[ViewVC] Diff of: 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 1.105 by root, Mon Sep 26 18:23:18 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 X_STACKSIZE EIO_STACKSIZE
48	#endif	49	#endif
49	#include "xthread.h"	50	#include "xthread.h"
50		51
51	#include <errno.h>	52	#include <errno.h>
52	#include <stddef.h>	53	#include <stddef.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	/* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
		111
		112	#ifdef EIO_STRUCT_STATI64
		113	/* look at perl's non-sloppy stat */
		114	#define stat(path,buf) _stati64 (path,buf)
		115	#define fstat(fd,buf) _fstati64 (fd,buf)
		116	#endif
		117	#define lstat(path,buf) stat (path,buf)
		118	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		119	#define mkdir(path,mode) _mkdir (path)
		120	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		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 ()
		127	#define chown(path,uid,gid) EIO_ENOSYS ()
		128	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		129	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		130	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		131	#define mknod(path,mode,dev) 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	}
		167
		168	/* we could even stat and see if it exists */
		169	static int
		170	symlink (const char old, const char neu)
		171	{
		172	#if WINVER >= 0x0600
		173	if (CreateSymbolicLink (neu, old, 1))
		174	return 0;
		175
		176	if (CreateSymbolicLink (neu, old, 0))
		177	return 0;
		178	#endif
		179
		180	return EIO_ERRNO (ENOENT, -1);
		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)
		196
84	#else	197	#else
85		198
86	# include <sys/time.h>	199	#include <sys/time.h>
87	# include <sys/select.h>	200	#include <sys/select.h>
		201	#include <sys/statvfs.h>
88	# include <unistd.h>	202	#include <unistd.h>
		203	#include <signal.h>
		204	#include <dirent.h>
		205
		206	#if _POSIX_MEMLOCK \|\| _POSIX_MEMLOCK_RANGE \|\| _POSIX_MAPPED_FILES
		207	#include <sys/mman.h>
		208	#endif
		209
		210	#define D_NAME(entp) entp->d_name
		211
		212	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		213	#if __FreeBSD__ \|\| defined __NetBSD__ \|\| defined __OpenBSD__
		214	#define _DIRENT_HAVE_D_TYPE /* sigh */
		215	#define D_INO(de) (de)->d_fileno
		216	#define D_NAMLEN(de) (de)->d_namlen
		217	#elif __linux \|\| defined d_ino \|\| _XOPEN_SOURCE >= 600
		218	#define D_INO(de) (de)->d_ino
		219	#endif
		220
		221	#ifdef _D_EXACT_NAMLEN
		222	#undef D_NAMLEN
		223	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		224	#endif
		225
		226	#ifdef _DIRENT_HAVE_D_TYPE
		227	#define D_TYPE(de) (de)->d_type
		228	#endif
		229
		230	#ifndef EIO_STRUCT_DIRENT
		231	#define EIO_STRUCT_DIRENT struct dirent
		232	#endif
		233
		234	#endif
		235
		236	#if HAVE_UTIMES
89	# include <utime.h>	237	# 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	238	#endif
105		239
106	#ifdef _D_EXACT_NAMLEN	240	#if HAVE_SYS_SYSCALL_H
107	# undef D_NAMLEN	241	# include <sys/syscall.h>
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	242	#endif
114		243
115	# ifndef EIO_STRUCT_DIRENT	244	#if HAVE_SYS_PRCTL_H
116	# define EIO_STRUCT_DIRENT struct dirent	245	# include <sys/prctl.h>
117	# endif
118
119	#endif	246	#endif
120		247
121	#if HAVE_SENDFILE	248	#if HAVE_SENDFILE
122	# if __linux	249	# if __linux
123	# include <sys/sendfile.h>	250	# include <sys/sendfile.h>
…		…
138	#endif	265	#endif
139	#ifndef D_INO	266	#ifndef D_INO
140	# define D_INO(de) 0	267	# define D_INO(de) 0
141	#endif	268	#endif
142	#ifndef D_NAMLEN	269	#ifndef D_NAMLEN
143	# define D_NAMLEN(de) strlen ((de)->d_name)	270	# define D_NAMLEN(entp) strlen (D_NAME (entp))
144	#endif	271	#endif
145		272
146	/* used for struct dirent, AIX doesn't provide it */	273	/* used for struct dirent, AIX doesn't provide it */
147	#ifndef NAME_MAX	274	#ifndef NAME_MAX
148	# define NAME_MAX 4096	275	# define NAME_MAX 4096
…		…
155		282
156	/* buffer size for various temporary buffers */	283	/* buffer size for various temporary buffers */
157	#define EIO_BUFSIZE 65536	284	#define EIO_BUFSIZE 65536
158		285
159	#define dBUF \	286	#define dBUF \
160	char *eio_buf; \
161	ETP_WORKER_LOCK (self); \
162	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	287	char *eio_buf = malloc (EIO_BUFSIZE); \
163	ETP_WORKER_UNLOCK (self); \
164	errno = ENOMEM; \	288	errno = ENOMEM; \
165	if (!eio_buf) \	289	if (!eio_buf) \
166	return -1;	290	return -1
		291
		292	#define FUBd \
		293	free (eio_buf)
167		294
168	#define EIO_TICKS ((1000000 + 1023) >> 10)	295	#define EIO_TICKS ((1000000 + 1023) >> 10)
169		296
170	/*****************************************************************************/	297	/*****************************************************************************/
171		298
172	#if __GNUC__ >= 3	299	struct tmpbuf
173	# define expect(expr,value) __builtin_expect ((expr),(value))	300	{
174	#else	301	void *ptr;
175	# define expect(expr,value) (expr)	302	int len;
176	#endif	303	};
177		304
178	#define expect_false(expr) expect ((expr) != 0, 0)	305	static void *
179	#define expect_true(expr) expect ((expr) != 0, 1)	306	tmpbuf_get (struct tmpbuf *buf, int len)
		307	{
		308	if (buf->len < len)
		309	{
		310	free (buf->ptr);
		311	buf->ptr = malloc (buf->len = len);
		312	}
		313
		314	return buf->ptr;
		315	}
180		316
181	/*****************************************************************************/	317	/*****************************************************************************/
182		318
183	#define ETP_PRI_MIN EIO_PRI_MIN	319	#define ETP_PRI_MIN EIO_PRI_MIN
184	#define ETP_PRI_MAX EIO_PRI_MAX	320	#define ETP_PRI_MAX EIO_PRI_MAX
…		…
190	static int eio_finish (eio_req *req);	326	static int eio_finish (eio_req *req);
191	#define ETP_FINISH(req) eio_finish (req)	327	#define ETP_FINISH(req) eio_finish (req)
192	static void eio_execute (struct etp_worker self, eio_req req);	328	static void eio_execute (struct etp_worker self, eio_req req);
193	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	329	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194		330
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	/*****************************************************************************/	331	/*****************************************************************************/
213		332
214	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	333	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215		334
216	/* calculate time difference in ~1/EIO_TICKS of a second */	335	/* calculate time difference in ~1/EIO_TICKS of a second */
		336	ecb_inline int
217	static int tvdiff (struct timeval tv1, struct timeval tv2)	337	tvdiff (struct timeval tv1, struct timeval tv2)
218	{	338	{
219	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	339	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	340	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
221	}	341	}
222		342
…		…
226	static void (*done_poll_cb) (void);	346	static void (*done_poll_cb) (void);
227		347
228	static unsigned int max_poll_time; /* reslock */	348	static unsigned int max_poll_time; /* reslock */
229	static unsigned int max_poll_reqs; /* reslock */	349	static unsigned int max_poll_reqs; /* reslock */
230		350
231	static volatile unsigned int nreqs; /* reqlock */	351	static unsigned int nreqs; /* reqlock */
232	static volatile unsigned int nready; /* reqlock */	352	static unsigned int nready; /* reqlock */
233	static volatile unsigned int npending; /* reqlock */	353	static unsigned int npending; /* reqlock */
234	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */	354	static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
235	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */	355	static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
236		356
237	static xmutex_t wrklock;	357	static xmutex_t wrklock;
238	static xmutex_t reslock;	358	static xmutex_t reslock;
239	static xmutex_t reqlock;	359	static xmutex_t reqlock;
240	static xcond_t reqwait;	360	static xcond_t reqwait;
…		…
243	/*	363	/*
244	* make our pread/pwrite emulation safe against themselves, but not against	364	* make our pread/pwrite emulation safe against themselves, but not against
245	* normal read/write by using a mutex. slows down execution a lot,	365	* normal read/write by using a mutex. slows down execution a lot,
246	* but that's your problem, not mine.	366	* but that's your problem, not mine.
