ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.2 by root, Sat May 10 23:33:40 2008 UTC vs.
Revision 1.102 by root, Sun Aug 21 01:03:09 2011 UTC

		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
		5	* All rights reserved.
		6	*
		7	* Redistribution and use in source and binary forms, with or without modifica-
		8	* tion, are permitted provided that the following conditions are met:
		9	*
		10	* 1. Redistributions of source code must retain the above copyright notice,
		11	* this list of conditions and the following disclaimer.
		12	*
		13	* 2. Redistributions in binary form must reproduce the above copyright
		14	* notice, this list of conditions and the following disclaimer in the
		15	* documentation and/or other materials provided with the distribution.
		16	*
		17	* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
		18	* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
		19	* CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
		20	* EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
		21	* CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
		22	* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
		23	* OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
		24	* WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
		25	* ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
		26	* OF THE POSSIBILITY OF SUCH DAMAGE.
		27	*
		28	* Alternatively, the contents of this file may be used under the terms of
		29	* the GNU General Public License ("GPL") version 2 or any later version,
		30	* in which case the provisions of the GPL are applicable instead of
		31	* the above. If you wish to allow the use of your version of this file
		32	* only under the terms of the GPL and not to allow others to use your
		33	* version of this file under the BSD license, indicate your decision
		34	* by deleting the provisions above and replace them with the notice
		35	* and other provisions required by the GPL. If you do not delete the
		36	* provisions above, a recipient may use your version of this file under
		37	* either the BSD or the GPL.
		38	*/
		39
		40	#ifndef _WIN32
		41	# include "config.h"
		42	#endif
		43
1	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
		46
		47	#ifdef EIO_STACKSIZE
		48	# define X_STACKSIZE EIO_STACKSIZE
		49	#endif
2	#include "xthread.h"	50	#include "xthread.h"
3		51
4	#include <errno.h>	52	#include <errno.h>
5
6	#include "EXTERN.h"
7	#include "perl.h"
8	#include "XSUB.h"
9
10	#include <stddef.h>	53	#include <stddef.h>
11	#include <stdlib.h>	54	#include <stdlib.h>
		55	#include <string.h>
12	#include <errno.h>	56	#include <errno.h>
13	#include <sys/types.h>	57	#include <sys/types.h>
14	#include <sys/stat.h>	58	#include <sys/stat.h>
15	#include <limits.h>	59	#include <limits.h>
16	#include <fcntl.h>	60	#include <fcntl.h>
17	#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);
18		81
19	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
20	# 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
21	#endif	84	#endif
22		85
…		…
26		89
27	#ifndef EIO_FEED	90	#ifndef EIO_FEED
28	# 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)
29	#endif	92	#endif
30		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
31	#ifdef _WIN32	105	#ifdef _WIN32
32		106
33	/*doh*/	107	#undef PAGESIZE
		108	#define PAGESIZE 4096 /* GetSystemInfo? */
		109
		110	/* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
		111
		112	#ifdef EIO_STRUCT_STATI64
		113	/* look at perl's non-sloppy stat */
		114	#define stat(path,buf) _stati64 (path,buf)
		115	#define fstat(fd,buf) _fstati64 (fd,buf)
		116	#endif
		117	#define lstat(path,buf) stat (path,buf)
		118	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		119	#define mkdir(path,mode) _mkdir (path)
		120	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		121
		122	#define chmod(path,mode) _chmod (path, mode)
		123	#define dup(fd) _dup (fd)
		124	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		125
		126	#define fchmod(fd,mode) EIO_ENOSYS ()
		127	#define chown(path,uid,gid) EIO_ENOSYS ()
		128	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		129	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		130	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		131	#define mknod(path,mode,dev) EIO_ENOSYS ()
		132	#define sync() EIO_ENOSYS ()
		133	#define readlink(path,buf,s) EIO_ENOSYS ()
		134	#define statvfs(path,buf) EIO_ENOSYS ()
		135	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		136
		137	/* rename() uses MoveFile, which fails to overwrite */
		138	#define rename(old,neu) eio__rename (old, neu)
		139
		140	static int
		141	eio__rename (const char *old, const char *neu)
		142	{
		143	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		144	return 0;
		145
		146	/* should steal _dosmaperr */
		147	switch (GetLastError ())
		148	{
		149	case ERROR_FILE_NOT_FOUND:
		150	case ERROR_PATH_NOT_FOUND:
		151	case ERROR_INVALID_DRIVE:
		152	case ERROR_NO_MORE_FILES:
		153	case ERROR_BAD_NETPATH:
		154	case ERROR_BAD_NET_NAME:
		155	case ERROR_BAD_PATHNAME:
		156	case ERROR_FILENAME_EXCED_RANGE:
		157	errno = ENOENT;
		158	break;
		159
		160	default:
		161	errno = EACCES;
		162	break;
		163	}
		164
		165	return -1;
		166	}
		167
		168	/* we could even stat and see if it exists */
		169	static int
		170	symlink (const char *old, const char *neu)
		171	{
		172	#if WINVER >= 0x0600
		173	if (CreateSymbolicLink (neu, old, 1))
		174	return 0;
		175
		176	if (CreateSymbolicLink (neu, old, 0))
		177	return 0;
		178	#endif
		179
		180	return EIO_ERRNO (ENOENT, -1);
		181	}
		182
		183	/* POSIX API only */
		184	#define CreateHardLink(neu,old,flags) 0
		185	#define CreateSymbolicLink(neu,old,flags) 0
		186
		187	struct statvfs
		188	{
		189	int dummy;
		190	};
		191
		192	#define DT_DIR EIO_DT_DIR
		193	#define DT_REG EIO_DT_REG
		194	#define D_NAME(entp) entp.cFileName
		195	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
34		196
35	#else	197	#else
36		198
37	# include "config.h"
38	# include <sys/time.h>	199	#include <sys/time.h>
39	# include <sys/select.h>	200	#include <sys/select.h>
		201	#include <sys/statvfs.h>
40	# include <unistd.h>	202	#include <unistd.h>
		203	#include <signal.h>
		204	#include <dirent.h>
		205
		206	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		207	#include <sys/mman.h>
		208	#endif
		209
		210	#define D_NAME(entp) entp->d_name
		211
		212	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		213	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		214	#define _DIRENT_HAVE_D_TYPE /* sigh */
		215	#define D_INO(de) (de)->d_fileno
		216	#define D_NAMLEN(de) (de)->d_namlen
		217	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		218	#define D_INO(de) (de)->d_ino
		219	#endif
		220
		221	#ifdef _D_EXACT_NAMLEN
		222	#undef D_NAMLEN
		223	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		224	#endif
		225
		226	#ifdef _DIRENT_HAVE_D_TYPE
		227	#define D_TYPE(de) (de)->d_type
		228	#endif
		229
		230	#ifndef EIO_STRUCT_DIRENT
		231	#define EIO_STRUCT_DIRENT struct dirent
		232	#endif
		233
		234	#endif
		235
		236	#if HAVE_UTIMES
41	# include <utime.h>	237	# include <utime.h>
42	# include <signal.h>
43
44	# ifndef EIO_STRUCT_DIRENT
45	# define EIO_STRUCT_DIRENT struct dirent
46	# endif	238	#endif
47		239
48	#endif	240	#if HAVE_SYS_SYSCALL_H
49		241	# include <sys/syscall.h>
50	# ifndef EIO_STRUCT_STAT
51	# define EIO_STRUCT_STAT struct stat
52	# endif	242	#endif
		243
		244	#if HAVE_SYS_PRCTL_H
		245	# include <sys/prctl.h>
		246	#endif
53		247
54	#if HAVE_SENDFILE	248	#if HAVE_SENDFILE
55	# if __linux	249	# if __linux
56	# include <sys/sendfile.h>	250	# include <sys/sendfile.h>
57	# elif __freebsd	251	# elif __FreeBSD__ || defined __APPLE__
58	# include <sys/socket.h>	252	# include <sys/socket.h>
59	# include <sys/uio.h>	253	# include <sys/uio.h>
60	# elif __hpux	254	# elif __hpux
61	# include <sys/socket.h>	255	# include <sys/socket.h>
62	# elif __solaris /* not yet */	256	# elif __solaris
63	# include <sys/sendfile.h>	257	# include <sys/sendfile.h>
64	# else	258	# else
65	# error sendfile support requested but not available	259	# error sendfile support requested but not available
66	# endif	260	# endif
67	#endif	261	#endif
68		262
69	/* number of seconds after which an idle threads exit */	263	#ifndef D_TYPE
70	#define IDLE_TIMEOUT 10	264	# define D_TYPE(de) 0
		265	#endif
		266	#ifndef D_INO
		267	# define D_INO(de) 0
		268	#endif
		269	#ifndef D_NAMLEN
		270	# define D_NAMLEN(entp) strlen (D_NAME (entp))
		271	#endif
71		272
72	/* used for struct dirent, AIX doesn't provide it */	273	/* used for struct dirent, AIX doesn't provide it */
73	#ifndef NAME_MAX	274	#ifndef NAME_MAX
74	# define NAME_MAX 4096	275	# define NAME_MAX 4096
75	#endif	276	#endif
76		277
		278	/* used for readlink etc. */
		279	#ifndef PATH_MAX
		280	# define PATH_MAX 4096
		281	#endif
		282
77	/* buffer size for various temporary buffers */	283	/* buffer size for various temporary buffers */
78	#define EIO_BUFSIZE 65536	284	#define EIO_BUFSIZE 65536
79		285
80	#define dBUF \	286	#define dBUF \
81	char *eio_buf; \
82	X_LOCK (wrklock); \
83	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	287	char *eio_buf = malloc (EIO_BUFSIZE); \
84	X_UNLOCK (wrklock); \	288	errno = ENOMEM; \
85	if (!eio_buf) \	289	if (!eio_buf) \
86	return -1;	290	return -1
		291
		292	#define FUBd \
		293	free (eio_buf)
87		294
88	#define EIO_TICKS ((1000000 + 1023) >> 10)	295	#define EIO_TICKS ((1000000 + 1023) >> 10)
		296
		297	#define ETP_PRI_MIN EIO_PRI_MIN
		298	#define ETP_PRI_MAX EIO_PRI_MAX
		299
		300	struct etp_worker;
		301
		302	#define ETP_REQ eio_req
		303	#define ETP_DESTROY(req) eio_destroy (req)
		304	static int eio_finish (eio_req *req);
		305	#define ETP_FINISH(req) eio_finish (req)
		306	static void eio_execute (struct etp_worker *self, eio_req *req);
		307	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		308
		309	/*****************************************************************************/
		310
		311	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		312
		313	/* calculate time difference in ~1/EIO_TICKS of a second */
		314	ecb_inline int
		315	tvdiff (struct timeval *tv1, struct timeval *tv2)
		316	{
		317	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
		318	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
		319	}
		320
		321	static unsigned int started, idle, wanted = 4;
89		322
90	static void (*want_poll_cb) (void);	323	static void (*want_poll_cb) (void);
91	static void (*done_poll_cb) (void);	324	static void (*done_poll_cb) (void);
92		325
93	static unsigned int max_poll_time = 0;	326	static unsigned int max_poll_time; /* reslock */
94	static unsigned int max_poll_reqs = 0;	327	static unsigned int max_poll_reqs; /* reslock */
95		328
96	/* calculcate time difference in ~1/EIO_TICKS of a second */	329	static unsigned int nreqs; /* reqlock */
97	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	330	static unsigned int nready; /* reqlock */
98	{	331	static unsigned int npending; /* reqlock */
99	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	332	static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
100	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	333	static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
101	}
102		334
103	static unsigned int started, idle, wanted;	335	static xmutex_t wrklock;
		336	static xmutex_t reslock;
		337	static xmutex_t reqlock;
		338	static xcond_t reqwait;
		339
		340	#if !HAVE_PREADWRITE
		341	/*
		342	* make our pread/pwrite emulation safe against themselves, but not against
		343	* normal read/write by using a mutex. slows down execution a lot,
		344	* but that's your problem, not mine.
