ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.73 by root, Fri Jun 17 18:55:42 2011 UTC vs.
Revision 1.120 by root, Tue Apr 24 18:47:50 2012 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>	4	* Copyright (c) 2007,2008,2009,2010,2011,2012 Marc Alexander Lehmann <libeio@schmorp.de>
5	* All rights reserved.	5	* All rights reserved.
6	*	6	*
7	* Redistribution and use in source and binary forms, with or without modifica-	7	* Redistribution and use in source and binary forms, with or without modifica-
8	* tion, are permitted provided that the following conditions are met:	8	* tion, are permitted provided that the following conditions are met:
9	*	9	*
…		…
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
64	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */	63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65	/* intptr_t only comes form stdint.h, says idiot openbsd coder */	64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66	#if HAVE_STDINT_H	65	#if HAVE_STDINT_H
67	# include <stdint.h>	66	# include <stdint.h>
68	#endif	67	#endif
69		68
		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
70	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
71	# 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
72	#endif	84	#endif
73		85
74	#ifndef EIO_DESTROY	86	#ifndef EIO_DESTROY
…		…
77		89
78	#ifndef EIO_FEED	90	#ifndef EIO_FEED
79	# 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)
80	#endif	92	#endif
81		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
82	#ifdef _WIN32	105	#ifdef _WIN32
83		106
84	/*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
85	#else	197	#else
86		198
87	# include <sys/time.h>	199	#include <sys/time.h>
88	# include <sys/select.h>	200	#include <sys/select.h>
		201	#include <sys/statvfs.h>
89	# 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__ || __NetBSD__ || __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
90	# include <utime.h>	237	# 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	238	#endif
106		239
107	#ifdef _D_EXACT_NAMLEN	240	#if HAVE_SYS_SYSCALL_H
108	# undef D_NAMLEN	241	# 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	242	#endif
115		243
116	# ifndef EIO_STRUCT_DIRENT	244	#if HAVE_SYS_PRCTL_H
117	# define EIO_STRUCT_DIRENT struct dirent	245	# include <sys/prctl.h>
118	# endif
119
120	#endif	246	#endif
121		247
122	#if HAVE_SENDFILE	248	#if HAVE_SENDFILE
123	# if __linux	249	# if __linux
124	# include <sys/sendfile.h>	250	# include <sys/sendfile.h>
…		…
139	#endif	265	#endif
140	#ifndef D_INO	266	#ifndef D_INO
141	# define D_INO(de) 0	267	# define D_INO(de) 0
142	#endif	268	#endif
143	#ifndef D_NAMLEN	269	#ifndef D_NAMLEN
144	# define D_NAMLEN(de) strlen ((de)->d_name)	270	# define D_NAMLEN(entp) strlen (D_NAME (entp))
145	#endif	271	#endif
146		272
147	/* used for struct dirent, AIX doesn't provide it */	273	/* used for struct dirent, AIX doesn't provide it */
148	#ifndef NAME_MAX	274	#ifndef NAME_MAX
149	# define NAME_MAX 4096	275	# define NAME_MAX 4096
…		…
156		282
157	/* buffer size for various temporary buffers */	283	/* buffer size for various temporary buffers */
158	#define EIO_BUFSIZE 65536	284	#define EIO_BUFSIZE 65536
159		285
160	#define dBUF \	286	#define dBUF \
161	char *eio_buf; \
162	ETP_WORKER_LOCK (self); \
163	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	287	char *eio_buf = malloc (EIO_BUFSIZE); \
164	ETP_WORKER_UNLOCK (self); \
165	errno = ENOMEM; \	288	errno = ENOMEM; \
166	if (!eio_buf) \	289	if (!eio_buf) \
167	return -1;	290	return -1
		291
		292	#define FUBd \
		293	free (eio_buf)
168		294
169	#define EIO_TICKS ((1000000 + 1023) >> 10)	295	#define EIO_TICKS ((1000000 + 1023) >> 10)
		296
		297	/*****************************************************************************/
		298
		299	struct tmpbuf
		300	{
		301	void *ptr;
		302	int len;
		303	};
		304
		305	static void *
		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	}
		316
		317	struct tmpbuf;
		318
		319	#if _POSIX_VERSION >= 200809L
		320	#define HAVE_AT 1
		321	#define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
		322	#ifndef O_SEARCH
		323	#define O_SEARCH O_RDONLY
		324	#endif
		325	#else
		326	#define HAVE_AT 0
		327	static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
		328	#endif
		329
		330	struct eio_pwd
		331	{
		332	#if HAVE_AT
		333	int fd;
		334	#endif
		335	int len;
		336	char str[1]; /* actually, a 0-terminated canonical path */
		337	};
		338
		339	/*****************************************************************************/
170		340
171	#define ETP_PRI_MIN EIO_PRI_MIN	341	#define ETP_PRI_MIN EIO_PRI_MIN
172	#define ETP_PRI_MAX EIO_PRI_MAX	342	#define ETP_PRI_MAX EIO_PRI_MAX
173		343
174	struct etp_worker;	344	struct etp_worker;
…		…
178	static int eio_finish (eio_req *req);	348	static int eio_finish (eio_req *req);
179	#define ETP_FINISH(req) eio_finish (req)	349	#define ETP_FINISH(req) eio_finish (req)
180	static void eio_execute (struct etp_worker *self, eio_req *req);	350	static void eio_execute (struct etp_worker *self, eio_req *req);
181	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	351	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
182		352
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	/*****************************************************************************/	353	/*****************************************************************************/
201		354
202	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	355	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
203		356
204	/* calculate time difference in ~1/EIO_TICKS of a second */	357	/* calculate time difference in ~1/EIO_TICKS of a second */
…		…
215	static void (*done_poll_cb) (void);	368	static void (*done_poll_cb) (void);
216		369
217	static unsigned int max_poll_time; /* reslock */	370	static unsigned int max_poll_time; /* reslock */
218	static unsigned int max_poll_reqs; /* reslock */	371	static unsigned int max_poll_reqs; /* reslock */
219		372
220	static volatile unsigned int nreqs; /* reqlock */	373	static unsigned int nreqs; /* reqlock */
221	static volatile unsigned int nready; /* reqlock */	374	static unsigned int nready; /* reqlock */
222	static volatile unsigned int npending; /* reqlock */	375	static unsigned int npending; /* reqlock */
223	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */	376	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 */	377	static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
225		378
226	static xmutex_t wrklock;	379	static xmutex_t wrklock;
227	static xmutex_t reslock;	380	static xmutex_t reslock;
228	static xmutex_t reqlock;	381	static xmutex_t reqlock;
229	static xcond_t reqwait;	382	static xcond_t reqwait;
…		…
232	/*	385	/*
233	* make our pread/pwrite emulation safe against themselves, but not against	386	* make our pread/pwrite emulation safe against themselves, but not against
234	* normal read/write by using a mutex. slows down execution a lot,	387	* normal read/write by using a mutex. slows down execution a lot,
235	* but that's your problem, not mine.	388	* but that's your problem, not mine.
