ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.5 by root, Sun May 11 01:08:05 2008 UTC vs.
Revision 1.99 by root, Tue Jul 26 11:07:08 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 XTHREAD_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>
…		…
8	#include <errno.h>	56	#include <errno.h>
9	#include <sys/types.h>	57	#include <sys/types.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 <sched.h>	61	#include <assert.h>
		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);
14		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
…		…
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	#ifdef EIO_STRUCT_STATI64
		111	#define stat(path,buf) _stati64 (path,buf)
		112	#define fstat(fd,buf) _fstati64 (fd,buf)
		113	#endif
		114	#define lstat(path,buf) stat (path,buf)
		115	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		116	#define mkdir(path,mode) _mkdir (path)
		117	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		118
		119	#define chmod(path,mode) _chmod (path, mode)
		120	#define dup(fd) _dup (fd)
		121	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		122
		123	#define fchmod(fd,mode) EIO_ENOSYS ()
		124	#define chown(path,uid,gid) EIO_ENOSYS ()
		125	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		126	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		127	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		128	#define mknod(path,mode,dev) EIO_ENOSYS ()
		129	#define sync() EIO_ENOSYS ()
		130	#define readlink(path,buf,s) EIO_ENOSYS ()
		131	#define statvfs(path,buf) EIO_ENOSYS ()
		132	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		133
		134	/* rename() uses MoveFile, which fails to overwrite */
		135	#define rename(old,neu) eio__rename (old, neu)
		136
		137	static int
		138	eio__rename (const char *old, const char *neu)
		139	{
		140	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		141	return 0;
		142
		143	/* should steal _dosmaperr */
		144	switch (GetLastError ())
		145	{
		146	case ERROR_FILE_NOT_FOUND:
		147	case ERROR_PATH_NOT_FOUND:
		148	case ERROR_INVALID_DRIVE:
		149	case ERROR_NO_MORE_FILES:
		150	case ERROR_BAD_NETPATH:
		151	case ERROR_BAD_NET_NAME:
		152	case ERROR_BAD_PATHNAME:
		153	case ERROR_FILENAME_EXCED_RANGE:
		154	errno = ENOENT;
		155	break;
		156
		157	default:
		158	errno = EACCES;
		159	break;
		160	}
		161
		162	return -1;
		163	}
		164
		165	/* we could even stat and see if it exists */
		166	static int
		167	symlink (const char *old, const char *neu)
		168	{
		169	#if WINVER >= 0x0600
		170	if (CreateSymbolicLink (neu, old, 1))
		171	return 0;
		172
		173	if (CreateSymbolicLink (neu, old, 0))
		174	return 0;
		175	#endif
		176
		177	return EIO_ERRNO (ENOENT, -1);
		178	}
		179
		180	/* POSIX API only */
		181	#define CreateHardLink(neu,old,flags) 0
		182	#define CreateSymbolicLink(neu,old,flags) 0
		183
		184	struct statvfs
		185	{
		186	int dummy;
		187	};
		188
		189	#define DT_DIR EIO_DT_DIR
		190	#define DT_REG EIO_DT_REG
		191	#define D_NAME(entp) entp.cFileName
		192	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
30		193
31	#else	194	#else
32		195
33	# include "config.h"
34	# include <sys/time.h>	196	#include <sys/time.h>
35	# include <sys/select.h>	197	#include <sys/select.h>
		198	#include <sys/statvfs.h>
36	# include <unistd.h>	199	#include <unistd.h>
		200	#include <signal.h>
		201	#include <dirent.h>
		202
		203	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		204	#include <sys/mman.h>
		205	#endif
		206
		207	#define D_NAME(entp) entp->d_name
		208
		209	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		210	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		211	#define _DIRENT_HAVE_D_TYPE /* sigh */
		212	#define D_INO(de) (de)->d_fileno
		213	#define D_NAMLEN(de) (de)->d_namlen
		214	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		215	#define D_INO(de) (de)->d_ino
		216	#endif
		217
		218	#ifdef _D_EXACT_NAMLEN
		219	#undef D_NAMLEN
		220	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		221	#endif
		222
		223	#ifdef _DIRENT_HAVE_D_TYPE
		224	#define D_TYPE(de) (de)->d_type
		225	#endif
		226
		227	#ifndef EIO_STRUCT_DIRENT
		228	#define EIO_STRUCT_DIRENT struct dirent
		229	#endif
		230
		231	#endif
		232
		233	#if HAVE_UTIMES
37	# include <utime.h>	234	# 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	235	#endif
44		236
45	#endif	237	#if HAVE_SYS_SYSCALL_H
46		238	# include <sys/syscall.h>
47	# ifndef EIO_STRUCT_STAT
48	# define EIO_STRUCT_STAT struct stat
49	# endif	239	#endif
		240
		241	#if HAVE_SYS_PRCTL_H
		242	# include <sys/prctl.h>
		243	#endif
50		244
51	#if HAVE_SENDFILE	245	#if HAVE_SENDFILE
52	# if __linux	246	# if __linux
53	# include <sys/sendfile.h>	247	# include <sys/sendfile.h>
54	# elif __freebsd	248	# elif __FreeBSD__ || defined __APPLE__
55	# include <sys/socket.h>	249	# include <sys/socket.h>
56	# include <sys/uio.h>	250	# include <sys/uio.h>
57	# elif __hpux	251	# elif __hpux
58	# include <sys/socket.h>	252	# include <sys/socket.h>
59	# elif __solaris /* not yet */	253	# elif __solaris
60	# include <sys/sendfile.h>	254	# include <sys/sendfile.h>
61	# else	255	# else
62	# error sendfile support requested but not available	256	# error sendfile support requested but not available
63	# endif	257	# endif
64	#endif	258	#endif
65		259
66	/* number of seconds after which an idle threads exit */	260	#ifndef D_TYPE
67	#define IDLE_TIMEOUT 10	261	# define D_TYPE(de) 0
		262	#endif
		263	#ifndef D_INO
		264	# define D_INO(de) 0
		265	#endif
		266	#ifndef D_NAMLEN
		267	# define D_NAMLEN(entp) strlen (D_NAME (entp))
		268	#endif
68		269
69	/* used for struct dirent, AIX doesn't provide it */	270	/* used for struct dirent, AIX doesn't provide it */
70	#ifndef NAME_MAX	271	#ifndef NAME_MAX
71	# define NAME_MAX 4096	272	# define NAME_MAX 4096
72	#endif	273	#endif
73		274
		275	/* used for readlink etc. */
		276	#ifndef PATH_MAX
		277	# define PATH_MAX 4096
		278	#endif
		279
74	/* buffer size for various temporary buffers */	280	/* buffer size for various temporary buffers */
75	#define EIO_BUFSIZE 65536	281	#define EIO_BUFSIZE 65536
76		282
77	#define dBUF \	283	#define dBUF \
78	char *eio_buf; \
79	X_LOCK (wrklock); \
80	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	284	char *eio_buf = malloc (EIO_BUFSIZE); \
81	X_UNLOCK (wrklock); \	285	errno = ENOMEM; \
82	if (!eio_buf) \	286	if (!eio_buf) \
83	return -1;	287	return -1
		288
		289	#define FUBd \
		290	free (eio_buf)
84		291
85	#define EIO_TICKS ((1000000 + 1023) >> 10)	292	#define EIO_TICKS ((1000000 + 1023) >> 10)
		293
		294	#define ETP_PRI_MIN EIO_PRI_MIN
		295	#define ETP_PRI_MAX EIO_PRI_MAX
		296
		297	struct etp_worker;
		298
		299	#define ETP_REQ eio_req
		300	#define ETP_DESTROY(req) eio_destroy (req)
		301	static int eio_finish (eio_req *req);
		302	#define ETP_FINISH(req) eio_finish (req)
		303	static void eio_execute (struct etp_worker *self, eio_req *req);
		304	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		305
		306	/*****************************************************************************/
		307
		308	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		309
		310	/* calculate time difference in ~1/EIO_TICKS of a second */
		311	ecb_inline int
		312	tvdiff (struct timeval *tv1, struct timeval *tv2)
		313	{
		314	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
		315	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
		316	}
		317
		318	static unsigned int started, idle, wanted = 4;
86		319
87	static void (*want_poll_cb) (void);	320	static void (*want_poll_cb) (void);
88	static void (*done_poll_cb) (void);	321	static void (*done_poll_cb) (void);
89		322
90	static unsigned int max_poll_time = 0;	323	static unsigned int max_poll_time; /* reslock */
91	static unsigned int max_poll_reqs = 0;	324	static unsigned int max_poll_reqs; /* reslock */
92		325
93	/* calculcate time difference in ~1/EIO_TICKS of a second */	326	static volatile unsigned int nreqs; /* reqlock */
94	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	327	static volatile unsigned int nready; /* reqlock */
95	{	328	static volatile unsigned int npending; /* reqlock */
96	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	329	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	330	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98	}
99		331
100	static unsigned int started, idle, wanted;	332	static xmutex_t wrklock;
		333	static xmutex_t reslock;
		334	static xmutex_t reqlock;
		335	static xcond_t reqwait;
		336
		337	#if !HAVE_PREADWRITE
		338	/*
		339	* make our pread/pwrite emulation safe against themselves, but not against
		340	* normal read/write by using a mutex. slows down execution a lot,
		341	* but that's your problem, not mine.
		342	*/
		343	static xmutex_t preadwritelock;
		344	#endif
		345
		346	typedef struct etp_worker
		347	{
		348	/* locked by wrklock */
		349	struct etp_worker *prev, *next;
		350
		351	xthread_t tid;
		352
		353	/* locked by reslock, reqlock or wrklock */
		354	ETP_REQ *req; /* currently processed request */
		355
		356	#ifdef ETP_WORKER_COMMON
		357	ETP_WORKER_COMMON
		358	#endif
		359	} etp_worker;
		360
		361	static etp_worker wrk_first; /* NOT etp */
		362
		363	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		364	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
101		365
102	/* worker threads management */	366	/* worker threads management */
103	static mutex_t wrklock = X_MUTEX_INIT;
104		367
105	typedef struct worker {	368	static void ecb_cold
106	/* locked by wrklock */
107	struct worker *prev, *next;
108
109	thread_t tid;
110
111	/* locked by reslock, reqlock or wrklock */
112	eio_req *req; /* currently processed request */
113	void *dbuf;
114	DIR *dirp;
115	} worker;
116
117	static worker wrk_first = { &wrk_first, &wrk_first, 0 };
118
119	static void worker_clear (worker *wrk)	369	etp_worker_clear (etp_worker *wrk)
120	{	370	{
121	if (wrk->dirp)
122	{
123	closedir (wrk->dirp);
124	wrk->dirp = 0;
125	}
126
127	if (wrk->dbuf)
128	{
129	free (wrk->dbuf);
130	wrk->dbuf = 0;
131	}
132	}	371	}
133		372
134	static void worker_free (worker *wrk)	373	static void ecb_cold
		374	etp_worker_free (etp_worker *wrk)
135	{	375	{
136	wrk->next->prev = wrk->prev;	376	wrk->next->prev = wrk->prev;
137	wrk->prev->next = wrk->next;	377	wrk->prev->next = wrk->next;
138		378
139	free (wrk);	379	free (wrk);
140	}	380	}
141		381
142	static volatile unsigned int nreqs, nready, npending;	382	static unsigned int
143	static volatile unsigned int max_idle = 4;	383	etp_nreqs (void)
144
145	static mutex_t reslock = X_MUTEX_INIT;
146	static mutex_t reqlock = X_MUTEX_INIT;
147	static cond_t reqwait = X_COND_INIT;
148
149	unsigned int eio_nreqs (void)
150	{	384	{
		385	int retval;
		386	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		387	retval = nreqs;
		388	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151	return nreqs;	389	return retval;
152	}	390	}
153		391
154	unsigned int eio_nready (void)	392	static unsigned int
		393	etp_nready (void)
155	{	394	{
156	unsigned int retval;	395	unsigned int retval;
157		396
158	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	397	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159	retval = nready;	398	retval = nready;
160	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	399	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161		400
162	return retval;	401	return retval;
163	}	402	}
164		403
165	unsigned int eio_npending (void)	404	static unsigned int
		405	etp_npending (void)
166	{	406	{
167	unsigned int retval;	407	unsigned int retval;
168		408
169	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	409	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170	retval = npending;	410	retval = npending;
171	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	411	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172		412
173	return retval;	413	return retval;
174	}	414	}
175		415
176	unsigned int eio_nthreads (void)	416	static unsigned int
		417	etp_nthreads (void)
177	{	418	{
178	unsigned int retval;	419	unsigned int retval;
179		420
180	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	421	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181	retval = started;	422	retval = started;
…		…
188	* a somewhat faster data structure might be nice, but	429	* a somewhat faster data structure might be nice, but
189	* with 8 priorities this actually needs <20 insns	430	* with 8 priorities this actually needs <20 insns
190	* per shift, the most expensive operation.	431	* per shift, the most expensive operation.
