ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.49 by root, Sat Jan 2 14:24:32 2010 UTC vs.
Revision 1.111 by root, Fri Sep 30 20:49:25 2011 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008,2009,2010 Marc Alexander Lehmann <libeio@schmorp.de>	4	* Copyright (c) 2007,2008,2009,2010,2011 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	*
…		…
35	* and other provisions required by the GPL. If you do not delete the	35	* and other provisions required by the GPL. If you do not delete the
36	* provisions above, a recipient may use your version of this file under	36	* provisions above, a recipient may use your version of this file under
37	* either the BSD or the GPL.	37	* either the BSD or the GPL.
38	*/	38	*/
39		39
		40	#ifndef _WIN32
		41	# include "config.h"
		42	#endif
		43
40	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
41		46
42	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
43	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define X_STACKSIZE EIO_STACKSIZE
44	#endif	49	#endif
45	#include "xthread.h"	50	#include "xthread.h"
46		51
47	#include <errno.h>	52	#include <errno.h>
48	#include <stddef.h>	53	#include <stddef.h>
…		…
53	#include <sys/stat.h>	58	#include <sys/stat.h>
54	#include <limits.h>	59	#include <limits.h>
55	#include <fcntl.h>	60	#include <fcntl.h>
56	#include <assert.h>	61	#include <assert.h>
57		62
		63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
		64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
		65	#if HAVE_STDINT_H
		66	# include <stdint.h>
		67	#endif
		68
		69	#ifndef ECANCELED
		70	# define ECANCELED EDOM
		71	#endif
		72	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
		77	# define ENOTSOCK WSAENOTSOCK
		78	#endif
		79
		80	static void eio_destroy (eio_req *req);
		81
58	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
59	# 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
60	#endif	84	#endif
61		85
62	#ifndef EIO_DESTROY	86	#ifndef EIO_DESTROY
…		…
65		89
66	#ifndef EIO_FEED	90	#ifndef EIO_FEED
67	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	91	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68	#endif	92	#endif
69		93
		94	#ifndef EIO_FD_TO_WIN32_HANDLE
		95	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		96	#endif
		97	#ifndef EIO_WIN32_HANDLE_TO_FD
		98	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		99	#endif
		100
		101	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		102
		103	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		104
70	#ifdef _WIN32	105	#ifdef _WIN32
71		106
72	/*doh*/	107	#undef PAGESIZE
		108	#define PAGESIZE 4096 /* GetSystemInfo? */
		109
		110	/* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
		111
		112	#ifdef EIO_STRUCT_STATI64
		113	/* look at perl's non-sloppy stat */
		114	#define stat(path,buf) _stati64 (path,buf)
		115	#define fstat(fd,buf) _fstati64 (fd,buf)
		116	#endif
		117	#define lstat(path,buf) stat (path,buf)
		118	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		119	#define mkdir(path,mode) _mkdir (path)
		120	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		121
		122	#define chmod(path,mode) _chmod (path, mode)
		123	#define dup(fd) _dup (fd)
		124	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		125
		126	#define fchmod(fd,mode) EIO_ENOSYS ()
		127	#define chown(path,uid,gid) EIO_ENOSYS ()
		128	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		129	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		130	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		131	#define mknod(path,mode,dev) EIO_ENOSYS ()
		132	#define sync() EIO_ENOSYS ()
		133	#define readlink(path,buf,s) EIO_ENOSYS ()
		134	#define statvfs(path,buf) EIO_ENOSYS ()
		135	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		136
		137	/* rename() uses MoveFile, which fails to overwrite */
		138	#define rename(old,neu) eio__rename (old, neu)
		139
		140	static int
		141	eio__rename (const char *old, const char *neu)
		142	{
		143	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		144	return 0;
		145
		146	/* should steal _dosmaperr */
		147	switch (GetLastError ())
		148	{
		149	case ERROR_FILE_NOT_FOUND:
		150	case ERROR_PATH_NOT_FOUND:
		151	case ERROR_INVALID_DRIVE:
		152	case ERROR_NO_MORE_FILES:
		153	case ERROR_BAD_NETPATH:
		154	case ERROR_BAD_NET_NAME:
		155	case ERROR_BAD_PATHNAME:
		156	case ERROR_FILENAME_EXCED_RANGE:
		157	errno = ENOENT;
		158	break;
		159
		160	default:
		161	errno = EACCES;
		162	break;
		163	}
		164
		165	return -1;
		166	}
		167
		168	/* we could even stat and see if it exists */
		169	static int
		170	symlink (const char *old, const char *neu)
		171	{
		172	#if WINVER >= 0x0600
		173	if (CreateSymbolicLink (neu, old, 1))
		174	return 0;
		175
		176	if (CreateSymbolicLink (neu, old, 0))
		177	return 0;
		178	#endif
		179
		180	return EIO_ERRNO (ENOENT, -1);
		181	}
		182
		183	/* POSIX API only */
		184	#define CreateHardLink(neu,old,flags) 0
		185	#define CreateSymbolicLink(neu,old,flags) 0
		186
		187	struct statvfs
		188	{
		189	int dummy;
		190	};
		191
		192	#define DT_DIR EIO_DT_DIR
		193	#define DT_REG EIO_DT_REG
		194	#define D_NAME(entp) entp.cFileName
		195	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
		196
73	#else	197	#else
74		198
75	# include "config.h"
76	# include <sys/time.h>	199	#include <sys/time.h>
77	# include <sys/select.h>	200	#include <sys/select.h>
78	# include <sys/mman.h>	201	#include <sys/statvfs.h>
79	# include <unistd.h>	202	#include <unistd.h>
		203	#include <signal.h>
		204	#include <dirent.h>
		205
		206	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		207	#include <sys/mman.h>
		208	#endif
		209
		210	#define D_NAME(entp) entp->d_name
		211
		212	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		213	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		214	#define _DIRENT_HAVE_D_TYPE /* sigh */
		215	#define D_INO(de) (de)->d_fileno
		216	#define D_NAMLEN(de) (de)->d_namlen
		217	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		218	#define D_INO(de) (de)->d_ino
		219	#endif
		220
		221	#ifdef _D_EXACT_NAMLEN
		222	#undef D_NAMLEN
		223	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		224	#endif
		225
		226	#ifdef _DIRENT_HAVE_D_TYPE
		227	#define D_TYPE(de) (de)->d_type
		228	#endif
		229
		230	#ifndef EIO_STRUCT_DIRENT
		231	#define EIO_STRUCT_DIRENT struct dirent
		232	#endif
		233
		234	#endif
		235
		236	#if HAVE_UTIMES
80	# include <utime.h>	237	# include <utime.h>
81	# include <signal.h>
82	# include <dirent.h>
83
84	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85	# if __freebsd || defined __NetBSD__ || defined __OpenBSD__
86	# define _DIRENT_HAVE_D_TYPE /* sigh */
87	# define D_INO(de) (de)->d_fileno
88	# define D_NAMLEN(de) (de)->d_namlen
89	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
90	# define D_INO(de) (de)->d_ino
91	# endif	238	#endif
92		239
93	#ifdef _D_EXACT_NAMLEN	240	#if HAVE_SYS_SYSCALL_H
94	# undef D_NAMLEN	241	# include <sys/syscall.h>
95	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96	#endif
97
98	# ifdef _DIRENT_HAVE_D_TYPE
99	# define D_TYPE(de) (de)->d_type
100	# endif	242	#endif
101		243
102	# ifndef EIO_STRUCT_DIRENT	244	#if HAVE_SYS_PRCTL_H
103	# define EIO_STRUCT_DIRENT struct dirent	245	# include <sys/prctl.h>
104	# endif
105
106	#endif	246	#endif
107		247
108	#if HAVE_SENDFILE	248	#if HAVE_SENDFILE
109	# if __linux	249	# if __linux
110	# include <sys/sendfile.h>	250	# include <sys/sendfile.h>
111	# elif __freebsd || defined __APPLE__	251	# elif __FreeBSD__ || defined __APPLE__
112	# include <sys/socket.h>	252	# include <sys/socket.h>
113	# include <sys/uio.h>	253	# include <sys/uio.h>
114	# elif __hpux	254	# elif __hpux
115	# include <sys/socket.h>	255	# include <sys/socket.h>
116	# elif __solaris	256	# elif __solaris
…		…
125	#endif	265	#endif
126	#ifndef D_INO	266	#ifndef D_INO
127	# define D_INO(de) 0	267	# define D_INO(de) 0
128	#endif	268	#endif
129	#ifndef D_NAMLEN	269	#ifndef D_NAMLEN
130	# define D_NAMLEN(de) strlen ((de)->d_name)	270	# define D_NAMLEN(entp) strlen (D_NAME (entp))
131	#endif	271	#endif
132
133	/* number of seconds after which an idle threads exit */
134	#define IDLE_TIMEOUT 10
135		272
136	/* used for struct dirent, AIX doesn't provide it */	273	/* used for struct dirent, AIX doesn't provide it */
137	#ifndef NAME_MAX	274	#ifndef NAME_MAX
138	# define NAME_MAX 4096	275	# define NAME_MAX 4096
139	#endif	276	#endif
140		277
		278	/* used for readlink etc. */
		279	#ifndef PATH_MAX
		280	# define PATH_MAX 4096
		281	#endif
		282
141	/* buffer size for various temporary buffers */	283	/* buffer size for various temporary buffers */
142	#define EIO_BUFSIZE 65536	284	#define EIO_BUFSIZE 65536
143		285
144	#define dBUF \	286	#define dBUF \
145	char *eio_buf; \
146	ETP_WORKER_LOCK (self); \
147	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	287	char *eio_buf = malloc (EIO_BUFSIZE); \
148	ETP_WORKER_UNLOCK (self); \
149	errno = ENOMEM; \	288	errno = ENOMEM; \
150	if (!eio_buf) \	289	if (!eio_buf) \
151	return -1;	290	return -1
		291
		292	#define FUBd \
		293	free (eio_buf)
152		294
153	#define EIO_TICKS ((1000000 + 1023) >> 10)	295	#define EIO_TICKS ((1000000 + 1023) >> 10)
154		296
155	/*****************************************************************************/	297	/*****************************************************************************/
156		298
157	#if __GNUC__ >= 3	299	struct tmpbuf
158	# define expect(expr,value) __builtin_expect ((expr),(value))	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
159	#else	325	#else
160	# define expect(expr,value) (expr)	326	#define HAVE_AT 0
		327	static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
161	#endif	328	#endif
162		329
163	#define expect_false(expr) expect ((expr) != 0, 0)	330	struct eio_pwd
164	#define expect_true(expr) expect ((expr) != 0, 1)	331	{
		332	#if HAVE_AT
		333	int fd;
		334	#endif
		335	int len;
		336	char str[1]; /* actually, a 0-terminated canonical path */
		337	};
165		338
166	/*****************************************************************************/	339	/*****************************************************************************/
167		340
168	#define ETP_PRI_MIN EIO_PRI_MIN	341	#define ETP_PRI_MIN EIO_PRI_MIN
169	#define ETP_PRI_MAX EIO_PRI_MAX	342	#define ETP_PRI_MAX EIO_PRI_MAX
…		…
175	static int eio_finish (eio_req *req);	348	static int eio_finish (eio_req *req);
176	#define ETP_FINISH(req) eio_finish (req)	349	#define ETP_FINISH(req) eio_finish (req)
177	static void eio_execute (struct etp_worker *self, eio_req *req);	350	static void eio_execute (struct etp_worker *self, eio_req *req);
178	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	351	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
179		352
180	#define ETP_WORKER_CLEAR(req) \
181	if (wrk->dbuf) \
182	{ \
183	free (wrk->dbuf); \
184	wrk->dbuf = 0; \
185	} \
186	\
187	if (wrk->dirp) \
188	{ \
189	closedir (wrk->dirp); \
190	wrk->dirp = 0; \
191	}
192
193	#define ETP_WORKER_COMMON \
194	void *dbuf; \
195	DIR *dirp;
196
197	/*****************************************************************************/	353	/*****************************************************************************/
198		354
199	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	355	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200		356
201	/* calculate time difference in ~1/EIO_TICKS of a second */	357	/* calculate time difference in ~1/EIO_TICKS of a second */
		358	ecb_inline int
202	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	359	tvdiff (struct timeval *tv1, struct timeval *tv2)
203	{	360	{
204	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	361	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	362	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
206	}	363	}
207		364
…		…
211	static void (*done_poll_cb) (void);	368	static void (*done_poll_cb) (void);
212		369
213	static unsigned int max_poll_time; /* reslock */	370	static unsigned int max_poll_time; /* reslock */
214	static unsigned int max_poll_reqs; /* reslock */	371	static unsigned int max_poll_reqs; /* reslock */
215		372
216	static volatile unsigned int nreqs; /* reqlock */	373	static unsigned int nreqs; /* reqlock */
217	static volatile unsigned int nready; /* reqlock */	374	static unsigned int nready; /* reqlock */
218	static volatile unsigned int npending; /* reqlock */	375	static unsigned int npending; /* reqlock */
219	static volatile unsigned int max_idle = 4;	376	static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		377	static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
220		378
221	static mutex_t wrklock = X_MUTEX_INIT;	379	static xmutex_t wrklock;
222	static mutex_t reslock = X_MUTEX_INIT;	380	static xmutex_t reslock;
223	static mutex_t reqlock = X_MUTEX_INIT;	381	static xmutex_t reqlock;
224	static cond_t reqwait = X_COND_INIT;	382	static xcond_t reqwait;
225		383
226	#if !HAVE_PREADWRITE	384	#if !HAVE_PREADWRITE
227	/*	385	/*
228	* make our pread/pwrite emulation safe against themselves, but not against	386	* make our pread/pwrite emulation safe against themselves, but not against
229	* 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,
230	* but that's your problem, not mine.	388	* but that's your problem, not mine.
