ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.86 by root, Thu Jul 14 18:30:10 2011 UTC vs.
Revision 1.118 by root, Mon Apr 2 17:53:27 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>
…		…
58	#include <sys/stat.h>	58	#include <sys/stat.h>
59	#include <limits.h>	59	#include <limits.h>
60	#include <fcntl.h>	60	#include <fcntl.h>
61	#include <assert.h>	61	#include <assert.h>
62		62
63	#include <sys/statvfs.h>
64	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */	63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65	/* intptr_t only comes from stdint.h, says idiot openbsd coder */	64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66	#if HAVE_STDINT_H	65	#if HAVE_STDINT_H
67	# include <stdint.h>	66	# include <stdint.h>
68	#endif	67	#endif
69		68
70	#ifndef ECANCELED	69	#ifndef ECANCELED
71	# define ECANCELED EDOM	70	# define ECANCELED EDOM
72	#endif	71	#endif
		72	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
		77	# define ENOTSOCK WSAENOTSOCK
		78	#endif
73		79
74	static void eio_destroy (eio_req *req);	80	static void eio_destroy (eio_req *req);
75		81
76	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
77	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	83	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
…		…
96		102
97	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)	103	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
98		104
99	#ifdef _WIN32	105	#ifdef _WIN32
100		106
		107	#undef PAGESIZE
101	#define PAGESIZE 4096 /* GetSystemInfo? */	108	#define PAGESIZE 4096 /* GetSystemInfo? */
102		109
		110	/* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
		111
		112	#ifdef EIO_STRUCT_STATI64
		113	/* look at perl's non-sloppy stat */
103	#define stat(path,buf) _stati64 (path,buf)	114	#define stat(path,buf) _stati64 (path,buf)
		115	#define fstat(fd,buf) _fstati64 (fd,buf)
		116	#endif
104	#define lstat(path,buf) stat (path,buf)	117	#define lstat(path,buf) stat (path,buf)
105	#define fstat(fd,buf) _fstati64 (path,buf)
106	#define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))	118	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
107	#define mkdir(path,mode) _mkdir (path)	119	#define mkdir(path,mode) _mkdir (path)
108	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))	120	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
109		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 ()
110	#define chown(path,uid,gid) EIO_ENOSYS ()	127	#define chown(path,uid,gid) EIO_ENOSYS ()
111	#define fchown(fd,uid,gid) EIO_ENOSYS ()	128	#define fchown(fd,uid,gid) EIO_ENOSYS ()
112	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */	129	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
113	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */	130	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
114	#define mknod(path,mode,dev) EIO_ENOSYS ()	131	#define mknod(path,mode,dev) EIO_ENOSYS ()
115	#define sync() EIO_ENOSYS ()	132	#define sync() EIO_ENOSYS ()
		133	#define readlink(path,buf,s) EIO_ENOSYS ()
		134	#define statvfs(path,buf) EIO_ENOSYS ()
		135	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		136
		137	/* rename() uses MoveFile, which fails to overwrite */
		138	#define rename(old,neu) eio__rename (old, neu)
		139
		140	static int
		141	eio__rename (const char *old, const char *neu)
		142	{
		143	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		144	return 0;
		145
		146	/* should steal _dosmaperr */
		147	switch (GetLastError ())
		148	{
		149	case ERROR_FILE_NOT_FOUND:
		150	case ERROR_PATH_NOT_FOUND:
		151	case ERROR_INVALID_DRIVE:
		152	case ERROR_NO_MORE_FILES:
		153	case ERROR_BAD_NETPATH:
		154	case ERROR_BAD_NET_NAME:
		155	case ERROR_BAD_PATHNAME:
		156	case ERROR_FILENAME_EXCED_RANGE:
		157	errno = ENOENT;
		158	break;
		159
		160	default:
		161	errno = EACCES;
		162	break;
		163	}
		164
		165	return -1;
		166	}
116		167
117	/* we could even stat and see if it exists */	168	/* we could even stat and see if it exists */
118	static int	169	static int
119	symlink (const char *old, const char *neu)	170	symlink (const char *old, const char *neu)
120	{	171	{
		172	#if WINVER >= 0x0600
121	if (CreateSymbolicLink (neu, old, 1))	173	if (CreateSymbolicLink (neu, old, 1))
122	return 0;	174	return 0;
123		175
124	if (CreateSymbolicLink (neu, old, 0))	176	if (CreateSymbolicLink (neu, old, 0))
125	return 0;	177	return 0;
		178	#endif
126		179
127	return EIO_ERRNO (ENOENT, -1);	180	return EIO_ERRNO (ENOENT, -1);
128	}	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)
129		196
130	#else	197	#else
131		198
132	#include <sys/time.h>	199	#include <sys/time.h>
133	#include <sys/select.h>	200	#include <sys/select.h>
134	#include <sys/statvfs.h>	201	#include <sys/statvfs.h>
135	#include <unistd.h>	202	#include <unistd.h>
136	#include <utime.h>
137	#include <signal.h>	203	#include <signal.h>
138	#include <dirent.h>	204	#include <dirent.h>
139		205
140	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES	206	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
141	#include <sys/mman.h>	207	#include <sys/mman.h>
142	#endif	208	#endif
143		209
		210	#define D_NAME(entp) entp->d_name
		211
144	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */	212	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
145	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__	213	#if __FreeBSD__ || __NetBSD__ || __OpenBSD__
146	#define _DIRENT_HAVE_D_TYPE /* sigh */	214	#define _DIRENT_HAVE_D_TYPE /* sigh */
147	#define D_INO(de) (de)->d_fileno	215	#define D_INO(de) (de)->d_fileno
148	#define D_NAMLEN(de) (de)->d_namlen	216	#define D_NAMLEN(de) (de)->d_namlen
149	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600	217	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
150	#define D_INO(de) (de)->d_ino	218	#define D_INO(de) (de)->d_ino
…		…
161		229
162	#ifndef EIO_STRUCT_DIRENT	230	#ifndef EIO_STRUCT_DIRENT
163	#define EIO_STRUCT_DIRENT struct dirent	231	#define EIO_STRUCT_DIRENT struct dirent
164	#endif	232	#endif
165		233
		234	#endif
		235
		236	#if HAVE_UTIMES
		237	# include <utime.h>
		238	#endif
		239
		240	#if HAVE_SYS_SYSCALL_H
		241	# include <sys/syscall.h>
		242	#endif
		243
		244	#if HAVE_SYS_PRCTL_H
		245	# include <sys/prctl.h>
166	#endif	246	#endif
167		247
168	#if HAVE_SENDFILE	248	#if HAVE_SENDFILE
169	# if __linux	249	# if __linux
170	# include <sys/sendfile.h>	250	# include <sys/sendfile.h>
…		…
185	#endif	265	#endif
186	#ifndef D_INO	266	#ifndef D_INO
187	# define D_INO(de) 0	267	# define D_INO(de) 0
188	#endif	268	#endif
189	#ifndef D_NAMLEN	269	#ifndef D_NAMLEN
190	# define D_NAMLEN(de) strlen ((de)->d_name)	270	# define D_NAMLEN(entp) strlen (D_NAME (entp))
191	#endif	271	#endif
192		272
193	/* used for struct dirent, AIX doesn't provide it */	273	/* used for struct dirent, AIX doesn't provide it */
194	#ifndef NAME_MAX	274	#ifndef NAME_MAX
195	# define NAME_MAX 4096	275	# define NAME_MAX 4096
…		…
202		282
203	/* buffer size for various temporary buffers */	283	/* buffer size for various temporary buffers */
204	#define EIO_BUFSIZE 65536	284	#define EIO_BUFSIZE 65536
205		285
206	#define dBUF \	286	#define dBUF \
207	char *eio_buf; \
208	ETP_WORKER_LOCK (self); \
209	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	287	char *eio_buf = malloc (EIO_BUFSIZE); \
210	ETP_WORKER_UNLOCK (self); \
211	errno = ENOMEM; \	288	errno = ENOMEM; \
212	if (!eio_buf) \	289	if (!eio_buf) \
213	return -1;	290	return -1
		291
		292	#define FUBd \
		293	free (eio_buf)
214		294
