ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.4 by root, Sun May 11 00:10:15 2008 UTC vs.
Revision 1.45 by root, Sat Jan 2 12:50:22 2010 UTC

		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008,2009 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
1	#include "eio.h"	40	#include "eio.h"
		41
		42	#ifdef EIO_STACKSIZE
		43	# define XTHREAD_STACKSIZE EIO_STACKSIZE
		44	#endif
2	#include "xthread.h"	45	#include "xthread.h"
3		46
4	#include <errno.h>	47	#include <errno.h>
5	#include <stddef.h>	48	#include <stddef.h>
6	#include <stdlib.h>	49	#include <stdlib.h>
…		…
8	#include <errno.h>	51	#include <errno.h>
9	#include <sys/types.h>	52	#include <sys/types.h>
10	#include <sys/stat.h>	53	#include <sys/stat.h>
11	#include <limits.h>	54	#include <limits.h>
12	#include <fcntl.h>	55	#include <fcntl.h>
13	#include <sched.h>	56	#include <assert.h>
14		57
15	#ifndef EIO_FINISH	58	#ifndef EIO_FINISH
16	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	59	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17	#endif	60	#endif
18		61
…		…
25	#endif	68	#endif
26		69
27	#ifdef _WIN32	70	#ifdef _WIN32
28		71
29	/*doh*/	72	/*doh*/
30
31	#else	73	#else
32		74
33	# include "config.h"	75	# include "config.h"
34	# include <sys/time.h>	76	# include <sys/time.h>
35	# include <sys/select.h>	77	# include <sys/select.h>
		78	# include <sys/mman.h>
36	# include <unistd.h>	79	# include <unistd.h>
37	# include <utime.h>	80	# include <utime.h>
38	# include <signal.h>	81	# include <signal.h>
39	# include <dirent.h>	82	# include <dirent.h>
40		83
		84	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		85	# if __freebsd || defined __NetBSD__ || defined __OpenBSD__
		86	# define _DIRENT_HAVE_D_TYPE /* sigh */
		87	# define D_INO(de) (de)->d_fileno
		88	# define D_NAMLEN(de) (de)->d_namlen
		89	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		90	# define D_INO(de) (de)->d_ino
		91	# endif
		92
		93	#ifdef _D_EXACT_NAMLEN
		94	# undef D_NAMLEN
		95	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		96	#endif
		97
		98	# ifdef _DIRENT_HAVE_D_TYPE
		99	# define D_TYPE(de) (de)->d_type
		100	# endif
		101
41	# ifndef EIO_STRUCT_DIRENT	102	# ifndef EIO_STRUCT_DIRENT
42	# define EIO_STRUCT_DIRENT struct dirent	103	# define EIO_STRUCT_DIRENT struct dirent
43	# endif	104	# endif
44		105
45	#endif	106	#endif
46		107
47	# ifndef EIO_STRUCT_STAT
48	# define EIO_STRUCT_STAT struct stat
49	# endif
50
51	#if HAVE_SENDFILE	108	#if HAVE_SENDFILE
52	# if __linux	109	# if __linux
53	# include <sys/sendfile.h>	110	# include <sys/sendfile.h>
54	# elif __freebsd	111	# elif __freebsd || defined __APPLE__
55	# include <sys/socket.h>	112	# include <sys/socket.h>
56	# include <sys/uio.h>	113	# include <sys/uio.h>
57	# elif __hpux	114	# elif __hpux
58	# include <sys/socket.h>	115	# include <sys/socket.h>
59	# elif __solaris /* not yet */	116	# elif __solaris
60	# include <sys/sendfile.h>	117	# include <sys/sendfile.h>
		118	# elif defined _WIN32
61	# else	119	# else
62	# error sendfile support requested but not available	120	# error sendfile support requested but not available
63	# endif	121	# endif
64	#endif	122	#endif
65		123
		124	#ifndef D_TYPE
		125	# define D_TYPE(de) 0
		126	#endif
		127	#ifndef D_INO
		128	# define D_INO(de) 0
		129	#endif
		130	#ifndef D_NAMLEN
		131	# define D_NAMLEN(de) strlen ((de)->d_name)
		132	#endif
		133
66	/* number of seconds after which an idle threads exit */	134	/* number of seconds after which an idle threads exit */
67	#define IDLE_TIMEOUT 10	135	#define IDLE_TIMEOUT 10
68		136
69	/* used for struct dirent, AIX doesn't provide it */	137	/* used for struct dirent, AIX doesn't provide it */
70	#ifndef NAME_MAX	138	#ifndef NAME_MAX
…		…
74	/* buffer size for various temporary buffers */	142	/* buffer size for various temporary buffers */
75	#define EIO_BUFSIZE 65536	143	#define EIO_BUFSIZE 65536
76		144
77	#define dBUF \	145	#define dBUF \
78	char *eio_buf; \	146	char *eio_buf; \
79	X_LOCK (wrklock); \	147	ETP_WORKER_LOCK (self); \
80	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	148	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81	X_UNLOCK (wrklock); \	149	ETP_WORKER_UNLOCK (self); \
		150	errno = ENOMEM; \
82	if (!eio_buf) \	151	if (!eio_buf) \
83	return -1;	152	return -1;
84		153
85	#define EIO_TICKS ((1000000 + 1023) >> 10)	154	#define EIO_TICKS ((1000000 + 1023) >> 10)
86		155
		156	/*****************************************************************************/
		157
		158	#if __GNUC__ >= 3
		159	# define expect(expr,value) __builtin_expect ((expr),(value))
		160	#else
		161	# define expect(expr,value) (expr)
		162	#endif
		163
		164	#define expect_false(expr) expect ((expr) != 0, 0)
		165	#define expect_true(expr) expect ((expr) != 0, 1)
		166
		167	/*****************************************************************************/
		168
		169	#define ETP_PRI_MIN EIO_PRI_MIN
		170	#define ETP_PRI_MAX EIO_PRI_MAX
		171
		172	struct etp_worker;
		173
		174	#define ETP_REQ eio_req
		175	#define ETP_DESTROY(req) eio_destroy (req)
		176	static int eio_finish (eio_req *req);
		177	#define ETP_FINISH(req) eio_finish (req)
		178	static void eio_execute (struct etp_worker *self, eio_req *req);
		179	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		180
		181	#define ETP_WORKER_CLEAR(req) \
		182	if (wrk->dbuf) \
		183	{ \
		184	free (wrk->dbuf); \
		185	wrk->dbuf = 0; \
		186	} \
		187	\
		188	if (wrk->dirp) \
		189	{ \
		190	closedir (wrk->dirp); \
		191	wrk->dirp = 0; \
		192	}
		193
		194	#define ETP_WORKER_COMMON \
		195	void *dbuf; \
		196	DIR *dirp;
		197