231	*/	389	*/
232	static mutex_t preadwritelock = X_MUTEX_INIT;	390	static xmutex_t preadwritelock;
233	#endif	391	#endif
234		392
235	typedef struct etp_worker	393	typedef struct etp_worker
236	{	394	{
		395	struct tmpbuf tmpbuf;
		396
237	/* locked by wrklock */	397	/* locked by wrklock */
238	struct etp_worker *prev, *next;	398	struct etp_worker *prev, *next;
239		399
240	thread_t tid;	400	xthread_t tid;
241		401
242	/* locked by reslock, reqlock or wrklock */	402	#ifdef ETP_WORKER_COMMON
243	ETP_REQ *req; /* currently processed request */
244
245	ETP_WORKER_COMMON	403	ETP_WORKER_COMMON
		404	#endif
246	} etp_worker;	405	} etp_worker;
247		406
248	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	407	static etp_worker wrk_first; /* NOT etp */
249		408
250	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	409	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
251	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	410	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
252		411
253	/* worker threads management */	412	/* worker threads management */
254		413
		414	static void
255	static void etp_worker_clear (etp_worker *wrk)	415	etp_worker_clear (etp_worker *wrk)
256	{	416	{
257	ETP_WORKER_CLEAR (wrk);
258	}	417	}
259		418
		419	static void ecb_cold
260	static void etp_worker_free (etp_worker *wrk)	420	etp_worker_free (etp_worker *wrk)
261	{	421	{
		422	free (wrk->tmpbuf.ptr);
		423
262	wrk->next->prev = wrk->prev;	424	wrk->next->prev = wrk->prev;
263	wrk->prev->next = wrk->next;	425	wrk->prev->next = wrk->next;
264		426
265	free (wrk);	427	free (wrk);
266	}	428	}
267		429
268	static unsigned int etp_nreqs (void)	430	static unsigned int
		431	etp_nreqs (void)
269	{	432	{
270	int retval;	433	int retval;
271	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	434	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
272	retval = nreqs;	435	retval = nreqs;
273	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	436	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
274	return retval;	437	return retval;
275	}	438	}
276		439
277	static unsigned int etp_nready (void)	440	static unsigned int
		441	etp_nready (void)
278	{	442	{
279	unsigned int retval;	443	unsigned int retval;
280		444
281	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	445	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282	retval = nready;	446	retval = nready;
283	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	447	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
284		448
285	return retval;	449	return retval;
286	}	450	}
287		451
288	static unsigned int etp_npending (void)	452	static unsigned int
		453	etp_npending (void)
289	{	454	{
290	unsigned int retval;	455	unsigned int retval;
291		456
292	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	457	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
293	retval = npending;	458	retval = npending;
294	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	459	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
295		460
296	return retval;	461	return retval;
297	}	462	}
298		463
299	static unsigned int etp_nthreads (void)	464	static unsigned int
		465	etp_nthreads (void)
300	{	466	{
301	unsigned int retval;	467	unsigned int retval;
302		468
303	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	469	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
304	retval = started;	470	retval = started;
…		…
318	} etp_reqq;	484	} etp_reqq;
319		485
320	static etp_reqq req_queue;	486	static etp_reqq req_queue;
321	static etp_reqq res_queue;	487	static etp_reqq res_queue;
322		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
		500	static int ecb_noinline
323	static int reqq_push (etp_reqq *q, ETP_REQ *req)	501	reqq_push (etp_reqq *q, ETP_REQ *req)
324	{	502	{
325	int pri = req->pri;	503	int pri = req->pri;
326	req->next = 0;	504	req->next = 0;
327		505
328	if (q->qe[pri])	506	if (q->qe[pri])
…		…
334	q->qe[pri] = q->qs[pri] = req;	512	q->qe[pri] = q->qs[pri] = req;
335		513
336	return q->size++;	514	return q->size++;
337	}	515	}
338		516
		517	static ETP_REQ * ecb_noinline
339	static ETP_REQ *reqq_shift (etp_reqq *q)	518	reqq_shift (etp_reqq *q)
340	{	519	{
341	int pri;	520	int pri;
342		521
343	if (!q->size)	522	if (!q->size)
344	return 0;	523	return 0;
…		…
359	}	538	}
360		539
361	abort ();	540	abort ();
362	}	541	}
363		542
364	static void etp_atfork_prepare (void)	543	static int ecb_cold
		544	etp_init (void (*want_poll)(void), void (*done_poll)(void))
365	{	545	{
366	X_LOCK (wrklock);	546	X_MUTEX_CREATE (wrklock);
367	X_LOCK (reqlock);	547	X_MUTEX_CREATE (reslock);
368	X_LOCK (reslock);	548	X_MUTEX_CREATE (reqlock);
369	#if !HAVE_PREADWRITE	549	X_COND_CREATE (reqwait);
370	X_LOCK (preadwritelock);
371	#endif
372	}
373		550
374	static void etp_atfork_parent (void)	551	reqq_init (&req_queue);
375	{	552	reqq_init (&res_queue);
376	#if !HAVE_PREADWRITE
377	X_UNLOCK (preadwritelock);
378	#endif
379	X_UNLOCK (reslock);
380	X_UNLOCK (reqlock);
381	X_UNLOCK (wrklock);
382	}
383		553
384	static void etp_atfork_child (void)	554	wrk_first.next =
385	{	555	wrk_first.prev = &wrk_first;
386	ETP_REQ *prv;
387
388	while ((prv = reqq_shift (&req_queue)))
389	ETP_DESTROY (prv);
390
391	while ((prv = reqq_shift (&res_queue)))
392	ETP_DESTROY (prv);
393
394	while (wrk_first.next != &wrk_first)
395	{
396	etp_worker *wrk = wrk_first.next;
397
398	if (wrk->req)
399	ETP_DESTROY (wrk->req);
400
401	etp_worker_clear (wrk);
402	etp_worker_free (wrk);
403	}
404		556
405	started = 0;	557	started = 0;
406	idle = 0;	558	idle = 0;
407	nreqs = 0;	559	nreqs = 0;
408	nready = 0;	560	nready = 0;
409	npending = 0;	561	npending = 0;
410		562
411	etp_atfork_parent ();
412	}
413
414	static void
415	etp_once_init (void)
416	{
417	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
418	}
419
420	static int
421	etp_init (void (*want_poll)(void), void (*done_poll)(void))
422	{
423	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
424
425	pthread_once (&doinit, etp_once_init);
426
427	want_poll_cb = want_poll;	563	want_poll_cb = want_poll;
428	done_poll_cb = done_poll;	564	done_poll_cb = done_poll;
429		565
430	return 0;	566	return 0;
431	}	567	}
432		568
433	X_THREAD_PROC (etp_proc);	569	X_THREAD_PROC (etp_proc);
434		570
		571	static void ecb_cold
435	static void etp_start_thread (void)	572	etp_start_thread (void)
436	{	573	{
437	etp_worker *wrk = calloc (1, sizeof (etp_worker));	574	etp_worker *wrk = calloc (1, sizeof (etp_worker));
438		575
439	/*TODO*/	576	/*TODO*/
440	assert (("unable to allocate worker thread data", wrk));	577	assert (("unable to allocate worker thread data", wrk));
…		…
453	free (wrk);	590	free (wrk);
454		591
455	X_UNLOCK (wrklock);	592	X_UNLOCK (wrklock);
456	}	593	}
457		594
		595	static void
458	static void etp_maybe_start_thread (void)	596	etp_maybe_start_thread (void)
459	{	597	{
460	if (expect_true (etp_nthreads () >= wanted))	598	if (ecb_expect_true (etp_nthreads () >= wanted))
461	return;	599	return;
462		600
463	/* todo: maybe use idle here, but might be less exact */	601	/* todo: maybe use idle here, but might be less exact */
464	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))	602	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
465	return;	603	return;
466		604
467	etp_start_thread ();	605	etp_start_thread ();
468	}	606	}
469		607
		608	static void ecb_cold
470	static void etp_end_thread (void)	609	etp_end_thread (void)
471	{	610	{
472	eio_req *req = calloc (1, sizeof (eio_req));	611	eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
473		612
474	req->type = -1;	613	req->type = -1;
475	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	614	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
476		615
477	X_LOCK (reqlock);	616	X_LOCK (reqlock);
…		…
482	X_LOCK (wrklock);	621	X_LOCK (wrklock);
483	--started;	622	--started;
484	X_UNLOCK (wrklock);	623	X_UNLOCK (wrklock);
485	}	624	}
486		625
487	static int etp_poll (void)	626	static int
		627	etp_poll (void)
488	{	628	{
489	unsigned int maxreqs;	629	unsigned int maxreqs;
490	unsigned int maxtime;	630	unsigned int maxtime;
491	struct timeval tv_start, tv_now;	631	struct timeval tv_start, tv_now;
492		632
…		…
522		662
523	X_LOCK (reqlock);	663	X_LOCK (reqlock);
524	--nreqs;	664	--nreqs;
525	X_UNLOCK (reqlock);	665	X_UNLOCK (reqlock);
526		666
527	if (expect_false (req->type == EIO_GROUP && req->size))	667	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
528	{	668	{
529	req->int1 = 1; /* mark request as delayed */	669	req->int1 = 1; /* mark request as delayed */
530	continue;	670	continue;
531	}	671	}
532	else	672	else
533	{	673	{
534	int res = ETP_FINISH (req);	674	int res = ETP_FINISH (req);
535	if (expect_false (res))	675	if (ecb_expect_false (res))
536	return res;	676	return res;
537	}	677	}
538		678
539	if (expect_false (maxreqs && !--maxreqs))	679	if (ecb_expect_false (maxreqs && !--maxreqs))
540	break;	680	break;
541		681
542	if (maxtime)	682	if (maxtime)
543	{	683	{
544	gettimeofday (&tv_now, 0);	684	gettimeofday (&tv_now, 0);
…		…
550		690
551	errno = EAGAIN;	691	errno = EAGAIN;
552	return -1;	692	return -1;
553	}	693	}
554		694
		695	static void
555	static void etp_cancel (ETP_REQ *req)	696	etp_cancel (ETP_REQ *req)
556	{	697	{
557	X_LOCK (wrklock);	698	req->cancelled = 1;
558	req->flags |= EIO_FLAG_CANCELLED;
559	X_UNLOCK (wrklock);
560		699
561	eio_grp_cancel (req);	700	eio_grp_cancel (req);
562	}	701	}
563		702
		703	static void
564	static void etp_submit (ETP_REQ *req)	704	etp_submit (ETP_REQ *req)
565	{	705	{
566	req->pri -= ETP_PRI_MIN;	706	req->pri -= ETP_PRI_MIN;
567		707
568	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	708	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
569	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	709	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
570		710
571	if (expect_false (req->type == EIO_GROUP))	711	if (ecb_expect_false (req->type == EIO_GROUP))
572	{	712	{
573	/* I hope this is worth it :/ */	713	/* I hope this is worth it :/ */
574	X_LOCK (reqlock);	714	X_LOCK (reqlock);
575	++nreqs;	715	++nreqs;
576	X_UNLOCK (reqlock);	716	X_UNLOCK (reqlock);
…		…
595		735
596	etp_maybe_start_thread ();	736	etp_maybe_start_thread ();
597	}	737	}
598	}	738	}
599		739
		740	static void ecb_cold
600	static void etp_set_max_poll_time (double nseconds)	741	etp_set_max_poll_time (double nseconds)
601	{	742	{
602	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	743	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603	max_poll_time = nseconds * EIO_TICKS;	744	max_poll_time = nseconds * EIO_TICKS;
