ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.89 by root, Thu Jul 14 22:36:17 2011 UTC vs.
Revision 1.120 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>
…		…
67	#endif	67	#endif
68		68
69	#ifndef ECANCELED	69	#ifndef ECANCELED
70	# define ECANCELED EDOM	70	# define ECANCELED EDOM
71	#endif	71	#endif
		72	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
		77	# define ENOTSOCK WSAENOTSOCK
		78	#endif
72		79
73	static void eio_destroy (eio_req *req);	80	static void eio_destroy (eio_req *req);
74		81
75	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
76	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	83	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
…		…
95		102
96	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)	103	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
97		104
98	#ifdef _WIN32	105	#ifdef _WIN32
99		106
		107	#undef PAGESIZE
100	#define PAGESIZE 4096 /* GetSystemInfo? */	108	#define PAGESIZE 4096 /* GetSystemInfo? */
101		109
		110	/* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
		111
102	#ifdef EIO_STRUCT_STATI64	112	#ifdef EIO_STRUCT_STATI64
		113	/* look at perl's non-sloppy stat */
103	#define stat(path,buf) _stati64 (path,buf)	114	#define stat(path,buf) _stati64 (path,buf)
104	#define fstat(fd,buf) _fstati64 (path,buf)	115	#define fstat(fd,buf) _fstati64 (fd,buf)
105	#endif	116	#endif
106	#define lstat(path,buf) stat (path,buf)	117	#define lstat(path,buf) stat (path,buf)
107	#define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))	118	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
108	#define mkdir(path,mode) _mkdir (path)	119	#define mkdir(path,mode) _mkdir (path)
109	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))	120	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
110		121
		122	#define chmod(path,mode) _chmod (path, mode)
		123	#define dup(fd) _dup (fd)
		124	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		125
		126	#define fchmod(fd,mode) EIO_ENOSYS ()
111	#define chown(path,uid,gid) EIO_ENOSYS ()	127	#define chown(path,uid,gid) EIO_ENOSYS ()
112	#define fchown(fd,uid,gid) EIO_ENOSYS ()	128	#define fchown(fd,uid,gid) EIO_ENOSYS ()
113	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */	129	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
114	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */	130	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
115	#define mknod(path,mode,dev) EIO_ENOSYS ()	131	#define mknod(path,mode,dev) EIO_ENOSYS ()
116	#define sync() EIO_ENOSYS ()	132	#define sync() EIO_ENOSYS ()
		133	#define readlink(path,buf,s) EIO_ENOSYS ()
		134	#define statvfs(path,buf) EIO_ENOSYS ()
		135	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		136
		137	/* rename() uses MoveFile, which fails to overwrite */
		138	#define rename(old,neu) eio__rename (old, neu)
		139
		140	static int
		141	eio__rename (const char *old, const char *neu)
		142	{
		143	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		144	return 0;
		145
		146	/* should steal _dosmaperr */
		147	switch (GetLastError ())
		148	{
		149	case ERROR_FILE_NOT_FOUND:
		150	case ERROR_PATH_NOT_FOUND:
		151	case ERROR_INVALID_DRIVE:
		152	case ERROR_NO_MORE_FILES:
		153	case ERROR_BAD_NETPATH:
		154	case ERROR_BAD_NET_NAME:
		155	case ERROR_BAD_PATHNAME:
		156	case ERROR_FILENAME_EXCED_RANGE:
		157	errno = ENOENT;
		158	break;
		159
		160	default:
		161	errno = EACCES;
		162	break;
		163	}
		164
		165	return -1;
		166	}
117		167
118	/* we could even stat and see if it exists */	168	/* we could even stat and see if it exists */
119	static int	169	static int
120	symlink (const char *old, const char *neu)	170	symlink (const char *old, const char *neu)
121	{	171	{
		172	#if WINVER >= 0x0600
122	if (CreateSymbolicLink (neu, old, 1))	173	if (CreateSymbolicLink (neu, old, 1))
123	return 0;	174	return 0;
124		175
125	if (CreateSymbolicLink (neu, old, 0))	176	if (CreateSymbolicLink (neu, old, 0))
126	return 0;	177	return 0;
		178	#endif
127		179
128	return EIO_ERRNO (ENOENT, -1);	180	return EIO_ERRNO (ENOENT, -1);
129	}	181	}
		182
		183	/* POSIX API only */
		184	#define CreateHardLink(neu,old,flags) 0
		185	#define CreateSymbolicLink(neu,old,flags) 0
		186
		187	struct statvfs
		188	{
		189	int dummy;
		190	};
		191
		192	#define DT_DIR EIO_DT_DIR
		193	#define DT_REG EIO_DT_REG
		194	#define D_NAME(entp) entp.cFileName
		195	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
130		196
131	#else	197	#else
132		198
133	#include <sys/time.h>	199	#include <sys/time.h>
134	#include <sys/select.h>	200	#include <sys/select.h>
135	#include <sys/statvfs.h>	201	#include <sys/statvfs.h>
136	#include <unistd.h>	202	#include <unistd.h>
137	#include <utime.h>
138	#include <signal.h>	203	#include <signal.h>
139	#include <dirent.h>	204	#include <dirent.h>
140		205
141	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES	206	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
142	#include <sys/mman.h>	207	#include <sys/mman.h>
143	#endif	208	#endif
144		209
		210	#define D_NAME(entp) entp->d_name
		211
145	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */	212	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