		345	*/
		346	static xmutex_t preadwritelock;
		347	#endif
		348
		349	typedef struct etp_worker
		350	{
		351	/* locked by wrklock */
		352	struct etp_worker *prev, *next;
		353
		354	xthread_t tid;
		355
		356	/* locked by reslock, reqlock or wrklock */
		357	ETP_REQ *req; /* currently processed request */
		358
		359	#ifdef ETP_WORKER_COMMON
		360	ETP_WORKER_COMMON
		361	#endif
		362	} etp_worker;
		363
		364	static etp_worker wrk_first; /* NOT etp */
		365
		366	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		367	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
104		368
105	/* worker threads management */	369	/* worker threads management */
106	static mutex_t wrklock = X_MUTEX_INIT;
107		370
108	typedef struct worker {	371	static void ecb_cold
109	/* locked by wrklock */
110	struct worker *prev, *next;
111
112	thread_t tid;
113
114	/* locked by reslock, reqlock or wrklock */
115	eio_req *req; /* currently processed request */
116	void *dbuf;
117	DIR *dirp;
118	} worker;
119
120	static worker wrk_first = { &wrk_first, &wrk_first, 0 };
121
122	static void worker_clear (worker *wrk)	372	etp_worker_clear (etp_worker *wrk)
123	{	373	{
124	if (wrk->dirp)
125	{
126	closedir (wrk->dirp);
127	wrk->dirp = 0;
128	}
129
130	if (wrk->dbuf)
131	{
132	free (wrk->dbuf);
133	wrk->dbuf = 0;
134	}
135	}	374	}
136		375
137	static void worker_free (worker *wrk)	376	static void ecb_cold
		377	etp_worker_free (etp_worker *wrk)
138	{	378	{
139	wrk->next->prev = wrk->prev;	379	wrk->next->prev = wrk->prev;
140	wrk->prev->next = wrk->next;	380	wrk->prev->next = wrk->next;
141		381
142	free (wrk);	382	free (wrk);
143	}	383	}
144		384
145	static volatile unsigned int nreqs, nready, npending;	385	static unsigned int
146	static volatile unsigned int max_idle = 4;	386	etp_nreqs (void)
147
148	static mutex_t reslock = X_MUTEX_INIT;
149	static mutex_t reqlock = X_MUTEX_INIT;
150	static cond_t reqwait = X_COND_INIT;
151
152	unsigned int eio_nreqs (void)
153	{	387	{
		388	int retval;
		389	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		390	retval = nreqs;
		391	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154	return nreqs;	392	return retval;
155	}	393	}
156		394
157	unsigned int eio_nready (void)	395	static unsigned int
		396	etp_nready (void)
158	{	397	{
159	unsigned int retval;	398	unsigned int retval;
160		399
161	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	400	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162	retval = nready;	401	retval = nready;
163	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	402	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164		403
165	return retval;	404	return retval;
166	}	405	}
167		406
168	unsigned int eio_npending (void)	407	static unsigned int
		408	etp_npending (void)
169	{	409	{
170	unsigned int retval;	410	unsigned int retval;
171		411
172	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	412	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173	retval = npending;	413	retval = npending;
174	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	414	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175		415
176	return retval;	416	return retval;
177	}	417	}
178		418
179	unsigned int eio_nthreads (void)	419	static unsigned int
		420	etp_nthreads (void)
180	{	421	{
181	unsigned int retval;	422	unsigned int retval;
182		423
183	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	424	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184	retval = started;	425	retval = started;
…		…
191	* a somewhat faster data structure might be nice, but	432	* a somewhat faster data structure might be nice, but
192	* with 8 priorities this actually needs <20 insns	433	* with 8 priorities this actually needs <20 insns
193	* per shift, the most expensive operation.	434	* per shift, the most expensive operation.
194	*/	435	*/
195	typedef struct {	436	typedef struct {
196	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	437	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197	int size;	438	int size;
198	} reqq;	439	} etp_reqq;
199		440
200	static reqq req_queue;	441	static etp_reqq req_queue;
201	static reqq res_queue;	442	static etp_reqq res_queue;
202		443
203	static int reqq_push (reqq *q, eio_req *req)	444	static void ecb_noinline ecb_cold
		445	reqq_init (etp_reqq *q)
		446	{
		447	int pri;
		448
		449	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		450	q->qs[pri] = q->qe[pri] = 0;
		451
		452	q->size = 0;
		453	}
		454
		455	static int ecb_noinline
		456	reqq_push (etp_reqq *q, ETP_REQ *req)
204	{	457	{
205	int pri = req->pri;	458	int pri = req->pri;
206	req->next = 0;	459	req->next = 0;
207		460
208	if (q->qe[pri])	461	if (q->qe[pri])
…		…
214	q->qe[pri] = q->qs[pri] = req;	467	q->qe[pri] = q->qs[pri] = req;
215		468
216	return q->size++;	469	return q->size++;
217	}	470	}
218		471
219	static eio_req *reqq_shift (reqq *q)	472	static ETP_REQ * ecb_noinline
		473	reqq_shift (etp_reqq *q)
220	{	474	{
221	int pri;	475	int pri;
222		476
223	if (!q->size)	477	if (!q->size)
224	return 0;	478	return 0;
225		479
226	--q->size;	480	--q->size;
227		481
228	for (pri = EIO_NUM_PRI; pri--; )	482	for (pri = ETP_NUM_PRI; pri--; )
229	{	483	{
230	eio_req *req = q->qs[pri];	484	eio_req *req = q->qs[pri];
231		485
232	if (req)	486	if (req)
233	{	487	{
…		…
239	}	493	}
240		494
241	abort ();	495	abort ();
242	}	496	}
243		497
244	static void grp_try_feed (eio_req *grp)	498	static int ecb_cold
		499	etp_init (void (*want_poll)(void), void (*done_poll)(void))
245	{	500	{
246	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	501	X_MUTEX_CREATE (wrklock);
247	{	502	X_MUTEX_CREATE (reslock);
248	int old_len = grp->size;	503	X_MUTEX_CREATE (reqlock);
		504	X_COND_CREATE (reqwait);
249		505
250	EIO_FEED (grp);	506	reqq_init (&req_queue);
		507	reqq_init (&res_queue);
251		508
252	/* stop if no progress has been made */	509	wrk_first.next =
253	if (old_len == grp->size)	510	wrk_first.prev = &wrk_first;
254	{
255	grp->feed = 0;
256	break;
257	}
258	}
259	}
260		511
261	static int eio_finish (eio_req *req);	512	started = 0;
		513	idle = 0;
		514	nreqs = 0;
		515	nready = 0;
		516	npending = 0;
262		517
263	static int grp_dec (eio_req *grp)	518	want_poll_cb = want_poll;
264	{	519	done_poll_cb = done_poll;
265	--grp->size;
266		520
267	/* call feeder, if applicable */
268	grp_try_feed (grp);
269
270	/* finish, if done */
271	if (!grp->size && grp->int1)
272	return eio_finish (grp);
273	else
274	return 0;	521	return 0;
275	}	522	}
276		523
277	void eio_destroy (eio_req *req)
278	{
279	if ((req)->flags & EIO_FLAG_PTR2_FREE)
280	free (req->ptr2);
281
282	EIO_DESTROY (req);
283	}
284
285	static int eio_finish (eio_req *req)
286	{
287	int res = EIO_FINISH (req);
288
289	if (req->grp)
290	{
291	int res2;
292	eio_req *grp = req->grp;
293
294	/* unlink request */
295	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
296	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
297
298	if (grp->grp_first == req)
299	grp->grp_first = req->grp_next;
300
301	res2 = grp_dec (grp);
302
303	if (!res && res2)
304	res = res2;
305	}
306
307	eio_destroy (req);
308
309	return res;
310	}
311
312	void eio_grp_cancel (eio_req *grp)
313	{
314	for (grp = grp->grp_first; grp; grp = grp->grp_next)
315	eio_cancel (grp);
316	}
317
318	void eio_cancel (eio_req *req)
319	{
320	req->flags |= EIO_FLAG_CANCELLED;
321
322	eio_grp_cancel (req);
323	}
324
325	X_THREAD_PROC (eio_proc);	524	X_THREAD_PROC (etp_proc);
326		525
327	static void start_thread (void)	526	static void ecb_cold
		527	etp_start_thread (void)
328	{	528	{
329	worker *wrk = calloc (1, sizeof (worker));	529	etp_worker *wrk = calloc (1, sizeof (etp_worker));
330		530
331	if (!wrk)	531	/*TODO*/
332	croak ("unable to allocate worker thread data");	532	assert (("unable to allocate worker thread data", wrk));
333		533
334	X_LOCK (wrklock);	534	X_LOCK (wrklock);
335		535
336	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	536	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
337	{	537	{
338	wrk->prev = &wrk_first;	538	wrk->prev = &wrk_first;
339	wrk->next = wrk_first.next;	539	wrk->next = wrk_first.next;
340	wrk_first.next->prev = wrk;	540	wrk_first.next->prev = wrk;
341	wrk_first.next = wrk;	541	wrk_first.next = wrk;
…		…
345	free (wrk);	545	free (wrk);
346		546
347	X_UNLOCK (wrklock);	547	X_UNLOCK (wrklock);
348	}	548	}
349		549
		550	static void
350	static void maybe_start_thread (void)	551	etp_maybe_start_thread (void)
351	{	552	{
352	if (eio_nthreads () >= wanted)	553	if (ecb_expect_true (etp_nthreads () >= wanted))
353	return;	554	return;
354		555
355	/* todo: maybe use idle here, but might be less exact */	556	/* todo: maybe use idle here, but might be less exact */
356	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	557	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
357	return;	558	return;
358		559
359	start_thread ();	560	etp_start_thread ();
360	}	561	}
361		562
362	void eio_submit (eio_req *req)	563	static void ecb_cold
363	{	564	etp_end_thread (void)
364	++nreqs;
365
366	X_LOCK (reqlock);
367	++nready;
368	reqq_push (&req_queue, req);
369	X_COND_SIGNAL (reqwait);
370	X_UNLOCK (reqlock);
371
372	maybe_start_thread ();
373	}
374
375	static void end_thread (void)
376	{	565	{
377	eio_req *req = calloc (1, sizeof (eio_req));	566	eio_req *req = calloc (1, sizeof (eio_req));
378		567
379	req->type = EIO_QUIT;	568	req->type = -1;