247	*/	367	*/
248	static xmutex_t preadwritelock = X_MUTEX_INIT;	368	static xmutex_t preadwritelock;
249	#endif	369	#endif
250		370
251	typedef struct etp_worker	371	typedef struct etp_worker
252	{	372	{
		373	struct tmpbuf tmpbuf;
		374
253	/* locked by wrklock */	375	/* locked by wrklock */
254	struct etp_worker prev, next;	376	struct etp_worker prev, next;
255		377
256	xthread_t tid;	378	xthread_t tid;
257		379
258	/* locked by reslock, reqlock or wrklock */	380	#ifdef ETP_WORKER_COMMON
259	ETP_REQ req; / currently processed request */
260
261	ETP_WORKER_COMMON	381	ETP_WORKER_COMMON
		382	#endif
262	} etp_worker;	383	} etp_worker;
263		384
264	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	385	static etp_worker wrk_first; /* NOT etp */
265		386
266	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	387	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	388	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268		389
269	/* worker threads management */	390	/* worker threads management */
270		391
		392	static void ecb_cold
271	static void etp_worker_clear (etp_worker *wrk)	393	etp_worker_clear (etp_worker *wrk)
272	{	394	{
273	ETP_WORKER_CLEAR (wrk);
274	}	395	}
275		396
		397	static void ecb_cold
276	static void etp_worker_free (etp_worker *wrk)	398	etp_worker_free (etp_worker *wrk)
277	{	399	{
		400	free (wrk->tmpbuf.ptr);
		401
278	wrk->next->prev = wrk->prev;	402	wrk->next->prev = wrk->prev;
279	wrk->prev->next = wrk->next;	403	wrk->prev->next = wrk->next;
280		404
281	free (wrk);	405	free (wrk);
282	}	406	}
283		407
284	static unsigned int etp_nreqs (void)	408	static unsigned int
		409	etp_nreqs (void)
285	{	410	{
286	int retval;	411	int retval;
287	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	412	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288	retval = nreqs;	413	retval = nreqs;
289	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	414	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290	return retval;	415	return retval;
291	}	416	}
292		417
293	static unsigned int etp_nready (void)	418	static unsigned int
		419	etp_nready (void)
294	{	420	{
295	unsigned int retval;	421	unsigned int retval;
296		422
297	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	423	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298	retval = nready;	424	retval = nready;
299	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	425	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300		426
301	return retval;	427	return retval;
302	}	428	}
303		429
304	static unsigned int etp_npending (void)	430	static unsigned int
		431	etp_npending (void)
305	{	432	{
306	unsigned int retval;	433	unsigned int retval;
307		434
308	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	435	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309	retval = npending;	436	retval = npending;
310	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	437	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311		438
312	return retval;	439	return retval;
313	}	440	}
314		441
315	static unsigned int etp_nthreads (void)	442	static unsigned int
		443	etp_nthreads (void)
316	{	444	{
317	unsigned int retval;	445	unsigned int retval;
318		446
319	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	447	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320	retval = started;	448	retval = started;
…		…
334	} etp_reqq;	462	} etp_reqq;
335		463
336	static etp_reqq req_queue;	464	static etp_reqq req_queue;
337	static etp_reqq res_queue;	465	static etp_reqq res_queue;
338		466
		467	static void ecb_noinline ecb_cold
		468	reqq_init (etp_reqq *q)
		469	{
		470	int pri;
		471
		472	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		473	q->qs[pri] = q->qe[pri] = 0;
		474
		475	q->size = 0;
		476	}
		477
		478	static int ecb_noinline
339	static int reqq_push (etp_reqq q, ETP_REQ req)	479	reqq_push (etp_reqq q, ETP_REQ req)
340	{	480	{
341	int pri = req->pri;	481	int pri = req->pri;
342	req->next = 0;	482	req->next = 0;
343		483
344	if (q->qe[pri])	484	if (q->qe[pri])
…		…
350	q->qe[pri] = q->qs[pri] = req;	490	q->qe[pri] = q->qs[pri] = req;
351		491
352	return q->size++;	492	return q->size++;
353	}	493	}
354		494
		495	static ETP_REQ * ecb_noinline
355	static ETP_REQ reqq_shift (etp_reqq q)	496	reqq_shift (etp_reqq *q)
356	{	497	{
357	int pri;	498	int pri;
358		499
359	if (!q->size)	500	if (!q->size)
360	return 0;	501	return 0;
…		…
375	}	516	}
376		517
377	abort ();	518	abort ();
378	}	519	}
379		520
380	static void etp_thread_init (void)	521	static int ecb_cold
		522	etp_init (void (want_poll)(void), void (done_poll)(void))
381	{	523	{
382	X_MUTEX_CREATE (wrklock);	524	X_MUTEX_CREATE (wrklock);
383	X_MUTEX_CREATE (reslock);	525	X_MUTEX_CREATE (reslock);
384	X_MUTEX_CREATE (reqlock);	526	X_MUTEX_CREATE (reqlock);
385	X_COND_CREATE (reqwait);	527	X_COND_CREATE (reqwait);
386	}
387		528
388	static void etp_atfork_prepare (void)	529	reqq_init (&req_queue);
389	{	530	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		531
398	static void etp_atfork_parent (void)	532	wrk_first.next =
399	{	533	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		534
429	started = 0;	535	started = 0;
430	idle = 0;	536	idle = 0;
431	nreqs = 0;	537	nreqs = 0;
432	nready = 0;	538	nready = 0;
433	npending = 0;	539	npending = 0;
434		540
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;	541	want_poll_cb = want_poll;
453	done_poll_cb = done_poll;	542	done_poll_cb = done_poll;
454		543
455	return 0;	544	return 0;
456	}	545	}
457		546
458	X_THREAD_PROC (etp_proc);	547	X_THREAD_PROC (etp_proc);
459		548
		549	static void ecb_cold
460	static void etp_start_thread (void)	550	etp_start_thread (void)
461	{	551	{
462	etp_worker *wrk = calloc (1, sizeof (etp_worker));	552	etp_worker *wrk = calloc (1, sizeof (etp_worker));
463		553
464	/TODO/	554	/TODO/
465	assert (("unable to allocate worker thread data", wrk));	555	assert (("unable to allocate worker thread data", wrk));
…		…
478	free (wrk);	568	free (wrk);
479		569
480	X_UNLOCK (wrklock);	570	X_UNLOCK (wrklock);
481	}	571	}
482		572
		573	static void
483	static void etp_maybe_start_thread (void)	574	etp_maybe_start_thread (void)
484	{	575	{
485	if (expect_true (etp_nthreads () >= wanted))	576	if (ecb_expect_true (etp_nthreads () >= wanted))
486	return;	577	return;
487		578
488	/* todo: maybe use idle here, but might be less exact */	579	/* todo: maybe use idle here, but might be less exact */
489	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	580	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
490	return;	581	return;
491		582
492	etp_start_thread ();	583	etp_start_thread ();
493	}	584	}
494		585
		586	static void ecb_cold
495	static void etp_end_thread (void)	587	etp_end_thread (void)
496	{	588	{
497	eio_req *req = calloc (1, sizeof (eio_req));	589	eio_req req = calloc (1, sizeof (eio_req)); / will be freed by worker */
498		590
499	req->type = -1;	591	req->type = -1;
500	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	592	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
501		593
502	X_LOCK (reqlock);	594	X_LOCK (reqlock);
…		…
507	X_LOCK (wrklock);	599	X_LOCK (wrklock);
508	--started;	600	--started;
509	X_UNLOCK (wrklock);	601	X_UNLOCK (wrklock);
510	}	602	}
511		603
512	static int etp_poll (void)	604	static int
		605	etp_poll (void)
513	{	606	{
514	unsigned int maxreqs;	607	unsigned int maxreqs;
515	unsigned int maxtime;	608	unsigned int maxtime;
516	struct timeval tv_start, tv_now;	609	struct timeval tv_start, tv_now;
517		610
…		…
547		640
548	X_LOCK (reqlock);	641	X_LOCK (reqlock);
549	--nreqs;	642	--nreqs;
550	X_UNLOCK (reqlock);	643	X_UNLOCK (reqlock);
551		644
552	if (expect_false (req->type == EIO_GROUP && req->size))	645	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
553	{	646	{
554	req->int1 = 1; /* mark request as delayed */	647	req->int1 = 1; /* mark request as delayed */
555	continue;	648	continue;
556	}	649	}
557	else	650	else
558	{	651	{
559	int res = ETP_FINISH (req);	652	int res = ETP_FINISH (req);
560	if (expect_false (res))	653	if (ecb_expect_false (res))
561	return res;	654	return res;
562	}	655	}
563		656
564	if (expect_false (maxreqs && !--maxreqs))	657	if (ecb_expect_false (maxreqs && !--maxreqs))
565	break;	658	break;
566		659
567	if (maxtime)	660	if (maxtime)
568	{	661	{
569	gettimeofday (&tv_now, 0);	662	gettimeofday (&tv_now, 0);
…		…
575		668
576	errno = EAGAIN;	669	errno = EAGAIN;
577	return -1;	670	return -1;
578	}	671	}
579		672
		673	static void
580	static void etp_cancel (ETP_REQ *req)	674	etp_cancel (ETP_REQ *req)
581	{	675	{
582	X_LOCK (wrklock);	676	req->cancelled = 1;
583	req->flags \|= EIO_FLAG_CANCELLED;
584	X_UNLOCK (wrklock);
585		677
586	eio_grp_cancel (req);	678	eio_grp_cancel (req);
587	}	679	}
588		680
		681	static void
589	static void etp_submit (ETP_REQ *req)	682	etp_submit (ETP_REQ *req)
590	{	683	{
591	req->pri -= ETP_PRI_MIN;	684	req->pri -= ETP_PRI_MIN;
592		685
593	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	686	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;	687	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595		688
596	if (expect_false (req->type == EIO_GROUP))	689	if (ecb_expect_false (req->type == EIO_GROUP))
597	{	690	{
598	/* I hope this is worth it :/ */	691	/* I hope this is worth it :/ */
599	X_LOCK (reqlock);	692	X_LOCK (reqlock);
600	++nreqs;	693	++nreqs;
601	X_UNLOCK (reqlock);	694	X_UNLOCK (reqlock);
…		…
620		713
621	etp_maybe_start_thread ();	714	etp_maybe_start_thread ();
622	}	715	}
623	}	716	}
624		717
		718	static void ecb_cold
625	static void etp_set_max_poll_time (double nseconds)	719	etp_set_max_poll_time (double nseconds)
626	{	720	{
627	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	721	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628	max_poll_time = nseconds * EIO_TICKS;	722	max_poll_time = nseconds * EIO_TICKS;
629	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	723	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630	}	724	}
631		725
		726	static void ecb_cold
632	static void etp_set_max_poll_reqs (unsigned int maxreqs)	727	etp_set_max_poll_reqs (unsigned int maxreqs)
633	{	728	{
634	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	729	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635	max_poll_reqs = maxreqs;	730	max_poll_reqs = maxreqs;
636	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	731	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637	}	732	}
638		733
		734	static void ecb_cold
639	static void etp_set_max_idle (unsigned int nthreads)	735	etp_set_max_idle (unsigned int nthreads)
640	{	736	{
641	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	737	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642	max_idle = nthreads;	738	max_idle = nthreads;