146	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__	213	#if __FreeBSD__ || __NetBSD__ || __OpenBSD__
147	#define _DIRENT_HAVE_D_TYPE /* sigh */	214	#define _DIRENT_HAVE_D_TYPE /* sigh */
148	#define D_INO(de) (de)->d_fileno	215	#define D_INO(de) (de)->d_fileno
149	#define D_NAMLEN(de) (de)->d_namlen	216	#define D_NAMLEN(de) (de)->d_namlen
150	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600	217	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
151	#define D_INO(de) (de)->d_ino	218	#define D_INO(de) (de)->d_ino
…		…
162		229
163	#ifndef EIO_STRUCT_DIRENT	230	#ifndef EIO_STRUCT_DIRENT
164	#define EIO_STRUCT_DIRENT struct dirent	231	#define EIO_STRUCT_DIRENT struct dirent
165	#endif	232	#endif
166		233
		234	#endif
		235
		236	#if HAVE_UTIMES
		237	# include <utime.h>
		238	#endif
		239
		240	#if HAVE_SYS_SYSCALL_H
		241	# include <sys/syscall.h>
		242	#endif
		243
		244	#if HAVE_SYS_PRCTL_H
		245	# include <sys/prctl.h>
167	#endif	246	#endif
168		247
169	#if HAVE_SENDFILE	248	#if HAVE_SENDFILE
170	# if __linux	249	# if __linux
171	# include <sys/sendfile.h>	250	# include <sys/sendfile.h>
…		…
186	#endif	265	#endif
187	#ifndef D_INO	266	#ifndef D_INO
188	# define D_INO(de) 0	267	# define D_INO(de) 0
189	#endif	268	#endif
190	#ifndef D_NAMLEN	269	#ifndef D_NAMLEN
191	# define D_NAMLEN(de) strlen ((de)->d_name)	270	# define D_NAMLEN(entp) strlen (D_NAME (entp))
192	#endif	271	#endif
193		272
194	/* used for struct dirent, AIX doesn't provide it */	273	/* used for struct dirent, AIX doesn't provide it */
195	#ifndef NAME_MAX	274	#ifndef NAME_MAX
196	# define NAME_MAX 4096	275	# define NAME_MAX 4096
…		…
203		282
204	/* buffer size for various temporary buffers */	283	/* buffer size for various temporary buffers */
205	#define EIO_BUFSIZE 65536	284	#define EIO_BUFSIZE 65536
206		285
207	#define dBUF \	286	#define dBUF \
208	char *eio_buf; \
209	ETP_WORKER_LOCK (self); \
210	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	287	char *eio_buf = malloc (EIO_BUFSIZE); \
211	ETP_WORKER_UNLOCK (self); \
212	errno = ENOMEM; \	288	errno = ENOMEM; \
213	if (!eio_buf) \	289	if (!eio_buf) \
214	return -1;	290	return -1
		291
		292	#define FUBd \
		293	free (eio_buf)
215		294
216	#define EIO_TICKS ((1000000 + 1023) >> 10)	295	#define EIO_TICKS ((1000000 + 1023) >> 10)
		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	/*****************************************************************************/
217		340
218	#define ETP_PRI_MIN EIO_PRI_MIN	341	#define ETP_PRI_MIN EIO_PRI_MIN
219	#define ETP_PRI_MAX EIO_PRI_MAX	342	#define ETP_PRI_MAX EIO_PRI_MAX
220		343
221	struct etp_worker;	344	struct etp_worker;
…		…
225	static int eio_finish (eio_req *req);	348	static int eio_finish (eio_req *req);
226	#define ETP_FINISH(req) eio_finish (req)	349	#define ETP_FINISH(req) eio_finish (req)
227	static void eio_execute (struct etp_worker *self, eio_req *req);	350	static void eio_execute (struct etp_worker *self, eio_req *req);
228	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	351	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
229		352
230	#define ETP_WORKER_CLEAR(req) \
231	if (wrk->dbuf) \
232	{ \
233	free (wrk->dbuf); \
234	wrk->dbuf = 0; \
235	} \
236	\
237	if (wrk->dirp) \
238	{ \
239	closedir (wrk->dirp); \
240	wrk->dirp = 0; \
241	}
242
243	#define ETP_WORKER_COMMON \
244	void *dbuf; \
245	DIR *dirp;
246
247	/*****************************************************************************/	353	/*****************************************************************************/
248		354
249	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	355	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
250		356
251	/* calculate time difference in ~1/EIO_TICKS of a second */	357	/* calculate time difference in ~1/EIO_TICKS of a second */
…		…
262	static void (*done_poll_cb) (void);	368	static void (*done_poll_cb) (void);
263		369
264	static unsigned int max_poll_time; /* reslock */	370	static unsigned int max_poll_time; /* reslock */
265	static unsigned int max_poll_reqs; /* reslock */	371	static unsigned int max_poll_reqs; /* reslock */
266		372
267	static volatile unsigned int nreqs; /* reqlock */	373	static unsigned int nreqs; /* reqlock */
268	static volatile unsigned int nready; /* reqlock */	374	static unsigned int nready; /* reqlock */
269	static volatile unsigned int npending; /* reqlock */	375	static unsigned int npending; /* reqlock */
270	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 */
271	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 */
272		378
273	static xmutex_t wrklock;	379	static xmutex_t wrklock;
274	static xmutex_t reslock;	380	static xmutex_t reslock;
275	static xmutex_t reqlock;	381	static xmutex_t reqlock;
276	static xcond_t reqwait;	382	static xcond_t reqwait;
…		…
279	/*	385	/*
280	* make our pread/pwrite emulation safe against themselves, but not against	386	* make our pread/pwrite emulation safe against themselves, but not against
281	* 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,
282	* but that's your problem, not mine.	388	* but that's your problem, not mine.
