580	static void etp_cancel (ETP_REQ *req)	609	etp_cancel (ETP_REQ *req)
581	{	610	{
582	X_LOCK (wrklock);	611	req->cancelled = 1;
583	req->flags |= EIO_FLAG_CANCELLED;
584	X_UNLOCK (wrklock);
585		612
586	eio_grp_cancel (req);	613	eio_grp_cancel (req);
587	}	614	}
588		615
		616	static void
589	static void etp_submit (ETP_REQ *req)	617	etp_submit (ETP_REQ *req)
590	{	618	{
591	req->pri -= ETP_PRI_MIN;	619	req->pri -= ETP_PRI_MIN;
592		620
593	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	621	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
594	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	622	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595		623
596	if (expect_false (req->type == EIO_GROUP))	624	if (ecb_expect_false (req->type == EIO_GROUP))
597	{	625	{
598	/* I hope this is worth it :/ */	626	/* I hope this is worth it :/ */
599	X_LOCK (reqlock);	627	X_LOCK (reqlock);
600	++nreqs;	628	++nreqs;
601	X_UNLOCK (reqlock);	629	X_UNLOCK (reqlock);
…		…
620		648
621	etp_maybe_start_thread ();	649	etp_maybe_start_thread ();
622	}	650	}
623	}	651	}
624		652
		653	static void ecb_cold
625	static void etp_set_max_poll_time (double nseconds)	654	etp_set_max_poll_time (double nseconds)
626	{	655	{
627	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	656	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628	max_poll_time = nseconds * EIO_TICKS;	657	max_poll_time = nseconds * EIO_TICKS;
629	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	658	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630	}	659	}
631		660
		661	static void ecb_cold
632	static void etp_set_max_poll_reqs (unsigned int maxreqs)	662	etp_set_max_poll_reqs (unsigned int maxreqs)
633	{	663	{
634	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	664	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635	max_poll_reqs = maxreqs;	665	max_poll_reqs = maxreqs;
636	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	666	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637	}	667	}
638		668
		669	static void ecb_cold
639	static void etp_set_max_idle (unsigned int nthreads)	670	etp_set_max_idle (unsigned int nthreads)
640	{	671	{
641	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	672	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642	max_idle = nthreads;	673	max_idle = nthreads;
643	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	674	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644	}	675	}
645		676
		677	static void ecb_cold
646	static void etp_set_idle_timeout (unsigned int seconds)	678	etp_set_idle_timeout (unsigned int seconds)
647	{	679	{
648	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	680	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649	idle_timeout = seconds;	681	idle_timeout = seconds;
650	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	682	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651	}	683	}
652		684
		685	static void ecb_cold
653	static void etp_set_min_parallel (unsigned int nthreads)	686	etp_set_min_parallel (unsigned int nthreads)
654	{	687	{
655	if (wanted < nthreads)	688	if (wanted < nthreads)
656	wanted = nthreads;	689	wanted = nthreads;
657	}	690	}
658		691
		692	static void ecb_cold
659	static void etp_set_max_parallel (unsigned int nthreads)	693	etp_set_max_parallel (unsigned int nthreads)
660	{	694	{
661	if (wanted > nthreads)	695	if (wanted > nthreads)
662	wanted = nthreads;	696	wanted = nthreads;
663		697
664	while (started > wanted)	698	while (started > wanted)
665	etp_end_thread ();	699	etp_end_thread ();
666	}	700	}
667		701
668	/*****************************************************************************/	702	/*****************************************************************************/
669		703
		704	static void
670	static void grp_try_feed (eio_req *grp)	705	grp_try_feed (eio_req *grp)
671	{	706	{
672	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	707	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673	{	708	{
674	grp->flags &= ~EIO_FLAG_GROUPADD;	709	grp->flags &= ~EIO_FLAG_GROUPADD;
675		710
…		…
682	break;	717	break;
683	}	718	}
684	}	719	}
685	}	720	}
686		721
		722	static int
687	static int grp_dec (eio_req *grp)	723	grp_dec (eio_req *grp)
688	{	724	{
689	--grp->size;	725	--grp->size;
690		726
691	/* call feeder, if applicable */	727	/* call feeder, if applicable */
692	grp_try_feed (grp);	728	grp_try_feed (grp);
…		…
696	return eio_finish (grp);	732	return eio_finish (grp);
697	else	733	else
698	return 0;	734	return 0;
699	}	735	}
700		736
		737	static void
701	void eio_destroy (eio_req *req)	738	eio_destroy (eio_req *req)
702	{	739	{
703	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	740	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	741	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705		742
706	EIO_DESTROY (req);	743	EIO_DESTROY (req);
707	}	744	}
708		745
		746	static int
709	static int eio_finish (eio_req *req)	747	eio_finish (eio_req *req)
710	{	748	{
711	int res = EIO_FINISH (req);	749	int res = EIO_FINISH (req);
712		750
713	if (req->grp)	751	if (req->grp)
714	{	752	{
…		…
722	if (grp->grp_first == req)	760	if (grp->grp_first == req)
723	grp->grp_first = req->grp_next;	761	grp->grp_first = req->grp_next;
724		762
725	res2 = grp_dec (grp);	763	res2 = grp_dec (grp);
726		764
727	if (!res && res2)	765	if (!res)
728	res = res2;	766	res = res2;
729	}	767	}
730		768
731	eio_destroy (req);	769	eio_destroy (req);
732		770
733	return res;	771	return res;
734	}	772	}
735		773
		774	void
736	void eio_grp_cancel (eio_req *grp)	775	eio_grp_cancel (eio_req *grp)
737	{	776	{
738	for (grp = grp->grp_first; grp; grp = grp->grp_next)	777	for (grp = grp->grp_first; grp; grp = grp->grp_next)
