ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.77 by root, Tue Jul 5 14:02:15 2011 UTC vs.
Revision 1.106 by root, Mon Sep 26 20:19:08 2011 UTC

…		…
43		43
44	#include "eio.h"	44	#include "eio.h"
45	#include "ecb.h"	45	#include "ecb.h"
46		46
47	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
48	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define X_STACKSIZE EIO_STACKSIZE
49	#endif	49	#endif
50	#include "xthread.h"	50	#include "xthread.h"
51		51
52	#include <errno.h>	52	#include <errno.h>
53	#include <stddef.h>	53	#include <stddef.h>
54	#include <stdlib.h>	54	#include <stdlib.h>
55	#include <string.h>	55	#include <string.h>
56	#include <errno.h>	56	#include <errno.h>
57	#include <sys/types.h>	57	#include <sys/types.h>
58	#include <sys/stat.h>	58	#include <sys/stat.h>
59	#include <sys/statvfs.h>
60	#include <limits.h>	59	#include <limits.h>
61	#include <fcntl.h>	60	#include <fcntl.h>
62	#include <assert.h>	61	#include <assert.h>
63		62
		63	#if _POSIX_VERSION >= 200809L
		64	# define HAVE_AT 1
		65	#else
		66	# define HAVE_AT 0
		67	#endif
		68
64	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */	69	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65	/* intptr_t only comes form stdint.h, says idiot openbsd coder */	70	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66	#if HAVE_STDINT_H	71	#if HAVE_STDINT_H
67	# include <stdint.h>	72	# include <stdint.h>
68	#endif	73	#endif
69		74
		75	#ifndef ECANCELED
		76	# define ECANCELED EDOM
		77	#endif
		78	#ifndef ELOOP
		79	# define ELOOP EDOM
		80	#endif
		81
		82	#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
		83	# define ENOTSOCK WSAENOTSOCK
		84	#endif
		85
		86	static void eio_destroy (eio_req *req);
		87
70	#ifndef EIO_FINISH	88	#ifndef EIO_FINISH
71	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	89	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
72	#endif	90	#endif
73		91
74	#ifndef EIO_DESTROY	92	#ifndef EIO_DESTROY
…		…
77		95
78	#ifndef EIO_FEED	96	#ifndef EIO_FEED
79	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	97	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
80	#endif	98	#endif
81		99
		100	#ifndef EIO_FD_TO_WIN32_HANDLE
		101	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		102	#endif
		103	#ifndef EIO_WIN32_HANDLE_TO_FD
		104	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		105	#endif
		106
		107	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		108
		109	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		110
82	#ifdef _WIN32	111	#ifdef _WIN32
83		112
84	/*doh*/	113	#undef PAGESIZE
		114	#define PAGESIZE 4096 /* GetSystemInfo? */
		115
		116	/* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
		117
		118	#ifdef EIO_STRUCT_STATI64
		119	/* look at perl's non-sloppy stat */
		120	#define stat(path,buf) _stati64 (path,buf)
		121	#define fstat(fd,buf) _fstati64 (fd,buf)
		122	#endif
		123	#define lstat(path,buf) stat (path,buf)
		124	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		125	#define mkdir(path,mode) _mkdir (path)
		126	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		127
		128	#define chmod(path,mode) _chmod (path, mode)
		129	#define dup(fd) _dup (fd)
		130	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		131
		132	#define fchmod(fd,mode) EIO_ENOSYS ()
		133	#define chown(path,uid,gid) EIO_ENOSYS ()
		134	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		135	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		136	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		137	#define mknod(path,mode,dev) EIO_ENOSYS ()
		138	#define sync() EIO_ENOSYS ()
		139	#define readlink(path,buf,s) EIO_ENOSYS ()
		140	#define statvfs(path,buf) EIO_ENOSYS ()
		141	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		142
		143	/* rename() uses MoveFile, which fails to overwrite */
		144	#define rename(old,neu) eio__rename (old, neu)
		145
		146	static int
		147	eio__rename (const char *old, const char *neu)
		148	{
		149	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		150	return 0;
		151
		152	/* should steal _dosmaperr */
		153	switch (GetLastError ())
		154	{
		155	case ERROR_FILE_NOT_FOUND:
		156	case ERROR_PATH_NOT_FOUND:
		157	case ERROR_INVALID_DRIVE:
		158	case ERROR_NO_MORE_FILES:
		159	case ERROR_BAD_NETPATH:
		160	case ERROR_BAD_NET_NAME:
		161	case ERROR_BAD_PATHNAME:
		162	case ERROR_FILENAME_EXCED_RANGE:
		163	errno = ENOENT;
		164	break;
		165
		166	default:
		167	errno = EACCES;
		168	break;
		169	}
		170
		171	return -1;
		172	}
		173
		174	/* we could even stat and see if it exists */
		175	static int
		176	symlink (const char *old, const char *neu)
		177	{
		178	#if WINVER >= 0x0600
		179	if (CreateSymbolicLink (neu, old, 1))
		180	return 0;
		181
		182	if (CreateSymbolicLink (neu, old, 0))
		183	return 0;
		184	#endif
		185
		186	return EIO_ERRNO (ENOENT, -1);
		187	}
		188
		189	/* POSIX API only */
		190	#define CreateHardLink(neu,old,flags) 0
		191	#define CreateSymbolicLink(neu,old,flags) 0
		192
		193	struct statvfs
		194	{
		195	int dummy;
		196	};
		197
		198	#define DT_DIR EIO_DT_DIR
		199	#define DT_REG EIO_DT_REG
		200	#define D_NAME(entp) entp.cFileName
		201	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
		202
85	#else	203	#else
86		204
87	# include <sys/time.h>	205	#include <sys/time.h>
88	# include <sys/select.h>	206	#include <sys/select.h>
		207	#include <sys/statvfs.h>
89	# include <unistd.h>	208	#include <unistd.h>
		209	#include <signal.h>
		210	#include <dirent.h>
		211
		212	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		213	#include <sys/mman.h>
		214	#endif
		215
		216	#define D_NAME(entp) entp->d_name
		217
		218	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		219	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		220	#define _DIRENT_HAVE_D_TYPE /* sigh */
		221	#define D_INO(de) (de)->d_fileno
		222	#define D_NAMLEN(de) (de)->d_namlen
		223	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		224	#define D_INO(de) (de)->d_ino
		225	#endif
		226
		227	#ifdef _D_EXACT_NAMLEN
		228	#undef D_NAMLEN
		229	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		230	#endif
		231
		232	#ifdef _DIRENT_HAVE_D_TYPE
		233	#define D_TYPE(de) (de)->d_type
		234	#endif
		235
		236	#ifndef EIO_STRUCT_DIRENT
		237	#define EIO_STRUCT_DIRENT struct dirent
		238	#endif
		239
		240	#endif
		241
		242	#if HAVE_UTIMES
90	# include <utime.h>	243	# include <utime.h>
91	# include <signal.h>
92	# include <dirent.h>
93
94	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
95	# include <sys/mman.h>
96	#endif
97
98	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
99	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
100	# define _DIRENT_HAVE_D_TYPE /* sigh */
101	# define D_INO(de) (de)->d_fileno
102	# define D_NAMLEN(de) (de)->d_namlen
103	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
104	# define D_INO(de) (de)->d_ino
105	# endif	244	#endif
106		245
107	#ifdef _D_EXACT_NAMLEN	246	#if HAVE_SYS_SYSCALL_H
108	# undef D_NAMLEN	247	# include <sys/syscall.h>
109	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
110	#endif
111
112	# ifdef _DIRENT_HAVE_D_TYPE
113	# define D_TYPE(de) (de)->d_type
114	# endif	248	#endif
115		249
116	# ifndef EIO_STRUCT_DIRENT	250	#if HAVE_SYS_PRCTL_H
117	# define EIO_STRUCT_DIRENT struct dirent	251	# include <sys/prctl.h>
118	# endif
119
120	#endif	252	#endif
121		253
122	#if HAVE_SENDFILE	254	#if HAVE_SENDFILE
123	# if __linux	255	# if __linux
124	# include <sys/sendfile.h>	256	# include <sys/sendfile.h>
…		…
139	#endif	271	#endif
140	#ifndef D_INO	272	#ifndef D_INO
141	# define D_INO(de) 0	273	# define D_INO(de) 0
142	#endif	274	#endif
143	#ifndef D_NAMLEN	275	#ifndef D_NAMLEN
144	# define D_NAMLEN(de) strlen ((de)->d_name)	276	# define D_NAMLEN(entp) strlen (D_NAME (entp))
145	#endif	277	#endif
146		278
147	/* used for struct dirent, AIX doesn't provide it */	279	/* used for struct dirent, AIX doesn't provide it */
148	#ifndef NAME_MAX	280	#ifndef NAME_MAX
149	# define NAME_MAX 4096	281	# define NAME_MAX 4096
…		…
156		288
157	/* buffer size for various temporary buffers */	289	/* buffer size for various temporary buffers */
158	#define EIO_BUFSIZE 65536	290	#define EIO_BUFSIZE 65536
159		291
160	#define dBUF \	292	#define dBUF \
161	char *eio_buf; \
162	ETP_WORKER_LOCK (self); \
163	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	293	char *eio_buf = malloc (EIO_BUFSIZE); \