283	*/	389	*/
284	static xmutex_t preadwritelock = X_MUTEX_INIT;	390	static xmutex_t preadwritelock;
285	#endif	391	#endif
286		392
287	typedef struct etp_worker	393	typedef struct etp_worker
288	{	394	{
		395	struct tmpbuf tmpbuf;
		396
289	/* locked by wrklock */	397	/* locked by wrklock */
290	struct etp_worker *prev, *next;	398	struct etp_worker *prev, *next;
291		399
292	xthread_t tid;	400	xthread_t tid;
293		401
294	/* locked by reslock, reqlock or wrklock */	402	#ifdef ETP_WORKER_COMMON
295	ETP_REQ *req; /* currently processed request */
296
297	ETP_WORKER_COMMON	403	ETP_WORKER_COMMON
		404	#endif
298	} etp_worker;	405	} etp_worker;
299		406
300	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	407	static etp_worker wrk_first; /* NOT etp */
301		408
302	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	409	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
303	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	410	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
304		411
305	/* worker threads management */	412	/* worker threads management */
306		413
307	static void ecb_cold	414	static void
308	etp_worker_clear (etp_worker *wrk)	415	etp_worker_clear (etp_worker *wrk)
309	{	416	{
310	ETP_WORKER_CLEAR (wrk);
311	}	417	}
312		418
313	static void ecb_cold	419	static void ecb_cold
314	etp_worker_free (etp_worker *wrk)	420	etp_worker_free (etp_worker *wrk)
315	{	421	{
		422	free (wrk->tmpbuf.ptr);
		423
316	wrk->next->prev = wrk->prev;	424	wrk->next->prev = wrk->prev;
317	wrk->prev->next = wrk->next;	425	wrk->prev->next = wrk->next;
318		426
319	free (wrk);	427	free (wrk);
320	}	428	}
…		…
376	} etp_reqq;	484	} etp_reqq;
377		485
378	static etp_reqq req_queue;	486	static etp_reqq req_queue;
379	static etp_reqq res_queue;	487	static etp_reqq res_queue;
380		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
381	static int ecb_noinline	500	static int ecb_noinline
382	reqq_push (etp_reqq *q, ETP_REQ *req)	501	reqq_push (etp_reqq *q, ETP_REQ *req)
383	{	502	{
384	int pri = req->pri;	503	int pri = req->pri;
385	req->next = 0;	504	req->next = 0;
…		…
419	}	538	}
420		539
421	abort ();	540	abort ();
422	}	541	}
423		542
424	static void ecb_cold	543	static int ecb_cold
425	etp_thread_init (void)	544	etp_init (void (*want_poll)(void), void (*done_poll)(void))
426	{	545	{
427	#if !HAVE_PREADWRITE
428	X_MUTEX_CREATE (preadwritelock);
429	#endif
430	X_MUTEX_CREATE (wrklock);	546	X_MUTEX_CREATE (wrklock);
431	X_MUTEX_CREATE (reslock);	547	X_MUTEX_CREATE (reslock);
432	X_MUTEX_CREATE (reqlock);	548	X_MUTEX_CREATE (reqlock);
433	X_COND_CREATE (reqwait);	549	X_COND_CREATE (reqwait);
434	}
435		550
436	static void ecb_cold	551	reqq_init (&req_queue);
437	etp_atfork_prepare (void)	552	reqq_init (&res_queue);
438	{
439	}
440		553
441	static void ecb_cold	554	wrk_first.next =
442	etp_atfork_parent (void)	555	wrk_first.prev = &wrk_first;
443	{
444	}
445
446	static void ecb_cold
447	etp_atfork_child (void)
448	{
449	ETP_REQ *prv;
450
451	while ((prv = reqq_shift (&req_queue)))
452	ETP_DESTROY (prv);
453
454	while ((prv = reqq_shift (&res_queue)))
455	ETP_DESTROY (prv);
456
457	while (wrk_first.next != &wrk_first)
458	{
459	etp_worker *wrk = wrk_first.next;
460
461	if (wrk->req)
462	ETP_DESTROY (wrk->req);
463
464	etp_worker_clear (wrk);
465	etp_worker_free (wrk);
466	}
467		556
468	started = 0;	557	started = 0;
469	idle = 0;	558	idle = 0;
470	nreqs = 0;	559	nreqs = 0;
471	nready = 0;	560	nready = 0;
472	npending = 0;	561	npending = 0;
473
474	etp_thread_init ();
475	}
476
477	static void ecb_cold
478	etp_once_init (void)
479	{
480	etp_thread_init ();
481	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
482	}
483
484	static int ecb_cold
485	etp_init (void (*want_poll)(void), void (*done_poll)(void))
486	{
487	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
488
489	pthread_once (&doinit, etp_once_init);
490		562
491	want_poll_cb = want_poll;	563	want_poll_cb = want_poll;
492	done_poll_cb = done_poll;	564	done_poll_cb = done_poll;
493		565
494	return 0;	566	return 0;
…		…
534	}	606	}
535		607
536	static void ecb_cold	608	static void ecb_cold
537	etp_end_thread (void)	609	etp_end_thread (void)
538	{	610	{
539	eio_req *req = calloc (1, sizeof (eio_req));	611	eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
540		612
541	req->type = -1;	613	req->type = -1;
542	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	614	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
543		615
544	X_LOCK (reqlock);	616	X_LOCK (reqlock);
…		…
952	#if !HAVE_FDATASYNC	1024	#if !HAVE_FDATASYNC
953	# undef fdatasync	1025	# undef fdatasync
954	# define fdatasync(fd) fsync (fd)	1026	# define fdatasync(fd) fsync (fd)
955	#endif	1027	#endif
956		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	}
		1045
957	/* sync_file_range always needs emulation */	1046	/* sync_file_range always needs emulation */
958	static int	1047	static int
959	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)
960	{	1049	{
961	#if HAVE_SYNC_FILE_RANGE	1050	#if HAVE_SYNC_FILE_RANGE
…		…
983	}	1072	}
984		1073
985	static int	1074	static int
986	eio__fallocate (int fd, int mode, off_t offset, size_t len)	1075	eio__fallocate (int fd, int mode, off_t offset, size_t len)
987	{	1076	{
988	#if HAVE_FALLOCATE	1077	#if HAVE_LINUX_FALLOCATE
989	return fallocate (fd, mode, offset, len);	1078	return fallocate (fd, mode, offset, len);
990	#else	1079	#else
991	errno = ENOSYS;	1080	return EIO_ENOSYS ();
992	return -1;
993	#endif	1081	#endif
994	}	1082	}
995		1083
996	#if !HAVE_READAHEAD	1084	#if !HAVE_READAHEAD
997	# undef readahead	1085	# undef readahead
…		…
1010	pread (fd, eio_buf, len, offset);	1098	pread (fd, eio_buf, len, offset);
1011	offset += len;	1099	offset += len;
1012	todo -= len;	1100	todo -= len;
1013	}	1101	}
1014		1102
1015	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 */
1016	return count;	1107	return 0;
1017	}	1108	}
1018		1109
1019	#endif	1110	#endif
1020		1111
1021	/* sendfile always needs emulation */	1112	/* sendfile always needs emulation */
1022	static eio_ssize_t	1113	static eio_ssize_t
1023	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)
1024	{	1115	{
1025	eio_ssize_t written = 0;	1116	eio_ssize_t written = 0;
1026	eio_ssize_t res;	1117	eio_ssize_t res;
1027		1118
