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

Comparing libeio/eio.c (file contents):
Revision 1.89 by root, Thu Jul 14 22:36:17 2011 UTC vs.
Revision 1.102 by root, Sun Aug 21 01:03:09 2011 UTC

…		…
43		43
44	#include "eio.h"	44	#include "eio.h"
45	#include "ecb.h"	45	#include "ecb.h"
46		46
47	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
48	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define X_STACKSIZE EIO_STACKSIZE
49	#endif	49	#endif
50	#include "xthread.h"	50	#include "xthread.h"
51		51
52	#include <errno.h>	52	#include <errno.h>
53	#include <stddef.h>	53	#include <stddef.h>
…		…
67	#endif	67	#endif
68		68
69	#ifndef ECANCELED	69	#ifndef ECANCELED
70	# define ECANCELED EDOM	70	# define ECANCELED EDOM
71	#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
72		79
73	static void eio_destroy (eio_req *req);	80	static void eio_destroy (eio_req *req);
74		81
75	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
76	# 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
…		…
95		102
96	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)	103	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
97		104
98	#ifdef _WIN32	105	#ifdef _WIN32
99		106
		107	#undef PAGESIZE
100	#define PAGESIZE 4096 /* GetSystemInfo? */	108	#define PAGESIZE 4096 /* GetSystemInfo? */
101		109
		110	/* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
		111
102	#ifdef EIO_STRUCT_STATI64	112	#ifdef EIO_STRUCT_STATI64
		113	/* look at perl's non-sloppy stat */
103	#define stat(path,buf) _stati64 (path,buf)	114	#define stat(path,buf) _stati64 (path,buf)
104	#define fstat(fd,buf) _fstati64 (path,buf)	115	#define fstat(fd,buf) _fstati64 (fd,buf)
105	#endif	116	#endif
106	#define lstat(path,buf) stat (path,buf)	117	#define lstat(path,buf) stat (path,buf)
107	#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))
108	#define mkdir(path,mode) _mkdir (path)	119	#define mkdir(path,mode) _mkdir (path)
109	#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))
110		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 ()
111	#define chown(path,uid,gid) EIO_ENOSYS ()	127	#define chown(path,uid,gid) EIO_ENOSYS ()
112	#define fchown(fd,uid,gid) EIO_ENOSYS ()	128	#define fchown(fd,uid,gid) EIO_ENOSYS ()
113	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */	129	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
114	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */	130	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
115	#define mknod(path,mode,dev) EIO_ENOSYS ()	131	#define mknod(path,mode,dev) EIO_ENOSYS ()
116	#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	}
117		167
118	/* we could even stat and see if it exists */	168	/* we could even stat and see if it exists */
119	static int	169	static int
120	symlink (const char old, const char neu)	170	symlink (const char old, const char neu)
121	{	171	{
		172	#if WINVER >= 0x0600
122	if (CreateSymbolicLink (neu, old, 1))	173	if (CreateSymbolicLink (neu, old, 1))
123	return 0;	174	return 0;
124		175
125	if (CreateSymbolicLink (neu, old, 0))	176	if (CreateSymbolicLink (neu, old, 0))
126	return 0;	177	return 0;
		178	#endif
127		179
128	return EIO_ERRNO (ENOENT, -1);	180	return EIO_ERRNO (ENOENT, -1);
129	}	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)
130		196
131	#else	197	#else
132		198
133	#include <sys/time.h>	199	#include <sys/time.h>
134	#include <sys/select.h>	200	#include <sys/select.h>
135	#include <sys/statvfs.h>	201	#include <sys/statvfs.h>
136	#include <unistd.h>	202	#include <unistd.h>
137	#include <utime.h>
138	#include <signal.h>	203	#include <signal.h>
139	#include <dirent.h>	204	#include <dirent.h>
140		205
141	#if _POSIX_MEMLOCK \|\| _POSIX_MEMLOCK_RANGE \|\| _POSIX_MAPPED_FILES	206	#if _POSIX_MEMLOCK \|\| _POSIX_MEMLOCK_RANGE \|\| _POSIX_MAPPED_FILES
142	#include <sys/mman.h>	207	#include <sys/mman.h>
143	#endif	208	#endif
		209
		210	#define D_NAME(entp) entp->d_name
144		211
145	/* 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 */
146	#if __FreeBSD__ \|\| defined __NetBSD__ \|\| defined __OpenBSD__	213	#if __FreeBSD__ \|\| defined __NetBSD__ \|\| defined __OpenBSD__
147	#define _DIRENT_HAVE_D_TYPE /* sigh */	214	#define _DIRENT_HAVE_D_TYPE /* sigh */