164	ETP_WORKER_UNLOCK (self); \
165	errno = ENOMEM; \	294	errno = ENOMEM; \
166	if (!eio_buf) \	295	if (!eio_buf) \
167	return -1;	296	return -1
		297
		298	#define FUBd \
		299	free (eio_buf)
168		300
169	#define EIO_TICKS ((1000000 + 1023) >> 10)	301	#define EIO_TICKS ((1000000 + 1023) >> 10)
		302
		303	/*****************************************************************************/
		304
		305	struct tmpbuf
		306	{
		307	void *ptr;
		308	int len;
		309	};
		310
		311	static void *
		312	tmpbuf_get (struct tmpbuf *buf, int len)
		313	{
		314	if (buf->len < len)
		315	{
		316	free (buf->ptr);
		317	buf->ptr = malloc (buf->len = len);
		318	}
		319
		320	return buf->ptr;
		321	}
		322
		323	/*****************************************************************************/
170		324
171	#define ETP_PRI_MIN EIO_PRI_MIN	325	#define ETP_PRI_MIN EIO_PRI_MIN
172	#define ETP_PRI_MAX EIO_PRI_MAX	326	#define ETP_PRI_MAX EIO_PRI_MAX
173		327
174	struct etp_worker;	328	struct etp_worker;
…		…
178	static int eio_finish (eio_req *req);	332	static int eio_finish (eio_req *req);
179	#define ETP_FINISH(req) eio_finish (req)	333	#define ETP_FINISH(req) eio_finish (req)
180	static void eio_execute (struct etp_worker *self, eio_req *req);	334	static void eio_execute (struct etp_worker *self, eio_req *req);
181	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	335	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
182		336
183	#define ETP_WORKER_CLEAR(req) \
184	if (wrk->dbuf) \
185	{ \
186	free (wrk->dbuf); \
187	wrk->dbuf = 0; \
188	} \
189	\
190	if (wrk->dirp) \
191	{ \
192	closedir (wrk->dirp); \
193	wrk->dirp = 0; \
194	}
195
196	#define ETP_WORKER_COMMON \
197	void *dbuf; \
198	DIR *dirp;
199
200	/*****************************************************************************/	337	/*****************************************************************************/
201		338
202	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	339	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
203		340
204	/* calculate time difference in ~1/EIO_TICKS of a second */	341	/* calculate time difference in ~1/EIO_TICKS of a second */
…		…
215	static void (*done_poll_cb) (void);	352	static void (*done_poll_cb) (void);
216		353
217	static unsigned int max_poll_time; /* reslock */	354	static unsigned int max_poll_time; /* reslock */
218	static unsigned int max_poll_reqs; /* reslock */	355	static unsigned int max_poll_reqs; /* reslock */
219		356
220	static volatile unsigned int nreqs; /* reqlock */	357	static unsigned int nreqs; /* reqlock */
221	static volatile unsigned int nready; /* reqlock */	358	static unsigned int nready; /* reqlock */
222	static volatile unsigned int npending; /* reqlock */	359	static unsigned int npending; /* reqlock */
223	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */	360	static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
224	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */	361	static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
225		362
226	static xmutex_t wrklock;	363	static xmutex_t wrklock;
227	static xmutex_t reslock;	364	static xmutex_t reslock;
228	static xmutex_t reqlock;	365	static xmutex_t reqlock;
229	static xcond_t reqwait;	366	static xcond_t reqwait;
…		…
232	/*	369	/*
233	* make our pread/pwrite emulation safe against themselves, but not against	370	* make our pread/pwrite emulation safe against themselves, but not against
234	* normal read/write by using a mutex. slows down execution a lot,	371	* normal read/write by using a mutex. slows down execution a lot,
235	* but that's your problem, not mine.	372	* but that's your problem, not mine.
236	*/	373	*/
237	static xmutex_t preadwritelock = X_MUTEX_INIT;	374	static xmutex_t preadwritelock;
238	#endif	375	#endif
239		376
240	typedef struct etp_worker	377	typedef struct etp_worker
241	{	378	{
		379	struct tmpbuf tmpbuf;
		380
242	/* locked by wrklock */	381	/* locked by wrklock */
243	struct etp_worker *prev, *next;	382	struct etp_worker *prev, *next;
244		383
245	xthread_t tid;	384	xthread_t tid;
246		385
247	/* locked by reslock, reqlock or wrklock */	386	#ifdef ETP_WORKER_COMMON
248	ETP_REQ *req; /* currently processed request */
249
250	ETP_WORKER_COMMON	387	ETP_WORKER_COMMON
		388	#endif
251	} etp_worker;	389	} etp_worker;
252		390
253	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	391	static etp_worker wrk_first; /* NOT etp */
254		392
255	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	393	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
256	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	394	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
257		395
258	/* worker threads management */	396	/* worker threads management */
259		397
260	static void ecb_cold	398	static void ecb_cold
261	etp_worker_clear (etp_worker *wrk)	399	etp_worker_clear (etp_worker *wrk)
262	{	400	{
263	ETP_WORKER_CLEAR (wrk);
264	}	401	}
265		402
266	static void ecb_cold	403	static void ecb_cold
267	etp_worker_free (etp_worker *wrk)	404	etp_worker_free (etp_worker *wrk)
268	{	405	{
		406	free (wrk->tmpbuf.ptr);
		407
269	wrk->next->prev = wrk->prev;	408	wrk->next->prev = wrk->prev;
270	wrk->prev->next = wrk->next;	409	wrk->prev->next = wrk->next;
271		410
272	free (wrk);	411	free (wrk);
273	}	412	}
…		…
329	} etp_reqq;	468	} etp_reqq;
330		469
331	static etp_reqq req_queue;	470	static etp_reqq req_queue;
332	static etp_reqq res_queue;	471	static etp_reqq res_queue;
333		472
		473	static void ecb_noinline ecb_cold
		474	reqq_init (etp_reqq *q)
		475	{
		476	int pri;
		477
		478	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		479	q->qs[pri] = q->qe[pri] = 0;
		480
		481	q->size = 0;
		482	}
		483
334	static int ecb_noinline	484	static int ecb_noinline
335	reqq_push (etp_reqq *q, ETP_REQ *req)	485	reqq_push (etp_reqq *q, ETP_REQ *req)
336	{	486	{
337	int pri = req->pri;	487	int pri = req->pri;
338	req->next = 0;	488	req->next = 0;
…		…
372	}	522	}
373		523
374	abort ();	524	abort ();
375	}	525	}
376		526
377	static void ecb_cold	527	static int ecb_cold
378	etp_thread_init (void)	528	etp_init (void (*want_poll)(void), void (*done_poll)(void))
379	{	529	{
380	X_MUTEX_CREATE (wrklock);	530	X_MUTEX_CREATE (wrklock);
381	X_MUTEX_CREATE (reslock);	531	X_MUTEX_CREATE (reslock);
382	X_MUTEX_CREATE (reqlock);	532	X_MUTEX_CREATE (reqlock);
383	X_COND_CREATE (reqwait);	533	X_COND_CREATE (reqwait);
384	}
385		534
386	static void ecb_cold	535	reqq_init (&req_queue);
387	etp_atfork_prepare (void)	536	reqq_init (&res_queue);
388	{
389	X_LOCK (wrklock);
390	X_LOCK (reqlock);
391	X_LOCK (reslock);
392	#if !HAVE_PREADWRITE
393	X_LOCK (preadwritelock);
394	#endif
395	}
396		537
397	static void ecb_cold	538	wrk_first.next =
398	etp_atfork_parent (void)	539	wrk_first.prev = &wrk_first;
399	{
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 ecb_cold
409	etp_atfork_child (void)
410	{
411	ETP_REQ *prv;
412
413	while ((prv = reqq_shift (&req_queue)))
414	ETP_DESTROY (prv);
415
416	while ((prv = reqq_shift (&res_queue)))
417	ETP_DESTROY (prv);
418
419	while (wrk_first.next != &wrk_first)
420	{
421	etp_worker *wrk = wrk_first.next;
422
423	if (wrk->req)
424	ETP_DESTROY (wrk->req);
425
426	etp_worker_clear (wrk);
427	etp_worker_free (wrk);
428	}
429		540
430	started = 0;	541	started = 0;
431	idle = 0;	542	idle = 0;
432	nreqs = 0;	543	nreqs = 0;
433	nready = 0;	544	nready = 0;
434	npending = 0;	545	npending = 0;
435
436	etp_thread_init ();
437	}
438
439	static void ecb_cold
440	etp_once_init (void)
441	{
442	etp_thread_init ();
443	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
444	}
445
446	static int ecb_cold
447	etp_init (void (*want_poll)(void), void (*done_poll)(void))
448	{
449	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
450
451	pthread_once (&doinit, etp_once_init);
452		546
453	want_poll_cb = want_poll;	547	want_poll_cb = want_poll;
454	done_poll_cb = done_poll;	548	done_poll_cb = done_poll;
455		549
456	return 0;	550	return 0;
…		…
496	}	590	}
497		591
498	static void ecb_cold	592	static void ecb_cold
499	etp_end_thread (void)	593	etp_end_thread (void)
500	{	594	{
501	eio_req *req = calloc (1, sizeof (eio_req));	595	eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
502		596
503	req->type = -1;	597	req->type = -1;
504	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	598	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
505		599
506	X_LOCK (reqlock);	600	X_LOCK (reqlock);