1028	if (!count)	1119	if (!count)
…		…
1056		1147
1057	/* according to source inspection, this is correct, and useful behaviour */	1148	/* according to source inspection, this is correct, and useful behaviour */
1058	if (sbytes)	1149	if (sbytes)
1059	res = sbytes;	1150	res = sbytes;
1060		1151
1061	# elif defined (__APPLE__)	1152	# elif defined __APPLE__
1062	off_t sbytes = count;	1153	off_t sbytes = count;
1063	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	1154	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1064		1155
1065	/* according to the manpage, sbytes is always valid */	1156	/* according to the manpage, sbytes is always valid */
1066	if (sbytes)	1157	if (sbytes)
…		…
1093	HANDLE h = TO_SOCKET (ifd);	1184	HANDLE h = TO_SOCKET (ifd);
1094	SetFilePointer (h, offset, 0, FILE_BEGIN);	1185	SetFilePointer (h, offset, 0, FILE_BEGIN);
1095	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);	1186	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1096		1187
1097	#else	1188	#else
1098	res = -1;	1189	res = EIO_ENOSYS ();
1099	errno = ENOSYS;
1100	#endif	1190	#endif
1101		1191
1102	/* we assume sendfile can copy at least 128mb in one go */	1192	/* we assume sendfile can copy at least 128mb in one go */
1103	if (res <= 128 * 1024 * 1024)	1193	if (res <= 128 * 1024 * 1024)
1104	{	1194	{
…		…
1126	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1216	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1127	/* BSDs */	1217	/* BSDs */
1128	#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... */
1129	|| errno == ENOTSUP	1219	|| errno == ENOTSUP
1130	#endif	1220	#endif
		1221	#ifdef EOPNOTSUPP /* windows */
1131	|| errno == EOPNOTSUPP /* BSDs */	1222	|| errno == EOPNOTSUPP /* BSDs */
		1223	#endif
1132	#if __solaris	1224	#if __solaris
1133	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1225	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1134	#endif	1226	#endif
1135	)	1227	)
1136	)	1228	)
…		…
1162		1254
1163	offset += cnt;	1255	offset += cnt;
1164	res += cnt;	1256	res += cnt;
1165	count -= cnt;	1257	count -= cnt;
1166	}	1258	}
		1259
		1260	FUBd;
1167	}	1261	}
1168		1262
1169	return res;	1263	return res;
1170	}	1264	}
1171		1265
…		…
1287		1381
1288	/*****************************************************************************/	1382	/*****************************************************************************/
1289	/* requests implemented outside eio_execute, because they are so large */	1383	/* requests implemented outside eio_execute, because they are so large */
1290		1384
1291	static void	1385	static void
1292	eio__realpath (eio_req *req, etp_worker *self)	1386	eio__lseek (eio_req *req)
1293	{	1387	{
1294	char *rel = req->ptr1;	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;
1295	char *res;	1404	char *res;
1296	char *tmp1, *tmp2;	1405	char *tmp1, *tmp2;
1297	#if SYMLOOP_MAX > 32	1406	#if SYMLOOP_MAX > 32
1298	int symlinks = SYMLOOP_MAX;	1407	int symlinks = SYMLOOP_MAX;
1299	#else	1408	#else
1300	int symlinks = 32;	1409	int symlinks = 32;
1301	#endif	1410	#endif
1302		1411
1303	req->result = -1;
1304
1305	errno = EINVAL;	1412	errno = EINVAL;
1306	if (!rel)	1413	if (!rel)
1307	return;	1414	return -1;
1308		1415
1309	errno = ENOENT;	1416	errno = ENOENT;
1310	if (!*rel)	1417	if (!*rel)
1311	return;	1418	return -1;
1312		1419
1313	if (!req->ptr2)	1420	res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1314	{
1315	X_LOCK (wrklock);
1316	req->flags |= EIO_FLAG_PTR2_FREE;
1317	X_UNLOCK (wrklock);
1318	req->ptr2 = malloc (PATH_MAX * 3);
1319
1320	errno = ENOMEM;
1321	if (!req->ptr2)
1322	return;
1323	}
1324
1325	res = req->ptr2;
1326	tmp1 = res + PATH_MAX;	1421	tmp1 = res + PATH_MAX;
1327	tmp2 = tmp1 + PATH_MAX;	1422	tmp2 = tmp1 + PATH_MAX;
1328		1423
1329	#if 0 /* disabled, the musl way to do things is just too racy */	1424	#if 0 /* disabled, the musl way to do things is just too racy */
1330	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)	1425	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
…		…
1349	#endif	1444	#endif
1350	#endif	1445	#endif
1351		1446
1352	if (*rel != '/')	1447	if (*rel != '/')
1353	{	1448	{
		1449	int len;
		1450
		1451	errno = ENOENT;
		1452	if (wd == EIO_INVALID_WD)
		1453	return -1;
		1454
		1455	if (wd == EIO_CWD)
		1456	{
1354	if (!getcwd (res, PATH_MAX))	1457	if (!getcwd (res, PATH_MAX))
1355	return;	1458	return -1;
		1459
		1460	len = strlen (res);
		1461	}
		1462	else
		1463	memcpy (res, wd->str, len = wd->len);
1356		1464
1357	if (res [1]) /* only use if not / */	1465	if (res [1]) /* only use if not / */
1358	res += strlen (res);	1466	res += len;
1359	}	1467	}
1360		1468
1361	while (*rel)	1469	while (*rel)
1362	{	1470	{
1363	eio_ssize_t len, linklen;	1471	eio_ssize_t len, linklen;
1364	char *beg = rel;	1472	const char *beg = rel;
1365		1473
1366	while (*rel && *rel != '/')	1474	while (*rel && *rel != '/')
1367	++rel;	1475	++rel;
1368		1476
1369	len = rel - beg;	1477	len = rel - beg;
…		…
1381		1489
1382	if (beg [1] == '.' && len == 2)	1490	if (beg [1] == '.' && len == 2)
1383	{	1491	{
1384	/* .. - back up one component, if possible */	1492	/* .. - back up one component, if possible */
1385		1493
1386	while (res != req->ptr2)	1494	while (res != tmpbuf->ptr)
1387	if (*--res == '/')	1495	if (*--res == '/')
1388	break;	1496	break;
1389		1497
1390	continue;	1498	continue;
1391	}	1499	}
1392	}	1500	}
1393		1501
1394	errno = ENAMETOOLONG;	1502	errno = ENAMETOOLONG;
1395	if (res + 1 + len + 1 >= tmp1)	1503	if (res + 1 + len + 1 >= tmp1)
1396	return;	1504	return -1;
1397		1505
1398	/* copy one component */	1506	/* copy one component */
1399	*res = '/';	1507	*res = '/';
1400	memcpy (res + 1, beg, len);	1508	memcpy (res + 1, beg, len);
1401		1509
1402	/* zero-terminate, for readlink */	1510	/* zero-terminate, for readlink */
1403	res [len + 1] = 0;	1511	res [len + 1] = 0;
1404		1512
1405	/* now check if it's a symlink */	1513	/* now check if it's a symlink */
1406	linklen = readlink (req->ptr2, tmp1, PATH_MAX);	1514	linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1407		1515
1408	if (linklen < 0)	1516	if (linklen < 0)
1409	{	1517	{
1410	if (errno != EINVAL)	1518	if (errno != EINVAL)
1411	return;	1519	return -1;
1412		1520
1413	/* it's a normal directory. hopefully */	1521	/* it's a normal directory. hopefully */