380	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	569	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381		570
382	X_LOCK (reqlock);	571	X_LOCK (reqlock);
383	reqq_push (&req_queue, req);	572	reqq_push (&req_queue, req);
384	X_COND_SIGNAL (reqwait);	573	X_COND_SIGNAL (reqwait);
385	X_UNLOCK (reqlock);	574	X_UNLOCK (reqlock);
…		…
387	X_LOCK (wrklock);	576	X_LOCK (wrklock);
388	--started;	577	--started;
389	X_UNLOCK (wrklock);	578	X_UNLOCK (wrklock);
390	}	579	}
391		580
392	void eio_set_max_poll_time (double nseconds)	581	static int
393	{
394	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
395	max_poll_time = nseconds;
396	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397	}
398
399	void eio_set_max_poll_reqs (unsigned int maxreqs)
400	{
401	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
402	max_poll_reqs = maxreqs;
403	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404	}
405
406	void eio_set_max_idle (unsigned int nthreads)
407	{
408	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
409	max_idle = nthreads <= 0 ? 1 : nthreads;
410	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411	}
412
413	void eio_set_min_parallel (unsigned int nthreads)
414	{
415	if (wanted < nthreads)
416	wanted = nthreads;
417	}
418
419	void eio_set_max_parallel (unsigned int nthreads)
420	{
421	if (wanted > nthreads)
422	wanted = nthreads;
423
424	while (started > wanted)
425	end_thread ();
426	}
427
428	int eio_poll (void)	582	etp_poll (void)
429	{	583	{
430	int maxreqs = max_poll_reqs;	584	unsigned int maxreqs;
		585	unsigned int maxtime;
431	struct timeval tv_start, tv_now;	586	struct timeval tv_start, tv_now;
432	eio_req *req;
433		587
		588	X_LOCK (reslock);
		589	maxreqs = max_poll_reqs;
		590	maxtime = max_poll_time;
		591	X_UNLOCK (reslock);
		592
434	if (max_poll_time)	593	if (maxtime)
435	gettimeofday (&tv_start, 0);	594	gettimeofday (&tv_start, 0);
436		595
437	for (;;)	596	for (;;)
438	{	597	{
		598	ETP_REQ *req;
		599
439	maybe_start_thread ();	600	etp_maybe_start_thread ();
440		601
441	X_LOCK (reslock);	602	X_LOCK (reslock);
442	req = reqq_shift (&res_queue);	603	req = reqq_shift (&res_queue);
443		604
444	if (req)	605	if (req)
…		…
452	X_UNLOCK (reslock);	613	X_UNLOCK (reslock);
453		614
454	if (!req)	615	if (!req)
455	return 0;	616	return 0;
456		617
		618	X_LOCK (reqlock);
457	--nreqs;	619	--nreqs;
		620	X_UNLOCK (reqlock);
458		621
459	if (req->type == EIO_GROUP && req->size)	622	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
460	{	623	{
461	req->int1 = 1; /* mark request as delayed */	624	req->int1 = 1; /* mark request as delayed */
462	continue;	625	continue;
463	}	626	}
464	else	627	else
465	{	628	{
466	int res = eio_finish (req);	629	int res = ETP_FINISH (req);
467	if (res)	630	if (ecb_expect_false (res))
468	return res;	631	return res;
469	}	632	}
470		633
471	if (maxreqs && !--maxreqs)	634	if (ecb_expect_false (maxreqs && !--maxreqs))
472	break;	635	break;
473		636
474	if (max_poll_time)	637	if (maxtime)
475	{	638	{
476	gettimeofday (&tv_now, 0);	639	gettimeofday (&tv_now, 0);
477		640
478	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	641	if (tvdiff (&tv_start, &tv_now) >= maxtime)
479	break;	642	break;
480	}	643	}
481	}	644	}
482		645
483	errno = EAGAIN;	646	errno = EAGAIN;
484	return -1;	647	return -1;
485	}	648	}
486		649
		650	static void
		651	etp_cancel (ETP_REQ *req)
		652	{
		653	req->cancelled = 1;
		654
		655	eio_grp_cancel (req);
		656	}
		657
		658	static void
		659	etp_submit (ETP_REQ *req)
		660	{
		661	req->pri -= ETP_PRI_MIN;
		662
		663	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		664	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		665
		666	if (ecb_expect_false (req->type == EIO_GROUP))
		667	{
		668	/* I hope this is worth it :/ */
		669	X_LOCK (reqlock);
		670	++nreqs;
		671	X_UNLOCK (reqlock);
		672
		673	X_LOCK (reslock);
		674
		675	++npending;
		676
		677	if (!reqq_push (&res_queue, req) && want_poll_cb)
		678	want_poll_cb ();
		679
		680	X_UNLOCK (reslock);
		681	}
		682	else
		683	{
		684	X_LOCK (reqlock);
		685	++nreqs;
		686	++nready;
		687	reqq_push (&req_queue, req);
		688	X_COND_SIGNAL (reqwait);
		689	X_UNLOCK (reqlock);
		690
		691	etp_maybe_start_thread ();
		692	}
		693	}
		694
		695	static void ecb_cold
		696	etp_set_max_poll_time (double nseconds)
		697	{
		698	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		699	max_poll_time = nseconds * EIO_TICKS;
		700	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		701	}
		702
		703	static void ecb_cold
		704	etp_set_max_poll_reqs (unsigned int maxreqs)
		705	{
		706	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		707	max_poll_reqs = maxreqs;
		708	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		709	}
		710
		711	static void ecb_cold
		712	etp_set_max_idle (unsigned int nthreads)
		713	{
		714	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		715	max_idle = nthreads;
		716	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		717	}
		718
		719	static void ecb_cold
		720	etp_set_idle_timeout (unsigned int seconds)
		721	{
		722	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		723	idle_timeout = seconds;
		724	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		725	}
		726
		727	static void ecb_cold
		728	etp_set_min_parallel (unsigned int nthreads)
		729	{
		730	if (wanted < nthreads)
		731	wanted = nthreads;
		732	}
		733
		734	static void ecb_cold
		735	etp_set_max_parallel (unsigned int nthreads)
		736	{
		737	if (wanted > nthreads)
		738	wanted = nthreads;
		739
		740	while (started > wanted)
		741	etp_end_thread ();
		742	}
		743
		744	/*****************************************************************************/
		745
		746	static void
		747	grp_try_feed (eio_req *grp)
		748	{
		749	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		750	{
		751	grp->flags &= ~EIO_FLAG_GROUPADD;
		752
		753	EIO_FEED (grp);
		754
		755	/* stop if no progress has been made */
		756	if (!(grp->flags & EIO_FLAG_GROUPADD))
		757	{
		758	grp->feed = 0;
		759	break;
		760	}
		761	}
		762	}
		763
		764	static int
		765	grp_dec (eio_req *grp)
		766	{
		767	--grp->size;
		768
		769	/* call feeder, if applicable */
		770	grp_try_feed (grp);
		771
		772	/* finish, if done */
		773	if (!grp->size && grp->int1)
		774	return eio_finish (grp);
		775	else
		776	return 0;
		777	}
		778
		779	static void
		780	eio_destroy (eio_req *req)
		781	{
		782	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		783	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		784
		785	EIO_DESTROY (req);
		786	}
		787
		788	static int
		789	eio_finish (eio_req *req)
		790	{
		791	int res = EIO_FINISH (req);
		792
		793	if (req->grp)
		794	{
		795	int res2;
		796	eio_req *grp = req->grp;
		797
		798	/* unlink request */
		799	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		800	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		801
		802	if (grp->grp_first == req)
		803	grp->grp_first = req->grp_next;
		804
		805	res2 = grp_dec (grp);
		806
		807	if (!res)
		808	res = res2;
		809	}
		810
		811	eio_destroy (req);
		812
		813	return res;
		814	}
		815
		816	void
		817	eio_grp_cancel (eio_req *grp)
		818	{
		819	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		820	eio_cancel (grp);
		821	}
		822
		823	void
		824	eio_cancel (eio_req *req)
		825	{
		826	etp_cancel (req);
		827	}
		828
		829	void
		830	eio_submit (eio_req *req)
		831	{
		832	etp_submit (req);
		833	}
		834
		835	unsigned int
		836	eio_nreqs (void)
		837	{
		838	return etp_nreqs ();
		839	}
		840
		841	unsigned int
		842	eio_nready (void)
		843	{
		844	return etp_nready ();
		845	}
		846
		847	unsigned int
		848	eio_npending (void)
		849	{
		850	return etp_npending ();
		851	}
		852
		853	unsigned int ecb_cold
		854	eio_nthreads (void)
		855	{
		856	return etp_nthreads ();
		857	}
		858
		859	void ecb_cold
		860	eio_set_max_poll_time (double nseconds)
		861	{
		862	etp_set_max_poll_time (nseconds);
		863	}
		864
		865	void ecb_cold
		866	eio_set_max_poll_reqs (unsigned int maxreqs)
		867	{
		868	etp_set_max_poll_reqs (maxreqs);
		869	}
		870
		871	void ecb_cold
		872	eio_set_max_idle (unsigned int nthreads)
		873	{
		874	etp_set_max_idle (nthreads);
		875	}
		876
		877	void ecb_cold
		878	eio_set_idle_timeout (unsigned int seconds)
		879	{
		880	etp_set_idle_timeout (seconds);
		881	}
		882
		883	void ecb_cold
		884	eio_set_min_parallel (unsigned int nthreads)
		885	{
		886	etp_set_min_parallel (nthreads);
		887	}
		888
		889	void ecb_cold
		890	eio_set_max_parallel (unsigned int nthreads)
		891	{
		892	etp_set_max_parallel (nthreads);
		893	}
		894
		895	int eio_poll (void)
		896	{
		897	return etp_poll ();
		898	}
		899
487	/*****************************************************************************/	900	/*****************************************************************************/
488	/* work around various missing functions */	901	/* work around various missing functions */
489		902
490	#if !HAVE_PREADWRITE	903	#if !HAVE_PREADWRITE
		904	# undef pread
		905	# undef pwrite
491	# define pread aio_pread	906	# define pread eio__pread
492	# define pwrite aio_pwrite	907	# define pwrite eio__pwrite
493		908
494	/*	909	static eio_ssize_t
495	* make our pread/pwrite safe against themselves, but not against	910	eio__pread (int fd, void *buf, size_t count, off_t offset)
496	* normal read/write by using a mutex. slows down execution a lot,
497	* but that's your problem, not mine.