…		…
583	}	677	}
584		678
585	static void	679	static void
586	etp_cancel (ETP_REQ *req)	680	etp_cancel (ETP_REQ *req)
587	{	681	{
588	X_LOCK (wrklock);	682	req->cancelled = 1;
589	req->flags |= EIO_FLAG_CANCELLED;
590	X_UNLOCK (wrklock);
591		683
592	eio_grp_cancel (req);	684	eio_grp_cancel (req);
593	}	685	}
594		686
595	static void	687	static void
…		…
711	return eio_finish (grp);	803	return eio_finish (grp);
712	else	804	else
713	return 0;	805	return 0;
714	}	806	}
715		807
716	void	808	static void
717	eio_destroy (eio_req *req)	809	eio_destroy (eio_req *req)
718	{	810	{
719	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	811	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
720	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	812	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
721		813
…		…
739	if (grp->grp_first == req)	831	if (grp->grp_first == req)
740	grp->grp_first = req->grp_next;	832	grp->grp_first = req->grp_next;
741		833
742	res2 = grp_dec (grp);	834	res2 = grp_dec (grp);
743		835
744	if (!res && res2)	836	if (!res)
745	res = res2;	837	res = res2;
746	}	838	}
747		839
748	eio_destroy (req);	840	eio_destroy (req);
749		841
…		…
841	# undef pread	933	# undef pread
842	# undef pwrite	934	# undef pwrite
843	# define pread eio__pread	935	# define pread eio__pread
844	# define pwrite eio__pwrite	936	# define pwrite eio__pwrite
845		937
846	static ssize_t	938	static eio_ssize_t
847	eio__pread (int fd, void *buf, size_t count, off_t offset)	939	eio__pread (int fd, void *buf, size_t count, off_t offset)
848	{	940	{
849	ssize_t res;	941	eio_ssize_t res;
850	off_t ooffset;	942	off_t ooffset;
851		943
852	X_LOCK (preadwritelock);	944	X_LOCK (preadwritelock);
853	ooffset = lseek (fd, 0, SEEK_CUR);	945	ooffset = lseek (fd, 0, SEEK_CUR);
854	lseek (fd, offset, SEEK_SET);	946	lseek (fd, offset, SEEK_SET);
…		…
857	X_UNLOCK (preadwritelock);	949	X_UNLOCK (preadwritelock);
858		950
859	return res;	951	return res;
860	}	952	}
861		953
862	static ssize_t	954	static eio_ssize_t
863	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	955	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
864	{	956	{
865	ssize_t res;	957	eio_ssize_t res;
866	off_t ooffset;	958	off_t ooffset;
867		959
868	X_LOCK (preadwritelock);	960	X_LOCK (preadwritelock);
869	ooffset = lseek (fd, 0, SEEK_CUR);	961	ooffset = lseek (fd, 0, SEEK_CUR);
870	lseek (fd, offset, SEEK_SET);	962	lseek (fd, offset, SEEK_SET);
…		…
915		1007
916	#if !HAVE_FDATASYNC	1008	#if !HAVE_FDATASYNC
917	# undef fdatasync	1009	# undef fdatasync
918	# define fdatasync(fd) fsync (fd)	1010	# define fdatasync(fd) fsync (fd)
919	#endif	1011	#endif
		1012
		1013	static int
		1014	eio__syncfs (int fd)
		1015	{
		1016	int res;
		1017
		1018	#if HAVE_SYS_SYNCFS
		1019	res = (int)syscall (__NR_syncfs, (int)(fd));
		1020	#else
		1021	res = -1;
		1022	errno = ENOSYS;
		1023	#endif
		1024
		1025	if (res < 0 && errno == ENOSYS && fd >= 0)
		1026	sync ();
		1027
		1028	return res;
		1029	}
920		1030
921	/* sync_file_range always needs emulation */	1031	/* sync_file_range always needs emulation */
922	static int	1032	static int
923	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	1033	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
924	{	1034	{
…		…
944	/* even though we could play tricks with the flags, it's better to always	1054	/* even though we could play tricks with the flags, it's better to always
945	* call fdatasync, as that matches the expectation of its users best */	1055	* call fdatasync, as that matches the expectation of its users best */
946	return fdatasync (fd);	1056	return fdatasync (fd);
947	}	1057	}
948		1058
		1059	static int
		1060	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		1061	{
		1062	#if HAVE_FALLOCATE
		1063	return fallocate (fd, mode, offset, len);
		1064	#else
		1065	errno = ENOSYS;
		1066	return -1;
		1067	#endif
		1068	}
		1069
949	#if !HAVE_READAHEAD	1070	#if !HAVE_READAHEAD
950	# undef readahead	1071	# undef readahead
951	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1072	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
952		1073
953	static ssize_t	1074	static eio_ssize_t
954	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1075	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
955	{	1076	{
956	size_t todo = count;	1077	size_t todo = count;
957	dBUF;	1078	dBUF;
958		1079
…		…
963	pread (fd, eio_buf, len, offset);	1084	pread (fd, eio_buf, len, offset);
964	offset += len;	1085	offset += len;
965	todo -= len;	1086	todo -= len;
966	}	1087	}
967		1088
		1089	FUBd;
		1090
968	errno = 0;	1091	errno = 0;
969	return count;	1092	return count;
970	}	1093	}
971		1094
972	#endif	1095	#endif
973		1096
974	/* sendfile always needs emulation */	1097	/* sendfile always needs emulation */
975	static ssize_t	1098	static eio_ssize_t
976	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1099	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
977	{	1100	{
978	ssize_t written = 0;	1101	eio_ssize_t written = 0;
979	ssize_t res;	1102	eio_ssize_t res;
980		1103
981	if (!count)	1104	if (!count)
982	return 0;	1105	return 0;
983		1106
984	for (;;)	1107	for (;;)
985	{	1108	{
		1109	#ifdef __APPLE__
		1110	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1111	#endif
986	#if HAVE_SENDFILE	1112	#if HAVE_SENDFILE
987	# if __linux	1113	# if __linux
988	off_t soffset = offset;	1114	off_t soffset = offset;
989	res = sendfile (ofd, ifd, &soffset, count);	1115	res = sendfile (ofd, ifd, &soffset, count);
990		1116
…		…
1006		1132
1007	/* according to source inspection, this is correct, and useful behaviour */	1133	/* according to source inspection, this is correct, and useful behaviour */
1008	if (sbytes)	1134	if (sbytes)
1009	res = sbytes;	1135	res = sbytes;
1010		1136
1011	# elif defined (__APPLE__) && 0 /* broken, as everything on os x */	1137	# elif defined (__APPLE__)
1012	off_t sbytes = count;	1138	off_t sbytes = count;
1013	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	1139	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1014		1140
1015	/* according to the manpage, sbytes is always valid */	1141	/* according to the manpage, sbytes is always valid */
1016	if (sbytes)	1142	if (sbytes)
…		…
1033	if (res < 0 && sbytes)	1159	if (res < 0 && sbytes)
1034	res = sbytes;	1160	res = sbytes;
1035		1161
1036	# endif	1162	# endif
1037		1163
1038	#elif defined (_WIN32)	1164	#elif defined (_WIN32) && 0
1039	/* does not work, just for documentation of what would need to be done */	1165	/* does not work, just for documentation of what would need to be done */
1040	/* actually, cannot be done like this, as TransmitFile changes the file offset, */	1166	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
1041	/* libeio guarantees that the file offset does not change, and windows */	1167	/* libeio guarantees that the file offset does not change, and windows */
1042	/* has no way to get an independent handle to the same file description */	1168	/* has no way to get an independent handle to the same file description */
1043	HANDLE h = TO_SOCKET (ifd);	1169	HANDLE h = TO_SOCKET (ifd);
…		…
1076	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1202	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1077	/* BSDs */	1203	/* BSDs */
1078	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1204	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1079	|| errno == ENOTSUP	1205	|| errno == ENOTSUP
1080	#endif	1206	#endif
		1207	#ifdef EOPNOTSUPP /* windows */
1081	|| errno == EOPNOTSUPP /* BSDs */	1208	|| errno == EOPNOTSUPP /* BSDs */
		1209	#endif
1082	#if __solaris	1210	#if __solaris
1083	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1211	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1084	#endif	1212	#endif
1085	)	1213	)
1086	)	1214	)
…		…
1090		1218
1091	res = 0;	1219	res = 0;
1092		1220
1093	while (count)	1221	while (count)
1094	{	1222	{
1095	ssize_t cnt;	1223	eio_ssize_t cnt;
1096		1224
1097	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1225	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1098		1226
1099	if (cnt <= 0)	1227	if (cnt <= 0)
1100	{	1228	{
…		…
1112		1240
1113	offset += cnt;	1241	offset += cnt;
1114	res += cnt;	1242	res += cnt;
1115	count -= cnt;	1243	count -= cnt;
1116	}	1244	}
		1245
		1246	FUBd;
1117	}	1247	}
1118		1248
1119	return res;	1249	return res;
1120	}	1250	}
1121		1251
…		…