236	*/	389	*/
237	static xmutex_t preadwritelock = X_MUTEX_INIT;	390	static xmutex_t preadwritelock;
238	#endif	391	#endif
239		392
240	typedef struct etp_worker	393	typedef struct etp_worker
241	{	394	{
		395	struct tmpbuf tmpbuf;
		396
242	/* locked by wrklock */	397	/* locked by wrklock */
243	struct etp_worker *prev, *next;	398	struct etp_worker *prev, *next;
244		399
245	xthread_t tid;	400	xthread_t tid;
246		401
247	/* locked by reslock, reqlock or wrklock */	402	#ifdef ETP_WORKER_COMMON
248	ETP_REQ *req; /* currently processed request */
249
250	ETP_WORKER_COMMON	403	ETP_WORKER_COMMON
		404	#endif
251	} etp_worker;	405	} etp_worker;
252		406
253	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	407	static etp_worker wrk_first; /* NOT etp */
254		408
255	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	409	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
256	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	410	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
257		411
258	/* worker threads management */	412	/* worker threads management */
259		413
260	static void ecb_cold	414	static void
261	etp_worker_clear (etp_worker *wrk)	415	etp_worker_clear (etp_worker *wrk)
262	{	416	{
263	ETP_WORKER_CLEAR (wrk);
264	}	417	}
265		418
266	static void ecb_cold	419	static void ecb_cold
267	etp_worker_free (etp_worker *wrk)	420	etp_worker_free (etp_worker *wrk)
268	{	421	{
		422	free (wrk->tmpbuf.ptr);
		423
269	wrk->next->prev = wrk->prev;	424	wrk->next->prev = wrk->prev;
270	wrk->prev->next = wrk->next;	425	wrk->prev->next = wrk->next;
271		426
272	free (wrk);	427	free (wrk);
273	}	428	}
…		…
329	} etp_reqq;	484	} etp_reqq;
330		485
331	static etp_reqq req_queue;	486	static etp_reqq req_queue;
332	static etp_reqq res_queue;	487	static etp_reqq res_queue;
333		488
		489	static void ecb_noinline ecb_cold
		490	reqq_init (etp_reqq *q)
		491	{
		492	int pri;
		493
		494	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		495	q->qs[pri] = q->qe[pri] = 0;
		496
		497	q->size = 0;
		498	}
		499
334	static int ecb_noinline	500	static int ecb_noinline
335	reqq_push (etp_reqq *q, ETP_REQ *req)	501	reqq_push (etp_reqq *q, ETP_REQ *req)
336	{	502	{
337	int pri = req->pri;	503	int pri = req->pri;
338	req->next = 0;	504	req->next = 0;
…		…
372	}	538	}
373		539
374	abort ();	540	abort ();
375	}	541	}
376		542
377	static void ecb_cold	543	static int ecb_cold
378	etp_thread_init (void)	544	etp_init (void (*want_poll)(void), void (*done_poll)(void))
379	{	545	{
380	X_MUTEX_CREATE (wrklock);	546	X_MUTEX_CREATE (wrklock);
381	X_MUTEX_CREATE (reslock);	547	X_MUTEX_CREATE (reslock);
382	X_MUTEX_CREATE (reqlock);	548	X_MUTEX_CREATE (reqlock);
383	X_COND_CREATE (reqwait);	549	X_COND_CREATE (reqwait);
384	}
385		550
386	static void ecb_cold	551	reqq_init (&req_queue);
387	etp_atfork_prepare (void)	552	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		553
397	static void ecb_cold	554	wrk_first.next =
398	etp_atfork_parent (void)	555	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		556
430	started = 0;	557	started = 0;
431	idle = 0;	558	idle = 0;
432	nreqs = 0;	559	nreqs = 0;
433	nready = 0;	560	nready = 0;
434	npending = 0;	561	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		562
453	want_poll_cb = want_poll;	563	want_poll_cb = want_poll;
454	done_poll_cb = done_poll;	564	done_poll_cb = done_poll;
455		565
456	return 0;	566	return 0;
…		…
496	}	606	}
497		607
498	static void ecb_cold	608	static void ecb_cold
499	etp_end_thread (void)	609	etp_end_thread (void)
500	{	610	{
501	eio_req *req = calloc (1, sizeof (eio_req));	611	eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
502		612
503	req->type = -1;	613	req->type = -1;
504	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	614	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
505		615
506	X_LOCK (reqlock);	616	X_LOCK (reqlock);
…		…
583	}	693	}
584		694
585	static void	695	static void
586	etp_cancel (ETP_REQ *req)	696	etp_cancel (ETP_REQ *req)
587	{	697	{
588	X_LOCK (wrklock);	698	req->cancelled = 1;
589	req->flags |= EIO_FLAG_CANCELLED;
590	X_UNLOCK (wrklock);
591		699
592	eio_grp_cancel (req);	700	eio_grp_cancel (req);
593	}	701	}
594		702
595	static void	703	static void
…		…
711	return eio_finish (grp);	819	return eio_finish (grp);
712	else	820	else
713	return 0;	821	return 0;
714	}	822	}
715		823
716	void	824	static void
717	eio_destroy (eio_req *req)	825	eio_destroy (eio_req *req)
718	{	826	{
719	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	827	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
720	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	828	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
721		829
…		…
739	if (grp->grp_first == req)	847	if (grp->grp_first == req)
740	grp->grp_first = req->grp_next;	848	grp->grp_first = req->grp_next;
741		849
742	res2 = grp_dec (grp);	850	res2 = grp_dec (grp);
743		851
744	if (!res && res2)	852	if (!res)
745	res = res2;	853	res = res2;
746	}	854	}
747		855
748	eio_destroy (req);	856	eio_destroy (req);
749		857
…		…
841	# undef pread	949	# undef pread
842	# undef pwrite	950	# undef pwrite
843	# define pread eio__pread	951	# define pread eio__pread
844	# define pwrite eio__pwrite	952	# define pwrite eio__pwrite
845		953
846	static ssize_t	954	static eio_ssize_t
847	eio__pread (int fd, void *buf, size_t count, off_t offset)	955	eio__pread (int fd, void *buf, size_t count, off_t offset)
848	{	956	{
849	ssize_t res;	957	eio_ssize_t res;
850	off_t ooffset;	958	off_t ooffset;
851		959
852	X_LOCK (preadwritelock);	960	X_LOCK (preadwritelock);
853	ooffset = lseek (fd, 0, SEEK_CUR);	961	ooffset = lseek (fd, 0, SEEK_CUR);
854	lseek (fd, offset, SEEK_SET);	962	lseek (fd, offset, SEEK_SET);
…		…
857	X_UNLOCK (preadwritelock);	965	X_UNLOCK (preadwritelock);
858		966
859	return res;	967	return res;
860	}	968	}
861		969
862	static ssize_t	970	static eio_ssize_t
863	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	971	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
864	{	972	{
865	ssize_t res;	973	eio_ssize_t res;
866	off_t ooffset;	974	off_t ooffset;
867		975
868	X_LOCK (preadwritelock);	976	X_LOCK (preadwritelock);
869	ooffset = lseek (fd, 0, SEEK_CUR);	977	ooffset = lseek (fd, 0, SEEK_CUR);
870	lseek (fd, offset, SEEK_SET);	978	lseek (fd, offset, SEEK_SET);
…		…
915		1023
916	#if !HAVE_FDATASYNC	1024	#if !HAVE_FDATASYNC
917	# undef fdatasync	1025	# undef fdatasync
918	# define fdatasync(fd) fsync (fd)	1026	# define fdatasync(fd) fsync (fd)
919	#endif	1027	#endif
		1028
		1029	static int
		1030	eio__syncfs (int fd)
		1031	{
		1032	int res;
		1033
		1034	#if HAVE_SYS_SYNCFS
		1035	res = (int)syscall (__NR_syncfs, (int)(fd));
		1036	#else
		1037	res = EIO_ENOSYS ();
		1038	#endif
		1039
		1040	if (res < 0 && errno == ENOSYS && fd >= 0)
		1041	sync ();
		1042
		1043	return res;
		1044	}
920		1045
921	/* sync_file_range always needs emulation */	1046	/* sync_file_range always needs emulation */
922	static int	1047	static int
923	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	1048	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
924	{	1049	{
…		…
944	/* even though we could play tricks with the flags, it's better to always	1069	/* 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 */	1070	* call fdatasync, as that matches the expectation of its users best */
946	return fdatasync (fd);	1071	return fdatasync (fd);
947	}	1072	}
948		1073
		1074	static int
		1075	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		1076	{
		1077	#if HAVE_LINUX_FALLOCATE
		1078	return fallocate (fd, mode, offset, len);
		1079	#else
		1080	return EIO_ENOSYS ();
		1081	#endif
		1082	}
		1083
949	#if !HAVE_READAHEAD	1084	#if !HAVE_READAHEAD
950	# undef readahead	1085	# undef readahead
951	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1086	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
952		1087
953	static ssize_t	1088	static eio_ssize_t
954	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1089	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
955	{	1090	{
956	size_t todo = count;	1091	size_t todo = count;
957	dBUF;	1092	dBUF;
958		1093
…		…
963	pread (fd, eio_buf, len, offset);	1098	pread (fd, eio_buf, len, offset);
964	offset += len;	1099	offset += len;
965	todo -= len;	1100	todo -= len;
966	}	1101	}
967		1102
968	errno = 0;	1103	FUBd;
		1104
		1105	/* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
		1106	/* but not for e.g. EIO or eof, so we also never fail */
969	return count;	1107	return 0;
970	}	1108	}
971		1109
972	#endif	1110	#endif
973		1111
974	/* sendfile always needs emulation */	1112	/* sendfile always needs emulation */
975	static ssize_t	1113	static eio_ssize_t
976	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1114	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
977	{	1115	{
978	ssize_t written = 0;	1116	eio_ssize_t written = 0;
979	ssize_t res;	1117	eio_ssize_t res;
980		1118
981	if (!count)	1119	if (!count)
982	return 0;	1120	return 0;
983		1121
984	for (;;)	1122	for (;;)
985	{	1123	{
