[ViewVC] Diff of: cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.88 by root, Thu Jul 14 19:31:43 2011 UTC vs.
Revision 1.101 by root, Wed Aug 3 15:25:38 2011 UTC

…		…
58	#include <sys/stat.h>	58	#include <sys/stat.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	#include <sys/statvfs.h>
64	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */	63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65	/* intptr_t only comes from stdint.h, says idiot openbsd coder */	64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66	#if HAVE_STDINT_H	65	#if HAVE_STDINT_H
67	# include <stdint.h>	66	# include <stdint.h>
68	#endif	67	#endif
69		68
70	#ifndef ECANCELED	69	#ifndef ECANCELED
71	# define ECANCELED EDOM	70	# define ECANCELED EDOM
72	#endif	71	#endif
		72	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
		77	# define ENOTSOCK WSAENOTSOCK
		78	#endif
73		79
74	static void eio_destroy (eio_req *req);	80	static void eio_destroy (eio_req *req);
75		81
76	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
77	# 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
…		…
96		102
97	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)	103	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
98		104
99	#ifdef _WIN32	105	#ifdef _WIN32
100		106
		107	#undef PAGESIZE
101	#define PAGESIZE 4096 /* GetSystemInfo? */	108	#define PAGESIZE 4096 /* GetSystemInfo? */
102		109
		110	/* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
		111
103	#ifdef EIO_STRUCT_STATI64	112	#ifdef EIO_STRUCT_STATI64
		113	/* look at perl's non-sloppy stat */
104	#define stat(path,buf) _stati64 (path,buf)	114	#define stat(path,buf) _stati64 (path,buf)
105	#define fstat(fd,buf) _fstati64 (path,buf)	115	#define fstat(fd,buf) _fstati64 (fd,buf)
106	#endif	116	#endif
107	#define lstat(path,buf) stat (path,buf)	117	#define lstat(path,buf) stat (path,buf)
108	#define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))	118	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
109	#define mkdir(path,mode) _mkdir (path)	119	#define mkdir(path,mode) _mkdir (path)
110	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))	120	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
111		121
		122	#define chmod(path,mode) _chmod (path, mode)
		123	#define dup(fd) _dup (fd)
		124	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		125
		126	#define fchmod(fd,mode) EIO_ENOSYS ()
112	#define chown(path,uid,gid) EIO_ENOSYS ()	127	#define chown(path,uid,gid) EIO_ENOSYS ()
113	#define fchown(fd,uid,gid) EIO_ENOSYS ()	128	#define fchown(fd,uid,gid) EIO_ENOSYS ()
114	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */	129	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
115	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */	130	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
116	#define mknod(path,mode,dev) EIO_ENOSYS ()	131	#define mknod(path,mode,dev) EIO_ENOSYS ()
117	#define sync() EIO_ENOSYS ()	132	#define sync() EIO_ENOSYS ()
		133	#define readlink(path,buf,s) EIO_ENOSYS ()
		134	#define statvfs(path,buf) EIO_ENOSYS ()
		135	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		136
		137	/* rename() uses MoveFile, which fails to overwrite */
		138	#define rename(old,neu) eio__rename (old, neu)
		139
		140	static int
		141	eio__rename (const char old, const char neu)
		142	{
		143	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		144	return 0;
		145
		146	/* should steal _dosmaperr */
		147	switch (GetLastError ())
		148	{
		149	case ERROR_FILE_NOT_FOUND:
		150	case ERROR_PATH_NOT_FOUND:
		151	case ERROR_INVALID_DRIVE:
		152	case ERROR_NO_MORE_FILES:
		153	case ERROR_BAD_NETPATH:
		154	case ERROR_BAD_NET_NAME:
		155	case ERROR_BAD_PATHNAME:
		156	case ERROR_FILENAME_EXCED_RANGE:
		157	errno = ENOENT;
		158	break;
		159
		160	default:
		161	errno = EACCES;
		162	break;
		163	}
		164
		165	return -1;
		166	}
118		167
119	/* we could even stat and see if it exists */	168	/* we could even stat and see if it exists */
120	static int	169	static int
121	symlink (const char old, const char neu)	170	symlink (const char old, const char neu)
122	{	171	{
		172	#if WINVER >= 0x0600
123	if (CreateSymbolicLink (neu, old, 1))	173	if (CreateSymbolicLink (neu, old, 1))
124	return 0;	174	return 0;
125		175