		198	/*****************************************************************************/
		199
		200	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		201
		202	/* calculate time difference in ~1/EIO_TICKS of a second */
		203	static int tvdiff (struct timeval *tv1, struct timeval *tv2)
		204	{
		205	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
		206	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
		207	}
		208
		209	static unsigned int started, idle, wanted = 4;
		210
87	static void (*want_poll_cb) (void);	211	static void (*want_poll_cb) (void);
88	static void (*done_poll_cb) (void);	212	static void (*done_poll_cb) (void);
89		213
90	static unsigned int max_poll_time = 0;	214	static unsigned int max_poll_time; /* reslock */
91	static unsigned int max_poll_reqs = 0;	215	static unsigned int max_poll_reqs; /* reslock */
92		216
93	/* calculcate time difference in ~1/EIO_TICKS of a second */	217	static volatile unsigned int nreqs; /* reqlock */
94	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	218	static volatile unsigned int nready; /* reqlock */
95	{	219	static volatile unsigned int npending; /* reqlock */
96	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	220	static volatile unsigned int max_idle = 4;
97	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
98	}
99		221
100	static unsigned int started, idle, wanted;
101
102	/* worker threads management */
103	static mutex_t wrklock = X_MUTEX_INIT;	222	static mutex_t wrklock = X_MUTEX_INIT;
104
105	typedef struct worker {
106	/* locked by wrklock */
107	struct worker *prev, *next;
108
109	thread_t tid;
110
111	/* locked by reslock, reqlock or wrklock */
112	eio_req *req; /* currently processed request */
113	void *dbuf;
114	DIR *dirp;
115	} worker;
116
117	static worker wrk_first = { &wrk_first, &wrk_first, 0 };
118
119	static void worker_clear (worker *wrk)
120	{
121	if (wrk->dirp)
122	{
123	closedir (wrk->dirp);
124	wrk->dirp = 0;
125	}
126
127	if (wrk->dbuf)
128	{
129	free (wrk->dbuf);
130	wrk->dbuf = 0;
131	}
132	}
133
134	static void worker_free (worker *wrk)
135	{
136	wrk->next->prev = wrk->prev;
137	wrk->prev->next = wrk->next;
138
139	free (wrk);
140	}
141
142	static volatile unsigned int nreqs, nready, npending;
143	static volatile unsigned int max_idle = 4;
144
145	static mutex_t reslock = X_MUTEX_INIT;	223	static mutex_t reslock = X_MUTEX_INIT;
146	static mutex_t reqlock = X_MUTEX_INIT;	224	static mutex_t reqlock = X_MUTEX_INIT;
147	static cond_t reqwait = X_COND_INIT;	225	static cond_t reqwait = X_COND_INIT;
148		226
		227	#if !HAVE_PREADWRITE
		228	/*
		229	* make our pread/pwrite emulation safe against themselves, but not against
		230	* normal read/write by using a mutex. slows down execution a lot,
		231	* but that's your problem, not mine.
		232	*/
		233	static mutex_t preadwritelock = X_MUTEX_INIT;
		234	#endif
		235
		236	typedef struct etp_worker
		237	{
		238	/* locked by wrklock */
		239	struct etp_worker *prev, *next;
		240
		241	thread_t tid;
		242
		243	/* locked by reslock, reqlock or wrklock */
		244	ETP_REQ *req; /* currently processed request */
		245
		246	ETP_WORKER_COMMON
		247	} etp_worker;
		248
		249	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
		250
		251	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		252	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
		253
		254	/* worker threads management */
		255
		256	static void etp_worker_clear (etp_worker *wrk)
		257	{
		258	ETP_WORKER_CLEAR (wrk);
		259	}
		260
		261	static void etp_worker_free (etp_worker *wrk)
		262	{
		263	wrk->next->prev = wrk->prev;
		264	wrk->prev->next = wrk->next;
		265
		266	free (wrk);
		267	}
		268
149	unsigned int eio_nreqs (void)	269	static unsigned int etp_nreqs (void)
150	{	270	{
		271	int retval;
		272	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		273	retval = nreqs;
		274	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151	return nreqs;	275	return retval;
152	}	276	}
153		277
154	unsigned int eio_nready (void)	278	static unsigned int etp_nready (void)
155	{	279	{
156	unsigned int retval;	280	unsigned int retval;
157		281
158	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	282	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159	retval = nready;	283	retval = nready;
160	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	284	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161		285
162	return retval;	286	return retval;
163	}	287	}
164		288
165	unsigned int eio_npending (void)	289	static unsigned int etp_npending (void)
166	{	290	{
167	unsigned int retval;	291	unsigned int retval;
168		292
169	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	293	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170	retval = npending;	294	retval = npending;
171	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	295	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172		296
173	return retval;	297	return retval;
174	}	298	}
175		299
176	unsigned int eio_nthreads (void)	300	static unsigned int etp_nthreads (void)
177	{	301	{
178	unsigned int retval;	302	unsigned int retval;
179		303
180	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	304	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181	retval = started;	305	retval = started;
…		…
188	* a somewhat faster data structure might be nice, but	312	* a somewhat faster data structure might be nice, but
189	* with 8 priorities this actually needs <20 insns	313	* with 8 priorities this actually needs <20 insns
190	* per shift, the most expensive operation.	314	* per shift, the most expensive operation.
