ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/cvsroot/libeio/eio.c

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.11 by root, Tue May 13 18:50:27 2008 UTC vs.
Revision 1.114 by root, Fri Dec 30 07:38:00 2011 UTC

		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
		5	* All rights reserved.
		6	*
		7	* Redistribution and use in source and binary forms, with or without modifica-
		8	* tion, are permitted provided that the following conditions are met:
		9	*
		10	* 1. Redistributions of source code must retain the above copyright notice,
		11	* this list of conditions and the following disclaimer.
		12	*
		13	* 2. Redistributions in binary form must reproduce the above copyright
		14	* notice, this list of conditions and the following disclaimer in the
		15	* documentation and/or other materials provided with the distribution.
		16	*
		17	* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
		18	* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
		19	* CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
		20	* EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
		21	* CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
		22	* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
		23	* OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
		24	* WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
		25	* ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
		26	* OF THE POSSIBILITY OF SUCH DAMAGE.
		27	*
		28	* Alternatively, the contents of this file may be used under the terms of
		29	* the GNU General Public License ("GPL") version 2 or any later version,
		30	* in which case the provisions of the GPL are applicable instead of
		31	* the above. If you wish to allow the use of your version of this file
		32	* only under the terms of the GPL and not to allow others to use your
		33	* version of this file under the BSD license, indicate your decision
		34	* by deleting the provisions above and replace them with the notice
		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
		37	* either the BSD or the GPL.
		38	*/
		39
		40	#ifndef _WIN32
		41	# include "config.h"
		42	#endif
		43
1	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
		46
		47	#ifdef EIO_STACKSIZE
		48	# define X_STACKSIZE EIO_STACKSIZE
		49	#endif
2	#include "xthread.h"	50	#include "xthread.h"
3		51
4	#include <errno.h>	52	#include <errno.h>
5	#include <stddef.h>	53	#include <stddef.h>
6	#include <stdlib.h>	54	#include <stdlib.h>
…		…
10	#include <sys/stat.h>	58	#include <sys/stat.h>
11	#include <limits.h>	59	#include <limits.h>
12	#include <fcntl.h>	60	#include <fcntl.h>
13	#include <assert.h>	61	#include <assert.h>
14		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
15	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
16	# 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
17	#endif	84	#endif
18		85
19	#ifndef EIO_DESTROY	86	#ifndef EIO_DESTROY
…		…
22		89
23	#ifndef EIO_FEED	90	#ifndef EIO_FEED
24	# 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)
25	#endif	92	#endif
26		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
27	#ifdef _WIN32	105	#ifdef _WIN32
28		106
29	/*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)
30		196
31	#else	197	#else
32		198
33	# include "config.h"
34	# include <sys/time.h>	199	#include <sys/time.h>
35	# include <sys/select.h>	200	#include <sys/select.h>
		201	#include <sys/statvfs.h>
36	# 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
37	# include <utime.h>	237	# include <utime.h>
38	# include <signal.h>
39	# include <dirent.h>
40
41	# ifndef EIO_STRUCT_DIRENT
42	# define EIO_STRUCT_DIRENT struct dirent
43	# endif	238	#endif
44		239
		240	#if HAVE_SYS_SYSCALL_H
		241	# include <sys/syscall.h>
		242	#endif
		243
		244	#if HAVE_SYS_PRCTL_H
		245	# include <sys/prctl.h>
45	#endif	246	#endif
46		247
47	#if HAVE_SENDFILE	248	#if HAVE_SENDFILE
48	# if __linux	249	# if __linux
49	# include <sys/sendfile.h>	250	# include <sys/sendfile.h>
50	# elif __freebsd	251	# elif __FreeBSD__ || defined __APPLE__
51	# include <sys/socket.h>	252	# include <sys/socket.h>
52	# include <sys/uio.h>	253	# include <sys/uio.h>
53	# elif __hpux	254	# elif __hpux
54	# include <sys/socket.h>	255	# include <sys/socket.h>
55	# elif __solaris /* not yet */	256	# elif __solaris
56	# include <sys/sendfile.h>	257	# include <sys/sendfile.h>
57	# else	258	# else
58	# error sendfile support requested but not available	259	# error sendfile support requested but not available
59	# endif	260	# endif
60	#endif	261	#endif
61		262
62	/* number of seconds after which an idle threads exit */	263	#ifndef D_TYPE
63	#define IDLE_TIMEOUT 10	264	# define D_TYPE(de) 0
		265	#endif
		266	#ifndef D_INO
		267	# define D_INO(de) 0
		268	#endif
		269	#ifndef D_NAMLEN
		270	# define D_NAMLEN(entp) strlen (D_NAME (entp))
		271	#endif
64		272
65	/* used for struct dirent, AIX doesn't provide it */	273	/* used for struct dirent, AIX doesn't provide it */
66	#ifndef NAME_MAX	274	#ifndef NAME_MAX
67	# define NAME_MAX 4096	275	# define NAME_MAX 4096
68	#endif	276	#endif
69		277
		278	/* used for readlink etc. */
		279	#ifndef PATH_MAX
		280	# define PATH_MAX 4096
		281	#endif
		282
70	/* buffer size for various temporary buffers */	283	/* buffer size for various temporary buffers */
71	#define EIO_BUFSIZE 65536	284	#define EIO_BUFSIZE 65536
72		285
73	#define dBUF \	286	#define dBUF \
74	char *eio_buf; \
75	X_LOCK (wrklock); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	287	char *eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (wrklock); \
78	errno = ENOMEM; \	288	errno = ENOMEM; \
79	if (!eio_buf) \	289	if (!eio_buf) \
80	return -1;	290	return -1
		291
		292	#define FUBd \
		293	free (eio_buf)
81		294
82	#define EIO_TICKS ((1000000 + 1023) >> 10)	295	#define EIO_TICKS ((1000000 + 1023) >> 10)
		296
		297	/*****************************************************************************/
		298
		299	struct tmpbuf
		300	{
		301	void *ptr;
		302	int len;
		303	};
		304
		305	static void *
		306	tmpbuf_get (struct tmpbuf *buf, int len)
		307	{
		308	if (buf->len < len)
		309	{
		310	free (buf->ptr);
		311	buf->ptr = malloc (buf->len = len);
		312	}
		313
		314	return buf->ptr;
		315	}
		316
		317	struct tmpbuf;
		318
		319	#if _POSIX_VERSION >= 200809L
		320	#define HAVE_AT 1
		321	#define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
		322	#ifndef O_SEARCH
		323	#define O_SEARCH O_RDONLY
		324	#endif
		325	#else
		326	#define HAVE_AT 0
		327	static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
		328	#endif
		329
		330	struct eio_pwd
		331	{
		332	#if HAVE_AT
		333	int fd;
		334	#endif
		335	int len;
		336	char str[1]; /* actually, a 0-terminated canonical path */
		337	};
		338
		339	/*****************************************************************************/
		340
		341	#define ETP_PRI_MIN EIO_PRI_MIN
		342	#define ETP_PRI_MAX EIO_PRI_MAX
		343
		344	struct etp_worker;
		345
		346	#define ETP_REQ eio_req
		347	#define ETP_DESTROY(req) eio_destroy (req)
		348	static int eio_finish (eio_req *req);
		349	#define ETP_FINISH(req) eio_finish (req)
		350	static void eio_execute (struct etp_worker *self, eio_req *req);
		351	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		352
		353	/*****************************************************************************/
		354
		355	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		356
		357	/* calculate time difference in ~1/EIO_TICKS of a second */
		358	ecb_inline int
		359	tvdiff (struct timeval *tv1, struct timeval *tv2)
		360	{
		361	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
		362	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
		363	}
		364
		365	static unsigned int started, idle, wanted = 4;
83		366
84	static void (*want_poll_cb) (void);	367	static void (*want_poll_cb) (void);
85	static void (*done_poll_cb) (void);	368	static void (*done_poll_cb) (void);
86		369
87	static unsigned int max_poll_time = 0;	370	static unsigned int max_poll_time; /* reslock */
88	static unsigned int max_poll_reqs = 0;	371	static unsigned int max_poll_reqs; /* reslock */
89		372
90	/* calculcate time difference in ~1/EIO_TICKS of a second */	373	static unsigned int nreqs; /* reqlock */
91	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	374	static unsigned int nready; /* reqlock */
92	{	375	static unsigned int npending; /* reqlock */
93	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	376	static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
94	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	377	static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
95	}
96		378
97	static unsigned int started, idle, wanted = 4;	379	static xmutex_t wrklock;
		380	static xmutex_t reslock;
		381	static xmutex_t reqlock;
		382	static xcond_t reqwait;
		383
		384	#if !HAVE_PREADWRITE
		385	/*
		386	* make our pread/pwrite emulation safe against themselves, but not against
		387	* normal read/write by using a mutex. slows down execution a lot,
		388	* but that's your problem, not mine.