1148	/* round up length */	1278	/* round up length */
1149	*length = (*length + mask) & ~mask;	1279	*length = (*length + mask) & ~mask;
1150	}	1280	}
1151		1281
1152	#if !_POSIX_MEMLOCK	1282	#if !_POSIX_MEMLOCK
1153	# define eio__mlockall(a) ((errno = ENOSYS), -1)	1283	# define eio__mlockall(a) EIO_ENOSYS ()
1154	#else	1284	#else
1155		1285
1156	static int	1286	static int
1157	eio__mlockall (int flags)	1287	eio__mlockall (int flags)
1158	{	1288	{
…		…
1172	return mlockall (flags);	1302	return mlockall (flags);
1173	}	1303	}
1174	#endif	1304	#endif
1175		1305
1176	#if !_POSIX_MEMLOCK_RANGE	1306	#if !_POSIX_MEMLOCK_RANGE
1177	# define eio__mlock(a,b) ((errno = ENOSYS), -1)	1307	# define eio__mlock(a,b) EIO_ENOSYS ()
1178	#else	1308	#else
1179		1309
1180	static int	1310	static int
1181	eio__mlock (void *addr, size_t length)	1311	eio__mlock (void *addr, size_t length)
1182	{	1312	{
…		…
1186	}	1316	}
1187		1317
1188	#endif	1318	#endif
1189		1319
1190	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)	1320	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1191	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)	1321	# define eio__msync(a,b,c) EIO_ENOSYS ()
1192	#else	1322	#else
1193		1323
1194	static int	1324	static int
1195	eio__msync (void *mem, size_t len, int flags)	1325	eio__msync (void *mem, size_t len, int flags)
1196	{	1326	{
…		…
1236	}	1366	}
1237		1367
1238	/*****************************************************************************/	1368	/*****************************************************************************/
1239	/* requests implemented outside eio_execute, because they are so large */	1369	/* requests implemented outside eio_execute, because they are so large */
1240		1370
		1371	/* copies some absolute path to tmpbuf */
		1372	static char *
		1373	eio__getwd (struct tmpbuf *tmpbuf, eio_wd wd)
		1374	{
		1375	if (wd == EIO_CWD)
		1376	return getcwd (tmpbuf->ptr, PATH_MAX);
		1377
		1378	#if HAVE_AT
		1379	abort (); /*TODO*/
		1380	#else
		1381	strcpy (tmpbuf->ptr, wd);
		1382	#endif
		1383	return tmpbuf->ptr;
		1384	}
		1385
		1386	/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1241	static void	1387	static int
1242	eio__realpath (eio_req *req, etp_worker *self)	1388	eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1243	{	1389	{
1244	char *rel = req->ptr1;	1390	const char *rel = path;
1245	char *res;	1391	char *res;
1246	char *tmp1, *tmp2;	1392	char *tmp1, *tmp2;
1247	#if SYMLOOP_MAX > 32	1393	#if SYMLOOP_MAX > 32
1248	int links = SYMLOOP_MAX;	1394	int symlinks = SYMLOOP_MAX;
1249	#else	1395	#else
1250	int links = 32;	1396	int symlinks = 32;
1251	#endif	1397	#endif
1252		1398
1253	req->result = -1;	1399	/*D*/ /*TODO: wd ignored */
1254		1400
1255	errno = EINVAL;	1401	errno = EINVAL;
1256	if (!rel)	1402	if (!rel)
1257	return;	1403	return -1;
1258		1404
1259	errno = ENOENT;	1405	errno = ENOENT;
1260	if (!*rel)	1406	if (!*rel)
1261	return;	1407	return -1;
1262		1408
1263	if (!req->ptr2)	1409	res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1264	{
1265	X_LOCK (wrklock);
1266	req->flags |= EIO_FLAG_PTR2_FREE;
1267	X_UNLOCK (wrklock);
1268	req->ptr2 = malloc (PATH_MAX * 3);
1269
1270	errno = ENOMEM;
1271	if (!req->ptr2)
1272	return;
1273	}
1274
1275	res = req->ptr2;
1276	tmp1 = res + PATH_MAX;	1410	tmp1 = res + PATH_MAX;
1277	tmp2 = tmp1 + PATH_MAX;	1411	tmp2 = tmp1 + PATH_MAX;
1278		1412
		1413	#if 0 /* disabled, the musl way to do things is just too racy */
		1414	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1415	/* on linux we may be able to ask the kernel */
		1416	{
		1417	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1418
		1419	if (fd >= 0)
		1420	{
		1421	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1422	req->result = readlink (tmp1, res, PATH_MAX);
		1423	close (fd);
		1424
		1425	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1426
		1427	if (req->result > 0)
		1428	goto done;
		1429	}
		1430	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1431	return;
		1432	}
		1433	#endif
		1434	#endif
		1435
1279	if (*rel != '/')	1436	if (*rel != '/')
1280	{	1437	{
1281	if (!getcwd (res, PATH_MAX))	1438	if (!eio__getwd (tmpbuf, wd))
1282	return;	1439	return -1;
1283		1440
1284	if (res [1]) /* only use if not / */	1441	if (res [1]) /* only use if not / */
1285	res += strlen (res);	1442	res += strlen (res);
1286	}	1443	}
1287		1444
1288	while (*rel)	1445	while (*rel)
1289	{	1446	{
1290	ssize_t len, linklen;	1447	eio_ssize_t len, linklen;
1291	char *beg = rel;	1448	const char *beg = rel;
1292		1449
1293	while (*rel && *rel != '/')	1450	while (*rel && *rel != '/')
1294	++rel;	1451	++rel;
1295		1452
1296	len = rel - beg;	1453	len = rel - beg;
…		…
1308		1465
1309	if (beg [1] == '.' && len == 2)	1466	if (beg [1] == '.' && len == 2)
1310	{	1467	{
1311	/* .. - back up one component, if possible */	1468	/* .. - back up one component, if possible */
1312		1469
1313	while (res != req->ptr2)	1470	while (res != tmpbuf->ptr)
1314	if (*--res == '/')	1471	if (*--res == '/')
1315	break;	1472	break;
1316		1473
1317	continue;	1474	continue;
1318	}	1475	}
…		…
1328		1485
1329	/* zero-terminate, for readlink */	1486	/* zero-terminate, for readlink */
1330	res [len + 1] = 0;	1487	res [len + 1] = 0;
1331		1488
1332	/* now check if it's a symlink */	1489	/* now check if it's a symlink */
1333	linklen = readlink (req->ptr2, tmp1, PATH_MAX);	1490	linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1334		1491
1335	if (linklen < 0)	1492	if (linklen < 0)
1336	{	1493	{
1337	if (errno != EINVAL)	1494	if (errno != EINVAL)
1338	return;	1495	return -1;
1339		1496
1340	/* it's a normal directory. hopefully */	1497	/* it's a normal directory. hopefully */
1341	res += len + 1;	1498	res += len + 1;
1342	}	1499	}
1343	else	1500	else
…		…
1345	/* yay, it was a symlink - build new path in tmp2 */	1502	/* yay, it was a symlink - build new path in tmp2 */
1346	int rellen = strlen (rel);	1503	int rellen = strlen (rel);
1347		1504
1348	errno = ENAMETOOLONG;	1505	errno = ENAMETOOLONG;
1349	if (linklen + 1 + rellen >= PATH_MAX)	1506	if (linklen + 1 + rellen >= PATH_MAX)
1350	return;	1507	return -1;
		1508
		1509	errno = ELOOP;
		1510	if (!--symlinks)
		1511	return -1;
1351		1512
1352	if (*tmp1 == '/')	1513	if (*tmp1 == '/')
1353	res = req->ptr2; /* symlink resolves to an absolute path */	1514	res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1354		1515
1355	/* we need to be careful, as rel might point into tmp2 already */	1516	/* we need to be careful, as rel might point into tmp2 already */
1356	memmove (tmp2 + linklen + 1, rel, rellen + 1);	1517	memmove (tmp2 + linklen + 1, rel, rellen + 1);
1357	tmp2 [linklen] = '/';	1518	tmp2 [linklen] = '/';
1358	memcpy (tmp2, tmp1, linklen);	1519	memcpy (tmp2, tmp1, linklen);
…		…
1360	rel = tmp2;	1521	rel = tmp2;
1361	}	1522	}
1362	}	1523	}
1363		1524
1364	/* special case for the lone root path */	1525	/* special case for the lone root path */
1365	if (res == req->ptr2)	1526	if (res == tmpbuf->ptr)
1366	*res++ = '/';	1527	*res++ = '/';
1367		1528
1368	req->result = res - (char *)req->ptr2;	1529	return res - (char *)tmpbuf->ptr;
1369	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1370	}	1530	}
1371		1531
1372	static signed char	1532	static signed char
1373	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1533	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1374	{	1534	{
…		…
1382		1542
1383	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1543	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1384	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1544	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1385		1545
1386	static void	1546	static void
1387	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1547	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1388	{	1548	{
1389	unsigned char bits [9 + sizeof (ino_t) * 8];	1549	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1390	unsigned char *bit = bits;	1550	unsigned char *bit = bits;
1391		1551
1392	assert (CHAR_BIT == 8);	1552	assert (CHAR_BIT == 8);
1393	assert (sizeof (eio_dirent) * 8 < 256);	1553	assert (sizeof (eio_dirent) * 8 < 256);
1394	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1554	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1396		1556
1397	if (size <= EIO_SORT_FAST)	1557	if (size <= EIO_SORT_FAST)
1398	return;	1558	return;
1399		1559