126	if (CreateSymbolicLink (neu, old, 0))	176	if (CreateSymbolicLink (neu, old, 0))
127	return 0;	177	return 0;
		178	#endif
128		179
129	return EIO_ERRNO (ENOENT, -1);	180	return EIO_ERRNO (ENOENT, -1);
130	}	181	}
		182
		183	/* POSIX API only */
		184	#define CreateHardLink(neu,old,flags) 0
		185	#define CreateSymbolicLink(neu,old,flags) 0
		186
		187	struct statvfs
		188	{
		189	int dummy;
		190	};
		191
		192	#define DT_DIR EIO_DT_DIR
		193	#define DT_REG EIO_DT_REG
		194	#define D_NAME(entp) entp.cFileName
		195	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
131		196
132	#else	197	#else
133		198
134	#include <sys/time.h>	199	#include <sys/time.h>
135	#include <sys/select.h>	200	#include <sys/select.h>
136	#include <sys/statvfs.h>	201	#include <sys/statvfs.h>
137	#include <unistd.h>	202	#include <unistd.h>
138	#include <utime.h>
139	#include <signal.h>	203	#include <signal.h>
140	#include <dirent.h>	204	#include <dirent.h>
141		205
142	#if _POSIX_MEMLOCK \|\| _POSIX_MEMLOCK_RANGE \|\| _POSIX_MAPPED_FILES	206	#if _POSIX_MEMLOCK \|\| _POSIX_MEMLOCK_RANGE \|\| _POSIX_MAPPED_FILES
143	#include <sys/mman.h>	207	#include <sys/mman.h>
144	#endif	208	#endif
		209
		210	#define D_NAME(entp) entp->d_name
145		211
146	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */	212	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
147	#if __FreeBSD__ \|\| defined __NetBSD__ \|\| defined __OpenBSD__	213	#if __FreeBSD__ \|\| defined __NetBSD__ \|\| defined __OpenBSD__
148	#define _DIRENT_HAVE_D_TYPE /* sigh */	214	#define _DIRENT_HAVE_D_TYPE /* sigh */
149	#define D_INO(de) (de)->d_fileno	215	#define D_INO(de) (de)->d_fileno
…		…
163		229
164	#ifndef EIO_STRUCT_DIRENT	230	#ifndef EIO_STRUCT_DIRENT
165	#define EIO_STRUCT_DIRENT struct dirent	231	#define EIO_STRUCT_DIRENT struct dirent
166	#endif	232	#endif
167		233
		234	#endif
		235
		236	#if HAVE_UTIMES
		237	# include <utime.h>
		238	#endif
		239
		240	#if HAVE_SYS_SYSCALL_H
		241	# include <sys/syscall.h>
		242	#endif
		243
		244	#if HAVE_SYS_PRCTL_H
		245	# include <sys/prctl.h>
168	#endif	246	#endif
169		247
170	#if HAVE_SENDFILE	248	#if HAVE_SENDFILE
171	# if __linux	249	# if __linux
172	# include <sys/sendfile.h>	250	# include <sys/sendfile.h>
…		…
187	#endif	265	#endif
188	#ifndef D_INO	266	#ifndef D_INO
189	# define D_INO(de) 0	267	# define D_INO(de) 0
190	#endif	268	#endif
191	#ifndef D_NAMLEN	269	#ifndef D_NAMLEN
192	# define D_NAMLEN(de) strlen ((de)->d_name)	270	# define D_NAMLEN(entp) strlen (D_NAME (entp))
193	#endif	271	#endif
194		272
195	/* used for struct dirent, AIX doesn't provide it */	273	/* used for struct dirent, AIX doesn't provide it */
196	#ifndef NAME_MAX	274	#ifndef NAME_MAX
197	# define NAME_MAX 4096	275	# define NAME_MAX 4096
…		…
204		282
205	/* buffer size for various temporary buffers */	283	/* buffer size for various temporary buffers */
206	#define EIO_BUFSIZE 65536	284	#define EIO_BUFSIZE 65536
207		285
208	#define dBUF \	286	#define dBUF \
209	char *eio_buf; \
210	ETP_WORKER_LOCK (self); \
211	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	287	char *eio_buf = malloc (EIO_BUFSIZE); \
212	ETP_WORKER_UNLOCK (self); \
213	errno = ENOMEM; \	288	errno = ENOMEM; \
214	if (!eio_buf) \	289	if (!eio_buf) \
215	return -1;	290	return -1
		291
		292	#define FUBd \
		293	free (eio_buf)
216		294
217	#define EIO_TICKS ((1000000 + 1023) >> 10)	295	#define EIO_TICKS ((1000000 + 1023) >> 10)
218		296
219	#define ETP_PRI_MIN EIO_PRI_MIN	297	#define ETP_PRI_MIN EIO_PRI_MIN
220	#define ETP_PRI_MAX EIO_PRI_MAX	298	#define ETP_PRI_MAX EIO_PRI_MAX
…		…
226	static int eio_finish (eio_req *req);	304	static int eio_finish (eio_req *req);
227	#define ETP_FINISH(req) eio_finish (req)	305	#define ETP_FINISH(req) eio_finish (req)
228	static void eio_execute (struct etp_worker self, eio_req req);	306	static void eio_execute (struct etp_worker self, eio_req req);
229	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	307	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
230		308
231	#define ETP_WORKER_CLEAR(req) \
232	if (wrk->dbuf) \
233	{ \
234	free (wrk->dbuf); \
235	wrk->dbuf = 0; \
236	} \
237	\
238	if (wrk->dirp) \
239	{ \
240	closedir (wrk->dirp); \
241	wrk->dirp = 0; \
242	}
243
244	#define ETP_WORKER_COMMON \
245	void *dbuf; \