498	*/
499	static mutex_t preadwritelock = X_MUTEX_INIT;
500
501	static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
502	{	911	{
503	ssize_t res;	912	eio_ssize_t res;
504	off_t ooffset;	913	off_t ooffset;
505		914
506	X_LOCK (preadwritelock);	915	X_LOCK (preadwritelock);
507	ooffset = lseek (fd, 0, SEEK_CUR);	916	ooffset = lseek (fd, 0, SEEK_CUR);
508	lseek (fd, offset, SEEK_SET);	917	lseek (fd, offset, SEEK_SET);
…		…
511	X_UNLOCK (preadwritelock);	920	X_UNLOCK (preadwritelock);
512		921
513	return res;	922	return res;
514	}	923	}
515		924
516	static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset)	925	static eio_ssize_t
		926	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
517	{	927	{
518	ssize_t res;	928	eio_ssize_t res;
519	off_t ooffset;	929	off_t ooffset;
520		930
521	X_LOCK (preadwritelock);	931	X_LOCK (preadwritelock);
522	ooffset = lseek (fd, 0, SEEK_CUR);	932	ooffset = lseek (fd, 0, SEEK_CUR);
523	lseek (fd, offset, SEEK_SET);	933	lseek (fd, offset, SEEK_SET);
524	res = write (fd, buf, count);	934	res = write (fd, buf, count);
525	lseek (fd, offset, SEEK_SET);	935	lseek (fd, ooffset, SEEK_SET);
526	X_UNLOCK (preadwritelock);	936	X_UNLOCK (preadwritelock);
527		937
528	return res;	938	return res;
529	}	939	}
530	#endif	940	#endif
531		941
532	#ifndef HAVE_FUTIMES	942	#ifndef HAVE_UTIMES
533		943
		944	# undef utimes
534	# define utimes(path,times) aio_utimes (path, times)	945	# define utimes(path,times) eio__utimes (path, times)
535	# define futimes(fd,times) aio_futimes (fd, times)
536		946
		947	static int
537	static int aio_utimes (const char *filename, const struct timeval times[2])	948	eio__utimes (const char *filename, const struct timeval times[2])
538	{	949	{
539	if (times)	950	if (times)
540	{	951	{
541	struct utimbuf buf;	952	struct utimbuf buf;
542		953
…		…
547	}	958	}
548	else	959	else
549	return utime (filename, 0);	960	return utime (filename, 0);
550	}	961	}
551		962
		963	#endif
		964
		965	#ifndef HAVE_FUTIMES
		966
		967	# undef futimes
		968	# define futimes(fd,times) eio__futimes (fd, times)
		969
		970	static int
552	static int aio_futimes (int fd, const struct timeval tv[2])	971	eio__futimes (int fd, const struct timeval tv[2])
553	{	972	{
554	errno = ENOSYS;	973	errno = ENOSYS;
555	return -1;	974	return -1;
556	}	975	}
557		976
558	#endif	977	#endif
559		978
560	#if !HAVE_FDATASYNC	979	#if !HAVE_FDATASYNC
		980	# undef fdatasync
561	# define fdatasync fsync	981	# define fdatasync(fd) fsync (fd)
562	#endif	982	#endif
		983
		984	static int
		985	eio__syncfs (int fd)
		986	{
		987	int res;
		988
		989	#if HAVE_SYS_SYNCFS
		990	res = (int)syscall (__NR_syncfs, (int)(fd));
		991	#else
		992	res = -1;
		993	errno = ENOSYS;
		994	#endif
		995
		996	if (res < 0 && errno == ENOSYS && fd >= 0)
		997	sync ();
		998
		999	return res;
		1000	}
		1001
		1002	/* sync_file_range always needs emulation */
		1003	static int
		1004	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		1005	{
		1006	#if HAVE_SYNC_FILE_RANGE
		1007	int res;
		1008
		1009	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		1010	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		1011	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		1012	{
		1013	flags = 0
		1014	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		1015	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		1016	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		1017	}
		1018
		1019	res = sync_file_range (fd, offset, nbytes, flags);
		1020
		1021	if (!res || errno != ENOSYS)
		1022	return res;
		1023	#endif
		1024
		1025	/* even though we could play tricks with the flags, it's better to always
		1026	* call fdatasync, as that matches the expectation of its users best */
		1027	return fdatasync (fd);
		1028	}
		1029
		1030	static int
		1031	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		1032	{
		1033	#if HAVE_FALLOCATE
		1034	return fallocate (fd, mode, offset, len);
		1035	#else
		1036	errno = ENOSYS;
		1037	return -1;
		1038	#endif
		1039	}
563		1040
564	#if !HAVE_READAHEAD	1041	#if !HAVE_READAHEAD
		1042	# undef readahead
565	# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self)	1043	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566		1044
		1045	static eio_ssize_t
567	static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self)	1046	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568	{	1047	{
569	size_t todo = count;	1048	size_t todo = count;
570	dBUF;	1049	dBUF;
571		1050
572	while (todo > 0)	1051	while (todo > 0)
573	{	1052	{
574	size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;	1053	size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575		1054
576	pread (fd, aio_buf, len, offset);	1055	pread (fd, eio_buf, len, offset);
577	offset += len;	1056	offset += len;
578	todo -= len;	1057	todo -= len;
579	}	1058	}
580		1059
		1060	FUBd;
		1061
581	errno = 0;	1062	errno = 0;
582	return count;	1063	return count;
583	}	1064	}
584		1065
585	#endif	1066	#endif
586		1067
587	#if !HAVE_READDIR_R
588	# define readdir_r aio_readdir_r
589
590	static mutex_t readdirlock = X_MUTEX_INIT;
591
592	static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
593	{
594	X_DIRENT *e;
595	int errorno;
596
597	X_LOCK (readdirlock);
598
599	e = readdir (dirp);
600	errorno = errno;
601
602	if (e)
603	{
604	*res = ent;
605	strcpy (ent->d_name, e->d_name);
606	}
607	else
608	*res = 0;
609
610	X_UNLOCK (readdirlock);
611
612	errno = errorno;
613	return e ? 0 : -1;
614	}
615	#endif
616
617	/* sendfile always needs emulation */	1068	/* sendfile always needs emulation */
		1069	static eio_ssize_t
618	static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self)	1070	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
619	{	1071	{
		1072	eio_ssize_t written = 0;
620	ssize_t res;	1073	eio_ssize_t res;
621		1074
622	if (!count)	1075	if (!count)
623	return 0;	1076	return 0;
624		1077
		1078	for (;;)
		1079	{
		1080	#ifdef __APPLE__
		1081	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1082	#endif
625	#if HAVE_SENDFILE	1083	#if HAVE_SENDFILE
626	# if __linux	1084	# if __linux
		1085	off_t soffset = offset;
627	res = sendfile (ofd, ifd, &offset, count);	1086	res = sendfile (ofd, ifd, &soffset, count);
628		1087
629	# elif __freebsd	1088	# elif __FreeBSD__
630	/*	1089	/*
631	* Of course, the freebsd sendfile is a dire hack with no thoughts	1090	* Of course, the freebsd sendfile is a dire hack with no thoughts
632	* wasted on making it similar to other I/O functions.	1091	* wasted on making it similar to other I/O functions.
633	*/	1092	*/
634	{
635	off_t sbytes;	1093	off_t sbytes;
636	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1094	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637		1095
638	if (res < 0 && sbytes)	1096	#if 0 /* according to the manpage, this is correct, but broken behaviour */
639	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1097	/* freebsd' sendfile will return 0 on success */
		1098	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1099	/* not on e.g. EIO or EPIPE - sounds broken */
		1100	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
640	res = sbytes;	1101	res = sbytes;
641	}	1102	#endif
		1103
		1104	/* according to source inspection, this is correct, and useful behaviour */
		1105	if (sbytes)
		1106	res = sbytes;
		1107
		1108	# elif defined (__APPLE__)
		1109	off_t sbytes = count;
		1110	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1111
		1112	/* according to the manpage, sbytes is always valid */
		1113	if (sbytes)
		1114	res = sbytes;
642		1115
643	# elif __hpux	1116	# elif __hpux
644	res = sendfile (ofd, ifd, offset, count, 0, 0);	1117	res = sendfile (ofd, ifd, offset, count, 0, 0);
645		1118
646	# elif __solaris	1119	# elif __solaris
647	{
648	struct sendfilevec vec;	1120	struct sendfilevec vec;
649	size_t sbytes;	1121	size_t sbytes;
650		1122
651	vec.sfv_fd = ifd;	1123	vec.sfv_fd = ifd;
652	vec.sfv_flag = 0;	1124	vec.sfv_flag = 0;
653	vec.sfv_off = offset;	1125	vec.sfv_off = offset;
654	vec.sfv_len = count;	1126	vec.sfv_len = count;
655		1127
656	res = sendfilev (ofd, &vec, 1, &sbytes);	1128	res = sendfilev (ofd, &vec, 1, &sbytes);
657		1129
658	if (res < 0 && sbytes)	1130	if (res < 0 && sbytes)
659	res = sbytes;	1131	res = sbytes;
660	}
661		1132
662	# endif	1133	# endif
		1134
		1135	#elif defined (_WIN32) && 0
		1136	/* does not work, just for documentation of what would need to be done */
		1137	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1138	/* libeio guarantees that the file offset does not change, and windows */
		1139	/* has no way to get an independent handle to the same file description */
		1140	HANDLE h = TO_SOCKET (ifd);
		1141	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1142	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1143
663	#else	1144	#else
664	res = -1;	1145	res = -1;
665	errno = ENOSYS;	1146	errno = ENOSYS;
666	#endif	1147	#endif
667		1148
		1149	/* we assume sendfile can copy at least 128mb in one go */
		1150	if (res <= 128 * 1024 * 1024)
		1151	{
		1152	if (res > 0)
		1153	written += res;
		1154
		1155	if (written)
		1156	return written;
		1157
		1158	break;
		1159	}
		1160	else
		1161	{
		1162	/* if we requested more, then probably the kernel was lazy */
		1163	written += res;
		1164	offset += res;
		1165	count -= res;
		1166
		1167	if (!count)
		1168	return written;
		1169	}
		1170	}
		1171
668	if (res < 0	1172	if (res < 0
669	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1173	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1174	/* BSDs */
		1175	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1176	|| errno == ENOTSUP
		1177	#endif
		1178	#ifdef EOPNOTSUPP /* windows */
		1179	|| errno == EOPNOTSUPP /* BSDs */
		1180	#endif
670	#if __solaris	1181	#if __solaris
671	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1182	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
672	#endif	1183	#endif
673	)	1184	)
674	)	1185	)
…		…
678		1189
679	res = 0;	1190	res = 0;
680		1191
681	while (count)	1192	while (count)
682	{	1193	{
683	ssize_t cnt;	1194	eio_ssize_t cnt;
684		1195
685	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1196	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
686		1197
687	if (cnt <= 0)	1198	if (cnt <= 0)
688	{	1199	{
…		…
700		1211
701	offset += cnt;	1212	offset += cnt;
702	res += cnt;	1213	res += cnt;
703	count -= cnt;	1214	count -= cnt;
704	}	1215	}
		1216
		1217	FUBd;
705	}	1218	}
706		1219
707	return res;	1220	return res;
708	}	1221	}
709		1222
		1223	#ifdef PAGESIZE
		1224	# define eio_pagesize() PAGESIZE
		1225	#else
		1226	static intptr_t
		1227	eio_pagesize (void)
		1228	{
		1229	static intptr_t page;
		1230
		1231	if (!page)
		1232	page = sysconf (_SC_PAGESIZE);
		1233
		1234	return page;
		1235	}
		1236	#endif
		1237
		1238	static void
		1239	eio_page_align (void **addr, size_t *length)
		1240	{
		1241	intptr_t mask = eio_pagesize () - 1;
		1242
		1243	/* round down addr */
		1244	intptr_t adj = mask & (intptr_t)*addr;
		1245
		1246	*addr = (void *)((intptr_t)*addr - adj);
		1247	*length += adj;
		1248
		1249	/* round up length */
		1250	*length = (*length + mask) & ~mask;
		1251	}
		1252
		1253	#if !_POSIX_MEMLOCK
		1254	# define eio__mlockall(a) EIO_ENOSYS ()
		1255	#else
		1256
		1257	static int
		1258	eio__mlockall (int flags)
		1259	{
		1260	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1261	extern int mallopt (int, int);
		1262	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1263	#endif
		1264
		1265	if (EIO_MCL_CURRENT != MCL_CURRENT
		1266	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1267	{
		1268	flags = 0
		1269	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1270	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1271	}
		1272
		1273	return mlockall (flags);
		1274	}
		1275	#endif
		1276
		1277	#if !_POSIX_MEMLOCK_RANGE
		1278	# define eio__mlock(a,b) EIO_ENOSYS ()
		1279	#else
		1280
		1281	static int
		1282	eio__mlock (void *addr, size_t length)
		1283	{
		1284	eio_page_align (&addr, &length);
		1285
		1286	return mlock (addr, length);
		1287	}
		1288
		1289	#endif
		1290
		1291	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1292	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1293	#else
		1294
		1295	static int
		1296	eio__msync (void *mem, size_t len, int flags)
		1297	{
		1298	eio_page_align (&mem, &len);
		1299
		1300	if (EIO_MS_ASYNC != MS_SYNC
		1301	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1302	|| EIO_MS_SYNC != MS_SYNC)
		1303	{
		1304	flags = 0
		1305	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1306	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1307	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1308	}
		1309
		1310	return msync (mem, len, flags);
		1311	}
		1312
		1313	#endif
		1314
		1315	static int
		1316	eio__mtouch (eio_req *req)
		1317	{
		1318	void *mem = req->ptr2;
		1319	size_t len = req->size;
		1320	int flags = req->int1;
		1321
		1322	eio_page_align (&mem, &len);
		1323
		1324	{
		1325	intptr_t addr = (intptr_t)mem;
		1326	intptr_t end = addr + len;
		1327	intptr_t page = eio_pagesize ();
		1328
		1329	if (addr < end)
		1330	if (flags & EIO_MT_MODIFY) /* modify */
		1331	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1332	else
		1333	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1334	}
		1335
		1336	return 0;
		1337	}
		1338
		1339	/*****************************************************************************/
		1340	/* requests implemented outside eio_execute, because they are so large */
		1341
		1342	static void
		1343	eio__realpath (eio_req *req, etp_worker *self)
		1344	{
		1345	char *rel = req->ptr1;
		1346	char *res;
		1347	char *tmp1, *tmp2;
		1348	#if SYMLOOP_MAX > 32
		1349	int symlinks = SYMLOOP_MAX;
		1350	#else
		1351	int symlinks = 32;
		1352	#endif
		1353
		1354	req->result = -1;
		1355
		1356	errno = EINVAL;
		1357	if (!rel)
		1358	return;
		1359
		1360	errno = ENOENT;
		1361	if (!*rel)
		1362	return;
		1363
		1364	if (!req->ptr2)
		1365	{
		1366	X_LOCK (wrklock);
		1367	req->flags |= EIO_FLAG_PTR2_FREE;
		1368	X_UNLOCK (wrklock);
		1369	req->ptr2 = malloc (PATH_MAX * 3);
		1370
		1371	errno = ENOMEM;
		1372	if (!req->ptr2)
		1373	return;
		1374	}
		1375
		1376	res = req->ptr2;
		1377	tmp1 = res + PATH_MAX;
		1378	tmp2 = tmp1 + PATH_MAX;
		1379
		1380	#if 0 /* disabled, the musl way to do things is just too racy */
		1381	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1382	/* on linux we may be able to ask the kernel */
		1383	{
		1384	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1385
		1386	if (fd >= 0)
		1387	{
		1388	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1389	req->result = readlink (tmp1, res, PATH_MAX);
		1390	close (fd);
		1391
		1392	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1393
		1394	if (req->result > 0)
		1395	goto done;
		1396	}
		1397	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1398	return;
		1399	}
		1400	#endif
		1401	#endif
		1402
		1403	if (*rel != '/')
		1404	{
		1405	if (!getcwd (res, PATH_MAX))
		1406	return;
		1407
		1408	if (res [1]) /* only use if not / */
		1409	res += strlen (res);
		1410	}
		1411
		1412	while (*rel)
		1413	{
		1414	eio_ssize_t len, linklen;
		1415	char *beg = rel;
		1416
		1417	while (*rel && *rel != '/')
		1418	++rel;
		1419
		1420	len = rel - beg;
		1421
		1422	if (!len) /* skip slashes */
		1423	{
		1424	++rel;
		1425	continue;
		1426	}
		1427
		1428	if (beg [0] == '.')