191	*/	315	*/
192	typedef struct {	316	typedef struct {
193	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	317	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194	int size;	318	int size;
195	} reqq;	319	} etp_reqq;
196		320
197	static reqq req_queue;	321	static etp_reqq req_queue;
198	static reqq res_queue;	322	static etp_reqq res_queue;
199		323
200	static int reqq_push (reqq *q, eio_req *req)	324	static int reqq_push (etp_reqq *q, ETP_REQ *req)
201	{	325	{
202	int pri = req->pri;	326	int pri = req->pri;
203	req->next = 0;	327	req->next = 0;
204		328
205	if (q->qe[pri])	329	if (q->qe[pri])
…		…
211	q->qe[pri] = q->qs[pri] = req;	335	q->qe[pri] = q->qs[pri] = req;
212		336
213	return q->size++;	337	return q->size++;
214	}	338	}
215		339
216	static eio_req *reqq_shift (reqq *q)	340	static ETP_REQ *reqq_shift (etp_reqq *q)
217	{	341	{
218	int pri;	342	int pri;
219		343
220	if (!q->size)	344	if (!q->size)
221	return 0;	345	return 0;
222		346
223	--q->size;	347	--q->size;
224		348
225	for (pri = EIO_NUM_PRI; pri--; )	349	for (pri = ETP_NUM_PRI; pri--; )
226	{	350	{
227	eio_req *req = q->qs[pri];	351	eio_req *req = q->qs[pri];
228		352
229	if (req)	353	if (req)
230	{	354	{
…		…
236	}	360	}
237		361
238	abort ();	362	abort ();
239	}	363	}
240		364
241	static void grp_try_feed (eio_req *grp)	365	static void etp_atfork_prepare (void)
242	{	366	{
243	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	367	X_LOCK (wrklock);
		368	X_LOCK (reqlock);
		369	X_LOCK (reslock);
		370	#if !HAVE_PREADWRITE
		371	X_LOCK (preadwritelock);
		372	#endif
		373	}
		374
		375	static void etp_atfork_parent (void)
		376	{
		377	#if !HAVE_PREADWRITE
		378	X_UNLOCK (preadwritelock);
		379	#endif
		380	X_UNLOCK (reslock);
		381	X_UNLOCK (reqlock);
		382	X_UNLOCK (wrklock);
		383	}
		384
		385	static void etp_atfork_child (void)
		386	{
		387	ETP_REQ *prv;
		388
		389	while ((prv = reqq_shift (&req_queue)))
		390	ETP_DESTROY (prv);
		391
		392	while ((prv = reqq_shift (&res_queue)))
		393	ETP_DESTROY (prv);
		394
		395	while (wrk_first.next != &wrk_first)
244	{	396	{
245	int old_len = grp->size;	397	etp_worker *wrk = wrk_first.next;
246		398
247	EIO_FEED (grp);	399	if (wrk->req)
		400	ETP_DESTROY (wrk->req);
248		401
249	/* stop if no progress has been made */	402	etp_worker_clear (wrk);
250	if (old_len == grp->size)	403	etp_worker_free (wrk);
251	{
252	grp->feed = 0;
253	break;
254	}
255	}	404	}
256	}
257		405
258	static int eio_finish (eio_req *req);	406	started = 0;
		407	idle = 0;
		408	nreqs = 0;
		409	nready = 0;
		410	npending = 0;
259		411
260	static int grp_dec (eio_req *grp)	412	etp_atfork_parent ();
261	{	413	}
262	--grp->size;
263		414
264	/* call feeder, if applicable */	415	static void
265	grp_try_feed (grp);	416	etp_once_init (void)
		417	{
		418	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
		419	}
266		420
267	/* finish, if done */	421	static int
268	if (!grp->size && grp->int1)	422	etp_init (void (*want_poll)(void), void (*done_poll)(void))
269	return eio_finish (grp);	423	{
270	else	424	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
		425
		426	pthread_once (&doinit, etp_once_init);
		427
		428	want_poll_cb = want_poll;
		429	done_poll_cb = done_poll;
		430
271	return 0;	431	return 0;
272	}	432	}
273		433
274	void eio_destroy (eio_req *req)
275	{
276	if ((req)->flags & EIO_FLAG_PTR2_FREE)
277	free (req->ptr2);
278
279	EIO_DESTROY (req);
280	}
281
282	static int eio_finish (eio_req *req)
283	{
284	int res = EIO_FINISH (req);
285
286	if (req->grp)
287	{
288	int res2;
289	eio_req *grp = req->grp;
290
291	/* unlink request */
292	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
293	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
294
295	if (grp->grp_first == req)
296	grp->grp_first = req->grp_next;
297
298	res2 = grp_dec (grp);
299
300	if (!res && res2)
301	res = res2;
302	}
303
304	eio_destroy (req);
305
306	return res;
307	}
308
309	void eio_grp_cancel (eio_req *grp)
310	{
311	for (grp = grp->grp_first; grp; grp = grp->grp_next)
312	eio_cancel (grp);
313	}
314
315	void eio_cancel (eio_req *req)
316	{
317	req->flags |= EIO_FLAG_CANCELLED;
318
319	eio_grp_cancel (req);
320	}
321
322	X_THREAD_PROC (eio_proc);	434	X_THREAD_PROC (etp_proc);
323		435
324	static void start_thread (void)	436	static void etp_start_thread (void)
325	{	437	{
326	worker *wrk = calloc (1, sizeof (worker));	438	etp_worker *wrk = calloc (1, sizeof (etp_worker));
327		439
328	if (!wrk)	440	/*TODO*/
329	croak ("unable to allocate worker thread data");	441	assert (("unable to allocate worker thread data", wrk));
330		442
331	X_LOCK (wrklock);	443	X_LOCK (wrklock);
332		444
333	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	445	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334	{	446	{
335	wrk->prev = &wrk_first;	447	wrk->prev = &wrk_first;
336	wrk->next = wrk_first.next;	448	wrk->next = wrk_first.next;
337	wrk_first.next->prev = wrk;	449	wrk_first.next->prev = wrk;
338	wrk_first.next = wrk;	450	wrk_first.next = wrk;
…		…
342	free (wrk);	454	free (wrk);
343		455
344	X_UNLOCK (wrklock);	456	X_UNLOCK (wrklock);
345	}	457	}
346		458
347	static void maybe_start_thread (void)	459	static void etp_maybe_start_thread (void)
348	{	460	{
349	if (eio_nthreads () >= wanted)	461	if (expect_true (etp_nthreads () >= wanted))
350	return;	462	return;
351		463
352	/* todo: maybe use idle here, but might be less exact */	464	/* todo: maybe use idle here, but might be less exact */
353	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	465	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354	return;	466	return;
355		467
356	start_thread ();	468	etp_start_thread ();
357	}	469	}
358		470
359	void eio_submit (eio_req *req)
360	{
361	++nreqs;
362
363	X_LOCK (reqlock);
364	++nready;
365	reqq_push (&req_queue, req);
366	X_COND_SIGNAL (reqwait);
367	X_UNLOCK (reqlock);
368
369	maybe_start_thread ();
370	}
371
372	static void end_thread (void)	471	static void etp_end_thread (void)
373	{	472	{
374	eio_req *req = calloc (1, sizeof (eio_req));	473	eio_req *req = calloc (1, sizeof (eio_req));
375		474
376	req->type = EIO_QUIT;	475	req->type = -1;
377	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	476	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
378		477
379	X_LOCK (reqlock);	478	X_LOCK (reqlock);
380	reqq_push (&req_queue, req);	479	reqq_push (&req_queue, req);
381	X_COND_SIGNAL (reqwait);	480	X_COND_SIGNAL (reqwait);
382	X_UNLOCK (reqlock);	481	X_UNLOCK (reqlock);
…		…
384	X_LOCK (wrklock);	483	X_LOCK (wrklock);