246	DIR *dirp;
247
248	/*****************************************************************************/	309	/*****************************************************************************/
249		310
250	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	311	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
251		312
252	/* calculate time difference in ~1/EIO_TICKS of a second */	313	/* calculate time difference in ~1/EIO_TICKS of a second */
…		…
263	static void (*done_poll_cb) (void);	324	static void (*done_poll_cb) (void);
264		325
265	static unsigned int max_poll_time; /* reslock */	326	static unsigned int max_poll_time; /* reslock */
266	static unsigned int max_poll_reqs; /* reslock */	327	static unsigned int max_poll_reqs; /* reslock */
267		328
268	static volatile unsigned int nreqs; /* reqlock */	329	static unsigned int nreqs; /* reqlock */
269	static volatile unsigned int nready; /* reqlock */	330	static unsigned int nready; /* reqlock */
270	static volatile unsigned int npending; /* reqlock */	331	static unsigned int npending; /* reqlock */
271	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */	332	static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
272	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */	333	static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
273		334
274	static xmutex_t wrklock;	335	static xmutex_t wrklock;
275	static xmutex_t reslock;	336	static xmutex_t reslock;
276	static xmutex_t reqlock;	337	static xmutex_t reqlock;
277	static xcond_t reqwait;	338	static xcond_t reqwait;
…		…
280	/*	341	/*
281	* make our pread/pwrite emulation safe against themselves, but not against	342	* make our pread/pwrite emulation safe against themselves, but not against
282	* normal read/write by using a mutex. slows down execution a lot,	343	* normal read/write by using a mutex. slows down execution a lot,
283	* but that's your problem, not mine.	344	* but that's your problem, not mine.
284	*/	345	*/
285	static xmutex_t preadwritelock = X_MUTEX_INIT;	346	static xmutex_t preadwritelock;
286	#endif	347	#endif
287		348
288	typedef struct etp_worker	349	typedef struct etp_worker
289	{	350	{
290	/* locked by wrklock */	351	/* locked by wrklock */
…		…
293	xthread_t tid;	354	xthread_t tid;
294		355
295	/* locked by reslock, reqlock or wrklock */	356	/* locked by reslock, reqlock or wrklock */
296	ETP_REQ req; / currently processed request */	357	ETP_REQ req; / currently processed request */
297		358
		359	#ifdef ETP_WORKER_COMMON
298	ETP_WORKER_COMMON	360	ETP_WORKER_COMMON
		361	#endif
299	} etp_worker;	362	} etp_worker;
300		363
301	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	364	static etp_worker wrk_first; /* NOT etp */
302		365
303	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	366	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
304	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	367	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
305		368
306	/* worker threads management */	369	/* worker threads management */
307		370
308	static void ecb_cold	371	static void ecb_cold
309	etp_worker_clear (etp_worker *wrk)	372	etp_worker_clear (etp_worker *wrk)
310	{	373	{
311	ETP_WORKER_CLEAR (wrk);
312	}	374	}
313		375
314	static void ecb_cold	376	static void ecb_cold
315	etp_worker_free (etp_worker *wrk)	377	etp_worker_free (etp_worker *wrk)
316	{	378	{
…		…
377	} etp_reqq;	439	} etp_reqq;
378		440
379	static etp_reqq req_queue;	441	static etp_reqq req_queue;
380	static etp_reqq res_queue;	442	static etp_reqq res_queue;
381		443
		444	static void ecb_noinline ecb_cold
		445	reqq_init (etp_reqq *q)
		446	{
		447	int pri;
		448
		449	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		450	q->qs[pri] = q->qe[pri] = 0;
		451
		452	q->size = 0;
		453	}
		454
382	static int ecb_noinline	455	static int ecb_noinline
383	reqq_push (etp_reqq q, ETP_REQ req)	456	reqq_push (etp_reqq q, ETP_REQ req)
384	{	457	{
385	int pri = req->pri;	458	int pri = req->pri;
386	req->next = 0;	459	req->next = 0;
…		…
420	}	493	}
421		494
422	abort ();	495	abort ();
423	}	496	}
424		497
425	static void ecb_cold	498	static int ecb_cold
426	etp_thread_init (void)	499	etp_init (void (want_poll)(void), void (done_poll)(void))
427	{	500	{
428	#if !HAVE_PREADWRITE