604	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	745	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605	}	746	}
606		747
		748	static void ecb_cold
607	static void etp_set_max_poll_reqs (unsigned int maxreqs)	749	etp_set_max_poll_reqs (unsigned int maxreqs)
608	{	750	{
609	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	751	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
610	max_poll_reqs = maxreqs;	752	max_poll_reqs = maxreqs;
611	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	753	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
612	}	754	}
613		755
		756	static void ecb_cold
614	static void etp_set_max_idle (unsigned int nthreads)	757	etp_set_max_idle (unsigned int nthreads)
615	{	758	{
616	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	759	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617	max_idle = nthreads <= 0 ? 1 : nthreads;	760	max_idle = nthreads;
618	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	761	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619	}	762	}
620		763
		764	static void ecb_cold
		765	etp_set_idle_timeout (unsigned int seconds)
		766	{
		767	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		768	idle_timeout = seconds;
		769	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		770	}
		771
		772	static void ecb_cold
621	static void etp_set_min_parallel (unsigned int nthreads)	773	etp_set_min_parallel (unsigned int nthreads)
622	{	774	{
623	if (wanted < nthreads)	775	if (wanted < nthreads)
624	wanted = nthreads;	776	wanted = nthreads;
625	}	777	}
626		778
		779	static void ecb_cold
627	static void etp_set_max_parallel (unsigned int nthreads)	780	etp_set_max_parallel (unsigned int nthreads)
628	{	781	{
629	if (wanted > nthreads)	782	if (wanted > nthreads)
630	wanted = nthreads;	783	wanted = nthreads;
631		784
632	while (started > wanted)	785	while (started > wanted)
633	etp_end_thread ();	786	etp_end_thread ();
634	}	787	}
635		788
636	/*****************************************************************************/	789	/*****************************************************************************/
637		790
		791	static void
638	static void grp_try_feed (eio_req *grp)	792	grp_try_feed (eio_req *grp)
639	{	793	{
640	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	794	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
641	{	795	{
642	grp->flags &= ~EIO_FLAG_GROUPADD;	796	grp->flags &= ~EIO_FLAG_GROUPADD;
643		797
…		…
650	break;	804	break;
651	}	805	}
652	}	806	}
653	}	807	}
654		808
		809	static int
655	static int grp_dec (eio_req *grp)	810	grp_dec (eio_req *grp)
656	{	811	{
657	--grp->size;	812	--grp->size;
658		813
659	/* call feeder, if applicable */	814	/* call feeder, if applicable */
660	grp_try_feed (grp);	815	grp_try_feed (grp);
…		…
664	return eio_finish (grp);	819	return eio_finish (grp);
665	else	820	else
666	return 0;	821	return 0;
667	}	822	}
668		823
		824	static void
669	void eio_destroy (eio_req *req)	825	eio_destroy (eio_req *req)
670	{	826	{
671	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	827	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
672	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	828	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
673		829
674	EIO_DESTROY (req);	830	EIO_DESTROY (req);
675	}	831	}
676		832
		833	static int
677	static int eio_finish (eio_req *req)	834	eio_finish (eio_req *req)
678	{	835	{
679	int res = EIO_FINISH (req);	836	int res = EIO_FINISH (req);
680		837
681	if (req->grp)	838	if (req->grp)
682	{	839	{
…		…
690	if (grp->grp_first == req)	847	if (grp->grp_first == req)
691	grp->grp_first = req->grp_next;	848	grp->grp_first = req->grp_next;
692		849
693	res2 = grp_dec (grp);	850	res2 = grp_dec (grp);
694		851
695	if (!res && res2)	852	if (!res)
696	res = res2;	853	res = res2;
697	}	854	}
698		855
699	eio_destroy (req);	856	eio_destroy (req);
700		857
701	return res;	858	return res;
702	}	859	}
703		860
		861	void
704	void eio_grp_cancel (eio_req *grp)	862	eio_grp_cancel (eio_req *grp)
705	{	863	{
706	for (grp = grp->grp_first; grp; grp = grp->grp_next)	864	for (grp = grp->grp_first; grp; grp = grp->grp_next)
707	eio_cancel (grp);	865	eio_cancel (grp);
708	}	866	}
709		867
		868	void
710	void eio_cancel (eio_req *req)	869	eio_cancel (eio_req *req)
711	{	870	{
712	etp_cancel (req);	871	etp_cancel (req);
713	}	872	}
714		873
		874	void
715	void eio_submit (eio_req *req)	875	eio_submit (eio_req *req)
716	{	876	{
717	etp_submit (req);	877	etp_submit (req);
718	}	878	}
719		879
720	unsigned int eio_nreqs (void)	880	unsigned int
		881	eio_nreqs (void)
721	{	882	{
722	return etp_nreqs ();	883	return etp_nreqs ();
723	}	884	}
724		885
725	unsigned int eio_nready (void)	886	unsigned int
		887	eio_nready (void)
726	{	888	{
727	return etp_nready ();	889	return etp_nready ();
728	}	890	}
729		891
730	unsigned int eio_npending (void)	892	unsigned int
		893	eio_npending (void)
731	{	894	{
732	return etp_npending ();	895	return etp_npending ();
733	}	896	}
734		897
735	unsigned int eio_nthreads (void)	898	unsigned int ecb_cold
		899	eio_nthreads (void)
736	{	900	{
737	return etp_nthreads ();	901	return etp_nthreads ();
738	}	902	}
739		903
		904	void ecb_cold
740	void eio_set_max_poll_time (double nseconds)	905	eio_set_max_poll_time (double nseconds)
741	{	906	{
742	etp_set_max_poll_time (nseconds);	907	etp_set_max_poll_time (nseconds);
743	}	908	}
744		909
		910	void ecb_cold
745	void eio_set_max_poll_reqs (unsigned int maxreqs)	911	eio_set_max_poll_reqs (unsigned int maxreqs)
746	{	912	{
747	etp_set_max_poll_reqs (maxreqs);	913	etp_set_max_poll_reqs (maxreqs);
748	}	914	}
749		915
		916	void ecb_cold
750	void eio_set_max_idle (unsigned int nthreads)	917	eio_set_max_idle (unsigned int nthreads)
751	{	918	{
752	etp_set_max_idle (nthreads);	919	etp_set_max_idle (nthreads);
753	}	920	}
754		921
		922	void ecb_cold
		923	eio_set_idle_timeout (unsigned int seconds)
		924	{
		925	etp_set_idle_timeout (seconds);
		926	}
		927
		928	void ecb_cold
755	void eio_set_min_parallel (unsigned int nthreads)	929	eio_set_min_parallel (unsigned int nthreads)
756	{	930	{
757	etp_set_min_parallel (nthreads);	931	etp_set_min_parallel (nthreads);
758	}	932	}
759		933
		934	void ecb_cold
760	void eio_set_max_parallel (unsigned int nthreads)	935	eio_set_max_parallel (unsigned int nthreads)
761	{	936	{
762	etp_set_max_parallel (nthreads);	937	etp_set_max_parallel (nthreads);
763	}	938	}
764		939
765	int eio_poll (void)	940	int eio_poll (void)
…		…
774	# undef pread	949	# undef pread
775	# undef pwrite	950	# undef pwrite
776	# define pread eio__pread	951	# define pread eio__pread
777	# define pwrite eio__pwrite	952	# define pwrite eio__pwrite
778		953
779	static ssize_t	954	static eio_ssize_t
780	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)
781	{	956	{
782	ssize_t res;	957	eio_ssize_t res;
783	off_t ooffset;	958	off_t ooffset;
784		959
785	X_LOCK (preadwritelock);	960	X_LOCK (preadwritelock);
786	ooffset = lseek (fd, 0, SEEK_CUR);	961	ooffset = lseek (fd, 0, SEEK_CUR);
787	lseek (fd, offset, SEEK_SET);	962	lseek (fd, offset, SEEK_SET);
…		…
790	X_UNLOCK (preadwritelock);	965	X_UNLOCK (preadwritelock);
791		966
792	return res;	967	return res;
793	}	968	}
794		969
795	static ssize_t	970	static eio_ssize_t
796	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)
797	{	972	{
798	ssize_t res;	973	eio_ssize_t res;
799	off_t ooffset;	974	off_t ooffset;
800		975
801	X_LOCK (preadwritelock);	976	X_LOCK (preadwritelock);
802	ooffset = lseek (fd, 0, SEEK_CUR);	977	ooffset = lseek (fd, 0, SEEK_CUR);
803	lseek (fd, offset, SEEK_SET);	978	lseek (fd, offset, SEEK_SET);
…		…
807		982
808	return res;	983	return res;
809	}	984	}
810	#endif	985	#endif
811		986
812	#ifndef HAVE_FUTIMES	987	#ifndef HAVE_UTIMES
813		988
814	# undef utimes	989	# undef utimes
815	# undef futimes
816	# define utimes(path,times) eio__utimes (path, times)	990	# define utimes(path,times) eio__utimes (path, times)
817	# define futimes(fd,times) eio__futimes (fd, times)
818		991
819	static int	992	static int
820	eio__utimes (const char *filename, const struct timeval times[2])	993	eio__utimes (const char *filename, const struct timeval times[2])
821	{	994	{
822	if (times)	995	if (times)
…		…
830	}	1003	}
831	else	1004	else
832	return utime (filename, 0);	1005	return utime (filename, 0);
833	}	1006	}
834		1007
		1008	#endif
		1009
		1010	#ifndef HAVE_FUTIMES
		1011
		1012	# undef futimes
		1013	# define futimes(fd,times) eio__futimes (fd, times)
		1014
		1015	static int
835	static int eio__futimes (int fd, const struct timeval tv[2])	1016	eio__futimes (int fd, const struct timeval tv[2])
836	{	1017	{
837	errno = ENOSYS;	1018	errno = ENOSYS;
838	return -1;	1019	return -1;
839	}	1020	}
840		1021
…		…
843	#if !HAVE_FDATASYNC	1024	#if !HAVE_FDATASYNC
844	# undef fdatasync	1025	# undef fdatasync
845	# define fdatasync(fd) fsync (fd)	1026	# define fdatasync(fd) fsync (fd)
846	#endif	1027	#endif
847		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	}
		1046
848	/* sync_file_range always needs emulation */	1047	/* sync_file_range always needs emulation */
849	int	1048	static int
850	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)
851	{	1050	{
852	#if HAVE_SYNC_FILE_RANGE	1051	#if HAVE_SYNC_FILE_RANGE
853	int res;	1052	int res;
854		1053
…		…
871	/* even though we could play tricks with the flags, it's better to always	1070	/* even though we could play tricks with the flags, it's better to always
872	* call fdatasync, as that matches the expectation of its users best */	1071	* call fdatasync, as that matches the expectation of its users best */
873	return fdatasync (fd);	1072	return fdatasync (fd);
874	}	1073	}
875		1074
		1075	static int
		1076	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		1077	{
		1078	#if HAVE_FALLOCATE
		1079	return fallocate (fd, mode, offset, len);
		1080	#else
		1081	errno = ENOSYS;
		1082	return -1;
		1083	#endif
		1084	}
		1085
876	#if !HAVE_READAHEAD	1086	#if !HAVE_READAHEAD
877	# undef readahead	1087	# undef readahead
878	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1088	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
879		1089
880	static ssize_t	1090	static eio_ssize_t
881	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)
882	{	1092	{
883	size_t todo = count;	1093	size_t todo = count;
884	dBUF;	1094	dBUF;
885		1095
…		…
890	pread (fd, eio_buf, len, offset);	1100	pread (fd, eio_buf, len, offset);
891	offset += len;	1101	offset += len;
892	todo -= len;	1102	todo -= len;
893	}	1103	}
894		1104
		1105	FUBd;
		1106
895	errno = 0;	1107	errno = 0;
896	return count;	1108	return count;
897	}	1109	}
898		1110
899	#endif	1111	#endif
900		1112