215	#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	/*****************************************************************************/
216		340
217	#define ETP_PRI_MIN EIO_PRI_MIN	341	#define ETP_PRI_MIN EIO_PRI_MIN
218	#define ETP_PRI_MAX EIO_PRI_MAX	342	#define ETP_PRI_MAX EIO_PRI_MAX
219		343
220	struct etp_worker;	344	struct etp_worker;
…		…
224	static int eio_finish (eio_req *req);	348	static int eio_finish (eio_req *req);
225	#define ETP_FINISH(req) eio_finish (req)	349	#define ETP_FINISH(req) eio_finish (req)
226	static void eio_execute (struct etp_worker *self, eio_req *req);	350	static void eio_execute (struct etp_worker *self, eio_req *req);
227	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	351	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
228		352
229	#define ETP_WORKER_CLEAR(req) \
230	if (wrk->dbuf) \
231	{ \
232	free (wrk->dbuf); \
233	wrk->dbuf = 0; \
234	} \
235	\
236	if (wrk->dirp) \
237	{ \
238	closedir (wrk->dirp); \
239	wrk->dirp = 0; \
240	}
241
242	#define ETP_WORKER_COMMON \
243	void *dbuf; \
244	DIR *dirp;
245
246	/*****************************************************************************/	353	/*****************************************************************************/
247		354
248	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	355	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
249		356
250	/* calculate time difference in ~1/EIO_TICKS of a second */	357	/* calculate time difference in ~1/EIO_TICKS of a second */
…		…
261	static void (*done_poll_cb) (void);	368	static void (*done_poll_cb) (void);
262		369
263	static unsigned int max_poll_time; /* reslock */	370	static unsigned int max_poll_time; /* reslock */
264	static unsigned int max_poll_reqs; /* reslock */	371	static unsigned int max_poll_reqs; /* reslock */
265		372
266	static volatile unsigned int nreqs; /* reqlock */	373	static unsigned int nreqs; /* reqlock */
267	static volatile unsigned int nready; /* reqlock */	374	static unsigned int nready; /* reqlock */
268	static volatile unsigned int npending; /* reqlock */	375	static unsigned int npending; /* reqlock */
269	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 */
270	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 */
271		378
272	static xmutex_t wrklock;	379	static xmutex_t wrklock;
273	static xmutex_t reslock;	380	static xmutex_t reslock;
274	static xmutex_t reqlock;	381	static xmutex_t reqlock;
275	static xcond_t reqwait;	382	static xcond_t reqwait;
…		…
278	/*	385	/*
279	* make our pread/pwrite emulation safe against themselves, but not against	386	* make our pread/pwrite emulation safe against themselves, but not against
280	* 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,
281	* but that's your problem, not mine.	388	* but that's your problem, not mine.
282	*/	389	*/
283	static xmutex_t preadwritelock = X_MUTEX_INIT;	390	static xmutex_t preadwritelock;
284	#endif	391	#endif
285		392
286	typedef struct etp_worker	393	typedef struct etp_worker
287	{	394	{
		395	struct tmpbuf tmpbuf;
		396
288	/* locked by wrklock */	397	/* locked by wrklock */
289	struct etp_worker *prev, *next;	398	struct etp_worker *prev, *next;
290		399
291	xthread_t tid;	400	xthread_t tid;
292		401
293	/* locked by reslock, reqlock or wrklock */	402	#ifdef ETP_WORKER_COMMON
294	ETP_REQ *req; /* currently processed request */
295
296	ETP_WORKER_COMMON	403	ETP_WORKER_COMMON
		404	#endif
297	} etp_worker;	405	} etp_worker;
298		406
299	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	407	static etp_worker wrk_first; /* NOT etp */
300		408
301	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	409	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
302	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	410	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
303		411
304	/* worker threads management */	412	/* worker threads management */
305		413
306	static void ecb_cold	414	static void
307	etp_worker_clear (etp_worker *wrk)	415	etp_worker_clear (etp_worker *wrk)
308	{	416	{
309	ETP_WORKER_CLEAR (wrk);
310	}	417	}
311		418
312	static void ecb_cold	419	static void ecb_cold
313	etp_worker_free (etp_worker *wrk)	420	etp_worker_free (etp_worker *wrk)
314	{	421	{
		422	free (wrk->tmpbuf.ptr);
		423
315	wrk->next->prev = wrk->prev;	424	wrk->next->prev = wrk->prev;
316	wrk->prev->next = wrk->next;	425	wrk->prev->next = wrk->next;
317		426
318	free (wrk);	427	free (wrk);
319	}	428	}
…		…
375	} etp_reqq;	484	} etp_reqq;
376		485
377	static etp_reqq req_queue;	486	static etp_reqq req_queue;
378	static etp_reqq res_queue;	487	static etp_reqq res_queue;
379		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
380	static int ecb_noinline	500	static int ecb_noinline
381	reqq_push (etp_reqq *q, ETP_REQ *req)	501	reqq_push (etp_reqq *q, ETP_REQ *req)
382	{	502	{
383	int pri = req->pri;	503	int pri = req->pri;
384	req->next = 0;	504	req->next = 0;
…		…
418	}	538	}
419		539
420	abort ();	540	abort ();
421	}	541	}
422		542
423	static void ecb_cold	543	static int ecb_cold
424	etp_thread_init (void)	544	etp_init (void (*want_poll)(void), void (*done_poll)(void))
425	{	545	{
426	#if !HAVE_PREADWRITE
427	X_MUTEX_CREATE (preadwritelock);
428	#endif
429	X_MUTEX_CREATE (wrklock);	546	X_MUTEX_CREATE (wrklock);
430	X_MUTEX_CREATE (reslock);	547	X_MUTEX_CREATE (reslock);
431	X_MUTEX_CREATE (reqlock);	548	X_MUTEX_CREATE (reqlock);
432	X_COND_CREATE (reqwait);	549	X_COND_CREATE (reqwait);
433	}
434		550
435	static void ecb_cold	551	reqq_init (&req_queue);
436	etp_atfork_prepare (void)	552	reqq_init (&res_queue);
437	{
438	}
439		553
440	static void ecb_cold	554	wrk_first.next =
441	etp_atfork_parent (void)	555	wrk_first.prev = &wrk_first;
442	{
443	}
444
445	static void ecb_cold
446	etp_atfork_child (void)
447	{
448	ETP_REQ *prv;
449
450	while ((prv = reqq_shift (&req_queue)))
451	ETP_DESTROY (prv);
452
453	while ((prv = reqq_shift (&res_queue)))
454	ETP_DESTROY (prv);
455
456	while (wrk_first.next != &wrk_first)
457	{
458	etp_worker *wrk = wrk_first.next;
459
460	if (wrk->req)
461	ETP_DESTROY (wrk->req);
462
463	etp_worker_clear (wrk);
464	etp_worker_free (wrk);
465	}
466		556
467	started = 0;	557	started = 0;
468	idle = 0;	558	idle = 0;
469	nreqs = 0;	559	nreqs = 0;
470	nready = 0;	560	nready = 0;
471	npending = 0;	561	npending = 0;
472
473	etp_thread_init ();
474	}
475
476	static void ecb_cold
477	etp_once_init (void)
478	{
479	etp_thread_init ();
480	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
481	}
482
483	static int ecb_cold
484	etp_init (void (*want_poll)(void), void (*done_poll)(void))
485	{
486	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
487
488	pthread_once (&doinit, etp_once_init);
489		562
490	want_poll_cb = want_poll;	563	want_poll_cb = want_poll;
491	done_poll_cb = done_poll;	564	done_poll_cb = done_poll;
492		565
493	return 0;	566	return 0;
…		…
533	}	606	}
534		607
535	static void ecb_cold	608	static void ecb_cold
536	etp_end_thread (void)	609	etp_end_thread (void)
537	{	610	{
538	eio_req *req = calloc (1, sizeof (eio_req));	611	eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
539		612
540	req->type = -1;	613	req->type = -1;
541	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	614	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
542		615
543	X_LOCK (reqlock);	616	X_LOCK (reqlock);
…		…
876	# undef pread	949	# undef pread
877	# undef pwrite	950	# undef pwrite
878	# define pread eio__pread	951	# define pread eio__pread
879	# define pwrite eio__pwrite	952	# define pwrite eio__pwrite
880		953
881	static ssize_t	954	static eio_ssize_t