191	*/	432	*/
192	typedef struct {	433	typedef struct {
193	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	434	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194	int size;	435	int size;
195	} reqq;	436	} etp_reqq;
196		437
197	static reqq req_queue;	438	static etp_reqq req_queue;
198	static reqq res_queue;	439	static etp_reqq res_queue;
199		440
200	static int reqq_push (reqq *q, eio_req *req)	441	static void ecb_noinline ecb_cold
		442	reqq_init (etp_reqq *q)
		443	{
		444	int pri;
		445
		446	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		447	q->qs[pri] = q->qe[pri] = 0;
		448
		449	q->size = 0;
		450	}
		451
		452	static int ecb_noinline
		453	reqq_push (etp_reqq *q, ETP_REQ *req)
201	{	454	{
202	int pri = req->pri;	455	int pri = req->pri;
203	req->next = 0;	456	req->next = 0;
204		457
205	if (q->qe[pri])	458	if (q->qe[pri])
…		…
211	q->qe[pri] = q->qs[pri] = req;	464	q->qe[pri] = q->qs[pri] = req;
212		465
213	return q->size++;	466	return q->size++;
214	}	467	}
215		468
216	static eio_req *reqq_shift (reqq *q)	469	static ETP_REQ * ecb_noinline
		470	reqq_shift (etp_reqq *q)
217	{	471	{
218	int pri;	472	int pri;
219		473
220	if (!q->size)	474	if (!q->size)
221	return 0;	475	return 0;
222		476
223	--q->size;	477	--q->size;
224		478
225	for (pri = EIO_NUM_PRI; pri--; )	479	for (pri = ETP_NUM_PRI; pri--; )
226	{	480	{
227	eio_req *req = q->qs[pri];	481	eio_req *req = q->qs[pri];
228		482
229	if (req)	483	if (req)
230	{	484	{
…		…
236	}	490	}
237		491
238	abort ();	492	abort ();
239	}	493	}
240		494
241	static void grp_try_feed (eio_req *grp)	495	static int ecb_cold
		496	etp_init (void (*want_poll)(void), void (*done_poll)(void))
242	{	497	{
243	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	498	X_MUTEX_CREATE (wrklock);
244	{	499	X_MUTEX_CREATE (reslock);
245	int old_len = grp->size;	500	X_MUTEX_CREATE (reqlock);
		501	X_COND_CREATE (reqwait);
246		502
247	EIO_FEED (grp);	503	reqq_init (&req_queue);
		504	reqq_init (&res_queue);
248		505
249	/* stop if no progress has been made */	506	wrk_first.next =
250	if (old_len == grp->size)	507	wrk_first.prev = &wrk_first;
251	{
252	grp->feed = 0;
253	break;
254	}
255	}
256	}
257		508
258	static int eio_finish (eio_req *req);	509	started = 0;
		510	idle = 0;
		511	nreqs = 0;
		512	nready = 0;
		513	npending = 0;
259		514
260	static int grp_dec (eio_req *grp)	515	want_poll_cb = want_poll;
261	{	516	done_poll_cb = done_poll;
262	--grp->size;
263		517
264	/* call feeder, if applicable */
265	grp_try_feed (grp);
266
267	/* finish, if done */
268	if (!grp->size && grp->int1)
269	return eio_finish (grp);
270	else
271	return 0;	518	return 0;
272	}	519	}
273		520
274	void eio_destroy (eio_req *req)
275	{
276	if ((req)->flags & EIO_FLAG_PTR2_FREE)
277	free (req->ptr2);
278
279	EIO_DESTROY (req);
280	}
281
282	static int eio_finish (eio_req *req)
283	{
284	int res = EIO_FINISH (req);
285
286	if (req->grp)
287	{
288	int res2;
289	eio_req *grp = req->grp;
290
291	/* unlink request */
292	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
293	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
294
295	if (grp->grp_first == req)
296	grp->grp_first = req->grp_next;
297
298	res2 = grp_dec (grp);
299
300	if (!res && res2)
301	res = res2;
302	}
303
304	eio_destroy (req);
305
306	return res;
307	}
308
309	void eio_grp_cancel (eio_req *grp)
310	{
311	for (grp = grp->grp_first; grp; grp = grp->grp_next)
312	eio_cancel (grp);
313	}
314
315	void eio_cancel (eio_req *req)
316	{
317	req->flags |= EIO_FLAG_CANCELLED;
318
319	eio_grp_cancel (req);
320	}
321
322	X_THREAD_PROC (eio_proc);	521	X_THREAD_PROC (etp_proc);
323		522
324	static void start_thread (void)	523	static void ecb_cold
		524	etp_start_thread (void)
325	{	525	{
326	worker *wrk = calloc (1, sizeof (worker));	526	etp_worker *wrk = calloc (1, sizeof (etp_worker));
327		527
328	if (!wrk)	528	/*TODO*/
329	croak ("unable to allocate worker thread data");	529	assert (("unable to allocate worker thread data", wrk));
330		530
331	X_LOCK (wrklock);	531	X_LOCK (wrklock);
332		532
333	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	533	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334	{	534	{
335	wrk->prev = &wrk_first;	535	wrk->prev = &wrk_first;
336	wrk->next = wrk_first.next;	536	wrk->next = wrk_first.next;
337	wrk_first.next->prev = wrk;	537	wrk_first.next->prev = wrk;
338	wrk_first.next = wrk;	538	wrk_first.next = wrk;
…		…
342	free (wrk);	542	free (wrk);
343		543
344	X_UNLOCK (wrklock);	544	X_UNLOCK (wrklock);
345	}	545	}
346		546
		547	static void
347	static void maybe_start_thread (void)	548	etp_maybe_start_thread (void)
348	{	549	{
349	if (eio_nthreads () >= wanted)	550	if (ecb_expect_true (etp_nthreads () >= wanted))
350	return;	551	return;
351		552
352	/* todo: maybe use idle here, but might be less exact */	553	/* todo: maybe use idle here, but might be less exact */
353	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	554	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354	return;	555	return;
355		556
356	start_thread ();	557	etp_start_thread ();
357	}	558	}
358		559
359	void eio_submit (eio_req *req)	560	static void ecb_cold
360	{	561	etp_end_thread (void)
361	++nreqs;
362
363	X_LOCK (reqlock);
364	++nready;
365	reqq_push (&req_queue, req);
366	X_COND_SIGNAL (reqwait);
367	X_UNLOCK (reqlock);
368
369	maybe_start_thread ();
370	}
371
372	static void end_thread (void)
373	{	562	{
374	eio_req *req = calloc (1, sizeof (eio_req));	563	eio_req *req = calloc (1, sizeof (eio_req));
375		564
376	req->type = EIO_QUIT;	565	req->type = -1;
377	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	566	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
378		567
379	X_LOCK (reqlock);	568	X_LOCK (reqlock);
380	reqq_push (&req_queue, req);	569	reqq_push (&req_queue, req);
381	X_COND_SIGNAL (reqwait);	570	X_COND_SIGNAL (reqwait);
382	X_UNLOCK (reqlock);	571	X_UNLOCK (reqlock);
…		…
384	X_LOCK (wrklock);	573	X_LOCK (wrklock);
385	--started;	574	--started;
386	X_UNLOCK (wrklock);	575	X_UNLOCK (wrklock);
387	}	576	}
388		577
389	void eio_set_max_poll_time (double nseconds)	578	static int
390	{
391	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
392	max_poll_time = nseconds;
393	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
394	}
395
396	void eio_set_max_poll_reqs (unsigned int maxreqs)
397	{
398	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
399	max_poll_reqs = maxreqs;
400	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
401	}
402
403	void eio_set_max_idle (unsigned int nthreads)
404	{
405	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
406	max_idle = nthreads <= 0 ? 1 : nthreads;
407	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
408	}
409
410	void eio_set_min_parallel (unsigned int nthreads)
411	{
412	if (wanted < nthreads)
413	wanted = nthreads;
414	}
415
416	void eio_set_max_parallel (unsigned int nthreads)
417	{
418	if (wanted > nthreads)
419	wanted = nthreads;
420
421	while (started > wanted)
422	end_thread ();
423	}
424
425	int eio_poll (void)	579	etp_poll (void)
426	{	580	{
427	int maxreqs = max_poll_reqs;	581	unsigned int maxreqs;
		582	unsigned int maxtime;
428	struct timeval tv_start, tv_now;	583	struct timeval tv_start, tv_now;
429	eio_req *req;
430		584
		585	X_LOCK (reslock);
		586	maxreqs = max_poll_reqs;
		587	maxtime = max_poll_time;
		588	X_UNLOCK (reslock);
		589
431	if (max_poll_time)	590	if (maxtime)
432	gettimeofday (&tv_start, 0);	591	gettimeofday (&tv_start, 0);
433		592
434	for (;;)	593	for (;;)
435	{	594	{
		595	ETP_REQ *req;
		596
436	maybe_start_thread ();	597	etp_maybe_start_thread ();
437		598
438	X_LOCK (reslock);	599	X_LOCK (reslock);
439	req = reqq_shift (&res_queue);	600	req = reqq_shift (&res_queue);
440		601
441	if (req)	602	if (req)
…		…
449	X_UNLOCK (reslock);	610	X_UNLOCK (reslock);
450		611
451	if (!req)	612	if (!req)
452	return 0;	613	return 0;
453		614
		615	X_LOCK (reqlock);
454	--nreqs;	616	--nreqs;
		617	X_UNLOCK (reqlock);
455		618
456	if (req->type == EIO_GROUP && req->size)	619	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
457	{	620	{
458	req->int1 = 1; /* mark request as delayed */	621	req->int1 = 1; /* mark request as delayed */
459	continue;	622	continue;
460	}	623	}
461	else	624	else
462	{	625	{
463	int res = eio_finish (req);	626	int res = ETP_FINISH (req);
464	if (res)	627	if (ecb_expect_false (res))
465	return res;	628	return res;
466	}	629	}
467		630
468	if (maxreqs && !--maxreqs)	631	if (ecb_expect_false (maxreqs && !--maxreqs))
469	break;	632	break;
470		633
471	if (max_poll_time)	634	if (maxtime)
472	{	635	{
473	gettimeofday (&tv_now, 0);	636	gettimeofday (&tv_now, 0);
474		637
475	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	638	if (tvdiff (&tv_start, &tv_now) >= maxtime)
476	break;	639	break;
477	}	640	}
478	}	641	}
479		642
480	errno = EAGAIN;	643	errno = EAGAIN;
481	return -1;	644	return -1;
482	}	645	}
483		646
		647	static void
		648	etp_cancel (ETP_REQ *req)
		649	{
		650	req->cancelled = 1;
		651
		652	eio_grp_cancel (req);
		653	}
		654
		655	static void
		656	etp_submit (ETP_REQ *req)
		657	{
		658	req->pri -= ETP_PRI_MIN;
		659
		660	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		661	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		662
		663	if (ecb_expect_false (req->type == EIO_GROUP))
		664	{
		665	/* I hope this is worth it :/ */
		666	X_LOCK (reqlock);
		667	++nreqs;
		668	X_UNLOCK (reqlock);
		669
		670	X_LOCK (reslock);
		671
		672	++npending;
		673
		674	if (!reqq_push (&res_queue, req) && want_poll_cb)
		675	want_poll_cb ();
		676
		677	X_UNLOCK (reslock);
		678	}
		679	else
		680	{
		681	X_LOCK (reqlock);
		682	++nreqs;
		683	++nready;
		684	reqq_push (&req_queue, req);
		685	X_COND_SIGNAL (reqwait);
		686	X_UNLOCK (reqlock);
		687
		688	etp_maybe_start_thread ();
		689	}
		690	}
		691
		692	static void ecb_cold
		693	etp_set_max_poll_time (double nseconds)
		694	{
		695	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		696	max_poll_time = nseconds * EIO_TICKS;
		697	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		698	}
		699
		700	static void ecb_cold
		701	etp_set_max_poll_reqs (unsigned int maxreqs)
		702	{
		703	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		704	max_poll_reqs = maxreqs;
		705	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		706	}
		707
		708	static void ecb_cold
		709	etp_set_max_idle (unsigned int nthreads)
		710	{
		711	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		712	max_idle = nthreads;
		713	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		714	}
		715
		716	static void ecb_cold
		717	etp_set_idle_timeout (unsigned int seconds)
		718	{
		719	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		720	idle_timeout = seconds;