148	#define D_INO(de) (de)->d_fileno	215	#define D_INO(de) (de)->d_fileno
…		…
162		229
163	#ifndef EIO_STRUCT_DIRENT	230	#ifndef EIO_STRUCT_DIRENT
164	#define EIO_STRUCT_DIRENT struct dirent	231	#define EIO_STRUCT_DIRENT struct dirent
165	#endif	232	#endif
166		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>
167	#endif	246	#endif
168		247
169	#if HAVE_SENDFILE	248	#if HAVE_SENDFILE
170	# if __linux	249	# if __linux
171	# include <sys/sendfile.h>	250	# include <sys/sendfile.h>
…		…
186	#endif	265	#endif
187	#ifndef D_INO	266	#ifndef D_INO
188	# define D_INO(de) 0	267	# define D_INO(de) 0
189	#endif	268	#endif
190	#ifndef D_NAMLEN	269	#ifndef D_NAMLEN
191	# define D_NAMLEN(de) strlen ((de)->d_name)	270	# define D_NAMLEN(entp) strlen (D_NAME (entp))
192	#endif	271	#endif
193		272
194	/* used for struct dirent, AIX doesn't provide it */	273	/* used for struct dirent, AIX doesn't provide it */
195	#ifndef NAME_MAX	274	#ifndef NAME_MAX
196	# define NAME_MAX 4096	275	# define NAME_MAX 4096
…		…
203		282
204	/* buffer size for various temporary buffers */	283	/* buffer size for various temporary buffers */
205	#define EIO_BUFSIZE 65536	284	#define EIO_BUFSIZE 65536
206		285
207	#define dBUF \	286	#define dBUF \
208	char *eio_buf; \
209	ETP_WORKER_LOCK (self); \
210	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	287	char *eio_buf = malloc (EIO_BUFSIZE); \
211	ETP_WORKER_UNLOCK (self); \
212	errno = ENOMEM; \	288	errno = ENOMEM; \
213	if (!eio_buf) \	289	if (!eio_buf) \
214	return -1;	290	return -1
		291
		292	#define FUBd \
		293	free (eio_buf)
215		294
216	#define EIO_TICKS ((1000000 + 1023) >> 10)	295	#define EIO_TICKS ((1000000 + 1023) >> 10)
217		296
218	#define ETP_PRI_MIN EIO_PRI_MIN	297	#define ETP_PRI_MIN EIO_PRI_MIN
219	#define ETP_PRI_MAX EIO_PRI_MAX	298	#define ETP_PRI_MAX EIO_PRI_MAX
…		…
225	static int eio_finish (eio_req *req);	304	static int eio_finish (eio_req *req);
226	#define ETP_FINISH(req) eio_finish (req)	305	#define ETP_FINISH(req) eio_finish (req)
227	static void eio_execute (struct etp_worker self, eio_req req);	306	static void eio_execute (struct etp_worker self, eio_req req);
228	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	307	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
229		308
230	#define ETP_WORKER_CLEAR(req) \
231	if (wrk->dbuf) \
232	{ \
233	free (wrk->dbuf); \
234	wrk->dbuf = 0; \
235	} \
236	\
237	if (wrk->dirp) \
238	{ \
239	closedir (wrk->dirp); \
240	wrk->dirp = 0; \
241	}
242
243	#define ETP_WORKER_COMMON \
244	void *dbuf; \
245	DIR *dirp;
246
247	/*****************************************************************************/	309	/*****************************************************************************/
248		310
249	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	311	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
250		312
251	/* calculate time difference in ~1/EIO_TICKS of a second */	313	/* calculate time difference in ~1/EIO_TICKS of a second */
…		…
262	static void (*done_poll_cb) (void);	324	static void (*done_poll_cb) (void);
263		325
264	static unsigned int max_poll_time; /* reslock */	326	static unsigned int max_poll_time; /* reslock */
265	static unsigned int max_poll_reqs; /* reslock */	327	static unsigned int max_poll_reqs; /* reslock */
266		328
267	static volatile unsigned int nreqs; /* reqlock */	329	static unsigned int nreqs; /* reqlock */
268	static volatile unsigned int nready; /* reqlock */	330	static unsigned int nready; /* reqlock */
269	static volatile unsigned int npending; /* reqlock */	331	static unsigned int npending; /* reqlock */
270	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 */
271	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 */
272		334
273	static xmutex_t wrklock;	335	static xmutex_t wrklock;
274	static xmutex_t reslock;	336	static xmutex_t reslock;
275	static xmutex_t reqlock;	337	static xmutex_t reqlock;
276	static xcond_t reqwait;	338	static xcond_t reqwait;
…		…
279	/*	341	/*
280	* make our pread/pwrite emulation safe against themselves, but not against	342	* make our pread/pwrite emulation safe against themselves, but not against
281	* 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,
282	* but that's your problem, not mine.	344	* but that's your problem, not mine.