739	eio_cancel (grp);	778	eio_cancel (grp);
740	}	779	}
741		780
		781	void
742	void eio_cancel (eio_req *req)	782	eio_cancel (eio_req *req)
743	{	783	{
744	etp_cancel (req);	784	etp_cancel (req);
745	}	785	}
746		786
		787	void
747	void eio_submit (eio_req *req)	788	eio_submit (eio_req *req)
748	{	789	{
749	etp_submit (req);	790	etp_submit (req);
750	}	791	}
751		792
752	unsigned int eio_nreqs (void)	793	unsigned int
		794	eio_nreqs (void)
753	{	795	{
754	return etp_nreqs ();	796	return etp_nreqs ();
755	}	797	}
756		798
757	unsigned int eio_nready (void)	799	unsigned int
		800	eio_nready (void)
758	{	801	{
759	return etp_nready ();	802	return etp_nready ();
760	}	803	}
761		804
762	unsigned int eio_npending (void)	805	unsigned int
		806	eio_npending (void)
763	{	807	{
764	return etp_npending ();	808	return etp_npending ();
765	}	809	}
766		810
767	unsigned int eio_nthreads (void)	811	unsigned int ecb_cold
		812	eio_nthreads (void)
768	{	813	{
769	return etp_nthreads ();	814	return etp_nthreads ();
770	}	815	}
771		816
		817	void ecb_cold
772	void eio_set_max_poll_time (double nseconds)	818	eio_set_max_poll_time (double nseconds)
773	{	819	{
774	etp_set_max_poll_time (nseconds);	820	etp_set_max_poll_time (nseconds);
775	}	821	}
776		822
		823	void ecb_cold
777	void eio_set_max_poll_reqs (unsigned int maxreqs)	824	eio_set_max_poll_reqs (unsigned int maxreqs)
778	{	825	{
779	etp_set_max_poll_reqs (maxreqs);	826	etp_set_max_poll_reqs (maxreqs);
780	}	827	}
781		828
		829	void ecb_cold
782	void eio_set_max_idle (unsigned int nthreads)	830	eio_set_max_idle (unsigned int nthreads)
783	{	831	{
784	etp_set_max_idle (nthreads);	832	etp_set_max_idle (nthreads);
785	}	833	}
786		834
		835	void ecb_cold
787	void eio_set_idle_timeout (unsigned int seconds)	836	eio_set_idle_timeout (unsigned int seconds)
788	{	837	{
789	etp_set_idle_timeout (seconds);	838	etp_set_idle_timeout (seconds);
790	}	839	}
791		840
		841	void ecb_cold
792	void eio_set_min_parallel (unsigned int nthreads)	842	eio_set_min_parallel (unsigned int nthreads)
793	{	843	{
794	etp_set_min_parallel (nthreads);	844	etp_set_min_parallel (nthreads);
795	}	845	}
796		846
		847	void ecb_cold
797	void eio_set_max_parallel (unsigned int nthreads)	848	eio_set_max_parallel (unsigned int nthreads)
798	{	849	{
799	etp_set_max_parallel (nthreads);	850	etp_set_max_parallel (nthreads);
800	}	851	}
801		852
802	int eio_poll (void)	853	int eio_poll (void)
…		…
811	# undef pread	862	# undef pread
812	# undef pwrite	863	# undef pwrite
813	# define pread eio__pread	864	# define pread eio__pread
814	# define pwrite eio__pwrite	865	# define pwrite eio__pwrite
815		866
816	static ssize_t	867	static eio_ssize_t
817	eio__pread (int fd, void *buf, size_t count, off_t offset)	868	eio__pread (int fd, void *buf, size_t count, off_t offset)
818	{	869	{
819	ssize_t res;	870	eio_ssize_t res;
820	off_t ooffset;	871	off_t ooffset;
821		872
822	X_LOCK (preadwritelock);	873	X_LOCK (preadwritelock);
823	ooffset = lseek (fd, 0, SEEK_CUR);	874	ooffset = lseek (fd, 0, SEEK_CUR);
824	lseek (fd, offset, SEEK_SET);	875	lseek (fd, offset, SEEK_SET);
…		…
827	X_UNLOCK (preadwritelock);	878	X_UNLOCK (preadwritelock);
828		879
829	return res;	880	return res;
830	}	881	}
831		882
832	static ssize_t	883	static eio_ssize_t
833	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	884	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834	{	885	{
835	ssize_t res;	886	eio_ssize_t res;
836	off_t ooffset;	887	off_t ooffset;
837		888
838	X_LOCK (preadwritelock);	889	X_LOCK (preadwritelock);
839	ooffset = lseek (fd, 0, SEEK_CUR);	890	ooffset = lseek (fd, 0, SEEK_CUR);
840	lseek (fd, offset, SEEK_SET);	891	lseek (fd, offset, SEEK_SET);
…		…
872	#ifndef HAVE_FUTIMES	923	#ifndef HAVE_FUTIMES
873		924
874	# undef futimes	925	# undef futimes
875	# define futimes(fd,times) eio__futimes (fd, times)	926	# define futimes(fd,times) eio__futimes (fd, times)
876		927
		928	static int
877	static int eio__futimes (int fd, const struct timeval tv[2])	929	eio__futimes (int fd, const struct timeval tv[2])
878	{	930	{
879	errno = ENOSYS;	931	errno = ENOSYS;
880	return -1;	932	return -1;
881	}	933	}
882		934
…		…
886	# undef fdatasync	938	# undef fdatasync
887	# define fdatasync(fd) fsync (fd)	939	# define fdatasync(fd) fsync (fd)
888	#endif	940	#endif
889		941
890	/* sync_file_range always needs emulation */	942	/* sync_file_range always needs emulation */
891	int	943	static int
892	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	944	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893	{	945	{
894	#if HAVE_SYNC_FILE_RANGE	946	#if HAVE_SYNC_FILE_RANGE
895	int res;	947	int res;
896		948
…		…
913	/* even though we could play tricks with the flags, it's better to always	965	/* even though we could play tricks with the flags, it's better to always
914	* call fdatasync, as that matches the expectation of its users best */	966	* call fdatasync, as that matches the expectation of its users best */
915	return fdatasync (fd);	967	return fdatasync (fd);
916	}	968	}
917		969
		970	static int
		971	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		972	{
		973	#if HAVE_FALLOCATE
		974	return fallocate (fd, mode, offset, len);
		975	#else
		976	errno = ENOSYS;
		977	return -1;
		978	#endif
		979	}
		980
918	#if !HAVE_READAHEAD	981	#if !HAVE_READAHEAD
919	# undef readahead	982	# undef readahead
920	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	983	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921		984
922	static ssize_t	985	static eio_ssize_t
923	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	986	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
924	{	987	{