901	/* sendfile always needs emulation */	1113	/* sendfile always needs emulation */
902	static ssize_t	1114	static eio_ssize_t
903	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1115	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
904	{	1116	{
		1117	eio_ssize_t written = 0;
905	ssize_t res;	1118	eio_ssize_t res;
906		1119
907	if (!count)	1120	if (!count)
908	return 0;	1121	return 0;
909		1122
		1123	for (;;)
		1124	{
		1125	#ifdef __APPLE__
		1126	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1127	#endif
910	#if HAVE_SENDFILE	1128	#if HAVE_SENDFILE
911	# if __linux	1129	# if __linux
		1130	off_t soffset = offset;
912	res = sendfile (ofd, ifd, &offset, count);	1131	res = sendfile (ofd, ifd, &soffset, count);
913		1132
914	# elif __freebsd	1133	# elif __FreeBSD__
915	/*	1134	/*
916	* Of course, the freebsd sendfile is a dire hack with no thoughts	1135	* Of course, the freebsd sendfile is a dire hack with no thoughts
917	* wasted on making it similar to other I/O functions.	1136	* wasted on making it similar to other I/O functions.
918	*/	1137	*/
919	{
920	off_t sbytes;	1138	off_t sbytes;
921	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1139	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922		1140
923	if (res < 0 && sbytes)	1141	#if 0 /* according to the manpage, this is correct, but broken behaviour */
924	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1142	/* freebsd' sendfile will return 0 on success */
		1143	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1144	/* not on e.g. EIO or EPIPE - sounds broken */
		1145	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
925	res = sbytes;	1146	res = sbytes;
926	}	1147	#endif
		1148
		1149	/* according to source inspection, this is correct, and useful behaviour */
		1150	if (sbytes)
		1151	res = sbytes;
927		1152
928	# elif defined (__APPLE__)	1153	# elif defined (__APPLE__)
929
930	{
931	off_t sbytes = count;	1154	off_t sbytes = count;
932	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	1155	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
933		1156
934	if (res < 0 && errno == EAGAIN && sbytes)	1157	/* according to the manpage, sbytes is always valid */
		1158	if (sbytes)
935	res = sbytes;	1159	res = sbytes;
936	}
937		1160
938	# elif __hpux	1161	# elif __hpux
939	res = sendfile (ofd, ifd, offset, count, 0, 0);	1162	res = sendfile (ofd, ifd, offset, count, 0, 0);
940		1163
941	# elif __solaris	1164	# elif __solaris
942	{
943	struct sendfilevec vec;	1165	struct sendfilevec vec;
944	size_t sbytes;	1166	size_t sbytes;
945		1167
946	vec.sfv_fd = ifd;	1168	vec.sfv_fd = ifd;
947	vec.sfv_flag = 0;	1169	vec.sfv_flag = 0;
948	vec.sfv_off = offset;	1170	vec.sfv_off = offset;
949	vec.sfv_len = count;	1171	vec.sfv_len = count;
950		1172
951	res = sendfilev (ofd, &vec, 1, &sbytes);	1173	res = sendfilev (ofd, &vec, 1, &sbytes);
952		1174
953	if (res < 0 && sbytes)	1175	if (res < 0 && sbytes)
954	res = sbytes;	1176	res = sbytes;
955	}
956		1177
957	# endif	1178	# endif
958		1179
959	#elif defined (_WIN32)	1180	#elif defined (_WIN32) && 0
960
961	/* does not work, just for documentation of what would need to be done */	1181	/* does not work, just for documentation of what would need to be done */
962	{	1182	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1183	/* libeio guarantees that the file offset does not change, and windows */
		1184	/* has no way to get an independent handle to the same file description */
963	HANDLE h = TO_SOCKET (ifd);	1185	HANDLE h = TO_SOCKET (ifd);
964	SetFilePointer (h, offset, 0, FILE_BEGIN);	1186	SetFilePointer (h, offset, 0, FILE_BEGIN);
965	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);	1187	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
966	}
967		1188
968	#else	1189	#else
969	res = -1;	1190	res = -1;
970	errno = ENOSYS;	1191	errno = ENOSYS;
971	#endif	1192	#endif
972		1193
		1194	/* we assume sendfile can copy at least 128mb in one go */
		1195	if (res <= 128 * 1024 * 1024)
		1196	{
		1197	if (res > 0)
		1198	written += res;
		1199
		1200	if (written)
		1201	return written;
		1202
		1203	break;
		1204	}
		1205	else
		1206	{
		1207	/* if we requested more, then probably the kernel was lazy */
		1208	written += res;
		1209	offset += res;
		1210	count -= res;
		1211
		1212	if (!count)
		1213	return written;
		1214	}
		1215	}
		1216
973	if (res < 0	1217	if (res < 0
974	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1218	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
975	/* BSDs */	1219	/* BSDs */
976	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1220	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
977	|| errno == ENOTSUP	1221	|| errno == ENOTSUP
978	#endif	1222	#endif
		1223	#ifdef EOPNOTSUPP /* windows */
979	|| errno == EOPNOTSUPP /* BSDs */	1224	|| errno == EOPNOTSUPP /* BSDs */
		1225	#endif
980	#if __solaris	1226	#if __solaris
981	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1227	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
982	#endif	1228	#endif
983	)	1229	)
984	)	1230	)
…		…
988		1234
989	res = 0;	1235	res = 0;
990		1236
991	while (count)	1237	while (count)
992	{	1238	{
993	ssize_t cnt;	1239	eio_ssize_t cnt;
994		1240
995	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1241	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
996		1242
997	if (cnt <= 0)	1243	if (cnt <= 0)
998	{	1244	{
…		…
1010		1256
1011	offset += cnt;	1257	offset += cnt;
1012	res += cnt;	1258	res += cnt;
1013	count -= cnt;	1259	count -= cnt;
1014	}	1260	}
		1261
		1262	FUBd;
1015	}	1263	}
1016		1264
1017	return res;	1265	return res;
		1266	}
		1267
		1268	#ifdef PAGESIZE
		1269	# define eio_pagesize() PAGESIZE
		1270	#else
		1271	static intptr_t
		1272	eio_pagesize (void)
		1273	{
		1274	static intptr_t page;
		1275
		1276	if (!page)
		1277	page = sysconf (_SC_PAGESIZE);
		1278
		1279	return page;
		1280	}
		1281	#endif
		1282
		1283	static void
		1284	eio_page_align (void **addr, size_t *length)
		1285	{
		1286	intptr_t mask = eio_pagesize () - 1;
		1287
		1288	/* round down addr */
		1289	intptr_t adj = mask & (intptr_t)*addr;
		1290
		1291	*addr = (void *)((intptr_t)*addr - adj);
		1292	*length += adj;
		1293
		1294	/* round up length */
		1295	*length = (*length + mask) & ~mask;
		1296	}
		1297
		1298	#if !_POSIX_MEMLOCK
		1299	# define eio__mlockall(a) EIO_ENOSYS ()
		1300	#else
		1301
		1302	static int
		1303	eio__mlockall (int flags)
		1304	{
		1305	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1306	extern int mallopt (int, int);
		1307	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1308	#endif
		1309
		1310	if (EIO_MCL_CURRENT != MCL_CURRENT
		1311	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1312	{
		1313	flags = 0
		1314	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1315	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1316	}
		1317
		1318	return mlockall (flags);
		1319	}
		1320	#endif
		1321
		1322	#if !_POSIX_MEMLOCK_RANGE
		1323	# define eio__mlock(a,b) EIO_ENOSYS ()
		1324	#else
		1325
		1326	static int
		1327	eio__mlock (void *addr, size_t length)
		1328	{
		1329	eio_page_align (&addr, &length);
		1330
		1331	return mlock (addr, length);
		1332	}
		1333
		1334	#endif
		1335
		1336	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1337	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1338	#else
		1339
		1340	static int
		1341	eio__msync (void *mem, size_t len, int flags)
		1342	{
		1343	eio_page_align (&mem, &len);
		1344
		1345	if (EIO_MS_ASYNC != MS_SYNC
		1346	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1347	|| EIO_MS_SYNC != MS_SYNC)
		1348	{
		1349	flags = 0
		1350	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1351	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1352	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1353	}
		1354
		1355	return msync (mem, len, flags);
		1356	}
		1357
		1358	#endif
		1359
		1360	static int
		1361	eio__mtouch (eio_req *req)
		1362	{
		1363	void *mem = req->ptr2;
		1364	size_t len = req->size;
		1365	int flags = req->int1;
		1366
		1367	eio_page_align (&mem, &len);
		1368
		1369	{
		1370	intptr_t addr = (intptr_t)mem;
		1371	intptr_t end = addr + len;
		1372	intptr_t page = eio_pagesize ();
		1373
		1374	if (addr < end)
		1375	if (flags & EIO_MT_MODIFY) /* modify */
		1376	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1377	else
		1378	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1379	}
		1380
		1381	return 0;
		1382	}
		1383
		1384	/*****************************************************************************/
		1385	/* requests implemented outside eio_execute, because they are so large */
		1386
		1387	/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
		1388	static int
		1389	eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1390	{
		1391	const char *rel = path;
		1392	char *res;
		1393	char *tmp1, *tmp2;
		1394	#if SYMLOOP_MAX > 32
		1395	int symlinks = SYMLOOP_MAX;
		1396	#else
		1397	int symlinks = 32;
		1398	#endif
		1399
		1400	errno = EINVAL;
		1401	if (!rel)
		1402	return -1;
		1403
		1404	errno = ENOENT;
		1405	if (!*rel)
		1406	return -1;
		1407
		1408	res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
		1409	tmp1 = res + PATH_MAX;
		1410	tmp2 = tmp1 + PATH_MAX;
		1411
		1412	#if 0 /* disabled, the musl way to do things is just too racy */
		1413	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1414	/* on linux we may be able to ask the kernel */
		1415	{
		1416	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1417
		1418	if (fd >= 0)
		1419	{
		1420	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1421	req->result = readlink (tmp1, res, PATH_MAX);
		1422	close (fd);
		1423
		1424	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1425
		1426	if (req->result > 0)
		1427	goto done;
		1428	}
		1429	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1430	return;
		1431	}
		1432	#endif
		1433	#endif
		1434
		1435	if (*rel != '/')
		1436	{
		1437	int len;
		1438
		1439	errno = ENOENT;
		1440	if (wd == EIO_INVALID_WD)
		1441	return -1;
		1442
		1443	if (wd == EIO_CWD)
		1444	{
		1445	if (!getcwd (res, PATH_MAX))
		1446	return -1;
		1447
		1448	len = strlen (res);
		1449	}
		1450	else
		1451	memcpy (res, wd->str, len = wd->len);
		1452
		1453	if (res [1]) /* only use if not / */
		1454	res += len;
		1455	}
		1456
		1457	while (*rel)
		1458	{
		1459	eio_ssize_t len, linklen;
		1460	const char *beg = rel;
		1461
		1462	while (*rel && *rel != '/')
		1463	++rel;
		1464
		1465	len = rel - beg;
		1466
		1467	if (!len) /* skip slashes */
		1468	{
		1469	++rel;
		1470	continue;
		1471	}
		1472
		1473	if (beg [0] == '.')