283	*/	345	*/
284	static xmutex_t preadwritelock = X_MUTEX_INIT;	346	static xmutex_t preadwritelock;
285	#endif	347	#endif
286		348
287	typedef struct etp_worker	349	typedef struct etp_worker
288	{	350	{
289	/* locked by wrklock */	351	/* locked by wrklock */
…		…
292	xthread_t tid;	354	xthread_t tid;
293		355
294	/* locked by reslock, reqlock or wrklock */	356	/* locked by reslock, reqlock or wrklock */
295	ETP_REQ req; / currently processed request */	357	ETP_REQ req; / currently processed request */
296		358
		359	#ifdef ETP_WORKER_COMMON
297	ETP_WORKER_COMMON	360	ETP_WORKER_COMMON
		361	#endif
298	} etp_worker;	362	} etp_worker;
299		363
300	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	364	static etp_worker wrk_first; /* NOT etp */
301		365
302	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	366	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
303	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	367	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
304		368
305	/* worker threads management */	369	/* worker threads management */
306		370
307	static void ecb_cold	371	static void ecb_cold
308	etp_worker_clear (etp_worker *wrk)	372	etp_worker_clear (etp_worker *wrk)
309	{	373	{
310	ETP_WORKER_CLEAR (wrk);
311	}	374	}
312		375
313	static void ecb_cold	376	static void ecb_cold
314	etp_worker_free (etp_worker *wrk)	377	etp_worker_free (etp_worker *wrk)
315	{	378	{
…		…
376	} etp_reqq;	439	} etp_reqq;
377		440
378	static etp_reqq req_queue;	441	static etp_reqq req_queue;
379	static etp_reqq res_queue;	442	static etp_reqq res_queue;
380		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
381	static int ecb_noinline	455	static int ecb_noinline
382	reqq_push (etp_reqq q, ETP_REQ req)	456	reqq_push (etp_reqq q, ETP_REQ req)
383	{	457	{
384	int pri = req->pri;	458	int pri = req->pri;
385	req->next = 0;	459	req->next = 0;
…		…
419	}	493	}
420		494
421	abort ();	495	abort ();
422	}	496	}
423		497
424	static void ecb_cold	498	static int ecb_cold
425	etp_thread_init (void)	499	etp_init (void (want_poll)(void), void (done_poll)(void))
426	{	500	{
427	#if !HAVE_PREADWRITE
428	X_MUTEX_CREATE (preadwritelock);
429	#endif
430	X_MUTEX_CREATE (wrklock);	501	X_MUTEX_CREATE (wrklock);
431	X_MUTEX_CREATE (reslock);	502	X_MUTEX_CREATE (reslock);
432	X_MUTEX_CREATE (reqlock);	503	X_MUTEX_CREATE (reqlock);
433	X_COND_CREATE (reqwait);	504	X_COND_CREATE (reqwait);
434	}
435		505
436	static void ecb_cold	506	reqq_init (&req_queue);
437	etp_atfork_prepare (void)	507	reqq_init (&res_queue);
438	{
439	}
440		508
441	static void ecb_cold	509	wrk_first.next =
442	etp_atfork_parent (void)	510	wrk_first.prev = &wrk_first;
443	{
444	}
445
446	static void ecb_cold
447	etp_atfork_child (void)
448	{
449	ETP_REQ *prv;
450
451	while ((prv = reqq_shift (&req_queue)))
452	ETP_DESTROY (prv);
453
454	while ((prv = reqq_shift (&res_queue)))
455	ETP_DESTROY (prv);
456
457	while (wrk_first.next != &wrk_first)
458	{
459	etp_worker *wrk = wrk_first.next;
460
461	if (wrk->req)
462	ETP_DESTROY (wrk->req);
463
464	etp_worker_clear (wrk);
465	etp_worker_free (wrk);
466	}
467		511
468	started = 0;	512	started = 0;
469	idle = 0;	513	idle = 0;
470	nreqs = 0;	514	nreqs = 0;
471	nready = 0;	515	nready = 0;
472	npending = 0;	516	npending = 0;
473
474	etp_thread_init ();
475	}
476
477	static void ecb_cold
478	etp_once_init (void)
479	{
480	etp_thread_init ();
481	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
482	}
483
484	static int ecb_cold
485	etp_init (void (want_poll)(void), void (done_poll)(void))
486	{
487	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
488
489	pthread_once (&doinit, etp_once_init);
490		517
491	want_poll_cb = want_poll;	518	want_poll_cb = want_poll;
492	done_poll_cb = done_poll;	519	done_poll_cb = done_poll;
493		520
494	return 0;	521	return 0;
…		…
952	#if !HAVE_FDATASYNC	979	#if !HAVE_FDATASYNC