		389	*/
		390	static xmutex_t preadwritelock;
		391	#endif
		392
		393	typedef struct etp_worker
		394	{
		395	struct tmpbuf tmpbuf;
		396
		397	/* locked by wrklock */
		398	struct etp_worker *prev, *next;
		399
		400	xthread_t tid;
		401
		402	#ifdef ETP_WORKER_COMMON
		403	ETP_WORKER_COMMON
		404	#endif
		405	} etp_worker;
		406
		407	static etp_worker wrk_first; /* NOT etp */
		408
		409	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		410	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
98		411
99	/* worker threads management */	412	/* worker threads management */
100	static mutex_t wrklock = X_MUTEX_INIT;
101		413
102	typedef struct worker	414	static void
103	{
104	/* locked by wrklock */
105	struct worker *prev, *next;
106
107	thread_t tid;
108
109	/* locked by reslock, reqlock or wrklock */
110	eio_req *req; /* currently processed request */
111	void *dbuf;
112	DIR *dirp;
113	} worker;
114
115	static worker wrk_first = { &wrk_first, &wrk_first, 0 };
116
117	static void worker_clear (worker *wrk)	415	etp_worker_clear (etp_worker *wrk)
118	{	416	{
119	if (wrk->dirp)	417	}
120	{
121	closedir (wrk->dirp);
122	wrk->dirp = 0;
123	}
124		418
125	if (wrk->dbuf)	419	static void ecb_cold
126	{	420	etp_worker_free (etp_worker *wrk)
		421	{
127	free (wrk->dbuf);	422	free (wrk->tmpbuf.ptr);
128	wrk->dbuf = 0;
129	}
130	}
131		423
132	static void worker_free (worker *wrk)
133	{
134	wrk->next->prev = wrk->prev;	424	wrk->next->prev = wrk->prev;
135	wrk->prev->next = wrk->next;	425	wrk->prev->next = wrk->next;
136		426
137	free (wrk);	427	free (wrk);
138	}	428	}
139		429
140	static volatile unsigned int nreqs, nready, npending;	430	static unsigned int
141	static volatile unsigned int max_idle = 4;	431	etp_nreqs (void)
142
143	static mutex_t reslock = X_MUTEX_INIT;
144	static mutex_t reqlock = X_MUTEX_INIT;
145	static cond_t reqwait = X_COND_INIT;
146
147	unsigned int eio_nreqs (void)
148	{	432	{
		433	int retval;
		434	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		435	retval = nreqs;
		436	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
149	return nreqs;	437	return retval;
150	}	438	}
151		439
152	unsigned int eio_nready (void)	440	static unsigned int
		441	etp_nready (void)
153	{	442	{
154	unsigned int retval;	443	unsigned int retval;
155		444
156	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	445	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157	retval = nready;	446	retval = nready;
158	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	447	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159		448
160	return retval;	449	return retval;
161	}	450	}
162		451
163	unsigned int eio_npending (void)	452	static unsigned int
		453	etp_npending (void)
164	{	454	{
165	unsigned int retval;	455	unsigned int retval;
166		456
167	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	457	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168	retval = npending;	458	retval = npending;
169	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	459	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170		460
171	return retval;	461	return retval;
172	}	462	}
173		463
174	unsigned int eio_nthreads (void)	464	static unsigned int
		465	etp_nthreads (void)
175	{	466	{
176	unsigned int retval;	467	unsigned int retval;
177		468
178	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	469	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179	retval = started;	470	retval = started;
…		…
186	* a somewhat faster data structure might be nice, but	477	* a somewhat faster data structure might be nice, but
187	* with 8 priorities this actually needs <20 insns	478	* with 8 priorities this actually needs <20 insns
188	* per shift, the most expensive operation.	479	* per shift, the most expensive operation.
189	*/	480	*/
190	typedef struct {	481	typedef struct {
191	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	482	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
192	int size;	483	int size;
193	} reqq;	484	} etp_reqq;
194		485
195	static reqq req_queue;	486	static etp_reqq req_queue;
196	static reqq res_queue;	487	static etp_reqq res_queue;
197		488
198	static int reqq_push (reqq *q, eio_req *req)	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
		501	reqq_push (etp_reqq *q, ETP_REQ *req)
199	{	502	{
200	int pri = req->pri;	503	int pri = req->pri;
201	req->next = 0;	504	req->next = 0;
202		505
203	if (q->qe[pri])	506	if (q->qe[pri])
…		…
209	q->qe[pri] = q->qs[pri] = req;	512	q->qe[pri] = q->qs[pri] = req;
210		513
211	return q->size++;	514	return q->size++;
212	}	515	}
213		516
214	static eio_req *reqq_shift (reqq *q)	517	static ETP_REQ * ecb_noinline
		518	reqq_shift (etp_reqq *q)
215	{	519	{
216	int pri;	520	int pri;
217		521
218	if (!q->size)	522	if (!q->size)
219	return 0;	523	return 0;
220		524
221	--q->size;	525	--q->size;
222		526
223	for (pri = EIO_NUM_PRI; pri--; )	527	for (pri = ETP_NUM_PRI; pri--; )
224	{	528	{
225	eio_req *req = q->qs[pri];	529	eio_req *req = q->qs[pri];
226		530
227	if (req)	531	if (req)
228	{	532	{
…		…
234	}	538	}
235		539
236	abort ();	540	abort ();
237	}	541	}
238		542
239	static void grp_try_feed (eio_req *grp)	543	static int ecb_cold
		544	etp_init (void (*want_poll)(void), void (*done_poll)(void))
240	{	545	{
241	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	546	X_MUTEX_CREATE (wrklock);
242	{	547	X_MUTEX_CREATE (reslock);
243	int old_len = grp->size;	548	X_MUTEX_CREATE (reqlock);
		549	X_COND_CREATE (reqwait);
244		550
245	EIO_FEED (grp);	551	reqq_init (&req_queue);
		552	reqq_init (&res_queue);
246		553
247	/* stop if no progress has been made */	554	wrk_first.next =
248	if (old_len == grp->size)	555	wrk_first.prev = &wrk_first;
249	{
250	grp->feed = 0;
251	break;
252	}
253	}
254	}
255		556
256	static int eio_finish (eio_req *req);	557	started = 0;
		558	idle = 0;
		559	nreqs = 0;
		560	nready = 0;
		561	npending = 0;
257		562
258	static int grp_dec (eio_req *grp)	563	want_poll_cb = want_poll;
259	{	564	done_poll_cb = done_poll;
260	--grp->size;
261		565
262	/* call feeder, if applicable */
263	grp_try_feed (grp);
264
265	/* finish, if done */
266	if (!grp->size && grp->int1)
267	return eio_finish (grp);
268	else
269	return 0;	566	return 0;
270	}	567	}
271		568
272	void eio_destroy (eio_req *req)
273	{
274	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
275	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
276
277	EIO_DESTROY (req);
278	}
279
280	static int eio_finish (eio_req *req)
281	{
282	int res = EIO_FINISH (req);
283
284	if (req->grp)
285	{
286	int res2;
287	eio_req *grp = req->grp;
288
289	/* unlink request */
290	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
291	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
292
293	if (grp->grp_first == req)
294	grp->grp_first = req->grp_next;
295
296	res2 = grp_dec (grp);
297
298	if (!res && res2)
299	res = res2;
300	}
301
302	eio_destroy (req);
303
304	return res;
305	}
306
307	void eio_grp_cancel (eio_req *grp)
308	{
309	for (grp = grp->grp_first; grp; grp = grp->grp_next)
310	eio_cancel (grp);
311	}
312
313	void eio_cancel (eio_req *req)
314	{
315	X_LOCK (wrklock);
316	req->flags |= EIO_FLAG_CANCELLED;
317	X_UNLOCK (wrklock);
318
319	eio_grp_cancel (req);
320	}
321
322	X_THREAD_PROC (eio_proc);	569	X_THREAD_PROC (etp_proc);
323		570
324	static void start_thread (void)	571	static void ecb_cold
		572	etp_start_thread (void)
325	{	573	{
326	worker *wrk = calloc (1, sizeof (worker));	574	etp_worker *wrk = calloc (1, sizeof (etp_worker));
327		575
328	/*TODO*/	576	/*TODO*/
329	assert (("unable to allocate worker thread data", wrk));	577	assert (("unable to allocate worker thread data", wrk));
330		578
331	X_LOCK (wrklock);	579	X_LOCK (wrklock);
332		580
333	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	581	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334	{	582	{
335	wrk->prev = &wrk_first;	583	wrk->prev = &wrk_first;
336	wrk->next = wrk_first.next;	584	wrk->next = wrk_first.next;
337	wrk_first.next->prev = wrk;	585	wrk_first.next->prev = wrk;
338	wrk_first.next = wrk;	586	wrk_first.next = wrk;
…		…
342	free (wrk);	590	free (wrk);
343		591
344	X_UNLOCK (wrklock);	592	X_UNLOCK (wrklock);
345	}	593	}
346		594