1414	res += len + 1;	1522	res += len + 1;
1415	}	1523	}
1416	else	1524	else
…		…
1418	/* yay, it was a symlink - build new path in tmp2 */	1526	/* yay, it was a symlink - build new path in tmp2 */
1419	int rellen = strlen (rel);	1527	int rellen = strlen (rel);
1420		1528
1421	errno = ENAMETOOLONG;	1529	errno = ENAMETOOLONG;
1422	if (linklen + 1 + rellen >= PATH_MAX)	1530	if (linklen + 1 + rellen >= PATH_MAX)
1423	return;	1531	return -1;
1424		1532
1425	errno = ELOOP;	1533	errno = ELOOP;
1426	if (!--symlinks)	1534	if (!--symlinks)
1427	return;	1535	return -1;
1428		1536
1429	if (*tmp1 == '/')	1537	if (*tmp1 == '/')
1430	res = req->ptr2; /* symlink resolves to an absolute path */	1538	res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1431		1539
1432	/* we need to be careful, as rel might point into tmp2 already */	1540	/* we need to be careful, as rel might point into tmp2 already */
1433	memmove (tmp2 + linklen + 1, rel, rellen + 1);	1541	memmove (tmp2 + linklen + 1, rel, rellen + 1);
1434	tmp2 [linklen] = '/';	1542	tmp2 [linklen] = '/';
1435	memcpy (tmp2, tmp1, linklen);	1543	memcpy (tmp2, tmp1, linklen);
…		…
1437	rel = tmp2;	1545	rel = tmp2;
1438	}	1546	}
1439	}	1547	}
1440		1548
1441	/* special case for the lone root path */	1549	/* special case for the lone root path */
1442	if (res == req->ptr2)	1550	if (res == tmpbuf->ptr)
1443	*res++ = '/';	1551	*res++ = '/';
1444		1552
1445	req->result = res - (char *)req->ptr2;	1553	return res - (char *)tmpbuf->ptr;
1446
1447	done:
1448	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1449	}	1554	}
1450		1555
1451	static signed char	1556	static signed char
1452	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1557	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1453	{	1558	{
…		…
1461		1566
1462	#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 */
1463	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1568	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1464		1569
1465	static void	1570	static void
1466	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)
1467	{	1572	{
1468	unsigned char bits [9 + sizeof (ino_t) * 8];	1573	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1469	unsigned char *bit = bits;	1574	unsigned char *bit = bits;
1470		1575
1471	assert (CHAR_BIT == 8);	1576	assert (CHAR_BIT == 8);
1472	assert (sizeof (eio_dirent) * 8 < 256);	1577	assert (sizeof (eio_dirent) * 8 < 256);
1473	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1578	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1475		1580
1476	if (size <= EIO_SORT_FAST)	1581	if (size <= EIO_SORT_FAST)
1477	return;	1582	return;
1478		1583
1479	/* 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 */
1480	/* 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 */
1481	/* inode_bits must contain all inodes ORed together */	1586	/* inode_bits must contain all inodes ORed together */
1482	/* 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 */
1483	{	1588	{
1484	ino_t endianness;	1589	eio_ino_t endianness;
1485	int i, j;	1590	int i, j;
1486		1591
1487	/* 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" */
1488	for (i = 0; i < sizeof (ino_t); ++i)	1593	for (i = 0; i < sizeof (eio_ino_t); ++i)
1489	((unsigned char *)&endianness)[i] = i;	1594	((unsigned char *)&endianness)[i] = i;
1490		1595
1491	*bit++ = 0;	1596	*bit++ = 0;
1492		1597
1493	for (i = 0; i < sizeof (ino_t); ++i)	1598	for (i = 0; i < sizeof (eio_ino_t); ++i)
1494	{	1599	{
1495	/* shifting off the byte offsets out of "endianness" */	1600	/* shifting off the byte offsets out of "endianness" */
1496	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1601	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1497	endianness >>= 8;	1602	endianness >>= 8;
1498		1603
1499	for (j = 0; j < 8; ++j)	1604	for (j = 0; j < 8; ++j)
1500	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1605	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1501	*bit++ = offs + j;	1606	*bit++ = offs + j;
1502	}	1607	}
1503		1608
1504	for (j = 0; j < 8; ++j)	1609	for (j = 0; j < 8; ++j)
1505	if (score_bits & (1 << j))	1610	if (score_bits & (1 << j))
1506	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1611	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1507	}	1612	}
1508		1613
1509	/* now actually do the sorting (a variant of MSD radix sort) */	1614	/* now actually do the sorting (a variant of MSD radix sort) */
1510	{	1615	{
1511	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1616	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1512	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1617	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1513	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1618	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1514	int stk_idx = 0;	1619	int stk_idx = 0;
1515		1620
1516	base_stk [stk_idx] = dents;	1621	base_stk [stk_idx] = dents;
1517	end_stk [stk_idx] = dents + size;	1622	end_stk [stk_idx] = dents + size;
1518	bit_stk [stk_idx] = bit - 1;	1623	bit_stk [stk_idx] = bit - 1;
…		…
1597	}	1702	}
1598	}	1703	}
1599	}	1704	}
1600		1705
1601	static void	1706	static void
1602	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)
1603	{	1708	{
1604	if (size <= 1)	1709	if (size <= 1)
1605	return; /* our insertion sort relies on size > 0 */	1710	return; /* our insertion sort relies on size > 0 */
1606		1711
1607	/* 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 */
…		…
1615		1720
1616	/* read a full directory */	1721	/* read a full directory */
1617	static void	1722	static void
1618	eio__scandir (eio_req *req, etp_worker *self)	1723	eio__scandir (eio_req *req, etp_worker *self)
1619	{	1724	{
1620	DIR *dirp;
1621	EIO_STRUCT_DIRENT *entp;
1622	char *name, *names;	1725	char *name, *names;
1623	int namesalloc = 4096;	1726	int namesalloc = 4096 - sizeof (void *) * 4;
1624	int namesoffs = 0;	1727	int namesoffs = 0;
1625	int flags = req->int1;	1728	int flags = req->int1;
1626	eio_dirent *dents = 0;	1729	eio_dirent *dents = 0;
1627	int dentalloc = 128;	1730	int dentalloc = 128;
1628	int dentoffs = 0;	1731	int dentoffs = 0;
1629	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
1630		1740
1631	req->result = -1;	1741	req->result = -1;
1632		1742
1633	if (!(flags & EIO_READDIR_DENTS))	1743	if (!(flags & EIO_READDIR_DENTS))