		1124	#ifdef __APPLE__
		1125	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1126	#endif
986	#if HAVE_SENDFILE	1127	#if HAVE_SENDFILE
987	# if __linux	1128	# if __linux
988	off_t soffset = offset;	1129	off_t soffset = offset;
989	res = sendfile (ofd, ifd, &soffset, count);	1130	res = sendfile (ofd, ifd, &soffset, count);
990		1131
…		…
1006		1147
1007	/* according to source inspection, this is correct, and useful behaviour */	1148	/* according to source inspection, this is correct, and useful behaviour */
1008	if (sbytes)	1149	if (sbytes)
1009	res = sbytes;	1150	res = sbytes;
1010		1151
1011	# elif defined (__APPLE__)	1152	# elif defined __APPLE__
1012	off_t sbytes = count;	1153	off_t sbytes = count;
1013	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	1154	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1014		1155
1015	/* according to the manpage, sbytes is always valid */	1156	/* according to the manpage, sbytes is always valid */
1016	if (sbytes)	1157	if (sbytes)
…		…
1033	if (res < 0 && sbytes)	1174	if (res < 0 && sbytes)
1034	res = sbytes;	1175	res = sbytes;
1035		1176
1036	# endif	1177	# endif
1037		1178
1038	#elif defined (_WIN32)	1179	#elif defined (_WIN32) && 0
1039	/* does not work, just for documentation of what would need to be done */	1180	/* 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, */	1181	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
1041	/* libeio guarantees that the file offset does not change, and windows */	1182	/* 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 */	1183	/* has no way to get an independent handle to the same file description */
1043	HANDLE h = TO_SOCKET (ifd);	1184	HANDLE h = TO_SOCKET (ifd);
1044	SetFilePointer (h, offset, 0, FILE_BEGIN);	1185	SetFilePointer (h, offset, 0, FILE_BEGIN);
1045	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);	1186	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1046		1187
1047	#else	1188	#else
1048	res = -1;	1189	res = EIO_ENOSYS ();
1049	errno = ENOSYS;
1050	#endif	1190	#endif
1051		1191
1052	/* we assume sendfile can copy at least 128mb in one go */	1192	/* we assume sendfile can copy at least 128mb in one go */
1053	if (res <= 128 * 1024 * 1024)	1193	if (res <= 128 * 1024 * 1024)
1054	{	1194	{
…		…
1076	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1216	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1077	/* BSDs */	1217	/* BSDs */
1078	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1218	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1079	|| errno == ENOTSUP	1219	|| errno == ENOTSUP
1080	#endif	1220	#endif
		1221	#ifdef EOPNOTSUPP /* windows */
1081	|| errno == EOPNOTSUPP /* BSDs */	1222	|| errno == EOPNOTSUPP /* BSDs */
		1223	#endif
1082	#if __solaris	1224	#if __solaris
1083	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1225	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1084	#endif	1226	#endif
1085	)	1227	)
1086	)	1228	)
…		…
1090		1232
1091	res = 0;	1233	res = 0;
1092		1234
1093	while (count)	1235	while (count)
1094	{	1236	{
1095	ssize_t cnt;	1237	eio_ssize_t cnt;
1096		1238
1097	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1239	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1098		1240
1099	if (cnt <= 0)	1241	if (cnt <= 0)
1100	{	1242	{
…		…
1112		1254
1113	offset += cnt;	1255	offset += cnt;
1114	res += cnt;	1256	res += cnt;
1115	count -= cnt;	1257	count -= cnt;
1116	}	1258	}
		1259
		1260	FUBd;
1117	}	1261	}
1118		1262
1119	return res;	1263	return res;
		1264	}
		1265
		1266	#ifdef PAGESIZE
		1267	# define eio_pagesize() PAGESIZE
		1268	#else
		1269	static intptr_t
		1270	eio_pagesize (void)
		1271	{
		1272	static intptr_t page;
		1273
		1274	if (!page)
		1275	page = sysconf (_SC_PAGESIZE);
		1276
		1277	return page;
		1278	}
		1279	#endif
		1280
		1281	static void
		1282	eio_page_align (void **addr, size_t *length)
		1283	{
		1284	intptr_t mask = eio_pagesize () - 1;
		1285
		1286	/* round down addr */
		1287	intptr_t adj = mask & (intptr_t)*addr;
		1288
		1289	*addr = (void *)((intptr_t)*addr - adj);
		1290	*length += adj;
		1291
		1292	/* round up length */
		1293	*length = (*length + mask) & ~mask;
		1294	}
		1295
		1296	#if !_POSIX_MEMLOCK
		1297	# define eio__mlockall(a) EIO_ENOSYS ()
		1298	#else
		1299
		1300	static int
		1301	eio__mlockall (int flags)
		1302	{
		1303	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1304	extern int mallopt (int, int);
		1305	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1306	#endif
		1307
		1308	if (EIO_MCL_CURRENT != MCL_CURRENT
		1309	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1310	{
		1311	flags = 0
		1312	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1313	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1314	}
		1315
		1316	return mlockall (flags);
		1317	}
		1318	#endif
		1319
		1320	#if !_POSIX_MEMLOCK_RANGE
		1321	# define eio__mlock(a,b) EIO_ENOSYS ()
		1322	#else
		1323
		1324	static int
		1325	eio__mlock (void *addr, size_t length)
		1326	{
		1327	eio_page_align (&addr, &length);
		1328
		1329	return mlock (addr, length);
		1330	}
		1331
		1332	#endif
		1333
		1334	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1335	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1336	#else
		1337
		1338	static int
		1339	eio__msync (void *mem, size_t len, int flags)
		1340	{
		1341	eio_page_align (&mem, &len);
		1342
		1343	if (EIO_MS_ASYNC != MS_SYNC
		1344	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1345	|| EIO_MS_SYNC != MS_SYNC)
		1346	{
		1347	flags = 0
		1348	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1349	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1350	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1351	}
		1352
		1353	return msync (mem, len, flags);
		1354	}
		1355
		1356	#endif
		1357
		1358	static int
		1359	eio__mtouch (eio_req *req)
		1360	{
		1361	void *mem = req->ptr2;
		1362	size_t len = req->size;
		1363	int flags = req->int1;
		1364
		1365	eio_page_align (&mem, &len);
		1366
		1367	{
		1368	intptr_t addr = (intptr_t)mem;
		1369	intptr_t end = addr + len;
		1370	intptr_t page = eio_pagesize ();
		1371
		1372	if (addr < end)
		1373	if (flags & EIO_MT_MODIFY) /* modify */
		1374	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1375	else
		1376	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1377	}
		1378
		1379	return 0;
		1380	}
		1381
		1382	/*****************************************************************************/
		1383	/* requests implemented outside eio_execute, because they are so large */
		1384
		1385	static void
		1386	eio__lseek (eio_req *req)
		1387	{
		1388	/* this usually gets optimised away completely, or your compiler sucks, */
		1389	/* or the whence constants really are not 0, 1, 2 */
		1390	int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
		1391	: req->int2 == EIO_SEEK_CUR ? SEEK_CUR
		1392	: req->int2 == EIO_SEEK_END ? SEEK_END
		1393	: req->int2;
		1394
		1395	req->offs = lseek (req->int1, req->offs, whence);
		1396	req->result = req->offs == (off_t)-1 ? -1 : 0;
		1397	}
		1398
		1399	/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
		1400	static int
		1401	eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1402	{
		1403	const char *rel = path;
		1404	char *res;
		1405	char *tmp1, *tmp2;
		1406	#if SYMLOOP_MAX > 32
		1407	int symlinks = SYMLOOP_MAX;
		1408	#else
		1409	int symlinks = 32;
		1410	#endif
		1411
		1412	errno = EINVAL;
		1413	if (!rel)
		1414	return -1;
		1415
		1416	errno = ENOENT;
		1417	if (!*rel)
		1418	return -1;
		1419
		1420	res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
		1421	tmp1 = res + PATH_MAX;
		1422	tmp2 = tmp1 + PATH_MAX;
		1423
		1424	#if 0 /* disabled, the musl way to do things is just too racy */
		1425	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1426	/* on linux we may be able to ask the kernel */
		1427	{
		1428	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1429
		1430	if (fd >= 0)
		1431	{
		1432	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1433	req->result = readlink (tmp1, res, PATH_MAX);
		1434	close (fd);
		1435
		1436	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1437
		1438	if (req->result > 0)
		1439	goto done;
		1440	}
		1441	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1442	return;
		1443	}
		1444	#endif
		1445	#endif
		1446
		1447	if (*rel != '/')
		1448	{
		1449	int len;
		1450
		1451	errno = ENOENT;
		1452	if (wd == EIO_INVALID_WD)
		1453	return -1;
		1454
		1455	if (wd == EIO_CWD)
		1456	{
		1457	if (!getcwd (res, PATH_MAX))
		1458	return -1;
		1459
		1460	len = strlen (res);
		1461	}
		1462	else
		1463	memcpy (res, wd->str, len = wd->len);
		1464
		1465	if (res [1]) /* only use if not / */
		1466	res += len;
		1467	}
		1468
		1469	while (*rel)
		1470	{
		1471	eio_ssize_t len, linklen;
		1472	const char *beg = rel;
		1473
		1474	while (*rel && *rel != '/')
		1475	++rel;
		1476
		1477	len = rel - beg;
		1478
		1479	if (!len) /* skip slashes */
		1480	{
		1481	++rel;
		1482	continue;
		1483	}
		1484
		1485	if (beg [0] == '.')