		721	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		722	}
		723
		724	static void ecb_cold
		725	etp_set_min_parallel (unsigned int nthreads)
		726	{
		727	if (wanted < nthreads)
		728	wanted = nthreads;
		729	}
		730
		731	static void ecb_cold
		732	etp_set_max_parallel (unsigned int nthreads)
		733	{
		734	if (wanted > nthreads)
		735	wanted = nthreads;
		736
		737	while (started > wanted)
		738	etp_end_thread ();
		739	}
		740
		741	/*****************************************************************************/
		742
		743	static void
		744	grp_try_feed (eio_req *grp)
		745	{
		746	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		747	{
		748	grp->flags &= ~EIO_FLAG_GROUPADD;
		749
		750	EIO_FEED (grp);
		751
		752	/* stop if no progress has been made */
		753	if (!(grp->flags & EIO_FLAG_GROUPADD))
		754	{
		755	grp->feed = 0;
		756	break;
		757	}
		758	}
		759	}
		760
		761	static int
		762	grp_dec (eio_req *grp)
		763	{
		764	--grp->size;
		765
		766	/* call feeder, if applicable */
		767	grp_try_feed (grp);
		768
		769	/* finish, if done */
		770	if (!grp->size && grp->int1)
		771	return eio_finish (grp);
		772	else
		773	return 0;
		774	}
		775
		776	static void
		777	eio_destroy (eio_req *req)
		778	{
		779	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		780	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		781
		782	EIO_DESTROY (req);
		783	}
		784
		785	static int
		786	eio_finish (eio_req *req)
		787	{
		788	int res = EIO_FINISH (req);
		789
		790	if (req->grp)
		791	{
		792	int res2;
		793	eio_req *grp = req->grp;
		794
		795	/* unlink request */
		796	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		797	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		798
		799	if (grp->grp_first == req)
		800	grp->grp_first = req->grp_next;
		801
		802	res2 = grp_dec (grp);
		803
		804	if (!res)
		805	res = res2;
		806	}
		807
		808	eio_destroy (req);
		809
		810	return res;
		811	}
		812
		813	void
		814	eio_grp_cancel (eio_req *grp)
		815	{
		816	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		817	eio_cancel (grp);
		818	}
		819
		820	void
		821	eio_cancel (eio_req *req)
		822	{
		823	etp_cancel (req);
		824	}
		825
		826	void
		827	eio_submit (eio_req *req)
		828	{
		829	etp_submit (req);
		830	}
		831
		832	unsigned int
		833	eio_nreqs (void)
		834	{
		835	return etp_nreqs ();
		836	}
		837
		838	unsigned int
		839	eio_nready (void)
		840	{
		841	return etp_nready ();
		842	}
		843
		844	unsigned int
		845	eio_npending (void)
		846	{
		847	return etp_npending ();
		848	}
		849
		850	unsigned int ecb_cold
		851	eio_nthreads (void)
		852	{
		853	return etp_nthreads ();
		854	}
		855
		856	void ecb_cold
		857	eio_set_max_poll_time (double nseconds)
		858	{
		859	etp_set_max_poll_time (nseconds);
		860	}
		861
		862	void ecb_cold
		863	eio_set_max_poll_reqs (unsigned int maxreqs)
		864	{
		865	etp_set_max_poll_reqs (maxreqs);
		866	}
		867
		868	void ecb_cold
		869	eio_set_max_idle (unsigned int nthreads)
		870	{
		871	etp_set_max_idle (nthreads);
		872	}
		873
		874	void ecb_cold
		875	eio_set_idle_timeout (unsigned int seconds)
		876	{
		877	etp_set_idle_timeout (seconds);
		878	}
		879
		880	void ecb_cold
		881	eio_set_min_parallel (unsigned int nthreads)
		882	{
		883	etp_set_min_parallel (nthreads);
		884	}
		885
		886	void ecb_cold
		887	eio_set_max_parallel (unsigned int nthreads)
		888	{
		889	etp_set_max_parallel (nthreads);
		890	}
		891
		892	int eio_poll (void)
		893	{
		894	return etp_poll ();
		895	}
		896
484	/*****************************************************************************/	897	/*****************************************************************************/
485	/* work around various missing functions */	898	/* work around various missing functions */
486		899
487	#if !HAVE_PREADWRITE	900	#if !HAVE_PREADWRITE
		901	# undef pread
		902	# undef pwrite
488	# define pread aio_pread	903	# define pread eio__pread
489	# define pwrite aio_pwrite	904	# define pwrite eio__pwrite
490		905
491	/*	906	static eio_ssize_t
492	* make our pread/pwrite safe against themselves, but not against	907	eio__pread (int fd, void *buf, size_t count, off_t offset)
493	* normal read/write by using a mutex. slows down execution a lot,
494	* but that's your problem, not mine.
495	*/
496	static mutex_t preadwritelock = X_MUTEX_INIT;
497
498	static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
499	{	908	{
500	ssize_t res;	909	eio_ssize_t res;
501	off_t ooffset;	910	off_t ooffset;
502		911
503	X_LOCK (preadwritelock);	912	X_LOCK (preadwritelock);
504	ooffset = lseek (fd, 0, SEEK_CUR);	913	ooffset = lseek (fd, 0, SEEK_CUR);
505	lseek (fd, offset, SEEK_SET);	914	lseek (fd, offset, SEEK_SET);
…		…
508	X_UNLOCK (preadwritelock);	917	X_UNLOCK (preadwritelock);
509		918
510	return res;	919	return res;
511	}	920	}
512		921
513	static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset)	922	static eio_ssize_t
		923	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
514	{	924	{
515	ssize_t res;	925	eio_ssize_t res;
516	off_t ooffset;	926	off_t ooffset;
517		927
518	X_LOCK (preadwritelock);	928	X_LOCK (preadwritelock);
519	ooffset = lseek (fd, 0, SEEK_CUR);	929	ooffset = lseek (fd, 0, SEEK_CUR);
520	lseek (fd, offset, SEEK_SET);	930	lseek (fd, offset, SEEK_SET);
521	res = write (fd, buf, count);	931	res = write (fd, buf, count);
522	lseek (fd, offset, SEEK_SET);	932	lseek (fd, ooffset, SEEK_SET);
523	X_UNLOCK (preadwritelock);	933	X_UNLOCK (preadwritelock);
524		934
525	return res;	935	return res;
526	}	936	}
527	#endif	937	#endif
528		938
529	#ifndef HAVE_FUTIMES	939	#ifndef HAVE_UTIMES
530		940
		941	# undef utimes
531	# define utimes(path,times) aio_utimes (path, times)	942	# define utimes(path,times) eio__utimes (path, times)
532	# define futimes(fd,times) aio_futimes (fd, times)
533		943
		944	static int
534	static int aio_utimes (const char *filename, const struct timeval times[2])	945	eio__utimes (const char *filename, const struct timeval times[2])
535	{	946	{
536	if (times)	947	if (times)
537	{	948	{
538	struct utimbuf buf;	949	struct utimbuf buf;
539		950
…		…
544	}	955	}
545	else	956	else
546	return utime (filename, 0);	957	return utime (filename, 0);
547	}	958	}
548		959
		960	#endif
		961
		962	#ifndef HAVE_FUTIMES
		963
		964	# undef futimes
		965	# define futimes(fd,times) eio__futimes (fd, times)
		966
		967	static int
549	static int aio_futimes (int fd, const struct timeval tv[2])	968	eio__futimes (int fd, const struct timeval tv[2])
550	{	969	{
551	errno = ENOSYS;	970	errno = ENOSYS;
552	return -1;	971	return -1;
553	}	972	}
554		973
555	#endif	974	#endif
556		975
557	#if !HAVE_FDATASYNC	976	#if !HAVE_FDATASYNC
		977	# undef fdatasync
558	# define fdatasync fsync	978	# define fdatasync(fd) fsync (fd)
559	#endif	979	#endif
		980
		981	static int
		982	eio__syncfs (int fd)
		983	{
		984	int res;
		985
		986	#if HAVE_SYS_SYNCFS
		987	res = (int)syscall (__NR_syncfs, (int)(fd));
		988	#else
		989	res = -1;
		990	errno = ENOSYS;
		991	#endif
		992
		993	if (res < 0 && errno == ENOSYS && fd >= 0)
		994	sync ();
		995
		996	return res;
		997	}
		998
		999	/* sync_file_range always needs emulation */
		1000	static int
		1001	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		1002	{
		1003	#if HAVE_SYNC_FILE_RANGE
		1004	int res;
		1005
		1006	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		1007	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		1008	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		1009	{
		1010	flags = 0
		1011	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		1012	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		1013	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		1014	}
		1015
		1016	res = sync_file_range (fd, offset, nbytes, flags);
		1017
		1018	if (!res || errno != ENOSYS)
		1019	return res;
		1020	#endif
		1021
		1022	/* even though we could play tricks with the flags, it's better to always
		1023	* call fdatasync, as that matches the expectation of its users best */
		1024	return fdatasync (fd);
		1025	}
		1026
		1027	static int
		1028	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		1029	{
		1030	#if HAVE_FALLOCATE
		1031	return fallocate (fd, mode, offset, len);
		1032	#else
		1033	errno = ENOSYS;
		1034	return -1;
		1035	#endif
		1036	}
560		1037
561	#if !HAVE_READAHEAD	1038	#if !HAVE_READAHEAD
		1039	# undef readahead
562	# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self)	1040	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
563		1041
		1042	static eio_ssize_t
564	static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self)	1043	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
565	{	1044	{
566	size_t todo = count;	1045	size_t todo = count;
567	dBUF;	1046	dBUF;
568		1047
569	while (todo > 0)	1048	while (todo > 0)
…		…
573	pread (fd, eio_buf, len, offset);	1052	pread (fd, eio_buf, len, offset);
574	offset += len;	1053	offset += len;
575	todo -= len;	1054	todo -= len;
576	}	1055	}
577		1056
		1057	FUBd;
		1058
578	errno = 0;	1059	errno = 0;
579	return count;	1060	return count;
580	}	1061	}
581		1062
582	#endif	1063	#endif
583		1064
584	#if !HAVE_READDIR_R
585	# define readdir_r aio_readdir_r
586
587	static mutex_t readdirlock = X_MUTEX_INIT;
588
589	static int readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
590	{
591	EIO_STRUCT_DIRENT *e;
592	int errorno;
593
594	X_LOCK (readdirlock);
595
596	e = readdir (dirp);
597	errorno = errno;
598
599	if (e)
600	{
601	*res = ent;
602	strcpy (ent->d_name, e->d_name);
603	}
604	else
605	*res = 0;
606
607	X_UNLOCK (readdirlock);
608
609	errno = errorno;
610	return e ? 0 : -1;
611	}
612	#endif
613
614	/* sendfile always needs emulation */	1065	/* sendfile always needs emulation */
		1066	static eio_ssize_t
615	static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self)	1067	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
616	{	1068	{
		1069	eio_ssize_t written = 0;
617	ssize_t res;	1070	eio_ssize_t res;
618		1071
619	if (!count)	1072	if (!count)
620	return 0;	1073	return 0;
621		1074
		1075	for (;;)
		1076	{
		1077	#ifdef __APPLE__
		1078	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1079	#endif
622	#if HAVE_SENDFILE	1080	#if HAVE_SENDFILE
623	# if __linux	1081	# if __linux
		1082	off_t soffset = offset;
624	res = sendfile (ofd, ifd, &offset, count);	1083	res = sendfile (ofd, ifd, &soffset, count);
625		1084
626	# elif __freebsd	1085	# elif __FreeBSD__
627	/*	1086	/*
628	* Of course, the freebsd sendfile is a dire hack with no thoughts	1087	* Of course, the freebsd sendfile is a dire hack with no thoughts
629	* wasted on making it similar to other I/O functions.	1088	* wasted on making it similar to other I/O functions.