1634	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1744	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1635		1745
1636	X_LOCK (wrklock);	1746	#ifdef _WIN32
1637	/* 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
1638	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
1639		1814
1640	if (req->flags & EIO_FLAG_PTR1_FREE)	1815	if (req->flags & EIO_FLAG_PTR1_FREE)
1641	free (req->ptr1);	1816	free (req->ptr1);
1642		1817
1643	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1818	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1644	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1819	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1645	req->ptr2 = names = malloc (namesalloc);	1820	req->ptr2 = names = malloc (namesalloc);
1646	X_UNLOCK (wrklock);
1647		1821
1648	if (dirp && names && (!flags || dents))	1822	if (!names || (flags && !dents))
		1823	return;
		1824
1649	for (;;)	1825	for (;;)
1650	{	1826	{
		1827	int done;
		1828
		1829	#ifdef _WIN32
		1830	done = !dirp;
		1831	#else
1651	errno = 0;	1832	errno = 0;
1652	entp = readdir (dirp);	1833	entp = readdir (dirp);
		1834	done = !entp;
		1835	#endif
1653		1836
1654	if (!entp)	1837	if (done)
1655	{	1838	{
		1839	#ifndef _WIN32
		1840	int old_errno = errno;
		1841	closedir (dirp);
		1842	errno = old_errno;
		1843
1656	if (errno)	1844	if (errno)
1657	break;	1845	break;
		1846	#endif
1658		1847
1659	/* sort etc. */	1848	/* sort etc. */
1660	req->int1 = flags;	1849	req->int1 = flags;
1661	req->result = dentoffs;	1850	req->result = dentoffs;
1662		1851
1663	if (flags & EIO_READDIR_STAT_ORDER)	1852	if (flags & EIO_READDIR_STAT_ORDER)
1664	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);
1665	else if (flags & EIO_READDIR_DIRS_FIRST)	1854	else if (flags & EIO_READDIR_DIRS_FIRST)
1666	if (flags & EIO_READDIR_FOUND_UNKNOWN)	1855	if (flags & EIO_READDIR_FOUND_UNKNOWN)
1667	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 */
1668	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))
1669	{	1908	{
1670	/* 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
1671	eio_dirent *oth = dents + dentoffs;	1916	ent = dents + dentoffs;
1672	eio_dirent *dir = dents;
1673		1917
1674	/* 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 */
1675	/* by walking from both sides and swapping if necessary */	1919	ent->namelen = len - 1;
1676	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)
1677	{	1980	{
1678	if (dir->type == EIO_DT_DIR)	1981	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1679	++dir;	1982	ent->score = 1;
1680	else if ((--oth)->type == EIO_DT_DIR)	1983	else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1681	{	1984	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1682	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1683
1684	++dir;
1685	}
1686	}	1985	}
1687		1986	else if (ent->type == EIO_DT_DIR)
1688	/* now sort the dirs only (dirs all have the same score) */	1987	ent->score = 0;
1689	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1690	}	1988	}
1691
1692	break;
1693	}
1694
1695	/* now add the entry to our list(s) */
1696	name = entp->d_name;
1697
1698	/* skip . and .. entries */
1699	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1700	{
1701	int len = D_NAMLEN (entp) + 1;
1702
1703	while (ecb_expect_false (namesoffs + len > namesalloc))
1704	{
1705	namesalloc *= 2;
1706	X_LOCK (wrklock);
1707	req->ptr2 = names = realloc (names, namesalloc);
1708	X_UNLOCK (wrklock);
1709
1710	if (!names)
1711	break;
1712	}	1989	}
1713		1990
1714	memcpy (names + namesoffs, name, len);
1715
1716	if (dents)
1717	{
1718	struct eio_dirent *ent;
1719
1720	if (ecb_expect_false (dentoffs == dentalloc))
1721	{
1722	dentalloc *= 2;
1723	X_LOCK (wrklock);
1724	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1725	X_UNLOCK (wrklock);
1726
1727	if (!dents)
1728	break;
1729	}
1730
1731	ent = dents + dentoffs;
1732
1733	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1734	ent->namelen = len - 1;
1735	ent->inode = D_INO (entp);
1736
1737	inode_bits |= ent->inode;
1738
1739	switch (D_TYPE (entp))
1740	{
1741	default:
1742	ent->type = EIO_DT_UNKNOWN;
1743	flags |= EIO_READDIR_FOUND_UNKNOWN;
1744	break;
1745
1746	#ifdef DT_FIFO
1747	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1748	#endif
1749	#ifdef DT_CHR
1750	case DT_CHR: ent->type = EIO_DT_CHR; break;
1751	#endif
1752	#ifdef DT_MPC
1753	case DT_MPC: ent->type = EIO_DT_MPC; break;
1754	#endif
1755	#ifdef DT_DIR
1756	case DT_DIR: ent->type = EIO_DT_DIR; break;
1757	#endif
1758	#ifdef DT_NAM
1759	case DT_NAM: ent->type = EIO_DT_NAM; break;
1760	#endif
1761	#ifdef DT_BLK
1762	case DT_BLK: ent->type = EIO_DT_BLK; break;
1763	#endif
1764	#ifdef DT_MPB
1765	case DT_MPB: ent->type = EIO_DT_MPB; break;
1766	#endif
1767	#ifdef DT_REG
1768	case DT_REG: ent->type = EIO_DT_REG; break;
1769	#endif
1770	#ifdef DT_NWK
1771	case DT_NWK: ent->type = EIO_DT_NWK; break;
1772	#endif
1773	#ifdef DT_CMP
1774	case DT_CMP: ent->type = EIO_DT_CMP; break;
1775	#endif
1776	#ifdef DT_LNK
1777	case DT_LNK: ent->type = EIO_DT_LNK; break;
1778	#endif
1779	#ifdef DT_SOCK
1780	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1781	#endif
1782	#ifdef DT_DOOR
1783	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1784	#endif
1785	#ifdef DT_WHT
1786	case DT_WHT: ent->type = EIO_DT_WHT; break;
1787	#endif
1788	}
1789
1790	ent->score = 7;
1791
1792	if (flags & EIO_READDIR_DIRS_FIRST)
1793	{
1794	if (ent->type == EIO_DT_UNKNOWN)
1795	{
1796	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1797	ent->score = 1;
1798	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1799	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1800	}
1801	else if (ent->type == EIO_DT_DIR)
1802	ent->score = 0;
1803	}
1804	}
1805
1806	namesoffs += len;	1991	namesoffs += len;
1807	++dentoffs;	1992	++dentoffs;
1808	}	1993	}
1809		1994
1810	if (EIO_CANCELLED (req))	1995	if (EIO_CANCELLED (req))
1811	{	1996	{
1812	errno = ECANCELED;	1997	errno = ECANCELED;
1813	break;	1998	break;
1814	}	1999	}
		2000
		2001	#ifdef _WIN32
		2002	if (!FindNextFile (dirp, &entp))
		2003	{
		2004	FindClose (dirp);
		2005	dirp = 0;
1815	}	2006	}
		2007	#endif
		2008	}
1816	}	2009	}
		2010
		2011	/*****************************************************************************/