429	X_MUTEX_CREATE (preadwritelock);
430	#endif
431	X_MUTEX_CREATE (wrklock);	501	X_MUTEX_CREATE (wrklock);
432	X_MUTEX_CREATE (reslock);	502	X_MUTEX_CREATE (reslock);
433	X_MUTEX_CREATE (reqlock);	503	X_MUTEX_CREATE (reqlock);
434	X_COND_CREATE (reqwait);	504	X_COND_CREATE (reqwait);
435	}
436		505
437	static void ecb_cold	506	reqq_init (&req_queue);
438	etp_atfork_prepare (void)	507	reqq_init (&res_queue);
439	{
440	}
441		508
442	static void ecb_cold	509	wrk_first.next =
443	etp_atfork_parent (void)	510	wrk_first.prev = &wrk_first;
444	{
445	}
446
447	static void ecb_cold
448	etp_atfork_child (void)
449	{
450	ETP_REQ *prv;
451
452	while ((prv = reqq_shift (&req_queue)))
453	ETP_DESTROY (prv);
454
455	while ((prv = reqq_shift (&res_queue)))
456	ETP_DESTROY (prv);
457
458	while (wrk_first.next != &wrk_first)
459	{
460	etp_worker *wrk = wrk_first.next;
461
462	if (wrk->req)
463	ETP_DESTROY (wrk->req);
464
465	etp_worker_clear (wrk);
466	etp_worker_free (wrk);
467	}
468		511
469	started = 0;	512	started = 0;
470	idle = 0;	513	idle = 0;
471	nreqs = 0;	514	nreqs = 0;
472	nready = 0;	515	nready = 0;
473	npending = 0;	516	npending = 0;
474
475	etp_thread_init ();
476	}
477
478	static void ecb_cold
479	etp_once_init (void)
480	{
481	etp_thread_init ();
482	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
483	}
484
485	static int ecb_cold
486	etp_init (void (want_poll)(void), void (done_poll)(void))
487	{
488	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
489
490	pthread_once (&doinit, etp_once_init);
491		517
492	want_poll_cb = want_poll;	518	want_poll_cb = want_poll;
493	done_poll_cb = done_poll;	519	done_poll_cb = done_poll;
494		520
495	return 0;	521	return 0;
…		…
953	#if !HAVE_FDATASYNC	979	#if !HAVE_FDATASYNC
954	# undef fdatasync	980	# undef fdatasync
955	# define fdatasync(fd) fsync (fd)	981	# define fdatasync(fd) fsync (fd)
956	#endif	982	#endif
957		983
		984	static int
		985	eio__syncfs (int fd)
		986	{
		987	int res;
		988
		989	#if HAVE_SYS_SYNCFS
		990	res = (int)syscall (__NR_syncfs, (int)(fd));
		991	#else
		992	res = -1;
		993	errno = ENOSYS;
		994	#endif
		995
		996	if (res < 0 && errno == ENOSYS && fd >= 0)
		997	sync ();
		998
		999	return res;
		1000	}
		1001
958	/* sync_file_range always needs emulation */	1002	/* sync_file_range always needs emulation */
959	static int	1003	static int
960	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	1004	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
961	{	1005	{
962	#if HAVE_SYNC_FILE_RANGE	1006	#if HAVE_SYNC_FILE_RANGE
…		…
1011	pread (fd, eio_buf, len, offset);	1055	pread (fd, eio_buf, len, offset);
1012	offset += len;	1056	offset += len;
1013	todo -= len;	1057	todo -= len;
1014	}	1058	}
1015		1059
		1060	FUBd;
		1061
1016	errno = 0;	1062	errno = 0;
1017	return count;	1063	return count;
1018	}	1064	}
1019		1065
1020	#endif	1066	#endif
1021		1067
1022	/* sendfile always needs emulation */	1068	/* sendfile always needs emulation */
1023	static eio_ssize_t	1069	static eio_ssize_t
1024	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1070	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
1025	{	1071	{
1026	eio_ssize_t written = 0;	1072	eio_ssize_t written = 0;
1027	eio_ssize_t res;	1073	eio_ssize_t res;
1028		1074
1029	if (!count)	1075	if (!count)
…		…
1127	&& (errno == ENOSYS \|\| errno == EINVAL \|\| errno == ENOTSOCK	1173	&& (errno == ENOSYS \|\| errno == EINVAL \|\| errno == ENOTSOCK
1128	/* BSDs */	1174	/* BSDs */
1129	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1175	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1130	\|\| errno == ENOTSUP	1176	\|\| errno == ENOTSUP
1131	#endif	1177	#endif
		1178	#ifdef EOPNOTSUPP /* windows */
1132	\|\| errno == EOPNOTSUPP /* BSDs */	1179	\|\| errno == EOPNOTSUPP /* BSDs */
		1180	#endif
1133	#if __solaris	1181	#if __solaris
1134	\|\| errno == EAFNOSUPPORT \|\| errno == EPROTOTYPE	1182	\|\| errno == EAFNOSUPPORT \|\| errno == EPROTOTYPE
1135	#endif	1183	#endif
1136	)	1184	)
1137	)	1185	)
…		…
1163		1211
1164	offset += cnt;	1212	offset += cnt;
1165	res += cnt;	1213	res += cnt;
1166	count -= cnt;	1214	count -= cnt;
1167	}	1215	}
		1216
		1217	FUBd;
1168	}	1218	}
1169		1219
1170	return res;	1220	return res;
1171	}	1221	}
1172		1222
…		…