1400	/* first prepare an array of bits to test in our radix sort */	1560	/* first prepare an array of bits to test in our radix sort */
1401	/* try to take endianness into account, as well as differences in ino_t sizes */	1561	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1402	/* inode_bits must contain all inodes ORed together */	1562	/* inode_bits must contain all inodes ORed together */
1403	/* which is used to skip bits that are 0 everywhere, which is very common */	1563	/* which is used to skip bits that are 0 everywhere, which is very common */
1404	{	1564	{
1405	ino_t endianness;	1565	eio_ino_t endianness;
1406	int i, j;	1566	int i, j;
1407		1567
1408	/* we store the byte offset of byte n into byte n of "endianness" */	1568	/* we store the byte offset of byte n into byte n of "endianness" */
1409	for (i = 0; i < sizeof (ino_t); ++i)	1569	for (i = 0; i < sizeof (eio_ino_t); ++i)
1410	((unsigned char *)&endianness)[i] = i;	1570	((unsigned char *)&endianness)[i] = i;
1411		1571
1412	*bit++ = 0;	1572	*bit++ = 0;
1413		1573
1414	for (i = 0; i < sizeof (ino_t); ++i)	1574	for (i = 0; i < sizeof (eio_ino_t); ++i)
1415	{	1575	{
1416	/* shifting off the byte offsets out of "endianness" */	1576	/* shifting off the byte offsets out of "endianness" */
1417	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1577	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1418	endianness >>= 8;	1578	endianness >>= 8;
1419		1579
1420	for (j = 0; j < 8; ++j)	1580	for (j = 0; j < 8; ++j)
1421	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1581	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1422	*bit++ = offs + j;	1582	*bit++ = offs + j;
1423	}	1583	}
1424		1584
1425	for (j = 0; j < 8; ++j)	1585	for (j = 0; j < 8; ++j)
1426	if (score_bits & (1 << j))	1586	if (score_bits & (1 << j))
1427	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1587	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1428	}	1588	}
1429		1589
1430	/* now actually do the sorting (a variant of MSD radix sort) */	1590	/* now actually do the sorting (a variant of MSD radix sort) */
1431	{	1591	{
1432	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1592	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1433	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1593	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1434	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1594	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1435	int stk_idx = 0;	1595	int stk_idx = 0;
1436		1596
1437	base_stk [stk_idx] = dents;	1597	base_stk [stk_idx] = dents;
1438	end_stk [stk_idx] = dents + size;	1598	end_stk [stk_idx] = dents + size;
1439	bit_stk [stk_idx] = bit - 1;	1599	bit_stk [stk_idx] = bit - 1;
…		…
1518	}	1678	}
1519	}	1679	}
1520	}	1680	}
1521		1681
1522	static void	1682	static void
1523	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1683	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1524	{	1684	{
1525	if (size <= 1)	1685	if (size <= 1)
1526	return; /* our insertion sort relies on size > 0 */	1686	return; /* our insertion sort relies on size > 0 */
1527		1687
1528	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1688	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1534	eio_dent_insertion_sort (dents, size);	1694	eio_dent_insertion_sort (dents, size);
1535	}	1695	}
1536		1696
1537	/* read a full directory */	1697	/* read a full directory */
1538	static void	1698	static void
1539	eio__scandir (eio_req *req, etp_worker *self)	1699	eio__scandir (eio_req *req)
1540	{	1700	{
1541	DIR *dirp;
1542	EIO_STRUCT_DIRENT *entp;
1543	char *name, *names;	1701	char *name, *names;
1544	int namesalloc = 4096;	1702	int namesalloc = 4096 - sizeof (void *) * 4;
1545	int namesoffs = 0;	1703	int namesoffs = 0;
1546	int flags = req->int1;	1704	int flags = req->int1;
1547	eio_dirent *dents = 0;	1705	eio_dirent *dents = 0;
1548	int dentalloc = 128;	1706	int dentalloc = 128;
1549	int dentoffs = 0;	1707	int dentoffs = 0;
1550	ino_t inode_bits = 0;	1708	eio_ino_t inode_bits = 0;
		1709	#ifdef _WIN32
		1710	HANDLE dirp;
		1711	WIN32_FIND_DATA entp;
		1712	#else
		1713	DIR *dirp;
		1714	EIO_STRUCT_DIRENT *entp;
		1715	#endif
1551		1716
1552	req->result = -1;	1717	req->result = -1;
1553		1718
1554	if (!(flags & EIO_READDIR_DENTS))	1719	if (!(flags & EIO_READDIR_DENTS))
1555	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1720	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1556		1721
1557	X_LOCK (wrklock);	1722	#ifdef _WIN32
1558	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1723	{
		1724	int len = strlen ((const char *)req->ptr1);
		1725	char *path = malloc (MAX_PATH);
		1726	const char *fmt;
		1727
		1728	if (!len)
		1729	fmt = "./*";
		1730	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1731	fmt = "%s*";
		1732	else
		1733	fmt = "%s/*";
		1734
		1735	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1736	dirp = FindFirstFile (path, &entp);
		1737	free (path);
		1738
		1739	if (dirp == INVALID_HANDLE_VALUE)
		1740	{
		1741	/* should steal _dosmaperr */
		1742	switch (GetLastError ())
		1743	{
		1744	case ERROR_FILE_NOT_FOUND:
		1745	req->result = 0;
		1746	break;
		1747
		1748	case ERROR_INVALID_NAME:
		1749	case ERROR_PATH_NOT_FOUND:
		1750	case ERROR_NO_MORE_FILES:
		1751	errno = ENOENT;
		1752	break;
		1753
		1754	case ERROR_NOT_ENOUGH_MEMORY:
		1755	errno = ENOMEM;
		1756	break;
		1757
		1758	default:
		1759	errno = EINVAL;
		1760	break;
		1761	}
		1762
		1763	return;
		1764	}
		1765	}
		1766	#else
1559	self->dirp = dirp = opendir (req->ptr1);	1767	dirp = opendir (req->ptr1);
		1768
		1769	if (!dirp)
		1770	return;
		1771	#endif
1560		1772
1561	if (req->flags & EIO_FLAG_PTR1_FREE)	1773	if (req->flags & EIO_FLAG_PTR1_FREE)
1562	free (req->ptr1);	1774	free (req->ptr1);
1563		1775
1564	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1776	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1565	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1777	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1566	req->ptr2 = names = malloc (namesalloc);	1778	req->ptr2 = names = malloc (namesalloc);
1567	X_UNLOCK (wrklock);
1568		1779
1569	if (dirp && names && (!flags || dents))	1780	if (!names || (flags && !dents))
		1781	return;
		1782
1570	for (;;)	1783	for (;;)
1571	{	1784	{
		1785	int done;
		1786
		1787	#ifdef _WIN32
		1788	done = !dirp;
		1789	#else
1572	errno = 0;	1790	errno = 0;
1573	entp = readdir (dirp);	1791	entp = readdir (dirp);
		1792	done = !entp;
		1793	#endif
1574		1794
1575	if (!entp)	1795	if (done)
1576	{	1796	{
		1797	#ifndef _WIN32
		1798	int old_errno = errno;
		1799	closedir (dirp);
		1800	errno = old_errno;
		1801
1577	if (errno)	1802	if (errno)
1578	break;	1803	break;
		1804	#endif
1579		1805
1580	/* sort etc. */	1806	/* sort etc. */
1581	req->int1 = flags;	1807	req->int1 = flags;
1582	req->result = dentoffs;	1808	req->result = dentoffs;
1583		1809
1584	if (flags & EIO_READDIR_STAT_ORDER)	1810	if (flags & EIO_READDIR_STAT_ORDER)
1585	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);	1811	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1586	else if (flags & EIO_READDIR_DIRS_FIRST)	1812	else if (flags & EIO_READDIR_DIRS_FIRST)
1587	if (flags & EIO_READDIR_FOUND_UNKNOWN)	1813	if (flags & EIO_READDIR_FOUND_UNKNOWN)
1588	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */	1814	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1589	else	1815	else
		1816	{
		1817	/* in this case, all is known, and we just put dirs first and sort them */
		1818	eio_dirent *oth = dents + dentoffs;
		1819	eio_dirent *dir = dents;
		1820
		1821	/* now partition dirs to the front, and non-dirs to the back */
		1822	/* by walking from both sides and swapping if necessary */
		1823	while (oth > dir)
		1824	{
		1825	if (dir->type == EIO_DT_DIR)
		1826	++dir;
		1827	else if ((--oth)->type == EIO_DT_DIR)
		1828	{
		1829	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1830
		1831	++dir;
		1832	}
		1833	}
		1834
		1835	/* now sort the dirs only (dirs all have the same score) */
		1836	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1837	}
		1838
		1839	break;
		1840	}
		1841
		1842	/* now add the entry to our list(s) */
		1843	name = D_NAME (entp);
		1844
		1845	/* skip . and .. entries */
		1846	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
		1847	{
		1848	int len = D_NAMLEN (entp) + 1;
		1849
		1850	while (ecb_expect_false (namesoffs + len > namesalloc))
		1851	{
		1852	namesalloc *= 2;