953	# undef fdatasync	980	# undef fdatasync
954	# define fdatasync(fd) fsync (fd)	981	# define fdatasync(fd) fsync (fd)
955	#endif	982	#endif
956		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
957	/* sync_file_range always needs emulation */	1002	/* sync_file_range always needs emulation */
958	static int	1003	static int
959	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)
960	{	1005	{
961	#if HAVE_SYNC_FILE_RANGE	1006	#if HAVE_SYNC_FILE_RANGE
…		…
1010	pread (fd, eio_buf, len, offset);	1055	pread (fd, eio_buf, len, offset);
1011	offset += len;	1056	offset += len;
1012	todo -= len;	1057	todo -= len;
1013	}	1058	}
1014		1059
		1060	FUBd;
		1061
1015	errno = 0;	1062	errno = 0;
1016	return count;	1063	return count;
1017	}	1064	}
1018		1065
1019	#endif	1066	#endif
1020		1067
1021	/* sendfile always needs emulation */	1068	/* sendfile always needs emulation */
1022	static eio_ssize_t	1069	static eio_ssize_t
1023	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)
1024	{	1071	{
1025	eio_ssize_t written = 0;	1072	eio_ssize_t written = 0;
1026	eio_ssize_t res;	1073	eio_ssize_t res;
1027		1074
1028	if (!count)	1075	if (!count)
…		…
1126	&& (errno == ENOSYS \|\| errno == EINVAL \|\| errno == ENOTSOCK	1173	&& (errno == ENOSYS \|\| errno == EINVAL \|\| errno == ENOTSOCK
1127	/* BSDs */	1174	/* BSDs */
1128	#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... */
1129	\|\| errno == ENOTSUP	1176	\|\| errno == ENOTSUP
1130	#endif	1177	#endif
		1178	#ifdef EOPNOTSUPP /* windows */
1131	\|\| errno == EOPNOTSUPP /* BSDs */	1179	\|\| errno == EOPNOTSUPP /* BSDs */
		1180	#endif
1132	#if __solaris	1181	#if __solaris
1133	\|\| errno == EAFNOSUPPORT \|\| errno == EPROTOTYPE	1182	\|\| errno == EAFNOSUPPORT \|\| errno == EPROTOTYPE
1134	#endif	1183	#endif
1135	)	1184	)
1136	)	1185	)
…		…
1162		1211
1163	offset += cnt;	1212	offset += cnt;
1164	res += cnt;	1213	res += cnt;
1165	count -= cnt;	1214	count -= cnt;
1166	}	1215	}
		1216
		1217	FUBd;
1167	}	1218	}
1168		1219
1169	return res;	1220	return res;
1170	}	1221	}
1171		1222
…		…
1461		1512
1462	#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 */
1463	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1514	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1464		1515
1465	static void	1516	static void
1466	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)
1467	{	1518	{
1468	unsigned char bits [9 + sizeof (ino_t) * 8];	1519	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1469	unsigned char *bit = bits;	1520	unsigned char *bit = bits;
1470		1521
1471	assert (CHAR_BIT == 8);	1522	assert (CHAR_BIT == 8);
1472	assert (sizeof (eio_dirent) * 8 < 256);	1523	assert (sizeof (eio_dirent) * 8 < 256);
1473	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1524	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1475		1526
1476	if (size <= EIO_SORT_FAST)	1527	if (size <= EIO_SORT_FAST)
1477	return;	1528	return;
1478		1529
1479	/* 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 */
1480	/* 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 */
1481	/* inode_bits must contain all inodes ORed together */	1532	/* inode_bits must contain all inodes ORed together */
1482	/* 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 */
1483	{	1534	{
1484	ino_t endianness;	1535	eio_ino_t endianness;
1485	int i, j;	1536	int i, j;
1486		1537
1487	/* 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" */
1488	for (i = 0; i < sizeof (ino_t); ++i)	1539	for (i = 0; i < sizeof (eio_ino_t); ++i)
1489	((unsigned char *)&endianness)[i] = i;	1540	((unsigned char *)&endianness)[i] = i;
1490		1541
1491	*bit++ = 0;	1542	*bit++ = 0;
1492		1543
1493	for (i = 0; i < sizeof (ino_t); ++i)	1544	for (i = 0; i < sizeof (eio_ino_t); ++i)
1494	{	1545	{
1495	/* shifting off the byte offsets out of "endianness" */	1546	/* shifting off the byte offsets out of "endianness" */