		595	static void
347	static void maybe_start_thread (void)	596	etp_maybe_start_thread (void)
348	{	597	{
349	if (eio_nthreads () >= wanted)	598	if (ecb_expect_true (etp_nthreads () >= wanted))
350	return;	599	return;
351		600
352	/* todo: maybe use idle here, but might be less exact */	601	/* todo: maybe use idle here, but might be less exact */
353	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	602	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354	return;	603	return;
355		604
356	start_thread ();	605	etp_start_thread ();
357	}	606	}
358		607
359	void eio_submit (eio_req *req)	608	static void ecb_cold
		609	etp_end_thread (void)
360	{	610	{
361	req->pri += EIO_PRI_BIAS;
362
363	if (req->pri < EIO_PRI_MIN + EIO_PRI_BIAS) req->pri = EIO_PRI_MIN + EIO_PRI_BIAS;
364	if (req->pri > EIO_PRI_MAX + EIO_PRI_BIAS) req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
365
366	++nreqs;
367
368	X_LOCK (reqlock);
369	++nready;
370	reqq_push (&req_queue, req);
371	X_COND_SIGNAL (reqwait);
372	X_UNLOCK (reqlock);
373
374	maybe_start_thread ();
375	}
376
377	static void end_thread (void)
378	{
379	eio_req *req = calloc (1, sizeof (eio_req));	611	eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
380		612
381	req->type = EIO_QUIT;	613	req->type = -1;
382	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	614	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
383		615
384	X_LOCK (reqlock);	616	X_LOCK (reqlock);
385	reqq_push (&req_queue, req);	617	reqq_push (&req_queue, req);
386	X_COND_SIGNAL (reqwait);	618	X_COND_SIGNAL (reqwait);
387	X_UNLOCK (reqlock);	619	X_UNLOCK (reqlock);
…		…
389	X_LOCK (wrklock);	621	X_LOCK (wrklock);
390	--started;	622	--started;
391	X_UNLOCK (wrklock);	623	X_UNLOCK (wrklock);
392	}	624	}
393		625
394	void eio_set_max_poll_time (double nseconds)	626	static int
395	{
396	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
397	max_poll_time = nseconds;
398	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
399	}
400
401	void eio_set_max_poll_reqs (unsigned int maxreqs)
402	{
403	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
404	max_poll_reqs = maxreqs;
405	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
406	}
407
408	void eio_set_max_idle (unsigned int nthreads)
409	{
410	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
411	max_idle = nthreads <= 0 ? 1 : nthreads;
412	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
413	}
414
415	void eio_set_min_parallel (unsigned int nthreads)
416	{
417	if (wanted < nthreads)
418	wanted = nthreads;
419	}
420
421	void eio_set_max_parallel (unsigned int nthreads)
422	{
423	if (wanted > nthreads)
424	wanted = nthreads;
425
426	while (started > wanted)
427	end_thread ();
428	}
429
430	int eio_poll (void)	627	etp_poll (void)
431	{	628	{
432	int maxreqs = max_poll_reqs;	629	unsigned int maxreqs;
		630	unsigned int maxtime;
433	struct timeval tv_start, tv_now;	631	struct timeval tv_start, tv_now;
434	eio_req *req;
435		632
		633	X_LOCK (reslock);
		634	maxreqs = max_poll_reqs;
		635	maxtime = max_poll_time;
		636	X_UNLOCK (reslock);
		637
436	if (max_poll_time)	638	if (maxtime)
437	gettimeofday (&tv_start, 0);	639	gettimeofday (&tv_start, 0);
438		640
439	for (;;)	641	for (;;)
440	{	642	{
		643	ETP_REQ *req;
		644
441	maybe_start_thread ();	645	etp_maybe_start_thread ();
442		646
443	X_LOCK (reslock);	647	X_LOCK (reslock);
444	req = reqq_shift (&res_queue);	648	req = reqq_shift (&res_queue);
445		649
446	if (req)	650	if (req)
…		…
454	X_UNLOCK (reslock);	658	X_UNLOCK (reslock);
455		659
456	if (!req)	660	if (!req)
457	return 0;	661	return 0;
458		662
		663	X_LOCK (reqlock);
459	--nreqs;	664	--nreqs;
		665	X_UNLOCK (reqlock);
460		666
461	if (req->type == EIO_GROUP && req->size)	667	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
462	{	668	{
463	req->int1 = 1; /* mark request as delayed */	669	req->int1 = 1; /* mark request as delayed */
464	continue;	670	continue;
465	}	671	}
466	else	672	else
467	{	673	{
468	int res = eio_finish (req);	674	int res = ETP_FINISH (req);
469	if (res)	675	if (ecb_expect_false (res))
470	return res;	676	return res;
471	}	677	}
472		678
473	if (maxreqs && !--maxreqs)	679	if (ecb_expect_false (maxreqs && !--maxreqs))
474	break;	680	break;
475		681
476	if (max_poll_time)	682	if (maxtime)
477	{	683	{
478	gettimeofday (&tv_now, 0);	684	gettimeofday (&tv_now, 0);
479		685
480	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	686	if (tvdiff (&tv_start, &tv_now) >= maxtime)
481	break;	687	break;
482	}	688	}
483	}	689	}
484		690
485	errno = EAGAIN;	691	errno = EAGAIN;
486	return -1;	692	return -1;
487	}	693	}
488		694
		695	static void
		696	etp_cancel (ETP_REQ *req)
		697	{
		698	req->cancelled = 1;
		699
		700	eio_grp_cancel (req);
		701	}
		702
		703	static void
		704	etp_submit (ETP_REQ *req)
		705	{
		706	req->pri -= ETP_PRI_MIN;
		707
		708	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		709	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		710
		711	if (ecb_expect_false (req->type == EIO_GROUP))
		712	{
		713	/* I hope this is worth it :/ */
		714	X_LOCK (reqlock);
		715	++nreqs;
		716	X_UNLOCK (reqlock);
		717
		718	X_LOCK (reslock);
		719
		720	++npending;
		721
		722	if (!reqq_push (&res_queue, req) && want_poll_cb)
		723	want_poll_cb ();
		724
		725	X_UNLOCK (reslock);
		726	}
		727	else
		728	{
		729	X_LOCK (reqlock);
		730	++nreqs;
		731	++nready;
		732	reqq_push (&req_queue, req);
		733	X_COND_SIGNAL (reqwait);
		734	X_UNLOCK (reqlock);
		735
		736	etp_maybe_start_thread ();
		737	}
		738	}
		739
		740	static void ecb_cold
		741	etp_set_max_poll_time (double nseconds)
		742	{
		743	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		744	max_poll_time = nseconds * EIO_TICKS;
		745	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		746	}
		747
		748	static void ecb_cold
		749	etp_set_max_poll_reqs (unsigned int maxreqs)
		750	{
		751	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		752	max_poll_reqs = maxreqs;
		753	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		754	}
		755
		756	static void ecb_cold
		757	etp_set_max_idle (unsigned int nthreads)
		758	{
		759	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		760	max_idle = nthreads;
		761	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		762	}
		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
		773	etp_set_min_parallel (unsigned int nthreads)
		774	{
		775	if (wanted < nthreads)
		776	wanted = nthreads;
		777	}
		778
		779	static void ecb_cold
		780	etp_set_max_parallel (unsigned int nthreads)
		781	{
		782	if (wanted > nthreads)
		783	wanted = nthreads;
		784
		785	while (started > wanted)
		786	etp_end_thread ();
		787	}
		788
		789	/*****************************************************************************/
		790
		791	static void
		792	grp_try_feed (eio_req *grp)
		793	{
		794	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		795	{
		796	grp->flags &= ~EIO_FLAG_GROUPADD;
		797
		798	EIO_FEED (grp);
		799
		800	/* stop if no progress has been made */
		801	if (!(grp->flags & EIO_FLAG_GROUPADD))
		802	{
		803	grp->feed = 0;
		804	break;
		805	}
		806	}
		807	}
		808
		809	static int
		810	grp_dec (eio_req *grp)
		811	{
		812	--grp->size;
		813
		814	/* call feeder, if applicable */
		815	grp_try_feed (grp);
		816
		817	/* finish, if done */
		818	if (!grp->size && grp->int1)
		819	return eio_finish (grp);
		820	else
		821	return 0;
		822	}
		823
		824	static void
		825	eio_destroy (eio_req *req)
		826	{
		827	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		828	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		829
		830	EIO_DESTROY (req);
		831	}
		832
		833	static int
		834	eio_finish (eio_req *req)
		835	{
		836	int res = EIO_FINISH (req);
		837
		838	if (req->grp)
		839	{
		840	int res2;
		841	eio_req *grp = req->grp;
		842
		843	/* unlink request */
		844	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		845	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		846
		847	if (grp->grp_first == req)
		848	grp->grp_first = req->grp_next;
		849
		850	res2 = grp_dec (grp);
		851
		852	if (!res)
		853	res = res2;
		854	}
		855
		856	eio_destroy (req);
		857
		858	return res;
		859	}
		860
		861	void
		862	eio_grp_cancel (eio_req *grp)