		1853	req->ptr2 = names = realloc (names, namesalloc);
		1854
		1855	if (!names)
		1856	break;
		1857	}
		1858
		1859	memcpy (names + namesoffs, name, len);
		1860
		1861	if (dents)
		1862	{
		1863	struct eio_dirent *ent;
		1864
		1865	if (ecb_expect_false (dentoffs == dentalloc))
1590	{	1866	{
1591	/* in this case, all is known, and we just put dirs first and sort them */	1867	dentalloc *= 2;
		1868	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1869
		1870	if (!dents)
		1871	break;
		1872	}
		1873
1592	eio_dirent *oth = dents + dentoffs;	1874	ent = dents + dentoffs;
1593	eio_dirent *dir = dents;
1594		1875
1595	/* now partition dirs to the front, and non-dirs to the back */	1876	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1596	/* by walking from both sides and swapping if necessary */	1877	ent->namelen = len - 1;
1597	while (oth > dir)	1878	ent->inode = D_INO (entp);
		1879
		1880	inode_bits |= ent->inode;
		1881
		1882	switch (D_TYPE (entp))
		1883	{
		1884	default:
		1885	ent->type = EIO_DT_UNKNOWN;
		1886	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1887	break;
		1888
		1889	#ifdef DT_FIFO
		1890	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1891	#endif
		1892	#ifdef DT_CHR
		1893	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1894	#endif
		1895	#ifdef DT_MPC
		1896	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1897	#endif
		1898	#ifdef DT_DIR
		1899	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1900	#endif
		1901	#ifdef DT_NAM
		1902	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1903	#endif
		1904	#ifdef DT_BLK
		1905	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1906	#endif
		1907	#ifdef DT_MPB
		1908	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1909	#endif
		1910	#ifdef DT_REG
		1911	case DT_REG: ent->type = EIO_DT_REG; break;
		1912	#endif
		1913	#ifdef DT_NWK
		1914	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1915	#endif
		1916	#ifdef DT_CMP
		1917	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1918	#endif
		1919	#ifdef DT_LNK
		1920	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1921	#endif
		1922	#ifdef DT_SOCK
		1923	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1924	#endif
		1925	#ifdef DT_DOOR
		1926	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1927	#endif
		1928	#ifdef DT_WHT
		1929	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1930	#endif
		1931	}
		1932
		1933	ent->score = 7;
		1934
		1935	if (flags & EIO_READDIR_DIRS_FIRST)
		1936	{
		1937	if (ent->type == EIO_DT_UNKNOWN)
1598	{	1938	{
1599	if (dir->type == EIO_DT_DIR)	1939	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1600	++dir;	1940	ent->score = 1;
1601	else if ((--oth)->type == EIO_DT_DIR)	1941	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1602	{	1942	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1603	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1604
1605	++dir;
1606	}
1607	}	1943	}
1608		1944	else if (ent->type == EIO_DT_DIR)
1609	/* now sort the dirs only (dirs all have the same score) */	1945	ent->score = 0;
1610	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1611	}	1946	}
1612
1613	break;
1614	}
1615
1616	/* now add the entry to our list(s) */
1617	name = entp->d_name;
1618
1619	/* skip . and .. entries */
1620	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1621	{
1622	int len = D_NAMLEN (entp) + 1;
1623
1624	while (ecb_expect_false (namesoffs + len > namesalloc))
1625	{
1626	namesalloc *= 2;
1627	X_LOCK (wrklock);
1628	req->ptr2 = names = realloc (names, namesalloc);
1629	X_UNLOCK (wrklock);
1630
1631	if (!names)
1632	break;
1633	}	1947	}
1634		1948
1635	memcpy (names + namesoffs, name, len);
1636
1637	if (dents)
1638	{
1639	struct eio_dirent *ent;
1640
1641	if (ecb_expect_false (dentoffs == dentalloc))
1642	{
1643	dentalloc *= 2;
1644	X_LOCK (wrklock);
1645	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1646	X_UNLOCK (wrklock);
1647
1648	if (!dents)
1649	break;
1650	}
1651
1652	ent = dents + dentoffs;
1653
1654	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1655	ent->namelen = len - 1;
1656	ent->inode = D_INO (entp);
1657
1658	inode_bits |= ent->inode;
1659
1660	switch (D_TYPE (entp))
1661	{
1662	default:
1663	ent->type = EIO_DT_UNKNOWN;
1664	flags |= EIO_READDIR_FOUND_UNKNOWN;
1665	break;
1666
1667	#ifdef DT_FIFO
1668	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1669	#endif
1670	#ifdef DT_CHR
1671	case DT_CHR: ent->type = EIO_DT_CHR; break;
1672	#endif
1673	#ifdef DT_MPC
1674	case DT_MPC: ent->type = EIO_DT_MPC; break;
1675	#endif
1676	#ifdef DT_DIR
1677	case DT_DIR: ent->type = EIO_DT_DIR; break;
1678	#endif
1679	#ifdef DT_NAM
1680	case DT_NAM: ent->type = EIO_DT_NAM; break;
1681	#endif
1682	#ifdef DT_BLK
1683	case DT_BLK: ent->type = EIO_DT_BLK; break;
1684	#endif
1685	#ifdef DT_MPB
1686	case DT_MPB: ent->type = EIO_DT_MPB; break;
1687	#endif
1688	#ifdef DT_REG
1689	case DT_REG: ent->type = EIO_DT_REG; break;
1690	#endif
1691	#ifdef DT_NWK
1692	case DT_NWK: ent->type = EIO_DT_NWK; break;
1693	#endif
1694	#ifdef DT_CMP
1695	case DT_CMP: ent->type = EIO_DT_CMP; break;
1696	#endif
1697	#ifdef DT_LNK
1698	case DT_LNK: ent->type = EIO_DT_LNK; break;
1699	#endif
1700	#ifdef DT_SOCK
1701	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1702	#endif
1703	#ifdef DT_DOOR
1704	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1705	#endif
1706	#ifdef DT_WHT
1707	case DT_WHT: ent->type = EIO_DT_WHT; break;
1708	#endif
1709	}
1710
1711	ent->score = 7;
1712
1713	if (flags & EIO_READDIR_DIRS_FIRST)
1714	{
1715	if (ent->type == EIO_DT_UNKNOWN)
1716	{
1717	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1718	ent->score = 1;
1719	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1720	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1721	}
1722	else if (ent->type == EIO_DT_DIR)
1723	ent->score = 0;
1724	}
1725	}
1726
1727	namesoffs += len;	1949	namesoffs += len;
1728	++dentoffs;	1950	++dentoffs;
1729	}	1951	}
1730		1952
1731	if (EIO_CANCELLED (req))	1953	if (EIO_CANCELLED (req))
1732	{	1954	{
1733	errno = ECANCELED;	1955	errno = ECANCELED;
1734	break;	1956	break;
1735	}	1957	}
		1958
		1959	#ifdef _WIN32
		1960	if (!FindNextFile (dirp, &entp))
		1961	{
		1962	FindClose (dirp);
		1963	dirp = 0;
1736	}	1964	}
		1965	#endif
		1966	}
1737	}	1967	}
		1968
		1969	/*****************************************************************************/
		1970	/* working directory stuff */
		1971
		1972	#if HAVE_AT
		1973
		1974	#define WD2FD(wd) ((wd) ? ((int)(long)(wd)) - 1 : AT_FDCWD)
		1975
		1976	#ifndef O_SEARCH
		1977	# define O_SEARCH O_RDONLY
		1978	#endif
		1979
		1980	eio_wd
		1981	eio_wd_open_sync (eio_wd wd, const char *path)
		1982	{
		1983	int fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1984
		1985	return fd >= 0 ? (eio_wd)(long)(fd + 1) : EIO_INVALID_WD;
		1986	}
		1987
		1988	static eio_wd
		1989	eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1990	{
		1991	return eio_wd_open_sync (wd, path);
		1992	}
		1993
		1994	void
		1995	eio_wd_close_sync (eio_wd wd)
		1996	{
		1997	int fd = WD2FD (wd);
		1998
		1999	if (fd >= 0)
		2000	close (fd);
		2001	}
		2002
		2003	static int
		2004	eio__truncateat (int dirfd, const char *path, off_t length)
		2005	{
		2006	int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
		2007	int res;
		2008
		2009	if (fd < 0)
		2010	return fd;
		2011
		2012	res = ftruncate (fd, length);
		2013	close (fd);
		2014	return res;
		2015	}
		2016
		2017	static int
		2018	eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
		2019	{
		2020	int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
		2021	int res;
		2022
		2023	if (fd < 0)
		2024	return fd;
		2025
		2026	res = fstatvfs (fd, buf);
		2027	close (fd);
		2028	return res;
		2029
		2030	}
		2031
		2032	#else
		2033
		2034	/* on legacy systems, we represent the working directories simply by their path strings */
		2035
		2036	static const char *
		2037	wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2038	{
		2039	if (!wd || *path == '/')
		2040	return path;
		2041
		2042	{
		2043	int l1 = strlen ((const char *)wd);
		2044	int l2 = strlen (path);
		2045
		2046	char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
		2047
		2048	memcpy (res, wd, l1);
		2049	res [l1] = '/';
		2050	memcpy (res + l1 + 1, path, l2 + 1);
		2051
		2052	return res;
		2053	}
		2054	}
		2055
		2056	static eio_wd