630	*/	1089	*/
631	{
632	off_t sbytes;	1090	off_t sbytes;
633	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1091	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
634		1092
635	if (res < 0 && sbytes)	1093	#if 0 /* according to the manpage, this is correct, but broken behaviour */
636	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1094	/* freebsd' sendfile will return 0 on success */
		1095	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1096	/* not on e.g. EIO or EPIPE - sounds broken */
		1097	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
637	res = sbytes;	1098	res = sbytes;
638	}	1099	#endif
		1100
		1101	/* according to source inspection, this is correct, and useful behaviour */
		1102	if (sbytes)
		1103	res = sbytes;
		1104
		1105	# elif defined (__APPLE__)
		1106	off_t sbytes = count;
		1107	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1108
		1109	/* according to the manpage, sbytes is always valid */
		1110	if (sbytes)
		1111	res = sbytes;
639		1112
640	# elif __hpux	1113	# elif __hpux
641	res = sendfile (ofd, ifd, offset, count, 0, 0);	1114	res = sendfile (ofd, ifd, offset, count, 0, 0);
642		1115
643	# elif __solaris	1116	# elif __solaris
644	{
645	struct sendfilevec vec;	1117	struct sendfilevec vec;
646	size_t sbytes;	1118	size_t sbytes;
647		1119
648	vec.sfv_fd = ifd;	1120	vec.sfv_fd = ifd;
649	vec.sfv_flag = 0;	1121	vec.sfv_flag = 0;
650	vec.sfv_off = offset;	1122	vec.sfv_off = offset;
651	vec.sfv_len = count;	1123	vec.sfv_len = count;
652		1124
653	res = sendfilev (ofd, &vec, 1, &sbytes);	1125	res = sendfilev (ofd, &vec, 1, &sbytes);
654		1126
655	if (res < 0 && sbytes)	1127	if (res < 0 && sbytes)
656	res = sbytes;	1128	res = sbytes;
657	}
658		1129
659	# endif	1130	# endif
		1131
		1132	#elif defined (_WIN32) && 0
		1133	/* does not work, just for documentation of what would need to be done */
		1134	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1135	/* libeio guarantees that the file offset does not change, and windows */
		1136	/* has no way to get an independent handle to the same file description */
		1137	HANDLE h = TO_SOCKET (ifd);
		1138	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1139	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1140
660	#else	1141	#else
661	res = -1;	1142	res = -1;
662	errno = ENOSYS;	1143	errno = ENOSYS;
663	#endif	1144	#endif
664		1145
		1146	/* we assume sendfile can copy at least 128mb in one go */
		1147	if (res <= 128 * 1024 * 1024)
		1148	{
		1149	if (res > 0)
		1150	written += res;
		1151
		1152	if (written)
		1153	return written;
		1154
		1155	break;
		1156	}
		1157	else
		1158	{
		1159	/* if we requested more, then probably the kernel was lazy */
		1160	written += res;
		1161	offset += res;
		1162	count -= res;
		1163
		1164	if (!count)
		1165	return written;
		1166	}
		1167	}
		1168
665	if (res < 0	1169	if (res < 0
666	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1170	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1171	/* BSDs */
		1172	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1173	|| errno == ENOTSUP
		1174	#endif
		1175	#ifdef EOPNOTSUPP /* windows */
		1176	|| errno == EOPNOTSUPP /* BSDs */
		1177	#endif
667	#if __solaris	1178	#if __solaris
668	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1179	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
669	#endif	1180	#endif
670	)	1181	)
671	)	1182	)
…		…
675		1186
676	res = 0;	1187	res = 0;
677		1188
678	while (count)	1189	while (count)
679	{	1190	{
680	ssize_t cnt;	1191	eio_ssize_t cnt;
681		1192
682	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1193	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
683		1194
684	if (cnt <= 0)	1195	if (cnt <= 0)
685	{	1196	{
…		…
697		1208
698	offset += cnt;	1209	offset += cnt;
699	res += cnt;	1210	res += cnt;
700	count -= cnt;	1211	count -= cnt;
701	}	1212	}
		1213
		1214	FUBd;
702	}	1215	}
703		1216
704	return res;	1217	return res;
705	}	1218	}
706		1219
		1220	#ifdef PAGESIZE
		1221	# define eio_pagesize() PAGESIZE
		1222	#else
		1223	static intptr_t
		1224	eio_pagesize (void)
		1225	{
		1226	static intptr_t page;
		1227
		1228	if (!page)
		1229	page = sysconf (_SC_PAGESIZE);
		1230
		1231	return page;
		1232	}
		1233	#endif
		1234
		1235	static void
		1236	eio_page_align (void **addr, size_t *length)
		1237	{
		1238	intptr_t mask = eio_pagesize () - 1;
		1239
		1240	/* round down addr */
		1241	intptr_t adj = mask & (intptr_t)*addr;
		1242
		1243	*addr = (void *)((intptr_t)*addr - adj);
		1244	*length += adj;
		1245
		1246	/* round up length */
		1247	*length = (*length + mask) & ~mask;
		1248	}
		1249
		1250	#if !_POSIX_MEMLOCK
		1251	# define eio__mlockall(a) EIO_ENOSYS ()
		1252	#else
		1253
		1254	static int
		1255	eio__mlockall (int flags)
		1256	{
		1257	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1258	extern int mallopt (int, int);
		1259	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1260	#endif
		1261
		1262	if (EIO_MCL_CURRENT != MCL_CURRENT
		1263	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1264	{
		1265	flags = 0
		1266	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1267	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1268	}
		1269
		1270	return mlockall (flags);
		1271	}
		1272	#endif
		1273
		1274	#if !_POSIX_MEMLOCK_RANGE
		1275	# define eio__mlock(a,b) EIO_ENOSYS ()
		1276	#else
		1277
		1278	static int
		1279	eio__mlock (void *addr, size_t length)
		1280	{
		1281	eio_page_align (&addr, &length);
		1282
		1283	return mlock (addr, length);
		1284	}
		1285
		1286	#endif
		1287
		1288	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1289	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1290	#else
		1291
		1292	static int
		1293	eio__msync (void *mem, size_t len, int flags)
		1294	{
		1295	eio_page_align (&mem, &len);
		1296
		1297	if (EIO_MS_ASYNC != MS_SYNC
		1298	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1299	|| EIO_MS_SYNC != MS_SYNC)
		1300	{
		1301	flags = 0
		1302	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1303	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1304	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1305	}
		1306
		1307	return msync (mem, len, flags);
		1308	}
		1309
		1310	#endif
		1311
		1312	static int
		1313	eio__mtouch (eio_req *req)
		1314	{
		1315	void *mem = req->ptr2;
		1316	size_t len = req->size;
		1317	int flags = req->int1;
		1318
		1319	eio_page_align (&mem, &len);
		1320
		1321	{
		1322	intptr_t addr = (intptr_t)mem;
		1323	intptr_t end = addr + len;
		1324	intptr_t page = eio_pagesize ();
		1325
		1326	if (addr < end)
		1327	if (flags & EIO_MT_MODIFY) /* modify */
		1328	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1329	else
		1330	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1331	}
		1332
		1333	return 0;
		1334	}
		1335
		1336	/*****************************************************************************/
		1337	/* requests implemented outside eio_execute, because they are so large */
		1338
		1339	static void
		1340	eio__realpath (eio_req *req, etp_worker *self)
		1341	{
		1342	char *rel = req->ptr1;
		1343	char *res;
		1344	char *tmp1, *tmp2;
		1345	#if SYMLOOP_MAX > 32
		1346	int symlinks = SYMLOOP_MAX;
		1347	#else
		1348	int symlinks = 32;
		1349	#endif
		1350
		1351	req->result = -1;
		1352
		1353	errno = EINVAL;
		1354	if (!rel)
		1355	return;
		1356
		1357	errno = ENOENT;
		1358	if (!*rel)
		1359	return;
		1360
		1361	if (!req->ptr2)
		1362	{
		1363	X_LOCK (wrklock);
		1364	req->flags |= EIO_FLAG_PTR2_FREE;
		1365	X_UNLOCK (wrklock);
		1366	req->ptr2 = malloc (PATH_MAX * 3);
		1367
		1368	errno = ENOMEM;
		1369	if (!req->ptr2)
		1370	return;
		1371	}
		1372
		1373	res = req->ptr2;
		1374	tmp1 = res + PATH_MAX;
		1375	tmp2 = tmp1 + PATH_MAX;
		1376
		1377	#if 0 /* disabled, the musl way to do things is just too racy */
		1378	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1379	/* on linux we may be able to ask the kernel */
		1380	{
		1381	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1382
		1383	if (fd >= 0)
		1384	{
		1385	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1386	req->result = readlink (tmp1, res, PATH_MAX);
		1387	close (fd);
		1388
		1389	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1390
		1391	if (req->result > 0)
		1392	goto done;
		1393	}
		1394	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1395	return;
		1396	}
		1397	#endif
		1398	#endif
		1399
		1400	if (*rel != '/')
		1401	{
		1402	if (!getcwd (res, PATH_MAX))
		1403	return;
		1404
		1405	if (res [1]) /* only use if not / */
		1406	res += strlen (res);
		1407	}
		1408
		1409	while (*rel)
		1410	{
		1411	eio_ssize_t len, linklen;
		1412	char *beg = rel;
		1413
		1414	while (*rel && *rel != '/')
		1415	++rel;
		1416
		1417	len = rel - beg;
		1418
		1419	if (!len) /* skip slashes */
		1420	{
		1421	++rel;
		1422	continue;
		1423	}
		1424
		1425	if (beg [0] == '.')