		1474	{
		1475	if (len == 1)
		1476	continue; /* . - nop */
		1477
		1478	if (beg [1] == '.' && len == 2)
		1479	{
		1480	/* .. - back up one component, if possible */
		1481
		1482	while (res != tmpbuf->ptr)
		1483	if (*--res == '/')
		1484	break;
		1485
		1486	continue;
		1487	}
		1488	}
		1489
		1490	errno = ENAMETOOLONG;
		1491	if (res + 1 + len + 1 >= tmp1)
		1492	return;
		1493
		1494	/* copy one component */
		1495	*res = '/';
		1496	memcpy (res + 1, beg, len);
		1497
		1498	/* zero-terminate, for readlink */
		1499	res [len + 1] = 0;
		1500
		1501	/* now check if it's a symlink */
		1502	linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
		1503
		1504	if (linklen < 0)
		1505	{
		1506	if (errno != EINVAL)
		1507	return -1;
		1508
		1509	/* it's a normal directory. hopefully */
		1510	res += len + 1;
		1511	}
		1512	else
		1513	{
		1514	/* yay, it was a symlink - build new path in tmp2 */
		1515	int rellen = strlen (rel);
		1516
		1517	errno = ENAMETOOLONG;
		1518	if (linklen + 1 + rellen >= PATH_MAX)
		1519	return -1;
		1520
		1521	errno = ELOOP;
		1522	if (!--symlinks)
		1523	return -1;
		1524
		1525	if (*tmp1 == '/')
		1526	res = tmpbuf->ptr; /* symlink resolves to an absolute path */
		1527
		1528	/* we need to be careful, as rel might point into tmp2 already */
		1529	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1530	tmp2 [linklen] = '/';
		1531	memcpy (tmp2, tmp1, linklen);
		1532
		1533	rel = tmp2;
		1534	}
		1535	}
		1536
		1537	/* special case for the lone root path */
		1538	if (res == tmpbuf->ptr)
		1539	*res++ = '/';
		1540
		1541	return res - (char *)tmpbuf->ptr;
1018	}	1542	}
1019		1543
1020	static signed char	1544	static signed char
1021	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1545	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1022	{	1546	{
1023	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */	1547	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1024	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;	1548	: a->inode < b->inode ? -1
		1549	: a->inode > b->inode ? 1
		1550	: 0;
1025	}	1551	}
1026		1552
1027	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0	1553	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1028		1554
1029	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1555	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1030	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1556	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1031		1557
1032	static void	1558	static void
1033	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1559	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1034	{	1560	{
1035	unsigned char bits [9 + sizeof (ino_t) * 8];	1561	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1036	unsigned char *bit = bits;	1562	unsigned char *bit = bits;
1037		1563
1038	assert (CHAR_BIT == 8);	1564	assert (CHAR_BIT == 8);
1039	assert (sizeof (eio_dirent) * 8 < 256);	1565	assert (sizeof (eio_dirent) * 8 < 256);
1040	assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */	1566	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1041	assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */	1567	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1042		1568
1043	if (size <= EIO_SORT_FAST)	1569	if (size <= EIO_SORT_FAST)
1044	return;	1570	return;
1045		1571
1046	/* first prepare an array of bits to test in our radix sort */	1572	/* first prepare an array of bits to test in our radix sort */
1047	/* try to take endianness into account, as well as differences in ino_t sizes */	1573	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1048	/* inode_bits must contain all inodes ORed together */	1574	/* inode_bits must contain all inodes ORed together */
1049	/* which is used to skip bits that are 0 everywhere, which is very common */	1575	/* which is used to skip bits that are 0 everywhere, which is very common */
1050	{	1576	{
1051	ino_t endianness;	1577	eio_ino_t endianness;
1052	int i, j;	1578	int i, j;
1053		1579
1054	/* we store the byte offset of byte n into byte n of "endianness" */	1580	/* we store the byte offset of byte n into byte n of "endianness" */
1055	for (i = 0; i < sizeof (ino_t); ++i)	1581	for (i = 0; i < sizeof (eio_ino_t); ++i)
1056	((unsigned char *)&endianness)[i] = i;	1582	((unsigned char *)&endianness)[i] = i;
1057		1583
1058	*bit++ = 0;	1584	*bit++ = 0;
1059		1585
1060	for (i = 0; i < sizeof (ino_t); ++i)	1586	for (i = 0; i < sizeof (eio_ino_t); ++i)
1061	{	1587	{
1062	/* shifting off the byte offsets out of "endianness" */	1588	/* shifting off the byte offsets out of "endianness" */
1063	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1589	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1064	endianness >>= 8;	1590	endianness >>= 8;
1065		1591
1066	for (j = 0; j < 8; ++j)	1592	for (j = 0; j < 8; ++j)
1067	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1593	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1068	*bit++ = offs + j;	1594	*bit++ = offs + j;
1069	}	1595	}
1070		1596
1071	for (j = 0; j < 8; ++j)	1597	for (j = 0; j < 8; ++j)
1072	if (score_bits & (1 << j))	1598	if (score_bits & (1 << j))
1073	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1599	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1074	}	1600	}
1075		1601
1076	/* now actually do the sorting (a variant of MSD radix sort) */	1602	/* now actually do the sorting (a variant of MSD radix sort) */
1077	{	1603	{
1078	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1604	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1079	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1605	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1080	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1606	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1081	int stk_idx = 0;	1607	int stk_idx = 0;
1082		1608
1083	base_stk [stk_idx] = dents;	1609	base_stk [stk_idx] = dents;
1084	end_stk [stk_idx] = dents + size;	1610	end_stk [stk_idx] = dents + size;
1085	bit_stk [stk_idx] = bit - 1;	1611	bit_stk [stk_idx] = bit - 1;
…		…
1164	}	1690	}
1165	}	1691	}
1166	}	1692	}
1167		1693
1168	static void	1694	static void
1169	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1695	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1170	{	1696	{
1171	if (size <= 1)	1697	if (size <= 1)
1172	return; /* our insertion sort relies on size > 0 */	1698	return; /* our insertion sort relies on size > 0 */
1173		1699
1174	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1700	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1182		1708
1183	/* read a full directory */	1709	/* read a full directory */
1184	static void	1710	static void
1185	eio__scandir (eio_req *req, etp_worker *self)	1711	eio__scandir (eio_req *req, etp_worker *self)
1186	{	1712	{
1187	DIR *dirp;
1188	EIO_STRUCT_DIRENT *entp;
1189	char *name, *names;	1713	char *name, *names;
1190	int namesalloc = 4096;	1714	int namesalloc = 4096 - sizeof (void *) * 4;
1191	int namesoffs = 0;	1715	int namesoffs = 0;
1192	int flags = req->int1;	1716	int flags = req->int1;
1193	eio_dirent *dents = 0;	1717	eio_dirent *dents = 0;
1194	int dentalloc = 128;	1718	int dentalloc = 128;
1195	int dentoffs = 0;	1719	int dentoffs = 0;
1196	ino_t inode_bits = 0;	1720	eio_ino_t inode_bits = 0;
		1721	#ifdef _WIN32
		1722	HANDLE dirp;
		1723	WIN32_FIND_DATA entp;
		1724	#else
		1725	DIR *dirp;
		1726	EIO_STRUCT_DIRENT *entp;
		1727	#endif
1197		1728
1198	req->result = -1;	1729	req->result = -1;
1199		1730
1200	if (!(flags & EIO_READDIR_DENTS))	1731	if (!(flags & EIO_READDIR_DENTS))
1201	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1732	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1202		1733
1203	X_LOCK (wrklock);	1734	#ifdef _WIN32
1204	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1735	{
		1736	int len = strlen ((const char *)req->ptr1);
		1737	char *path = malloc (MAX_PATH);
		1738	const char *fmt;
		1739	const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		1740
		1741	if (!len)
		1742	fmt = "./*";
		1743	else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
		1744	fmt = "%s*";
		1745	else
		1746	fmt = "%s/*";
		1747
		1748	_snprintf (path, MAX_PATH, fmt, reqpath);
		1749	dirp = FindFirstFile (path, &entp);
		1750	free (path);
		1751
		1752	if (dirp == INVALID_HANDLE_VALUE)
		1753	{
		1754	/* should steal _dosmaperr */
		1755	switch (GetLastError ())
		1756	{
		1757	case ERROR_FILE_NOT_FOUND:
		1758	req->result = 0;
		1759	break;
		1760
		1761	case ERROR_INVALID_NAME:
		1762	case ERROR_PATH_NOT_FOUND:
		1763	case ERROR_NO_MORE_FILES:
		1764	errno = ENOENT;
		1765	break;
		1766
		1767	case ERROR_NOT_ENOUGH_MEMORY:
		1768	errno = ENOMEM;
		1769	break;
		1770
		1771	default:
		1772	errno = EINVAL;
		1773	break;
		1774	}
		1775
		1776	return;
		1777	}
		1778	}
		1779	#else
		1780	#if HAVE_AT
		1781	if (req->wd)
		1782	{
		1783	int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1784
		1785	if (fd < 0)
		1786	return;
		1787
		1788	dirp = fdopendir (fd);
		1789
		1790	if (!dirp)
		1791	close (fd);
		1792	}
		1793	else
1205	self->dirp = dirp = opendir (req->ptr1);	1794	dirp = opendir (req->ptr1);
		1795	#else
		1796	dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
		1797	#endif
		1798
		1799	if (!dirp)
		1800	return;
		1801	#endif
		1802
		1803	if (req->flags & EIO_FLAG_PTR1_FREE)
		1804	free (req->ptr1);
		1805
1206	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1806	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1207	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1807	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1208	req->ptr2 = names = malloc (namesalloc);	1808	req->ptr2 = names = malloc (namesalloc);
1209	X_UNLOCK (wrklock);
1210		1809
1211	if (dirp && names && (!flags || dents))	1810	if (!names || (flags && !dents))
		1811	return;
		1812
1212	for (;;)	1813	for (;;)
1213	{	1814	{
		1815	int done;
		1816
		1817	#ifdef _WIN32
		1818	done = !dirp;
		1819	#else
1214	errno = 0;	1820	errno = 0;
1215	entp = readdir (dirp);	1821	entp = readdir (dirp);
		1822	done = !entp;
		1823	#endif
1216		1824
1217	if (!entp)	1825	if (done)
1218	{	1826	{
		1827	#ifndef _WIN32
		1828	int old_errno = errno;
		1829	closedir (dirp);
		1830	errno = old_errno;
		1831
1219	if (errno)	1832	if (errno)
1220	break;	1833	break;
		1834	#endif
1221		1835
1222	/* sort etc. */	1836	/* sort etc. */
1223	req->int1 = flags;	1837	req->int1 = flags;
1224	req->result = dentoffs;	1838	req->result = dentoffs;
1225		1839
1226	if (flags & EIO_READDIR_STAT_ORDER)	1840	if (flags & EIO_READDIR_STAT_ORDER)
1227	eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */	1841	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1228	else if (flags & EIO_READDIR_DIRS_FIRST)	1842	else if (flags & EIO_READDIR_DIRS_FIRST)