385	--started;	484	--started;
386	X_UNLOCK (wrklock);	485	X_UNLOCK (wrklock);
387	}	486	}
388		487
389	void eio_set_max_poll_time (double nseconds)
390	{
391	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
392	max_poll_time = nseconds;
393	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
394	}
395
396	void eio_set_max_poll_reqs (unsigned int maxreqs)
397	{
398	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
399	max_poll_reqs = maxreqs;
400	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
401	}
402
403	void eio_set_max_idle (unsigned int nthreads)
404	{
405	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
406	max_idle = nthreads <= 0 ? 1 : nthreads;
407	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
408	}
409
410	void eio_set_min_parallel (unsigned int nthreads)
411	{
412	if (wanted < nthreads)
413	wanted = nthreads;
414	}
415
416	void eio_set_max_parallel (unsigned int nthreads)
417	{
418	if (wanted > nthreads)
419	wanted = nthreads;
420
421	while (started > wanted)
422	end_thread ();
423	}
424
425	int eio_poll (void)	488	static int etp_poll (void)
426	{	489	{
427	int maxreqs = max_poll_reqs;	490	unsigned int maxreqs;
		491	unsigned int maxtime;
428	struct timeval tv_start, tv_now;	492	struct timeval tv_start, tv_now;
429	eio_req *req;
430		493
		494	X_LOCK (reslock);
		495	maxreqs = max_poll_reqs;
		496	maxtime = max_poll_time;
		497	X_UNLOCK (reslock);
		498
431	if (max_poll_time)	499	if (maxtime)
432	gettimeofday (&tv_start, 0);	500	gettimeofday (&tv_start, 0);
433		501
434	for (;;)	502	for (;;)
435	{	503	{
		504	ETP_REQ *req;
		505
436	maybe_start_thread ();	506	etp_maybe_start_thread ();
437		507
438	X_LOCK (reslock);	508	X_LOCK (reslock);
439	req = reqq_shift (&res_queue);	509	req = reqq_shift (&res_queue);
440		510
441	if (req)	511	if (req)
…		…
449	X_UNLOCK (reslock);	519	X_UNLOCK (reslock);
450		520
451	if (!req)	521	if (!req)
452	return 0;	522	return 0;
453		523
		524	X_LOCK (reqlock);
454	--nreqs;	525	--nreqs;
		526	X_UNLOCK (reqlock);
455		527
456	if (req->type == EIO_GROUP && req->size)	528	if (expect_false (req->type == EIO_GROUP && req->size))
457	{	529	{
458	req->int1 = 1; /* mark request as delayed */	530	req->int1 = 1; /* mark request as delayed */
459	continue;	531	continue;
460	}	532	}
461	else	533	else
462	{	534	{
463	int res = eio_finish (req);	535	int res = ETP_FINISH (req);
464	if (res)	536	if (expect_false (res))
465	return res;	537	return res;
466	}	538	}
467		539
468	if (maxreqs && !--maxreqs)	540	if (expect_false (maxreqs && !--maxreqs))
469	break;	541	break;
470		542
471	if (max_poll_time)	543	if (maxtime)
472	{	544	{
473	gettimeofday (&tv_now, 0);	545	gettimeofday (&tv_now, 0);
474		546
475	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	547	if (tvdiff (&tv_start, &tv_now) >= maxtime)
476	break;	548	break;
477	}	549	}
478	}	550	}
479		551
480	errno = EAGAIN;	552	errno = EAGAIN;
481	return -1;	553	return -1;
482	}	554	}
483		555
		556	static void etp_cancel (ETP_REQ *req)
		557	{
		558	X_LOCK (wrklock);
		559	req->flags |= EIO_FLAG_CANCELLED;
		560	X_UNLOCK (wrklock);
		561
		562	eio_grp_cancel (req);
		563	}
		564
		565	static void etp_submit (ETP_REQ *req)
		566	{
		567	req->pri -= ETP_PRI_MIN;
		568
		569	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		570	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		571
		572	if (expect_false (req->type == EIO_GROUP))
		573	{
		574	/* I hope this is worth it :/ */
		575	X_LOCK (reqlock);
		576	++nreqs;
		577	X_UNLOCK (reqlock);
		578
		579	X_LOCK (reslock);
		580
		581	++npending;
		582
		583	if (!reqq_push (&res_queue, req) && want_poll_cb)
		584	want_poll_cb ();
		585
		586	X_UNLOCK (reslock);
		587	}
		588	else
		589	{
		590	X_LOCK (reqlock);
		591	++nreqs;
		592	++nready;
		593	reqq_push (&req_queue, req);
		594	X_COND_SIGNAL (reqwait);
		595	X_UNLOCK (reqlock);
		596
		597	etp_maybe_start_thread ();
		598	}
		599	}
		600
		601	static void etp_set_max_poll_time (double nseconds)
		602	{
		603	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		604	max_poll_time = nseconds * EIO_TICKS;
		605	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		606	}
		607
		608	static void etp_set_max_poll_reqs (unsigned int maxreqs)
		609	{
		610	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		611	max_poll_reqs = maxreqs;
		612	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		613	}
		614
		615	static void etp_set_max_idle (unsigned int nthreads)
		616	{
		617	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		618	max_idle = nthreads <= 0 ? 1 : nthreads;
		619	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		620	}
		621
		622	static void etp_set_min_parallel (unsigned int nthreads)
		623	{
		624	if (wanted < nthreads)
		625	wanted = nthreads;
		626	}
		627
		628	static void etp_set_max_parallel (unsigned int nthreads)
		629	{
		630	if (wanted > nthreads)
		631	wanted = nthreads;
		632
		633	while (started > wanted)
		634	etp_end_thread ();
		635	}
		636
		637	/*****************************************************************************/
		638
		639	static void grp_try_feed (eio_req *grp)
		640	{
		641	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		642	{
		643	grp->flags &= ~EIO_FLAG_GROUPADD;
		644
		645	EIO_FEED (grp);
		646
		647	/* stop if no progress has been made */
		648	if (!(grp->flags & EIO_FLAG_GROUPADD))
		649	{
		650	grp->feed = 0;
		651	break;
		652	}
		653	}
		654	}
		655
		656	static int grp_dec (eio_req *grp)
		657	{
		658	--grp->size;
		659
		660	/* call feeder, if applicable */
		661	grp_try_feed (grp);
		662
		663	/* finish, if done */
		664	if (!grp->size && grp->int1)
		665	return eio_finish (grp);
		666	else
		667	return 0;
		668	}
		669
		670	void eio_destroy (eio_req *req)
		671	{
		672	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		673	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		674
		675	EIO_DESTROY (req);
		676	}
		677
		678	static int eio_finish (eio_req *req)
		679	{
		680	int res = EIO_FINISH (req);
		681
		682	if (req->grp)
		683	{
		684	int res2;
		685	eio_req *grp = req->grp;
		686
		687	/* unlink request */
		688	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		689	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		690
		691	if (grp->grp_first == req)
		692	grp->grp_first = req->grp_next;
		693
		694	res2 = grp_dec (grp);
		695
		696	if (!res && res2)
		697	res = res2;
		698	}
		699
		700	eio_destroy (req);
		701
		702	return res;
		703	}
		704
		705	void eio_grp_cancel (eio_req *grp)
		706	{