		863	{
		864	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		865	eio_cancel (grp);
		866	}
		867
		868	void
		869	eio_cancel (eio_req *req)
		870	{
		871	etp_cancel (req);
		872	}
		873
		874	void
		875	eio_submit (eio_req *req)
		876	{
		877	etp_submit (req);
		878	}
		879
		880	unsigned int
		881	eio_nreqs (void)
		882	{
		883	return etp_nreqs ();
		884	}
		885
		886	unsigned int
		887	eio_nready (void)
		888	{
		889	return etp_nready ();
		890	}
		891
		892	unsigned int
		893	eio_npending (void)
		894	{
		895	return etp_npending ();
		896	}
		897
		898	unsigned int ecb_cold
		899	eio_nthreads (void)
		900	{
		901	return etp_nthreads ();
		902	}
		903
		904	void ecb_cold
		905	eio_set_max_poll_time (double nseconds)
		906	{
		907	etp_set_max_poll_time (nseconds);
		908	}
		909
		910	void ecb_cold
		911	eio_set_max_poll_reqs (unsigned int maxreqs)
		912	{
		913	etp_set_max_poll_reqs (maxreqs);
		914	}
		915
		916	void ecb_cold
		917	eio_set_max_idle (unsigned int nthreads)
		918	{
		919	etp_set_max_idle (nthreads);
		920	}
		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
		929	eio_set_min_parallel (unsigned int nthreads)
		930	{
		931	etp_set_min_parallel (nthreads);
		932	}
		933
		934	void ecb_cold
		935	eio_set_max_parallel (unsigned int nthreads)
		936	{
		937	etp_set_max_parallel (nthreads);
		938	}
		939
		940	int eio_poll (void)
		941	{
		942	return etp_poll ();
		943	}
		944
489	/*****************************************************************************/	945	/*****************************************************************************/
490	/* work around various missing functions */	946	/* work around various missing functions */
491		947
492	#if !HAVE_PREADWRITE	948	#if !HAVE_PREADWRITE
		949	# undef pread
		950	# undef pwrite
493	# define pread eio__pread	951	# define pread eio__pread
494	# define pwrite eio__pwrite	952	# define pwrite eio__pwrite
495		953
496	/*
497	* make our pread/pwrite safe against themselves, but not against
498	* normal read/write by using a mutex. slows down execution a lot,
499	* but that's your problem, not mine.
500	*/
501	static mutex_t preadwritelock = X_MUTEX_INIT;
502
503	static ssize_t	954	static eio_ssize_t
504	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)
505	{	956	{
506	ssize_t res;	957	eio_ssize_t res;
507	off_t ooffset;	958	off_t ooffset;
508		959
509	X_LOCK (preadwritelock);	960	X_LOCK (preadwritelock);
510	ooffset = lseek (fd, 0, SEEK_CUR);	961	ooffset = lseek (fd, 0, SEEK_CUR);
511	lseek (fd, offset, SEEK_SET);	962	lseek (fd, offset, SEEK_SET);
…		…
514	X_UNLOCK (preadwritelock);	965	X_UNLOCK (preadwritelock);
515		966
516	return res;	967	return res;
517	}	968	}
518		969
519	static ssize_t	970	static eio_ssize_t
520	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)
521	{	972	{
522	ssize_t res;	973	eio_ssize_t res;
523	off_t ooffset;	974	off_t ooffset;
524		975
525	X_LOCK (preadwritelock);	976	X_LOCK (preadwritelock);
526	ooffset = lseek (fd, 0, SEEK_CUR);	977	ooffset = lseek (fd, 0, SEEK_CUR);
527	lseek (fd, offset, SEEK_SET);	978	lseek (fd, offset, SEEK_SET);
528	res = write (fd, buf, count);	979	res = write (fd, buf, count);
529	lseek (fd, offset, SEEK_SET);	980	lseek (fd, ooffset, SEEK_SET);
530	X_UNLOCK (preadwritelock);	981	X_UNLOCK (preadwritelock);
531		982
532	return res;	983	return res;
533	}	984	}
534	#endif	985	#endif
535		986
536	#ifndef HAVE_FUTIMES	987	#ifndef HAVE_UTIMES
537		988
		989	# undef utimes
538	# define utimes(path,times) eio__utimes (path, times)	990	# define utimes(path,times) eio__utimes (path, times)
539	# define futimes(fd,times) eio__futimes (fd, times)
540		991
541	static int	992	static int
542	eio__utimes (const char *filename, const struct timeval times[2])	993	eio__utimes (const char *filename, const struct timeval times[2])
543	{	994	{
544	if (times)	995	if (times)
…		…
552	}	1003	}
553	else	1004	else
554	return utime (filename, 0);	1005	return utime (filename, 0);
555	}	1006	}
556		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
557	static int eio__futimes (int fd, const struct timeval tv[2])	1016	eio__futimes (int fd, const struct timeval tv[2])
558	{	1017	{
559	errno = ENOSYS;	1018	errno = ENOSYS;
560	return -1;	1019	return -1;
561	}	1020	}
562		1021
563	#endif	1022	#endif
564		1023
565	#if !HAVE_FDATASYNC	1024	#if !HAVE_FDATASYNC
		1025	# undef fdatasync
566	# define fdatasync fsync	1026	# define fdatasync(fd) fsync (fd)
567	#endif	1027	#endif
		1028
		1029	static int
		1030	eio__syncfs (int fd)
		1031	{
		1032	int res;
		1033
		1034	#if HAVE_SYS_SYNCFS
		1035	res = (int)syscall (__NR_syncfs, (int)(fd));
		1036	#else
		1037	res = -1;
		1038	errno = ENOSYS;
		1039	#endif
		1040
		1041	if (res < 0 && errno == ENOSYS && fd >= 0)
		1042	sync ();
		1043
		1044	return res;
		1045	}
		1046
		1047	/* sync_file_range always needs emulation */
		1048	static int
		1049	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		1050	{
		1051	#if HAVE_SYNC_FILE_RANGE
		1052	int res;
		1053
		1054	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		1055	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		1056	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		1057	{
		1058	flags = 0
		1059	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		1060	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		1061	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		1062	}
		1063
		1064	res = sync_file_range (fd, offset, nbytes, flags);
		1065
		1066	if (!res || errno != ENOSYS)
		1067	return res;
		1068	#endif
		1069
		1070	/* even though we could play tricks with the flags, it's better to always
		1071	* call fdatasync, as that matches the expectation of its users best */
		1072	return fdatasync (fd);
		1073	}
		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	}
568		1085
569	#if !HAVE_READAHEAD	1086	#if !HAVE_READAHEAD
		1087	# undef readahead
570	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1088	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
571		1089
572	static ssize_t	1090	static eio_ssize_t
573	eio__readahead (int fd, off_t offset, size_t count, worker *self)	1091	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
574	{	1092	{
575	size_t todo = count;	1093	size_t todo = count;
576	dBUF;	1094	dBUF;
577		1095
578	while (todo > 0)	1096	while (todo > 0)
…		…
582	pread (fd, eio_buf, len, offset);	1100	pread (fd, eio_buf, len, offset);
583	offset += len;	1101	offset += len;
584	todo -= len;	1102	todo -= len;
585	}	1103	}
586		1104
		1105	FUBd;
		1106
587	errno = 0;	1107	errno = 0;
588	return count;	1108	return count;
589	}	1109	}
590		1110
591	#endif	1111	#endif
592		1112
593	#if !HAVE_READDIR_R
594	# define readdir_r eio__readdir_r
595
596	static mutex_t readdirlock = X_MUTEX_INIT;
597
598	static int
599	eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
600	{
601	EIO_STRUCT_DIRENT *e;
602	int errorno;
603
604	X_LOCK (readdirlock);
605
606	e = readdir (dirp);
607	errorno = errno;
608
609	if (e)
610	{
611	*res = ent;
612	strcpy (ent->d_name, e->d_name);
613	}
614	else
615	*res = 0;
616
617	X_UNLOCK (readdirlock);
618
619	errno = errorno;
620	return e ? 0 : -1;
621	}
622	#endif
623
624	/* sendfile always needs emulation */	1113	/* sendfile always needs emulation */
625	static ssize_t	1114	static eio_ssize_t
626	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self)	1115	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
627	{	1116	{
		1117	eio_ssize_t written = 0;
628	ssize_t res;	1118	eio_ssize_t res;
629		1119
630	if (!count)	1120	if (!count)
631	return 0;	1121	return 0;
632		1122
		1123	for (;;)
		1124	{
		1125	#ifdef __APPLE__
		1126	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1127	#endif
633	#if HAVE_SENDFILE	1128	#if HAVE_SENDFILE
634	# if __linux	1129	# if __linux
		1130	off_t soffset = offset;
635	res = sendfile (ofd, ifd, &offset, count);	1131	res = sendfile (ofd, ifd, &soffset, count);
636		1132
637	# elif __freebsd	1133	# elif __FreeBSD__
638	/*	1134	/*
639	* 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
640	* wasted on making it similar to other I/O functions.	1136	* wasted on making it similar to other I/O functions.
641	*/	1137	*/
642	{
643	off_t sbytes;	1138	off_t sbytes;