1229	if (flags & EIO_READDIR_FOUND_UNKNOWN)	1843	if (flags & EIO_READDIR_FOUND_UNKNOWN)
1230	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */	1844	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1231	else	1845	else
		1846	{
		1847	/* in this case, all is known, and we just put dirs first and sort them */
		1848	eio_dirent *oth = dents + dentoffs;
		1849	eio_dirent *dir = dents;
		1850
		1851	/* now partition dirs to the front, and non-dirs to the back */
		1852	/* by walking from both sides and swapping if necessary */
		1853	while (oth > dir)
		1854	{
		1855	if (dir->type == EIO_DT_DIR)
		1856	++dir;
		1857	else if ((--oth)->type == EIO_DT_DIR)
		1858	{
		1859	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1860
		1861	++dir;
		1862	}
		1863	}
		1864
		1865	/* now sort the dirs only (dirs all have the same score) */
		1866	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1867	}
		1868
		1869	break;
		1870	}
		1871
		1872	/* now add the entry to our list(s) */
		1873	name = D_NAME (entp);
		1874
		1875	/* skip . and .. entries */
		1876	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
		1877	{
		1878	int len = D_NAMLEN (entp) + 1;
		1879
		1880	while (ecb_expect_false (namesoffs + len > namesalloc))
		1881	{
		1882	namesalloc *= 2;
		1883	req->ptr2 = names = realloc (names, namesalloc);
		1884
		1885	if (!names)
		1886	break;
		1887	}
		1888
		1889	memcpy (names + namesoffs, name, len);
		1890
		1891	if (dents)
		1892	{
		1893	struct eio_dirent *ent;
		1894
		1895	if (ecb_expect_false (dentoffs == dentalloc))
1232	{	1896	{
1233	/* in this case, all is known, and we just put dirs first and sort them */	1897	dentalloc *= 2;
		1898	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1899
		1900	if (!dents)
		1901	break;
		1902	}
		1903
1234	eio_dirent *oth = dents + dentoffs;	1904	ent = dents + dentoffs;
1235	eio_dirent *dir = dents;
1236		1905
1237	/* now partition dirs to the front, and non-dirs to the back */	1906	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1238	/* by walking from both sides and swapping if necessary */	1907	ent->namelen = len - 1;
1239	/* also clear score, so it doesn't influence sorting */	1908	ent->inode = D_INO (entp);
1240	while (oth > dir)	1909
		1910	inode_bits |= ent->inode;
		1911
		1912	switch (D_TYPE (entp))
		1913	{
		1914	default:
		1915	ent->type = EIO_DT_UNKNOWN;
		1916	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1917	break;
		1918
		1919	#ifdef DT_FIFO
		1920	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1921	#endif
		1922	#ifdef DT_CHR
		1923	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1924	#endif
		1925	#ifdef DT_MPC
		1926	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1927	#endif
		1928	#ifdef DT_DIR
		1929	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1930	#endif
		1931	#ifdef DT_NAM
		1932	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1933	#endif
		1934	#ifdef DT_BLK
		1935	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1936	#endif
		1937	#ifdef DT_MPB
		1938	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1939	#endif
		1940	#ifdef DT_REG
		1941	case DT_REG: ent->type = EIO_DT_REG; break;
		1942	#endif
		1943	#ifdef DT_NWK
		1944	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1945	#endif
		1946	#ifdef DT_CMP
		1947	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1948	#endif
		1949	#ifdef DT_LNK
		1950	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1951	#endif
		1952	#ifdef DT_SOCK
		1953	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1954	#endif
		1955	#ifdef DT_DOOR
		1956	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1957	#endif
		1958	#ifdef DT_WHT
		1959	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1960	#endif
		1961	}
		1962
		1963	ent->score = 7;
		1964
		1965	if (flags & EIO_READDIR_DIRS_FIRST)
		1966	{
		1967	if (ent->type == EIO_DT_UNKNOWN)
1241	{	1968	{
1242	if (dir->type == EIO_DT_DIR)	1969	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1243	++dir;	1970	ent->score = 1;
1244	else if ((--oth)->type == EIO_DT_DIR)	1971	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1245	{	1972	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1246	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1247
1248	++dir;
1249	}
1250	}	1973	}
1251		1974	else if (ent->type == EIO_DT_DIR)
1252	/* now sort the dirs only */	1975	ent->score = 0;
1253	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1254	}	1976	}
1255
1256	break;
1257	}
1258
1259	/* now add the entry to our list(s) */
1260	name = entp->d_name;
1261
1262	/* skip . and .. entries */
1263	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1264	{
1265	int len = D_NAMLEN (entp) + 1;
1266
1267	while (expect_false (namesoffs + len > namesalloc))
1268	{
1269	namesalloc *= 2;
1270	X_LOCK (wrklock);
1271	req->ptr2 = names = realloc (names, namesalloc);
1272	X_UNLOCK (wrklock);
1273
1274	if (!names)
1275	break;
1276	}	1977	}
1277		1978
1278	memcpy (names + namesoffs, name, len);
1279
1280	if (dents)
1281	{
1282	struct eio_dirent *ent;
1283
1284	if (expect_false (dentoffs == dentalloc))
1285	{
1286	dentalloc *= 2;
1287	X_LOCK (wrklock);
1288	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1289	X_UNLOCK (wrklock);
1290
1291	if (!dents)
1292	break;
1293	}
1294
1295	ent = dents + dentoffs;
1296
1297	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1298	ent->namelen = len - 1;
1299	ent->inode = D_INO (entp);
1300
1301	inode_bits |= ent->inode;
1302
1303	switch (D_TYPE (entp))
1304	{
1305	default:
1306	ent->type = EIO_DT_UNKNOWN;
1307	flags |= EIO_READDIR_FOUND_UNKNOWN;
1308	break;
1309
1310	#ifdef DT_FIFO
1311	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1312	#endif
1313	#ifdef DT_CHR
1314	case DT_CHR: ent->type = EIO_DT_CHR; break;
1315	#endif
1316	#ifdef DT_MPC
1317	case DT_MPC: ent->type = EIO_DT_MPC; break;
1318	#endif
1319	#ifdef DT_DIR
1320	case DT_DIR: ent->type = EIO_DT_DIR; break;
1321	#endif
1322	#ifdef DT_NAM
1323	case DT_NAM: ent->type = EIO_DT_NAM; break;
1324	#endif
1325	#ifdef DT_BLK
1326	case DT_BLK: ent->type = EIO_DT_BLK; break;
1327	#endif
1328	#ifdef DT_MPB
1329	case DT_MPB: ent->type = EIO_DT_MPB; break;
1330	#endif
1331	#ifdef DT_REG
1332	case DT_REG: ent->type = EIO_DT_REG; break;
1333	#endif
1334	#ifdef DT_NWK
1335	case DT_NWK: ent->type = EIO_DT_NWK; break;
1336	#endif
1337	#ifdef DT_CMP
1338	case DT_CMP: ent->type = EIO_DT_CMP; break;
1339	#endif
1340	#ifdef DT_LNK
1341	case DT_LNK: ent->type = EIO_DT_LNK; break;
1342	#endif
1343	#ifdef DT_SOCK
1344	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1345	#endif
1346	#ifdef DT_DOOR
1347	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1348	#endif
1349	#ifdef DT_WHT
1350	case DT_WHT: ent->type = EIO_DT_WHT; break;
1351	#endif
1352	}
1353
1354	ent->score = 7;
1355
1356	if (flags & EIO_READDIR_DIRS_FIRST)
1357	{
1358	if (ent->type == EIO_DT_UNKNOWN)
1359	{
1360	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1361	ent->score = 1;
1362	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1363	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1364	}
1365	else if (ent->type == EIO_DT_DIR)
1366	ent->score = 0;
1367	}
1368	}
1369
1370	namesoffs += len;	1979	namesoffs += len;
1371	++dentoffs;	1980	++dentoffs;
1372	}	1981	}
1373		1982
1374	if (EIO_CANCELLED (req))	1983	if (EIO_CANCELLED (req))
1375	{	1984	{
1376	errno = ECANCELED;	1985	errno = ECANCELED;
1377	break;	1986	break;
1378	}	1987	}
		1988
		1989	#ifdef _WIN32
		1990	if (!FindNextFile (dirp, &entp))
		1991	{
		1992	FindClose (dirp);
		1993	dirp = 0;
1379	}	1994	}
		1995	#endif
		1996	}
1380	}	1997	}
1381		1998
1382	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)	1999	/*****************************************************************************/
1383	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)	2000	/* working directory stuff */
1384	#else	2001	/* various deficiencies in the posix 2008 api force us to */
		2002	/* keep the absolute path in string form at all times */
		2003	/* fuck yeah. */
1385		2004
1386	int	2005	#if !HAVE_AT
1387	eio__msync (void *mem, size_t len, int flags)	2006
		2007	/* a bit like realpath, but usually faster because it doesn'T have to return */
		2008	/* an absolute or canonical path */
		2009	static const char *
		2010	wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1388	{	2011	{
1389	if (EIO_MS_ASYNC != MS_SYNC	2012	if (!wd || *path == '/')
1390	|| EIO_MS_INVALIDATE != MS_INVALIDATE	2013	return path;
1391	|| EIO_MS_SYNC != MS_SYNC)	2014
		2015	if (path [0] == '.' && !path [1])
		2016	return wd->str;
		2017
		2018	{
		2019	int l1 = wd->len;
		2020	int l2 = strlen (path);
		2021
		2022	char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
		2023
		2024	memcpy (res, wd->str, l1);
		2025	res [l1] = '/';
		2026	memcpy (res + l1 + 1, path, l2 + 1);
		2027
		2028	return res;
		2029	}
		2030	}
		2031
		2032	#endif
		2033
		2034	static eio_wd
		2035	eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2036	{
		2037	int fd;
		2038	eio_wd res;
		2039	int len = eio__realpath (tmpbuf, wd, path);
		2040
		2041	if (len < 0)
		2042	return EIO_INVALID_WD;
		2043
		2044	#if HAVE_AT
		2045	fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		2046
		2047	if (fd < 0)
		2048	return EIO_INVALID_WD;
		2049	#endif
		2050
		2051	res = malloc (sizeof (*res) + len); /* one extra 0-byte */
		2052
		2053	#if HAVE_AT
		2054	res->fd = fd;
		2055	#endif
		2056
		2057	res->len = len;
		2058	memcpy (res->str, tmpbuf->ptr, len);
		2059	res->str [len] = 0;
		2060
		2061	return res;
		2062	}
		2063
		2064	eio_wd
		2065	eio_wd_open_sync (eio_wd wd, const char *path)
		2066	{
		2067	struct tmpbuf tmpbuf = { 0 };
		2068	wd = eio__wd_open_sync (&tmpbuf, wd, path);
		2069	free (tmpbuf.ptr);
		2070
		2071	return wd;
		2072	}
		2073
		2074	void
		2075	eio_wd_close_sync (eio_wd wd)
		2076	{
		2077	if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1392	{	2078	{
1393	flags = 0	2079	#if HAVE_AT
1394	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)	2080	close (wd->fd);
1395	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)	2081	#endif
1396	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);	2082	free (wd);
1397	}	2083	}
1398
1399	return msync (mem, len, flags);
1400	}	2084	}
1401		2085
1402	#endif	2086	#if HAVE_AT
1403		2087
1404	int	2088	/* they forgot these */
1405	eio__mtouch (void *mem, size_t len, int flags)
1406	{
1407	intptr_t addr = (intptr_t)mem;
1408	intptr_t end = addr + len;
1409	#ifdef PAGESIZE
1410	const intptr_t page = PAGESIZE;
1411	#else
1412	static intptr_t page;
1413		2089
1414	if (!page)	2090	static int
1415	page = sysconf (_SC_PAGESIZE);	2091	eio__truncateat (int dirfd, const char *path, off_t length)
1416	#endif	2092	{
		2093	int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
		2094	int res;
1417		2095