		1486	{
		1487	if (len == 1)
		1488	continue; /* . - nop */
		1489
		1490	if (beg [1] == '.' && len == 2)
		1491	{
		1492	/* .. - back up one component, if possible */
		1493
		1494	while (res != tmpbuf->ptr)
		1495	if (*--res == '/')
		1496	break;
		1497
		1498	continue;
		1499	}
		1500	}
		1501
		1502	errno = ENAMETOOLONG;
		1503	if (res + 1 + len + 1 >= tmp1)
		1504	return -1;
		1505
		1506	/* copy one component */
		1507	*res = '/';
		1508	memcpy (res + 1, beg, len);
		1509
		1510	/* zero-terminate, for readlink */
		1511	res [len + 1] = 0;
		1512
		1513	/* now check if it's a symlink */
		1514	linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
		1515
		1516	if (linklen < 0)
		1517	{
		1518	if (errno != EINVAL)
		1519	return -1;
		1520
		1521	/* it's a normal directory. hopefully */
		1522	res += len + 1;
		1523	}
		1524	else
		1525	{
		1526	/* yay, it was a symlink - build new path in tmp2 */
		1527	int rellen = strlen (rel);
		1528
		1529	errno = ENAMETOOLONG;
		1530	if (linklen + 1 + rellen >= PATH_MAX)
		1531	return -1;
		1532
		1533	errno = ELOOP;
		1534	if (!--symlinks)
		1535	return -1;
		1536
		1537	if (*tmp1 == '/')
		1538	res = tmpbuf->ptr; /* symlink resolves to an absolute path */
		1539
		1540	/* we need to be careful, as rel might point into tmp2 already */
		1541	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1542	tmp2 [linklen] = '/';
		1543	memcpy (tmp2, tmp1, linklen);
		1544
		1545	rel = tmp2;
		1546	}
		1547	}
		1548
		1549	/* special case for the lone root path */
		1550	if (res == tmpbuf->ptr)
		1551	*res++ = '/';
		1552
		1553	return res - (char *)tmpbuf->ptr;
1120	}	1554	}
1121		1555
1122	static signed char	1556	static signed char
1123	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1557	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1124	{	1558	{
…		…
1132		1566
1133	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1567	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1134	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1568	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1135		1569
1136	static void	1570	static void
1137	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1571	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1138	{	1572	{
1139	unsigned char bits [9 + sizeof (ino_t) * 8];	1573	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1140	unsigned char *bit = bits;	1574	unsigned char *bit = bits;
1141		1575
1142	assert (CHAR_BIT == 8);	1576	assert (CHAR_BIT == 8);
1143	assert (sizeof (eio_dirent) * 8 < 256);	1577	assert (sizeof (eio_dirent) * 8 < 256);
1144	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1578	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1146		1580
1147	if (size <= EIO_SORT_FAST)	1581	if (size <= EIO_SORT_FAST)
1148	return;	1582	return;
1149		1583
1150	/* first prepare an array of bits to test in our radix sort */	1584	/* first prepare an array of bits to test in our radix sort */
1151	/* try to take endianness into account, as well as differences in ino_t sizes */	1585	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1152	/* inode_bits must contain all inodes ORed together */	1586	/* inode_bits must contain all inodes ORed together */
1153	/* which is used to skip bits that are 0 everywhere, which is very common */	1587	/* which is used to skip bits that are 0 everywhere, which is very common */
1154	{	1588	{
1155	ino_t endianness;	1589	eio_ino_t endianness;
1156	int i, j;	1590	int i, j;
1157		1591
1158	/* we store the byte offset of byte n into byte n of "endianness" */	1592	/* we store the byte offset of byte n into byte n of "endianness" */
1159	for (i = 0; i < sizeof (ino_t); ++i)	1593	for (i = 0; i < sizeof (eio_ino_t); ++i)
1160	((unsigned char *)&endianness)[i] = i;	1594	((unsigned char *)&endianness)[i] = i;
1161		1595
1162	*bit++ = 0;	1596	*bit++ = 0;
1163		1597
1164	for (i = 0; i < sizeof (ino_t); ++i)	1598	for (i = 0; i < sizeof (eio_ino_t); ++i)
1165	{	1599	{
1166	/* shifting off the byte offsets out of "endianness" */	1600	/* shifting off the byte offsets out of "endianness" */
1167	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1601	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1168	endianness >>= 8;	1602	endianness >>= 8;
1169		1603
1170	for (j = 0; j < 8; ++j)	1604	for (j = 0; j < 8; ++j)
1171	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1605	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1172	*bit++ = offs + j;	1606	*bit++ = offs + j;
1173	}	1607	}
1174		1608
1175	for (j = 0; j < 8; ++j)	1609	for (j = 0; j < 8; ++j)
1176	if (score_bits & (1 << j))	1610	if (score_bits & (1 << j))
1177	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1611	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1178	}	1612	}
1179		1613
1180	/* now actually do the sorting (a variant of MSD radix sort) */	1614	/* now actually do the sorting (a variant of MSD radix sort) */
1181	{	1615	{
1182	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1616	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1183	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1617	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1184	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1618	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1185	int stk_idx = 0;	1619	int stk_idx = 0;
1186		1620
1187	base_stk [stk_idx] = dents;	1621	base_stk [stk_idx] = dents;
1188	end_stk [stk_idx] = dents + size;	1622	end_stk [stk_idx] = dents + size;
1189	bit_stk [stk_idx] = bit - 1;	1623	bit_stk [stk_idx] = bit - 1;
…		…
1268	}	1702	}
1269	}	1703	}
1270	}	1704	}
1271		1705
1272	static void	1706	static void
1273	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1707	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1274	{	1708	{
1275	if (size <= 1)	1709	if (size <= 1)
1276	return; /* our insertion sort relies on size > 0 */	1710	return; /* our insertion sort relies on size > 0 */
1277		1711
1278	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1712	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1286		1720
1287	/* read a full directory */	1721	/* read a full directory */
1288	static void	1722	static void
1289	eio__scandir (eio_req *req, etp_worker *self)	1723	eio__scandir (eio_req *req, etp_worker *self)
1290	{	1724	{
1291	DIR *dirp;
1292	EIO_STRUCT_DIRENT *entp;
1293	char *name, *names;	1725	char *name, *names;
1294	int namesalloc = 4096;	1726	int namesalloc = 4096 - sizeof (void *) * 4;
1295	int namesoffs = 0;	1727	int namesoffs = 0;
1296	int flags = req->int1;	1728	int flags = req->int1;
1297	eio_dirent *dents = 0;	1729	eio_dirent *dents = 0;
1298	int dentalloc = 128;	1730	int dentalloc = 128;
1299	int dentoffs = 0;	1731	int dentoffs = 0;
1300	ino_t inode_bits = 0;	1732	eio_ino_t inode_bits = 0;
		1733	#ifdef _WIN32
		1734	HANDLE dirp;
		1735	WIN32_FIND_DATA entp;
		1736	#else
		1737	DIR *dirp;
		1738	EIO_STRUCT_DIRENT *entp;
		1739	#endif
1301		1740
1302	req->result = -1;	1741	req->result = -1;
1303		1742
1304	if (!(flags & EIO_READDIR_DENTS))	1743	if (!(flags & EIO_READDIR_DENTS))
1305	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1744	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1306		1745
1307	X_LOCK (wrklock);	1746	#ifdef _WIN32
1308	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1747	{
		1748	int len = strlen ((const char *)req->ptr1);
		1749	char *path = malloc (MAX_PATH);
		1750	const char *fmt;
		1751	const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		1752
		1753	if (!len)
		1754	fmt = "./*";
		1755	else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
		1756	fmt = "%s*";
		1757	else
		1758	fmt = "%s/*";
		1759
		1760	_snprintf (path, MAX_PATH, fmt, reqpath);
		1761	dirp = FindFirstFile (path, &entp);
		1762	free (path);
		1763
		1764	if (dirp == INVALID_HANDLE_VALUE)
		1765	{
		1766	/* should steal _dosmaperr */