1199	/* round up length */	1249	/* round up length */
1200	length = (length + mask) & ~mask;	1250	length = (length + mask) & ~mask;
1201	}	1251	}
1202		1252
1203	#if !_POSIX_MEMLOCK	1253	#if !_POSIX_MEMLOCK
1204	# define eio__mlockall(a) eio_nosyscall()	1254	# define eio__mlockall(a) EIO_ENOSYS ()
1205	#else	1255	#else
1206		1256
1207	static int	1257	static int
1208	eio__mlockall (int flags)	1258	eio__mlockall (int flags)
1209	{	1259	{
…		…
1462		1512
1463	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1513	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1464	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1514	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1465		1515
1466	static void	1516	static void
1467	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1517	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1468	{	1518	{
1469	unsigned char bits [9 + sizeof (ino_t) * 8];	1519	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1470	unsigned char *bit = bits;	1520	unsigned char *bit = bits;
1471		1521
1472	assert (CHAR_BIT == 8);	1522	assert (CHAR_BIT == 8);
1473	assert (sizeof (eio_dirent) * 8 < 256);	1523	assert (sizeof (eio_dirent) * 8 < 256);
1474	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1524	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1476		1526
1477	if (size <= EIO_SORT_FAST)	1527	if (size <= EIO_SORT_FAST)
1478	return;	1528	return;
1479		1529
1480	/* first prepare an array of bits to test in our radix sort */	1530	/* first prepare an array of bits to test in our radix sort */
1481	/* try to take endianness into account, as well as differences in ino_t sizes */	1531	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1482	/* inode_bits must contain all inodes ORed together */	1532	/* inode_bits must contain all inodes ORed together */
1483	/* which is used to skip bits that are 0 everywhere, which is very common */	1533	/* which is used to skip bits that are 0 everywhere, which is very common */
1484	{	1534	{
1485	ino_t endianness;	1535	eio_ino_t endianness;
1486	int i, j;	1536	int i, j;
1487		1537
1488	/* we store the byte offset of byte n into byte n of "endianness" */	1538	/* we store the byte offset of byte n into byte n of "endianness" */
1489	for (i = 0; i < sizeof (ino_t); ++i)	1539	for (i = 0; i < sizeof (eio_ino_t); ++i)
1490	((unsigned char *)&endianness)[i] = i;	1540	((unsigned char *)&endianness)[i] = i;
1491		1541
1492	*bit++ = 0;	1542	*bit++ = 0;
1493		1543
1494	for (i = 0; i < sizeof (ino_t); ++i)	1544	for (i = 0; i < sizeof (eio_ino_t); ++i)
1495	{	1545	{
1496	/* shifting off the byte offsets out of "endianness" */	1546	/* shifting off the byte offsets out of "endianness" */
1497	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1547	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1498	endianness >>= 8;	1548	endianness >>= 8;
1499		1549
1500	for (j = 0; j < 8; ++j)	1550	for (j = 0; j < 8; ++j)
1501	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1551	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1502	*bit++ = offs + j;	1552	*bit++ = offs + j;
1503	}	1553	}
1504		1554
1505	for (j = 0; j < 8; ++j)	1555	for (j = 0; j < 8; ++j)
1506	if (score_bits & (1 << j))	1556	if (score_bits & (1 << j))
1507	bit++ = offsetof (eio_dirent, score) 8 + j;	1557	bit++ = offsetof (eio_dirent, score) 8 + j;
1508	}	1558	}
1509		1559
1510	/* now actually do the sorting (a variant of MSD radix sort) */	1560	/* now actually do the sorting (a variant of MSD radix sort) */
1511	{	1561	{
1512	eio_dirent base_stk [9 + sizeof (ino_t) 8], *base;	1562	eio_dirent base_stk [9 + sizeof (eio_ino_t) 8], *base;
1513	eio_dirent end_stk [9 + sizeof (ino_t) 8], *end;	1563	eio_dirent end_stk [9 + sizeof (eio_ino_t) 8], *end;
1514	unsigned char bit_stk [9 + sizeof (ino_t) 8];	1564	unsigned char bit_stk [9 + sizeof (eio_ino_t) 8];
1515	int stk_idx = 0;	1565	int stk_idx = 0;
1516		1566
1517	base_stk [stk_idx] = dents;	1567	base_stk [stk_idx] = dents;
1518	end_stk [stk_idx] = dents + size;	1568	end_stk [stk_idx] = dents + size;
1519	bit_stk [stk_idx] = bit - 1;	1569	bit_stk [stk_idx] = bit - 1;
…		…
1598	}	1648	}
1599	}	1649	}
1600	}	1650	}
1601		1651
1602	static void	1652	static void
1603	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1653	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1604	{	1654	{