882	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)
883	{	956	{
884	ssize_t res;	957	eio_ssize_t res;
885	off_t ooffset;	958	off_t ooffset;
886		959
887	X_LOCK (preadwritelock);	960	X_LOCK (preadwritelock);
888	ooffset = lseek (fd, 0, SEEK_CUR);	961	ooffset = lseek (fd, 0, SEEK_CUR);
889	lseek (fd, offset, SEEK_SET);	962	lseek (fd, offset, SEEK_SET);
…		…
892	X_UNLOCK (preadwritelock);	965	X_UNLOCK (preadwritelock);
893		966
894	return res;	967	return res;
895	}	968	}
896		969
897	static ssize_t	970	static eio_ssize_t
898	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)
899	{	972	{
900	ssize_t res;	973	eio_ssize_t res;
901	off_t ooffset;	974	off_t ooffset;
902		975
903	X_LOCK (preadwritelock);	976	X_LOCK (preadwritelock);
904	ooffset = lseek (fd, 0, SEEK_CUR);	977	ooffset = lseek (fd, 0, SEEK_CUR);
905	lseek (fd, offset, SEEK_SET);	978	lseek (fd, offset, SEEK_SET);
…		…
950		1023
951	#if !HAVE_FDATASYNC	1024	#if !HAVE_FDATASYNC
952	# undef fdatasync	1025	# undef fdatasync
953	# define fdatasync(fd) fsync (fd)	1026	# define fdatasync(fd) fsync (fd)
954	#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 = -1;
		1038	errno = ENOSYS;
		1039	#endif
		1040
		1041	if (res < 0 && errno == ENOSYS && fd >= 0)
		1042	sync ();
		1043
		1044	return res;
		1045	}
955		1046
956	/* sync_file_range always needs emulation */	1047	/* sync_file_range always needs emulation */
957	static int	1048	static int
958	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	1049	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
959	{	1050	{
…		…
994		1085
995	#if !HAVE_READAHEAD	1086	#if !HAVE_READAHEAD
996	# undef readahead	1087	# undef readahead
997	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1088	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
998		1089
999	static ssize_t	1090	static eio_ssize_t
1000	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1091	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
1001	{	1092	{
1002	size_t todo = count;	1093	size_t todo = count;
1003	dBUF;	1094	dBUF;
1004		1095
…		…
1009	pread (fd, eio_buf, len, offset);	1100	pread (fd, eio_buf, len, offset);
1010	offset += len;	1101	offset += len;
1011	todo -= len;	1102	todo -= len;
1012	}	1103	}
1013		1104
1014	errno = 0;	1105	FUBd;
		1106
		1107	/* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
		1108	/* but not for e.g. EIO or eof, so we also never fail */
1015	return count;	1109	return 0;
1016	}	1110	}
1017		1111
1018	#endif	1112	#endif
1019		1113
1020	/* sendfile always needs emulation */	1114	/* sendfile always needs emulation */
1021	static ssize_t	1115	static eio_ssize_t
1022	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1116	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
1023	{	1117	{
1024	ssize_t written = 0;	1118	eio_ssize_t written = 0;
1025	ssize_t res;	1119	eio_ssize_t res;
1026		1120
1027	if (!count)	1121	if (!count)
1028	return 0;	1122	return 0;
1029		1123
1030	for (;;)	1124	for (;;)
…		…
1055		1149
1056	/* according to source inspection, this is correct, and useful behaviour */	1150	/* according to source inspection, this is correct, and useful behaviour */
1057	if (sbytes)	1151	if (sbytes)
1058	res = sbytes;	1152	res = sbytes;
1059		1153
1060	# elif defined (__APPLE__)	1154	# elif defined __APPLE__
1061	off_t sbytes = count;	1155	off_t sbytes = count;
1062	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	1156	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1063		1157
1064	/* according to the manpage, sbytes is always valid */	1158	/* according to the manpage, sbytes is always valid */
1065	if (sbytes)	1159	if (sbytes)
…		…
1125	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1219	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1126	/* BSDs */	1220	/* BSDs */
1127	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1221	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1128	|| errno == ENOTSUP	1222	|| errno == ENOTSUP
1129	#endif	1223	#endif
		1224	#ifdef EOPNOTSUPP /* windows */
1130	|| errno == EOPNOTSUPP /* BSDs */	1225	|| errno == EOPNOTSUPP /* BSDs */
		1226	#endif
1131	#if __solaris	1227	#if __solaris
1132	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1228	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1133	#endif	1229	#endif
1134	)	1230	)
1135	)	1231	)
…		…
1139		1235
1140	res = 0;	1236	res = 0;
1141		1237
1142	while (count)	1238	while (count)
1143	{	1239	{
1144	ssize_t cnt;	1240	eio_ssize_t cnt;
1145		1241
1146	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1242	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1147		1243
1148	if (cnt <= 0)	1244	if (cnt <= 0)
1149	{	1245	{
…		…
1161		1257
1162	offset += cnt;	1258	offset += cnt;
1163	res += cnt;	1259	res += cnt;
1164	count -= cnt;	1260	count -= cnt;
1165	}	1261	}
		1262
		1263	FUBd;
1166	}	1264	}
1167		1265
1168	return res;	1266	return res;
1169	}	1267	}
1170		1268
…		…
1197	/* round up length */	1295	/* round up length */
1198	*length = (*length + mask) & ~mask;	1296	*length = (*length + mask) & ~mask;
1199	}	1297	}
1200		1298
1201	#if !_POSIX_MEMLOCK	1299	#if !_POSIX_MEMLOCK
1202	# define eio__mlockall(a) eio_nosyscall()	1300	# define eio__mlockall(a) EIO_ENOSYS ()
1203	#else	1301	#else
1204		1302
1205	static int	1303	static int
1206	eio__mlockall (int flags)	1304	eio__mlockall (int flags)
1207	{	1305	{
…		…
1286		1384
1287	/*****************************************************************************/	1385	/*****************************************************************************/
1288	/* requests implemented outside eio_execute, because they are so large */	1386	/* requests implemented outside eio_execute, because they are so large */
1289		1387
1290	static void	1388	static void
1291	eio__realpath (eio_req *req, etp_worker *self)	1389	eio__lseek (eio_req *req)
1292	{	1390	{
1293	char *rel = req->ptr1;	1391	/* this usually gets optimised away completely, or your compiler sucks, */
		1392	/* or the whence constants really are not 0, 1, 2 */
		1393	int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
		1394	: req->int2 == EIO_SEEK_CUR ? SEEK_CUR
		1395	: req->int2 == EIO_SEEK_END ? SEEK_END
		1396	: req->int2;
		1397
		1398	req->offs = lseek (req->int1, req->offs, whence);
		1399	req->result = req->offs == (off_t)-1 ? -1 : 0;
		1400	}
		1401
		1402	/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
		1403	static int
		1404	eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1405	{
		1406	const char *rel = path;
1294	char *res;	1407	char *res;
1295	char *tmp1, *tmp2;	1408	char *tmp1, *tmp2;
1296	#if SYMLOOP_MAX > 32	1409	#if SYMLOOP_MAX > 32
1297	int symlinks = SYMLOOP_MAX;	1410	int symlinks = SYMLOOP_MAX;
1298	#else	1411	#else
1299	int symlinks = 32;	1412	int symlinks = 32;
1300	#endif	1413	#endif
1301		1414
1302	req->result = -1;
1303
1304	errno = EINVAL;	1415	errno = EINVAL;
1305	if (!rel)	1416	if (!rel)
1306	return;	1417	return -1;
1307		1418
1308	errno = ENOENT;	1419	errno = ENOENT;
1309	if (!*rel)	1420	if (!*rel)
1310	return;	1421	return -1;
1311		1422
1312	if (!req->ptr2)	1423	res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1313	{
1314	X_LOCK (wrklock);
1315	req->flags |= EIO_FLAG_PTR2_FREE;
1316	X_UNLOCK (wrklock);
1317	req->ptr2 = malloc (PATH_MAX * 3);
1318
1319	errno = ENOMEM;
1320	if (!req->ptr2)
1321	return;
1322	}
1323
1324	res = req->ptr2;
1325	tmp1 = res + PATH_MAX;	1424	tmp1 = res + PATH_MAX;
1326	tmp2 = tmp1 + PATH_MAX;	1425	tmp2 = tmp1 + PATH_MAX;
1327		1426