		1767	switch (GetLastError ())
		1768	{
		1769	case ERROR_FILE_NOT_FOUND:
		1770	req->result = 0;
		1771	break;
		1772
		1773	case ERROR_INVALID_NAME:
		1774	case ERROR_PATH_NOT_FOUND:
		1775	case ERROR_NO_MORE_FILES:
		1776	errno = ENOENT;
		1777	break;
		1778
		1779	case ERROR_NOT_ENOUGH_MEMORY:
		1780	errno = ENOMEM;
		1781	break;
		1782
		1783	default:
		1784	errno = EINVAL;
		1785	break;
		1786	}
		1787
		1788	return;
		1789	}
		1790	}
		1791	#else
		1792	#if HAVE_AT
		1793	if (req->wd)
		1794	{
		1795	int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1796
		1797	if (fd < 0)
		1798	return;
		1799
		1800	dirp = fdopendir (fd);
		1801
		1802	if (!dirp)
		1803	close (fd);
		1804	}
		1805	else
1309	self->dirp = dirp = opendir (req->ptr1);	1806	dirp = opendir (req->ptr1);
		1807	#else
		1808	dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
		1809	#endif
		1810
		1811	if (!dirp)
		1812	return;
		1813	#endif
		1814
		1815	if (req->flags & EIO_FLAG_PTR1_FREE)
		1816	free (req->ptr1);
1310		1817
1311	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1818	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1312	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1819	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1313	req->ptr2 = names = malloc (namesalloc);	1820	req->ptr2 = names = malloc (namesalloc);
1314	X_UNLOCK (wrklock);
1315		1821
1316	if (dirp && names && (!flags || dents))	1822	if (!names || (flags && !dents))
		1823	return;
		1824
1317	for (;;)	1825	for (;;)
1318	{	1826	{
		1827	int done;
		1828
		1829	#ifdef _WIN32
		1830	done = !dirp;
		1831	#else
1319	errno = 0;	1832	errno = 0;
1320	entp = readdir (dirp);	1833	entp = readdir (dirp);
		1834	done = !entp;
		1835	#endif
1321		1836
1322	if (!entp)	1837	if (done)
1323	{	1838	{
		1839	#ifndef _WIN32
		1840	int old_errno = errno;
		1841	closedir (dirp);
		1842	errno = old_errno;
		1843
1324	if (errno)	1844	if (errno)
1325	break;	1845	break;
		1846	#endif
1326		1847
1327	/* sort etc. */	1848	/* sort etc. */
1328	req->int1 = flags;	1849	req->int1 = flags;
1329	req->result = dentoffs;	1850	req->result = dentoffs;
1330		1851
1331	if (flags & EIO_READDIR_STAT_ORDER)	1852	if (flags & EIO_READDIR_STAT_ORDER)
1332	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);	1853	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1333	else if (flags & EIO_READDIR_DIRS_FIRST)	1854	else if (flags & EIO_READDIR_DIRS_FIRST)
1334	if (flags & EIO_READDIR_FOUND_UNKNOWN)	1855	if (flags & EIO_READDIR_FOUND_UNKNOWN)
1335	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */	1856	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1336	else	1857	else
		1858	{
		1859	/* in this case, all is known, and we just put dirs first and sort them */
		1860	eio_dirent *oth = dents + dentoffs;
		1861	eio_dirent *dir = dents;
		1862
		1863	/* now partition dirs to the front, and non-dirs to the back */
		1864	/* by walking from both sides and swapping if necessary */
		1865	while (oth > dir)
		1866	{
		1867	if (dir->type == EIO_DT_DIR)
		1868	++dir;
		1869	else if ((--oth)->type == EIO_DT_DIR)
		1870	{
		1871	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1872
		1873	++dir;
		1874	}
		1875	}
		1876
		1877	/* now sort the dirs only (dirs all have the same score) */
		1878	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1879	}
		1880
		1881	break;
		1882	}
		1883
		1884	/* now add the entry to our list(s) */
		1885	name = D_NAME (entp);
		1886
		1887	/* skip . and .. entries */
		1888	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
		1889	{
		1890	int len = D_NAMLEN (entp) + 1;
		1891
		1892	while (ecb_expect_false (namesoffs + len > namesalloc))
		1893	{
		1894	namesalloc *= 2;
		1895	req->ptr2 = names = realloc (names, namesalloc);
		1896
		1897	if (!names)
		1898	break;
		1899	}
		1900
		1901	memcpy (names + namesoffs, name, len);
		1902
		1903	if (dents)
		1904	{
		1905	struct eio_dirent *ent;
		1906
		1907	if (ecb_expect_false (dentoffs == dentalloc))
1337	{	1908	{
1338	/* in this case, all is known, and we just put dirs first and sort them */	1909	dentalloc *= 2;
		1910	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1911
		1912	if (!dents)
		1913	break;
		1914	}
		1915
1339	eio_dirent *oth = dents + dentoffs;	1916	ent = dents + dentoffs;
1340	eio_dirent *dir = dents;
1341		1917
1342	/* now partition dirs to the front, and non-dirs to the back */	1918	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1343	/* by walking from both sides and swapping if necessary */	1919	ent->namelen = len - 1;
1344	while (oth > dir)	1920	ent->inode = D_INO (entp);
		1921
		1922	inode_bits |= ent->inode;
		1923
		1924	switch (D_TYPE (entp))
		1925	{
		1926	default:
		1927	ent->type = EIO_DT_UNKNOWN;
		1928	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1929	break;
		1930
		1931	#ifdef DT_FIFO
		1932	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1933	#endif
		1934	#ifdef DT_CHR
		1935	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1936	#endif
		1937	#ifdef DT_MPC
		1938	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1939	#endif
		1940	#ifdef DT_DIR
		1941	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1942	#endif
		1943	#ifdef DT_NAM
		1944	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1945	#endif
		1946	#ifdef DT_BLK
		1947	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1948	#endif
		1949	#ifdef DT_MPB
		1950	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1951	#endif
		1952	#ifdef DT_REG
		1953	case DT_REG: ent->type = EIO_DT_REG; break;
		1954	#endif
		1955	#ifdef DT_NWK
		1956	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1957	#endif
		1958	#ifdef DT_CMP
		1959	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1960	#endif
		1961	#ifdef DT_LNK
		1962	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1963	#endif
		1964	#ifdef DT_SOCK
		1965	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1966	#endif
		1967	#ifdef DT_DOOR
		1968	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1969	#endif
		1970	#ifdef DT_WHT
		1971	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1972	#endif
		1973	}
		1974
		1975	ent->score = 7;
		1976
		1977	if (flags & EIO_READDIR_DIRS_FIRST)
		1978	{
		1979	if (ent->type == EIO_DT_UNKNOWN)
1345	{	1980	{
1346	if (dir->type == EIO_DT_DIR)	1981	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1347	++dir;	1982	ent->score = 1;
1348	else if ((--oth)->type == EIO_DT_DIR)	1983	else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1349	{	1984	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1350	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1351
1352	++dir;
1353	}
1354	}	1985	}
1355		1986	else if (ent->type == EIO_DT_DIR)
1356	/* now sort the dirs only (dirs all have the same score) */	1987	ent->score = 0;
1357	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1358	}	1988	}
1359
1360	break;
1361	}
1362
1363	/* now add the entry to our list(s) */
1364	name = entp->d_name;
1365
1366	/* skip . and .. entries */
1367	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1368	{
1369	int len = D_NAMLEN (entp) + 1;
1370
1371	while (ecb_expect_false (namesoffs + len > namesalloc))
1372	{
1373	namesalloc *= 2;
1374	X_LOCK (wrklock);
1375	req->ptr2 = names = realloc (names, namesalloc);
1376	X_UNLOCK (wrklock);
1377
1378	if (!names)
1379	break;
1380	}	1989	}
1381		1990
1382	memcpy (names + namesoffs, name, len);
1383
1384	if (dents)
1385	{
1386	struct eio_dirent *ent;
1387
1388	if (ecb_expect_false (dentoffs == dentalloc))
1389	{
1390	dentalloc *= 2;
1391	X_LOCK (wrklock);
1392	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1393	X_UNLOCK (wrklock);
1394
1395	if (!dents)
1396	break;
1397	}
1398
1399	ent = dents + dentoffs;
1400
1401	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1402	ent->namelen = len - 1;