1418	/* round down to start of page, although this is probably useless */	2096	if (fd < 0)
1419	addr &= ~(page - 1); /* assume page size is always a power of two */	2097	return fd;
1420		2098
1421	if (addr < end)	2099	res = ftruncate (fd, length);
1422	if (flags & EIO_MT_MODIFY) /* modify */	2100	close (fd);
1423	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1424	else
1425	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1426
1427	return 0;	2101	return res;
1428	}	2102	}
		2103
		2104	static int
		2105	eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
		2106	{
		2107	int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
		2108	int res;
		2109
		2110	if (fd < 0)
		2111	return fd;
		2112
		2113	res = fstatvfs (fd, buf);
		2114	close (fd);
		2115	return res;
		2116
		2117	}
		2118
		2119	#endif
1429		2120
1430	/*****************************************************************************/	2121	/*****************************************************************************/
1431		2122
1432	#define ALLOC(len) \	2123	#define ALLOC(len) \
1433	if (!req->ptr2) \	2124	if (!req->ptr2) \
…		…
1442	req->result = -1; \	2133	req->result = -1; \
1443	break; \	2134	break; \
1444	} \	2135	} \
1445	}	2136	}
1446		2137
		2138	static void ecb_noinline ecb_cold
		2139	etp_proc_init (void)
		2140	{
		2141	#if HAVE_PRCTL_SET_NAME
		2142	/* provide a more sensible "thread name" */
		2143	char name[16 + 1];
		2144	const int namelen = sizeof (name) - 1;
		2145	int len;
		2146
		2147	prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
		2148	name [namelen] = 0;
		2149	len = strlen (name);
		2150	strcpy (name + (len < namelen - 4 ? len : namelen - 4), "/eio");
		2151	prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
		2152	}
		2153	#endif
		2154
1447	X_THREAD_PROC (etp_proc)	2155	X_THREAD_PROC (etp_proc)
1448	{	2156	{
1449	ETP_REQ *req;	2157	ETP_REQ *req;
1450	struct timespec ts;	2158	struct timespec ts;
1451	etp_worker *self = (etp_worker *)thr_arg;	2159	etp_worker *self = (etp_worker *)thr_arg;
1452		2160
		2161	etp_proc_init ();
		2162
1453	/* try to distribute timeouts somewhat randomly */	2163	/* try to distribute timeouts somewhat evenly */
1454	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	2164	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1455		2165
1456	for (;;)	2166	for (;;)
1457	{	2167	{
		2168	ts.tv_sec = 0;
		2169
1458	X_LOCK (reqlock);	2170	X_LOCK (reqlock);
1459		2171
1460	for (;;)	2172	for (;;)
1461	{	2173	{
1462	self->req = req = reqq_shift (&req_queue);	2174	req = reqq_shift (&req_queue);
1463		2175
1464	if (req)	2176	if (req)
1465	break;	2177	break;
1466		2178
		2179	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		2180	{
		2181	X_UNLOCK (reqlock);
		2182	X_LOCK (wrklock);
		2183	--started;
		2184	X_UNLOCK (wrklock);
		2185	goto quit;
		2186	}
		2187
1467	++idle;	2188	++idle;
1468		2189
1469	ts.tv_sec = time (0) + IDLE_TIMEOUT;	2190	if (idle <= max_idle)
1470	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	2191	/* we are allowed to idle, so do so without any timeout */
		2192	X_COND_WAIT (reqwait, reqlock);
		2193	else
1471	{	2194	{
1472	if (idle > max_idle)	2195	/* initialise timeout once */
1473	{	2196	if (!ts.tv_sec)
1474	--idle;	2197	ts.tv_sec = time (0) + idle_timeout;
1475	X_UNLOCK (reqlock);
1476	X_LOCK (wrklock);
1477	--started;
1478	X_UNLOCK (wrklock);
1479	goto quit;
1480	}
1481		2198
1482	/* we are allowed to idle, so do so without any timeout */
1483	X_COND_WAIT (reqwait, reqlock);	2199	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		2200	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1484	}	2201	}
1485		2202
1486	--idle;	2203	--idle;
1487	}	2204	}
1488		2205
…		…
1491	X_UNLOCK (reqlock);	2208	X_UNLOCK (reqlock);
1492		2209
1493	if (req->type < 0)	2210	if (req->type < 0)
1494	goto quit;	2211	goto quit;
1495		2212
1496	if (!EIO_CANCELLED (req))
1497	ETP_EXECUTE (self, req);	2213	ETP_EXECUTE (self, req);
1498		2214
1499	X_LOCK (reslock);	2215	X_LOCK (reslock);
1500		2216
1501	++npending;	2217	++npending;
1502		2218
1503	if (!reqq_push (&res_queue, req) && want_poll_cb)	2219	if (!reqq_push (&res_queue, req) && want_poll_cb)
1504	want_poll_cb ();	2220	want_poll_cb ();
1505		2221
1506	self->req = 0;
1507	etp_worker_clear (self);	2222	etp_worker_clear (self);
1508		2223
1509	X_UNLOCK (reslock);	2224	X_UNLOCK (reslock);
1510	}	2225	}
1511		2226
1512	quit:	2227	quit:
		2228	free (req);
		2229
1513	X_LOCK (wrklock);	2230	X_LOCK (wrklock);
1514	etp_worker_free (self);	2231	etp_worker_free (self);
1515	X_UNLOCK (wrklock);	2232	X_UNLOCK (wrklock);
1516		2233
1517	return 0;	2234	return 0;
1518	}	2235	}
1519		2236
1520	/*****************************************************************************/	2237	/*****************************************************************************/
1521		2238
		2239	int ecb_cold
1522	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	2240	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1523	{	2241	{
		2242	#if !HAVE_PREADWRITE
		2243	X_MUTEX_CREATE (preadwritelock);
		2244	#endif
		2245
1524	return etp_init (want_poll, done_poll);	2246	return etp_init (want_poll, done_poll);
1525	}	2247	}
1526		2248
		2249	ecb_inline void
1527	static void eio_api_destroy (eio_req *req)	2250	eio_api_destroy (eio_req *req)
1528	{	2251	{
1529	free (req);	2252	free (req);
1530	}	2253	}
1531		2254
1532	#define REQ(rtype) \	2255	#define REQ(rtype) \
…		…
1551	{ \	2274	{ \
1552	eio_api_destroy (req); \	2275	eio_api_destroy (req); \
1553	return 0; \	2276	return 0; \
1554	}	2277	}
1555		2278
		2279	static void
1556	static void eio_execute (etp_worker *self, eio_req *req)	2280	eio_execute (etp_worker *self, eio_req *req)
1557	{	2281	{
1558	errno = 0;	2282	#if HAVE_AT
		2283	int dirfd;
		2284	#else
		2285	const char *path;
		2286	#endif
		2287
		2288	if (ecb_expect_false (EIO_CANCELLED (req)))
		2289	{
		2290	req->result = -1;
		2291	req->errorno = ECANCELED;
		2292	return;
		2293	}
		2294
		2295	if (ecb_expect_false (req->wd == EIO_INVALID_WD))
		2296	{
		2297	req->result = -1;
		2298	req->errorno = ENOENT;
		2299	return;
		2300	}
		2301
		2302	if (req->type >= EIO_OPEN)
		2303	{
		2304	#if HAVE_AT
		2305	dirfd = WD2FD (req->wd);
		2306	#else
		2307	path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		2308	#endif
		2309	}
1559		2310
1560	switch (req->type)	2311	switch (req->type)
1561	{	2312	{
		2313	case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
		2314	req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
		2315	break;
		2316	case EIO_WD_CLOSE: req->result = 0;
		2317	eio_wd_close_sync (req->wd); break;
		2318
1562	case EIO_READ: ALLOC (req->size);	2319	case EIO_READ: ALLOC (req->size);
1563	req->result = req->offs >= 0	2320	req->result = req->offs >= 0
1564	? pread (req->int1, req->ptr2, req->size, req->offs)	2321	? pread (req->int1, req->ptr2, req->size, req->offs)
1565	: read (req->int1, req->ptr2, req->size); break;	2322	: read (req->int1, req->ptr2, req->size); break;
1566	case EIO_WRITE: req->result = req->offs >= 0	2323	case EIO_WRITE: req->result = req->offs >= 0
1567	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2324	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1568	: write (req->int1, req->ptr2, req->size); break;	2325	: write (req->int1, req->ptr2, req->size); break;
1569		2326
1570	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2327	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1571	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2328	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
		2329
		2330	#if HAVE_AT
1572		2331
1573	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2332	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1574	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2333	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1575	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2334	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1576	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2335	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1577	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1578	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1579
1580	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	2336	case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1581	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1582	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	2337	case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1583	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1584	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;	2338	case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1585	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1586
1587	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;	2339	case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1588	case EIO_CLOSE: req->result = close (req->int1); break;	2340
1589	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1590	case EIO_UNLINK: req->result = unlink (req->ptr1); break;	2341	case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1591	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	2342	case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1592	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	2343	case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1593	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2344	case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1594	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2345	case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1595	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2346	case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1596	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;	2347	case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1597
1598	case EIO_READLINK: ALLOC (NAME_MAX);	2348	case EIO_READLINK: ALLOC (PATH_MAX);
1599	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	2349	req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1600		2350	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1601	case EIO_SYNC: req->result = 0; sync (); break;	2351	req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1602	case EIO_FSYNC: req->result = fsync (req->int1); break;
1603	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1604	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1605	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1606	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1607
1608	case EIO_READDIR: eio__scandir (req, self); break;
1609
1610	case EIO_BUSY:	2352	case EIO_UTIME:
1611	#ifdef _WIN32	2353	case EIO_FUTIME:
1612	Sleep (req->nv1 * 1000.);	2354	{
		2355	struct timespec ts[2];
		2356	struct timespec *times;
		2357
		2358	if (req->nv1 != -1. || req->nv2 != -1.)