644	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1139	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
645		1140
646	if (res < 0 && sbytes)	1141	#if 0 /* according to the manpage, this is correct, but broken behaviour */
647	/* 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)
648	res = sbytes;	1146	res = sbytes;
649	}	1147	#endif
		1148
		1149	/* according to source inspection, this is correct, and useful behaviour */
		1150	if (sbytes)
		1151	res = sbytes;
		1152
		1153	# elif defined (__APPLE__)
		1154	off_t sbytes = count;
		1155	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1156
		1157	/* according to the manpage, sbytes is always valid */
		1158	if (sbytes)
		1159	res = sbytes;
650		1160
651	# elif __hpux	1161	# elif __hpux
652	res = sendfile (ofd, ifd, offset, count, 0, 0);	1162	res = sendfile (ofd, ifd, offset, count, 0, 0);
653		1163
654	# elif __solaris	1164	# elif __solaris
655	{
656	struct sendfilevec vec;	1165	struct sendfilevec vec;
657	size_t sbytes;	1166	size_t sbytes;
658		1167
659	vec.sfv_fd = ifd;	1168	vec.sfv_fd = ifd;
660	vec.sfv_flag = 0;	1169	vec.sfv_flag = 0;
661	vec.sfv_off = offset;	1170	vec.sfv_off = offset;
662	vec.sfv_len = count;	1171	vec.sfv_len = count;
663		1172
664	res = sendfilev (ofd, &vec, 1, &sbytes);	1173	res = sendfilev (ofd, &vec, 1, &sbytes);
665		1174
666	if (res < 0 && sbytes)	1175	if (res < 0 && sbytes)
667	res = sbytes;	1176	res = sbytes;
668	}
669		1177
670	# endif	1178	# endif
		1179
		1180	#elif defined (_WIN32) && 0
		1181	/* does not work, just for documentation of what would need to be done */
		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 */
		1185	HANDLE h = TO_SOCKET (ifd);
		1186	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1187	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1188
671	#else	1189	#else
672	res = -1;	1190	res = -1;
673	errno = ENOSYS;	1191	errno = ENOSYS;
674	#endif	1192	#endif
675		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
676	if (res < 0	1217	if (res < 0
677	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1218	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1219	/* BSDs */
		1220	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1221	|| errno == ENOTSUP
		1222	#endif
		1223	#ifdef EOPNOTSUPP /* windows */
		1224	|| errno == EOPNOTSUPP /* BSDs */
		1225	#endif
678	#if __solaris	1226	#if __solaris
679	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1227	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
680	#endif	1228	#endif
681	)	1229	)
682	)	1230	)
…		…
686		1234
687	res = 0;	1235	res = 0;
688		1236
689	while (count)	1237	while (count)
690	{	1238	{
691	ssize_t cnt;	1239	eio_ssize_t cnt;
692		1240
693	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);
694		1242
695	if (cnt <= 0)	1243	if (cnt <= 0)
696	{	1244	{
…		…
708		1256
709	offset += cnt;	1257	offset += cnt;
710	res += cnt;	1258	res += cnt;
711	count -= cnt;	1259	count -= cnt;
712	}	1260	}
		1261
		1262	FUBd;
713	}	1263	}
714		1264
715	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 -1;
		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;
		1542	}
		1543
		1544	static signed char
		1545	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1546	{
		1547	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1548	: a->inode < b->inode ? -1
		1549	: a->inode > b->inode ? 1
		1550	: 0;
		1551	}
		1552
		1553	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1554
		1555	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1556	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1557
		1558	static void
		1559	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1560	{
		1561	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
		1562	unsigned char *bit = bits;
		1563
		1564	assert (CHAR_BIT == 8);
		1565	assert (sizeof (eio_dirent) * 8 < 256);
		1566	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1567	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1568
		1569	if (size <= EIO_SORT_FAST)
		1570	return;
		1571
		1572	/* first prepare an array of bits to test in our radix sort */
		1573	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
		1574	/* inode_bits must contain all inodes ORed together */
		1575	/* which is used to skip bits that are 0 everywhere, which is very common */
		1576	{
		1577	eio_ino_t endianness;
		1578	int i, j;
		1579
		1580	/* we store the byte offset of byte n into byte n of "endianness" */
		1581	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1582	((unsigned char *)&endianness)[i] = i;
		1583
		1584	*bit++ = 0;
		1585
		1586	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1587	{
		1588	/* shifting off the byte offsets out of "endianness" */
		1589	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1590	endianness >>= 8;
		1591
		1592	for (j = 0; j < 8; ++j)
		1593	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
		1594	*bit++ = offs + j;
		1595	}
		1596
		1597	for (j = 0; j < 8; ++j)
		1598	if (score_bits & (1 << j))
		1599	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1600	}
		1601
		1602	/* now actually do the sorting (a variant of MSD radix sort) */
		1603	{
		1604	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
		1605	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
		1606	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
		1607	int stk_idx = 0;
		1608
		1609	base_stk [stk_idx] = dents;
		1610	end_stk [stk_idx] = dents + size;
		1611	bit_stk [stk_idx] = bit - 1;
		1612
		1613	do
		1614	{
		1615	base = base_stk [stk_idx];
		1616	end = end_stk [stk_idx];
		1617	bit = bit_stk [stk_idx];
		1618
		1619	for (;;)
		1620	{
		1621	unsigned char O = *bit >> 3;
		1622	unsigned char M = 1 << (*bit & 7);
		1623
		1624	eio_dirent *a = base;
		1625	eio_dirent *b = end;
		1626
		1627	if (b - a < EIO_SORT_CUTOFF)
		1628	break;
		1629
		1630	/* now bit-partition the array on the bit */
		1631	/* this ugly asymmetric loop seems to perform much better than typical */
		1632	/* partition algos found in the literature */
		1633	do
		1634	if (!(((unsigned char *)a)[O] & M))
		1635	++a;
		1636	else if (!(((unsigned char *)--b)[O] & M))
		1637	{
		1638	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1639	++a;
		1640	}
		1641	while (b > a);
		1642
		1643	/* next bit, or stop, if no bits left in this path */
		1644	if (!*--bit)
		1645	break;
		1646
		1647	base_stk [stk_idx] = a;
		1648	end_stk [stk_idx] = end;
		1649	bit_stk [stk_idx] = bit;
		1650	++stk_idx;
		1651
		1652	end = a;
		1653	}
		1654	}
		1655	while (stk_idx--);
		1656	}
		1657	}
		1658
		1659	static void
		1660	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1661	{
		1662	/* first move the smallest element to the front, to act as a sentinel */
		1663	{
		1664	int i;
		1665	eio_dirent *min = dents;
		1666
		1667	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1668	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1669	if (EIO_DENT_CMP (dents [i], <, *min))
		1670	min = &dents [i];
		1671
		1672	/* swap elements 0 and j (minimum) */
		1673	{
		1674	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1675	}
		1676	}
		1677
		1678	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1679	{
		1680	eio_dirent *i, *j;
		1681
		1682	for (i = dents + 1; i < dents + size; ++i)
		1683	{
		1684	eio_dirent value = *i;
		1685
		1686	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1687	j [1] = j [0];
		1688
		1689	j [1] = value;
		1690	}
		1691	}
		1692	}
		1693
		1694	static void
		1695	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1696	{
		1697	if (size <= 1)
		1698	return; /* our insertion sort relies on size > 0 */
		1699
		1700	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1701	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1702	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1703
		1704	/* use an insertion sort at the end, or for small arrays, */
		1705	/* as insertion sort is more efficient for small partitions */
		1706	eio_dent_insertion_sort (dents, size);
716	}	1707	}
717		1708
718	/* read a full directory */	1709	/* read a full directory */
719	static void	1710	static void
720	eio__scandir (eio_req *req, worker *self)	1711	eio__scandir (eio_req *req, etp_worker *self)
721	{	1712	{