		1426	{
		1427	if (len == 1)
		1428	continue; /* . - nop */
		1429
		1430	if (beg [1] == '.' && len == 2)
		1431	{
		1432	/* .. - back up one component, if possible */
		1433
		1434	while (res != req->ptr2)
		1435	if (*--res == '/')
		1436	break;
		1437
		1438	continue;
		1439	}
		1440	}
		1441
		1442	errno = ENAMETOOLONG;
		1443	if (res + 1 + len + 1 >= tmp1)
		1444	return;
		1445
		1446	/* copy one component */
		1447	*res = '/';
		1448	memcpy (res + 1, beg, len);
		1449
		1450	/* zero-terminate, for readlink */
		1451	res [len + 1] = 0;
		1452
		1453	/* now check if it's a symlink */
		1454	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1455
		1456	if (linklen < 0)
		1457	{
		1458	if (errno != EINVAL)
		1459	return;
		1460
		1461	/* it's a normal directory. hopefully */
		1462	res += len + 1;
		1463	}
		1464	else
		1465	{
		1466	/* yay, it was a symlink - build new path in tmp2 */
		1467	int rellen = strlen (rel);
		1468
		1469	errno = ENAMETOOLONG;
		1470	if (linklen + 1 + rellen >= PATH_MAX)
		1471	return;
		1472
		1473	errno = ELOOP;
		1474	if (!--symlinks)
		1475	return;
		1476
		1477	if (*tmp1 == '/')
		1478	res = req->ptr2; /* symlink resolves to an absolute path */
		1479
		1480	/* we need to be careful, as rel might point into tmp2 already */
		1481	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1482	tmp2 [linklen] = '/';
		1483	memcpy (tmp2, tmp1, linklen);
		1484
		1485	rel = tmp2;
		1486	}
		1487	}
		1488
		1489	/* special case for the lone root path */
		1490	if (res == req->ptr2)
		1491	*res++ = '/';
		1492
		1493	req->result = res - (char *)req->ptr2;
		1494
		1495	done:
		1496	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1497	}
		1498
		1499	static signed char
		1500	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1501	{
		1502	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1503	: a->inode < b->inode ? -1
		1504	: a->inode > b->inode ? 1
		1505	: 0;
		1506	}
		1507
		1508	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1509
		1510	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1511	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1512
		1513	static void
		1514	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1515	{
		1516	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
		1517	unsigned char *bit = bits;
		1518
		1519	assert (CHAR_BIT == 8);
		1520	assert (sizeof (eio_dirent) * 8 < 256);
		1521	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1522	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1523
		1524	if (size <= EIO_SORT_FAST)
		1525	return;
		1526
		1527	/* first prepare an array of bits to test in our radix sort */
		1528	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
		1529	/* inode_bits must contain all inodes ORed together */
		1530	/* which is used to skip bits that are 0 everywhere, which is very common */
		1531	{
		1532	eio_ino_t endianness;
		1533	int i, j;
		1534
		1535	/* we store the byte offset of byte n into byte n of "endianness" */
		1536	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1537	((unsigned char *)&endianness)[i] = i;
		1538
		1539	*bit++ = 0;
		1540
		1541	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1542	{
		1543	/* shifting off the byte offsets out of "endianness" */
		1544	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1545	endianness >>= 8;
		1546
		1547	for (j = 0; j < 8; ++j)
		1548	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
		1549	*bit++ = offs + j;
		1550	}
		1551
		1552	for (j = 0; j < 8; ++j)
		1553	if (score_bits & (1 << j))
		1554	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1555	}
		1556
		1557	/* now actually do the sorting (a variant of MSD radix sort) */
		1558	{
		1559	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
		1560	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
		1561	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
		1562	int stk_idx = 0;
		1563
		1564	base_stk [stk_idx] = dents;
		1565	end_stk [stk_idx] = dents + size;
		1566	bit_stk [stk_idx] = bit - 1;
		1567
		1568	do
		1569	{
		1570	base = base_stk [stk_idx];
		1571	end = end_stk [stk_idx];
		1572	bit = bit_stk [stk_idx];
		1573
		1574	for (;;)
		1575	{
		1576	unsigned char O = *bit >> 3;
		1577	unsigned char M = 1 << (*bit & 7);
		1578
		1579	eio_dirent *a = base;
		1580	eio_dirent *b = end;
		1581
		1582	if (b - a < EIO_SORT_CUTOFF)
		1583	break;
		1584
		1585	/* now bit-partition the array on the bit */
		1586	/* this ugly asymmetric loop seems to perform much better than typical */
		1587	/* partition algos found in the literature */
		1588	do
		1589	if (!(((unsigned char *)a)[O] & M))
		1590	++a;
		1591	else if (!(((unsigned char *)--b)[O] & M))
		1592	{
		1593	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1594	++a;
		1595	}
		1596	while (b > a);
		1597
		1598	/* next bit, or stop, if no bits left in this path */
		1599	if (!*--bit)
		1600	break;
		1601
		1602	base_stk [stk_idx] = a;
		1603	end_stk [stk_idx] = end;
		1604	bit_stk [stk_idx] = bit;
		1605	++stk_idx;
		1606
		1607	end = a;
		1608	}
		1609	}
		1610	while (stk_idx--);
		1611	}
		1612	}
		1613
		1614	static void
		1615	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1616	{
		1617	/* first move the smallest element to the front, to act as a sentinel */
		1618	{
		1619	int i;
		1620	eio_dirent *min = dents;
		1621
		1622	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1623	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1624	if (EIO_DENT_CMP (dents [i], <, *min))
		1625	min = &dents [i];
		1626
		1627	/* swap elements 0 and j (minimum) */
		1628	{
		1629	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1630	}
		1631	}
		1632
		1633	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1634	{
		1635	eio_dirent *i, *j;
		1636
		1637	for (i = dents + 1; i < dents + size; ++i)
		1638	{
		1639	eio_dirent value = *i;
		1640
		1641	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1642	j [1] = j [0];
		1643
		1644	j [1] = value;
		1645	}
		1646	}
		1647	}
		1648
		1649	static void
		1650	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1651	{
		1652	if (size <= 1)
		1653	return; /* our insertion sort relies on size > 0 */
		1654
		1655	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1656	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1657	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1658
		1659	/* use an insertion sort at the end, or for small arrays, */
		1660	/* as insertion sort is more efficient for small partitions */
		1661	eio_dent_insertion_sort (dents, size);
		1662	}
		1663
707	/* read a full directory */	1664	/* read a full directory */
		1665	static void
708	static void scandir_ (eio_req *req, worker *self)	1666	eio__scandir (eio_req *req, etp_worker *self)
709	{	1667	{
		1668	char *name, *names;
		1669	int namesalloc = 4096 - sizeof (void *) * 4;
		1670	int namesoffs = 0;
		1671	int flags = req->int1;
		1672	eio_dirent *dents = 0;
		1673	int dentalloc = 128;
		1674	int dentoffs = 0;
		1675	eio_ino_t inode_bits = 0;
		1676	#ifdef _WIN32
		1677	HANDLE dirp;
		1678	WIN32_FIND_DATA entp;
		1679	#else
710	DIR *dirp;	1680	DIR *dirp;
711	union
712	{
713	EIO_STRUCT_DIRENT d;
714	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
715	} *u;
716	EIO_STRUCT_DIRENT *entp;	1681	EIO_STRUCT_DIRENT *entp;
717	char *name, *names;	1682	#endif
718	int memlen = 4096;
719	int memofs = 0;
720	int res = 0;
721		1683
722	X_LOCK (wrklock);	1684	req->result = -1;
		1685
		1686	if (!(flags & EIO_READDIR_DENTS))
		1687	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
		1688
		1689	#ifdef _WIN32
		1690	{
		1691	int len = strlen ((const char *)req->ptr1);
		1692	char *path = malloc (MAX_PATH);
		1693	const char *fmt;
		1694
		1695	if (!len)
		1696	fmt = "./*";
		1697	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1698	fmt = "%s*";
		1699	else
		1700	fmt = "%s/*";
		1701
		1702	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1703	dirp = FindFirstFile (path, &entp);
		1704	free (path);
		1705
		1706	if (dirp == INVALID_HANDLE_VALUE)
		1707	{
		1708	dirp = 0;
		1709
		1710	/* should steal _dosmaperr */
		1711	switch (GetLastError ())
		1712	{
		1713	case ERROR_FILE_NOT_FOUND:
		1714	req->result = 0;
		1715	break;
		1716
		1717	case ERROR_INVALID_NAME:
		1718	case ERROR_PATH_NOT_FOUND:
		1719	case ERROR_NO_MORE_FILES:
		1720	errno = ENOENT;
		1721	break;
		1722
		1723	case ERROR_NOT_ENOUGH_MEMORY:
		1724	errno = ENOMEM;
		1725	break;
		1726
		1727	default:
		1728	errno = EINVAL;
		1729	break;
		1730	}
		1731	}
		1732	}
		1733	#else
723	self->dirp = dirp = opendir (req->ptr1);	1734	dirp = opendir (req->ptr1);
724	self->dbuf = u = malloc (sizeof (*u));	1735	#endif
		1736
725	req->flags |= EIO_FLAG_PTR2_FREE;	1737	if (req->flags & EIO_FLAG_PTR1_FREE)
		1738	free (req->ptr1);
		1739
		1740	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1741	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
726	req->ptr2 = names = malloc (memlen);	1742	req->ptr2 = names = malloc (namesalloc);
727	X_UNLOCK (wrklock);
728		1743
729	if (dirp && u && names)	1744	if (dirp && names && (!flags || dents))
730	for (;;)	1745	for (;;)
731	{	1746	{
		1747	int done;
		1748
		1749	#ifdef _WIN32
		1750	done = !dirp;
		1751	#else
732	errno = 0;	1752	errno = 0;
733	readdir_r (dirp, &u->d, &entp);	1753	entp = readdir (dirp);
		1754	done = !entp;
		1755	#endif
734		1756
735	if (!entp)	1757	if (done)
		1758	{
		1759	#ifndef _WIN32
		1760	int old_errno = errno;
		1761	closedir (dirp);
		1762	errno = old_errno;
		1763
		1764	if (errno)
		1765	break;
		1766	#endif
		1767
		1768	/* sort etc. */
		1769	req->int1 = flags;
		1770	req->result = dentoffs;
		1771