		2359	{
		2360	ts[0].tv_sec = req->nv1;
		2361	ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
		2362	ts[1].tv_sec = req->nv2;
		2363	ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
		2364
		2365	times = ts;
		2366	}
		2367	else
		2368	times = 0;
		2369
		2370	req->result = req->type == EIO_FUTIME
		2371	? futimens (req->int1, times)
		2372	: utimensat (dirfd, req->ptr1, times, 0);
		2373	}
		2374	break;
		2375
1613	#else	2376	#else
1614	{
1615	struct timeval tv;
1616		2377
1617	tv.tv_sec = req->nv1;	2378	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1618	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	2379	req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2380	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2381	req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2382	case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
		2383	case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
		2384	case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
		2385	case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1619		2386
1620	req->result = select (0, 0, 0, 0, &tv);	2387	case EIO_UNLINK: req->result = unlink (path ); break;
1621	}	2388	case EIO_RMDIR: req->result = rmdir (path ); break;
1622	#endif	2389	case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1623	break;	2390	case EIO_RENAME: req->result = rename (path , req->ptr2); break;
		2391	case EIO_LINK: req->result = link (path , req->ptr2); break;
		2392	case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
		2393	case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
		2394	case EIO_READLINK: ALLOC (PATH_MAX);
		2395	req->result = readlink (path, req->ptr2, PATH_MAX); break;
		2396	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2397	req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1624		2398
1625	case EIO_UTIME:	2399	case EIO_UTIME:
1626	case EIO_FUTIME:	2400	case EIO_FUTIME:
1627	{	2401	{
1628	struct timeval tv[2];	2402	struct timeval tv[2];
1629	struct timeval *times;	2403	struct timeval *times;
1630		2404
1631	if (req->nv1 != -1. || req->nv2 != -1.)	2405	if (req->nv1 != -1. || req->nv2 != -1.)
1632	{	2406	{
1633	tv[0].tv_sec = req->nv1;	2407	tv[0].tv_sec = req->nv1;
1634	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;	2408	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1635	tv[1].tv_sec = req->nv2;	2409	tv[1].tv_sec = req->nv2;
1636	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;	2410	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1637		2411
1638	times = tv;	2412	times = tv;
1639	}	2413	}
1640	else	2414	else
1641	times = 0;	2415	times = 0;
1642
1643		2416
1644	req->result = req->type == EIO_FUTIME	2417	req->result = req->type == EIO_FUTIME
1645	? futimes (req->int1, times)	2418	? futimes (req->int1, times)
1646	: utimes (req->ptr1, times);	2419	: utimes (req->ptr1, times);
1647	}	2420	}
1648	break;	2421	break;
1649		2422
		2423	#endif
		2424
		2425	case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
		2426	{
		2427	ALLOC (req->result);
		2428	memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
		2429	}
		2430	break;
		2431
		2432	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2433	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2434
		2435	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2436	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2437
		2438	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		2439	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		2440	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		2441
		2442	case EIO_CLOSE: req->result = close (req->int1); break;
		2443	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		2444	case EIO_SYNC: req->result = 0; sync (); break;
		2445	case EIO_FSYNC: req->result = fsync (req->int1); break;
		2446	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2447	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2448	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2449	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2450	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		2451	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2452	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		2453	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
		2454
		2455	case EIO_READDIR: eio__scandir (req, self); break;
		2456
		2457	case EIO_BUSY:
		2458	#ifdef _WIN32
		2459	Sleep (req->nv1 * 1e3);
		2460	#else
		2461	{
		2462	struct timeval tv;
		2463
		2464	tv.tv_sec = req->nv1;
		2465	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
		2466
		2467	req->result = select (0, 0, 0, 0, &tv);
		2468	}
		2469	#endif
		2470	break;
		2471
1650	case EIO_GROUP:	2472	case EIO_GROUP:
1651	abort (); /* handled in eio_request */	2473	abort (); /* handled in eio_request */
1652		2474
1653	case EIO_NOP:	2475	case EIO_NOP:
1654	req->result = 0;	2476	req->result = 0;
1655	break;	2477	break;
1656		2478
1657	case EIO_CUSTOM:	2479	case EIO_CUSTOM:
1658	((void (*)(eio_req *))req->feed) (req);	2480	req->feed (req);
1659	break;	2481	break;
1660		2482
1661	default:	2483	default:
		2484	errno = ENOSYS;
1662	req->result = -1;	2485	req->result = -1;
1663	break;	2486	break;
1664	}	2487	}
1665		2488
1666	req->errorno = errno;	2489	req->errorno = errno;
1667	}	2490	}
1668		2491
1669	#ifndef EIO_NO_WRAPPERS	2492	#ifndef EIO_NO_WRAPPERS
1670		2493
		2494	eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
		2495	{
		2496	REQ (EIO_WD_OPEN); PATH; SEND;
		2497	}
		2498
		2499	eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
		2500	{
		2501	REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
		2502	}
		2503
1671	eio_req *eio_nop (int pri, eio_cb cb, void *data)	2504	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1672	{	2505	{
1673	REQ (EIO_NOP); SEND;	2506	REQ (EIO_NOP); SEND;
1674	}	2507	}
1675		2508
…		…
1691	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2524	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1692	{	2525	{
1693	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2526	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1694	}	2527	}
1695		2528
		2529	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		2530	{
		2531	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2532	}
		2533
		2534	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2535	{
		2536	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2537	}
		2538
		2539	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2540	{
		2541	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2542	}
		2543
1696	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2544	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1697	{	2545	{
1698	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2546	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1699	}	2547	}
1700		2548
1701	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2549	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1702	{	2550	{
1703	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2551	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1704	}	2552	}
1705		2553
1706	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2554	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1707	{	2555	{
1708	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2556	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2557	}
		2558
		2559	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2560	{
		2561	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1709	}	2562	}
1710		2563
1711	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2564	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1712	{	2565	{
1713	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2566	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
1731	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2584	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1732	{	2585	{
1733	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2586	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1734	}	2587	}
1735		2588
		2589	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2590	{
		2591	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2592	}
		2593
1736	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2594	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1737	{	2595	{
1738	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2596	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1739	}	2597	}
1740		2598
…		…
1746	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2604	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1747	{	2605	{
1748	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2606	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1749	}	2607	}
1750		2608
1751	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2609	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1752	{	2610	{
1753	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2611	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1754	}	2612	}
1755		2613
1756	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2614	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1776	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2634	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1777	{	2635	{
1778	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2636	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1779	}	2637	}
1780		2638
1781	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2639	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1782	{	2640	{
1783	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2641	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1784	}	2642	}
1785		2643
1786	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2644	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1802	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2660	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1803	{	2661	{
1804	return eio__1path (EIO_READLINK, path, pri, cb, data);	2662	return eio__1path (EIO_READLINK, path, pri, cb, data);
1805	}	2663	}
1806		2664
		2665	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2666	{
		2667	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2668	}
		2669
1807	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2670	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1808	{	2671	{
1809	return eio__1path (EIO_STAT, path, pri, cb, data);	2672	return eio__1path (EIO_STAT, path, pri, cb, data);
1810	}	2673	}
1811		2674
1812	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2675	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1813	{	2676	{
1814	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2677	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1815	}	2678	}
1816		2679
		2680	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2681	{
		2682	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2683	}
		2684
1817	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2685	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1818	{	2686	{
1819	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2687	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1820	}	2688	}
1821		2689
…		…
1829	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;	2697	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1830	}	2698	}
1831		2699
1832	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2700	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1833	{	2701	{
1834	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2702	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1835	}	2703	}
1836		2704
1837	static eio_req *	2705	static eio_req *
1838	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2706	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1839	{	2707	{
…		…
1863	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2731	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1864	{	2732	{
1865	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2733	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1866	}	2734	}
1867		2735
1868	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2736	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1869	{	2737	{
1870	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;	2738	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1871	}	2739	}
1872		2740
1873	#endif	2741	#endif
1874		2742
1875	eio_req *eio_grp (eio_cb cb, void *data)	2743	eio_req *eio_grp (eio_cb cb, void *data)
…		…
1884	#undef SEND	2752	#undef SEND
1885		2753
1886	/*****************************************************************************/	2754	/*****************************************************************************/
1887	/* grp functions */	2755	/* grp functions */
1888		2756
		2757	void
1889	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2758	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1890	{	2759	{
1891	grp->int2 = limit;	2760	grp->int2 = limit;
1892	grp->feed = feed;	2761	grp->feed = feed;
1893		2762
1894	grp_try_feed (grp);	2763	grp_try_feed (grp);
1895	}	2764	}
1896		2765
		2766	void
1897	void eio_grp_limit (eio_req *grp, int limit)	2767	eio_grp_limit (eio_req *grp, int limit)
1898	{	2768	{
1899	grp->int2 = limit;	2769	grp->int2 = limit;
1900		2770
1901	grp_try_feed (grp);	2771	grp_try_feed (grp);
1902	}	2772	}
1903		2773
		2774	void
1904	void eio_grp_add (eio_req *grp, eio_req *req)	2775	eio_grp_add (eio_req *grp, eio_req *req)
1905	{	2776	{
1906	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2777	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1907		2778
1908	grp->flags |= EIO_FLAG_GROUPADD;	2779	grp->flags |= EIO_FLAG_GROUPADD;
1909		2780
…		…
1920	}	2791	}
1921		2792
1922	/*****************************************************************************/	2793	/*****************************************************************************/
1923	/* misc garbage */	2794	/* misc garbage */
1924		2795
		2796	eio_ssize_t
1925	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2797	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1926	{	2798	{
1927	etp_worker wrk;
1928	ssize_t ret;
1929
1930	wrk.dbuf = 0;
1931
1932	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2799	return eio__sendfile (ofd, ifd, offset, count);
1933
1934	if (wrk.dbuf)
1935	free (wrk.dbuf);
1936
1937	return ret;
1938	}	2800	}
1939		2801

Diff Legend

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