		1713	char *name, *names;
		1714	int namesalloc = 4096 - sizeof (void *) * 4;
		1715	int namesoffs = 0;
		1716	int flags = req->int1;
		1717	eio_dirent *dents = 0;
		1718	int dentalloc = 128;
		1719	int dentoffs = 0;
		1720	eio_ino_t inode_bits = 0;
		1721	#ifdef _WIN32
		1722	HANDLE dirp;
		1723	WIN32_FIND_DATA entp;
		1724	#else
722	DIR *dirp;	1725	DIR *dirp;
723	union
724	{
725	EIO_STRUCT_DIRENT d;
726	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
727	} *u;
728	EIO_STRUCT_DIRENT *entp;	1726	EIO_STRUCT_DIRENT *entp;
729	char *name, *names;	1727	#endif
730	int memlen = 4096;
731	int memofs = 0;
732	int res = 0;
733		1728
734	X_LOCK (wrklock);	1729	req->result = -1;
735	self->dirp = dirp = opendir (req->ptr1);
736	self->dbuf = u = malloc (sizeof (*u));
737	req->flags |= EIO_FLAG_PTR2_FREE;
738	req->ptr2 = names = malloc (memlen);
739	X_UNLOCK (wrklock);
740		1730
741	if (dirp && u && names)	1731	if (!(flags & EIO_READDIR_DENTS))
742	for (;;)	1732	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
		1733
		1734	#ifdef _WIN32
		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)
743	{	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
		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
		1806	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1807	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
		1808	req->ptr2 = names = malloc (namesalloc);
		1809
		1810	if (!names || (flags && !dents))
		1811	return;
		1812
		1813	for (;;)
		1814	{
		1815	int done;
		1816
		1817	#ifdef _WIN32
		1818	done = !dirp;
		1819	#else
744	errno = 0;	1820	errno = 0;
745	readdir_r (dirp, &u->d, &entp);	1821	entp = readdir (dirp);
		1822	done = !entp;
		1823	#endif
746		1824
		1825	if (done)
		1826	{
		1827	#ifndef _WIN32
		1828	int old_errno = errno;
		1829	closedir (dirp);
		1830	errno = old_errno;
		1831
747	if (!entp)	1832	if (errno)
		1833	break;
		1834	#endif
		1835
		1836	/* sort etc. */
		1837	req->int1 = flags;
		1838	req->result = dentoffs;
		1839
		1840	if (flags & EIO_READDIR_STAT_ORDER)
		1841	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1842	else if (flags & EIO_READDIR_DIRS_FIRST)
		1843	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1844	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		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
748	break;	1869	break;
		1870	}
749		1871
750	name = entp->d_name;	1872	/* now add the entry to our list(s) */
		1873	name = D_NAME (entp);
751		1874
		1875	/* skip . and .. entries */
752	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1876	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
753	{	1877	{
754	int len = strlen (name) + 1;	1878	int len = D_NAMLEN (entp) + 1;
755		1879
756	res++;	1880	while (ecb_expect_false (namesoffs + len > namesalloc))
757
758	while (memofs + len > memlen)
759	{	1881	{
760	memlen *= 2;	1882	namesalloc *= 2;
761	X_LOCK (wrklock);
762	req->ptr2 = names = realloc (names, memlen);	1883	req->ptr2 = names = realloc (names, namesalloc);
763	X_UNLOCK (wrklock);
764		1884
765	if (!names)	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))
		1896	{
		1897	dentalloc *= 2;
		1898	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1899
		1900	if (!dents)
766	break;	1901	break;
767	}	1902	}
768		1903
769	memcpy (names + memofs, name, len);	1904	ent = dents + dentoffs;
770	memofs += len;	1905
		1906	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1907	ent->namelen = len - 1;
		1908	ent->inode = D_INO (entp);
		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)
		1968	{
		1969	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1970	ent->score = 1;
		1971	else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
		1972	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1973	}
		1974	else if (ent->type == EIO_DT_DIR)
		1975	ent->score = 0;
		1976	}
771	}	1977	}
		1978
		1979	namesoffs += len;
		1980	++dentoffs;
772	}	1981	}
773		1982
774	if (errno)	1983	if (EIO_CANCELLED (req))
775	res = -1;	1984	{
		1985	errno = ECANCELED;
		1986	break;
		1987	}
		1988
		1989	#ifdef _WIN32
		1990	if (!FindNextFile (dirp, &entp))
		1991	{
		1992	FindClose (dirp);
		1993	dirp = 0;
		1994	}
		1995	#endif
		1996	}
		1997	}
		1998
		1999	/*****************************************************************************/
		2000	/* working directory stuff */
		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. */
		2004
		2005	#if !HAVE_AT
		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)
		2011	{
		2012	if (!wd || *path == '/')
		2013	return path;
		2014
		2015	if (path [0] == '.' && !path [1])
		2016	return wd->str;
		2017
776		2018	{
777	req->result = res;	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	}
778	}	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)
		2078	{
		2079	#if HAVE_AT
		2080	close (wd->fd);
		2081	#endif
		2082	free (wd);
		2083	}
		2084	}
		2085
		2086	#if HAVE_AT
		2087
		2088	/* they forgot these */
		2089
		2090	static int
		2091	eio__truncateat (int dirfd, const char *path, off_t length)
		2092	{
		2093	int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
		2094	int res;
		2095
		2096	if (fd < 0)
		2097	return fd;
		2098
		2099	res = ftruncate (fd, length);
		2100	close (fd);
		2101	return res;
		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
779		2120
780	/*****************************************************************************/	2121	/*****************************************************************************/
781		2122
782	#define ALLOC(len) \	2123	#define ALLOC(len) \
783	if (!req->ptr2) \	2124	if (!req->ptr2) \
…		…
792	req->result = -1; \	2133	req->result = -1; \
793	break; \	2134	break; \
794	} \	2135	} \
795	}	2136	}
796		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	#endif
		2153	}
		2154
797	X_THREAD_PROC (eio_proc)	2155	X_THREAD_PROC (etp_proc)
798	{	2156	{
799	eio_req *req;	2157	ETP_REQ *req;
800	struct timespec ts;	2158	struct timespec ts;
801	worker *self = (worker *)thr_arg;	2159	etp_worker *self = (etp_worker *)thr_arg;
802		2160
		2161	etp_proc_init ();
		2162
803	/* try to distribute timeouts somewhat randomly */	2163	/* try to distribute timeouts somewhat evenly */
804	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	2164	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
805		2165
806	for (;;)	2166	for (;;)
807	{	2167	{
808	ts.tv_sec = time (0) + IDLE_TIMEOUT;	2168	ts.tv_sec = 0;
809		2169
810	X_LOCK (reqlock);	2170	X_LOCK (reqlock);
811		2171
812	for (;;)	2172	for (;;)
813	{	2173	{
814	self->req = req = reqq_shift (&req_queue);	2174	req = reqq_shift (&req_queue);
815		2175
816	if (req)	2176	if (req)
817	break;	2177	break;
818		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
819	++idle;	2188	++idle;
820		2189
821	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	2190	if (idle <= max_idle)
		2191	/* we are allowed to idle, so do so without any timeout */
		2192	X_COND_WAIT (reqwait, reqlock);
		2193	else
822	{	2194	{
823	if (idle > max_idle)	2195	/* initialise timeout once */
824	{	2196	if (!ts.tv_sec)
825	--idle;	2197	ts.tv_sec = time (0) + idle_timeout;
826	X_UNLOCK (reqlock);
827	X_LOCK (wrklock);
828	--started;
829	X_UNLOCK (wrklock);
830	goto quit;
831	}
832		2198
833	/* we are allowed to idle, so do so without any timeout */
834	X_COND_WAIT (reqwait, reqlock);	2199	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
835	ts.tv_sec = time (0) + IDLE_TIMEOUT;	2200	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
836	}	2201	}
837		2202
838	--idle;	2203	--idle;
839	}	2204	}
840		2205
841	--nready;	2206	--nready;
842		2207
843	X_UNLOCK (reqlock);	2208	X_UNLOCK (reqlock);
844		2209
845	errno = 0; /* strictly unnecessary */
846
847	if (!EIO_CANCELLED (req))
848	switch (req->type)	2210	if (req->type < 0)
849	{
850	case EIO_READ: ALLOC (req->size);
851	req->result = req->offs >= 0
852	? pread (req->int1, req->ptr2, req->size, req->offs)
853	: read (req->int1, req->ptr2, req->size); break;
854	case EIO_WRITE: req->result = req->offs >= 0
855	? pwrite (req->int1, req->ptr2, req->size, req->offs)
856	: write (req->int1, req->ptr2, req->size); break;
857
858	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
859	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
860