1496	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1547	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1497	endianness >>= 8;	1548	endianness >>= 8;
1498		1549
1499	for (j = 0; j < 8; ++j)	1550	for (j = 0; j < 8; ++j)
1500	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1551	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1501	*bit++ = offs + j;	1552	*bit++ = offs + j;
1502	}	1553	}
1503		1554
1504	for (j = 0; j < 8; ++j)	1555	for (j = 0; j < 8; ++j)
1505	if (score_bits & (1 << j))	1556	if (score_bits & (1 << j))
1506	bit++ = offsetof (eio_dirent, score) 8 + j;	1557	bit++ = offsetof (eio_dirent, score) 8 + j;
1507	}	1558	}
1508		1559
1509	/* now actually do the sorting (a variant of MSD radix sort) */	1560	/* now actually do the sorting (a variant of MSD radix sort) */
1510	{	1561	{
1511	eio_dirent base_stk [9 + sizeof (ino_t) 8], *base;	1562	eio_dirent base_stk [9 + sizeof (eio_ino_t) 8], *base;
1512	eio_dirent end_stk [9 + sizeof (ino_t) 8], *end;	1563	eio_dirent end_stk [9 + sizeof (eio_ino_t) 8], *end;
1513	unsigned char bit_stk [9 + sizeof (ino_t) 8];	1564	unsigned char bit_stk [9 + sizeof (eio_ino_t) 8];
1514	int stk_idx = 0;	1565	int stk_idx = 0;
1515		1566
1516	base_stk [stk_idx] = dents;	1567	base_stk [stk_idx] = dents;
1517	end_stk [stk_idx] = dents + size;	1568	end_stk [stk_idx] = dents + size;
1518	bit_stk [stk_idx] = bit - 1;	1569	bit_stk [stk_idx] = bit - 1;
…		…
1597	}	1648	}
1598	}	1649	}
1599	}	1650	}
1600		1651
1601	static void	1652	static void
1602	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)
1603	{	1654	{
1604	if (size <= 1)	1655	if (size <= 1)
1605	return; /* our insertion sort relies on size > 0 */	1656	return; /* our insertion sort relies on size > 0 */
1606		1657
1607	/* 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 */
…		…
1615		1666
1616	/* read a full directory */	1667	/* read a full directory */
1617	static void	1668	static void
1618	eio__scandir (eio_req req, etp_worker self)	1669	eio__scandir (eio_req req, etp_worker self)
1619	{	1670	{
1620	DIR *dirp;
1621	EIO_STRUCT_DIRENT *entp;
1622	char name, names;	1671	char name, names;
1623	int namesalloc = 4096;	1672	int namesalloc = 4096 - sizeof (void ) 4;
1624	int namesoffs = 0;	1673	int namesoffs = 0;
1625	int flags = req->int1;	1674	int flags = req->int1;
1626	eio_dirent *dents = 0;	1675	eio_dirent *dents = 0;
1627	int dentalloc = 128;	1676	int dentalloc = 128;
1628	int dentoffs = 0;	1677	int dentoffs = 0;
1629	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
1630		1686
1631	req->result = -1;	1687	req->result = -1;
1632		1688
1633	if (!(flags & EIO_READDIR_DENTS))	1689	if (!(flags & EIO_READDIR_DENTS))
1634	flags &= ~(EIO_READDIR_DIRS_FIRST \| EIO_READDIR_STAT_ORDER);	1690	flags &= ~(EIO_READDIR_DIRS_FIRST \| EIO_READDIR_STAT_ORDER);
1635		1691
1636	X_LOCK (wrklock);	1692	#ifdef _WIN32
1637	/* 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
1638	self->dirp = dirp = opendir (req->ptr1);	1737	dirp = opendir (req->ptr1);
		1738	#endif
1639		1739
1640	if (req->flags & EIO_FLAG_PTR1_FREE)	1740	if (req->flags & EIO_FLAG_PTR1_FREE)
1641	free (req->ptr1);	1741	free (req->ptr1);
1642		1742
1643	req->flags \|= EIO_FLAG_PTR1_FREE \| EIO_FLAG_PTR2_FREE;	1743	req->flags \|= EIO_FLAG_PTR1_FREE \| EIO_FLAG_PTR2_FREE;
1644	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1744	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1645	req->ptr2 = names = malloc (namesalloc);	1745	req->ptr2 = names = malloc (namesalloc);
1646	X_UNLOCK (wrklock);
1647		1746
1648	if (dirp && names && (!flags \|\| dents))	1747	if (dirp && names && (!flags \|\| dents))
1649	for (;;)	1748	for (;;)
1650	{	1749	{
		1750	int done;
		1751
		1752	#ifdef _WIN32
		1753	done = !dirp;
		1754	#else
1651	errno = 0;	1755	errno = 0;
1652	entp = readdir (dirp);	1756	entp = readdir (dirp);
		1757	done = !entp;
		1758	#endif
1653		1759