1328	#if 0 /* disabled, the musl way to do things is just too racy */	1427	#if 0 /* disabled, the musl way to do things is just too racy */
1329	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)	1428	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
…		…
1348	#endif	1447	#endif
1349	#endif	1448	#endif
1350		1449
1351	if (*rel != '/')	1450	if (*rel != '/')
1352	{	1451	{
		1452	int len;
		1453
		1454	errno = ENOENT;
		1455	if (wd == EIO_INVALID_WD)
		1456	return -1;
		1457
		1458	if (wd == EIO_CWD)
		1459	{
1353	if (!getcwd (res, PATH_MAX))	1460	if (!getcwd (res, PATH_MAX))
1354	return;	1461	return -1;
		1462
		1463	len = strlen (res);
		1464	}
		1465	else
		1466	memcpy (res, wd->str, len = wd->len);
1355		1467
1356	if (res [1]) /* only use if not / */	1468	if (res [1]) /* only use if not / */
1357	res += strlen (res);	1469	res += len;
1358	}	1470	}
1359		1471
1360	while (*rel)	1472	while (*rel)
1361	{	1473	{
1362	ssize_t len, linklen;	1474	eio_ssize_t len, linklen;
1363	char *beg = rel;	1475	const char *beg = rel;
1364		1476
1365	while (*rel && *rel != '/')	1477	while (*rel && *rel != '/')
1366	++rel;	1478	++rel;
1367		1479
1368	len = rel - beg;	1480	len = rel - beg;
…		…
1380		1492
1381	if (beg [1] == '.' && len == 2)	1493	if (beg [1] == '.' && len == 2)
1382	{	1494	{
1383	/* .. - back up one component, if possible */	1495	/* .. - back up one component, if possible */
1384		1496
1385	while (res != req->ptr2)	1497	while (res != tmpbuf->ptr)
1386	if (*--res == '/')	1498	if (*--res == '/')
1387	break;	1499	break;
1388		1500
1389	continue;	1501	continue;
1390	}	1502	}
1391	}	1503	}
1392		1504
1393	errno = ENAMETOOLONG;	1505	errno = ENAMETOOLONG;
1394	if (res + 1 + len + 1 >= tmp1)	1506	if (res + 1 + len + 1 >= tmp1)
1395	return;	1507	return -1;
1396		1508
1397	/* copy one component */	1509	/* copy one component */
1398	*res = '/';	1510	*res = '/';
1399	memcpy (res + 1, beg, len);	1511	memcpy (res + 1, beg, len);
1400		1512
1401	/* zero-terminate, for readlink */	1513	/* zero-terminate, for readlink */
1402	res [len + 1] = 0;	1514	res [len + 1] = 0;
1403		1515
1404	/* now check if it's a symlink */	1516	/* now check if it's a symlink */
1405	linklen = readlink (req->ptr2, tmp1, PATH_MAX);	1517	linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1406		1518
1407	if (linklen < 0)	1519	if (linklen < 0)
1408	{	1520	{
1409	if (errno != EINVAL)	1521	if (errno != EINVAL)
1410	return;	1522	return -1;
1411		1523
1412	/* it's a normal directory. hopefully */	1524	/* it's a normal directory. hopefully */
1413	res += len + 1;	1525	res += len + 1;
1414	}	1526	}
1415	else	1527	else
…		…
1417	/* yay, it was a symlink - build new path in tmp2 */	1529	/* yay, it was a symlink - build new path in tmp2 */
1418	int rellen = strlen (rel);	1530	int rellen = strlen (rel);
1419		1531
1420	errno = ENAMETOOLONG;	1532	errno = ENAMETOOLONG;
1421	if (linklen + 1 + rellen >= PATH_MAX)	1533	if (linklen + 1 + rellen >= PATH_MAX)
1422	return;	1534	return -1;
1423		1535
1424	errno = ELOOP;	1536	errno = ELOOP;
1425	if (!--symlinks)	1537	if (!--symlinks)
1426	return;	1538	return -1;
1427		1539
1428	if (*tmp1 == '/')	1540	if (*tmp1 == '/')
1429	res = req->ptr2; /* symlink resolves to an absolute path */	1541	res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1430		1542
1431	/* we need to be careful, as rel might point into tmp2 already */	1543	/* we need to be careful, as rel might point into tmp2 already */
1432	memmove (tmp2 + linklen + 1, rel, rellen + 1);	1544	memmove (tmp2 + linklen + 1, rel, rellen + 1);
1433	tmp2 [linklen] = '/';	1545	tmp2 [linklen] = '/';
1434	memcpy (tmp2, tmp1, linklen);	1546	memcpy (tmp2, tmp1, linklen);
…		…
1436	rel = tmp2;	1548	rel = tmp2;
1437	}	1549	}
1438	}	1550	}
1439		1551
1440	/* special case for the lone root path */	1552	/* special case for the lone root path */
1441	if (res == req->ptr2)	1553	if (res == tmpbuf->ptr)
1442	*res++ = '/';	1554	*res++ = '/';
1443		1555
1444	req->result = res - (char *)req->ptr2;	1556	return res - (char *)tmpbuf->ptr;
1445
1446	done:
1447	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1448	}	1557	}
1449		1558
1450	static signed char	1559	static signed char
1451	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1560	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1452	{	1561	{
…		…
1460		1569
1461	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1570	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1462	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1571	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1463		1572
1464	static void	1573	static void
1465	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1574	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1466	{	1575	{
1467	unsigned char bits [9 + sizeof (ino_t) * 8];	1576	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1468	unsigned char *bit = bits;	1577	unsigned char *bit = bits;
1469		1578
1470	assert (CHAR_BIT == 8);	1579	assert (CHAR_BIT == 8);
1471	assert (sizeof (eio_dirent) * 8 < 256);	1580	assert (sizeof (eio_dirent) * 8 < 256);
1472	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1581	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1474		1583
1475	if (size <= EIO_SORT_FAST)	1584	if (size <= EIO_SORT_FAST)
1476	return;	1585	return;
1477		1586
1478	/* first prepare an array of bits to test in our radix sort */	1587	/* first prepare an array of bits to test in our radix sort */
1479	/* try to take endianness into account, as well as differences in ino_t sizes */	1588	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1480	/* inode_bits must contain all inodes ORed together */	1589	/* inode_bits must contain all inodes ORed together */
1481	/* which is used to skip bits that are 0 everywhere, which is very common */	1590	/* which is used to skip bits that are 0 everywhere, which is very common */
1482	{	1591	{
1483	ino_t endianness;	1592	eio_ino_t endianness;
1484	int i, j;	1593	int i, j;
1485		1594
1486	/* we store the byte offset of byte n into byte n of "endianness" */	1595	/* we store the byte offset of byte n into byte n of "endianness" */
1487	for (i = 0; i < sizeof (ino_t); ++i)	1596	for (i = 0; i < sizeof (eio_ino_t); ++i)
1488	((unsigned char *)&endianness)[i] = i;	1597	((unsigned char *)&endianness)[i] = i;
1489		1598
1490	*bit++ = 0;	1599	*bit++ = 0;
1491		1600
1492	for (i = 0; i < sizeof (ino_t); ++i)	1601	for (i = 0; i < sizeof (eio_ino_t); ++i)
1493	{	1602	{
1494	/* shifting off the byte offsets out of "endianness" */	1603	/* shifting off the byte offsets out of "endianness" */
1495	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1604	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1496	endianness >>= 8;	1605	endianness >>= 8;
1497		1606
1498	for (j = 0; j < 8; ++j)	1607	for (j = 0; j < 8; ++j)
1499	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1608	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1500	*bit++ = offs + j;	1609	*bit++ = offs + j;
1501	}	1610	}
1502		1611
1503	for (j = 0; j < 8; ++j)	1612	for (j = 0; j < 8; ++j)
1504	if (score_bits & (1 << j))	1613	if (score_bits & (1 << j))
1505	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1614	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1506	}	1615	}
1507		1616
1508	/* now actually do the sorting (a variant of MSD radix sort) */	1617	/* now actually do the sorting (a variant of MSD radix sort) */
1509	{	1618	{