861	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
862	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
863	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
864	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
865	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
866	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
867
868	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
869	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
870	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
871	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
872	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
873	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
874
875	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
876	case EIO_CLOSE: req->result = close (req->int1); break;
877	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
878	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
879	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
880	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
881	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
882	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
883	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
884	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
885
886	case EIO_READLINK: ALLOC (NAME_MAX);
887	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
888
889	case EIO_SYNC: req->result = 0; sync (); break;
890	case EIO_FSYNC: req->result = fsync (req->int1); break;
891	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
892
893	case EIO_READDIR: eio__scandir (req, self); break;
894
895	case EIO_BUSY:
896	#ifdef _WIN32
897	Sleep (req->nv1 * 1000.);
898	#else
899	{
900	struct timeval tv;
901
902	tv.tv_sec = req->nv1;
903	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
904
905	req->result = select (0, 0, 0, 0, &tv);
906	}
907	#endif
908	break;
909
910	case EIO_UTIME:
911	case EIO_FUTIME:
912	{
913	struct timeval tv[2];
914	struct timeval *times;
915
916	if (req->nv1 != -1. || req->nv2 != -1.)
917	{
918	tv[0].tv_sec = req->nv1;
919	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
920	tv[1].tv_sec = req->nv2;
921	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
922
923	times = tv;
924	}
925	else
926	times = 0;
927
928
929	req->result = req->type == EIO_FUTIME
930	? futimes (req->int1, times)
931	: utimes (req->ptr1, times);
932	}
933
934	case EIO_GROUP:
935	case EIO_NOP:
936	req->result = 0;
937	break;
938
939	case EIO_QUIT:
940	goto quit;	2211	goto quit;
941		2212
942	default:	2213	ETP_EXECUTE (self, req);
943	req->result = -1;
944	break;
945	}
946
947	req->errorno = errno;
948		2214
949	X_LOCK (reslock);	2215	X_LOCK (reslock);
950		2216
951	++npending;	2217	++npending;
952		2218
953	if (!reqq_push (&res_queue, req) && want_poll_cb)	2219	if (!reqq_push (&res_queue, req) && want_poll_cb)
954	want_poll_cb ();	2220	want_poll_cb ();
955		2221
956	self->req = 0;
957	worker_clear (self);	2222	etp_worker_clear (self);
958		2223
959	X_UNLOCK (reslock);	2224	X_UNLOCK (reslock);
960	}	2225	}
961		2226
962	quit:	2227	quit:
		2228	free (req);
		2229
963	X_LOCK (wrklock);	2230	X_LOCK (wrklock);
964	worker_free (self);	2231	etp_worker_free (self);
965	X_UNLOCK (wrklock);	2232	X_UNLOCK (wrklock);
966		2233
967	return 0;	2234	return 0;
968	}	2235	}
969		2236
970	/*****************************************************************************/	2237	/*****************************************************************************/
971		2238
972	static void eio_atfork_prepare (void)	2239	int ecb_cold
		2240	eio_init (void (*want_poll)(void), void (*done_poll)(void))
973	{	2241	{
974	X_LOCK (wrklock);
975	X_LOCK (reqlock);
976	X_LOCK (reslock);
977	#if !HAVE_PREADWRITE	2242	#if !HAVE_PREADWRITE
978	X_LOCK (preadwritelock);
979	#endif
980	#if !HAVE_READDIR_R
981	X_LOCK (readdirlock);
982	#endif
983	}
984
985	static void eio_atfork_parent (void)
986	{
987	#if !HAVE_READDIR_R
988	X_UNLOCK (readdirlock);
989	#endif
990	#if !HAVE_PREADWRITE
991	X_UNLOCK (preadwritelock);
992	#endif
993	X_UNLOCK (reslock);
994	X_UNLOCK (reqlock);
995	X_UNLOCK (wrklock);
996	}
997
998	static void eio_atfork_child (void)
999	{
1000	eio_req *prv;
1001
1002	while (prv = reqq_shift (&req_queue))
1003	eio_destroy (prv);
1004
1005	while (prv = reqq_shift (&res_queue))
1006	eio_destroy (prv);
1007
1008	while (wrk_first.next != &wrk_first)
1009	{
1010	worker *wrk = wrk_first.next;
1011
1012	if (wrk->req)
1013	eio_destroy (wrk->req);
1014
1015	worker_clear (wrk);
1016	worker_free (wrk);
1017	}
1018
1019	started = 0;
1020	idle = 0;
1021	nreqs = 0;
1022	nready = 0;
1023	npending = 0;
1024
1025	eio_atfork_parent ();
1026	}
1027
1028	int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1029	{
1030	want_poll_cb = want_poll;
1031	done_poll_cb = done_poll;
1032
1033	#ifdef _WIN32
1034	X_MUTEX_CHECK (wrklock);
1035	X_MUTEX_CHECK (reslock);
1036	X_MUTEX_CHECK (reqlock);
1037	X_MUTEX_CHECK (reqwait);
1038	X_MUTEX_CHECK (preadwritelock);	2243	X_MUTEX_CREATE (preadwritelock);
1039	X_MUTEX_CHECK (readdirlock);
1040
1041	X_COND_CHECK (reqwait);
1042	#endif	2244	#endif
1043		2245
1044	X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);	2246	return etp_init (want_poll, done_poll);
1045	}	2247	}
1046		2248
		2249	ecb_inline void
1047	static void eio_api_destroy (eio_req *req)	2250	eio_api_destroy (eio_req *req)
1048	{	2251	{
1049	free (req);	2252	free (req);
1050	}	2253	}
1051		2254
1052	#define REQ(rtype) \	2255	#define REQ(rtype) \
…		…
1071	{ \	2274	{ \
1072	eio_api_destroy (req); \	2275	eio_api_destroy (req); \
1073	return 0; \	2276	return 0; \
1074	}	2277	}
1075		2278
		2279	static void
		2280	eio_execute (etp_worker *self, eio_req *req)
		2281	{
		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	}
		2310
		2311	switch (req->type)
		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
		2319	case EIO_READ: ALLOC (req->size);
		2320	req->result = req->offs >= 0
		2321	? pread (req->int1, req->ptr2, req->size, req->offs)
		2322	: read (req->int1, req->ptr2, req->size); break;
		2323	case EIO_WRITE: req->result = req->offs >= 0
		2324	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		2325	: write (req->int1, req->ptr2, req->size); break;
		2326
		2327	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		2328	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
		2329
		2330	#if HAVE_AT
		2331
		2332	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2333	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
		2334	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2335	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
		2336	case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
		2337	case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
		2338	case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
		2339	case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
		2340
		2341	case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
		2342	case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
		2343	case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
		2344	case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
		2345	case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
		2346	case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
		2347	case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2348	case EIO_READLINK: ALLOC (PATH_MAX);
		2349	req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
		2350	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2351	req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2352	case EIO_UTIME:
		2353	case EIO_FUTIME:
		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
		2376	#else
		2377
		2378	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		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;
		2386
		2387	case EIO_UNLINK: req->result = unlink (path ); break;
		2388	case EIO_RMDIR: req->result = rmdir (path ); break;
		2389	case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); 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;
		2398
		2399	case EIO_UTIME:
		2400	case EIO_FUTIME:
		2401	{
		2402	struct timeval tv[2];
		2403	struct timeval *times;
		2404
		2405	if (req->nv1 != -1. || req->nv2 != -1.)