925	size_t todo = count;	988	size_t todo = count;
926	dBUF;	989	dBUF;
927		990
…		…
932	pread (fd, eio_buf, len, offset);	995	pread (fd, eio_buf, len, offset);
933	offset += len;	996	offset += len;
934	todo -= len;	997	todo -= len;
935	}	998	}
936		999
		1000	FUBd;
		1001
937	errno = 0;	1002	errno = 0;
938	return count;	1003	return count;
939	}	1004	}
940		1005
941	#endif	1006	#endif
942		1007
943	/* sendfile always needs emulation */	1008	/* sendfile always needs emulation */
944	static ssize_t	1009	static eio_ssize_t
945	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1010	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
946	{	1011	{
		1012	eio_ssize_t written = 0;
947	ssize_t res;	1013	eio_ssize_t res;
948		1014
949	if (!count)	1015	if (!count)
950	return 0;	1016	return 0;
951		1017
		1018	for (;;)
		1019	{
		1020	#ifdef __APPLE__
		1021	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1022	#endif
952	#if HAVE_SENDFILE	1023	#if HAVE_SENDFILE
953	# if __linux	1024	# if __linux
		1025	off_t soffset = offset;
954	res = sendfile (ofd, ifd, &offset, count);	1026	res = sendfile (ofd, ifd, &soffset, count);
955		1027
956	# elif __FreeBSD__	1028	# elif __FreeBSD__
957	/*	1029	/*
958	* Of course, the freebsd sendfile is a dire hack with no thoughts	1030	* Of course, the freebsd sendfile is a dire hack with no thoughts
959	* wasted on making it similar to other I/O functions.	1031	* wasted on making it similar to other I/O functions.
960	*/	1032	*/
961	{
962	off_t sbytes;	1033	off_t sbytes;
963	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1034	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
964		1035
965	#if 0 /* according to the manpage, this is correct, but broken behaviour */	1036	#if 0 /* according to the manpage, this is correct, but broken behaviour */
966	/* freebsd' sendfile will return 0 on success */	1037	/* freebsd' sendfile will return 0 on success */
967	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */	1038	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
968	/* not on e.g. EIO or EPIPE - sounds broken */	1039	/* not on e.g. EIO or EPIPE - sounds broken */
969	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)	1040	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
970	res = sbytes;	1041	res = sbytes;
971	#endif	1042	#endif
972		1043
973	/* according to source inspection, this is correct, and useful behaviour */	1044	/* according to source inspection, this is correct, and useful behaviour */
974	if (sbytes)	1045	if (sbytes)
975	res = sbytes;	1046	res = sbytes;
976	}
977		1047
978	# elif defined (__APPLE__)	1048	# elif defined (__APPLE__)
979
980	{
981	off_t sbytes = count;	1049	off_t sbytes = count;
982	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	1050	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983		1051
984	/* according to the manpage, sbytes is always valid */	1052	/* according to the manpage, sbytes is always valid */
985	if (sbytes)	1053	if (sbytes)
986	res = sbytes;	1054	res = sbytes;
987	}
988		1055
989	# elif __hpux	1056	# elif __hpux
990	res = sendfile (ofd, ifd, offset, count, 0, 0);	1057	res = sendfile (ofd, ifd, offset, count, 0, 0);
991		1058
992	# elif __solaris	1059	# elif __solaris
993	{
994	struct sendfilevec vec;	1060	struct sendfilevec vec;
995	size_t sbytes;	1061	size_t sbytes;
996		1062
997	vec.sfv_fd = ifd;	1063	vec.sfv_fd = ifd;
998	vec.sfv_flag = 0;	1064	vec.sfv_flag = 0;
999	vec.sfv_off = offset;	1065	vec.sfv_off = offset;
1000	vec.sfv_len = count;	1066	vec.sfv_len = count;
1001		1067
1002	res = sendfilev (ofd, &vec, 1, &sbytes);	1068	res = sendfilev (ofd, &vec, 1, &sbytes);
1003		1069
1004	if (res < 0 && sbytes)	1070	if (res < 0 && sbytes)
1005	res = sbytes;	1071	res = sbytes;
1006	}
1007		1072
1008	# endif	1073	# endif
1009		1074
1010	#elif defined (_WIN32)	1075	#elif defined (_WIN32) && 0
1011
1012	/* does not work, just for documentation of what would need to be done */	1076	/* does not work, just for documentation of what would need to be done */
1013	{	1077	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1078	/* libeio guarantees that the file offset does not change, and windows */
		1079	/* has no way to get an independent handle to the same file description */
1014	HANDLE h = TO_SOCKET (ifd);	1080	HANDLE h = TO_SOCKET (ifd);
1015	SetFilePointer (h, offset, 0, FILE_BEGIN);	1081	SetFilePointer (h, offset, 0, FILE_BEGIN);
1016	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);	1082	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1017	}
1018		1083
1019	#else	1084	#else
1020	res = -1;	1085	res = -1;
1021	errno = ENOSYS;	1086	errno = ENOSYS;
1022	#endif	1087	#endif
1023		1088
		1089	/* we assume sendfile can copy at least 128mb in one go */
		1090	if (res <= 128 * 1024 * 1024)
		1091	{
		1092	if (res > 0)
		1093	written += res;
		1094
		1095	if (written)
		1096	return written;
		1097
		1098	break;
		1099	}
		1100	else
		1101	{
		1102	/* if we requested more, then probably the kernel was lazy */
		1103	written += res;
		1104	offset += res;
		1105	count -= res;
		1106
		1107	if (!count)
		1108	return written;
		1109	}
		1110	}
		1111
1024	if (res < 0	1112	if (res < 0
1025	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1113	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1026	/* BSDs */	1114	/* BSDs */
1027	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1115	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1028	|| errno == ENOTSUP	1116	|| errno == ENOTSUP
1029	#endif	1117	#endif
		1118	#ifdef EOPNOTSUPP /* windows */
1030	|| errno == EOPNOTSUPP /* BSDs */	1119	|| errno == EOPNOTSUPP /* BSDs */
		1120	#endif
1031	#if __solaris	1121	#if __solaris