1403	ent->inode = D_INO (entp);
1404
1405	inode_bits |= ent->inode;
1406
1407	switch (D_TYPE (entp))
1408	{
1409	default:
1410	ent->type = EIO_DT_UNKNOWN;
1411	flags |= EIO_READDIR_FOUND_UNKNOWN;
1412	break;
1413
1414	#ifdef DT_FIFO
1415	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1416	#endif
1417	#ifdef DT_CHR
1418	case DT_CHR: ent->type = EIO_DT_CHR; break;
1419	#endif
1420	#ifdef DT_MPC
1421	case DT_MPC: ent->type = EIO_DT_MPC; break;
1422	#endif
1423	#ifdef DT_DIR
1424	case DT_DIR: ent->type = EIO_DT_DIR; break;
1425	#endif
1426	#ifdef DT_NAM
1427	case DT_NAM: ent->type = EIO_DT_NAM; break;
1428	#endif
1429	#ifdef DT_BLK
1430	case DT_BLK: ent->type = EIO_DT_BLK; break;
1431	#endif
1432	#ifdef DT_MPB
1433	case DT_MPB: ent->type = EIO_DT_MPB; break;
1434	#endif
1435	#ifdef DT_REG
1436	case DT_REG: ent->type = EIO_DT_REG; break;
1437	#endif
1438	#ifdef DT_NWK
1439	case DT_NWK: ent->type = EIO_DT_NWK; break;
1440	#endif
1441	#ifdef DT_CMP
1442	case DT_CMP: ent->type = EIO_DT_CMP; break;
1443	#endif
1444	#ifdef DT_LNK
1445	case DT_LNK: ent->type = EIO_DT_LNK; break;
1446	#endif
1447	#ifdef DT_SOCK
1448	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1449	#endif
1450	#ifdef DT_DOOR
1451	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1452	#endif
1453	#ifdef DT_WHT
1454	case DT_WHT: ent->type = EIO_DT_WHT; break;
1455	#endif
1456	}
1457
1458	ent->score = 7;
1459
1460	if (flags & EIO_READDIR_DIRS_FIRST)
1461	{
1462	if (ent->type == EIO_DT_UNKNOWN)
1463	{
1464	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1465	ent->score = 1;
1466	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1467	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1468	}
1469	else if (ent->type == EIO_DT_DIR)
1470	ent->score = 0;
1471	}
1472	}
1473
1474	namesoffs += len;	1991	namesoffs += len;
1475	++dentoffs;	1992	++dentoffs;
1476	}	1993	}
1477		1994
1478	if (EIO_CANCELLED (req))	1995	if (EIO_CANCELLED (req))
1479	{	1996	{
1480	errno = ECANCELED;	1997	errno = ECANCELED;
1481	break;	1998	break;
1482	}	1999	}
		2000
		2001	#ifdef _WIN32
		2002	if (!FindNextFile (dirp, &entp))
		2003	{
		2004	FindClose (dirp);
		2005	dirp = 0;
1483	}	2006	}
		2007	#endif
		2008	}
1484	}	2009	}
1485		2010
1486	#ifdef PAGESIZE	2011	/*****************************************************************************/
1487	# define eio_pagesize() PAGESIZE	2012	/* working directory stuff */
1488	#else	2013	/* various deficiencies in the posix 2008 api force us to */
1489	static intptr_t	2014	/* keep the absolute path in string form at all times */
1490	eio_pagesize (void)	2015	/* fuck yeah. */
1491	{
1492	static intptr_t page;
1493		2016
1494	if (!page)	2017	#if !HAVE_AT
1495	page = sysconf (_SC_PAGESIZE);
1496		2018
		2019	/* a bit like realpath, but usually faster because it doesn'T have to return */
		2020	/* an absolute or canonical path */
		2021	static const char *
		2022	wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2023	{
		2024	if (!wd || *path == '/')
		2025	return path;
		2026
		2027	if (path [0] == '.' && !path [1])
		2028	return wd->str;
		2029
		2030	{
		2031	int l1 = wd->len;
		2032	int l2 = strlen (path);
		2033
		2034	char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
		2035
		2036	memcpy (res, wd->str, l1);
		2037	res [l1] = '/';
		2038	memcpy (res + l1 + 1, path, l2 + 1);
		2039
		2040	return res;
		2041	}
		2042	}
		2043
		2044	#endif
		2045
		2046	static eio_wd
		2047	eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2048	{
		2049	int fd;
		2050	eio_wd res;
		2051	int len = eio__realpath (tmpbuf, wd, path);
		2052
		2053	if (len < 0)
		2054	return EIO_INVALID_WD;
		2055
		2056	#if HAVE_AT
		2057	fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		2058
		2059	if (fd < 0)
		2060	return EIO_INVALID_WD;
		2061	#endif
		2062
		2063	res = malloc (sizeof (*res) + len); /* one extra 0-byte */
		2064
		2065	#if HAVE_AT
		2066	res->fd = fd;
		2067	#endif
		2068
		2069	res->len = len;
		2070	memcpy (res->str, tmpbuf->ptr, len);
		2071	res->str [len] = 0;
		2072
1497	return page;	2073	return res;
1498	}	2074	}
1499	#endif
1500		2075
1501	static void	2076	eio_wd
1502	eio_page_align (void **addr, size_t *length)	2077	eio_wd_open_sync (eio_wd wd, const char *path)
1503	{	2078	{
1504	intptr_t mask = eio_pagesize () - 1;	2079	struct tmpbuf tmpbuf = { 0 };
		2080	wd = eio__wd_open_sync (&tmpbuf, wd, path);
		2081	free (tmpbuf.ptr);
1505		2082
1506	/* round down addr */	2083	return wd;
1507	intptr_t adj = mask & (intptr_t)*addr;
1508
1509	*addr = (void *)((intptr_t)*addr - adj);
1510	*length += adj;
1511
1512	/* round up length */
1513	*length = (*length + mask) & ~mask;
1514	}	2084	}
1515		2085
1516	#if !_POSIX_MEMLOCK	2086	void
1517	# define eio__mlockall(a) ((errno = ENOSYS), -1)	2087	eio_wd_close_sync (eio_wd wd)
1518	#else	2088	{
		2089	if (wd != EIO_INVALID_WD && wd != EIO_CWD)
		2090	{
		2091	#if HAVE_AT
		2092	close (wd->fd);
		2093	#endif
		2094	free (wd);
		2095	}
		2096	}
		2097
		2098	#if HAVE_AT
		2099
		2100	/* they forgot these */
1519		2101
1520	static int	2102	static int
1521	eio__mlockall (int flags)	2103	eio__truncateat (int dirfd, const char *path, off_t length)
1522	{	2104	{
1523	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7	2105	int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1524	extern int mallopt (int, int);	2106	int res;
1525	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1526	#endif
1527		2107
1528	if (EIO_MCL_CURRENT != MCL_CURRENT	2108	if (fd < 0)
1529	|| EIO_MCL_FUTURE != MCL_FUTURE)	2109	return fd;
1530	{
1531	flags = 0
1532	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1533	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1534	}
1535		2110
1536	return mlockall (flags);	2111	res = ftruncate (fd, length);
		2112	close (fd);
		2113	return res;
1537	}	2114	}
1538	#endif
1539
1540	#if !_POSIX_MEMLOCK_RANGE
1541	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1542	#else
1543		2115
1544	static int	2116	static int
1545	eio__mlock (void *addr, size_t length)	2117	eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1546	{	2118	{
1547	eio_page_align (&addr, &length);	2119	int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
		2120	int res;
1548		2121
1549	return mlock (addr, length);	2122	if (fd < 0)
1550	}	2123	return fd;
1551		2124
1552	#endif	2125	res = fstatvfs (fd, buf);
1553		2126	close (fd);
1554	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1555	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1556	#else
1557
1558	static int
1559	eio__msync (void *mem, size_t len, int flags)
1560	{
1561	eio_page_align (&mem, &len);
1562
1563	if (EIO_MS_ASYNC != MS_SYNC
1564	|| EIO_MS_INVALIDATE != MS_INVALIDATE
1565	|| EIO_MS_SYNC != MS_SYNC)
1566	{
1567	flags = 0
1568	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1569	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1570	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1571	}
1572
1573	return msync (mem, len, flags);
1574	}
1575
1576	#endif
1577
1578	static int
1579	eio__mtouch (eio_req *req)
1580	{
1581	void *mem = req->ptr2;
1582	size_t len = req->size;
1583	int flags = req->int1;
1584
1585	eio_page_align (&mem, &len);
1586
1587	{
1588	intptr_t addr = (intptr_t)mem;
1589	intptr_t end = addr + len;
1590	intptr_t page = eio_pagesize ();
1591
1592	if (addr < end)
1593	if (flags & EIO_MT_MODIFY) /* modify */
1594	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1595	else
1596	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1597	}
1598
1599	return 0;	2127	return res;
		2128
1600	}	2129	}
		2130
		2131	#endif
1601		2132
1602	/*****************************************************************************/	2133	/*****************************************************************************/
1603		2134
1604	#define ALLOC(len) \	2135	#define ALLOC(len) \