1510	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1619	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1511	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1620	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1512	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1621	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1513	int stk_idx = 0;	1622	int stk_idx = 0;
1514		1623
1515	base_stk [stk_idx] = dents;	1624	base_stk [stk_idx] = dents;
1516	end_stk [stk_idx] = dents + size;	1625	end_stk [stk_idx] = dents + size;
1517	bit_stk [stk_idx] = bit - 1;	1626	bit_stk [stk_idx] = bit - 1;
…		…
1596	}	1705	}
1597	}	1706	}
1598	}	1707	}
1599		1708
1600	static void	1709	static void
1601	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1710	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1602	{	1711	{
1603	if (size <= 1)	1712	if (size <= 1)
1604	return; /* our insertion sort relies on size > 0 */	1713	return; /* our insertion sort relies on size > 0 */
1605		1714
1606	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1715	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1614		1723
1615	/* read a full directory */	1724	/* read a full directory */
1616	static void	1725	static void
1617	eio__scandir (eio_req *req, etp_worker *self)	1726	eio__scandir (eio_req *req, etp_worker *self)
1618	{	1727	{
1619	DIR *dirp;
1620	EIO_STRUCT_DIRENT *entp;
1621	char *name, *names;	1728	char *name, *names;
1622	int namesalloc = 4096;	1729	int namesalloc = 4096 - sizeof (void *) * 4;
1623	int namesoffs = 0;	1730	int namesoffs = 0;
1624	int flags = req->int1;	1731	int flags = req->int1;
1625	eio_dirent *dents = 0;	1732	eio_dirent *dents = 0;
1626	int dentalloc = 128;	1733	int dentalloc = 128;
1627	int dentoffs = 0;	1734	int dentoffs = 0;
1628	ino_t inode_bits = 0;	1735	eio_ino_t inode_bits = 0;
		1736	#ifdef _WIN32
		1737	HANDLE dirp;
		1738	WIN32_FIND_DATA entp;
		1739	#else
		1740	DIR *dirp;
		1741	EIO_STRUCT_DIRENT *entp;
		1742	#endif
1629		1743
1630	req->result = -1;	1744	req->result = -1;
1631		1745
1632	if (!(flags & EIO_READDIR_DENTS))	1746	if (!(flags & EIO_READDIR_DENTS))
1633	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1747	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1634		1748
1635	X_LOCK (wrklock);	1749	#ifdef _WIN32
1636	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1750	{
		1751	int len = strlen ((const char *)req->ptr1);
		1752	char *path = malloc (MAX_PATH);
		1753	const char *fmt;
		1754	const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		1755
		1756	if (!len)
		1757	fmt = "./*";
		1758	else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
		1759	fmt = "%s*";
		1760	else
		1761	fmt = "%s/*";
		1762
		1763	_snprintf (path, MAX_PATH, fmt, reqpath);
		1764	dirp = FindFirstFile (path, &entp);
		1765	free (path);
		1766
		1767	if (dirp == INVALID_HANDLE_VALUE)
		1768	{
		1769	/* should steal _dosmaperr */
		1770	switch (GetLastError ())
		1771	{
		1772	case ERROR_FILE_NOT_FOUND:
		1773	req->result = 0;
		1774	break;
		1775
		1776	case ERROR_INVALID_NAME:
		1777	case ERROR_PATH_NOT_FOUND:
		1778	case ERROR_NO_MORE_FILES:
		1779	errno = ENOENT;
		1780	break;
		1781
		1782	case ERROR_NOT_ENOUGH_MEMORY:
		1783	errno = ENOMEM;
		1784	break;
		1785
		1786	default:
		1787	errno = EINVAL;
		1788	break;
		1789	}
		1790
		1791	return;
		1792	}
		1793	}
		1794	#else
		1795	#if HAVE_AT
		1796	if (req->wd)
		1797	{
		1798	int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1799
		1800	if (fd < 0)
		1801	return;
		1802
		1803	dirp = fdopendir (fd);
		1804
		1805	if (!dirp)
		1806	close (fd);
		1807	}
		1808	else
1637	self->dirp = dirp = opendir (req->ptr1);	1809	dirp = opendir (req->ptr1);
		1810	#else
		1811	dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
		1812	#endif
		1813
		1814	if (!dirp)
		1815	return;
		1816	#endif
1638		1817
1639	if (req->flags & EIO_FLAG_PTR1_FREE)	1818	if (req->flags & EIO_FLAG_PTR1_FREE)
1640	free (req->ptr1);	1819	free (req->ptr1);
1641		1820
1642	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1821	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1643	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1822	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1644	req->ptr2 = names = malloc (namesalloc);	1823	req->ptr2 = names = malloc (namesalloc);
1645	X_UNLOCK (wrklock);
1646		1824
1647	if (dirp && names && (!flags || dents))	1825	if (!names || (flags && !dents))
		1826	return;
		1827
1648	for (;;)	1828	for (;;)
1649	{	1829	{
		1830	int done;
		1831
		1832	#ifdef _WIN32
		1833	done = !dirp;
		1834	#else
1650	errno = 0;	1835	errno = 0;
1651	entp = readdir (dirp);	1836	entp = readdir (dirp);
		1837	done = !entp;
		1838	#endif
1652		1839
1653	if (!entp)	1840	if (done)
1654	{	1841	{
		1842	#ifndef _WIN32
		1843	int old_errno = errno;
		1844	closedir (dirp);
		1845	errno = old_errno;
		1846
1655	if (errno)	1847	if (errno)
1656	break;	1848	break;
		1849	#endif
1657		1850
1658	/* sort etc. */	1851	/* sort etc. */
1659	req->int1 = flags;	1852	req->int1 = flags;
1660	req->result = dentoffs;	1853	req->result = dentoffs;
1661		1854
1662	if (flags & EIO_READDIR_STAT_ORDER)	1855	if (flags & EIO_READDIR_STAT_ORDER)
1663	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);	1856	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1664	else if (flags & EIO_READDIR_DIRS_FIRST)	1857	else if (flags & EIO_READDIR_DIRS_FIRST)
1665	if (flags & EIO_READDIR_FOUND_UNKNOWN)	1858	if (flags & EIO_READDIR_FOUND_UNKNOWN)
1666	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */	1859	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1667	else	1860	else
		1861	{
		1862	/* in this case, all is known, and we just put dirs first and sort them */
		1863	eio_dirent *oth = dents + dentoffs;
		1864	eio_dirent *dir = dents;
		1865
		1866	/* now partition dirs to the front, and non-dirs to the back */
		1867	/* by walking from both sides and swapping if necessary */
		1868	while (oth > dir)
		1869	{
		1870	if (dir->type == EIO_DT_DIR)
		1871	++dir;
		1872	else if ((--oth)->type == EIO_DT_DIR)
		1873	{
		1874	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1875
		1876	++dir;
		1877	}
		1878	}
		1879
		1880	/* now sort the dirs only (dirs all have the same score) */
		1881	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1882	}
		1883
		1884	break;
		1885	}
		1886
		1887	/* now add the entry to our list(s) */
		1888	name = D_NAME (entp);
		1889
		1890	/* skip . and .. entries */
		1891	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
		1892	{
		1893	int len = D_NAMLEN (entp) + 1;
		1894
		1895	while (ecb_expect_false (namesoffs + len > namesalloc))
		1896	{
		1897	namesalloc *= 2;
		1898	req->ptr2 = names = realloc (names, namesalloc);
		1899
		1900	if (!names)
		1901	break;
		1902	}
		1903
		1904	memcpy (names + namesoffs, name, len);
		1905
		1906	if (dents)
		1907	{
		1908	struct eio_dirent *ent;
		1909
		1910	if (ecb_expect_false (dentoffs == dentalloc))
1668	{	1911	{
1669	/* in this case, all is known, and we just put dirs first and sort them */	1912	dentalloc *= 2;
		1913	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1914
		1915	if (!dents)
		1916	break;
		1917	}
		1918
1670	eio_dirent *oth = dents + dentoffs;	1919	ent = dents + dentoffs;
1671	eio_dirent *dir = dents;
1672		1920
1673	/* now partition dirs to the front, and non-dirs to the back */	1921	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1674	/* by walking from both sides and swapping if necessary */	1922	ent->namelen = len - 1;
1675	while (oth > dir)	1923	ent->inode = D_INO (entp);
		1924
		1925	inode_bits |= ent->inode;
		1926
		1927	switch (D_TYPE (entp))
		1928	{
		1929	default:
		1930	ent->type = EIO_DT_UNKNOWN;