		707	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		708	eio_cancel (grp);
		709	}
		710
		711	void eio_cancel (eio_req *req)
		712	{
		713	etp_cancel (req);
		714	}
		715
		716	void eio_submit (eio_req *req)
		717	{
		718	etp_submit (req);
		719	}
		720
		721	unsigned int eio_nreqs (void)
		722	{
		723	return etp_nreqs ();
		724	}
		725
		726	unsigned int eio_nready (void)
		727	{
		728	return etp_nready ();
		729	}
		730
		731	unsigned int eio_npending (void)
		732	{
		733	return etp_npending ();
		734	}
		735
		736	unsigned int eio_nthreads (void)
		737	{
		738	return etp_nthreads ();
		739	}
		740
		741	void eio_set_max_poll_time (double nseconds)
		742	{
		743	etp_set_max_poll_time (nseconds);
		744	}
		745
		746	void eio_set_max_poll_reqs (unsigned int maxreqs)
		747	{
		748	etp_set_max_poll_reqs (maxreqs);
		749	}
		750
		751	void eio_set_max_idle (unsigned int nthreads)
		752	{
		753	etp_set_max_idle (nthreads);
		754	}
		755
		756	void eio_set_min_parallel (unsigned int nthreads)
		757	{
		758	etp_set_min_parallel (nthreads);
		759	}
		760
		761	void eio_set_max_parallel (unsigned int nthreads)
		762	{
		763	etp_set_max_parallel (nthreads);
		764	}
		765
		766	int eio_poll (void)
		767	{
		768	return etp_poll ();
		769	}
		770
484	/*****************************************************************************/	771	/*****************************************************************************/
485	/* work around various missing functions */	772	/* work around various missing functions */
486		773
487	#if !HAVE_PREADWRITE	774	#if !HAVE_PREADWRITE
		775	# undef pread
		776	# undef pwrite
488	# define pread aio_pread	777	# define pread eio__pread
489	# define pwrite aio_pwrite	778	# define pwrite eio__pwrite
490		779
491	/*	780	static ssize_t
492	* make our pread/pwrite safe against themselves, but not against	781	eio__pread (int fd, void *buf, size_t count, off_t offset)
493	* normal read/write by using a mutex. slows down execution a lot,
494	* but that's your problem, not mine.
495	*/
496	static mutex_t preadwritelock = X_MUTEX_INIT;
497
498	static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
499	{	782	{
500	ssize_t res;	783	ssize_t res;
501	off_t ooffset;	784	off_t ooffset;
502		785
503	X_LOCK (preadwritelock);	786	X_LOCK (preadwritelock);
…		…
508	X_UNLOCK (preadwritelock);	791	X_UNLOCK (preadwritelock);
509		792
510	return res;	793	return res;
511	}	794	}
512		795
513	static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset)	796	static ssize_t
		797	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
514	{	798	{
515	ssize_t res;	799	ssize_t res;
516	off_t ooffset;	800	off_t ooffset;
517		801
518	X_LOCK (preadwritelock);	802	X_LOCK (preadwritelock);
519	ooffset = lseek (fd, 0, SEEK_CUR);	803	ooffset = lseek (fd, 0, SEEK_CUR);
520	lseek (fd, offset, SEEK_SET);	804	lseek (fd, offset, SEEK_SET);
521	res = write (fd, buf, count);	805	res = write (fd, buf, count);
522	lseek (fd, offset, SEEK_SET);	806	lseek (fd, ooffset, SEEK_SET);
523	X_UNLOCK (preadwritelock);	807	X_UNLOCK (preadwritelock);
524		808
525	return res;	809	return res;
526	}	810	}
527	#endif	811	#endif
528		812
529	#ifndef HAVE_FUTIMES	813	#ifndef HAVE_FUTIMES
530		814
		815	# undef utimes
		816	# undef futimes
531	# define utimes(path,times) aio_utimes (path, times)	817	# define utimes(path,times) eio__utimes (path, times)
532	# define futimes(fd,times) aio_futimes (fd, times)	818	# define futimes(fd,times) eio__futimes (fd, times)
533		819
		820	static int
534	static int aio_utimes (const char *filename, const struct timeval times[2])	821	eio__utimes (const char *filename, const struct timeval times[2])
535	{	822	{
536	if (times)	823	if (times)
537	{	824	{
538	struct utimbuf buf;	825	struct utimbuf buf;
539		826
…		…
544	}	831	}
545	else	832	else
546	return utime (filename, 0);	833	return utime (filename, 0);
547	}	834	}
548		835
549	static int aio_futimes (int fd, const struct timeval tv[2])	836	static int eio__futimes (int fd, const struct timeval tv[2])
550	{	837	{
551	errno = ENOSYS;	838	errno = ENOSYS;
552	return -1;	839	return -1;
553	}	840	}
554		841
555	#endif	842	#endif
556		843
557	#if !HAVE_FDATASYNC	844	#if !HAVE_FDATASYNC
		845	# undef fdatasync
558	# define fdatasync fsync	846	# define fdatasync(fd) fsync (fd)
559	#endif	847	#endif
		848
		849	/* sync_file_range always needs emulation */
		850	int
		851	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		852	{
		853	#if HAVE_SYNC_FILE_RANGE
		854	int res;
		855
		856	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		857	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		858	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		859	{
		860	flags = 0
		861	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		862	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		863	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		864	}
		865
		866	res = sync_file_range (fd, offset, nbytes, flags);
		867
		868	if (!res || errno != ENOSYS)
		869	return res;
		870	#endif
		871
		872	/* even though we could play tricks with the flags, it's better to always
		873	* call fdatasync, as that matches the expectation of its users best */
		874	return fdatasync (fd);
		875	}
560		876
561	#if !HAVE_READAHEAD	877	#if !HAVE_READAHEAD
		878	# undef readahead
562	# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self)	879	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
563		880
		881	static ssize_t
564	static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self)	882	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
565	{	883	{
566	size_t todo = count;	884	size_t todo = count;
567	dBUF;	885	dBUF;
568		886
569	while (todo > 0)	887	while (todo > 0)
…		…
579	return count;	897	return count;
580	}	898	}
581		899
582	#endif	900	#endif
583		901
584	#if !HAVE_READDIR_R
585	# define readdir_r aio_readdir_r
586
587	static mutex_t readdirlock = X_MUTEX_INIT;
588
589	static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
590	{
591	X_DIRENT *e;
592	int errorno;
593
594	X_LOCK (readdirlock);
595
596	e = readdir (dirp);
597	errorno = errno;
598
599	if (e)
600	{
601	*res = ent;
602	strcpy (ent->d_name, e->d_name);
603	}
604	else
605	*res = 0;
606
607	X_UNLOCK (readdirlock);
608
609	errno = errorno;
610	return e ? 0 : -1;
611	}
612	#endif
613
614	/* sendfile always needs emulation */	902	/* sendfile always needs emulation */
		903	static ssize_t
615	static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self)	904	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