		1429	{
		1430	if (len == 1)
		1431	continue; /* . - nop */
		1432
		1433	if (beg [1] == '.' && len == 2)
		1434	{
		1435	/* .. - back up one component, if possible */
		1436
		1437	while (res != req->ptr2)
		1438	if (*--res == '/')
		1439	break;
		1440
		1441	continue;
		1442	}
		1443	}
		1444
		1445	errno = ENAMETOOLONG;
		1446	if (res + 1 + len + 1 >= tmp1)
		1447	return;
		1448
		1449	/* copy one component */
		1450	*res = '/';
		1451	memcpy (res + 1, beg, len);
		1452
		1453	/* zero-terminate, for readlink */
		1454	res [len + 1] = 0;
		1455
		1456	/* now check if it's a symlink */
		1457	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1458
		1459	if (linklen < 0)
		1460	{
		1461	if (errno != EINVAL)
		1462	return;
		1463
		1464	/* it's a normal directory. hopefully */
		1465	res += len + 1;
		1466	}
		1467	else
		1468	{
		1469	/* yay, it was a symlink - build new path in tmp2 */
		1470	int rellen = strlen (rel);
		1471
		1472	errno = ENAMETOOLONG;
		1473	if (linklen + 1 + rellen >= PATH_MAX)
		1474	return;
		1475
		1476	errno = ELOOP;
		1477	if (!--symlinks)
		1478	return;
		1479
		1480	if (*tmp1 == '/')
		1481	res = req->ptr2; /* symlink resolves to an absolute path */
		1482
		1483	/* we need to be careful, as rel might point into tmp2 already */
		1484	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1485	tmp2 [linklen] = '/';
		1486	memcpy (tmp2, tmp1, linklen);
		1487
		1488	rel = tmp2;
		1489	}
		1490	}
		1491
		1492	/* special case for the lone root path */
		1493	if (res == req->ptr2)
		1494	*res++ = '/';
		1495
		1496	req->result = res - (char *)req->ptr2;
		1497
		1498	done:
		1499	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1500	}
		1501
		1502	static signed char
		1503	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1504	{
		1505	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1506	: a->inode < b->inode ? -1
		1507	: a->inode > b->inode ? 1
		1508	: 0;
		1509	}
		1510
		1511	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1512
		1513	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1514	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1515
		1516	static void
		1517	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1518	{
		1519	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
		1520	unsigned char *bit = bits;
		1521
		1522	assert (CHAR_BIT == 8);
		1523	assert (sizeof (eio_dirent) * 8 < 256);
		1524	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1525	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1526
		1527	if (size <= EIO_SORT_FAST)
		1528	return;
		1529
		1530	/* first prepare an array of bits to test in our radix sort */
		1531	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
		1532	/* inode_bits must contain all inodes ORed together */
		1533	/* which is used to skip bits that are 0 everywhere, which is very common */
		1534	{
		1535	eio_ino_t endianness;
		1536	int i, j;
		1537
		1538	/* we store the byte offset of byte n into byte n of "endianness" */
		1539	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1540	((unsigned char *)&endianness)[i] = i;
		1541
		1542	*bit++ = 0;
		1543
		1544	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1545	{
		1546	/* shifting off the byte offsets out of "endianness" */
		1547	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1548	endianness >>= 8;
		1549
		1550	for (j = 0; j < 8; ++j)
		1551	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
		1552	*bit++ = offs + j;
		1553	}
		1554
		1555	for (j = 0; j < 8; ++j)
		1556	if (score_bits & (1 << j))
		1557	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1558	}
		1559
		1560	/* now actually do the sorting (a variant of MSD radix sort) */
		1561	{
		1562	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
		1563	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
		1564	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
		1565	int stk_idx = 0;
		1566
		1567	base_stk [stk_idx] = dents;
		1568	end_stk [stk_idx] = dents + size;
		1569	bit_stk [stk_idx] = bit - 1;
		1570
		1571	do
		1572	{
		1573	base = base_stk [stk_idx];
		1574	end = end_stk [stk_idx];
		1575	bit = bit_stk [stk_idx];
		1576
		1577	for (;;)
		1578	{
		1579	unsigned char O = *bit >> 3;
		1580	unsigned char M = 1 << (*bit & 7);
		1581
		1582	eio_dirent *a = base;
		1583	eio_dirent *b = end;
		1584
		1585	if (b - a < EIO_SORT_CUTOFF)
		1586	break;
		1587
		1588	/* now bit-partition the array on the bit */
		1589	/* this ugly asymmetric loop seems to perform much better than typical */
		1590	/* partition algos found in the literature */
		1591	do
		1592	if (!(((unsigned char *)a)[O] & M))
		1593	++a;
		1594	else if (!(((unsigned char *)--b)[O] & M))
		1595	{
		1596	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1597	++a;
		1598	}
		1599	while (b > a);
		1600
		1601	/* next bit, or stop, if no bits left in this path */
		1602	if (!*--bit)
		1603	break;
		1604
		1605	base_stk [stk_idx] = a;
		1606	end_stk [stk_idx] = end;
		1607	bit_stk [stk_idx] = bit;
		1608	++stk_idx;
		1609
		1610	end = a;
		1611	}
		1612	}
		1613	while (stk_idx--);
		1614	}
		1615	}
		1616
		1617	static void
		1618	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1619	{
		1620	/* first move the smallest element to the front, to act as a sentinel */
		1621	{
		1622	int i;
		1623	eio_dirent *min = dents;
		1624
		1625	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1626	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1627	if (EIO_DENT_CMP (dents [i], <, *min))
		1628	min = &dents [i];
		1629
		1630	/* swap elements 0 and j (minimum) */
		1631	{
		1632	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1633	}
		1634	}
		1635
		1636	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1637	{
		1638	eio_dirent *i, *j;
		1639
		1640	for (i = dents + 1; i < dents + size; ++i)
		1641	{
		1642	eio_dirent value = *i;
		1643
		1644	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1645	j [1] = j [0];
		1646
		1647	j [1] = value;
		1648	}
		1649	}
		1650	}
		1651
		1652	static void
		1653	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1654	{
		1655	if (size <= 1)
		1656	return; /* our insertion sort relies on size > 0 */
		1657
		1658	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1659	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1660	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1661
		1662	/* use an insertion sort at the end, or for small arrays, */
		1663	/* as insertion sort is more efficient for small partitions */
		1664	eio_dent_insertion_sort (dents, size);
		1665	}
		1666
710	/* read a full directory */	1667	/* read a full directory */
		1668	static void
711	static void scandir_ (eio_req *req, worker *self)	1669	eio__scandir (eio_req *req, etp_worker *self)
712	{	1670	{
		1671	char *name, *names;
		1672	int namesalloc = 4096 - sizeof (void *) * 4;
		1673	int namesoffs = 0;
		1674	int flags = req->int1;
		1675	eio_dirent *dents = 0;
		1676	int dentalloc = 128;
		1677	int dentoffs = 0;
		1678	eio_ino_t inode_bits = 0;
		1679	#ifdef _WIN32
		1680	HANDLE dirp;
		1681	WIN32_FIND_DATA entp;
		1682	#else
713	DIR *dirp;	1683	DIR *dirp;
714	union
715	{
716	EIO_STRUCT_DIRENT d;
717	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
718	} *u;
719	EIO_STRUCT_DIRENT *entp;	1684	EIO_STRUCT_DIRENT *entp;
720	char *name, *names;	1685	#endif
721	int memlen = 4096;
722	int memofs = 0;
723	int res = 0;
724		1686
725	X_LOCK (wrklock);	1687	req->result = -1;
		1688
		1689	if (!(flags & EIO_READDIR_DENTS))
		1690	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
		1691
		1692	#ifdef _WIN32
		1693	{
		1694	int len = strlen ((const char *)req->ptr1);
		1695	char *path = malloc (MAX_PATH);
		1696	const char *fmt;
		1697
		1698	if (!len)
		1699	fmt = "./*";
		1700	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1701	fmt = "%s*";
		1702	else
		1703	fmt = "%s/*";
		1704
		1705	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1706	dirp = FindFirstFile (path, &entp);
		1707	free (path);
		1708
		1709	if (dirp == INVALID_HANDLE_VALUE)
		1710	{
		1711	dirp = 0;
		1712
		1713	/* should steal _dosmaperr */
		1714	switch (GetLastError ())
		1715	{
		1716	case ERROR_FILE_NOT_FOUND:
		1717	req->result = 0;
		1718	break;
		1719
		1720	case ERROR_INVALID_NAME:
		1721	case ERROR_PATH_NOT_FOUND:
		1722	case ERROR_NO_MORE_FILES:
		1723	errno = ENOENT;
		1724	break;
		1725
		1726	case ERROR_NOT_ENOUGH_MEMORY:
		1727	errno = ENOMEM;
		1728	break;
		1729
		1730	default:
		1731	errno = EINVAL;
		1732	break;
		1733	}
		1734	}
		1735	}
		1736	#else
726	self->dirp = dirp = opendir (req->ptr1);	1737	dirp = opendir (req->ptr1);
727	self->dbuf = u = malloc (sizeof (*u));	1738	#endif
		1739
728	req->flags |= EIO_FLAG_PTR2_FREE;	1740	if (req->flags & EIO_FLAG_PTR1_FREE)
		1741	free (req->ptr1);
		1742
		1743	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1744	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
729	req->ptr2 = names = malloc (memlen);	1745	req->ptr2 = names = malloc (namesalloc);
730	X_UNLOCK (wrklock);
731		1746
732	if (dirp && u && names)	1747	if (dirp && names && (!flags || dents))
733	for (;;)	1748	for (;;)
734	{	1749	{
		1750	int done;
		1751
		1752	#ifdef _WIN32
		1753	done = !dirp;
		1754	#else
735	errno = 0;	1755	errno = 0;
736	readdir_r (dirp, &u->d, &entp);	1756	entp = readdir (dirp);
		1757	done = !entp;
		1758	#endif
737		1759
738	if (!entp)	1760	if (done)
		1761	{
		1762	#ifndef _WIN32
		1763	int old_errno = errno;
		1764	closedir (dirp);
		1765	errno = old_errno;
		1766
		1767	if (errno)
		1768	break;
		1769	#endif
		1770
		1771	/* sort etc. */
		1772	req->int1 = flags;
		1773	req->result = dentoffs;
		1774