		2012	/* working directory stuff */
		2013	/* various deficiencies in the posix 2008 api force us to */
		2014	/* keep the absolute path in string form at all times */
		2015	/* fuck yeah. */
		2016
		2017	#if !HAVE_AT
		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
		2073	return res;
		2074	}
		2075
		2076	eio_wd
		2077	eio_wd_open_sync (eio_wd wd, const char *path)
		2078	{
		2079	struct tmpbuf tmpbuf = { 0 };
		2080	wd = eio__wd_open_sync (&tmpbuf, wd, path);
		2081	free (tmpbuf.ptr);
		2082
		2083	return wd;
		2084	}
		2085
		2086	void
		2087	eio_wd_close_sync (eio_wd wd)
		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 */
		2101
		2102	static int
		2103	eio__truncateat (int dirfd, const char *path, off_t length)
		2104	{
		2105	int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
		2106	int res;
		2107
		2108	if (fd < 0)
		2109	return fd;
		2110
		2111	res = ftruncate (fd, length);
		2112	close (fd);
		2113	return res;
		2114	}
		2115
		2116	static int
		2117	eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
		2118	{
		2119	int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
		2120	int res;
		2121
		2122	if (fd < 0)
		2123	return fd;
		2124
		2125	res = fstatvfs (fd, buf);
		2126	close (fd);
		2127	return res;
		2128
		2129	}
		2130
		2131	#endif
1817		2132
1818	/*****************************************************************************/	2133	/*****************************************************************************/
1819		2134
1820	#define ALLOC(len) \	2135	#define ALLOC(len) \
1821	if (!req->ptr2) \	2136	if (!req->ptr2) \
…		…
1830	req->result = -1; \	2145	req->result = -1; \
1831	break; \	2146	break; \
1832	} \	2147	} \
1833	}	2148	}
1834		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
1835	X_THREAD_PROC (etp_proc)	2167	X_THREAD_PROC (etp_proc)
1836	{	2168	{
1837	ETP_REQ *req;	2169	ETP_REQ *req;
1838	struct timespec ts;	2170	struct timespec ts;
1839	etp_worker *self = (etp_worker *)thr_arg;	2171	etp_worker *self = (etp_worker *)thr_arg;
1840	int timeout;	2172
		2173	etp_proc_init ();
1841		2174
1842	/* try to distribute timeouts somewhat evenly */	2175	/* try to distribute timeouts somewhat evenly */
1843	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	2176	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1844		2177
1845	for (;;)	2178	for (;;)
…		…
1848		2181
1849	X_LOCK (reqlock);	2182	X_LOCK (reqlock);
1850		2183
1851	for (;;)	2184	for (;;)
1852	{	2185	{
1853	self->req = req = reqq_shift (&req_queue);	2186	req = reqq_shift (&req_queue);
1854		2187
1855	if (req)	2188	if (req)
1856	break;	2189	break;
1857		2190
1858	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */	2191	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
…		…
1896	++npending;	2229	++npending;
1897		2230
1898	if (!reqq_push (&res_queue, req) && want_poll_cb)	2231	if (!reqq_push (&res_queue, req) && want_poll_cb)
1899	want_poll_cb ();	2232	want_poll_cb ();
1900		2233
1901	self->req = 0;
1902	etp_worker_clear (self);	2234	etp_worker_clear (self);
1903		2235
1904	X_UNLOCK (reslock);	2236	X_UNLOCK (reslock);
1905	}	2237	}
1906		2238
1907	quit:	2239	quit:
		2240	free (req);
		2241
1908	X_LOCK (wrklock);	2242	X_LOCK (wrklock);
1909	etp_worker_free (self);	2243	etp_worker_free (self);
1910	X_UNLOCK (wrklock);	2244	X_UNLOCK (wrklock);
1911
1912	return 0;
1913	}	2245	}
1914		2246
1915	/*****************************************************************************/	2247	/*****************************************************************************/
1916		2248
1917	int ecb_cold	2249	int ecb_cold
1918	eio_init (void (*want_poll)(void), void (*done_poll)(void))	2250	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1919	{	2251	{
		2252	#if !HAVE_PREADWRITE
		2253	X_MUTEX_CREATE (preadwritelock);
		2254	#endif
		2255
1920	return etp_init (want_poll, done_poll);	2256	return etp_init (want_poll, done_poll);
1921	}	2257	}
1922		2258
1923	ecb_inline void	2259	ecb_inline void
1924	eio_api_destroy (eio_req *req)	2260	eio_api_destroy (eio_req *req)
…		…
1951	}	2287	}
1952		2288
1953	static void	2289	static void
1954	eio_execute (etp_worker *self, eio_req *req)	2290	eio_execute (etp_worker *self, eio_req *req)
1955	{	2291	{
		2292	#if HAVE_AT
		2293	int dirfd;
		2294	#else
		2295	const char *path;
		2296	#endif
		2297
1956	if (ecb_expect_false (EIO_CANCELLED (req)))	2298	if (ecb_expect_false (EIO_CANCELLED (req)))
1957	{	2299	{
1958	req->result = -1;	2300	req->result = -1;
1959	req->errorno = ECANCELED;	2301	req->errorno = ECANCELED;
1960	return;	2302	return;
1961	}	2303	}
1962		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
1963	switch (req->type)	2321	switch (req->type)
1964	{	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;
1965	case EIO_READ: ALLOC (req->size);	2330	case EIO_READ: ALLOC (req->size);
1966	req->result = req->offs >= 0	2331	req->result = req->offs >= 0
1967	? pread (req->int1, req->ptr2, req->size, req->offs)	2332	? pread (req->int1, req->ptr2, req->size, req->offs)
1968	: read (req->int1, req->ptr2, req->size); break;	2333	: read (req->int1, req->ptr2, req->size); break;
1969	case EIO_WRITE: req->result = req->offs >= 0	2334	case EIO_WRITE: req->result = req->offs >= 0
1970	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2335	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1971	: write (req->int1, req->ptr2, req->size); break;	2336	: write (req->int1, req->ptr2, req->size); break;
1972		2337
1973	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;
1974	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
1975		2342
1976	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2343	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1977	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;
1978	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));
1979	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
1980	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2443	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1981	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2444	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1982		2445
1983	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1984	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1985	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));	2446	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1986	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;	2447	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1987		2448