616	{	905	{
617	ssize_t res;	906	ssize_t res;
618		907
619	if (!count)	908	if (!count)
620	return 0;	909	return 0;
…		…
635	if (res < 0 && sbytes)	924	if (res < 0 && sbytes)
636	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	925	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */
637	res = sbytes;	926	res = sbytes;
638	}	927	}
639		928
		929	# elif defined __APPLE__
		930
		931	{
		932	off_t bytes = count;
		933	res = sendfile (ifd, ofd, offset, &bytes, 0, 0);
		934
		935	if (res < 0 && errno == EAGAIN && bytes)
		936	res = sbytes;
		937	}
		938
640	# elif __hpux	939	# elif __hpux
641	res = sendfile (ofd, ifd, offset, count, 0, 0);	940	res = sendfile (ofd, ifd, offset, count, 0, 0);
642		941
643	# elif __solaris	942	# elif __solaris
644	{	943	{
…		…
655	if (res < 0 && sbytes)	954	if (res < 0 && sbytes)
656	res = sbytes;	955	res = sbytes;
657	}	956	}
658		957
659	# endif	958	# endif
		959	#elif defined _WIN32
		960
		961	/* does not work, just for documentation of what would need to be done */
		962	{
		963	HANDLE h = TO_SOCKET (ifd);
		964	SetFilePointer (h, offset, 0, FILE_BEGIN);
		965	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		966	}
		967
660	#else	968	#else
661	res = -1;	969	res = -1;
662	errno = ENOSYS;	970	errno = ENOSYS;
663	#endif	971	#endif
664		972
…		…
702	}	1010	}
703		1011
704	return res;	1012	return res;
705	}	1013	}
706		1014
		1015	static signed char
		1016	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1017	{
		1018	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1019	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
		1020	}
		1021
		1022	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1023
		1024	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1025	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1026
		1027	static void
		1028	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1029	{
		1030	unsigned char bits [9 + sizeof (ino_t) * 8];
		1031	unsigned char *bit = bits;
		1032
		1033	assert (CHAR_BIT == 8);
		1034	assert (sizeof (eio_dirent) * 8 < 256);
		1035	assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
		1036	assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
		1037
		1038	if (size <= EIO_SORT_FAST)
		1039	return;
		1040
		1041	/* first prepare an array of bits to test in our radix sort */
		1042	/* try to take endianness into account, as well as differences in ino_t sizes */
		1043	/* inode_bits must contain all inodes ORed together */
		1044	/* which is used to skip bits that are 0 everywhere, which is very common */
		1045	{
		1046	ino_t endianness;
		1047	int i, j;
		1048
		1049	/* we store the byte offset of byte n into byte n of "endianness" */
		1050	for (i = 0; i < sizeof (ino_t); ++i)
		1051	((unsigned char *)&endianness)[i] = i;
		1052
		1053	*bit++ = 0;
		1054
		1055	for (i = 0; i < sizeof (ino_t); ++i)
		1056	{
		1057	/* shifting off the byte offsets out of "endianness" */
		1058	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1059	endianness >>= 8;
		1060
		1061	for (j = 0; j < 8; ++j)
		1062	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1063	*bit++ = offs + j;
		1064	}
		1065
		1066	for (j = 0; j < 8; ++j)
		1067	if (score_bits & (1 << j))
		1068	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1069	}
		1070
		1071	/* now actually do the sorting (a variant of MSD radix sort) */
		1072	{
		1073	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1074	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1075	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1076	int stk_idx = 0;
		1077
		1078	base_stk [stk_idx] = dents;
		1079	end_stk [stk_idx] = dents + size;
		1080	bit_stk [stk_idx] = bit - 1;
		1081
		1082	do
		1083	{
		1084	base = base_stk [stk_idx];
		1085	end = end_stk [stk_idx];
		1086	bit = bit_stk [stk_idx];
		1087
		1088	for (;;)
		1089	{
		1090	unsigned char O = *bit >> 3;
		1091	unsigned char M = 1 << (*bit & 7);
		1092
		1093	eio_dirent *a = base;
		1094	eio_dirent *b = end;
		1095
		1096	if (b - a < EIO_SORT_CUTOFF)
		1097	break;
		1098
		1099	/* now bit-partition the array on the bit */
		1100	/* this ugly asymmetric loop seems to perform much better than typical */
		1101	/* partition algos found in the literature */
		1102	do
		1103	if (!(((unsigned char *)a)[O] & M))
		1104	++a;
		1105	else if (!(((unsigned char *)--b)[O] & M))
		1106	{
		1107	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1108	++a;
		1109	}
		1110	while (b > a);
		1111
		1112	/* next bit, or stop, if no bits left in this path */
		1113	if (!*--bit)
		1114	break;
		1115
		1116	base_stk [stk_idx] = a;
		1117	end_stk [stk_idx] = end;
		1118	bit_stk [stk_idx] = bit;
		1119	++stk_idx;
		1120
		1121	end = a;
		1122	}
		1123	}
		1124	while (stk_idx--);
		1125	}
		1126	}
		1127
		1128	static void
		1129	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1130	{
		1131	/* first move the smallest element to the front, to act as a sentinel */
		1132	{
		1133	int i;
		1134	eio_dirent *min = dents;
		1135
		1136	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1137	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1138	if (EIO_DENT_CMP (dents [i], <, *min))
		1139	min = &dents [i];
		1140
		1141	/* swap elements 0 and j (minimum) */
		1142	{
		1143	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1144	}
		1145	}
		1146
		1147	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1148	{
		1149	eio_dirent *i, *j;
		1150
		1151	for (i = dents + 1; i < dents + size; ++i)
		1152	{
		1153	eio_dirent value = *i;
		1154
		1155	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1156	j [1] = j [0];
		1157
		1158	j [1] = value;
		1159	}
		1160	}
		1161	}
		1162
		1163	static void
		1164	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1165	{
		1166	if (size <= 1)
		1167	return; /* our insertion sort relies on size > 0 */
		1168
		1169	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1170	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1171	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1172
		1173	/* use an insertion sort at the end, or for small arrays, */
		1174	/* as insertion sort is more efficient for small partitions */
		1175	eio_dent_insertion_sort (dents, size);
		1176	}
		1177
707	/* read a full directory */	1178	/* read a full directory */
		1179	static void
708	static void scandir_ (eio_req *req, worker *self)	1180	eio__scandir (eio_req *req, etp_worker *self)
709	{	1181	{
710	DIR *dirp;	1182	DIR *dirp;
711	union
712	{
713	EIO_STRUCT_DIRENT d;
714	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