1605	if (!req->ptr2) \	2136	if (!req->ptr2) \
…		…
1614	req->result = -1; \	2145	req->result = -1; \
1615	break; \	2146	break; \
1616	} \	2147	} \
1617	}	2148	}
1618		2149
		2150	static void ecb_noinline ecb_cold
		2151	etp_proc_init (void)
		2152	{
		2153	#if HAVE_PRCTL_SET_NAME
		2154	/* provide a more sensible "thread name" */
		2155	char name[16 + 1];
		2156	const int namelen = sizeof (name) - 1;
		2157	int len;
		2158
		2159	prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
		2160	name [namelen] = 0;
		2161	len = strlen (name);
		2162	strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
		2163	prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
		2164	#endif
		2165	}
		2166
1619	X_THREAD_PROC (etp_proc)	2167	X_THREAD_PROC (etp_proc)
1620	{	2168	{
1621	ETP_REQ *req;	2169	ETP_REQ *req;
1622	struct timespec ts;	2170	struct timespec ts;
1623	etp_worker *self = (etp_worker *)thr_arg;	2171	etp_worker *self = (etp_worker *)thr_arg;
1624		2172
		2173	etp_proc_init ();
		2174
1625	/* try to distribute timeouts somewhat randomly */	2175	/* try to distribute timeouts somewhat evenly */
1626	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	2176	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1627		2177
1628	for (;;)	2178	for (;;)
1629	{	2179	{
		2180	ts.tv_sec = 0;
		2181
1630	X_LOCK (reqlock);	2182	X_LOCK (reqlock);
1631		2183
1632	for (;;)	2184	for (;;)
1633	{	2185	{
1634	self->req = req = reqq_shift (&req_queue);	2186	req = reqq_shift (&req_queue);
1635		2187
1636	if (req)	2188	if (req)
1637	break;	2189	break;
1638		2190
		2191	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		2192	{
		2193	X_UNLOCK (reqlock);
		2194	X_LOCK (wrklock);
		2195	--started;
		2196	X_UNLOCK (wrklock);
		2197	goto quit;
		2198	}
		2199
1639	++idle;	2200	++idle;
1640		2201
1641	ts.tv_sec = time (0) + idle_timeout;	2202	if (idle <= max_idle)
1642	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	2203	/* we are allowed to idle, so do so without any timeout */
		2204	X_COND_WAIT (reqwait, reqlock);
		2205	else
1643	{	2206	{
1644	if (idle > max_idle)	2207	/* initialise timeout once */
1645	{	2208	if (!ts.tv_sec)
1646	--idle;	2209	ts.tv_sec = time (0) + idle_timeout;
1647	X_UNLOCK (reqlock);
1648	X_LOCK (wrklock);
1649	--started;
1650	X_UNLOCK (wrklock);
1651	goto quit;
1652	}
1653		2210
1654	/* we are allowed to idle, so do so without any timeout */
1655	X_COND_WAIT (reqwait, reqlock);	2211	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		2212	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1656	}	2213	}
1657		2214
1658	--idle;	2215	--idle;
1659	}	2216	}
1660		2217
…		…
1663	X_UNLOCK (reqlock);	2220	X_UNLOCK (reqlock);
1664		2221
1665	if (req->type < 0)	2222	if (req->type < 0)
1666	goto quit;	2223	goto quit;
1667		2224
1668	if (!EIO_CANCELLED (req))
1669	ETP_EXECUTE (self, req);	2225	ETP_EXECUTE (self, req);
1670		2226
1671	X_LOCK (reslock);	2227	X_LOCK (reslock);
1672		2228
1673	++npending;	2229	++npending;
1674		2230
1675	if (!reqq_push (&res_queue, req) && want_poll_cb)	2231	if (!reqq_push (&res_queue, req) && want_poll_cb)
1676	want_poll_cb ();	2232	want_poll_cb ();
1677		2233
1678	self->req = 0;
1679	etp_worker_clear (self);	2234	etp_worker_clear (self);
1680		2235
1681	X_UNLOCK (reslock);	2236	X_UNLOCK (reslock);
1682	}	2237	}
1683		2238
1684	quit:	2239	quit:
		2240	free (req);
		2241
1685	X_LOCK (wrklock);	2242	X_LOCK (wrklock);
1686	etp_worker_free (self);	2243	etp_worker_free (self);
1687	X_UNLOCK (wrklock);	2244	X_UNLOCK (wrklock);
1688
1689	return 0;
1690	}	2245	}
1691		2246
1692	/*****************************************************************************/	2247	/*****************************************************************************/
1693		2248
1694	int ecb_cold	2249	int ecb_cold
1695	eio_init (void (*want_poll)(void), void (*done_poll)(void))	2250	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1696	{	2251	{
		2252	#if !HAVE_PREADWRITE
		2253	X_MUTEX_CREATE (preadwritelock);
		2254	#endif
		2255
1697	return etp_init (want_poll, done_poll);	2256	return etp_init (want_poll, done_poll);
1698	}	2257	}
1699		2258
1700	ecb_inline void	2259	ecb_inline void
1701	eio_api_destroy (eio_req *req)	2260	eio_api_destroy (eio_req *req)
…		…
1728	}	2287	}
1729		2288
1730	static void	2289	static void
1731	eio_execute (etp_worker *self, eio_req *req)	2290	eio_execute (etp_worker *self, eio_req *req)
1732	{	2291	{
		2292	#if HAVE_AT
		2293	int dirfd;
		2294	#else
		2295	const char *path;
		2296	#endif
		2297
		2298	if (ecb_expect_false (EIO_CANCELLED (req)))
		2299	{
		2300	req->result = -1;
		2301	req->errorno = ECANCELED;
		2302	return;
		2303	}
		2304
		2305	if (ecb_expect_false (req->wd == EIO_INVALID_WD))
		2306	{
		2307	req->result = -1;
		2308	req->errorno = ENOENT;
		2309	return;
		2310	}
		2311
		2312	if (req->type >= EIO_OPEN)
		2313	{
		2314	#if HAVE_AT
		2315	dirfd = WD2FD (req->wd);
		2316	#else
		2317	path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		2318	#endif
		2319	}
		2320
1733	switch (req->type)	2321	switch (req->type)
1734	{	2322	{
		2323	case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
		2324	req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
		2325	break;
		2326	case EIO_WD_CLOSE: req->result = 0;
		2327	eio_wd_close_sync (req->wd); break;
		2328
		2329	case EIO_SEEK: eio__lseek (req); break;
1735	case EIO_READ: ALLOC (req->size);	2330	case EIO_READ: ALLOC (req->size);
1736	req->result = req->offs >= 0	2331	req->result = req->offs >= 0
1737	? pread (req->int1, req->ptr2, req->size, req->offs)	2332	? pread (req->int1, req->ptr2, req->size, req->offs)
1738	: read (req->int1, req->ptr2, req->size); break;	2333	: read (req->int1, req->ptr2, req->size); break;
1739	case EIO_WRITE: req->result = req->offs >= 0	2334	case EIO_WRITE: req->result = req->offs >= 0
1740	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2335	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1741	: write (req->int1, req->ptr2, req->size); break;	2336	: write (req->int1, req->ptr2, req->size); break;
1742		2337
1743	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2338	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1744	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2339	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
		2340
		2341	#if HAVE_AT
1745		2342
1746	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2343	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1747	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2344	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1748	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2345	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2346	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
		2347	case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
		2348	case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
		2349	case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
		2350	case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
		2351
		2352	case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
		2353	case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
		2354	case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
		2355	case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
		2356	case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
		2357	case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
		2358	case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2359	case EIO_READLINK: ALLOC (PATH_MAX);
		2360	req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
		2361	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2362	req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2363	case EIO_UTIME:
		2364	case EIO_FUTIME:
		2365	{
		2366	struct timespec ts[2];
		2367	struct timespec *times;
		2368
		2369	if (req->nv1 != -1. || req->nv2 != -1.)