		1931	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1932	break;
		1933
		1934	#ifdef DT_FIFO
		1935	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1936	#endif
		1937	#ifdef DT_CHR
		1938	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1939	#endif
		1940	#ifdef DT_MPC
		1941	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1942	#endif
		1943	#ifdef DT_DIR
		1944	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1945	#endif
		1946	#ifdef DT_NAM
		1947	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1948	#endif
		1949	#ifdef DT_BLK
		1950	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1951	#endif
		1952	#ifdef DT_MPB
		1953	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1954	#endif
		1955	#ifdef DT_REG
		1956	case DT_REG: ent->type = EIO_DT_REG; break;
		1957	#endif
		1958	#ifdef DT_NWK
		1959	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1960	#endif
		1961	#ifdef DT_CMP
		1962	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1963	#endif
		1964	#ifdef DT_LNK
		1965	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1966	#endif
		1967	#ifdef DT_SOCK
		1968	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1969	#endif
		1970	#ifdef DT_DOOR
		1971	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1972	#endif
		1973	#ifdef DT_WHT
		1974	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1975	#endif
		1976	}
		1977
		1978	ent->score = 7;
		1979
		1980	if (flags & EIO_READDIR_DIRS_FIRST)
		1981	{
		1982	if (ent->type == EIO_DT_UNKNOWN)
1676	{	1983	{
1677	if (dir->type == EIO_DT_DIR)	1984	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1678	++dir;	1985	ent->score = 1;
1679	else if ((--oth)->type == EIO_DT_DIR)	1986	else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1680	{	1987	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1681	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1682
1683	++dir;
1684	}
1685	}	1988	}
1686		1989	else if (ent->type == EIO_DT_DIR)
1687	/* now sort the dirs only (dirs all have the same score) */	1990	ent->score = 0;
1688	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1689	}	1991	}
1690
1691	break;
1692	}
1693
1694	/* now add the entry to our list(s) */
1695	name = entp->d_name;
1696
1697	/* skip . and .. entries */
1698	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1699	{
1700	int len = D_NAMLEN (entp) + 1;
1701
1702	while (ecb_expect_false (namesoffs + len > namesalloc))
1703	{
1704	namesalloc *= 2;
1705	X_LOCK (wrklock);
1706	req->ptr2 = names = realloc (names, namesalloc);
1707	X_UNLOCK (wrklock);
1708
1709	if (!names)
1710	break;
1711	}	1992	}
1712		1993
1713	memcpy (names + namesoffs, name, len);
1714
1715	if (dents)
1716	{
1717	struct eio_dirent *ent;
1718
1719	if (ecb_expect_false (dentoffs == dentalloc))
1720	{
1721	dentalloc *= 2;
1722	X_LOCK (wrklock);
1723	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1724	X_UNLOCK (wrklock);
1725
1726	if (!dents)
1727	break;
1728	}
1729
1730	ent = dents + dentoffs;
1731
1732	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1733	ent->namelen = len - 1;
1734	ent->inode = D_INO (entp);
1735
1736	inode_bits |= ent->inode;
1737
1738	switch (D_TYPE (entp))
1739	{
1740	default:
1741	ent->type = EIO_DT_UNKNOWN;
1742	flags |= EIO_READDIR_FOUND_UNKNOWN;
1743	break;
1744
1745	#ifdef DT_FIFO
1746	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1747	#endif
1748	#ifdef DT_CHR
1749	case DT_CHR: ent->type = EIO_DT_CHR; break;
1750	#endif
1751	#ifdef DT_MPC
1752	case DT_MPC: ent->type = EIO_DT_MPC; break;
1753	#endif
1754	#ifdef DT_DIR
1755	case DT_DIR: ent->type = EIO_DT_DIR; break;
1756	#endif
1757	#ifdef DT_NAM
1758	case DT_NAM: ent->type = EIO_DT_NAM; break;
1759	#endif
1760	#ifdef DT_BLK
1761	case DT_BLK: ent->type = EIO_DT_BLK; break;
1762	#endif
1763	#ifdef DT_MPB
1764	case DT_MPB: ent->type = EIO_DT_MPB; break;
1765	#endif
1766	#ifdef DT_REG
1767	case DT_REG: ent->type = EIO_DT_REG; break;
1768	#endif
1769	#ifdef DT_NWK
1770	case DT_NWK: ent->type = EIO_DT_NWK; break;
1771	#endif
1772	#ifdef DT_CMP
1773	case DT_CMP: ent->type = EIO_DT_CMP; break;
1774	#endif
1775	#ifdef DT_LNK
1776	case DT_LNK: ent->type = EIO_DT_LNK; break;
1777	#endif
1778	#ifdef DT_SOCK
1779	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1780	#endif
1781	#ifdef DT_DOOR
1782	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1783	#endif
1784	#ifdef DT_WHT
1785	case DT_WHT: ent->type = EIO_DT_WHT; break;
1786	#endif
1787	}
1788
1789	ent->score = 7;
1790
1791	if (flags & EIO_READDIR_DIRS_FIRST)
1792	{
1793	if (ent->type == EIO_DT_UNKNOWN)
1794	{
1795	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1796	ent->score = 1;
1797	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1798	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1799	}
1800	else if (ent->type == EIO_DT_DIR)
1801	ent->score = 0;
1802	}
1803	}
1804
1805	namesoffs += len;	1994	namesoffs += len;
1806	++dentoffs;	1995	++dentoffs;
1807	}	1996	}
1808		1997
1809	if (EIO_CANCELLED (req))	1998	if (EIO_CANCELLED (req))
1810	{	1999	{
1811	errno = ECANCELED;	2000	errno = ECANCELED;
1812	break;	2001	break;
1813	}	2002	}
		2003
		2004	#ifdef _WIN32
		2005	if (!FindNextFile (dirp, &entp))
		2006	{
		2007	FindClose (dirp);
		2008	dirp = 0;
1814	}	2009	}
		2010	#endif
		2011	}
1815	}	2012	}
		2013
		2014	/*****************************************************************************/
		2015	/* working directory stuff */
		2016	/* various deficiencies in the posix 2008 api force us to */
		2017	/* keep the absolute path in string form at all times */
		2018	/* fuck yeah. */
		2019
		2020	#if !HAVE_AT
		2021
		2022	/* a bit like realpath, but usually faster because it doesn'T have to return */
		2023	/* an absolute or canonical path */
		2024	static const char *
		2025	wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2026	{
		2027	if (!wd || *path == '/')
		2028	return path;
		2029
		2030	if (path [0] == '.' && !path [1])
		2031	return wd->str;
		2032
		2033	{
		2034	int l1 = wd->len;
		2035	int l2 = strlen (path);
		2036
		2037	char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
		2038
		2039	memcpy (res, wd->str, l1);
		2040	res [l1] = '/';
		2041	memcpy (res + l1 + 1, path, l2 + 1);
		2042
		2043	return res;
		2044	}
		2045	}
		2046
		2047	#endif
		2048
		2049	static eio_wd
		2050	eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2051	{
		2052	int fd;
		2053	eio_wd res;
		2054	int len = eio__realpath (tmpbuf, wd, path);
		2055
		2056	if (len < 0)
		2057	return EIO_INVALID_WD;
		2058
		2059	#if HAVE_AT
		2060	fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		2061
		2062	if (fd < 0)
		2063	return EIO_INVALID_WD;
		2064	#endif
		2065
		2066	res = malloc (sizeof (*res) + len); /* one extra 0-byte */
		2067
		2068	#if HAVE_AT
		2069	res->fd = fd;
		2070	#endif
		2071
		2072	res->len = len;
		2073	memcpy (res->str, tmpbuf->ptr, len);
		2074	res->str [len] = 0;
		2075
		2076	return res;
		2077	}
		2078
		2079	eio_wd
		2080	eio_wd_open_sync (eio_wd wd, const char *path)
		2081	{
		2082	struct tmpbuf tmpbuf = { 0 };
		2083	wd = eio__wd_open_sync (&tmpbuf, wd, path);
		2084	free (tmpbuf.ptr);
		2085
		2086	return wd;
		2087	}
		2088
		2089	void
		2090	eio_wd_close_sync (eio_wd wd)
		2091	{
		2092	if (wd != EIO_INVALID_WD && wd != EIO_CWD)
		2093	{
		2094	#if HAVE_AT
		2095	close (wd->fd);
		2096	#endif
		2097	free (wd);
		2098	}
		2099	}
		2100
		2101	#if HAVE_AT
		2102
		2103	/* they forgot these */
		2104
		2105	static int
		2106	eio__truncateat (int dirfd, const char *path, off_t length)