1032	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1122	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1033	#endif	1123	#endif
1034	)	1124	)
1035	)	1125	)
…		…
1039		1129
1040	res = 0;	1130	res = 0;
1041		1131
1042	while (count)	1132	while (count)
1043	{	1133	{
1044	ssize_t cnt;	1134	eio_ssize_t cnt;
1045		1135
1046	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1136	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1047		1137
1048	if (cnt <= 0)	1138	if (cnt <= 0)
1049	{	1139	{
…		…
1061		1151
1062	offset += cnt;	1152	offset += cnt;
1063	res += cnt;	1153	res += cnt;
1064	count -= cnt;	1154	count -= cnt;
1065	}	1155	}
		1156
		1157	FUBd;
1066	}	1158	}
1067		1159
1068	return res;	1160	return res;
		1161	}
		1162
		1163	#ifdef PAGESIZE
		1164	# define eio_pagesize() PAGESIZE
		1165	#else
		1166	static intptr_t
		1167	eio_pagesize (void)
		1168	{
		1169	static intptr_t page;
		1170
		1171	if (!page)
		1172	page = sysconf (_SC_PAGESIZE);
		1173
		1174	return page;
		1175	}
		1176	#endif
		1177
		1178	static void
		1179	eio_page_align (void **addr, size_t *length)
		1180	{
		1181	intptr_t mask = eio_pagesize () - 1;
		1182
		1183	/* round down addr */
		1184	intptr_t adj = mask & (intptr_t)*addr;
		1185
		1186	*addr = (void *)((intptr_t)*addr - adj);
		1187	*length += adj;
		1188
		1189	/* round up length */
		1190	*length = (*length + mask) & ~mask;
		1191	}
		1192
		1193	#if !_POSIX_MEMLOCK
		1194	# define eio__mlockall(a) EIO_ENOSYS ()
		1195	#else
		1196
		1197	static int
		1198	eio__mlockall (int flags)
		1199	{
		1200	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1201	extern int mallopt (int, int);
		1202	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1203	#endif
		1204
		1205	if (EIO_MCL_CURRENT != MCL_CURRENT
		1206	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1207	{
		1208	flags = 0
		1209	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1210	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1211	}
		1212
		1213	return mlockall (flags);
		1214	}
		1215	#endif
		1216
		1217	#if !_POSIX_MEMLOCK_RANGE
		1218	# define eio__mlock(a,b) EIO_ENOSYS ()
		1219	#else
		1220
		1221	static int
		1222	eio__mlock (void *addr, size_t length)
		1223	{
		1224	eio_page_align (&addr, &length);
		1225
		1226	return mlock (addr, length);
		1227	}
		1228
		1229	#endif
		1230
		1231	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1232	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1233	#else
		1234
		1235	static int
		1236	eio__msync (void *mem, size_t len, int flags)
		1237	{
		1238	eio_page_align (&mem, &len);
		1239
		1240	if (EIO_MS_ASYNC != MS_SYNC
		1241	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1242	|| EIO_MS_SYNC != MS_SYNC)
		1243	{
		1244	flags = 0
		1245	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1246	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1247	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1248	}
		1249
		1250	return msync (mem, len, flags);
		1251	}
		1252
		1253	#endif
		1254
		1255	static int
		1256	eio__mtouch (eio_req *req)
		1257	{
		1258	void *mem = req->ptr2;
		1259	size_t len = req->size;
		1260	int flags = req->int1;
		1261
		1262	eio_page_align (&mem, &len);
		1263
		1264	{
		1265	intptr_t addr = (intptr_t)mem;
		1266	intptr_t end = addr + len;
		1267	intptr_t page = eio_pagesize ();
		1268
		1269	if (addr < end)
		1270	if (flags & EIO_MT_MODIFY) /* modify */
		1271	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1272	else
		1273	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1274	}
		1275
		1276	return 0;
		1277	}
		1278
		1279	/*****************************************************************************/
		1280	/* requests implemented outside eio_execute, because they are so large */
		1281
		1282	static void
		1283	eio__realpath (eio_req *req, etp_worker *self)
		1284	{
		1285	char *rel = req->ptr1;
		1286	char *res;
		1287	char *tmp1, *tmp2;
		1288	#if SYMLOOP_MAX > 32
		1289	int symlinks = SYMLOOP_MAX;
		1290	#else
		1291	int symlinks = 32;
		1292	#endif
		1293
		1294	req->result = -1;
		1295
		1296	errno = EINVAL;
		1297	if (!rel)
		1298	return;
		1299
		1300	errno = ENOENT;
		1301	if (!*rel)
		1302	return;
		1303
		1304	if (!req->ptr2)
		1305	{
		1306	X_LOCK (wrklock);
		1307	req->flags |= EIO_FLAG_PTR2_FREE;
		1308	X_UNLOCK (wrklock);
		1309	req->ptr2 = malloc (PATH_MAX * 3);
		1310
		1311	errno = ENOMEM;
		1312	if (!req->ptr2)
		1313	return;
		1314	}
		1315
		1316	res = req->ptr2;
		1317	tmp1 = res + PATH_MAX;
		1318	tmp2 = tmp1 + PATH_MAX;
		1319
		1320	#if 0 /* disabled, the musl way to do things is just too racy */
		1321	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1322	/* on linux we may be able to ask the kernel */
		1323	{
		1324	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1325
		1326	if (fd >= 0)
		1327	{
		1328	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1329	req->result = readlink (tmp1, res, PATH_MAX);
		1330	close (fd);
		1331
		1332	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1333
		1334	if (req->result > 0)
		1335	goto done;
		1336	}
		1337	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1338	return;
		1339	}
		1340	#endif
		1341	#endif
		1342
		1343	if (*rel != '/')
		1344	{
		1345	if (!getcwd (res, PATH_MAX))
		1346	return;
		1347
		1348	if (res [1]) /* only use if not / */
		1349	res += strlen (res);
		1350	}
		1351
		1352	while (*rel)
		1353	{
		1354	eio_ssize_t len, linklen;
		1355	char *beg = rel;
		1356
		1357	while (*rel && *rel != '/')
		1358	++rel;
		1359
		1360	len = rel - beg;
		1361
		1362	if (!len) /* skip slashes */
		1363	{
		1364	++rel;
		1365	continue;
		1366	}
		1367
		1368	if (beg [0] == '.')