1605	if (size <= 1)	1655	if (size <= 1)
1606	return; /* our insertion sort relies on size > 0 */	1656	return; /* our insertion sort relies on size > 0 */
1607		1657
1608	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1658	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1616		1666
1617	/* read a full directory */	1667	/* read a full directory */
1618	static void	1668	static void
1619	eio__scandir (eio_req req, etp_worker self)	1669	eio__scandir (eio_req req, etp_worker self)
1620	{	1670	{
1621	DIR *dirp;
1622	EIO_STRUCT_DIRENT *entp;
1623	char name, names;	1671	char name, names;
1624	int namesalloc = 4096;	1672	int namesalloc = 4096 - sizeof (void ) 4;
1625	int namesoffs = 0;	1673	int namesoffs = 0;
1626	int flags = req->int1;	1674	int flags = req->int1;
1627	eio_dirent *dents = 0;	1675	eio_dirent *dents = 0;
1628	int dentalloc = 128;	1676	int dentalloc = 128;
1629	int dentoffs = 0;	1677	int dentoffs = 0;
1630	ino_t inode_bits = 0;	1678	eio_ino_t inode_bits = 0;
		1679	#ifdef _WIN32
		1680	HANDLE dirp;
		1681	WIN32_FIND_DATA entp;
		1682	#else
		1683	DIR *dirp;
		1684	EIO_STRUCT_DIRENT *entp;
		1685	#endif
1631		1686
1632	req->result = -1;	1687	req->result = -1;
1633		1688
1634	if (!(flags & EIO_READDIR_DENTS))	1689	if (!(flags & EIO_READDIR_DENTS))
1635	flags &= ~(EIO_READDIR_DIRS_FIRST \| EIO_READDIR_STAT_ORDER);	1690	flags &= ~(EIO_READDIR_DIRS_FIRST \| EIO_READDIR_STAT_ORDER);
1636		1691
1637	X_LOCK (wrklock);	1692	#ifdef _WIN32
1638	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1693	{
		1694	int len = strlen ((const char *)req->ptr1);
		1695	char *path = malloc (MAX_PATH);
		1696	const char *fmt;
		1697
		1698	if (!len)
		1699	fmt = "./*";
		1700	else if (((const char )req->ptr1)[len - 1] == '/' \|\| ((const char )req->ptr1)[len - 1] == '\\')
		1701	fmt = "%s*";
		1702	else
		1703	fmt = "%s/*";
		1704
		1705	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1706	dirp = FindFirstFile (path, &entp);
		1707	free (path);
		1708
		1709	if (dirp == INVALID_HANDLE_VALUE)
		1710	{
		1711	dirp = 0;
		1712
		1713	/* should steal _dosmaperr */
		1714	switch (GetLastError ())
		1715	{
		1716	case ERROR_FILE_NOT_FOUND:
		1717	req->result = 0;
		1718	break;
		1719
		1720	case ERROR_INVALID_NAME:
		1721	case ERROR_PATH_NOT_FOUND:
		1722	case ERROR_NO_MORE_FILES:
		1723	errno = ENOENT;
		1724	break;
		1725
		1726	case ERROR_NOT_ENOUGH_MEMORY:
		1727	errno = ENOMEM;
		1728	break;
		1729
		1730	default:
		1731	errno = EINVAL;
		1732	break;
		1733	}
		1734	}
		1735	}
		1736	#else
1639	self->dirp = dirp = opendir (req->ptr1);	1737	dirp = opendir (req->ptr1);
		1738	#endif
1640		1739
1641	if (req->flags & EIO_FLAG_PTR1_FREE)	1740	if (req->flags & EIO_FLAG_PTR1_FREE)
1642	free (req->ptr1);	1741	free (req->ptr1);
1643		1742
1644	req->flags \|= EIO_FLAG_PTR1_FREE \| EIO_FLAG_PTR2_FREE;	1743	req->flags \|= EIO_FLAG_PTR1_FREE \| EIO_FLAG_PTR2_FREE;
1645	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1744	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1646	req->ptr2 = names = malloc (namesalloc);	1745	req->ptr2 = names = malloc (namesalloc);
1647	X_UNLOCK (wrklock);
1648		1746
1649	if (dirp && names && (!flags \|\| dents))	1747	if (dirp && names && (!flags \|\| dents))
1650	for (;;)	1748	for (;;)
1651	{	1749	{
		1750	int done;
		1751
		1752	#ifdef _WIN32
		1753	done = !dirp;
		1754	#else
1652	errno = 0;	1755	errno = 0;
1653	entp = readdir (dirp);	1756	entp = readdir (dirp);
		1757	done = !entp;
		1758	#endif
1654		1759
1655	if (!entp)	1760	if (done)
1656	{	1761	{
		1762	#ifndef _WIN32
		1763	int old_errno = errno;
		1764	closedir (dirp);
		1765	errno = old_errno;
		1766
1657	if (errno)	1767	if (errno)
1658	break;	1768	break;
		1769	#endif
1659		1770
1660	/* sort etc. */	1771	/* sort etc. */
1661	req->int1 = flags;	1772	req->int1 = flags;
1662	req->result = dentoffs;	1773	req->result = dentoffs;
1663		1774
…		…
1692		1803
1693	break;	1804	break;
1694	}	1805	}
1695		1806
1696	/* now add the entry to our list(s) */	1807	/* now add the entry to our list(s) */
1697	name = entp->d_name;	1808	name = D_NAME (entp);
1698		1809
1699	/* skip . and .. entries */	1810	/* skip . and .. entries */
1700	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))	1811	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))