1654	if (!entp)	1760	if (done)
1655	{	1761	{
		1762	#ifndef _WIN32
		1763	int old_errno = errno;
		1764	closedir (dirp);
		1765	errno = old_errno;
		1766
1656	if (errno)	1767	if (errno)
1657	break;	1768	break;
		1769	#endif
1658		1770
1659	/* sort etc. */	1771	/* sort etc. */
1660	req->int1 = flags;	1772	req->int1 = flags;
1661	req->result = dentoffs;	1773	req->result = dentoffs;
1662		1774
…		…
1691		1803
1692	break;	1804	break;
1693	}	1805	}
1694		1806
1695	/* now add the entry to our list(s) */	1807	/* now add the entry to our list(s) */
1696	name = entp->d_name;	1808	name = D_NAME (entp);
1697		1809
1698	/* skip . and .. entries */	1810	/* skip . and .. entries */
1699	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))	1811	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))
1700	{	1812	{
1701	int len = D_NAMLEN (entp) + 1;	1813	int len = D_NAMLEN (entp) + 1;
1702		1814
1703	while (ecb_expect_false (namesoffs + len > namesalloc))	1815	while (ecb_expect_false (namesoffs + len > namesalloc))
1704	{	1816	{
1705	namesalloc *= 2;	1817	namesalloc *= 2;
1706	X_LOCK (wrklock);
1707	req->ptr2 = names = realloc (names, namesalloc);	1818	req->ptr2 = names = realloc (names, namesalloc);
1708	X_UNLOCK (wrklock);
1709		1819
1710	if (!names)	1820	if (!names)
1711	break;	1821	break;
1712	}	1822	}
1713		1823
…		…
1718	struct eio_dirent *ent;	1828	struct eio_dirent *ent;
1719		1829
1720	if (ecb_expect_false (dentoffs == dentalloc))	1830	if (ecb_expect_false (dentoffs == dentalloc))
1721	{	1831	{
1722	dentalloc *= 2;	1832	dentalloc *= 2;
1723	X_LOCK (wrklock);
1724	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));	1833	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1725	X_UNLOCK (wrklock);
1726		1834
1727	if (!dents)	1835	if (!dents)
1728	break;	1836	break;
1729	}	1837	}
1730		1838
…		…
1810	if (EIO_CANCELLED (req))	1918	if (EIO_CANCELLED (req))
1811	{	1919	{
1812	errno = ECANCELED;	1920	errno = ECANCELED;
1813	break;	1921	break;
1814	}	1922	}
		1923
		1924	#ifdef _WIN32
		1925	if (!FindNextFile (dirp, &entp))
		1926	{
		1927	FindClose (dirp);
		1928	dirp = 0;
		1929	}
		1930	#endif
1815	}	1931	}
1816	}	1932	}
1817		1933
1818	/*****************************************************************************/	1934	/*****************************************************************************/
1819		1935
…		…
1835	X_THREAD_PROC (etp_proc)	1951	X_THREAD_PROC (etp_proc)
1836	{	1952	{
1837	ETP_REQ *req;	1953	ETP_REQ *req;
1838	struct timespec ts;	1954	struct timespec ts;
1839	etp_worker self = (etp_worker )thr_arg;	1955	etp_worker self = (etp_worker )thr_arg;
1840	int timeout;	1956
		1957	#if HAVE_PRCTL_SET_NAME
		1958	prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
		1959	#endif
1841		1960
1842	/* try to distribute timeouts somewhat evenly */	1961	/* try to distribute timeouts somewhat evenly */
1843	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1962	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1844		1963
1845	for (;;)	1964	for (;;)
…		…
1915	/*****************************************************************************/	2034	/*****************************************************************************/
1916		2035
1917	int ecb_cold	2036	int ecb_cold
1918	eio_init (void (want_poll)(void), void (done_poll)(void))	2037	eio_init (void (want_poll)(void), void (done_poll)(void))
1919	{	2038	{
		2039	#if !HAVE_PREADWRITE
		2040	X_MUTEX_CREATE (preadwritelock);
		2041	#endif
		2042
1920	return etp_init (want_poll, done_poll);	2043	return etp_init (want_poll, done_poll);
1921	}	2044	}
1922		2045
1923	ecb_inline void	2046	ecb_inline void
1924	eio_api_destroy (eio_req *req)	2047	eio_api_destroy (eio_req *req)
…		…
1969	case EIO_WRITE: req->result = req->offs >= 0	2092	case EIO_WRITE: req->result = req->offs >= 0