		1369	{
		1370	if (len == 1)
		1371	continue; /* . - nop */
		1372
		1373	if (beg [1] == '.' && len == 2)
		1374	{
		1375	/* .. - back up one component, if possible */
		1376
		1377	while (res != req->ptr2)
		1378	if (*--res == '/')
		1379	break;
		1380
		1381	continue;
		1382	}
		1383	}
		1384
		1385	errno = ENAMETOOLONG;
		1386	if (res + 1 + len + 1 >= tmp1)
		1387	return;
		1388
		1389	/* copy one component */
		1390	*res = '/';
		1391	memcpy (res + 1, beg, len);
		1392
		1393	/* zero-terminate, for readlink */
		1394	res [len + 1] = 0;
		1395
		1396	/* now check if it's a symlink */
		1397	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1398
		1399	if (linklen < 0)
		1400	{
		1401	if (errno != EINVAL)
		1402	return;
		1403
		1404	/* it's a normal directory. hopefully */
		1405	res += len + 1;
		1406	}
		1407	else
		1408	{
		1409	/* yay, it was a symlink - build new path in tmp2 */
		1410	int rellen = strlen (rel);
		1411
		1412	errno = ENAMETOOLONG;
		1413	if (linklen + 1 + rellen >= PATH_MAX)
		1414	return;
		1415
		1416	errno = ELOOP;
		1417	if (!--symlinks)
		1418	return;
		1419
		1420	if (*tmp1 == '/')
		1421	res = req->ptr2; /* symlink resolves to an absolute path */
		1422
		1423	/* we need to be careful, as rel might point into tmp2 already */
		1424	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1425	tmp2 [linklen] = '/';
		1426	memcpy (tmp2, tmp1, linklen);
		1427
		1428	rel = tmp2;
		1429	}
		1430	}
		1431
		1432	/* special case for the lone root path */
		1433	if (res == req->ptr2)
		1434	*res++ = '/';
		1435
		1436	req->result = res - (char *)req->ptr2;
		1437
		1438	done:
		1439	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1069	}	1440	}
1070		1441
1071	static signed char	1442	static signed char
1072	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1443	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1073	{	1444	{
…		…
1081		1452
1082	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1453	#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 */	1454	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1084		1455
1085	static void	1456	static void
1086	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1457	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1087	{	1458	{
1088	unsigned char bits [9 + sizeof (ino_t) * 8];	1459	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1089	unsigned char *bit = bits;	1460	unsigned char *bit = bits;
1090		1461
1091	assert (CHAR_BIT == 8);	1462	assert (CHAR_BIT == 8);
1092	assert (sizeof (eio_dirent) * 8 < 256);	1463	assert (sizeof (eio_dirent) * 8 < 256);
1093	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1464	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1095		1466
1096	if (size <= EIO_SORT_FAST)	1467	if (size <= EIO_SORT_FAST)
1097	return;	1468	return;
1098		1469
1099	/* first prepare an array of bits to test in our radix sort */	1470	/* 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 */	1471	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1101	/* inode_bits must contain all inodes ORed together */	1472	/* inode_bits must contain all inodes ORed together */
1102	/* which is used to skip bits that are 0 everywhere, which is very common */	1473	/* which is used to skip bits that are 0 everywhere, which is very common */
1103	{	1474	{
1104	ino_t endianness;	1475	eio_ino_t endianness;
1105	int i, j;	1476	int i, j;
1106		1477
1107	/* we store the byte offset of byte n into byte n of "endianness" */	1478	/* we store the byte offset of byte n into byte n of "endianness" */
1108	for (i = 0; i < sizeof (ino_t); ++i)	1479	for (i = 0; i < sizeof (eio_ino_t); ++i)
1109	((unsigned char *)&endianness)[i] = i;	1480	((unsigned char *)&endianness)[i] = i;
1110		1481
1111	*bit++ = 0;	1482	*bit++ = 0;
1112		1483
1113	for (i = 0; i < sizeof (ino_t); ++i)	1484	for (i = 0; i < sizeof (eio_ino_t); ++i)
1114	{	1485	{
1115	/* shifting off the byte offsets out of "endianness" */	1486	/* shifting off the byte offsets out of "endianness" */
1116	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1487	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1117	endianness >>= 8;	1488	endianness >>= 8;
1118		1489
1119	for (j = 0; j < 8; ++j)	1490	for (j = 0; j < 8; ++j)
1120	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1491	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1121	*bit++ = offs + j;	1492	*bit++ = offs + j;
1122	}	1493	}
1123		1494
1124	for (j = 0; j < 8; ++j)	1495	for (j = 0; j < 8; ++j)
1125	if (score_bits & (1 << j))	1496	if (score_bits & (1 << j))
1126	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1497	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1127	}	1498	}
1128		1499
1129	/* now actually do the sorting (a variant of MSD radix sort) */	1500	/* now actually do the sorting (a variant of MSD radix sort) */
1130	{	1501	{
1131	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1502	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1132	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1503	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1133	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1504	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1134	int stk_idx = 0;	1505	int stk_idx = 0;
1135		1506
1136	base_stk [stk_idx] = dents;	1507	base_stk [stk_idx] = dents;
1137	end_stk [stk_idx] = dents + size;	1508	end_stk [stk_idx] = dents + size;
1138	bit_stk [stk_idx] = bit - 1;	1509	bit_stk [stk_idx] = bit - 1;
…		…
1217	}	1588	}
1218	}	1589	}
1219	}	1590	}
1220		1591
1221	static void	1592	static void
1222	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1593	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1223	{	1594	{
1224	if (size <= 1)	1595	if (size <= 1)
1225	return; /* our insertion sort relies on size > 0 */	1596	return; /* our insertion sort relies on size > 0 */
1226		1597