		2107	{
		2108	int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
		2109	int res;
		2110
		2111	if (fd < 0)
		2112	return fd;
		2113
		2114	res = ftruncate (fd, length);
		2115	close (fd);
		2116	return res;
		2117	}
		2118
		2119	static int
		2120	eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
		2121	{
		2122	int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
		2123	int res;
		2124
		2125	if (fd < 0)
		2126	return fd;
		2127
		2128	res = fstatvfs (fd, buf);
		2129	close (fd);
		2130	return res;
		2131
		2132	}
		2133
		2134	#endif
1816		2135
1817	/*****************************************************************************/	2136	/*****************************************************************************/
1818		2137
1819	#define ALLOC(len) \	2138	#define ALLOC(len) \
1820	if (!req->ptr2) \	2139	if (!req->ptr2) \
…		…
1829	req->result = -1; \	2148	req->result = -1; \
1830	break; \	2149	break; \
1831	} \	2150	} \
1832	}	2151	}
1833		2152
		2153	static void ecb_noinline ecb_cold
		2154	etp_proc_init (void)
		2155	{
		2156	#if HAVE_PRCTL_SET_NAME
		2157	/* provide a more sensible "thread name" */
		2158	char name[16 + 1];
		2159	const int namelen = sizeof (name) - 1;
		2160	int len;
		2161
		2162	prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
		2163	name [namelen] = 0;
		2164	len = strlen (name);
		2165	strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
		2166	prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
		2167	#endif
		2168	}
		2169
1834	X_THREAD_PROC (etp_proc)	2170	X_THREAD_PROC (etp_proc)
1835	{	2171	{
1836	ETP_REQ *req;	2172	ETP_REQ *req;
1837	struct timespec ts;	2173	struct timespec ts;
1838	etp_worker *self = (etp_worker *)thr_arg;	2174	etp_worker *self = (etp_worker *)thr_arg;
1839	int timeout;	2175
		2176	etp_proc_init ();
1840		2177
1841	/* try to distribute timeouts somewhat evenly */	2178	/* try to distribute timeouts somewhat evenly */
1842	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	2179	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1843		2180
1844	for (;;)	2181	for (;;)
…		…
1847		2184
1848	X_LOCK (reqlock);	2185	X_LOCK (reqlock);
1849		2186
1850	for (;;)	2187	for (;;)
1851	{	2188	{
1852	self->req = req = reqq_shift (&req_queue);	2189	req = reqq_shift (&req_queue);
1853		2190
1854	if (req)	2191	if (req)
1855	break;	2192	break;
1856		2193
1857	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */	2194	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
…		…
1895	++npending;	2232	++npending;
1896		2233
1897	if (!reqq_push (&res_queue, req) && want_poll_cb)	2234	if (!reqq_push (&res_queue, req) && want_poll_cb)
1898	want_poll_cb ();	2235	want_poll_cb ();
1899		2236
1900	self->req = 0;
1901	etp_worker_clear (self);	2237	etp_worker_clear (self);
1902		2238
1903	X_UNLOCK (reslock);	2239	X_UNLOCK (reslock);
1904	}	2240	}
1905		2241
1906	quit:	2242	quit:
		2243	free (req);
		2244
1907	X_LOCK (wrklock);	2245	X_LOCK (wrklock);
1908	etp_worker_free (self);	2246	etp_worker_free (self);
1909	X_UNLOCK (wrklock);	2247	X_UNLOCK (wrklock);
1910		2248
1911	return 0;	2249	return 0;
…		…
1914	/*****************************************************************************/	2252	/*****************************************************************************/
1915		2253
1916	int ecb_cold	2254	int ecb_cold
1917	eio_init (void (*want_poll)(void), void (*done_poll)(void))	2255	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1918	{	2256	{
		2257	#if !HAVE_PREADWRITE
		2258	X_MUTEX_CREATE (preadwritelock);
		2259	#endif
		2260
1919	return etp_init (want_poll, done_poll);	2261	return etp_init (want_poll, done_poll);
1920	}	2262	}
1921		2263
1922	ecb_inline void	2264	ecb_inline void
1923	eio_api_destroy (eio_req *req)	2265	eio_api_destroy (eio_req *req)
…		…
1950	}	2292	}
1951		2293
1952	static void	2294	static void
1953	eio_execute (etp_worker *self, eio_req *req)	2295	eio_execute (etp_worker *self, eio_req *req)
1954	{	2296	{
		2297	#if HAVE_AT
		2298	int dirfd;
		2299	#else
		2300	const char *path;
		2301	#endif
		2302
1955	if (ecb_expect_false (EIO_CANCELLED (req)))	2303	if (ecb_expect_false (EIO_CANCELLED (req)))
1956	{	2304	{
1957	req->result = -1;	2305	req->result = -1;
1958	req->errorno = ECANCELED;	2306	req->errorno = ECANCELED;
1959	return;	2307	return;
1960	}	2308	}
1961		2309
		2310	if (ecb_expect_false (req->wd == EIO_INVALID_WD))
		2311	{
		2312	req->result = -1;
		2313	req->errorno = ENOENT;
		2314	return;
		2315	}
		2316
		2317	if (req->type >= EIO_OPEN)
		2318	{
		2319	#if HAVE_AT
		2320	dirfd = WD2FD (req->wd);
		2321	#else
		2322	path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		2323	#endif
		2324	}
		2325
1962	switch (req->type)	2326	switch (req->type)
1963	{	2327	{
		2328	case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
		2329	req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
		2330	break;
		2331	case EIO_WD_CLOSE: req->result = 0;
		2332	eio_wd_close_sync (req->wd); break;
		2333
		2334	case EIO_SEEK: eio__lseek (req); break;
1964	case EIO_READ: ALLOC (req->size);	2335	case EIO_READ: ALLOC (req->size);
1965	req->result = req->offs >= 0	2336	req->result = req->offs >= 0
1966	? pread (req->int1, req->ptr2, req->size, req->offs)	2337	? pread (req->int1, req->ptr2, req->size, req->offs)
1967	: read (req->int1, req->ptr2, req->size); break;	2338	: read (req->int1, req->ptr2, req->size); break;
1968	case EIO_WRITE: req->result = req->offs >= 0	2339	case EIO_WRITE: req->result = req->offs >= 0
1969	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2340	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1970	: write (req->int1, req->ptr2, req->size); break;	2341	: write (req->int1, req->ptr2, req->size); break;
1971		2342
1972	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2343	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1973	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2344	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
		2345
		2346	#if HAVE_AT
1974		2347
1975	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2348	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1976	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2349	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1977	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2350	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2351	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
		2352	case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
		2353	case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
		2354	case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
		2355	case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
		2356
		2357	case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
		2358	case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
		2359	case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
		2360	case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
		2361	case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
		2362	case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
		2363	case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2364	case EIO_READLINK: ALLOC (PATH_MAX);
		2365	req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
		2366	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2367	req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2368	case EIO_UTIME:
		2369	case EIO_FUTIME:
		2370	{
		2371	struct timespec ts[2];
		2372	struct timespec *times;
		2373
		2374	if (req->nv1 != -1. || req->nv2 != -1.)