		1775	if (flags & EIO_READDIR_STAT_ORDER)
		1776	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1777	else if (flags & EIO_READDIR_DIRS_FIRST)
		1778	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1779	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1780	else
		1781	{
		1782	/* in this case, all is known, and we just put dirs first and sort them */
		1783	eio_dirent *oth = dents + dentoffs;
		1784	eio_dirent *dir = dents;
		1785
		1786	/* now partition dirs to the front, and non-dirs to the back */
		1787	/* by walking from both sides and swapping if necessary */
		1788	while (oth > dir)
		1789	{
		1790	if (dir->type == EIO_DT_DIR)
		1791	++dir;
		1792	else if ((--oth)->type == EIO_DT_DIR)
		1793	{
		1794	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1795
		1796	++dir;
		1797	}
		1798	}
		1799
		1800	/* now sort the dirs only (dirs all have the same score) */
		1801	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1802	}
		1803
739	break;	1804	break;
		1805	}
740		1806
741	name = entp->d_name;	1807	/* now add the entry to our list(s) */
		1808	name = D_NAME (entp);
742		1809
		1810	/* skip . and .. entries */
743	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1811	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
744	{	1812	{
745	int len = strlen (name) + 1;	1813	int len = D_NAMLEN (entp) + 1;
746		1814
747	res++;	1815	while (ecb_expect_false (namesoffs + len > namesalloc))
748
749	while (memofs + len > memlen)
750	{	1816	{
751	memlen *= 2;	1817	namesalloc *= 2;
752	X_LOCK (wrklock);
753	req->ptr2 = names = realloc (names, memlen);	1818	req->ptr2 = names = realloc (names, namesalloc);
754	X_UNLOCK (wrklock);
755		1819
756	if (!names)	1820	if (!names)
757	break;	1821	break;
758	}	1822	}
759		1823
760	memcpy (names + memofs, name, len);	1824	memcpy (names + namesoffs, name, len);
		1825
		1826	if (dents)
		1827	{
		1828	struct eio_dirent *ent;
		1829
		1830	if (ecb_expect_false (dentoffs == dentalloc))
		1831	{
		1832	dentalloc *= 2;
		1833	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1834
		1835	if (!dents)
		1836	break;
		1837	}
		1838
		1839	ent = dents + dentoffs;
		1840
		1841	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1842	ent->namelen = len - 1;
		1843	ent->inode = D_INO (entp);
		1844
		1845	inode_bits |= ent->inode;
		1846
		1847	switch (D_TYPE (entp))
		1848	{
		1849	default:
		1850	ent->type = EIO_DT_UNKNOWN;
		1851	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1852	break;
		1853
		1854	#ifdef DT_FIFO
		1855	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1856	#endif
		1857	#ifdef DT_CHR
		1858	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1859	#endif
		1860	#ifdef DT_MPC
		1861	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1862	#endif
		1863	#ifdef DT_DIR
		1864	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1865	#endif
		1866	#ifdef DT_NAM
		1867	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1868	#endif
		1869	#ifdef DT_BLK
		1870	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1871	#endif
		1872	#ifdef DT_MPB
		1873	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1874	#endif
		1875	#ifdef DT_REG
		1876	case DT_REG: ent->type = EIO_DT_REG; break;
		1877	#endif
		1878	#ifdef DT_NWK
		1879	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1880	#endif
		1881	#ifdef DT_CMP
		1882	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1883	#endif
		1884	#ifdef DT_LNK
		1885	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1886	#endif
		1887	#ifdef DT_SOCK
		1888	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1889	#endif
		1890	#ifdef DT_DOOR
		1891	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1892	#endif
		1893	#ifdef DT_WHT
		1894	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1895	#endif
		1896	}
		1897
		1898	ent->score = 7;
		1899
		1900	if (flags & EIO_READDIR_DIRS_FIRST)
		1901	{
		1902	if (ent->type == EIO_DT_UNKNOWN)
		1903	{
		1904	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1905	ent->score = 1;
		1906	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1907	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1908	}
		1909	else if (ent->type == EIO_DT_DIR)
		1910	ent->score = 0;
		1911	}
		1912	}
		1913
761	memofs += len;	1914	namesoffs += len;
		1915	++dentoffs;
762	}	1916	}
		1917
		1918	if (EIO_CANCELLED (req))
		1919	{
		1920	errno = ECANCELED;
		1921	break;
		1922	}
		1923
		1924	#ifdef _WIN32
		1925	if (!FindNextFile (dirp, &entp))
		1926	{
		1927	FindClose (dirp);
		1928	dirp = 0;
		1929	}
		1930	#endif
763	}	1931	}
764
765	if (errno)
766	res = -1;
767
768	req->result = res;
769	}	1932	}
770		1933
771	/*****************************************************************************/	1934	/*****************************************************************************/
772		1935
		1936	#define ALLOC(len) \
		1937	if (!req->ptr2) \
		1938	{ \
		1939	X_LOCK (wrklock); \
		1940	req->flags |= EIO_FLAG_PTR2_FREE; \
		1941	X_UNLOCK (wrklock); \
		1942	req->ptr2 = malloc (len); \
		1943	if (!req->ptr2) \
		1944	{ \
		1945	errno = ENOMEM; \
		1946	req->result = -1; \
		1947	break; \
		1948	} \
		1949	}
		1950
773	X_THREAD_PROC (eio_proc)	1951	X_THREAD_PROC (etp_proc)
774	{	1952	{
775	eio_req *req;	1953	ETP_REQ *req;
776	struct timespec ts;	1954	struct timespec ts;
777	worker *self = (worker *)thr_arg;	1955	etp_worker *self = (etp_worker *)thr_arg;
778		1956
		1957	#if HAVE_PRCTL_SET_NAME
		1958	prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
		1959	#endif
		1960
779	/* try to distribute timeouts somewhat randomly */	1961	/* try to distribute timeouts somewhat evenly */
780	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1962	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781		1963
782	for (;;)	1964	for (;;)
783	{	1965	{
784	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1966	ts.tv_sec = 0;
785		1967
786	X_LOCK (reqlock);	1968	X_LOCK (reqlock);
787		1969
788	for (;;)	1970	for (;;)
789	{	1971	{
790	self->req = req = reqq_shift (&req_queue);	1972	self->req = req = reqq_shift (&req_queue);
791		1973
792	if (req)	1974	if (req)
793	break;	1975	break;
794		1976
		1977	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1978	{
		1979	X_UNLOCK (reqlock);
		1980	X_LOCK (wrklock);
		1981	--started;
		1982	X_UNLOCK (wrklock);
		1983	goto quit;
		1984	}
		1985
795	++idle;	1986	++idle;
796		1987
797	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1988	if (idle <= max_idle)
		1989	/* we are allowed to idle, so do so without any timeout */
		1990	X_COND_WAIT (reqwait, reqlock);
		1991	else
798	{	1992	{
799	if (idle > max_idle)	1993	/* initialise timeout once */
800	{	1994	if (!ts.tv_sec)
801	--idle;	1995	ts.tv_sec = time (0) + idle_timeout;
802	X_UNLOCK (reqlock);
803	X_LOCK (wrklock);
804	--started;
805	X_UNLOCK (wrklock);
806	goto quit;
807	}
808		1996
809	/* we are allowed to idle, so do so without any timeout */
810	X_COND_WAIT (reqwait, reqlock);	1997	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
811	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1998	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
812	}	1999	}
813		2000
814	--idle;	2001	--idle;
815	}	2002	}
816		2003
817	--nready;	2004	--nready;
818		2005
819	X_UNLOCK (reqlock);	2006	X_UNLOCK (reqlock);
820		2007
821	errno = 0; /* strictly unnecessary */
822
823	if (!EIO_CANCELLED (req))
824	switch (req->type)	2008	if (req->type < 0)
825	{
826	case EIO_READ: req->result = req->offs >= 0
827	? pread (req->int1, req->ptr2, req->size, req->offs)
828	: read (req->int1, req->ptr2, req->size); break;
829	case EIO_WRITE: req->result = req->offs >= 0
830	? pwrite (req->int1, req->ptr2, req->size, req->offs)
831	: write (req->int1, req->ptr2, req->size); break;
832
833	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
834	case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
835
836	case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
837	case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
838	case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
839
840	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
841	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
842	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
843	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
844	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
845	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
846
847	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
848	case EIO_CLOSE: req->result = close (req->int1); break;
849	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
850	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
851	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
852	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
853	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
854	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
855	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
856	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
857	case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
858
859	case EIO_SYNC: req->result = 0; sync (); break;
860	case EIO_FSYNC: req->result = fsync (req->int1); break;
861	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
862
863	case EIO_READDIR: scandir_ (req, self); break;
864
865	case EIO_BUSY:
866	#ifdef _WIN32
867	Sleep (req->nv1 * 1000.);
868	#else
869	{
870	struct timeval tv;
871
872	tv.tv_sec = req->nv1;
873	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
874
875	req->result = select (0, 0, 0, 0, &tv);
876	}
877	#endif
878	break;
879
880	case EIO_UTIME:
881	case EIO_FUTIME:
882	{
883	struct timeval tv[2];
884	struct timeval *times;
885
886	if (req->nv1 != -1. || req->nv2 != -1.)