		1772	if (flags & EIO_READDIR_STAT_ORDER)
		1773	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1774	else if (flags & EIO_READDIR_DIRS_FIRST)
		1775	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1776	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1777	else
		1778	{
		1779	/* in this case, all is known, and we just put dirs first and sort them */
		1780	eio_dirent *oth = dents + dentoffs;
		1781	eio_dirent *dir = dents;
		1782
		1783	/* now partition dirs to the front, and non-dirs to the back */
		1784	/* by walking from both sides and swapping if necessary */
		1785	while (oth > dir)
		1786	{
		1787	if (dir->type == EIO_DT_DIR)
		1788	++dir;
		1789	else if ((--oth)->type == EIO_DT_DIR)
		1790	{
		1791	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1792
		1793	++dir;
		1794	}
		1795	}
		1796
		1797	/* now sort the dirs only (dirs all have the same score) */
		1798	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1799	}
		1800
736	break;	1801	break;
		1802	}
737		1803
738	name = entp->d_name;	1804	/* now add the entry to our list(s) */
		1805	name = D_NAME (entp);
739		1806
		1807	/* skip . and .. entries */
740	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1808	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
741	{	1809	{
742	int len = strlen (name) + 1;	1810	int len = D_NAMLEN (entp) + 1;
743		1811
744	res++;	1812	while (ecb_expect_false (namesoffs + len > namesalloc))
745
746	while (memofs + len > memlen)
747	{	1813	{
748	memlen *= 2;	1814	namesalloc *= 2;
749	X_LOCK (wrklock);
750	req->ptr2 = names = realloc (names, memlen);	1815	req->ptr2 = names = realloc (names, namesalloc);
751	X_UNLOCK (wrklock);
752		1816
753	if (!names)	1817	if (!names)
754	break;	1818	break;
755	}	1819	}
756		1820
757	memcpy (names + memofs, name, len);	1821	memcpy (names + namesoffs, name, len);
		1822
		1823	if (dents)
		1824	{
		1825	struct eio_dirent *ent;
		1826
		1827	if (ecb_expect_false (dentoffs == dentalloc))
		1828	{
		1829	dentalloc *= 2;
		1830	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1831
		1832	if (!dents)
		1833	break;
		1834	}
		1835
		1836	ent = dents + dentoffs;
		1837
		1838	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1839	ent->namelen = len - 1;
		1840	ent->inode = D_INO (entp);
		1841
		1842	inode_bits |= ent->inode;
		1843
		1844	switch (D_TYPE (entp))
		1845	{
		1846	default:
		1847	ent->type = EIO_DT_UNKNOWN;
		1848	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1849	break;
		1850
		1851	#ifdef DT_FIFO
		1852	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1853	#endif
		1854	#ifdef DT_CHR
		1855	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1856	#endif
		1857	#ifdef DT_MPC
		1858	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1859	#endif
		1860	#ifdef DT_DIR
		1861	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1862	#endif
		1863	#ifdef DT_NAM
		1864	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1865	#endif
		1866	#ifdef DT_BLK
		1867	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1868	#endif
		1869	#ifdef DT_MPB
		1870	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1871	#endif
		1872	#ifdef DT_REG
		1873	case DT_REG: ent->type = EIO_DT_REG; break;
		1874	#endif
		1875	#ifdef DT_NWK
		1876	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1877	#endif
		1878	#ifdef DT_CMP
		1879	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1880	#endif
		1881	#ifdef DT_LNK
		1882	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1883	#endif
		1884	#ifdef DT_SOCK
		1885	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1886	#endif
		1887	#ifdef DT_DOOR
		1888	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1889	#endif
		1890	#ifdef DT_WHT
		1891	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1892	#endif
		1893	}
		1894
		1895	ent->score = 7;
		1896
		1897	if (flags & EIO_READDIR_DIRS_FIRST)
		1898	{
		1899	if (ent->type == EIO_DT_UNKNOWN)
		1900	{
		1901	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1902	ent->score = 1;
		1903	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1904	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1905	}
		1906	else if (ent->type == EIO_DT_DIR)
		1907	ent->score = 0;
		1908	}
		1909	}
		1910
758	memofs += len;	1911	namesoffs += len;
		1912	++dentoffs;
759	}	1913	}
		1914
		1915	if (EIO_CANCELLED (req))
		1916	{
		1917	errno = ECANCELED;
		1918	break;
		1919	}
		1920
		1921	#ifdef _WIN32
		1922	if (!FindNextFile (dirp, &entp))
		1923	{
		1924	FindClose (dirp);
		1925	dirp = 0;
		1926	}
		1927	#endif
760	}	1928	}
761
762	if (errno)
763	res = -1;
764
765	req->result = res;
766	}	1929	}
767		1930
768	/*****************************************************************************/	1931	/*****************************************************************************/
769		1932
		1933	#define ALLOC(len) \
		1934	if (!req->ptr2) \
		1935	{ \
		1936	X_LOCK (wrklock); \
		1937	req->flags |= EIO_FLAG_PTR2_FREE; \
		1938	X_UNLOCK (wrklock); \
		1939	req->ptr2 = malloc (len); \
		1940	if (!req->ptr2) \
		1941	{ \
		1942	errno = ENOMEM; \
		1943	req->result = -1; \
		1944	break; \
		1945	} \
		1946	}
		1947
770	X_THREAD_PROC (eio_proc)	1948	X_THREAD_PROC (etp_proc)
771	{	1949	{
772	eio_req *req;	1950	ETP_REQ *req;
773	struct timespec ts;	1951	struct timespec ts;
774	worker *self = (worker *)thr_arg;	1952	etp_worker *self = (etp_worker *)thr_arg;
775		1953
		1954	#if HAVE_PRCTL_SET_NAME
		1955	prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
		1956	#endif
		1957
776	/* try to distribute timeouts somewhat randomly */	1958	/* try to distribute timeouts somewhat evenly */
777	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1959	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
778		1960
779	for (;;)	1961	for (;;)
780	{	1962	{
781	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1963	ts.tv_sec = 0;
782		1964
783	X_LOCK (reqlock);	1965	X_LOCK (reqlock);
784		1966
785	for (;;)	1967	for (;;)
786	{	1968	{
787	self->req = req = reqq_shift (&req_queue);	1969	self->req = req = reqq_shift (&req_queue);
788		1970
789	if (req)	1971	if (req)
790	break;	1972	break;
791		1973
		1974	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1975	{
		1976	X_UNLOCK (reqlock);
		1977	X_LOCK (wrklock);
		1978	--started;
		1979	X_UNLOCK (wrklock);
		1980	goto quit;
		1981	}
		1982
792	++idle;	1983	++idle;
793		1984
794	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1985	if (idle <= max_idle)
		1986	/* we are allowed to idle, so do so without any timeout */
		1987	X_COND_WAIT (reqwait, reqlock);
		1988	else
795	{	1989	{
796	if (idle > max_idle)	1990	/* initialise timeout once */
797	{	1991	if (!ts.tv_sec)
798	--idle;	1992	ts.tv_sec = time (0) + idle_timeout;
799	X_UNLOCK (reqlock);
800	X_LOCK (wrklock);
801	--started;
802	X_UNLOCK (wrklock);
803	goto quit;
804	}
805		1993
806	/* we are allowed to idle, so do so without any timeout */
807	X_COND_WAIT (reqwait, reqlock);	1994	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
808	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1995	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
809	}	1996	}
810		1997
811	--idle;	1998	--idle;
812	}	1999	}
813		2000
814	--nready;	2001	--nready;
815		2002
816	X_UNLOCK (reqlock);	2003	X_UNLOCK (reqlock);
817		2004
818	errno = 0; /* strictly unnecessary */
819
820	if (!EIO_CANCELLED (req))
821	switch (req->type)	2005	if (req->type < 0)
822	{
823	case EIO_READ: req->result = req->offs >= 0
824	? pread (req->int1, req->ptr2, req->size, req->offs)
825	: read (req->int1, req->ptr2, req->size); break;
826	case EIO_WRITE: req->result = req->offs >= 0
827	? pwrite (req->int1, req->ptr2, req->size, req->offs)
828	: write (req->int1, req->ptr2, req->size); break;
829
830	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
831	case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
832
833	case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
834	case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
835	case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
836
837	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
838	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
839	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
840	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
841	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
842	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
843
844	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
845	case EIO_CLOSE: req->result = close (req->int1); break;
846	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
847	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
848	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
849	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
850	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
851	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
852	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
853	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
854	case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
855
856	case EIO_SYNC: req->result = 0; sync (); break;
857	case EIO_FSYNC: req->result = fsync (req->int1); break;
858	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
859
860	case EIO_READDIR: scandir_ (req, self); break;
861
862	case EIO_BUSY:
863	#ifdef _WIN32
864	Sleep (req->nv1 * 1000.);
865	#else
866	{
867	struct timeval tv;
868
869	tv.tv_sec = req->nv1;
870	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
871
872	req->result = select (0, 0, 0, 0, &tv);
873	}
874	#endif
875	break;
876
877	case EIO_UTIME:
878	case EIO_FUTIME:
879	{
880	struct timeval tv[2];
881	struct timeval *times;
882
883	if (req->nv1 != -1. || req->nv2 != -1.)