		2375	{
		2376	ts[0].tv_sec = req->nv1;
		2377	ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
		2378	ts[1].tv_sec = req->nv2;
		2379	ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
		2380
		2381	times = ts;
		2382	}
		2383	else
		2384	times = 0;
		2385
		2386	req->result = req->type == EIO_FUTIME
		2387	? futimens (req->int1, times)
		2388	: utimensat (dirfd, req->ptr1, times, 0);
		2389	}
		2390	break;
		2391
		2392	#else
		2393
		2394	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2395	req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2396	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1978	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2397	req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2398	case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
		2399	case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
		2400	case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
		2401	case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
		2402
		2403	case EIO_UNLINK: req->result = unlink (path ); break;
		2404	case EIO_RMDIR: req->result = rmdir (path ); break;
		2405	case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
		2406	case EIO_RENAME: req->result = rename (path , req->ptr2); break;
		2407	case EIO_LINK: req->result = link (path , req->ptr2); break;
		2408	case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
		2409	case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
		2410	case EIO_READLINK: ALLOC (PATH_MAX);
		2411	req->result = readlink (path, req->ptr2, PATH_MAX); break;
		2412	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2413	req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2414
		2415	case EIO_UTIME:
		2416	case EIO_FUTIME:
		2417	{
		2418	struct timeval tv[2];
		2419	struct timeval *times;
		2420
		2421	if (req->nv1 != -1. || req->nv2 != -1.)
		2422	{
		2423	tv[0].tv_sec = req->nv1;
		2424	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
		2425	tv[1].tv_sec = req->nv2;
		2426	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
		2427
		2428	times = tv;
		2429	}
		2430	else
		2431	times = 0;
		2432
		2433	req->result = req->type == EIO_FUTIME
		2434	? futimes (req->int1, times)
		2435	: utimes (req->ptr1, times);
		2436	}
		2437	break;
		2438
		2439	#endif
		2440
		2441	case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
		2442	{
		2443	ALLOC (req->result);
		2444	memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
		2445	}
		2446	break;
		2447
1979	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2448	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1980	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2449	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1981		2450
1982	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1983	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1984	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));	2451	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1985	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;	2452	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1986		2453
1987	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1988	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	2454	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1989	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1990	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	2455	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1991	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1992	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;	2456	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1993		2457
1994	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1995	case EIO_CLOSE: req->result = close (req->int1); break;	2458	case EIO_CLOSE: req->result = close (req->int1); break;
1996	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;	2459	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1997	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1998	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1999	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2000	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2001	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2002	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2003	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2004
2005	case EIO_REALPATH: eio__realpath (req, self); break;
2006
2007	case EIO_READLINK: ALLOC (PATH_MAX);
2008	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2009
2010	case EIO_SYNC: req->result = 0; sync (); break;	2460	case EIO_SYNC: req->result = 0; sync (); break;
2011	case EIO_FSYNC: req->result = fsync (req->int1); break;	2461	case EIO_FSYNC: req->result = fsync (req->int1); break;
2012	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2462	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2463	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2464	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2013	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2465	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2014	case EIO_MTOUCH: req->result = eio__mtouch (req); break;	2466	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2015	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2467	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2016	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2468	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2017	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2018	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;	2469	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2019		2470
2020	case EIO_READDIR: eio__scandir (req, self); break;	2471	case EIO_READDIR: eio__scandir (req, self); break;
2021		2472
2022	case EIO_BUSY:	2473	case EIO_BUSY:
…		…
2032	req->result = select (0, 0, 0, 0, &tv);	2483	req->result = select (0, 0, 0, 0, &tv);
2033	}	2484	}
2034	#endif	2485	#endif
2035	break;	2486	break;
2036		2487
2037	case EIO_UTIME:
2038	case EIO_FUTIME:
2039	{
2040	struct timeval tv[2];
2041	struct timeval *times;
2042
2043	if (req->nv1 != -1. || req->nv2 != -1.)
2044	{
2045	tv[0].tv_sec = req->nv1;
2046	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2047	tv[1].tv_sec = req->nv2;
2048	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2049
2050	times = tv;
2051	}
2052	else
2053	times = 0;
2054
2055	req->result = req->type == EIO_FUTIME
2056	? futimes (req->int1, times)
2057	: utimes (req->ptr1, times);
2058	}
2059	break;
2060
2061	case EIO_GROUP:	2488	case EIO_GROUP:
2062	abort (); /* handled in eio_request */	2489	abort (); /* handled in eio_request */
2063		2490
2064	case EIO_NOP:	2491	case EIO_NOP:
2065	req->result = 0;	2492	req->result = 0;
…		…
2078	req->errorno = errno;	2505	req->errorno = errno;
2079	}	2506	}
2080		2507
2081	#ifndef EIO_NO_WRAPPERS	2508	#ifndef EIO_NO_WRAPPERS
2082		2509
		2510	eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
		2511	{
		2512	REQ (EIO_WD_OPEN); PATH; SEND;
		2513	}
		2514
		2515	eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
		2516	{
		2517	REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
		2518	}
		2519
2083	eio_req *eio_nop (int pri, eio_cb cb, void *data)	2520	eio_req *eio_nop (int pri, eio_cb cb, void *data)
2084	{	2521	{
2085	REQ (EIO_NOP); SEND;	2522	REQ (EIO_NOP); SEND;
2086	}	2523	}
2087		2524
…		…
2103	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2540	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2104	{	2541	{
2105	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2542	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2106	}	2543	}
2107		2544
		2545	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		2546	{
		2547	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2548	}
		2549
		2550	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2551	{
		2552	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2553	}
		2554
		2555	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2556	{
		2557	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2558	}
		2559
2108	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2560	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2109	{	2561	{
2110	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2562	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2111	}	2563	}
2112		2564
…		…
2118	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)	2570	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2119	{	2571	{
2120	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;	2572	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2121	}	2573	}
2122		2574
2123	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2124	{
2125	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2126	}
2127
2128	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)	2575	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2129	{	2576	{
2130	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;	2577	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2131	}	2578	}
2132		2579
2133	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2134	{
2135	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2136	}
2137
2138	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2580	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2139	{	2581	{
2140	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2582	REQ (EIO_CLOSE); req->int1 = fd; SEND;
2141	}	2583	}
2142		2584
2143	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)	2585	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2144	{	2586	{
2145	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;	2587	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2146	}	2588	}
2147		2589
		2590	eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
		2591	{
		2592	REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
		2593	}
		2594
2148	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)	2595	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2149	{	2596	{
2150	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;	2597	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2151	}	2598	}
2152		2599
…		…
2178	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2625	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2179	{	2626	{
2180	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2627	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2181	}	2628	}
2182		2629
2183	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2630	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2184	{	2631	{
2185	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2632	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2186	}	2633	}
2187		2634
2188	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2635	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
2208	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2655	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2209	{	2656	{
2210	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2657	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2211	}	2658	}
2212		2659
2213	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2660	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2214	{	2661	{
2215	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2662	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2216	}	2663	}
2217		2664
2218	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2665	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
2365	}	2812	}
2366		2813
2367	/*****************************************************************************/	2814	/*****************************************************************************/
2368	/* misc garbage */	2815	/* misc garbage */
2369		2816
2370	ssize_t	2817	eio_ssize_t
2371	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2818	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2372	{	2819	{
2373	etp_worker wrk;
2374	ssize_t ret;
2375
2376	wrk.dbuf = 0;
2377
2378	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2820	return eio__sendfile (ofd, ifd, offset, count);
2379
2380	if (wrk.dbuf)
2381	free (wrk.dbuf);
2382
2383	return ret;
2384	}	2821	}
2385		2822

Diff Legend

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