643	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	739	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644	}	740	}
645		741
		742	static void ecb_cold
646	static void etp_set_idle_timeout (unsigned int seconds)	743	etp_set_idle_timeout (unsigned int seconds)
647	{	744	{
648	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	745	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649	idle_timeout = seconds;	746	idle_timeout = seconds;
650	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	747	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651	}	748	}
652		749
		750	static void ecb_cold
653	static void etp_set_min_parallel (unsigned int nthreads)	751	etp_set_min_parallel (unsigned int nthreads)
654	{	752	{
655	if (wanted < nthreads)	753	if (wanted < nthreads)
656	wanted = nthreads;	754	wanted = nthreads;
657	}	755	}
658		756
		757	static void ecb_cold
659	static void etp_set_max_parallel (unsigned int nthreads)	758	etp_set_max_parallel (unsigned int nthreads)
660	{	759	{
661	if (wanted > nthreads)	760	if (wanted > nthreads)
662	wanted = nthreads;	761	wanted = nthreads;
663		762
664	while (started > wanted)	763	while (started > wanted)
665	etp_end_thread ();	764	etp_end_thread ();
666	}	765	}
667		766
668	/*****************************************************************************/	767	/*****************************************************************************/
669		768
		769	static void
670	static void grp_try_feed (eio_req *grp)	770	grp_try_feed (eio_req *grp)
671	{	771	{
672	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	772	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673	{	773	{
674	grp->flags &= ~EIO_FLAG_GROUPADD;	774	grp->flags &= ~EIO_FLAG_GROUPADD;
675		775
…		…
682	break;	782	break;
683	}	783	}
684	}	784	}
685	}	785	}
686		786
		787	static int
687	static int grp_dec (eio_req *grp)	788	grp_dec (eio_req *grp)
688	{	789	{
689	--grp->size;	790	--grp->size;
690		791
691	/* call feeder, if applicable */	792	/* call feeder, if applicable */
692	grp_try_feed (grp);	793	grp_try_feed (grp);
…		…
696	return eio_finish (grp);	797	return eio_finish (grp);
697	else	798	else
698	return 0;	799	return 0;
699	}	800	}
700		801
		802	static void
701	void eio_destroy (eio_req *req)	803	eio_destroy (eio_req *req)
702	{	804	{
703	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	805	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	806	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705		807
706	EIO_DESTROY (req);	808	EIO_DESTROY (req);
707	}	809	}
708		810
		811	static int
709	static int eio_finish (eio_req *req)	812	eio_finish (eio_req *req)
710	{	813	{
711	int res = EIO_FINISH (req);	814	int res = EIO_FINISH (req);
712		815
713	if (req->grp)	816	if (req->grp)
714	{	817	{
…		…
722	if (grp->grp_first == req)	825	if (grp->grp_first == req)
723	grp->grp_first = req->grp_next;	826	grp->grp_first = req->grp_next;
724		827
725	res2 = grp_dec (grp);	828	res2 = grp_dec (grp);
726		829
727	if (!res && res2)	830	if (!res)
728	res = res2;	831	res = res2;
729	}	832	}
730		833
731	eio_destroy (req);	834	eio_destroy (req);
732		835
733	return res;	836	return res;
734	}	837	}
735		838
		839	void
736	void eio_grp_cancel (eio_req *grp)	840	eio_grp_cancel (eio_req *grp)
737	{	841	{
738	for (grp = grp->grp_first; grp; grp = grp->grp_next)	842	for (grp = grp->grp_first; grp; grp = grp->grp_next)
739	eio_cancel (grp);	843	eio_cancel (grp);
740	}	844	}
741		845
		846	void
742	void eio_cancel (eio_req *req)	847	eio_cancel (eio_req *req)
743	{	848	{
744	etp_cancel (req);	849	etp_cancel (req);
745	}	850	}
746		851
		852	void
747	void eio_submit (eio_req *req)	853	eio_submit (eio_req *req)
748	{	854	{
749	etp_submit (req);	855	etp_submit (req);
750	}	856	}
751		857
752	unsigned int eio_nreqs (void)	858	unsigned int
		859	eio_nreqs (void)
753	{	860	{
754	return etp_nreqs ();	861	return etp_nreqs ();
755	}	862	}
756		863
757	unsigned int eio_nready (void)	864	unsigned int
		865	eio_nready (void)
758	{	866	{
759	return etp_nready ();	867	return etp_nready ();
760	}	868	}
761		869
762	unsigned int eio_npending (void)	870	unsigned int
		871	eio_npending (void)
763	{	872	{
764	return etp_npending ();	873	return etp_npending ();
765	}	874	}
766		875
767	unsigned int eio_nthreads (void)	876	unsigned int ecb_cold
		877	eio_nthreads (void)
768	{	878	{
769	return etp_nthreads ();	879	return etp_nthreads ();
770	}	880	}
771		881
		882	void ecb_cold
772	void eio_set_max_poll_time (double nseconds)	883	eio_set_max_poll_time (double nseconds)
773	{	884	{
774	etp_set_max_poll_time (nseconds);	885	etp_set_max_poll_time (nseconds);
775	}	886	}
776		887
		888	void ecb_cold
777	void eio_set_max_poll_reqs (unsigned int maxreqs)	889	eio_set_max_poll_reqs (unsigned int maxreqs)
778	{	890	{
779	etp_set_max_poll_reqs (maxreqs);	891	etp_set_max_poll_reqs (maxreqs);
780	}	892	}
781		893
		894	void ecb_cold
782	void eio_set_max_idle (unsigned int nthreads)	895	eio_set_max_idle (unsigned int nthreads)
783	{	896	{
784	etp_set_max_idle (nthreads);	897	etp_set_max_idle (nthreads);
785	}	898	}
786		899
		900	void ecb_cold
787	void eio_set_idle_timeout (unsigned int seconds)	901	eio_set_idle_timeout (unsigned int seconds)
788	{	902	{
789	etp_set_idle_timeout (seconds);	903	etp_set_idle_timeout (seconds);
790	}	904	}
791		905
		906	void ecb_cold
792	void eio_set_min_parallel (unsigned int nthreads)	907	eio_set_min_parallel (unsigned int nthreads)
793	{	908	{
794	etp_set_min_parallel (nthreads);	909	etp_set_min_parallel (nthreads);
795	}	910	}
796		911
		912	void ecb_cold
797	void eio_set_max_parallel (unsigned int nthreads)	913	eio_set_max_parallel (unsigned int nthreads)
798	{	914	{
799	etp_set_max_parallel (nthreads);	915	etp_set_max_parallel (nthreads);
800	}	916	}
801		917
802	int eio_poll (void)	918	int eio_poll (void)
…		…
811	# undef pread	927	# undef pread
812	# undef pwrite	928	# undef pwrite
813	# define pread eio__pread	929	# define pread eio__pread
814	# define pwrite eio__pwrite	930	# define pwrite eio__pwrite
815		931
816	static ssize_t	932	static eio_ssize_t
817	eio__pread (int fd, void *buf, size_t count, off_t offset)	933	eio__pread (int fd, void *buf, size_t count, off_t offset)
818	{	934	{
819	ssize_t res;	935	eio_ssize_t res;
820	off_t ooffset;	936	off_t ooffset;
821		937
822	X_LOCK (preadwritelock);	938	X_LOCK (preadwritelock);
823	ooffset = lseek (fd, 0, SEEK_CUR);	939	ooffset = lseek (fd, 0, SEEK_CUR);
824	lseek (fd, offset, SEEK_SET);	940	lseek (fd, offset, SEEK_SET);
…		…
827	X_UNLOCK (preadwritelock);	943	X_UNLOCK (preadwritelock);
828		944
829	return res;	945	return res;
830	}	946	}
831		947
832	static ssize_t	948	static eio_ssize_t
833	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	949	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834	{	950	{
835	ssize_t res;	951	eio_ssize_t res;
836	off_t ooffset;	952	off_t ooffset;
837		953
838	X_LOCK (preadwritelock);	954	X_LOCK (preadwritelock);
839	ooffset = lseek (fd, 0, SEEK_CUR);	955	ooffset = lseek (fd, 0, SEEK_CUR);
840	lseek (fd, offset, SEEK_SET);	956	lseek (fd, offset, SEEK_SET);
…		…
872	#ifndef HAVE_FUTIMES	988	#ifndef HAVE_FUTIMES
873		989
874	# undef futimes	990	# undef futimes
875	# define futimes(fd,times) eio__futimes (fd, times)	991	# define futimes(fd,times) eio__futimes (fd, times)
876		992
		993	static int
877	static int eio__futimes (int fd, const struct timeval tv[2])	994	eio__futimes (int fd, const struct timeval tv[2])
878	{	995	{
879	errno = ENOSYS;	996	errno = ENOSYS;
880	return -1;	997	return -1;
881	}	998	}
882		999
…		…
885	#if !HAVE_FDATASYNC	1002	#if !HAVE_FDATASYNC
886	# undef fdatasync	1003	# undef fdatasync
887	# define fdatasync(fd) fsync (fd)	1004	# define fdatasync(fd) fsync (fd)
888	#endif	1005	#endif
889		1006
		1007	static int
		1008	eio__syncfs (int fd)
		1009	{
		1010	int res;
		1011
		1012	#if HAVE_SYS_SYNCFS
		1013	res = (int)syscall (__NR_syncfs, (int)(fd));
		1014	#else
		1015	res = -1;
		1016	errno = ENOSYS;
		1017	#endif
		1018
		1019	if (res < 0 && errno == ENOSYS && fd >= 0)
		1020	sync ();
		1021
		1022	return res;
		1023	}
		1024
890	/* sync_file_range always needs emulation */	1025	/* sync_file_range always needs emulation */
891	int	1026	static int
892	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	1027	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893	{	1028	{
894	#if HAVE_SYNC_FILE_RANGE	1029	#if HAVE_SYNC_FILE_RANGE
895	int res;	1030	int res;
896		1031
…		…
913	/* even though we could play tricks with the flags, it's better to always	1048	/* 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 */	1049	* call fdatasync, as that matches the expectation of its users best */
915	return fdatasync (fd);	1050	return fdatasync (fd);
916	}	1051	}
917		1052
		1053	static int
		1054	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		1055	{
		1056	#if HAVE_FALLOCATE
		1057	return fallocate (fd, mode, offset, len);
		1058	#else
		1059	errno = ENOSYS;
		1060	return -1;
		1061	#endif
		1062	}
		1063
918	#if !HAVE_READAHEAD	1064	#if !HAVE_READAHEAD
919	# undef readahead	1065	# undef readahead
920	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1066	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921		1067
922	static ssize_t	1068	static eio_ssize_t
923	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1069	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
924	{	1070	{
925	size_t todo = count;	1071	size_t todo = count;
926	dBUF;	1072	dBUF;
927		1073
…		…
932	pread (fd, eio_buf, len, offset);	1078	pread (fd, eio_buf, len, offset);
933	offset += len;	1079	offset += len;
934	todo -= len;	1080	todo -= len;
935	}	1081	}
936		1082
		1083	FUBd;
		1084
937	errno = 0;	1085	errno = 0;
938	return count;	1086	return count;
939	}	1087	}
940		1088
941	#endif	1089	#endif
942		1090
943	/* sendfile always needs emulation */	1091	/* sendfile always needs emulation */
944	static ssize_t	1092	static eio_ssize_t
945	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1093	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
946	{	1094	{
		1095	eio_ssize_t written = 0;
947	ssize_t res;	1096	eio_ssize_t res;
948		1097
949	if (!count)	1098	if (!count)
950	return 0;	1099	return 0;
951		1100
		1101	for (;;)
		1102	{
		1103	#ifdef __APPLE__
		1104	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1105	#endif
952	#if HAVE_SENDFILE	1106	#if HAVE_SENDFILE
953	# if __linux	1107	# if __linux
		1108	off_t soffset = offset;
954	res = sendfile (ofd, ifd, &offset, count);	1109	res = sendfile (ofd, ifd, &soffset, count);
955		1110
956	# elif __FreeBSD__	1111	# elif __FreeBSD__
957	/*	1112	/*
958	* Of course, the freebsd sendfile is a dire hack with no thoughts	1113	* Of course, the freebsd sendfile is a dire hack with no thoughts
959	* wasted on making it similar to other I/O functions.	1114	* wasted on making it similar to other I/O functions.