1227	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1598	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1235		1606
1236	/* read a full directory */	1607	/* read a full directory */
1237	static void	1608	static void
1238	eio__scandir (eio_req *req, etp_worker *self)	1609	eio__scandir (eio_req *req, etp_worker *self)
1239	{	1610	{
1240	DIR *dirp;
1241	EIO_STRUCT_DIRENT *entp;
1242	char *name, *names;	1611	char *name, *names;
1243	int namesalloc = 4096;	1612	int namesalloc = 4096 - sizeof (void *) * 4;
1244	int namesoffs = 0;	1613	int namesoffs = 0;
1245	int flags = req->int1;	1614	int flags = req->int1;
1246	eio_dirent *dents = 0;	1615	eio_dirent *dents = 0;
1247	int dentalloc = 128;	1616	int dentalloc = 128;
1248	int dentoffs = 0;	1617	int dentoffs = 0;
1249	ino_t inode_bits = 0;	1618	eio_ino_t inode_bits = 0;
		1619	#ifdef _WIN32
		1620	HANDLE dirp;
		1621	WIN32_FIND_DATA entp;
		1622	#else
		1623	DIR *dirp;
		1624	EIO_STRUCT_DIRENT *entp;
		1625	#endif
1250		1626
1251	req->result = -1;	1627	req->result = -1;
1252		1628
1253	if (!(flags & EIO_READDIR_DENTS))	1629	if (!(flags & EIO_READDIR_DENTS))
1254	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1630	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1255		1631
1256	X_LOCK (wrklock);	1632	#ifdef _WIN32
1257	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1633	{
		1634	int len = strlen ((const char *)req->ptr1);
		1635	char *path = malloc (MAX_PATH);
		1636	const char *fmt;
		1637
		1638	if (!len)
		1639	fmt = "./*";
		1640	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1641	fmt = "%s*";
		1642	else
		1643	fmt = "%s/*";
		1644
		1645	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1646	dirp = FindFirstFile (path, &entp);
		1647	free (path);
		1648
		1649	if (dirp == INVALID_HANDLE_VALUE)
		1650	{
		1651	dirp = 0;
		1652
		1653	switch (GetLastError ())
		1654	{
		1655	case ERROR_FILE_NOT_FOUND:
		1656	req->result = 0;
		1657	break;
		1658
		1659	case ERROR_INVALID_NAME:
		1660	case ERROR_PATH_NOT_FOUND:
		1661	case ERROR_NO_MORE_FILES:
		1662	errno = ENOENT;
		1663	break;
		1664
		1665	case ERROR_NOT_ENOUGH_MEMORY:
		1666	errno = ENOMEM;
		1667	break;
		1668
		1669	default:
		1670	errno = EINVAL;
		1671	break;
		1672	}
		1673	}
		1674	}
		1675	#else
1258	self->dirp = dirp = opendir (req->ptr1);	1676	dirp = opendir (req->ptr1);
		1677	#endif
		1678
		1679	if (req->flags & EIO_FLAG_PTR1_FREE)
		1680	free (req->ptr1);
1259		1681
1260	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1682	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1261	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1683	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1262	req->ptr2 = names = malloc (namesalloc);	1684	req->ptr2 = names = malloc (namesalloc);
1263	X_UNLOCK (wrklock);
1264		1685
1265	if (dirp && names && (!flags || dents))	1686	if (dirp && names && (!flags || dents))
1266	for (;;)	1687	for (;;)
1267	{	1688	{
		1689	int done;
		1690
		1691	#ifdef _WIN32
		1692	done = !dirp;
		1693	#else
1268	errno = 0;	1694	errno = 0;
1269	entp = readdir (dirp);	1695	entp = readdir (dirp);
		1696	done = !entp;
		1697	#endif
1270		1698
1271	if (!entp)	1699	if (done)
1272	{	1700	{
		1701	#ifndef _WIN32
		1702	int old_errno = errno;
		1703	closedir (dirp);
		1704	errno = old_errno;
		1705
1273	if (errno)	1706	if (errno)
1274	break;	1707	break;
		1708	#endif
1275		1709
1276	/* sort etc. */	1710	/* sort etc. */
1277	req->int1 = flags;	1711	req->int1 = flags;
1278	req->result = dentoffs;	1712	req->result = dentoffs;
1279		1713
…		…
1308		1742
1309	break;	1743	break;
1310	}	1744	}
1311		1745
1312	/* now add the entry to our list(s) */	1746	/* now add the entry to our list(s) */
1313	name = entp->d_name;	1747	name = D_NAME (entp);
1314		1748
1315	/* skip . and .. entries */	1749	/* skip . and .. entries */
1316	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1750	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1317	{	1751	{
1318	int len = D_NAMLEN (entp) + 1;	1752	int len = D_NAMLEN (entp) + 1;
1319		1753
1320	while (expect_false (namesoffs + len > namesalloc))	1754	while (ecb_expect_false (namesoffs + len > namesalloc))
1321	{	1755	{
1322	namesalloc *= 2;	1756	namesalloc *= 2;
1323	X_LOCK (wrklock);
1324	req->ptr2 = names = realloc (names, namesalloc);	1757	req->ptr2 = names = realloc (names, namesalloc);
1325	X_UNLOCK (wrklock);
1326		1758
1327	if (!names)	1759	if (!names)
1328	break;	1760	break;
1329	}	1761	}
1330		1762
…		…
1332		1764
1333	if (dents)	1765	if (dents)
1334	{	1766	{
1335	struct eio_dirent *ent;	1767	struct eio_dirent *ent;
1336		1768
1337	if (expect_false (dentoffs == dentalloc))	1769	if (ecb_expect_false (dentoffs == dentalloc))
1338	{	1770	{
1339	dentalloc *= 2;	1771	dentalloc *= 2;
1340	X_LOCK (wrklock);
1341	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1772	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1342	X_UNLOCK (wrklock);
1343		1773
1344	if (!dents)	1774	if (!dents)
1345	break;	1775	break;
1346	}	1776	}
1347		1777
…		…
1427	if (EIO_CANCELLED (req))	1857	if (EIO_CANCELLED (req))
1428	{	1858	{
1429	errno = ECANCELED;	1859	errno = ECANCELED;
1430	break;	1860	break;
1431	}	1861	}
		1862
		1863	#ifdef _WIN32
		1864	if (!FindNextFile (dirp, &entp))
		1865	{
		1866	FindClose (dirp);
		1867	dirp = 0;
		1868	}
		1869	#endif
1432	}	1870	}
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	}	1871	}
1546		1872
1547	/*****************************************************************************/	1873	/*****************************************************************************/
1548		1874
1549	#define ALLOC(len) \	1875	#define ALLOC(len) \
…		…
1565	{	1891	{
1566	ETP_REQ *req;	1892	ETP_REQ *req;
1567	struct timespec ts;	1893	struct timespec ts;
1568	etp_worker *self = (etp_worker *)thr_arg;	1894	etp_worker *self = (etp_worker *)thr_arg;
1569		1895
1570	/* try to distribute timeouts somewhat randomly */	1896	/* try to distribute timeouts somewhat evenly */