715	} *u;
716	EIO_STRUCT_DIRENT *entp;	1183	EIO_STRUCT_DIRENT *entp;
717	char *name, *names;	1184	char *name, *names;
718	int memlen = 4096;	1185	int namesalloc = 4096;
719	int memofs = 0;	1186	int namesoffs = 0;
		1187	int flags = req->int1;
		1188	eio_dirent *dents = 0;
		1189	int dentalloc = 128;
720	int res = 0;	1190	int dentoffs = 0;
		1191	ino_t inode_bits = 0;
		1192
		1193	req->result = -1;
		1194
		1195	if (!(flags & EIO_READDIR_DENTS))
		1196	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
721		1197
722	X_LOCK (wrklock);	1198	X_LOCK (wrklock);
		1199	/* the corresponding closedir is in ETP_WORKER_CLEAR */
723	self->dirp = dirp = opendir (req->ptr1);	1200	self->dirp = dirp = opendir (req->ptr1);
724	self->dbuf = u = malloc (sizeof (*u));
725	req->flags |= EIO_FLAG_PTR2_FREE;	1201	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1202	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
726	req->ptr2 = names = malloc (memlen);	1203	req->ptr2 = names = malloc (namesalloc);
727	X_UNLOCK (wrklock);	1204	X_UNLOCK (wrklock);
728		1205
729	if (dirp && u && names)	1206	if (dirp && names && (!flags || dents))
730	for (;;)	1207	for (;;)
731	{	1208	{
732	errno = 0;	1209	errno = 0;
733	readdir_r (dirp, &u->d, &entp);	1210	entp = readdir (dirp);
734		1211
735	if (!entp)	1212	if (!entp)
		1213	{
		1214	if (errno)
		1215	break;
		1216
		1217	/* sort etc. */
		1218	req->int1 = flags;
		1219	req->result = dentoffs;
		1220
		1221	if (flags & EIO_READDIR_STAT_ORDER)
		1222	eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
		1223	else if (flags & EIO_READDIR_DIRS_FIRST)
		1224	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1225	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1226	else
		1227	{
		1228	/* in this case, all is known, and we just put dirs first and sort them */
		1229	eio_dirent *oth = dents + dentoffs;
		1230	eio_dirent *dir = dents;
		1231
		1232	/* now partition dirs to the front, and non-dirs to the back */
		1233	/* by walking from both sides and swapping if necessary */
		1234	/* also clear score, so it doesn't influence sorting */
		1235	while (oth > dir)
		1236	{
		1237	if (dir->type == EIO_DT_DIR)
		1238	++dir;
		1239	else if ((--oth)->type == EIO_DT_DIR)
		1240	{
		1241	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1242
		1243	++dir;
		1244	}
		1245	}
		1246
		1247	/* now sort the dirs only */
		1248	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1249	}
		1250
736	break;	1251	break;
		1252	}
737		1253
		1254	/* now add the entry to our list(s) */
738	name = entp->d_name;	1255	name = entp->d_name;
739		1256
		1257	/* skip . and .. entries */
740	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1258	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
741	{	1259	{
742	int len = strlen (name) + 1;	1260	int len = D_NAMLEN (entp) + 1;
743		1261
744	res++;	1262	while (expect_false (namesoffs + len > namesalloc))
745
746	while (memofs + len > memlen)
747	{	1263	{
748	memlen *= 2;	1264	namesalloc *= 2;
749	X_LOCK (wrklock);	1265	X_LOCK (wrklock);
750	req->ptr2 = names = realloc (names, memlen);	1266	req->ptr2 = names = realloc (names, namesalloc);
751	X_UNLOCK (wrklock);	1267	X_UNLOCK (wrklock);
752		1268
753	if (!names)	1269	if (!names)
754	break;	1270	break;
755	}	1271	}
756		1272
757	memcpy (names + memofs, name, len);	1273	memcpy (names + namesoffs, name, len);
		1274
		1275	if (dents)
		1276	{
		1277	struct eio_dirent *ent;
		1278
		1279	if (expect_false (dentoffs == dentalloc))
		1280	{
		1281	dentalloc *= 2;
		1282	X_LOCK (wrklock);
		1283	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1284	X_UNLOCK (wrklock);
		1285
		1286	if (!dents)
		1287	break;
		1288	}
		1289
		1290	ent = dents + dentoffs;
		1291
		1292	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1293	ent->namelen = len - 1;
		1294	ent->inode = D_INO (entp);
		1295
		1296	inode_bits |= ent->inode;
		1297
		1298	switch (D_TYPE (entp))
		1299	{
		1300	default:
		1301	ent->type = EIO_DT_UNKNOWN;
		1302	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1303	break;
		1304
		1305	#ifdef DT_FIFO
		1306	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1307	#endif
		1308	#ifdef DT_CHR
		1309	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1310	#endif
		1311	#ifdef DT_MPC
		1312	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1313	#endif
		1314	#ifdef DT_DIR
		1315	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1316	#endif
		1317	#ifdef DT_NAM
		1318	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1319	#endif
		1320	#ifdef DT_BLK
		1321	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1322	#endif
		1323	#ifdef DT_MPB
		1324	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1325	#endif
		1326	#ifdef DT_REG
		1327	case DT_REG: ent->type = EIO_DT_REG; break;
		1328	#endif
		1329	#ifdef DT_NWK
		1330	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1331	#endif
		1332	#ifdef DT_CMP
		1333	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1334	#endif
		1335	#ifdef DT_LNK
		1336	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1337	#endif
		1338	#ifdef DT_SOCK
		1339	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1340	#endif
		1341	#ifdef DT_DOOR
		1342	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1343	#endif
		1344	#ifdef DT_WHT
		1345	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1346	#endif
		1347	}
		1348
		1349	ent->score = 7;
		1350
		1351	if (flags & EIO_READDIR_DIRS_FIRST)
		1352	{
		1353	if (ent->type == EIO_DT_UNKNOWN)
		1354	{
		1355	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1356	ent->score = 1;
		1357	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1358	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1359	}
		1360	else if (ent->type == EIO_DT_DIR)
		1361	ent->score = 0;
		1362	}
		1363	}
		1364
758	memofs += len;	1365	namesoffs += len;
		1366	++dentoffs;
		1367	}
		1368
		1369	if (EIO_CANCELLED (req))
		1370	{
		1371	errno = ECANCELED;
		1372	break;
759	}	1373	}
760	}	1374	}
		1375	}
761		1376
762	if (errno)	1377	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
763	res = -1;	1378	# undef msync
764		1379	# define msync(a,b,c) ((errno = ENOSYS), -1)
765	req->result = res;	1380	#endif
		1381
		1382	int
		1383	eio__mtouch (void *mem, size_t len, int flags)
		1384	{
		1385	intptr_t addr = (intptr_t)mem;
		1386	intptr_t end = addr + len;
		1387	#ifdef PAGESIZE
		1388	const intptr_t page = PAGESIZE;
		1389	#else
		1390	static intptr_t page;
		1391
		1392	if (!page)