960	*/	1115	*/
961	{
962	off_t sbytes;	1116	off_t sbytes;
963	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1117	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
964		1118
965	#if 0 /* according to the manpage, this is correct, but broken behaviour */	1119	#if 0 /* according to the manpage, this is correct, but broken behaviour */
966	/* freebsd' sendfile will return 0 on success */	1120	/* freebsd' sendfile will return 0 on success */
967	/* freebsd 8 documents it as only setting sbytes on EINTR and EAGAIN, but /	1121	/* freebsd 8 documents it as only setting sbytes on EINTR and EAGAIN, but /
968	/* not on e.g. EIO or EPIPE - sounds broken */	1122	/* not on e.g. EIO or EPIPE - sounds broken */
969	if ((res < 0 && (errno == EAGAIN \|\| errno == EINTR) && sbytes) \|\| res == 0)	1123	if ((res < 0 && (errno == EAGAIN \|\| errno == EINTR) && sbytes) \|\| res == 0)
970	res = sbytes;	1124	res = sbytes;
971	#endif	1125	#endif
972		1126
973	/* according to source inspection, this is correct, and useful behaviour */	1127	/* according to source inspection, this is correct, and useful behaviour */
974	if (sbytes)	1128	if (sbytes)
975	res = sbytes;	1129	res = sbytes;
976	}
977		1130
978	# elif defined (__APPLE__)	1131	# elif defined (__APPLE__)
979
980	{
981	off_t sbytes = count;	1132	off_t sbytes = count;
982	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	1133	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983		1134
984	/* according to the manpage, sbytes is always valid */	1135	/* according to the manpage, sbytes is always valid */
985	if (sbytes)	1136	if (sbytes)
986	res = sbytes;	1137	res = sbytes;
987	}
988		1138
989	# elif __hpux	1139	# elif __hpux
990	res = sendfile (ofd, ifd, offset, count, 0, 0);	1140	res = sendfile (ofd, ifd, offset, count, 0, 0);
991		1141
992	# elif __solaris	1142	# elif __solaris
993	{
994	struct sendfilevec vec;	1143	struct sendfilevec vec;
995	size_t sbytes;	1144	size_t sbytes;
996		1145
997	vec.sfv_fd = ifd;	1146	vec.sfv_fd = ifd;
998	vec.sfv_flag = 0;	1147	vec.sfv_flag = 0;
999	vec.sfv_off = offset;	1148	vec.sfv_off = offset;
1000	vec.sfv_len = count;	1149	vec.sfv_len = count;
1001		1150
1002	res = sendfilev (ofd, &vec, 1, &sbytes);	1151	res = sendfilev (ofd, &vec, 1, &sbytes);
1003		1152
1004	if (res < 0 && sbytes)	1153	if (res < 0 && sbytes)
1005	res = sbytes;	1154	res = sbytes;
1006	}
1007		1155
1008	# endif	1156	# endif
1009		1157
1010	#elif defined (_WIN32)	1158	#elif defined (_WIN32) && 0
1011
1012	/* does not work, just for documentation of what would need to be done */	1159	/* does not work, just for documentation of what would need to be done */
1013	{	1160	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1161	/* libeio guarantees that the file offset does not change, and windows */
		1162	/* has no way to get an independent handle to the same file description */
1014	HANDLE h = TO_SOCKET (ifd);	1163	HANDLE h = TO_SOCKET (ifd);
1015	SetFilePointer (h, offset, 0, FILE_BEGIN);	1164	SetFilePointer (h, offset, 0, FILE_BEGIN);
1016	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);	1165	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1017	}
1018		1166
1019	#else	1167	#else
1020	res = -1;	1168	res = -1;
1021	errno = ENOSYS;	1169	errno = ENOSYS;
1022	#endif	1170	#endif
1023		1171
		1172	/* we assume sendfile can copy at least 128mb in one go */
		1173	if (res <= 128 * 1024 * 1024)
		1174	{
		1175	if (res > 0)
		1176	written += res;
		1177
		1178	if (written)
		1179	return written;
		1180
		1181	break;
		1182	}
		1183	else
		1184	{
		1185	/* if we requested more, then probably the kernel was lazy */
		1186	written += res;
		1187	offset += res;
		1188	count -= res;
		1189
		1190	if (!count)
		1191	return written;
		1192	}
		1193	}
		1194
1024	if (res < 0	1195	if (res < 0
1025	&& (errno == ENOSYS \|\| errno == EINVAL \|\| errno == ENOTSOCK	1196	&& (errno == ENOSYS \|\| errno == EINVAL \|\| errno == ENOTSOCK
1026	/* BSDs */	1197	/* BSDs */
1027	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1198	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1028	\|\| errno == ENOTSUP	1199	\|\| errno == ENOTSUP
1029	#endif	1200	#endif
		1201	#ifdef EOPNOTSUPP /* windows */
1030	\|\| errno == EOPNOTSUPP /* BSDs */	1202	\|\| errno == EOPNOTSUPP /* BSDs */
		1203	#endif
1031	#if __solaris	1204	#if __solaris
1032	\|\| errno == EAFNOSUPPORT \|\| errno == EPROTOTYPE	1205	\|\| errno == EAFNOSUPPORT \|\| errno == EPROTOTYPE
1033	#endif	1206	#endif
1034	)	1207	)
1035	)	1208	)
…		…
1039		1212
1040	res = 0;	1213	res = 0;
1041		1214
1042	while (count)	1215	while (count)
1043	{	1216	{
1044	ssize_t cnt;	1217	eio_ssize_t cnt;
1045		1218
1046	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1219	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1047		1220
1048	if (cnt <= 0)	1221	if (cnt <= 0)
1049	{	1222	{
…		…
1061		1234
1062	offset += cnt;	1235	offset += cnt;
1063	res += cnt;	1236	res += cnt;
1064	count -= cnt;	1237	count -= cnt;
1065	}	1238	}
		1239
		1240	FUBd;
1066	}	1241	}
1067		1242
1068	return res;	1243	return res;
		1244	}
		1245
		1246	#ifdef PAGESIZE
		1247	# define eio_pagesize() PAGESIZE
		1248	#else
		1249	static intptr_t
		1250	eio_pagesize (void)
		1251	{
		1252	static intptr_t page;
		1253
		1254	if (!page)
		1255	page = sysconf (_SC_PAGESIZE);
		1256
		1257	return page;
		1258	}
		1259	#endif
		1260
		1261	static void
		1262	eio_page_align (void *addr, size_t length)
		1263	{
		1264	intptr_t mask = eio_pagesize () - 1;
		1265
		1266	/* round down addr */
		1267	intptr_t adj = mask & (intptr_t)*addr;
		1268
		1269	addr = (void )((intptr_t)*addr - adj);
		1270	*length += adj;
		1271
		1272	/* round up length */
		1273	length = (length + mask) & ~mask;
		1274	}
		1275
		1276	#if !_POSIX_MEMLOCK
		1277	# define eio__mlockall(a) EIO_ENOSYS ()
		1278	#else
		1279
		1280	static int
		1281	eio__mlockall (int flags)
		1282	{
		1283	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1284	extern int mallopt (int, int);
		1285	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1286	#endif
		1287
		1288	if (EIO_MCL_CURRENT != MCL_CURRENT
		1289	\|\| EIO_MCL_FUTURE != MCL_FUTURE)
		1290	{
		1291	flags = 0
		1292	\| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1293	\| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1294	}
		1295
		1296	return mlockall (flags);
		1297	}
		1298	#endif
		1299
		1300	#if !_POSIX_MEMLOCK_RANGE
		1301	# define eio__mlock(a,b) EIO_ENOSYS ()
		1302	#else
		1303
		1304	static int
		1305	eio__mlock (void *addr, size_t length)
		1306	{
		1307	eio_page_align (&addr, &length);
		1308
		1309	return mlock (addr, length);
		1310	}
		1311
		1312	#endif
		1313
		1314	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1315	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1316	#else
		1317
		1318	static int
		1319	eio__msync (void *mem, size_t len, int flags)
		1320	{
		1321	eio_page_align (&mem, &len);
		1322
		1323	if (EIO_MS_ASYNC != MS_SYNC
		1324	\|\| EIO_MS_INVALIDATE != MS_INVALIDATE
		1325	\|\| EIO_MS_SYNC != MS_SYNC)
		1326	{
		1327	flags = 0
		1328	\| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1329	\| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1330	\| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1331	}
		1332
		1333	return msync (mem, len, flags);
		1334	}
		1335
		1336	#endif
		1337
		1338	static int
		1339	eio__mtouch (eio_req *req)
		1340	{
		1341	void *mem = req->ptr2;
		1342	size_t len = req->size;
		1343	int flags = req->int1;
		1344
		1345	eio_page_align (&mem, &len);
		1346
		1347	{
		1348	intptr_t addr = (intptr_t)mem;
		1349	intptr_t end = addr + len;
		1350	intptr_t page = eio_pagesize ();
		1351
		1352	if (addr < end)
		1353	if (flags & EIO_MT_MODIFY) /* modify */
		1354	do { ((volatile sig_atomic_t )addr) \|= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1355	else
		1356	do { ((volatile sig_atomic_t )addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1357	}
		1358
		1359	return 0;
		1360	}
		1361
		1362	/*****************************************************************************/
		1363	/* requests implemented outside eio_execute, because they are so large */
		1364
		1365	/* copies some absolute path to tmpbuf */
		1366	static char *
		1367	eio__getwd (struct tmpbuf *tmpbuf, eio_wd wd)
		1368	{
		1369	if (wd == EIO_CWD)
		1370	return getcwd (tmpbuf->ptr, PATH_MAX);
		1371
		1372	#if HAVE_AT
		1373	abort (); /TODO/
		1374	#else
		1375	strcpy (tmpbuf->ptr, wd);
		1376	#endif
		1377	return tmpbuf->ptr;
		1378	}
		1379
		1380	/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
		1381	static int
		1382	eio__realpath (struct tmpbuf tmpbuf, eio_wd wd, const char path)
		1383	{
		1384	const char *rel = path;
		1385	char *res;
		1386	char tmp1, tmp2;
		1387	#if SYMLOOP_MAX > 32
		1388	int symlinks = SYMLOOP_MAX;
		1389	#else
		1390	int symlinks = 32;
		1391	#endif
		1392
		1393	/D/ /TODO: wd ignored /
		1394
		1395	errno = EINVAL;
		1396	if (!rel)
		1397	return -1;
		1398
		1399	errno = ENOENT;
		1400	if (!*rel)
		1401	return -1;
		1402
		1403	res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
		1404	tmp1 = res + PATH_MAX;
		1405	tmp2 = tmp1 + PATH_MAX;
		1406
		1407	#if 0 /* disabled, the musl way to do things is just too racy */
		1408	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1409	/* on linux we may be able to ask the kernel */
		1410	{
		1411	int fd = open (rel, O_RDONLY \| O_NONBLOCK \| O_NOCTTY \| O_NOATIME);
		1412
		1413	if (fd >= 0)
		1414	{
		1415	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1416	req->result = readlink (tmp1, res, PATH_MAX);
		1417	close (fd);
		1418
		1419	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1420
		1421	if (req->result > 0)
		1422	goto done;
		1423	}
		1424	else if (errno == ELOOP \|\| errno == ENAMETOOLONG \|\| errno == ENOENT \|\| errno == ENOTDIR \|\| errno == EIO)
		1425	return;
		1426	}
		1427	#endif
		1428	#endif
		1429
		1430	if (*rel != '/')
		1431	{
		1432	if (!eio__getwd (tmpbuf, wd))
		1433	return -1;
		1434
		1435	if (res [1]) /* only use if not / */
		1436	res += strlen (res);
		1437	}
		1438
		1439	while (*rel)
		1440	{
		1441	eio_ssize_t len, linklen;
		1442	const char *beg = rel;
		1443
		1444	while (rel && rel != '/')
		1445	++rel;
		1446
		1447	len = rel - beg;
		1448
		1449	if (!len) /* skip slashes */
		1450	{
		1451	++rel;
		1452	continue;
		1453	}
		1454
		1455	if (beg [0] == '.')