		2406	{
		2407	tv[0].tv_sec = req->nv1;
		2408	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
		2409	tv[1].tv_sec = req->nv2;
		2410	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
		2411
		2412	times = tv;
		2413	}
		2414	else
		2415	times = 0;
		2416
		2417	req->result = req->type == EIO_FUTIME
		2418	? futimes (req->int1, times)
		2419	: utimes (req->ptr1, times);
		2420	}
		2421	break;
		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
		2472	case EIO_GROUP:
		2473	abort (); /* handled in eio_request */
		2474
		2475	case EIO_NOP:
		2476	req->result = 0;
		2477	break;
		2478
		2479	case EIO_CUSTOM:
		2480	req->feed (req);
		2481	break;
		2482
		2483	default:
		2484	errno = ENOSYS;
		2485	req->result = -1;
		2486	break;
		2487	}
		2488
		2489	req->errorno = errno;
		2490	}
		2491
		2492	#ifndef EIO_NO_WRAPPERS
		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
1076	eio_req *eio_nop (int pri, eio_cb cb, void *data)	2504	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1077	{	2505	{
1078	REQ (EIO_NOP); SEND;	2506	REQ (EIO_NOP); SEND;
1079	}	2507	}
1080		2508
…		…
1091	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	2519	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1092	{	2520	{
1093	REQ (EIO_FSYNC); req->int1 = fd; SEND;	2521	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1094	}	2522	}
1095		2523
		2524	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2525	{
		2526	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2527	}
		2528
1096	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2529	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1097	{	2530	{
1098	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2531	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1099	}	2532	}
1100		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
		2544	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2545	{
		2546	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2547	}
		2548
		2549	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2550	{
		2551	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2552	}
		2553
		2554	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2555	{
		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;
		2562	}
		2563
1101	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)
1102	{	2565	{
1103	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2566	REQ (EIO_CLOSE); req->int1 = fd; SEND;
1104	}	2567	}
1105		2568
…		…
1121	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)
1122	{	2585	{
1123	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2586	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1124	}	2587	}
1125		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
1126	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)
1127	{	2595	{
1128	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;
1129	}	2597	}
1130		2598
…		…
1136	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)
1137	{	2605	{
1138	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2606	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1139	}	2607	}
1140		2608
1141	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)
1142	{	2610	{
1143	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;
1144	}	2612	}
1145		2613
1146	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)
…		…
1166	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)
1167	{	2635	{
1168	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2636	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1169	}	2637	}
1170		2638
1171	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)
1172	{	2640	{
1173	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;
1174	}	2642	}
1175		2643
1176	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)
…		…
1192	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)
1193	{	2661	{
1194	return eio__1path (EIO_READLINK, path, pri, cb, data);	2662	return eio__1path (EIO_READLINK, path, pri, cb, data);
1195	}	2663	}
1196		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
1197	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)
1198	{	2671	{
1199	return eio__1path (EIO_STAT, path, pri, cb, data);	2672	return eio__1path (EIO_STAT, path, pri, cb, data);
1200	}	2673	}
1201		2674
1202	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)
1203	{	2676	{
1204	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2677	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1205	}	2678	}
1206		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
1207	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)
1208	{	2686	{
1209	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2687	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1210	}	2688	}
1211		2689
1212	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	2690	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1213	{	2691	{
1214	return eio__1path (EIO_RMDIR, path, pri, cb, data);	2692	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1215	}	2693	}
1216		2694
1217	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	2695	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1218	{	2696	{
1219	return eio__1path (EIO_READDIR, path, pri, cb, data);	2697	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1220	}	2698	}
1221		2699
1222	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)
1223	{	2701	{
1224	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	2702	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1225	}	2703	}
1226		2704
1227	static eio_req *	2705	static eio_req *
1228	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)
1229	{	2707	{
…		…
1253	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)
1254	{	2732	{
1255	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2733	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1256	}	2734	}
1257		2735
		2736	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
		2737	{
		2738	REQ (EIO_CUSTOM); req->feed = execute; SEND;
		2739	}
		2740
		2741	#endif
		2742
1258	eio_req *eio_grp (eio_cb cb, void *data)	2743	eio_req *eio_grp (eio_cb cb, void *data)
1259	{	2744	{
1260	const int pri = EIO_PRI_MAX;	2745	const int pri = EIO_PRI_MAX;
1261		2746
1262	REQ (EIO_GROUP); SEND;	2747	REQ (EIO_GROUP); SEND;
…		…
1267	#undef SEND	2752	#undef SEND
1268		2753
1269	/*****************************************************************************/	2754	/*****************************************************************************/
1270	/* grp functions */	2755	/* grp functions */
1271		2756
		2757	void
1272	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)
1273	{	2759	{
1274	grp->int2 = limit;	2760	grp->int2 = limit;
1275	grp->feed = feed;	2761	grp->feed = feed;
1276		2762
1277	grp_try_feed (grp);	2763	grp_try_feed (grp);
1278	}	2764	}
1279		2765
		2766	void
1280	void eio_grp_limit (eio_req *grp, int limit)	2767	eio_grp_limit (eio_req *grp, int limit)
1281	{	2768	{
1282	grp->int2 = limit;	2769	grp->int2 = limit;
1283		2770
1284	grp_try_feed (grp);	2771	grp_try_feed (grp);
1285	}	2772	}
1286		2773
		2774	void
1287	void eio_grp_add (eio_req *grp, eio_req *req)	2775	eio_grp_add (eio_req *grp, eio_req *req)
1288	{	2776	{
1289	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));
		2778
		2779	grp->flags |= EIO_FLAG_GROUPADD;
1290		2780
1291	++grp->size;	2781	++grp->size;
1292	req->grp = grp;	2782	req->grp = grp;
1293		2783
1294	req->grp_prev = 0;	2784	req->grp_prev = 0;
…		…
1301	}	2791	}
1302		2792
1303	/*****************************************************************************/	2793	/*****************************************************************************/
1304	/* misc garbage */	2794	/* misc garbage */
1305		2795
		2796	eio_ssize_t
1306	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)
1307	{	2798	{
1308	worker wrk;
1309
1310	wrk.dbuf = 0;
1311
1312	eio__sendfile (ofd, ifd, offset, count, &wrk);	2799	return eio__sendfile (ofd, ifd, offset, count);
1313
1314	if (wrk.dbuf)
1315	free (wrk.dbuf);
1316	}	2800	}
1317		2801

Diff Legend

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