884	{
885	tv[0].tv_sec = req->nv1;
886	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
887	tv[1].tv_sec = req->nv2;
888	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
889
890	times = tv;
891	}
892	else
893	times = 0;
894
895
896	req->result = req->type == EIO_FUTIME
897	? futimes (req->int1, times)
898	: utimes (req->ptr1, times);
899	}
900
901	case EIO_GROUP:
902	case EIO_NOP:
903	break;
904
905	case EIO_QUIT:
906	goto quit;	2006	goto quit;
907		2007
908	default:	2008	ETP_EXECUTE (self, req);
909	req->result = -1;
910	break;
911	}
912
913	req->errorno = errno;
914		2009
915	X_LOCK (reslock);	2010	X_LOCK (reslock);
916		2011
917	++npending;	2012	++npending;
918		2013
919	if (!reqq_push (&res_queue, req) && want_poll_cb)	2014	if (!reqq_push (&res_queue, req) && want_poll_cb)
920	want_poll_cb ();	2015	want_poll_cb ();
921		2016
922	self->req = 0;	2017	self->req = 0;
923	worker_clear (self);	2018	etp_worker_clear (self);
924		2019
925	X_UNLOCK (reslock);	2020	X_UNLOCK (reslock);
926	}	2021	}
927		2022
928	quit:	2023	quit:
929	X_LOCK (wrklock);	2024	X_LOCK (wrklock);
930	worker_free (self);	2025	etp_worker_free (self);
931	X_UNLOCK (wrklock);	2026	X_UNLOCK (wrklock);
932		2027
933	return 0;	2028	return 0;
934	}	2029	}
935		2030
936	/*****************************************************************************/	2031	/*****************************************************************************/
937		2032
938	static void eio_atfork_prepare (void)	2033	int ecb_cold
		2034	eio_init (void (*want_poll)(void), void (*done_poll)(void))
939	{	2035	{
940	X_LOCK (wrklock);
941	X_LOCK (reqlock);
942	X_LOCK (reslock);
943	#if !HAVE_PREADWRITE	2036	#if !HAVE_PREADWRITE
944	X_LOCK (preadwritelock);
945	#endif
946	#if !HAVE_READDIR_R
947	X_LOCK (readdirlock);
948	#endif
949	}
950
951	static void eio_atfork_parent (void)
952	{
953	#if !HAVE_READDIR_R
954	X_UNLOCK (readdirlock);
955	#endif
956	#if !HAVE_PREADWRITE
957	X_UNLOCK (preadwritelock);	2037	X_MUTEX_CREATE (preadwritelock);
958	#endif	2038	#endif
959	X_UNLOCK (reslock);
960	X_UNLOCK (reqlock);
961	X_UNLOCK (wrklock);
962	}
963		2039
964	static void eio_atfork_child (void)	2040	return etp_init (want_poll, done_poll);
965	{	2041	}
966	eio_req *prv;
967		2042
968	while (prv = reqq_shift (&req_queue))	2043	ecb_inline void
969	eio_destroy (prv);	2044	eio_api_destroy (eio_req *req)
		2045	{
		2046	free (req);
		2047	}
970		2048
971	while (prv = reqq_shift (&res_queue))	2049	#define REQ(rtype) \
972	eio_destroy (prv);	2050	eio_req *req; \
		2051	\
		2052	req = (eio_req *)calloc (1, sizeof *req); \
		2053	if (!req) \
		2054	return 0; \
		2055	\
		2056	req->type = rtype; \
		2057	req->pri = pri; \
		2058	req->finish = cb; \
		2059	req->data = data; \
		2060	req->destroy = eio_api_destroy;
973		2061
974	while (wrk_first.next != &wrk_first)	2062	#define SEND eio_submit (req); return req
		2063
		2064	#define PATH \
		2065	req->flags |= EIO_FLAG_PTR1_FREE; \
		2066	req->ptr1 = strdup (path); \
		2067	if (!req->ptr1) \
		2068	{ \
		2069	eio_api_destroy (req); \
		2070	return 0; \
		2071	}
		2072
		2073	static void
		2074	eio_execute (etp_worker *self, eio_req *req)
		2075	{
		2076	if (ecb_expect_false (EIO_CANCELLED (req)))
975	{	2077	{
976	worker *wrk = wrk_first.next;	2078	req->result = -1;
977		2079	req->errorno = ECANCELED;
978	if (wrk->req)	2080	return;
979	eio_destroy (wrk->req);
980
981	worker_clear (wrk);
982	worker_free (wrk);
983	}	2081	}
984		2082
985	started = 0;	2083	switch (req->type)
986	idle = 0;	2084	{
987	nreqs = 0;	2085	case EIO_READ: ALLOC (req->size);
988	nready = 0;	2086	req->result = req->offs >= 0
989	npending = 0;	2087	? pread (req->int1, req->ptr2, req->size, req->offs)
		2088	: read (req->int1, req->ptr2, req->size); break;
		2089	case EIO_WRITE: req->result = req->offs >= 0
		2090	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		2091	: write (req->int1, req->ptr2, req->size); break;
990		2092
991	eio_atfork_parent ();	2093	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
992	}	2094	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
993		2095
994	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	2096	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
995	{	2097	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
996	want_poll_cb = want_poll;	2098	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
997	done_poll_cb = done_poll;	2099	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2100	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2101	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
998		2102
		2103	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2104	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2105	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2106	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2107
		2108	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		2109	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		2110	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		2111	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		2112	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		2113	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		2114
		2115	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		2116	case EIO_CLOSE: req->result = close (req->int1); break;
		2117	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		2118	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		2119	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		2120	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		2121	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		2122	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		2123	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		2124	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2125
		2126	case EIO_REALPATH: eio__realpath (req, self); break;
		2127
		2128	case EIO_READLINK: ALLOC (PATH_MAX);
		2129	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
		2130
		2131	case EIO_SYNC: req->result = 0; sync (); break;
		2132	case EIO_FSYNC: req->result = fsync (req->int1); break;
		2133	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2134	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2135	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2136	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2137	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		2138	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2139	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		2140	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
		2141
		2142	case EIO_READDIR: eio__scandir (req, self); break;
		2143
		2144	case EIO_BUSY:
999	#ifdef _WIN32	2145	#ifdef _WIN32
1000	X_MUTEX_CHECK (wrklock);	2146	Sleep (req->nv1 * 1e3);
1001	X_MUTEX_CHECK (reslock);	2147	#else
1002	X_MUTEX_CHECK (reqlock);	2148	{
1003	X_MUTEX_CHECK (reqwait);	2149	struct timeval tv;
1004	X_MUTEX_CHECK (preadwritelock);
1005	X_MUTEX_CHECK (readdirlock);
1006		2150
1007	X_COND_CHECK (reqwait);	2151	tv.tv_sec = req->nv1;
1008	#endif	2152	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1009		2153
1010	X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);	2154	req->result = select (0, 0, 0, 0, &tv);
1011	}	2155	}
		2156	#endif
		2157	break;
1012		2158
1013	#if 0	2159	case EIO_UTIME:
		2160	case EIO_FUTIME:
		2161	{
		2162	struct timeval tv[2];
		2163	struct timeval *times;
1014		2164
		2165	if (req->nv1 != -1. || req->nv2 != -1.)
		2166	{
		2167	tv[0].tv_sec = req->nv1;
		2168	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		2169	tv[1].tv_sec = req->nv2;
		2170	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		2171
		2172	times = tv;
		2173	}
		2174	else
		2175	times = 0;
		2176
		2177	req->result = req->type == EIO_FUTIME
		2178	? futimes (req->int1, times)
		2179	: utimes (req->ptr1, times);
		2180	}
		2181	break;
		2182
		2183	case EIO_GROUP:
		2184	abort (); /* handled in eio_request */
		2185
		2186	case EIO_NOP:
		2187	req->result = 0;
		2188	break;
		2189
		2190	case EIO_CUSTOM:
		2191	req->feed (req);
		2192	break;
		2193
		2194	default:
		2195	errno = ENOSYS;
		2196	req->result = -1;
		2197	break;
		2198	}
		2199
		2200	req->errorno = errno;
		2201	}
		2202
		2203	#ifndef EIO_NO_WRAPPERS
		2204
		2205	eio_req *eio_nop (int pri, eio_cb cb, void *data)
		2206	{
		2207	REQ (EIO_NOP); SEND;
		2208	}
		2209
		2210	eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
		2211	{
		2212	REQ (EIO_BUSY); req->nv1 = delay; SEND;
		2213	}
		2214
1015	eio_req *eio_fsync (int fd, eio_cb cb);	2215	eio_req *eio_sync (int pri, eio_cb cb, void *data)
		2216	{
		2217	REQ (EIO_SYNC); SEND;
		2218	}
		2219
		2220	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
		2221	{
		2222	REQ (EIO_FSYNC); req->int1 = fd; SEND;
		2223	}
		2224
		2225	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2226	{
		2227	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2228	}
		2229
1016	eio_req *eio_fdatasync (int fd, eio_cb cb);	2230	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1017	eio_req *eio_dupclose (int fd, eio_cb cb);	2231	{
		2232	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2233	}
		2234
		2235	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2236	{
		2237	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2238	}
		2239
		2240	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2241	{
		2242	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2243	}
		2244
		2245	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2246	{
		2247	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2248	}
		2249
		2250	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2251	{
		2252	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2253	}
		2254
		2255	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2256	{
		2257	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2258	}
		2259
		2260	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2261	{
		2262	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2263	}
		2264
		2265	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
		2266	{
		2267	REQ (EIO_CLOSE); req->int1 = fd; SEND;
		2268	}
		2269
1018	eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb);	2270	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1019	eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb);	2271	{
1020	eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb);	2272	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1021	eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */	2273	}
		2274
		2275	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
		2276	{
		2277	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
		2278	}
		2279
		2280	eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
		2281	{
		2282	REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
		2283	}
		2284
		2285	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
		2286	{
		2287	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		2288	}
		2289
		2290	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2291	{
		2292	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2293	}
		2294
1022	eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb);	2295	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
		2296	{
		2297	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
		2298	}
		2299
1023	eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb);	2300	eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
		2301	{
		2302	REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
		2303	}
		2304
1024	eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb);	2305	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
		2306	{
		2307	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
		2308	}
		2309
1025	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb);	2310	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
		2311	{
		2312	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
		2313	}
		2314
1026	eio_req *eio_dup2 (int fd, int fd2, eio_cb cb);	2315	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
		2316	{
		2317	REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
		2318	}
		2319
1027	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb);	2320	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
		2321	{
		2322	REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
		2323	}
		2324
1028	eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb);	2325	eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1029	eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */	2326	{
1030	eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */	2327	REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1031	eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */	2328	}
		2329
1032	eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb);	2330	eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
		2331	{
		2332	REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
		2333	}
		2334
1033	eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb);	2335	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