		1456	{
		1457	if (len == 1)
		1458	continue; /* . - nop */
		1459
		1460	if (beg [1] == '.' && len == 2)
		1461	{
		1462	/* .. - back up one component, if possible */
		1463
		1464	while (res != tmpbuf->ptr)
		1465	if (*--res == '/')
		1466	break;
		1467
		1468	continue;
		1469	}
		1470	}
		1471
		1472	errno = ENAMETOOLONG;
		1473	if (res + 1 + len + 1 >= tmp1)
		1474	return;
		1475
		1476	/* copy one component */
		1477	*res = '/';
		1478	memcpy (res + 1, beg, len);
		1479
		1480	/* zero-terminate, for readlink */
		1481	res [len + 1] = 0;
		1482
		1483	/* now check if it's a symlink */
		1484	linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
		1485
		1486	if (linklen < 0)
		1487	{
		1488	if (errno != EINVAL)
		1489	return -1;
		1490
		1491	/* it's a normal directory. hopefully */
		1492	res += len + 1;
		1493	}
		1494	else
		1495	{
		1496	/* yay, it was a symlink - build new path in tmp2 */
		1497	int rellen = strlen (rel);
		1498
		1499	errno = ENAMETOOLONG;
		1500	if (linklen + 1 + rellen >= PATH_MAX)
		1501	return -1;
		1502
		1503	errno = ELOOP;
		1504	if (!--symlinks)
		1505	return -1;
		1506
		1507	if (*tmp1 == '/')
		1508	res = tmpbuf->ptr; /* symlink resolves to an absolute path */
		1509
		1510	/* we need to be careful, as rel might point into tmp2 already */
		1511	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1512	tmp2 [linklen] = '/';
		1513	memcpy (tmp2, tmp1, linklen);
		1514
		1515	rel = tmp2;
		1516	}
		1517	}
		1518
		1519	/* special case for the lone root path */
		1520	if (res == tmpbuf->ptr)
		1521	*res++ = '/';
		1522
		1523	return res - (char *)tmpbuf->ptr;
1069	}	1524	}
1070		1525
1071	static signed char	1526	static signed char
1072	eio_dent_cmp (const eio_dirent a, const eio_dirent b)	1527	eio_dent_cmp (const eio_dirent a, const eio_dirent b)
1073	{	1528	{
…		…
1081		1536
1082	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1537	#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 */	1538	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1084		1539
1085	static void	1540	static void
1086	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1541	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1087	{	1542	{
1088	unsigned char bits [9 + sizeof (ino_t) * 8];	1543	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1089	unsigned char *bit = bits;	1544	unsigned char *bit = bits;
1090		1545
1091	assert (CHAR_BIT == 8);	1546	assert (CHAR_BIT == 8);
1092	assert (sizeof (eio_dirent) * 8 < 256);	1547	assert (sizeof (eio_dirent) * 8 < 256);
1093	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1548	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1095		1550
1096	if (size <= EIO_SORT_FAST)	1551	if (size <= EIO_SORT_FAST)
1097	return;	1552	return;
1098		1553
1099	/* first prepare an array of bits to test in our radix sort */	1554	/* 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 */	1555	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1101	/* inode_bits must contain all inodes ORed together */	1556	/* inode_bits must contain all inodes ORed together */
1102	/* which is used to skip bits that are 0 everywhere, which is very common */	1557	/* which is used to skip bits that are 0 everywhere, which is very common */
1103	{	1558	{
1104	ino_t endianness;	1559	eio_ino_t endianness;
1105	int i, j;	1560	int i, j;
1106		1561
1107	/* we store the byte offset of byte n into byte n of "endianness" */	1562	/* we store the byte offset of byte n into byte n of "endianness" */
1108	for (i = 0; i < sizeof (ino_t); ++i)	1563	for (i = 0; i < sizeof (eio_ino_t); ++i)
1109	((unsigned char *)&endianness)[i] = i;	1564	((unsigned char *)&endianness)[i] = i;
1110		1565
1111	*bit++ = 0;	1566	*bit++ = 0;
1112		1567
1113	for (i = 0; i < sizeof (ino_t); ++i)	1568	for (i = 0; i < sizeof (eio_ino_t); ++i)
1114	{	1569	{
1115	/* shifting off the byte offsets out of "endianness" */	1570	/* shifting off the byte offsets out of "endianness" */
1116	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1571	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1117	endianness >>= 8;	1572	endianness >>= 8;
1118		1573
1119	for (j = 0; j < 8; ++j)	1574	for (j = 0; j < 8; ++j)
1120	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1575	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1121	*bit++ = offs + j;	1576	*bit++ = offs + j;
1122	}	1577	}
1123		1578
1124	for (j = 0; j < 8; ++j)	1579	for (j = 0; j < 8; ++j)
1125	if (score_bits & (1 << j))	1580	if (score_bits & (1 << j))
1126	bit++ = offsetof (eio_dirent, score) 8 + j;	1581	bit++ = offsetof (eio_dirent, score) 8 + j;
1127	}	1582	}
1128		1583
1129	/* now actually do the sorting (a variant of MSD radix sort) */	1584	/* now actually do the sorting (a variant of MSD radix sort) */
1130	{	1585	{
1131	eio_dirent base_stk [9 + sizeof (ino_t) 8], *base;	1586	eio_dirent base_stk [9 + sizeof (eio_ino_t) 8], *base;
1132	eio_dirent end_stk [9 + sizeof (ino_t) 8], *end;	1587	eio_dirent end_stk [9 + sizeof (eio_ino_t) 8], *end;
1133	unsigned char bit_stk [9 + sizeof (ino_t) 8];	1588	unsigned char bit_stk [9 + sizeof (eio_ino_t) 8];
1134	int stk_idx = 0;	1589	int stk_idx = 0;
1135		1590
1136	base_stk [stk_idx] = dents;	1591	base_stk [stk_idx] = dents;
1137	end_stk [stk_idx] = dents + size;	1592	end_stk [stk_idx] = dents + size;
1138	bit_stk [stk_idx] = bit - 1;	1593	bit_stk [stk_idx] = bit - 1;
…		…
1217	}	1672	}
1218	}	1673	}
1219	}	1674	}
1220		1675
1221	static void	1676	static void
1222	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1677	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1223	{	1678	{
1224	if (size <= 1)	1679	if (size <= 1)
1225	return; /* our insertion sort relies on size > 0 */	1680	return; /* our insertion sort relies on size > 0 */
1226		1681
1227	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1682	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1233	eio_dent_insertion_sort (dents, size);	1688	eio_dent_insertion_sort (dents, size);
1234	}	1689	}
1235		1690
1236	/* read a full directory */	1691	/* read a full directory */
1237	static void	1692	static void
1238	eio__scandir (eio_req req, etp_worker self)	1693	eio__scandir (eio_req *req)
1239	{	1694	{
1240	DIR *dirp;
1241	EIO_STRUCT_DIRENT *entp;
1242	char name, names;	1695	char name, names;
1243	int namesalloc = 4096;	1696	int namesalloc = 4096 - sizeof (void ) 4;
1244	int namesoffs = 0;	1697	int namesoffs = 0;
1245	int flags = req->int1;	1698	int flags = req->int1;
1246	eio_dirent *dents = 0;	1699	eio_dirent *dents = 0;
1247	int dentalloc = 128;	1700	int dentalloc = 128;
1248	int dentoffs = 0;	1701	int dentoffs = 0;
1249	ino_t inode_bits = 0;	1702	eio_ino_t inode_bits = 0;
		1703	#ifdef _WIN32
		1704	HANDLE dirp;
		1705	WIN32_FIND_DATA entp;
		1706	#else
		1707	DIR *dirp;
		1708	EIO_STRUCT_DIRENT *entp;
		1709	#endif
1250		1710
1251	req->result = -1;	1711	req->result = -1;
1252		1712
1253	if (!(flags & EIO_READDIR_DENTS))	1713	if (!(flags & EIO_READDIR_DENTS))
1254	flags &= ~(EIO_READDIR_DIRS_FIRST \| EIO_READDIR_STAT_ORDER);	1714	flags &= ~(EIO_READDIR_DIRS_FIRST \| EIO_READDIR_STAT_ORDER);
1255		1715
1256	X_LOCK (wrklock);	1716	#ifdef _WIN32
1257	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1717	{
		1718	int len = strlen ((const char *)req->ptr1);
		1719	char *path = malloc (MAX_PATH);
		1720	const char *fmt;
		1721
		1722	if (!len)
		1723	fmt = "./*";
		1724	else if (((const char )req->ptr1)[len - 1] == '/' \|\| ((const char )req->ptr1)[len - 1] == '\\')
		1725	fmt = "%s*";
		1726	else
		1727	fmt = "%s/*";
		1728
		1729	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1730	dirp = FindFirstFile (path, &entp);
		1731	free (path);
		1732
		1733	if (dirp == INVALID_HANDLE_VALUE)
		1734	{
		1735	/* should steal _dosmaperr */
		1736	switch (GetLastError ())
		1737	{
		1738	case ERROR_FILE_NOT_FOUND:
		1739	req->result = 0;
		1740	break;
		1741
		1742	case ERROR_INVALID_NAME:
		1743	case ERROR_PATH_NOT_FOUND:
		1744	case ERROR_NO_MORE_FILES:
		1745	errno = ENOENT;
		1746	break;
		1747
		1748	case ERROR_NOT_ENOUGH_MEMORY:
		1749	errno = ENOMEM;
		1750	break;
		1751
		1752	default:
		1753	errno = EINVAL;
		1754	break;
		1755	}
		1756
		1757	return;
		1758	}
		1759	}
		1760	#else
1258	self->dirp = dirp = opendir (req->ptr1);	1761	dirp = opendir (req->ptr1);
		1762
		1763	if (!dirp)
		1764	return;
		1765	#endif
		1766
		1767	if (req->flags & EIO_FLAG_PTR1_FREE)
		1768	free (req->ptr1);
1259		1769
1260	req->flags \|= EIO_FLAG_PTR1_FREE \| EIO_FLAG_PTR2_FREE;	1770	req->flags \|= EIO_FLAG_PTR1_FREE \| EIO_FLAG_PTR2_FREE;
1261	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1771	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1262	req->ptr2 = names = malloc (namesalloc);	1772	req->ptr2 = names = malloc (namesalloc);
1263	X_UNLOCK (wrklock);
1264		1773
1265	if (dirp && names && (!flags \|\| dents))	1774	if (!names \|\| (flags && !dents))
		1775	return;
		1776
1266	for (;;)	1777	for (;;)
1267	{	1778	{
		1779	int done;
		1780
		1781	#ifdef _WIN32
		1782	done = !dirp;
		1783	#else
1268	errno = 0;	1784	errno = 0;
1269	entp = readdir (dirp);	1785	entp = readdir (dirp);
		1786	done = !entp;
		1787	#endif
1270		1788
1271	if (!entp)	1789	if (done)
1272	{	1790	{
		1791	#ifndef _WIN32
		1792	int old_errno = errno;
		1793	closedir (dirp);
		1794	errno = old_errno;
		1795
1273	if (errno)	1796	if (errno)
1274	break;	1797	break;
		1798	#endif
1275		1799
1276	/* sort etc. */	1800	/* sort etc. */
1277	req->int1 = flags;	1801	req->int1 = flags;
1278	req->result = dentoffs;	1802	req->result = dentoffs;
1279		1803
1280	if (flags & EIO_READDIR_STAT_ORDER)	1804	if (flags & EIO_READDIR_STAT_ORDER)
1281	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);	1805	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1282	else if (flags & EIO_READDIR_DIRS_FIRST)	1806	else if (flags & EIO_READDIR_DIRS_FIRST)
1283	if (flags & EIO_READDIR_FOUND_UNKNOWN)	1807	if (flags & EIO_READDIR_FOUND_UNKNOWN)
1284	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */	1808	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1285	else	1809	else
		1810	{
		1811	/* in this case, all is known, and we just put dirs first and sort them */
		1812	eio_dirent *oth = dents + dentoffs;
		1813	eio_dirent *dir = dents;
		1814
		1815	/* now partition dirs to the front, and non-dirs to the back */