1988	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1989	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;
1990	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1991	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;
1992	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1993	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;	2451	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1994		2452
1995	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1996	case EIO_CLOSE: req->result = close (req->int1); break;	2453	case EIO_CLOSE: req->result = close (req->int1); break;
1997	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;	2454	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1998	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1999	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2000	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2001	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2002	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2003	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2004	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2005
2006	case EIO_REALPATH: eio__realpath (req, self); break;
2007
2008	case EIO_READLINK: ALLOC (PATH_MAX);
2009	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2010
2011	case EIO_SYNC: req->result = 0; sync (); break;	2455	case EIO_SYNC: req->result = 0; sync (); break;
2012	case EIO_FSYNC: req->result = fsync (req->int1); break;	2456	case EIO_FSYNC: req->result = fsync (req->int1); break;
2013	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;
2014	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;
2015	case EIO_MTOUCH: req->result = eio__mtouch (req); break;	2461	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2016	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;
2017	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2463	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2018	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2019	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;	2464	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2020		2465
2021	case EIO_READDIR: eio__scandir (req, self); break;	2466	case EIO_READDIR: eio__scandir (req, self); break;
2022		2467
2023	case EIO_BUSY:	2468	case EIO_BUSY:
…		…
2033	req->result = select (0, 0, 0, 0, &tv);	2478	req->result = select (0, 0, 0, 0, &tv);
2034	}	2479	}
2035	#endif	2480	#endif
2036	break;	2481	break;
2037		2482
2038	case EIO_UTIME:
2039	case EIO_FUTIME:
2040	{
2041	struct timeval tv[2];
2042	struct timeval *times;
2043
2044	if (req->nv1 != -1. || req->nv2 != -1.)
2045	{
2046	tv[0].tv_sec = req->nv1;
2047	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2048	tv[1].tv_sec = req->nv2;
2049	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2050
2051	times = tv;
2052	}
2053	else
2054	times = 0;
2055
2056	req->result = req->type == EIO_FUTIME
2057	? futimes (req->int1, times)
2058	: utimes (req->ptr1, times);
2059	}
2060	break;
2061
2062	case EIO_GROUP:	2483	case EIO_GROUP:
2063	abort (); /* handled in eio_request */	2484	abort (); /* handled in eio_request */
2064		2485
2065	case EIO_NOP:	2486	case EIO_NOP:
2066	req->result = 0;	2487	req->result = 0;
…		…
2069	case EIO_CUSTOM:	2490	case EIO_CUSTOM:
2070	req->feed (req);	2491	req->feed (req);
2071	break;	2492	break;
2072		2493
2073	default:	2494	default:
2074	errno = ENOSYS;
2075	req->result = -1;	2495	req->result = EIO_ENOSYS ();
2076	break;	2496	break;
2077	}	2497	}
2078		2498
2079	req->errorno = errno;	2499	req->errorno = errno;
2080	}	2500	}
2081		2501
2082	#ifndef EIO_NO_WRAPPERS	2502	#ifndef EIO_NO_WRAPPERS
2083		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
2084	eio_req *eio_nop (int pri, eio_cb cb, void *data)	2514	eio_req *eio_nop (int pri, eio_cb cb, void *data)
2085	{	2515	{
2086	REQ (EIO_NOP); SEND;	2516	REQ (EIO_NOP); SEND;
2087	}	2517	}
2088		2518
…		…
2104	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)
2105	{	2535	{
2106	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;
2107	}	2537	}
2108		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
2109	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)
2110	{	2555	{
2111	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;
2112	}	2557	}
2113		2558
…		…
2119	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)
2120	{	2565	{
2121	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;	2566	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2122	}	2567	}
2123		2568
2124	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2125	{
2126	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2127	}
2128
2129	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)	2569	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2130	{	2570	{
2131	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;	2571	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2132	}	2572	}
2133		2573
2134	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2135	{
2136	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2137	}
2138
2139	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)
2140	{	2575	{
2141	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2576	REQ (EIO_CLOSE); req->int1 = fd; SEND;
2142	}	2577	}
2143		2578
2144	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)
2145	{	2580	{
2146	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;
		2582	}
		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;
2147	}	2587	}
2148		2588
2149	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)
2150	{	2590	{
2151	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;
…		…
2369	/* misc garbage */	2809	/* misc garbage */
2370		2810
2371	eio_ssize_t	2811	eio_ssize_t
2372	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)
2373	{	2813	{
2374	etp_worker wrk;
2375	eio_ssize_t ret;
2376
2377	wrk.dbuf = 0;
2378
2379	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2814	return eio__sendfile (ofd, ifd, offset, count);
2380
2381	if (wrk.dbuf)
2382	free (wrk.dbuf);
2383
2384	return ret;
2385	}	2815	}
2386		2816

Diff Legend

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