		2336	{
		2337	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
		2338	}
		2339
		2340	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
		2341	{
		2342	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
		2343	}
		2344
1034	eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb);	2345	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
		2346	{
		2347	REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
		2348	}
		2349
1035	eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb);	2350	eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1036	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb);	2351	{
		2352	REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
		2353	}
		2354
		2355	static eio_req *
		2356	eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
		2357	{
		2358	REQ (type); PATH; SEND;
		2359	}
		2360
		2361	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
		2362	{
		2363	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2364	}
		2365
		2366	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2367	{
		2368	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2369	}
		2370
		2371	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
		2372	{
		2373	return eio__1path (EIO_STAT, path, pri, cb, data);
		2374	}
		2375
		2376	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
		2377	{
		2378	return eio__1path (EIO_LSTAT, path, pri, cb, data);
		2379	}
		2380
		2381	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2382	{
		2383	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2384	}
		2385
1037	eio_req *eio_unlink (const char *path, eio_cb cb);	2386	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
		2387	{
		2388	return eio__1path (EIO_UNLINK, path, pri, cb, data);
		2389	}
		2390
1038	eio_req *eio_rmdir (const char *path, eio_cb cb);	2391	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1039	eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */	2392	{
		2393	return eio__1path (EIO_RMDIR, path, pri, cb, data);
		2394	}
		2395
		2396	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
		2397	{
		2398	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
		2399	}
		2400
1040	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb);	2401	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1041	eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */	2402	{
1042	eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */	2403	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
		2404	}
		2405
		2406	static eio_req *
		2407	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2408	{
		2409	REQ (type); PATH;
		2410
		2411	req->flags |= EIO_FLAG_PTR2_FREE;
		2412	req->ptr2 = strdup (new_path);
		2413	if (!req->ptr2)
		2414	{
		2415	eio_api_destroy (req);
		2416	return 0;
		2417	}
		2418
		2419	SEND;
		2420	}
		2421
		2422	eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2423	{
		2424	return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
		2425	}
		2426
		2427	eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2428	{
		2429	return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
		2430	}
		2431
		2432	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2433	{
		2434	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
		2435	}
		2436
		2437	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
		2438	{
		2439	REQ (EIO_CUSTOM); req->feed = execute; SEND;
		2440	}
		2441
		2442	#endif
		2443
		2444	eio_req *eio_grp (eio_cb cb, void *data)
		2445	{
		2446	const int pri = EIO_PRI_MAX;
		2447
		2448	REQ (EIO_GROUP); SEND;
		2449	}
		2450
		2451	#undef REQ
		2452	#undef PATH
		2453	#undef SEND
		2454
		2455	/*****************************************************************************/
		2456	/* grp functions */
		2457
1043	void	2458	void
1044	aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1045	PROTOTYPE: $$$;$
1046	PPCODE:
1047	{
1048	dREQ;
1049
1050	req->type = EIO_OPEN;
1051	req->sv1 = newSVsv (pathname);
1052	req->ptr1 = SvPVbyte_nolen (req->sv1);
1053	req->int1 = flags;
1054	req->int2 = mode;
1055
1056	EIO_SEND;
1057	}
1058
1059	void
1060	aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1061	PROTOTYPE: $;$
1062	ALIAS:
1063	aio_fsync = EIO_FSYNC
1064	aio_fdatasync = EIO_FDATASYNC
1065	PPCODE:
1066	{
1067	dREQ;
1068
1069	req->type = ix;
1070	req->sv1 = newSVsv (fh);
1071	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1072
1073	EIO_SEND (req);
1074	}
1075
1076	void
1077	aio_close (SV *fh, SV *callback=&PL_sv_undef)
1078	PROTOTYPE: $;$
1079	PPCODE:
1080	{
1081	dREQ;
1082
1083	req->type = EIO_CLOSE;
1084	req->sv1 = newSVsv (fh);
1085	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1086
1087	EIO_SEND (req);
1088	}
1089
1090	void
1091	aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1092	ALIAS:
1093	aio_read = EIO_READ
1094	aio_write = EIO_WRITE
1095	PROTOTYPE: $$$$$;$
1096	PPCODE:
1097	{
1098	STRLEN svlen;
1099	char *svptr = SvPVbyte (data, svlen);
1100	UV len = SvUV (length);
1101
1102	SvUPGRADE (data, SVt_PV);
1103	SvPOK_on (data);
1104
1105	if (dataoffset < 0)
1106	dataoffset += svlen;
1107
1108	if (dataoffset < 0 || dataoffset > svlen)
1109	croak ("dataoffset outside of data scalar");
1110
1111	if (ix == EIO_WRITE)
1112	{
1113	/* write: check length and adjust. */
1114	if (!SvOK (length) || len + dataoffset > svlen)
1115	len = svlen - dataoffset;
1116	}
1117	else
1118	{
1119	/* read: grow scalar as necessary */
1120	svptr = SvGROW (data, len + dataoffset + 1);
1121	}
1122
1123	if (len < 0)
1124	croak ("length must not be negative");
1125
1126	{
1127	dREQ;
1128
1129	req->type = ix;
1130	req->sv1 = newSVsv (fh);
1131	req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1132	: IoOFP (sv_2io (fh)));
1133	req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1134	req->size = len;
1135	req->sv2 = SvREFCNT_inc (data);
1136	req->ptr2 = (char *)svptr + dataoffset;
1137	req->stroffset = dataoffset;
1138
1139	if (!SvREADONLY (data))
1140	{
1141	SvREADONLY_on (data);
1142	req->flags |= FLAG_SV2_RO_OFF;
1143	}
1144
1145	EIO_SEND;
1146	}
1147	}
1148
1149	void
1150	aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1151	PROTOTYPE: $$;$
1152	PPCODE:
1153	{
1154	SV *data;
1155	dREQ;
1156
1157	data = newSV (NAME_MAX);
1158	SvPOK_on (data);
1159
1160	req->type = EIO_READLINK;
1161	req->sv1 = newSVsv (path);
1162	req->ptr1 = SvPVbyte_nolen (req->sv1);
1163	req->sv2 = data;
1164	req->ptr2 = SvPVbyte_nolen (data);
1165
1166	EIO_SEND;
1167	}
1168
1169	void
1170	aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1171	PROTOTYPE: $$$$;$
1172	PPCODE:
1173	{
1174	dREQ;
1175
1176	req->type = EIO_SENDFILE;
1177	req->sv1 = newSVsv (out_fh);
1178	req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1179	req->sv2 = newSVsv (in_fh);
1180	req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1181	req->offs = SvVAL64 (in_offset);
1182	req->size = length;
1183
1184	EIO_SEND;
1185	}
1186
1187	void
1188	aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1189	PROTOTYPE: $$$;$
1190	PPCODE:
1191	{
1192	dREQ;
1193
1194	req->type = EIO_READAHEAD;
1195	req->sv1 = newSVsv (fh);
1196	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1197	req->offs = SvVAL64 (offset);
1198	req->size = length;
1199
1200	EIO_SEND;
1201	}
1202
1203	void
1204	aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1205	ALIAS:
1206	aio_stat = EIO_STAT
1207	aio_lstat = EIO_LSTAT
1208	PPCODE:
1209	{
1210	dREQ;
1211
1212	req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1213	if (!req->ptr2)
1214	{
1215	req_destroy (req);
1216	croak ("out of memory during aio_stat statdata allocation");
1217	}
1218
1219	req->flags |= FLAG_PTR2_FREE;
1220	req->sv1 = newSVsv (fh_or_path);
1221
1222	if (SvPOK (fh_or_path))
1223	{
1224	req->type = ix;
1225	req->ptr1 = SvPVbyte_nolen (req->sv1);
1226	}
1227	else
1228	{
1229	req->type = EIO_FSTAT;
1230	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1231	}
1232
1233	EIO_SEND;
1234	}
1235
1236	void
1237	aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1238	PPCODE:
1239	{
1240	dREQ;
1241
1242	req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1243	req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1244	req->sv1 = newSVsv (fh_or_path);
1245
1246	if (SvPOK (fh_or_path))
1247	{
1248	req->type = EIO_UTIME;
1249	req->ptr1 = SvPVbyte_nolen (req->sv1);
1250	}
1251	else
1252	{
1253	req->type = EIO_FUTIME;
1254	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1255	}
1256
1257	EIO_SEND;
1258	}
1259
1260	void
1261	aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1262	PPCODE:
1263	{
1264	dREQ;
1265
1266	req->sv1 = newSVsv (fh_or_path);
1267	req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1268
1269	if (SvPOK (fh_or_path))
1270	{
1271	req->type = EIO_TRUNCATE;
1272	req->ptr1 = SvPVbyte_nolen (req->sv1);
1273	}
1274	else
1275	{
1276	req->type = EIO_FTRUNCATE;
1277	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1278	}
1279
1280	EIO_SEND;
1281	}
1282
1283	void
1284	aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1285	ALIAS:
1286	aio_chmod = EIO_CHMOD
1287	aio_fchmod = EIO_FCHMOD
1288	aio_mkdir = EIO_MKDIR
1289	PPCODE:
1290	{
1291	dREQ;
1292
1293	req->type = type;
1294	req->int2 = mode;
1295	req->sv1 = newSVsv (fh_or_path);
1296
1297	if (ix == EIO_FCHMOD)
1298	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1299	else
1300	req->ptr1 = SvPVbyte_nolen (req->sv1);
1301
1302	EIO_SEND;
1303	}
1304
1305	void
1306	aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1307	PPCODE:
1308	{
1309	dREQ;
1310
1311	req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1312	req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1313	req->sv1 = newSVsv (fh_or_path);
1314
1315	if (SvPOK (fh_or_path))
1316	{
1317	req->type = EIO_CHOWN;
1318	req->ptr1 = SvPVbyte_nolen (req->sv1);
1319	}
1320	else
1321	{
1322	req->type = EIO_FCHOWN;
1323	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1324	}
1325
1326	EIO_SEND;
1327	}
1328
1329	void
1330	aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1331	ALIAS:
1332	aio_unlink = EIO_UNLINK
1333	aio_rmdir = EIO_RMDIR
1334	aio_readdir = EIO_READDIR
1335	PPCODE:
1336	{
1337	dREQ;
1338
1339	req->type = ix;
1340	req->sv1 = newSVsv (pathname);
1341	req->ptr1 = SvPVbyte_nolen (req->sv1);
1342
1343	EIO_SEND;
1344	}
1345
1346	void
1347	aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1348	ALIAS:
1349	aio_link = EIO_LINK
1350	aio_symlink = EIO_SYMLINK
1351	aio_rename = EIO_RENAME
1352	PPCODE:
1353	{
1354	dREQ;
1355
1356	req->type = ix;
1357	req->sv1 = newSVsv (oldpath);
1358	req->ptr1 = SvPVbyte_nolen (req->sv1);
1359	req->sv2 = newSVsv (newpath);
1360	req->ptr2 = SvPVbyte_nolen (req->sv2);
1361
1362	EIO_SEND;
1363	}
1364
1365	void
1366	aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1367	PPCODE:
1368	{
1369	dREQ;
1370
1371	req->type = EIO_MKNOD;
1372	req->sv1 = newSVsv (pathname);
1373	req->ptr1 = SvPVbyte_nolen (req->sv1);
1374	req->int2 = (mode_t)mode;
1375	req->offs = dev;
1376
1377	EIO_SEND;
1378	}
1379
1380	void
1381	aio_busy (double delay, SV *callback=&PL_sv_undef)
1382	PPCODE:
1383	{
1384	dREQ;
1385
1386	req->type = EIO_BUSY;
1387	req->nv1 = delay < 0. ? 0. : delay;
1388
1389	EIO_SEND;
1390	}
1391
1392	void
1393	aio_group (SV *callback=&PL_sv_undef)
1394	PROTOTYPE: ;$
1395	PPCODE:
1396	{
1397	dREQ;
1398
1399	req->type = EIO_GROUP;
1400
1401	req_send (req);
1402	XPUSHs (req_sv (req, AIO_GRP_KLASS));
1403	}
1404
1405	void
1406	aio_nop (SV *callback=&PL_sv_undef)
1407	ALIAS:
1408	aio_nop = EIO_NOP
1409	aio_sync = EIO_SYNC
1410	PPCODE:
1411	{
1412	dREQ;
1413
1414	#endif
1415
1416	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2459	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1417	{	2460	{
1418	grp->int2 = limit;	2461	grp->int2 = limit;
1419	grp->feed = feed;	2462	grp->feed = feed;
1420		2463
1421	grp_try_feed (grp);	2464	grp_try_feed (grp);
1422	}	2465	}
1423		2466
		2467	void
1424	void eio_grp_limit (eio_req *grp, int limit)	2468	eio_grp_limit (eio_req *grp, int limit)
1425	{	2469	{
1426	grp->int2 = limit;	2470	grp->int2 = limit;
1427		2471
1428	grp_try_feed (grp);	2472	grp_try_feed (grp);
1429	}	2473	}
1430		2474
		2475	void
1431	void eio_grp_add (eio_req *grp, eio_req *req)	2476	eio_grp_add (eio_req *grp, eio_req *req)
1432	{	2477	{
1433	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2478	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
		2479
		2480	grp->flags |= EIO_FLAG_GROUPADD;
1434		2481
1435	++grp->size;	2482	++grp->size;
1436	req->grp = grp;	2483	req->grp = grp;
1437		2484
1438	req->grp_prev = 0;	2485	req->grp_prev = 0;
…		…
1442	grp->grp_first->grp_prev = req;	2489	grp->grp_first->grp_prev = req;
1443		2490
1444	grp->grp_first = req;	2491	grp->grp_first = req;
1445	}	2492	}
1446		2493
		2494	/*****************************************************************************/
		2495	/* misc garbage */
1447		2496
		2497	eio_ssize_t
		2498	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
		2499	{
		2500	return eio__sendfile (ofd, ifd, offset, count);
		2501	}
		2502

Diff Legend

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