1571	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1897	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1572		1898
1573	for (;;)	1899	for (;;)
1574	{	1900	{
		1901	ts.tv_sec = 0;
		1902
1575	X_LOCK (reqlock);	1903	X_LOCK (reqlock);
1576		1904
1577	for (;;)	1905	for (;;)
1578	{	1906	{
1579	self->req = req = reqq_shift (&req_queue);	1907	self->req = req = reqq_shift (&req_queue);
1580		1908
1581	if (req)	1909	if (req)
1582	break;	1910	break;
1583		1911
		1912	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1913	{
		1914	X_UNLOCK (reqlock);
		1915	X_LOCK (wrklock);
		1916	--started;
		1917	X_UNLOCK (wrklock);
		1918	goto quit;
		1919	}
		1920
1584	++idle;	1921	++idle;
1585		1922
1586	ts.tv_sec = time (0) + idle_timeout;	1923	if (idle <= max_idle)
1587	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1924	/* we are allowed to idle, so do so without any timeout */
		1925	X_COND_WAIT (reqwait, reqlock);
		1926	else
1588	{	1927	{
1589	if (idle > max_idle)	1928	/* initialise timeout once */
1590	{	1929	if (!ts.tv_sec)
1591	--idle;	1930	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		1931
1599	/* we are allowed to idle, so do so without any timeout */
1600	X_COND_WAIT (reqwait, reqlock);	1932	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1933	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1601	}	1934	}
1602		1935
1603	--idle;	1936	--idle;
1604	}	1937	}
1605		1938
…		…
1608	X_UNLOCK (reqlock);	1941	X_UNLOCK (reqlock);
1609		1942
1610	if (req->type < 0)	1943	if (req->type < 0)
1611	goto quit;	1944	goto quit;
1612		1945
1613	if (!EIO_CANCELLED (req))
1614	ETP_EXECUTE (self, req);	1946	ETP_EXECUTE (self, req);
1615		1947
1616	X_LOCK (reslock);	1948	X_LOCK (reslock);
1617		1949
1618	++npending;	1950	++npending;
1619		1951
…		…
1634	return 0;	1966	return 0;
1635	}	1967	}
1636		1968
1637	/*****************************************************************************/	1969	/*****************************************************************************/
1638		1970
		1971	int ecb_cold
1639	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1972	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1640	{	1973	{
		1974	#if !HAVE_PREADWRITE
		1975	X_MUTEX_CREATE (preadwritelock);
		1976	#endif
		1977
1641	return etp_init (want_poll, done_poll);	1978	return etp_init (want_poll, done_poll);
1642	}	1979	}
1643		1980
		1981	ecb_inline void
1644	static void eio_api_destroy (eio_req *req)	1982	eio_api_destroy (eio_req *req)
1645	{	1983	{
1646	free (req);	1984	free (req);
1647	}	1985	}
1648		1986
1649	#define REQ(rtype) \	1987	#define REQ(rtype) \
…		…
1668	{ \	2006	{ \
1669	eio_api_destroy (req); \	2007	eio_api_destroy (req); \
1670	return 0; \	2008	return 0; \
1671	}	2009	}
1672		2010
		2011	static void
1673	static void eio_execute (etp_worker *self, eio_req *req)	2012	eio_execute (etp_worker *self, eio_req *req)
1674	{	2013	{
		2014	if (ecb_expect_false (EIO_CANCELLED (req)))
		2015	{
		2016	req->result = -1;
		2017	req->errorno = ECANCELED;
		2018	return;
		2019	}
		2020
1675	switch (req->type)	2021	switch (req->type)
1676	{	2022	{
1677	case EIO_READ: ALLOC (req->size);	2023	case EIO_READ: ALLOC (req->size);
1678	req->result = req->offs >= 0	2024	req->result = req->offs >= 0
1679	? pread (req->int1, req->ptr2, req->size, req->offs)	2025	? pread (req->int1, req->ptr2, req->size, req->offs)
…		…
1681	case EIO_WRITE: req->result = req->offs >= 0	2027	case EIO_WRITE: req->result = req->offs >= 0
1682	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2028	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1683	: write (req->int1, req->ptr2, req->size); break;	2029	: write (req->int1, req->ptr2, req->size); break;
1684		2030
1685	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2031	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1686	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2032	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1687		2033
1688	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2034	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1689	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2035	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1690	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2036	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1691	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2037	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
…		…
1713	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2059	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1714	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2060	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1715	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2061	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;	2062	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1717		2063
		2064	case EIO_REALPATH: eio__realpath (req, self); break;
		2065
1718	case EIO_READLINK: ALLOC (PATH_MAX);	2066	case EIO_READLINK: ALLOC (PATH_MAX);
1719	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;	2067	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1720		2068
1721	case EIO_SYNC: req->result = 0; sync (); break;	2069	case EIO_SYNC: req->result = 0; sync (); break;
1722	case EIO_FSYNC: req->result = fsync (req->int1); break;	2070	case EIO_FSYNC: req->result = fsync (req->int1); break;
1723	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2071	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1724	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2072	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;	2073	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1726	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2074	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1727	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2075	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;	2076	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2077	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1729		2078