887	{
888	tv[0].tv_sec = req->nv1;
889	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
890	tv[1].tv_sec = req->nv2;
891	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
892
893	times = tv;
894	}
895	else
896	times = 0;
897
898
899	req->result = req->type == EIO_FUTIME
900	? futimes (req->int1, times)
901	: utimes (req->ptr1, times);
902	}
903
904	case EIO_GROUP:
905	case EIO_NOP:
906	break;
907
908	case EIO_QUIT:
909	goto quit;	2009	goto quit;
910		2010
911	default:	2011	ETP_EXECUTE (self, req);
912	req->result = -1;
913	break;
914	}
915
916	req->errorno = errno;
917		2012
918	X_LOCK (reslock);	2013	X_LOCK (reslock);
919		2014
920	++npending;	2015	++npending;
921		2016
922	if (!reqq_push (&res_queue, req) && want_poll_cb)	2017	if (!reqq_push (&res_queue, req) && want_poll_cb)
923	want_poll_cb ();	2018	want_poll_cb ();
924		2019
925	self->req = 0;	2020	self->req = 0;
926	worker_clear (self);	2021	etp_worker_clear (self);
927		2022
928	X_UNLOCK (reslock);	2023	X_UNLOCK (reslock);
929	}	2024	}
930		2025
931	quit:	2026	quit:
932	X_LOCK (wrklock);	2027	X_LOCK (wrklock);
933	worker_free (self);	2028	etp_worker_free (self);
934	X_UNLOCK (wrklock);	2029	X_UNLOCK (wrklock);
935		2030
936	return 0;	2031	return 0;
937	}	2032	}
938		2033
939	/*****************************************************************************/	2034	/*****************************************************************************/
940		2035
941	static void eio_atfork_prepare (void)	2036	int ecb_cold
		2037	eio_init (void (*want_poll)(void), void (*done_poll)(void))
942	{	2038	{
943	X_LOCK (wrklock);
944	X_LOCK (reqlock);
945	X_LOCK (reslock);
946	#if !HAVE_PREADWRITE	2039	#if !HAVE_PREADWRITE
947	X_LOCK (preadwritelock);
948	#endif
949	#if !HAVE_READDIR_R
950	X_LOCK (readdirlock);
951	#endif
952	}
953
954	static void eio_atfork_parent (void)
955	{
956	#if !HAVE_READDIR_R
957	X_UNLOCK (readdirlock);
958	#endif
959	#if !HAVE_PREADWRITE
960	X_UNLOCK (preadwritelock);	2040	X_MUTEX_CREATE (preadwritelock);
961	#endif	2041	#endif
962	X_UNLOCK (reslock);
963	X_UNLOCK (reqlock);
964	X_UNLOCK (wrklock);
965	}
966		2042
967	static void eio_atfork_child (void)	2043	return etp_init (want_poll, done_poll);
968	{	2044	}
969	eio_req *prv;
970		2045
971	while (prv = reqq_shift (&req_queue))	2046	ecb_inline void
972	eio_destroy (prv);	2047	eio_api_destroy (eio_req *req)
		2048	{
		2049	free (req);
		2050	}
973		2051
974	while (prv = reqq_shift (&res_queue))	2052	#define REQ(rtype) \
975	eio_destroy (prv);	2053	eio_req *req; \
		2054	\
		2055	req = (eio_req *)calloc (1, sizeof *req); \
		2056	if (!req) \
		2057	return 0; \
		2058	\
		2059	req->type = rtype; \
		2060	req->pri = pri; \
		2061	req->finish = cb; \
		2062	req->data = data; \
		2063	req->destroy = eio_api_destroy;
976		2064
977	while (wrk_first.next != &wrk_first)	2065	#define SEND eio_submit (req); return req
		2066
		2067	#define PATH \
		2068	req->flags |= EIO_FLAG_PTR1_FREE; \
		2069	req->ptr1 = strdup (path); \
		2070	if (!req->ptr1) \
		2071	{ \
		2072	eio_api_destroy (req); \
		2073	return 0; \
		2074	}
		2075
		2076	static void
		2077	eio_execute (etp_worker *self, eio_req *req)
		2078	{
		2079	if (ecb_expect_false (EIO_CANCELLED (req)))
978	{	2080	{
979	worker *wrk = wrk_first.next;	2081	req->result = -1;
980		2082	req->errorno = ECANCELED;
981	if (wrk->req)	2083	return;
982	eio_destroy (wrk->req);
983
984	worker_clear (wrk);
985	worker_free (wrk);
986	}	2084	}
987		2085
988	started = 0;	2086	switch (req->type)
989	idle = 0;	2087	{
990	nreqs = 0;	2088	case EIO_READ: ALLOC (req->size);
991	nready = 0;	2089	req->result = req->offs >= 0
992	npending = 0;	2090	? pread (req->int1, req->ptr2, req->size, req->offs)
		2091	: read (req->int1, req->ptr2, req->size); break;
		2092	case EIO_WRITE: req->result = req->offs >= 0
		2093	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		2094	: write (req->int1, req->ptr2, req->size); break;
993		2095
994	eio_atfork_parent ();	2096	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
995	}	2097	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
996		2098
997	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	2099	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
998	{	2100	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
999	want_poll_cb = want_poll;	2101	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1000	done_poll_cb = done_poll;	2102	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2103	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2104	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1001		2105
		2106	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2107	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2108	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2109	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2110
		2111	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		2112	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		2113	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		2114	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		2115	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		2116	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		2117
		2118	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		2119	case EIO_CLOSE: req->result = close (req->int1); break;
		2120	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		2121	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		2122	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		2123	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		2124	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		2125	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		2126	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		2127	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2128
		2129	case EIO_REALPATH: eio__realpath (req, self); break;
		2130
		2131	case EIO_READLINK: ALLOC (PATH_MAX);
		2132	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
		2133
		2134	case EIO_SYNC: req->result = 0; sync (); break;
		2135	case EIO_FSYNC: req->result = fsync (req->int1); break;
		2136	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2137	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2138	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2139	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2140	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		2141	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2142	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		2143	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
		2144
		2145	case EIO_READDIR: eio__scandir (req, self); break;
		2146
		2147	case EIO_BUSY:
1002	#ifdef _WIN32	2148	#ifdef _WIN32
1003	X_MUTEX_CHECK (wrklock);	2149	Sleep (req->nv1 * 1e3);
1004	X_MUTEX_CHECK (reslock);	2150	#else
1005	X_MUTEX_CHECK (reqlock);	2151	{
1006	X_MUTEX_CHECK (reqwait);	2152	struct timeval tv;
1007	X_MUTEX_CHECK (preadwritelock);
1008	X_MUTEX_CHECK (readdirlock);
1009		2153
1010	X_COND_CHECK (reqwait);	2154	tv.tv_sec = req->nv1;
1011	#endif	2155	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1012		2156
1013	X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);	2157	req->result = select (0, 0, 0, 0, &tv);
1014	}	2158	}
		2159	#endif
		2160	break;
1015		2161
1016	#if 0	2162	case EIO_UTIME:
		2163	case EIO_FUTIME:
		2164	{
		2165	struct timeval tv[2];
		2166	struct timeval *times;
1017		2167
		2168	if (req->nv1 != -1. || req->nv2 != -1.)
		2169	{
		2170	tv[0].tv_sec = req->nv1;
		2171	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		2172	tv[1].tv_sec = req->nv2;
		2173	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		2174
		2175	times = tv;
		2176	}
		2177	else
		2178	times = 0;
		2179
		2180	req->result = req->type == EIO_FUTIME
		2181	? futimes (req->int1, times)
		2182	: utimes (req->ptr1, times);
		2183	}
		2184	break;
		2185
		2186	case EIO_GROUP:
		2187	abort (); /* handled in eio_request */
		2188
		2189	case EIO_NOP:
		2190	req->result = 0;
		2191	break;
		2192
		2193	case EIO_CUSTOM:
		2194	req->feed (req);
		2195	break;
		2196
		2197	default:
		2198	errno = ENOSYS;
		2199	req->result = -1;
		2200	break;
		2201	}
		2202
		2203	req->errorno = errno;
		2204	}
		2205
		2206	#ifndef EIO_NO_WRAPPERS
		2207
		2208	eio_req *eio_nop (int pri, eio_cb cb, void *data)
		2209	{
		2210	REQ (EIO_NOP); SEND;
		2211	}
		2212
		2213	eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
		2214	{
		2215	REQ (EIO_BUSY); req->nv1 = delay; SEND;
		2216	}
		2217
1018	eio_req *eio_fsync (int fd, eio_cb cb);	2218	eio_req *eio_sync (int pri, eio_cb cb, void *data)
		2219	{
		2220	REQ (EIO_SYNC); SEND;
		2221	}
		2222
		2223	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
		2224	{
		2225	REQ (EIO_FSYNC); req->int1 = fd; SEND;
		2226	}
		2227
		2228	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2229	{
		2230	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2231	}
		2232
1019	eio_req *eio_fdatasync (int fd, eio_cb cb);	2233	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1020	eio_req *eio_dupclose (int fd, eio_cb cb);	2234	{
		2235	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2236	}
		2237
		2238	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2239	{
		2240	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2241	}
		2242
		2243	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2244	{
		2245	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2246	}
		2247
		2248	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2249	{
		2250	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2251	}
		2252
		2253	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2254	{
		2255	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2256	}
		2257
		2258	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2259	{
		2260	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2261	}
		2262
		2263	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2264	{
		2265	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2266	}
		2267
		2268	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
		2269	{
		2270	REQ (EIO_CLOSE); req->int1 = fd; SEND;
		2271	}
		2272
1021	eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb);	2273	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1022	eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb);	2274	{
1023	eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb);	2275	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1024	eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */	2276	}
		2277
		2278	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
		2279	{
		2280	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
		2281	}
		2282
		2283	eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
		2284	{
		2285	REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
		2286	}
		2287
		2288	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
		2289	{
		2290	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		2291	}
		2292
		2293	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2294	{
		2295	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2296	}
		2297
1025	eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb);	2298	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
		2299	{
		2300	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
		2301	}
		2302
1026	eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb);	2303	eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
		2304	{
		2305	REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
		2306	}
		2307
1027	eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb);	2308	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
		2309	{
		2310	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
		2311	}
		2312
1028	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb);	2313	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
		2314	{
		2315	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
		2316	}
		2317
1029	eio_req *eio_dup2 (int fd, int fd2, eio_cb cb);	2318	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
		2319	{
		2320	REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
		2321	}
		2322
1030	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb);	2323	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
		2324	{
		2325	REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
		2326	}
		2327
1031	eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb);	2328	eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1032	eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */	2329	{
1033	eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */	2330	REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1034	eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */	2331	}
		2332
1035	eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb);	2333	eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
		2334	{
		2335	REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
		2336	}
		2337
1036	eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb);	2338	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