		1816	/* by walking from both sides and swapping if necessary */
		1817	while (oth > dir)
		1818	{
		1819	if (dir->type == EIO_DT_DIR)
		1820	++dir;
		1821	else if ((--oth)->type == EIO_DT_DIR)
		1822	{
		1823	eio_dirent tmp = dir; dir = oth; oth = tmp;
		1824
		1825	++dir;
		1826	}
		1827	}
		1828
		1829	/* now sort the dirs only (dirs all have the same score) */
		1830	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1831	}
		1832
		1833	break;
		1834	}
		1835
		1836	/* now add the entry to our list(s) */
		1837	name = D_NAME (entp);
		1838
		1839	/* skip . and .. entries */
		1840	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))
		1841	{
		1842	int len = D_NAMLEN (entp) + 1;
		1843
		1844	while (ecb_expect_false (namesoffs + len > namesalloc))
		1845	{
		1846	namesalloc *= 2;
		1847	req->ptr2 = names = realloc (names, namesalloc);
		1848
		1849	if (!names)
		1850	break;
		1851	}
		1852
		1853	memcpy (names + namesoffs, name, len);
		1854
		1855	if (dents)
		1856	{
		1857	struct eio_dirent *ent;
		1858
		1859	if (ecb_expect_false (dentoffs == dentalloc))
1286	{	1860	{
1287	/* in this case, all is known, and we just put dirs first and sort them */	1861	dentalloc *= 2;
		1862	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1863
		1864	if (!dents)
		1865	break;
		1866	}
		1867
1288	eio_dirent *oth = dents + dentoffs;	1868	ent = dents + dentoffs;
1289	eio_dirent *dir = dents;
1290		1869
1291	/* now partition dirs to the front, and non-dirs to the back */	1870	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1292	/* by walking from both sides and swapping if necessary */	1871	ent->namelen = len - 1;
1293	while (oth > dir)	1872	ent->inode = D_INO (entp);
		1873
		1874	inode_bits \|= ent->inode;
		1875
		1876	switch (D_TYPE (entp))
		1877	{
		1878	default:
		1879	ent->type = EIO_DT_UNKNOWN;
		1880	flags \|= EIO_READDIR_FOUND_UNKNOWN;
		1881	break;
		1882
		1883	#ifdef DT_FIFO
		1884	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1885	#endif
		1886	#ifdef DT_CHR
		1887	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1888	#endif
		1889	#ifdef DT_MPC
		1890	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1891	#endif
		1892	#ifdef DT_DIR
		1893	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1894	#endif
		1895	#ifdef DT_NAM
		1896	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1897	#endif
		1898	#ifdef DT_BLK
		1899	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1900	#endif
		1901	#ifdef DT_MPB
		1902	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1903	#endif
		1904	#ifdef DT_REG
		1905	case DT_REG: ent->type = EIO_DT_REG; break;
		1906	#endif
		1907	#ifdef DT_NWK
		1908	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1909	#endif
		1910	#ifdef DT_CMP
		1911	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1912	#endif
		1913	#ifdef DT_LNK
		1914	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1915	#endif
		1916	#ifdef DT_SOCK
		1917	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1918	#endif
		1919	#ifdef DT_DOOR
		1920	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1921	#endif
		1922	#ifdef DT_WHT
		1923	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1924	#endif
		1925	}
		1926
		1927	ent->score = 7;
		1928
		1929	if (flags & EIO_READDIR_DIRS_FIRST)
		1930	{
		1931	if (ent->type == EIO_DT_UNKNOWN)
1294	{	1932	{
1295	if (dir->type == EIO_DT_DIR)	1933	if (name == '.') / leading dots are likely directories, and, in any case, rare */
1296	++dir;	1934	ent->score = 1;
1297	else if ((--oth)->type == EIO_DT_DIR)	1935	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1298	{	1936	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1299	eio_dirent tmp = dir; dir = oth; oth = tmp;
1300
1301	++dir;
1302	}
1303	}	1937	}
1304		1938	else if (ent->type == EIO_DT_DIR)
1305	/* now sort the dirs only (dirs all have the same score) */	1939	ent->score = 0;
1306	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1307	}	1940	}
1308
1309	break;
1310	}
1311
1312	/* now add the entry to our list(s) */
1313	name = entp->d_name;
1314
1315	/* skip . and .. entries */
1316	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))
1317	{
1318	int len = D_NAMLEN (entp) + 1;
1319
1320	while (expect_false (namesoffs + len > namesalloc))
1321	{
1322	namesalloc *= 2;
1323	X_LOCK (wrklock);
1324	req->ptr2 = names = realloc (names, namesalloc);
1325	X_UNLOCK (wrklock);
1326
1327	if (!names)
1328	break;
1329	}	1941	}
1330		1942
1331	memcpy (names + namesoffs, name, len);
1332
1333	if (dents)
1334	{
1335	struct eio_dirent *ent;
1336
1337	if (expect_false (dentoffs == dentalloc))
1338	{
1339	dentalloc *= 2;
1340	X_LOCK (wrklock);
1341	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1342	X_UNLOCK (wrklock);
1343
1344	if (!dents)
1345	break;
1346	}
1347
1348	ent = dents + dentoffs;
1349
1350	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1351	ent->namelen = len - 1;
1352	ent->inode = D_INO (entp);
1353
1354	inode_bits \|= ent->inode;
1355
1356	switch (D_TYPE (entp))
1357	{
1358	default:
1359	ent->type = EIO_DT_UNKNOWN;
1360	flags \|= EIO_READDIR_FOUND_UNKNOWN;
1361	break;
1362
1363	#ifdef DT_FIFO
1364	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1365	#endif
1366	#ifdef DT_CHR
1367	case DT_CHR: ent->type = EIO_DT_CHR; break;
1368	#endif
1369	#ifdef DT_MPC
1370	case DT_MPC: ent->type = EIO_DT_MPC; break;
1371	#endif
1372	#ifdef DT_DIR
1373	case DT_DIR: ent->type = EIO_DT_DIR; break;
1374	#endif
1375	#ifdef DT_NAM
1376	case DT_NAM: ent->type = EIO_DT_NAM; break;
1377	#endif
1378	#ifdef DT_BLK
1379	case DT_BLK: ent->type = EIO_DT_BLK; break;
1380	#endif
1381	#ifdef DT_MPB
1382	case DT_MPB: ent->type = EIO_DT_MPB; break;
1383	#endif
1384	#ifdef DT_REG
1385	case DT_REG: ent->type = EIO_DT_REG; break;
1386	#endif
1387	#ifdef DT_NWK
1388	case DT_NWK: ent->type = EIO_DT_NWK; break;
1389	#endif
1390	#ifdef DT_CMP
1391	case DT_CMP: ent->type = EIO_DT_CMP; break;
1392	#endif
1393	#ifdef DT_LNK
1394	case DT_LNK: ent->type = EIO_DT_LNK; break;
1395	#endif
1396	#ifdef DT_SOCK
1397	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1398	#endif
1399	#ifdef DT_DOOR
1400	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1401	#endif
1402	#ifdef DT_WHT
1403	case DT_WHT: ent->type = EIO_DT_WHT; break;
1404	#endif
1405	}
1406
1407	ent->score = 7;
1408
1409	if (flags & EIO_READDIR_DIRS_FIRST)
1410	{
1411	if (ent->type == EIO_DT_UNKNOWN)
1412	{
1413	if (name == '.') / leading dots are likely directories, and, in any case, rare */
1414	ent->score = 1;
1415	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1416	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1417	}
1418	else if (ent->type == EIO_DT_DIR)
1419	ent->score = 0;
1420	}
1421	}
1422
1423	namesoffs += len;	1943	namesoffs += len;
1424	++dentoffs;	1944	++dentoffs;
1425	}	1945	}
1426		1946
1427	if (EIO_CANCELLED (req))	1947	if (EIO_CANCELLED (req))
1428	{	1948	{
1429	errno = ECANCELED;	1949	errno = ECANCELED;
1430	break;	1950	break;
1431	}	1951	}
		1952
		1953	#ifdef _WIN32
		1954	if (!FindNextFile (dirp, &entp))
		1955	{
		1956	FindClose (dirp);
		1957	dirp = 0;
1432	}	1958	}
		1959	#endif
		1960	}
1433	}	1961	}
1434		1962
1435	#ifdef PAGESIZE	1963	/*****************************************************************************/
1436	# define eio_pagesize() PAGESIZE	1964	/* working directory stuff */
		1965
		1966	#if HAVE_AT
		1967
		1968	#define WD2FD(wd) ((wd) ? ((int)(long)(wd)) - 1 : AT_FDCWD)
		1969
		1970	#ifndef O_SEARCH
		1971	# define O_SEARCH O_RDONLY
		1972	#endif
		1973
		1974	eio_wd
		1975	eio_wd_open_sync (eio_wd wd, const char *path)
		1976	{
		1977	int fd = openat (WD2FD (wd), path, O_CLOEXEC \| O_SEARCH \| O_DIRECTORY);
		1978
		1979	return fd >= 0 ? (eio_wd)(long)(fd + 1) : EIO_INVALID_WD;
		1980	}
		1981
		1982	static eio_wd
		1983	eio__wd_open_sync (struct tmpbuf tmpbuf, eio_wd wd, const char path)
		1984	{
		1985	return eio_wd_open_sync (wd, path);
		1986	}
		1987
		1988	void
		1989	eio_wd_close_sync (eio_wd wd)
		1990	{
		1991	int fd = WD2FD (wd);
		1992
		1993	if (fd >= 0)
		1994	close (fd);
		1995	}
		1996
		1997	static int
		1998	eio__truncateat (int dirfd, const char *path, off_t length)
		1999	{
		2000	int fd = openat (dirfd, path, O_WRONLY \| O_CLOEXEC);
		2001	int res;
		2002
		2003	if (fd < 0)
		2004	return fd;
		2005
		2006	res = ftruncate (fd, length);
		2007	close (fd);
		2008	return res;
		2009	}
		2010
		2011	static int
		2012	eio__statvfsat (int dirfd, const char path, struct statvfs buf)
		2013	{
		2014	int fd = openat (dirfd, path, O_SEARCH \| O_CLOEXEC);
		2015	int res;
		2016
		2017	if (fd < 0)
		2018	return fd;
		2019
		2020	res = fstatvfs (fd, buf);
		2021	close (fd);
		2022	return res;
		2023
		2024	}
		2025
1437	#else	2026	#else
1438	static intptr_t
1439	eio_pagesize (void)
1440	{
1441	static intptr_t page;
1442		2027
1443	if (!page)	2028	/* on legacy systems, we represent the working directories simply by their path strings */
1444	page = sysconf (_SC_PAGESIZE);
1445		2029
		2030	static const char *
		2031	wd_expand (struct tmpbuf tmpbuf, eio_wd wd, const char path)
		2032	{
		2033	if (!wd \|\| *path == '/')
1446	return page;	2034	return path;
1447	}
1448	#endif
1449		2035
		2036	{
		2037	int l1 = strlen ((const char *)wd);
		2038	int l2 = strlen (path);
		2039
		2040	char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
		2041
		2042	memcpy (res, wd, l1);
		2043	res [l1] = '/';
		2044	memcpy (res + l1 + 1, path, l2 + 1);
		2045
		2046	return res;
		2047	}
		2048	}
		2049
1450	static void	2050	static eio_wd
1451	eio_page_align (void *addr, size_t length)	2051	eio__wd_open_sync (struct tmpbuf tmpbuf, eio_wd wd, const char path)
1452	{	2052	{
1453	intptr_t mask = eio_pagesize () - 1;	2053	if (path == '/') / absolute paths ignore wd */
1454		2054	path = strdup (path);
1455	/* round down addr */	2055	else if (path [0] == '.' && !path [1]) /* special case '.', as it is common */
1456	intptr_t adj = mask & (intptr_t)*addr;	2056	return wd;
1457		2057	else
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	{	2058	{
1480	flags = 0	2059	int len = eio__realpath (tmpbuf, wd, path);
1481	\| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1482	\| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1483	}
1484		2060
1485	return mlockall (flags);	2061	path = EIO_INVALID_WD;
1486	}
1487	#endif
1488		2062