		2370	{
		2371	ts[0].tv_sec = req->nv1;
		2372	ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
		2373	ts[1].tv_sec = req->nv2;
		2374	ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
		2375
		2376	times = ts;
		2377	}
		2378	else
		2379	times = 0;
		2380
		2381	req->result = req->type == EIO_FUTIME
		2382	? futimens (req->int1, times)
		2383	: utimensat (dirfd, req->ptr1, times, 0);
		2384	}
		2385	break;
		2386
		2387	#else
		2388
		2389	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2390	req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2391	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1749	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2392	req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2393	case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
		2394	case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
		2395	case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
		2396	case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
		2397
		2398	case EIO_UNLINK: req->result = unlink (path ); break;
		2399	case EIO_RMDIR: req->result = rmdir (path ); break;
		2400	case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
		2401	case EIO_RENAME: req->result = rename (path , req->ptr2); break;
		2402	case EIO_LINK: req->result = link (path , req->ptr2); break;
		2403	case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
		2404	case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
		2405	case EIO_READLINK: ALLOC (PATH_MAX);
		2406	req->result = readlink (path, req->ptr2, PATH_MAX); break;
		2407	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2408	req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2409
		2410	case EIO_UTIME:
		2411	case EIO_FUTIME:
		2412	{
		2413	struct timeval tv[2];
		2414	struct timeval *times;
		2415
		2416	if (req->nv1 != -1. || req->nv2 != -1.)
		2417	{
		2418	tv[0].tv_sec = req->nv1;
		2419	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
		2420	tv[1].tv_sec = req->nv2;
		2421	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
		2422
		2423	times = tv;
		2424	}
		2425	else
		2426	times = 0;
		2427
		2428	req->result = req->type == EIO_FUTIME
		2429	? futimes (req->int1, times)
		2430	: utimes (req->ptr1, times);
		2431	}
		2432	break;
		2433
		2434	#endif
		2435
		2436	case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
		2437	{
		2438	ALLOC (req->result);
		2439	memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
		2440	}
		2441	break;
		2442
1750	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2443	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1751	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2444	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1752		2445
1753	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1754	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1755	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));	2446	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1756	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;	2447	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1757		2448
1758	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1759	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	2449	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1760	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1761	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	2450	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1762	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1763	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;	2451	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1764		2452
1765	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1766	case EIO_CLOSE: req->result = close (req->int1); break;	2453	case EIO_CLOSE: req->result = close (req->int1); break;
1767	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;	2454	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1768	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1769	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1770	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1771	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1772	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1773	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1774	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1775
1776	case EIO_READLINK: ALLOC (PATH_MAX);
1777	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1778
1779	case EIO_SYNC: req->result = 0; sync (); break;	2455	case EIO_SYNC: req->result = 0; sync (); break;
1780	case EIO_FSYNC: req->result = fsync (req->int1); break;	2456	case EIO_FSYNC: req->result = fsync (req->int1); break;
1781	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2457	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2458	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2459	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1782	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2460	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1783	case EIO_MTOUCH: req->result = eio__mtouch (req); break;	2461	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1784	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2462	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1785	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2463	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1786	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	2464	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1787		2465
1788	case EIO_READDIR: eio__scandir (req, self); break;	2466	case EIO_READDIR: eio__scandir (req, self); break;
1789		2467
1790	case EIO_BUSY:	2468	case EIO_BUSY:
1791	#ifdef _WIN32	2469	#ifdef _WIN32
…		…
1800	req->result = select (0, 0, 0, 0, &tv);	2478	req->result = select (0, 0, 0, 0, &tv);
1801	}	2479	}
1802	#endif	2480	#endif
1803	break;	2481	break;
1804		2482
1805	case EIO_UTIME:
1806	case EIO_FUTIME:
1807	{
1808	struct timeval tv[2];
1809	struct timeval *times;
1810
1811	if (req->nv1 != -1. || req->nv2 != -1.)
1812	{
1813	tv[0].tv_sec = req->nv1;
1814	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1815	tv[1].tv_sec = req->nv2;
1816	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1817
1818	times = tv;
1819	}
1820	else
1821	times = 0;
1822
1823	req->result = req->type == EIO_FUTIME
1824	? futimes (req->int1, times)
1825	: utimes (req->ptr1, times);
1826	}
1827	break;
1828
1829	case EIO_GROUP:	2483	case EIO_GROUP:
1830	abort (); /* handled in eio_request */	2484	abort (); /* handled in eio_request */
1831		2485
1832	case EIO_NOP:	2486	case EIO_NOP:
1833	req->result = 0;	2487	req->result = 0;
…		…
1836	case EIO_CUSTOM:	2490	case EIO_CUSTOM:
1837	req->feed (req);	2491	req->feed (req);
1838	break;	2492	break;
1839		2493
1840	default:	2494	default:
1841	errno = ENOSYS;
1842	req->result = -1;	2495	req->result = EIO_ENOSYS ();
1843	break;	2496	break;
1844	}	2497	}
1845		2498
1846	req->errorno = errno;	2499	req->errorno = errno;
1847	}	2500	}
1848		2501
1849	#ifndef EIO_NO_WRAPPERS	2502	#ifndef EIO_NO_WRAPPERS
1850		2503
		2504	eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
		2505	{
		2506	REQ (EIO_WD_OPEN); PATH; SEND;
		2507	}
		2508
		2509	eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
		2510	{
		2511	REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
		2512	}
		2513
1851	eio_req *eio_nop (int pri, eio_cb cb, void *data)	2514	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1852	{	2515	{
1853	REQ (EIO_NOP); SEND;	2516	REQ (EIO_NOP); SEND;
1854	}	2517	}
1855		2518
…		…
1871	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2534	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1872	{	2535	{
1873	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2536	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1874	}	2537	}
1875		2538
		2539	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		2540	{
		2541	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2542	}
		2543
		2544	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2545	{
		2546	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2547	}
		2548
		2549	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2550	{
		2551	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2552	}
		2553
1876	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2554	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1877	{	2555	{
1878	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2556	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1879	}	2557	}
1880		2558
…		…
1886	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)	2564	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1887	{	2565	{
1888	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;	2566	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1889	}	2567	}
1890		2568
1891	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2569	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1892	{	2570	{
1893	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2571	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1894	}
1895
1896	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1897	{
1898	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1899	}	2572	}
1900		2573
1901	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2574	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1902	{	2575	{
1903	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2576	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
1906	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)	2579	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1907	{	2580	{
1908	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;	2581	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1909	}	2582	}
1910		2583
		2584	eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
		2585	{
		2586	REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
		2587	}
		2588
1911	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)	2589	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1912	{	2590	{
1913	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;	2591	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1914	}	2592	}
1915		2593
…		…
1941	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2619	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1942	{	2620	{
1943	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2621	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1944	}	2622	}
1945		2623
1946	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2624	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1947	{	2625	{
1948	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2626	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1949	}	2627	}
1950		2628
1951	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2629	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1971	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2649	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1972	{	2650	{
1973	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2651	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1974	}	2652	}
1975		2653
1976	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2654	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1977	{	2655	{
1978	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2656	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1979	}	2657	}
1980		2658
1981	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2659	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1995	}	2673	}
1996		2674
1997	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2675	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1998	{	2676	{
1999	return eio__1path (EIO_READLINK, path, pri, cb, data);	2677	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2678	}
		2679
		2680	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2681	{
		2682	return eio__1path (EIO_REALPATH, path, pri, cb, data);
2000	}	2683	}
2001		2684
2002	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2685	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2003	{	2686	{
2004	return eio__1path (EIO_STAT, path, pri, cb, data);	2687	return eio__1path (EIO_STAT, path, pri, cb, data);
…		…
2123	}	2806	}
2124		2807
2125	/*****************************************************************************/	2808	/*****************************************************************************/
2126	/* misc garbage */	2809	/* misc garbage */
2127		2810
2128	ssize_t	2811	eio_ssize_t
2129	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2812	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2130	{	2813	{
2131	etp_worker wrk;
2132	ssize_t ret;
2133
2134	wrk.dbuf = 0;
2135
2136	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2814	return eio__sendfile (ofd, ifd, offset, count);
2137
2138	if (wrk.dbuf)
2139	free (wrk.dbuf);
2140
2141	return ret;
2142	}	2815	}
2143		2816

Diff Legend

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