		2057	eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2058	{
		2059	if (*path == '/') /* absolute paths ignore wd */
		2060	path = strdup (path);
		2061	else if (path [0] == '.' && !path [1]) /* special case '.', as it is common */
		2062	return wd;
		2063	else
		2064	{
		2065	int len = eio__realpath (tmpbuf, wd, path);
		2066
		2067	path = EIO_INVALID_WD;
		2068
		2069	if (len >= 0)
		2070	{
		2071	((char *)tmpbuf->ptr)[len] = 0;
		2072	path = strdup (tmpbuf->ptr);
		2073	}
		2074	}
		2075
		2076	if (!path)
		2077	path = EIO_INVALID_WD;
		2078
		2079	return (eio_wd)path;
		2080	}
		2081
		2082	eio_wd
		2083	eio_wd_open_sync (eio_wd wd, const char *path)
		2084	{
		2085	struct tmpbuf tmpbuf = { 0 };
		2086	wd = eio__wd_open_sync (&tmpbuf, wd, path);
		2087	free (tmpbuf.ptr);
		2088
		2089	return wd;
		2090	}
		2091
		2092	void
		2093	eio_wd_close_sync (eio_wd wd)
		2094	{
		2095	if (wd != EIO_INVALID_WD)
		2096	free (wd);
		2097	}
		2098
		2099	#endif
1738		2100
1739	/*****************************************************************************/	2101	/*****************************************************************************/
1740		2102
1741	#define ALLOC(len) \	2103	#define ALLOC(len) \
1742	if (!req->ptr2) \	2104	if (!req->ptr2) \
…		…
1757	{	2119	{
1758	ETP_REQ *req;	2120	ETP_REQ *req;
1759	struct timespec ts;	2121	struct timespec ts;
1760	etp_worker *self = (etp_worker *)thr_arg;	2122	etp_worker *self = (etp_worker *)thr_arg;
1761		2123
		2124	#if HAVE_PRCTL_SET_NAME
		2125	prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
		2126	#endif
		2127
1762	/* try to distribute timeouts somewhat randomly */	2128	/* try to distribute timeouts somewhat evenly */
1763	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	2129	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1764		2130
1765	for (;;)	2131	for (;;)
1766	{	2132	{
		2133	ts.tv_sec = 0;
		2134
1767	X_LOCK (reqlock);	2135	X_LOCK (reqlock);
1768		2136
1769	for (;;)	2137	for (;;)
1770	{	2138	{
1771	self->req = req = reqq_shift (&req_queue);	2139	req = reqq_shift (&req_queue);
1772		2140
1773	if (req)	2141	if (req)
1774	break;	2142	break;
1775		2143
		2144	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		2145	{
		2146	X_UNLOCK (reqlock);
		2147	X_LOCK (wrklock);
		2148	--started;
		2149	X_UNLOCK (wrklock);
		2150	goto quit;
		2151	}
		2152
1776	++idle;	2153	++idle;
1777		2154
1778	ts.tv_sec = time (0) + idle_timeout;	2155	if (idle <= max_idle)
1779	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	2156	/* we are allowed to idle, so do so without any timeout */
		2157	X_COND_WAIT (reqwait, reqlock);
		2158	else
1780	{	2159	{
1781	if (idle > max_idle)	2160	/* initialise timeout once */
1782	{	2161	if (!ts.tv_sec)
1783	--idle;	2162	ts.tv_sec = time (0) + idle_timeout;
1784	X_UNLOCK (reqlock);
1785	X_LOCK (wrklock);
1786	--started;
1787	X_UNLOCK (wrklock);
1788	goto quit;
1789	}
1790		2163
1791	/* we are allowed to idle, so do so without any timeout */
1792	X_COND_WAIT (reqwait, reqlock);	2164	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		2165	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1793	}	2166	}
1794		2167
1795	--idle;	2168	--idle;
1796	}	2169	}
1797		2170
…		…
1800	X_UNLOCK (reqlock);	2173	X_UNLOCK (reqlock);
1801		2174
1802	if (req->type < 0)	2175	if (req->type < 0)
1803	goto quit;	2176	goto quit;
1804		2177
1805	if (!EIO_CANCELLED (req))
1806	ETP_EXECUTE (self, req);	2178	ETP_EXECUTE (self, req);
1807		2179
1808	X_LOCK (reslock);	2180	X_LOCK (reslock);
1809		2181
1810	++npending;	2182	++npending;
1811		2183
1812	if (!reqq_push (&res_queue, req) && want_poll_cb)	2184	if (!reqq_push (&res_queue, req) && want_poll_cb)
1813	want_poll_cb ();	2185	want_poll_cb ();
1814		2186
1815	self->req = 0;
1816	etp_worker_clear (self);	2187	etp_worker_clear (self);
1817		2188
1818	X_UNLOCK (reslock);	2189	X_UNLOCK (reslock);
1819	}	2190	}
1820		2191
1821	quit:	2192	quit:
		2193	free (req);
		2194
1822	X_LOCK (wrklock);	2195	X_LOCK (wrklock);
1823	etp_worker_free (self);	2196	etp_worker_free (self);
1824	X_UNLOCK (wrklock);	2197	X_UNLOCK (wrklock);
1825		2198
1826	return 0;	2199	return 0;
…		…
1829	/*****************************************************************************/	2202	/*****************************************************************************/
1830		2203
1831	int ecb_cold	2204	int ecb_cold
1832	eio_init (void (*want_poll)(void), void (*done_poll)(void))	2205	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1833	{	2206	{
		2207	#if !HAVE_PREADWRITE
		2208	X_MUTEX_CREATE (preadwritelock);
		2209	#endif
		2210
1834	return etp_init (want_poll, done_poll);	2211	return etp_init (want_poll, done_poll);
1835	}	2212	}
1836		2213
1837	ecb_inline void	2214	ecb_inline void
1838	eio_api_destroy (eio_req *req)	2215	eio_api_destroy (eio_req *req)
…		…
1865	}	2242	}
1866		2243
1867	static void	2244	static void
1868	eio_execute (etp_worker *self, eio_req *req)	2245	eio_execute (etp_worker *self, eio_req *req)
1869	{	2246	{
		2247	#if HAVE_AT
		2248	int dirfd;
		2249	#else
		2250	const char *path;
		2251	#endif
		2252
		2253	if (ecb_expect_false (EIO_CANCELLED (req)))
		2254	{
		2255	req->result = -1;
		2256	req->errorno = ECANCELED;
		2257	return;
		2258	}
		2259
		2260	if (ecb_expect_false (req->wd == EIO_INVALID_WD))
		2261	{
		2262	req->result = -1;
		2263	req->errorno = ENOENT;
		2264	return;
		2265	}
		2266
		2267	if (req->type >= EIO_OPEN)
		2268	{
		2269	#if HAVE_AT
		2270	dirfd = WD2FD (req->wd);
		2271	#else
		2272	path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		2273	#endif
		2274	}
		2275
1870	switch (req->type)	2276	switch (req->type)
1871	{	2277	{
		2278	case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
		2279	req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
		2280	break;
		2281	case EIO_WD_CLOSE: req->result = 0;
		2282	eio_wd_close_sync (req->wd); break;
		2283
1872	case EIO_READ: ALLOC (req->size);	2284	case EIO_READ: ALLOC (req->size);
1873	req->result = req->offs >= 0	2285	req->result = req->offs >= 0
1874	? pread (req->int1, req->ptr2, req->size, req->offs)	2286	? pread (req->int1, req->ptr2, req->size, req->offs)
1875	: read (req->int1, req->ptr2, req->size); break;	2287	: read (req->int1, req->ptr2, req->size); break;
1876	case EIO_WRITE: req->result = req->offs >= 0	2288	case EIO_WRITE: req->result = req->offs >= 0
1877	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2289	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1878	: write (req->int1, req->ptr2, req->size); break;	2290	: write (req->int1, req->ptr2, req->size); break;
1879		2291
1880	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2292	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1881	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2293	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
		2294
		2295	#if HAVE_AT
1882		2296
1883	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2297	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1884	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2298	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1885	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2299	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2300	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
		2301	case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
		2302	case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
		2303	case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
		2304	case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
		2305
		2306	case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
		2307	case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
		2308	case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
		2309	case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2); break;
		2310	case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2, 0); break;
		2311	case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
		2312	case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2313	case EIO_READLINK: ALLOC (PATH_MAX);
		2314	req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
		2315	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2316	req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2317	case EIO_UTIME:
		2318	case EIO_FUTIME:
		2319	{
		2320	struct timespec ts[2];
		2321	struct timespec *times;
		2322
		2323	if (req->nv1 != -1. || req->nv2 != -1.)