1489	#if !_POSIX_MEMLOCK_RANGE	2063	if (len >= 0)
1490	# define eio__mlock(a,b) ((errno = ENOSYS), -1)	2064	{
1491	#else	2065	((char *)tmpbuf->ptr)[len] = 0;
1492		2066	path = strdup (tmpbuf->ptr);
1493	static int	2067	}
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	{	2068	}
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		2069
1522	return msync (mem, len, flags);	2070	if (!path)
1523	}	2071	path = EIO_INVALID_WD;
1524		2072
1525	#endif	2073	return (eio_wd)path;
		2074	}
1526		2075
1527	int	2076	eio_wd
1528	eio__mtouch (void *mem, size_t len, int flags)	2077	eio_wd_open_sync (eio_wd wd, const char *path)
1529	{	2078	{
1530	eio_page_align (&mem, &len);	2079	struct tmpbuf tmpbuf = { 0 };
		2080	wd = eio__wd_open_sync (&tmpbuf, wd, path);
		2081	free (tmpbuf.ptr);
1531		2082
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;	2083	return wd;
1545	}	2084	}
		2085
		2086	void
		2087	eio_wd_close_sync (eio_wd wd)
		2088	{
		2089	if (wd != EIO_INVALID_WD)
		2090	free (wd);
		2091	}
		2092
		2093	#endif
1546		2094
1547	/*****************************************************************************/	2095	/*****************************************************************************/
1548		2096
1549	#define ALLOC(len) \	2097	#define ALLOC(len) \
1550	if (!req->ptr2) \	2098	if (!req->ptr2) \
…		…
1565	{	2113	{
1566	ETP_REQ *req;	2114	ETP_REQ *req;
1567	struct timespec ts;	2115	struct timespec ts;
1568	etp_worker self = (etp_worker )thr_arg;	2116	etp_worker self = (etp_worker )thr_arg;
1569		2117
		2118	#if HAVE_PRCTL_SET_NAME
		2119	prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
		2120	#endif
		2121
1570	/* try to distribute timeouts somewhat randomly */	2122	/* try to distribute timeouts somewhat evenly */
1571	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	2123	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1572		2124
1573	for (;;)	2125	for (;;)
1574	{	2126	{
		2127	ts.tv_sec = 0;
		2128
1575	X_LOCK (reqlock);	2129	X_LOCK (reqlock);
1576		2130
1577	for (;;)	2131	for (;;)
1578	{	2132	{
1579	self->req = req = reqq_shift (&req_queue);	2133	req = reqq_shift (&req_queue);
1580		2134
1581	if (req)	2135	if (req)
1582	break;	2136	break;
1583		2137
		2138	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		2139	{
		2140	X_UNLOCK (reqlock);
		2141	X_LOCK (wrklock);
		2142	--started;
		2143	X_UNLOCK (wrklock);
		2144	goto quit;
		2145	}
		2146
1584	++idle;	2147	++idle;
1585		2148
1586	ts.tv_sec = time (0) + idle_timeout;	2149	if (idle <= max_idle)
1587	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	2150	/* we are allowed to idle, so do so without any timeout */
		2151	X_COND_WAIT (reqwait, reqlock);
		2152	else
1588	{	2153	{
1589	if (idle > max_idle)	2154	/* initialise timeout once */
1590	{	2155	if (!ts.tv_sec)
1591	--idle;	2156	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		2157
1599	/* we are allowed to idle, so do so without any timeout */
1600	X_COND_WAIT (reqwait, reqlock);	2158	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		2159	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1601	}	2160	}
1602		2161
1603	--idle;	2162	--idle;
1604	}	2163	}
1605		2164
…		…
1608	X_UNLOCK (reqlock);	2167	X_UNLOCK (reqlock);
1609		2168
1610	if (req->type < 0)	2169	if (req->type < 0)
1611	goto quit;	2170	goto quit;
1612		2171
1613	if (!EIO_CANCELLED (req))
1614	ETP_EXECUTE (self, req);	2172	ETP_EXECUTE (self, req);
1615		2173
1616	X_LOCK (reslock);	2174	X_LOCK (reslock);
1617		2175
1618	++npending;	2176	++npending;
1619		2177
1620	if (!reqq_push (&res_queue, req) && want_poll_cb)	2178	if (!reqq_push (&res_queue, req) && want_poll_cb)
1621	want_poll_cb ();	2179	want_poll_cb ();
1622		2180
1623	self->req = 0;
1624	etp_worker_clear (self);	2181	etp_worker_clear (self);
1625		2182
1626	X_UNLOCK (reslock);	2183	X_UNLOCK (reslock);
1627	}	2184	}
1628		2185
1629	quit:	2186	quit:
		2187	free (req);
		2188
1630	X_LOCK (wrklock);	2189	X_LOCK (wrklock);
1631	etp_worker_free (self);	2190	etp_worker_free (self);
1632	X_UNLOCK (wrklock);	2191	X_UNLOCK (wrklock);
1633		2192
1634	return 0;	2193	return 0;
1635	}	2194	}
1636		2195
1637	/*****************************************************************************/	2196	/*****************************************************************************/
1638		2197
		2198	int ecb_cold
1639	int eio_init (void (want_poll)(void), void (done_poll)(void))	2199	eio_init (void (want_poll)(void), void (done_poll)(void))
1640	{	2200	{
		2201	#if !HAVE_PREADWRITE
		2202	X_MUTEX_CREATE (preadwritelock);
		2203	#endif
		2204
1641	return etp_init (want_poll, done_poll);	2205	return etp_init (want_poll, done_poll);
1642	}	2206	}
1643		2207
		2208	ecb_inline void
1644	static void eio_api_destroy (eio_req *req)	2209	eio_api_destroy (eio_req *req)
1645	{	2210	{
1646	free (req);	2211	free (req);
1647	}	2212	}
1648		2213
1649	#define REQ(rtype) \	2214	#define REQ(rtype) \
…		…
1668	{ \	2233	{ \
1669	eio_api_destroy (req); \	2234	eio_api_destroy (req); \
1670	return 0; \	2235	return 0; \
1671	}	2236	}
1672		2237
		2238	static void
1673	static void eio_execute (etp_worker self, eio_req req)	2239	eio_execute (etp_worker self, eio_req req)
1674	{	2240	{
		2241	#if HAVE_AT
		2242	int dirfd;
		2243	#else
		2244	const char *path;
		2245	#endif
		2246
		2247	if (ecb_expect_false (EIO_CANCELLED (req)))
		2248	{
		2249	req->result = -1;
		2250	req->errorno = ECANCELED;
		2251	return;
		2252	}
		2253
		2254	if (ecb_expect_false (req->wd == EIO_INVALID_WD))
		2255	{
		2256	req->result = -1;
		2257	req->errorno = ENOENT;
		2258	return;
		2259	}
		2260
		2261	if (req->type >= EIO_OPEN)
		2262	{
		2263	#if HAVE_AT
		2264	dirfd = WD2FD (req->wd);
		2265	#else
		2266	path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		2267	#endif
		2268	}
		2269
1675	switch (req->type)	2270	switch (req->type)
1676	{	2271	{
		2272	case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
		2273	req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
		2274	break;
		2275	case EIO_WD_CLOSE: eio_wd_close_sync (req->wd); break;
		2276
1677	case EIO_READ: ALLOC (req->size);	2277	case EIO_READ: ALLOC (req->size);
1678	req->result = req->offs >= 0	2278	req->result = req->offs >= 0
1679	? pread (req->int1, req->ptr2, req->size, req->offs)	2279	? pread (req->int1, req->ptr2, req->size, req->offs)
1680	: read (req->int1, req->ptr2, req->size); break;	2280	: read (req->int1, req->ptr2, req->size); break;
1681	case EIO_WRITE: req->result = req->offs >= 0	2281	case EIO_WRITE: req->result = req->offs >= 0
1682	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2282	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1683	: write (req->int1, req->ptr2, req->size); break;	2283	: write (req->int1, req->ptr2, req->size); break;
1684		2284
1685	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2285	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;	2286	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1687		2287
		2288	#if HAVE_AT
1688	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2289	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1689	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2290	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1690	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2291	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2292	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
		2293	case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
		2294	case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
		2295	case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
		2296	case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
		2297
		2298	case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
		2299	case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
		2300	case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
		2301	case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2); break;
		2302	case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2, 0); break;
		2303	case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
		2304	case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2305	case EIO_READLINK: ALLOC (PATH_MAX);
		2306	req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
		2307	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2308	req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2309	#else
		2310	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2311	req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2312	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1691	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2313	req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2314	case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
		2315	case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
		2316	case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
		2317	case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
		2318
		2319	case EIO_UNLINK: req->result = unlink (path ); break;
		2320	case EIO_RMDIR: req->result = rmdir (path ); break;
		2321	case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
		2322	case EIO_RENAME: req->result = rename (path , req->ptr2); break;
		2323	case EIO_LINK: req->result = link (path , req->ptr2); break;
		2324	case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
		2325	case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
		2326	case EIO_READLINK: ALLOC (PATH_MAX);
		2327	req->result = readlink (path, req->ptr2, PATH_MAX); break;
		2328	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2329	req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2330	#endif
		2331
		2332	case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
		2333	{
		2334	ALLOC (req->result);
		2335	memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
		2336	}
		2337	break;
		2338
1692	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2339	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1693	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2340	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1694		2341
1695	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1696	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1697	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));	2342	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1698	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;	2343	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1699		2344