1730	case EIO_READDIR: eio__scandir (req, self); break;	2079	case EIO_READDIR: eio__scandir (req, self); break;
1731		2080
1732	case EIO_BUSY:	2081	case EIO_BUSY:
1733	#ifdef _WIN32	2082	#ifdef _WIN32
…		…
1774	case EIO_NOP:	2123	case EIO_NOP:
1775	req->result = 0;	2124	req->result = 0;
1776	break;	2125	break;
1777		2126
1778	case EIO_CUSTOM:	2127	case EIO_CUSTOM:
1779	((void (*)(eio_req *))req->feed) (req);	2128	req->feed (req);
1780	break;	2129	break;
1781		2130
1782	default:	2131	default:
1783	errno = ENOSYS;	2132	errno = ENOSYS;
1784	req->result = -1;	2133	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)	2182	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1834	{	2183	{
1835	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2184	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1836	}	2185	}
1837		2186
		2187	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2188	{
		2189	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2190	}
		2191
1838	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2192	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1839	{	2193	{
1840	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2194	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1841	}	2195	}
1842		2196
…		…
1883	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2237	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1884	{	2238	{
1885	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2239	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1886	}	2240	}
1887		2241
1888	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2242	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1889	{	2243	{
1890	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2244	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1891	}	2245	}
1892		2246
1893	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2247	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1913	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2267	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1914	{	2268	{
1915	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2269	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1916	}	2270	}
1917		2271
1918	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2272	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1919	{	2273	{
1920	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2274	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1921	}	2275	}
1922		2276
1923	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2277	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1937	}	2291	}
1938		2292
1939	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2293	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1940	{	2294	{
1941	return eio__1path (EIO_READLINK, path, pri, cb, data);	2295	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2296	}
		2297
		2298	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2299	{
		2300	return eio__1path (EIO_REALPATH, path, pri, cb, data);
1942	}	2301	}
1943		2302
1944	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2303	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1945	{	2304	{
1946	return eio__1path (EIO_STAT, path, pri, cb, data);	2305	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)	2364	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2006	{	2365	{
2007	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2366	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2008	}	2367	}
2009		2368
2010	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2369	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2011	{	2370	{
2012	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2371	REQ (EIO_CUSTOM); req->feed = execute; SEND;
2013	}	2372	}
2014		2373
2015	#endif	2374	#endif
2016		2375
2017	eio_req *eio_grp (eio_cb cb, void *data)	2376	eio_req *eio_grp (eio_cb cb, void *data)
…		…
2026	#undef SEND	2385	#undef SEND
2027		2386
2028	/*****************************************************************************/	2387	/*****************************************************************************/
2029	/* grp functions */	2388	/* grp functions */
2030		2389
		2390	void
2031	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2391	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2032	{	2392	{
2033	grp->int2 = limit;	2393	grp->int2 = limit;
2034	grp->feed = feed;	2394	grp->feed = feed;
2035		2395
2036	grp_try_feed (grp);	2396	grp_try_feed (grp);
2037	}	2397	}
2038		2398
		2399	void
2039	void eio_grp_limit (eio_req *grp, int limit)	2400	eio_grp_limit (eio_req *grp, int limit)
2040	{	2401	{
2041	grp->int2 = limit;	2402	grp->int2 = limit;
2042		2403
2043	grp_try_feed (grp);	2404	grp_try_feed (grp);
2044	}	2405	}
2045		2406
		2407	void
2046	void eio_grp_add (eio_req *grp, eio_req *req)	2408	eio_grp_add (eio_req *grp, eio_req *req)
2047	{	2409	{
2048	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2410	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2049		2411
2050	grp->flags |= EIO_FLAG_GROUPADD;	2412	grp->flags |= EIO_FLAG_GROUPADD;
2051		2413
…		…
2062	}	2424	}
2063		2425
2064	/*****************************************************************************/	2426	/*****************************************************************************/
2065	/* misc garbage */	2427	/* misc garbage */
2066		2428
		2429	eio_ssize_t
2067	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2430	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2068	{	2431	{
2069	etp_worker wrk;
2070	ssize_t ret;
2071
2072	wrk.dbuf = 0;
2073
2074	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2432	return eio__sendfile (ofd, ifd, offset, count);
2075
2076	if (wrk.dbuf)
2077	free (wrk.dbuf);
2078
2079	return ret;
2080	}	2433	}
2081		2434

Diff Legend

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