		2339	{
		2340	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
		2341	}
		2342
		2343	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
		2344	{
		2345	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
		2346	}
		2347
1037	eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb);	2348	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
		2349	{
		2350	REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
		2351	}
		2352
1038	eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb);	2353	eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1039	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb);	2354	{
		2355	REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
		2356	}
		2357
		2358	static eio_req *
		2359	eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
		2360	{
		2361	REQ (type); PATH; SEND;
		2362	}
		2363
		2364	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
		2365	{
		2366	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2367	}
		2368
		2369	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2370	{
		2371	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2372	}
		2373
		2374	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
		2375	{
		2376	return eio__1path (EIO_STAT, path, pri, cb, data);
		2377	}
		2378
		2379	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
		2380	{
		2381	return eio__1path (EIO_LSTAT, path, pri, cb, data);
		2382	}
		2383
		2384	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2385	{
		2386	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2387	}
		2388
1040	eio_req *eio_unlink (const char *path, eio_cb cb);	2389	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
		2390	{
		2391	return eio__1path (EIO_UNLINK, path, pri, cb, data);
		2392	}
		2393
1041	eio_req *eio_rmdir (const char *path, eio_cb cb);	2394	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1042	eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */	2395	{
		2396	return eio__1path (EIO_RMDIR, path, pri, cb, data);
		2397	}
		2398
		2399	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
		2400	{
		2401	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
		2402	}
		2403
1043	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb);	2404	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1044	eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */	2405	{
1045	eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */	2406	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
		2407	}
		2408
		2409	static eio_req *
		2410	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2411	{
		2412	REQ (type); PATH;
		2413
		2414	req->flags |= EIO_FLAG_PTR2_FREE;
		2415	req->ptr2 = strdup (new_path);
		2416	if (!req->ptr2)
		2417	{
		2418	eio_api_destroy (req);
		2419	return 0;
		2420	}
		2421
		2422	SEND;
		2423	}
		2424
		2425	eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2426	{
		2427	return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
		2428	}
		2429
		2430	eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2431	{
		2432	return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
		2433	}
		2434
		2435	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2436	{
		2437	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
		2438	}
		2439
		2440	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
		2441	{
		2442	REQ (EIO_CUSTOM); req->feed = execute; SEND;
		2443	}
		2444
		2445	#endif
		2446
		2447	eio_req *eio_grp (eio_cb cb, void *data)
		2448	{
		2449	const int pri = EIO_PRI_MAX;
		2450
		2451	REQ (EIO_GROUP); SEND;
		2452	}
		2453
		2454	#undef REQ
		2455	#undef PATH
		2456	#undef SEND
		2457
		2458	/*****************************************************************************/
		2459	/* grp functions */
		2460
1046	void	2461	void
1047	aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1048	PROTOTYPE: $$$;$
1049	PPCODE:
1050	{
1051	dREQ;
1052
1053	req->type = EIO_OPEN;
1054	req->sv1 = newSVsv (pathname);
1055	req->ptr1 = SvPVbyte_nolen (req->sv1);
1056	req->int1 = flags;
1057	req->int2 = mode;
1058
1059	EIO_SEND;
1060	}
1061
1062	void
1063	aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1064	PROTOTYPE: $;$
1065	ALIAS:
1066	aio_fsync = EIO_FSYNC
1067	aio_fdatasync = EIO_FDATASYNC
1068	PPCODE:
1069	{
1070	dREQ;
1071
1072	req->type = ix;
1073	req->sv1 = newSVsv (fh);
1074	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1075
1076	EIO_SEND (req);
1077	}
1078
1079	void
1080	aio_close (SV *fh, SV *callback=&PL_sv_undef)
1081	PROTOTYPE: $;$
1082	PPCODE:
1083	{
1084	dREQ;
1085
1086	req->type = EIO_CLOSE;
1087	req->sv1 = newSVsv (fh);
1088	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1089
1090	EIO_SEND (req);
1091	}
1092
1093	void
1094	aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1095	ALIAS:
1096	aio_read = EIO_READ
1097	aio_write = EIO_WRITE
1098	PROTOTYPE: $$$$$;$
1099	PPCODE:
1100	{
1101	STRLEN svlen;
1102	char *svptr = SvPVbyte (data, svlen);
1103	UV len = SvUV (length);
1104
1105	SvUPGRADE (data, SVt_PV);
1106	SvPOK_on (data);
1107
1108	if (dataoffset < 0)
1109	dataoffset += svlen;
1110
1111	if (dataoffset < 0 || dataoffset > svlen)
1112	croak ("dataoffset outside of data scalar");
1113
1114	if (ix == EIO_WRITE)
1115	{
1116	/* write: check length and adjust. */
1117	if (!SvOK (length) || len + dataoffset > svlen)
1118	len = svlen - dataoffset;
1119	}
1120	else
1121	{
1122	/* read: grow scalar as necessary */
1123	svptr = SvGROW (data, len + dataoffset + 1);
1124	}
1125
1126	if (len < 0)
1127	croak ("length must not be negative");
1128
1129	{
1130	dREQ;
1131
1132	req->type = ix;
1133	req->sv1 = newSVsv (fh);
1134	req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1135	: IoOFP (sv_2io (fh)));
1136	req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1137	req->size = len;
1138	req->sv2 = SvREFCNT_inc (data);
1139	req->ptr2 = (char *)svptr + dataoffset;
1140	req->stroffset = dataoffset;
1141
1142	if (!SvREADONLY (data))
1143	{
1144	SvREADONLY_on (data);
1145	req->flags |= FLAG_SV2_RO_OFF;
1146	}
1147
1148	EIO_SEND;
1149	}
1150	}
1151
1152	void
1153	aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1154	PROTOTYPE: $$;$
1155	PPCODE:
1156	{
1157	SV *data;
1158	dREQ;
1159
1160	data = newSV (NAME_MAX);
1161	SvPOK_on (data);
1162
1163	req->type = EIO_READLINK;
1164	req->sv1 = newSVsv (path);
1165	req->ptr1 = SvPVbyte_nolen (req->sv1);
1166	req->sv2 = data;
1167	req->ptr2 = SvPVbyte_nolen (data);
1168
1169	EIO_SEND;
1170	}
1171
1172	void
1173	aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174	PROTOTYPE: $$$$;$
1175	PPCODE:
1176	{
1177	dREQ;
1178
1179	req->type = EIO_SENDFILE;
1180	req->sv1 = newSVsv (out_fh);
1181	req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182	req->sv2 = newSVsv (in_fh);
1183	req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184	req->offs = SvVAL64 (in_offset);
1185	req->size = length;
1186
1187	EIO_SEND;
1188	}
1189
1190	void
1191	aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192	PROTOTYPE: $$$;$
1193	PPCODE:
1194	{
1195	dREQ;
1196
1197	req->type = EIO_READAHEAD;
1198	req->sv1 = newSVsv (fh);
1199	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1200	req->offs = SvVAL64 (offset);
1201	req->size = length;
1202
1203	EIO_SEND;
1204	}
1205
1206	void
1207	aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1208	ALIAS:
1209	aio_stat = EIO_STAT
1210	aio_lstat = EIO_LSTAT
1211	PPCODE:
1212	{
1213	dREQ;
1214
1215	req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1216	if (!req->ptr2)
1217	{
1218	req_destroy (req);
1219	croak ("out of memory during aio_stat statdata allocation");
1220	}
1221
1222	req->flags |= FLAG_PTR2_FREE;
1223	req->sv1 = newSVsv (fh_or_path);
1224
1225	if (SvPOK (fh_or_path))
1226	{
1227	req->type = ix;
1228	req->ptr1 = SvPVbyte_nolen (req->sv1);
1229	}
1230	else
1231	{
1232	req->type = EIO_FSTAT;
1233	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1234	}
1235
1236	EIO_SEND;
1237	}
1238
1239	void
1240	aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1241	PPCODE:
1242	{
1243	dREQ;
1244
1245	req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1246	req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1247	req->sv1 = newSVsv (fh_or_path);
1248
1249	if (SvPOK (fh_or_path))
1250	{
1251	req->type = EIO_UTIME;
1252	req->ptr1 = SvPVbyte_nolen (req->sv1);
1253	}
1254	else
1255	{
1256	req->type = EIO_FUTIME;
1257	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1258	}
1259
1260	EIO_SEND;
1261	}
1262
1263	void
1264	aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1265	PPCODE:
1266	{
1267	dREQ;
1268
1269	req->sv1 = newSVsv (fh_or_path);
1270	req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1271
1272	if (SvPOK (fh_or_path))
1273	{
1274	req->type = EIO_TRUNCATE;
1275	req->ptr1 = SvPVbyte_nolen (req->sv1);
1276	}
1277	else
1278	{
1279	req->type = EIO_FTRUNCATE;
1280	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1281	}
1282
1283	EIO_SEND;
1284	}
1285
1286	void
1287	aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1288	ALIAS:
1289	aio_chmod = EIO_CHMOD
1290	aio_fchmod = EIO_FCHMOD
1291	aio_mkdir = EIO_MKDIR
1292	PPCODE:
1293	{
1294	dREQ;
1295
1296	req->type = type;
1297	req->int2 = mode;
1298	req->sv1 = newSVsv (fh_or_path);
1299
1300	if (ix == EIO_FCHMOD)
1301	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1302	else
1303	req->ptr1 = SvPVbyte_nolen (req->sv1);
1304
1305	EIO_SEND;
1306	}
1307
1308	void
1309	aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1310	PPCODE:
1311	{
1312	dREQ;
1313
1314	req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1315	req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1316	req->sv1 = newSVsv (fh_or_path);
1317
1318	if (SvPOK (fh_or_path))
1319	{
1320	req->type = EIO_CHOWN;
1321	req->ptr1 = SvPVbyte_nolen (req->sv1);
1322	}
1323	else
1324	{
1325	req->type = EIO_FCHOWN;
1326	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1327	}
1328
1329	EIO_SEND;
1330	}
1331
1332	void
1333	aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1334	ALIAS:
1335	aio_unlink = EIO_UNLINK
1336	aio_rmdir = EIO_RMDIR
1337	aio_readdir = EIO_READDIR
1338	PPCODE:
1339	{
1340	dREQ;
1341
1342	req->type = ix;
1343	req->sv1 = newSVsv (pathname);
1344	req->ptr1 = SvPVbyte_nolen (req->sv1);
1345
1346	EIO_SEND;
1347	}
1348
1349	void
1350	aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1351	ALIAS:
1352	aio_link = EIO_LINK
1353	aio_symlink = EIO_SYMLINK
1354	aio_rename = EIO_RENAME
1355	PPCODE:
1356	{
1357	dREQ;
1358
1359	req->type = ix;
1360	req->sv1 = newSVsv (oldpath);
1361	req->ptr1 = SvPVbyte_nolen (req->sv1);
1362	req->sv2 = newSVsv (newpath);
1363	req->ptr2 = SvPVbyte_nolen (req->sv2);
1364
1365	EIO_SEND;
1366	}
1367
1368	void
1369	aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1370	PPCODE:
1371	{
1372	dREQ;
1373
1374	req->type = EIO_MKNOD;
1375	req->sv1 = newSVsv (pathname);
1376	req->ptr1 = SvPVbyte_nolen (req->sv1);
1377	req->int2 = (mode_t)mode;
1378	req->offs = dev;
1379
1380	EIO_SEND;
1381	}
1382
1383	void
1384	aio_busy (double delay, SV *callback=&PL_sv_undef)
1385	PPCODE:
1386	{
1387	dREQ;
1388
1389	req->type = EIO_BUSY;
1390	req->nv1 = delay < 0. ? 0. : delay;
1391
1392	EIO_SEND;
1393	}
1394
1395	void
1396	aio_group (SV *callback=&PL_sv_undef)
1397	PROTOTYPE: ;$
1398	PPCODE:
1399	{
1400	dREQ;
1401
1402	req->type = EIO_GROUP;
1403
1404	req_send (req);
1405	XPUSHs (req_sv (req, AIO_GRP_KLASS));
1406	}
1407
1408	void
1409	aio_nop (SV *callback=&PL_sv_undef)
1410	ALIAS:
1411	aio_nop = EIO_NOP
1412	aio_sync = EIO_SYNC
1413	PPCODE:
1414	{
1415	dREQ;
1416
1417	#endif
1418
1419	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2462	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420	{	2463	{
1421	grp->int2 = limit;	2464	grp->int2 = limit;
1422	grp->feed = feed;	2465	grp->feed = feed;
1423		2466
1424	grp_try_feed (grp);	2467	grp_try_feed (grp);
1425	}	2468	}
1426		2469
		2470	void
1427	void eio_grp_limit (eio_req *grp, int limit)	2471	eio_grp_limit (eio_req *grp, int limit)
1428	{	2472	{
1429	grp->int2 = limit;	2473	grp->int2 = limit;
1430		2474
1431	grp_try_feed (grp);	2475	grp_try_feed (grp);
1432	}	2476	}
1433		2477
		2478	void
1434	void eio_grp_add (eio_req *grp, eio_req *req)	2479	eio_grp_add (eio_req *grp, eio_req *req)
1435	{	2480	{
1436	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2481	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
		2482
		2483	grp->flags |= EIO_FLAG_GROUPADD;
1437		2484
1438	++grp->size;	2485	++grp->size;
1439	req->grp = grp;	2486	req->grp = grp;
1440		2487
1441	req->grp_prev = 0;	2488	req->grp_prev = 0;
…		…
1445	grp->grp_first->grp_prev = req;	2492	grp->grp_first->grp_prev = req;
1446		2493
1447	grp->grp_first = req;	2494	grp->grp_first = req;
1448	}	2495	}
1449		2496
		2497	/*****************************************************************************/
		2498	/* misc garbage */
1450		2499
		2500	eio_ssize_t
		2501	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
		2502	{
		2503	return eio__sendfile (ofd, ifd, offset, count);
		2504	}
		2505

Diff Legend

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