1701	{	1812	{
1702	int len = D_NAMLEN (entp) + 1;	1813	int len = D_NAMLEN (entp) + 1;
1703		1814
1704	while (ecb_expect_false (namesoffs + len > namesalloc))	1815	while (ecb_expect_false (namesoffs + len > namesalloc))
1705	{	1816	{
1706	namesalloc *= 2;	1817	namesalloc *= 2;
1707	X_LOCK (wrklock);
1708	req->ptr2 = names = realloc (names, namesalloc);	1818	req->ptr2 = names = realloc (names, namesalloc);
1709	X_UNLOCK (wrklock);
1710		1819
1711	if (!names)	1820	if (!names)
1712	break;	1821	break;
1713	}	1822	}
1714		1823
…		…
1719	struct eio_dirent *ent;	1828	struct eio_dirent *ent;
1720		1829
1721	if (ecb_expect_false (dentoffs == dentalloc))	1830	if (ecb_expect_false (dentoffs == dentalloc))
1722	{	1831	{
1723	dentalloc *= 2;	1832	dentalloc *= 2;
1724	X_LOCK (wrklock);
1725	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1833	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1726	X_UNLOCK (wrklock);
1727		1834
1728	if (!dents)	1835	if (!dents)
1729	break;	1836	break;
1730	}	1837	}
1731		1838
…		…
1811	if (EIO_CANCELLED (req))	1918	if (EIO_CANCELLED (req))
1812	{	1919	{
1813	errno = ECANCELED;	1920	errno = ECANCELED;
1814	break;	1921	break;
1815	}	1922	}
		1923
		1924	#ifdef _WIN32
		1925	if (!FindNextFile (dirp, &entp))
		1926	{
		1927	FindClose (dirp);
		1928	dirp = 0;
		1929	}
		1930	#endif
1816	}	1931	}
1817	}	1932	}
1818		1933
1819	/*****************************************************************************/	1934	/*****************************************************************************/
1820		1935
…		…
1836	X_THREAD_PROC (etp_proc)	1951	X_THREAD_PROC (etp_proc)
1837	{	1952	{
1838	ETP_REQ *req;	1953	ETP_REQ *req;
1839	struct timespec ts;	1954	struct timespec ts;
1840	etp_worker self = (etp_worker )thr_arg;	1955	etp_worker self = (etp_worker )thr_arg;
1841	int timeout;	1956
		1957	#if HAVE_PRCTL_SET_NAME
		1958	prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
		1959	#endif
1842		1960
1843	/* try to distribute timeouts somewhat evenly */	1961	/* try to distribute timeouts somewhat evenly */
1844	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1962	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1845		1963
1846	for (;;)	1964	for (;;)
…		…
1916	/*****************************************************************************/	2034	/*****************************************************************************/
1917		2035
1918	int ecb_cold	2036	int ecb_cold
1919	eio_init (void (want_poll)(void), void (done_poll)(void))	2037	eio_init (void (want_poll)(void), void (done_poll)(void))
1920	{	2038	{
		2039	#if !HAVE_PREADWRITE
		2040	X_MUTEX_CREATE (preadwritelock);
		2041	#endif
		2042
1921	return etp_init (want_poll, done_poll);	2043	return etp_init (want_poll, done_poll);
1922	}	2044	}
1923		2045
1924	ecb_inline void	2046	ecb_inline void
1925	eio_api_destroy (eio_req *req)	2047	eio_api_destroy (eio_req *req)
…		…
1970	case EIO_WRITE: req->result = req->offs >= 0	2092	case EIO_WRITE: req->result = req->offs >= 0
1971	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2093	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1972	: write (req->int1, req->ptr2, req->size); break;	2094	: write (req->int1, req->ptr2, req->size); break;
1973		2095
1974	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2096	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1975	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2097	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1976		2098
1977	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2099	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1978	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2100	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1979	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2101	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1980	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2102	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
…		…
2010	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;	2132	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2011		2133
2012	case EIO_SYNC: req->result = 0; sync (); break;	2134	case EIO_SYNC: req->result = 0; sync (); break;