		2324	{
		2325	ts[0].tv_sec = req->nv1;
		2326	ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
		2327	ts[1].tv_sec = req->nv2;
		2328	ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
		2329
		2330	times = ts;
		2331	}
		2332	else
		2333	times = 0;
		2334
		2335	req->result = req->type == EIO_FUTIME
		2336	? futimens (req->int1, times)
		2337	: utimensat (dirfd, req->ptr1, times, 0);
		2338	}
		2339	break;
		2340
		2341	#else
		2342
		2343	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2344	req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2345	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1886	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2346	req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2347	case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
		2348	case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
		2349	case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
		2350	case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
		2351
		2352	case EIO_UNLINK: req->result = unlink (path ); break;
		2353	case EIO_RMDIR: req->result = rmdir (path ); break;
		2354	case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
		2355	case EIO_RENAME: req->result = rename (path , req->ptr2); break;
		2356	case EIO_LINK: req->result = link (path , req->ptr2); break;
		2357	case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
		2358	case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
		2359	case EIO_READLINK: ALLOC (PATH_MAX);
		2360	req->result = readlink (path, req->ptr2, PATH_MAX); break;
		2361	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2362	req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2363
		2364	case EIO_UTIME:
		2365	case EIO_FUTIME:
		2366	{
		2367	struct timeval tv[2];
		2368	struct timeval *times;
		2369
		2370	if (req->nv1 != -1. || req->nv2 != -1.)
		2371	{
		2372	tv[0].tv_sec = req->nv1;
		2373	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
		2374	tv[1].tv_sec = req->nv2;
		2375	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
		2376
		2377	times = tv;
		2378	}
		2379	else
		2380	times = 0;
		2381
		2382	req->result = req->type == EIO_FUTIME
		2383	? futimes (req->int1, times)
		2384	: utimes (req->ptr1, times);
		2385	}
		2386	break;
		2387
		2388	#endif
		2389
		2390	case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
		2391	{
		2392	ALLOC (req->result);
		2393	memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
		2394	}
		2395	break;
		2396
1887	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2397	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1888	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2398	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1889		2399
1890	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1891	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1892	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));	2400	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1893	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;	2401	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1894		2402
1895	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1896	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	2403	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1897	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1898	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	2404	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1899	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1900	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;	2405	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1901		2406
1902	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1903	case EIO_CLOSE: req->result = close (req->int1); break;	2407	case EIO_CLOSE: req->result = close (req->int1); break;
1904	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;	2408	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1905	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1906	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1907	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1908	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1909	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1910	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1911	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1912
1913	case EIO_REALPATH: eio__realpath (req, self); break;
1914
1915	case EIO_READLINK: ALLOC (PATH_MAX);
1916	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1917
1918	case EIO_SYNC: req->result = 0; sync (); break;	2409	case EIO_SYNC: req->result = 0; sync (); break;
1919	case EIO_FSYNC: req->result = fsync (req->int1); break;	2410	case EIO_FSYNC: req->result = fsync (req->int1); break;
1920	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2411	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2412	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2413	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1921	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2414	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1922	case EIO_MTOUCH: req->result = eio__mtouch (req); break;	2415	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1923	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2416	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1924	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2417	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1925	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	2418	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1926		2419
1927	case EIO_READDIR: eio__scandir (req, self); break;	2420	case EIO_READDIR: eio__scandir (req); break;
1928		2421
1929	case EIO_BUSY:	2422	case EIO_BUSY:
1930	#ifdef _WIN32	2423	#ifdef _WIN32
1931	Sleep (req->nv1 * 1e3);	2424	Sleep (req->nv1 * 1e3);
1932	#else	2425	#else
…		…
1939	req->result = select (0, 0, 0, 0, &tv);	2432	req->result = select (0, 0, 0, 0, &tv);
1940	}	2433	}
1941	#endif	2434	#endif
1942	break;	2435	break;
1943		2436
1944	case EIO_UTIME:
1945	case EIO_FUTIME:
1946	{
1947	struct timeval tv[2];
1948	struct timeval *times;
1949
1950	if (req->nv1 != -1. || req->nv2 != -1.)
1951	{
1952	tv[0].tv_sec = req->nv1;
1953	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1954	tv[1].tv_sec = req->nv2;
1955	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1956
1957	times = tv;
1958	}
1959	else
1960	times = 0;
1961
1962	req->result = req->type == EIO_FUTIME
1963	? futimes (req->int1, times)
1964	: utimes (req->ptr1, times);
1965	}
1966	break;
1967
1968	case EIO_GROUP:	2437	case EIO_GROUP:
1969	abort (); /* handled in eio_request */	2438	abort (); /* handled in eio_request */
1970		2439
1971	case EIO_NOP:	2440	case EIO_NOP:
1972	req->result = 0;	2441	req->result = 0;
…		…
2010	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2479	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2011	{	2480	{
2012	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2481	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2013	}	2482	}
2014		2483
		2484	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		2485	{
		2486	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2487	}
		2488
		2489	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2490	{
		2491	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2492	}
		2493
		2494	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2495	{
		2496	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2497	}
		2498
2015	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2499	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2016	{	2500	{
2017	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2501	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2018	}	2502	}
2019		2503
…		…
2025	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)	2509	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2026	{	2510	{
2027	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;	2511	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2028	}	2512	}
2029		2513
2030	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2514	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2031	{	2515	{
2032	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2516	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2033	}
2034
2035	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2036	{
2037	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2038	}	2517	}
2039		2518
2040	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2519	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2041	{	2520	{
2042	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2521	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
2080	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2559	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2081	{	2560	{
2082	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2561	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2083	}	2562	}
2084		2563
2085	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2564	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2086	{	2565	{
2087	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2566	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2088	}	2567	}
2089		2568
2090	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2569	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
2110	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2589	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2111	{	2590	{
2112	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2591	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2113	}	2592	}
2114		2593
2115	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2594	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2116	{	2595	{
2117	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2596	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2118	}	2597	}
2119		2598
2120	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2599	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
2267	}	2746	}
2268		2747
2269	/*****************************************************************************/	2748	/*****************************************************************************/
2270	/* misc garbage */	2749	/* misc garbage */
2271		2750
2272	ssize_t	2751	eio_ssize_t
2273	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2752	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2274	{	2753	{
2275	etp_worker wrk;
2276	ssize_t ret;
2277
2278	wrk.dbuf = 0;
2279
2280	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2754	return eio__sendfile (ofd, ifd, offset, count);
2281
2282	if (wrk.dbuf)
2283	free (wrk.dbuf);
2284
2285	return ret;
2286	}	2755	}
2287		2756

Diff Legend

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