1700	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1701	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	2345	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1702	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1703	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	2346	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1704	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1705	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;	2347	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1706		2348
1707	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1708	case EIO_CLOSE: req->result = close (req->int1); break;	2349	case EIO_CLOSE: req->result = close (req->int1); break;
1709	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;	2350	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1710	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1711	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1712	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1713	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1714	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1715	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;
1717
1718	case EIO_READLINK: ALLOC (PATH_MAX);
1719	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1720
1721	case EIO_SYNC: req->result = 0; sync (); break;	2351	case EIO_SYNC: req->result = 0; sync (); break;
1722	case EIO_FSYNC: req->result = fsync (req->int1); break;	2352	case EIO_FSYNC: req->result = fsync (req->int1); break;
1723	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2353	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2354	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2355	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1724	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2356	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;	2357	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1726	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2358	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1727	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2359	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;	2360	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1729		2361
1730	case EIO_READDIR: eio__scandir (req, self); break;	2362	case EIO_READDIR: eio__scandir (req); break;
1731		2363
1732	case EIO_BUSY:	2364	case EIO_BUSY:
1733	#ifdef _WIN32	2365	#ifdef _WIN32
1734	Sleep (req->nv1 * 1e3);	2366	Sleep (req->nv1 * 1e3);
1735	#else	2367	#else
…		…
1774	case EIO_NOP:	2406	case EIO_NOP:
1775	req->result = 0;	2407	req->result = 0;
1776	break;	2408	break;
1777		2409
1778	case EIO_CUSTOM:	2410	case EIO_CUSTOM:
1779	((void ()(eio_req ))req->feed) (req);	2411	req->feed (req);
1780	break;	2412	break;
1781		2413
1782	default:	2414	default:
1783	errno = ENOSYS;	2415	errno = ENOSYS;
1784	req->result = -1;	2416	req->result = -1;
…		…
1813	eio_req eio_msync (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2445	eio_req eio_msync (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1814	{	2446	{
1815	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2447	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1816	}	2448	}
1817		2449
		2450	eio_req eio_fdatasync (int fd, int pri, eio_cb cb, void data)
		2451	{
		2452	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2453	}
		2454
		2455	eio_req eio_syncfs (int fd, int pri, eio_cb cb, void data)
		2456	{
		2457	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2458	}
		2459
		2460	eio_req eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void data)
		2461	{
		2462	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2463	}
		2464
1818	eio_req eio_mtouch (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2465	eio_req eio_mtouch (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1819	{	2466	{
1820	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2467	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1821	}	2468	}
1822		2469
…		…
1828	eio_req eio_mlockall (int flags, int pri, eio_cb cb, void data)	2475	eio_req eio_mlockall (int flags, int pri, eio_cb cb, void data)
1829	{	2476	{
1830	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;	2477	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1831	}	2478	}
1832		2479
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)	2480	eio_req eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void data)
1834	{	2481	{
1835	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2482	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1836	}
1837
1838	eio_req eio_fdatasync (int fd, int pri, eio_cb cb, void data)
1839	{
1840	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1841	}	2483	}
1842		2484
1843	eio_req eio_close (int fd, int pri, eio_cb cb, void data)	2485	eio_req eio_close (int fd, int pri, eio_cb cb, void data)
1844	{	2486	{
1845	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2487	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
1883	eio_req eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void data)	2525	eio_req eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void data)
1884	{	2526	{
1885	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2527	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1886	}	2528	}
1887		2529
1888	eio_req eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void data)	2530	eio_req eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void data)
1889	{	2531	{
1890	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2532	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1891	}	2533	}
1892		2534
1893	eio_req eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void data)	2535	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)	2555	eio_req eio_truncate (const char path, off_t offset, int pri, eio_cb cb, void *data)
1914	{	2556	{
1915	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2557	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1916	}	2558	}
1917		2559
1918	eio_req eio_chown (const char path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2560	eio_req eio_chown (const char path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1919	{	2561	{
1920	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2562	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1921	}	2563	}
1922		2564
1923	eio_req eio_chmod (const char path, mode_t mode, int pri, eio_cb cb, void *data)	2565	eio_req eio_chmod (const char path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1937	}	2579	}
1938		2580
1939	eio_req eio_readlink (const char path, int pri, eio_cb cb, void *data)	2581	eio_req eio_readlink (const char path, int pri, eio_cb cb, void *data)
1940	{	2582	{
1941	return eio__1path (EIO_READLINK, path, pri, cb, data);	2583	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2584	}
		2585
		2586	eio_req eio_realpath (const char path, int pri, eio_cb cb, void *data)
		2587	{
		2588	return eio__1path (EIO_REALPATH, path, pri, cb, data);
1942	}	2589	}
1943		2590
1944	eio_req eio_stat (const char path, int pri, eio_cb cb, void *data)	2591	eio_req eio_stat (const char path, int pri, eio_cb cb, void *data)
1945	{	2592	{
1946	return eio__1path (EIO_STAT, path, pri, cb, data);	2593	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)	2652	eio_req eio_rename (const char path, const char new_path, int pri, eio_cb cb, void data)
2006	{	2653	{
2007	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2654	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2008	}	2655	}
2009		2656
2010	eio_req eio_custom (eio_cb execute, int pri, eio_cb cb, void data)	2657	eio_req eio_custom (void (execute)(eio_req ), int pri, eio_cb cb, void data)
2011	{	2658	{
2012	REQ (EIO_CUSTOM); req->feed = (void ()(eio_req ))execute; SEND;	2659	REQ (EIO_CUSTOM); req->feed = execute; SEND;
2013	}	2660	}
2014		2661
2015	#endif	2662	#endif
2016		2663
2017	eio_req eio_grp (eio_cb cb, void data)	2664	eio_req eio_grp (eio_cb cb, void data)
…		…
2026	#undef SEND	2673	#undef SEND
2027		2674
2028	/*****************************************************************************/	2675	/*****************************************************************************/
2029	/* grp functions */	2676	/* grp functions */
2030		2677
		2678	void
2031	void eio_grp_feed (eio_req grp, void (feed)(eio_req *req), int limit)	2679	eio_grp_feed (eio_req grp, void (feed)(eio_req *req), int limit)
2032	{	2680	{
2033	grp->int2 = limit;	2681	grp->int2 = limit;
2034	grp->feed = feed;	2682	grp->feed = feed;
2035		2683
2036	grp_try_feed (grp);	2684	grp_try_feed (grp);
2037	}	2685	}
2038		2686
		2687	void
2039	void eio_grp_limit (eio_req *grp, int limit)	2688	eio_grp_limit (eio_req *grp, int limit)
2040	{	2689	{
2041	grp->int2 = limit;	2690	grp->int2 = limit;
2042		2691
2043	grp_try_feed (grp);	2692	grp_try_feed (grp);
2044	}	2693	}
2045		2694
		2695	void
2046	void eio_grp_add (eio_req grp, eio_req req)	2696	eio_grp_add (eio_req grp, eio_req req)
2047	{	2697	{
2048	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2698	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2049		2699
2050	grp->flags \|= EIO_FLAG_GROUPADD;	2700	grp->flags \|= EIO_FLAG_GROUPADD;
2051		2701
…		…
2062	}	2712	}
2063		2713
2064	/*****************************************************************************/	2714	/*****************************************************************************/
2065	/* misc garbage */	2715	/* misc garbage */
2066		2716
		2717	eio_ssize_t
2067	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2718	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2068	{	2719	{
2069	etp_worker wrk;
2070	ssize_t ret;
2071
2072	wrk.dbuf = 0;
2073
2074	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2720	return eio__sendfile (ofd, ifd, offset, count);
2075
2076	if (wrk.dbuf)
2077	free (wrk.dbuf);
2078
2079	return ret;
2080	}	2721	}
2081		2722

Diff Legend

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

Comparing cvsroot/libeio/eio.c (file contents): Revision 1.66 by root, Sun Jun 5 19:58:37 2011 UTC vs. Revision 1.105 by root, Mon Sep 26 18:23:18 2011 UTC

Diff Legend

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.66 by root, Sun Jun 5 19:58:37 2011 UTC vs.
Revision 1.105 by root, Mon Sep 26 18:23:18 2011 UTC