2013	case EIO_FSYNC: req->result = fsync (req->int1); break;	2135	case EIO_FSYNC: req->result = fsync (req->int1); break;
2014	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2136	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2137	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2138	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2015	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2139	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2016	case EIO_MTOUCH: req->result = eio__mtouch (req); break;	2140	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2017	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2141	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2018	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2142	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2019	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2020	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;	2143	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2021		2144
2022	case EIO_READDIR: eio__scandir (req, self); break;	2145	case EIO_READDIR: eio__scandir (req, self); break;
2023		2146
2024	case EIO_BUSY:	2147	case EIO_BUSY:
…		…
2105	eio_req eio_msync (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2228	eio_req eio_msync (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2106	{	2229	{
2107	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2230	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2108	}	2231	}
2109		2232
		2233	eio_req eio_fdatasync (int fd, int pri, eio_cb cb, void data)
		2234	{
		2235	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2236	}
		2237
		2238	eio_req eio_syncfs (int fd, int pri, eio_cb cb, void data)
		2239	{
		2240	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2241	}
		2242
		2243	eio_req eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void data)
		2244	{
		2245	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2246	}
		2247
2110	eio_req eio_mtouch (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2248	eio_req eio_mtouch (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2111	{	2249	{
2112	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2250	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2113	}	2251	}
2114		2252
…		…
2120	eio_req eio_mlockall (int flags, int pri, eio_cb cb, void data)	2258	eio_req eio_mlockall (int flags, int pri, eio_cb cb, void data)
2121	{	2259	{
2122	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;	2260	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2123	}	2261	}
2124		2262
2125	eio_req eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void data)
2126	{
2127	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2128	}
2129
2130	eio_req eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void data)	2263	eio_req eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void data)
2131	{	2264	{
2132	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;	2265	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2133	}
2134
2135	eio_req eio_fdatasync (int fd, int pri, eio_cb cb, void data)
2136	{
2137	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2138	}	2266	}
2139		2267
2140	eio_req eio_close (int fd, int pri, eio_cb cb, void data)	2268	eio_req eio_close (int fd, int pri, eio_cb cb, void data)
2141	{	2269	{
2142	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2270	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
2370	/* misc garbage */	2498	/* misc garbage */
2371		2499
2372	eio_ssize_t	2500	eio_ssize_t
2373	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2501	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2374	{	2502	{
2375	etp_worker wrk;
2376	eio_ssize_t ret;
2377
2378	wrk.dbuf = 0;
2379
2380	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2503	return eio__sendfile (ofd, ifd, offset, count);
2381
2382	if (wrk.dbuf)
2383	free (wrk.dbuf);
2384
2385	return ret;
2386	}	2504	}
2387		2505

Diff Legend

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

Comparing libeio/eio.c (file contents): Revision 1.88 by root, Thu Jul 14 19:31:43 2011 UTC vs. Revision 1.101 by root, Wed Aug 3 15:25:38 2011 UTC

Diff Legend

Comparing libeio/eio.c (file contents):
Revision 1.88 by root, Thu Jul 14 19:31:43 2011 UTC vs.
Revision 1.101 by root, Wed Aug 3 15:25:38 2011 UTC