1970	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2093	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1971	: write (req->int1, req->ptr2, req->size); break;	2094	: write (req->int1, req->ptr2, req->size); break;
1972		2095
1973	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;
1974	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;
1975		2098
1976	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2099	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1977	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2100	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1978	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2101	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1979	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2102	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
…		…
2009	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;	2132	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2010		2133
2011	case EIO_SYNC: req->result = 0; sync (); break;	2134	case EIO_SYNC: req->result = 0; sync (); break;
2012	case EIO_FSYNC: req->result = fsync (req->int1); break;	2135	case EIO_FSYNC: req->result = fsync (req->int1); break;
2013	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;
2014	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;
2015	case EIO_MTOUCH: req->result = eio__mtouch (req); break;	2140	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2016	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;
2017	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2142	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2018	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2019	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;
2020		2144
2021	case EIO_READDIR: eio__scandir (req, self); break;	2145	case EIO_READDIR: eio__scandir (req, self); break;
2022		2146
2023	case EIO_BUSY:	2147	case EIO_BUSY:
…		…
2104	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)
2105	{	2229	{
2106	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;
2107	}	2231	}
2108		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
2109	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)
2110	{	2249	{
2111	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;
2112	}	2251	}
2113		2252
…		…
2119	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)
2120	{	2259	{
2121	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;	2260	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2122	}	2261	}
2123		2262
2124	eio_req eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void data)
2125	{
2126	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2127	}
2128
2129	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)
2130	{	2264	{
2131	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;
2132	}
2133
2134	eio_req eio_fdatasync (int fd, int pri, eio_cb cb, void data)
2135	{
2136	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2137	}	2266	}
2138		2267
2139	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)
2140	{	2269	{
2141	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2270	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
2369	/* misc garbage */	2498	/* misc garbage */
2370		2499
2371	eio_ssize_t	2500	eio_ssize_t
2372	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)
2373	{	2502	{
2374	etp_worker wrk;
2375	eio_ssize_t ret;
2376
2377	wrk.dbuf = 0;
2378
2379	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2503	return eio__sendfile (ofd, ifd, offset, count);
2380
2381	if (wrk.dbuf)
2382	free (wrk.dbuf);
2383
2384	return ret;
2385	}	2504	}
2386		2505

Diff Legend

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

Comparing libeio/eio.c (file contents): Revision 1.89 by root, Thu Jul 14 22:36:17 2011 UTC vs. Revision 1.102 by root, Sun Aug 21 01:03:09 2011 UTC

Diff Legend

Comparing libeio/eio.c (file contents):
Revision 1.89 by root, Thu Jul 14 22:36:17 2011 UTC vs.
Revision 1.102 by root, Sun Aug 21 01:03:09 2011 UTC