		1393	page = sysconf (_SC_PAGESIZE);
		1394	#endif
		1395
		1396	/* round down to start of page, although this is probably useless */
		1397	addr &= ~(page - 1); /* assume page size is always a power of two */
		1398
		1399	if (addr < end)
		1400	if (flags) /* modify */
		1401	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
		1402	else
		1403	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
		1404
		1405	return 0;
766	}	1406	}
767		1407
768	/*****************************************************************************/	1408	/*****************************************************************************/
769		1409
		1410	#define ALLOC(len) \
		1411	if (!req->ptr2) \
		1412	{ \
		1413	X_LOCK (wrklock); \
		1414	req->flags |= EIO_FLAG_PTR2_FREE; \
		1415	X_UNLOCK (wrklock); \
		1416	req->ptr2 = malloc (len); \
		1417	if (!req->ptr2) \
		1418	{ \
		1419	errno = ENOMEM; \
		1420	req->result = -1; \
		1421	break; \
		1422	} \
		1423	}
		1424
770	X_THREAD_PROC (eio_proc)	1425	X_THREAD_PROC (etp_proc)
771	{	1426	{
772	eio_req *req;	1427	ETP_REQ *req;
773	struct timespec ts;	1428	struct timespec ts;
774	worker *self = (worker *)thr_arg;	1429	etp_worker *self = (etp_worker *)thr_arg;
775		1430
776	/* try to distribute timeouts somewhat randomly */	1431	/* try to distribute timeouts somewhat randomly */
777	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1432	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
778		1433
779	for (;;)	1434	for (;;)
780	{	1435	{
781	ts.tv_sec = time (0) + IDLE_TIMEOUT;
782
783	X_LOCK (reqlock);	1436	X_LOCK (reqlock);
784		1437
785	for (;;)	1438	for (;;)
786	{	1439	{
787	self->req = req = reqq_shift (&req_queue);	1440	self->req = req = reqq_shift (&req_queue);
…		…
789	if (req)	1442	if (req)
790	break;	1443	break;
791		1444
792	++idle;	1445	++idle;
793		1446
		1447	ts.tv_sec = time (0) + IDLE_TIMEOUT;
794	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1448	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
795	{	1449	{
796	if (idle > max_idle)	1450	if (idle > max_idle)
797	{	1451	{
798	--idle;	1452	--idle;
…		…
803	goto quit;	1457	goto quit;
804	}	1458	}
805		1459
806	/* we are allowed to idle, so do so without any timeout */	1460	/* we are allowed to idle, so do so without any timeout */
807	X_COND_WAIT (reqwait, reqlock);	1461	X_COND_WAIT (reqwait, reqlock);
808	ts.tv_sec = time (0) + IDLE_TIMEOUT;
809	}	1462	}
810		1463
811	--idle;	1464	--idle;
812	}	1465	}
813		1466
814	--nready;	1467	--nready;
815		1468
816	X_UNLOCK (reqlock);	1469	X_UNLOCK (reqlock);
817		1470
818	errno = 0; /* strictly unnecessary */	1471	if (req->type < 0)
		1472	goto quit;
819		1473
820	if (!EIO_CANCELLED (req))	1474	if (!EIO_CANCELLED (req))
821	switch (req->type)	1475	ETP_EXECUTE (self, req);
822	{
823	case EIO_READ: req->result = req->offs >= 0
824	? pread (req->int1, req->ptr2, req->size, req->offs)
825	: read (req->int1, req->ptr2, req->size); break;
826	case EIO_WRITE: req->result = req->offs >= 0
827	? pwrite (req->int1, req->ptr2, req->size, req->offs)
828	: write (req->int1, req->ptr2, req->size); break;
829
830	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
831	case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
832
833	case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
834	case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
835	case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
836
837	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
838	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
839	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
840	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
841	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
842	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
843
844	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
845	case EIO_CLOSE: req->result = close (req->int1); break;
846	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
847	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
848	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
849	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
850	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
851	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
852	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
853	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
854	case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
855
856	case EIO_SYNC: req->result = 0; sync (); break;
857	case EIO_FSYNC: req->result = fsync (req->int1); break;
858	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
859
860	case EIO_READDIR: scandir_ (req, self); break;
861
862	case EIO_BUSY:
863	#ifdef _WIN32
864	Sleep (req->nv1 * 1000.);
865	#else
866	{
867	struct timeval tv;
868
869	tv.tv_sec = req->nv1;
870	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
871
872	req->result = select (0, 0, 0, 0, &tv);
873	}
874	#endif
875	break;
876
877	case EIO_UTIME:
878	case EIO_FUTIME:
879	{
880	struct timeval tv[2];
881	struct timeval *times;
882
883	if (req->nv1 != -1. || req->nv2 != -1.)
884	{
885	tv[0].tv_sec = req->nv1;
886	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
887	tv[1].tv_sec = req->nv2;
888	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
889
890	times = tv;
891	}
892	else
893	times = 0;
894
895
896	req->result = req->type == EIO_FUTIME
897	? futimes (req->int1, times)
898	: utimes (req->ptr1, times);
899	}
900
901	case EIO_GROUP:
902	case EIO_NOP:
903	break;
904
905	case EIO_QUIT:
906	goto quit;
907
908	default:
909	req->result = -1;
910	break;
911	}
912
913	req->errorno = errno;
914		1476
915	X_LOCK (reslock);	1477	X_LOCK (reslock);
916		1478
917	++npending;	1479	++npending;
918		1480
919	if (!reqq_push (&res_queue, req) && want_poll_cb)	1481	if (!reqq_push (&res_queue, req) && want_poll_cb)
920	want_poll_cb ();	1482	want_poll_cb ();
921		1483
922	self->req = 0;	1484	self->req = 0;
923	worker_clear (self);	1485	etp_worker_clear (self);
924		1486
925	X_UNLOCK (reslock);	1487	X_UNLOCK (reslock);
926	}	1488	}
927		1489
928	quit:	1490	quit:
929	X_LOCK (wrklock);	1491	X_LOCK (wrklock);
930	worker_free (self);	1492	etp_worker_free (self);
931	X_UNLOCK (wrklock);	1493	X_UNLOCK (wrklock);
932		1494
933	return 0;	1495	return 0;
934	}	1496	}
935		1497
936	/*****************************************************************************/	1498	/*****************************************************************************/
937		1499
938	static void eio_atfork_prepare (void)	1500	int eio_init (void (*want_poll)(void), void (*done_poll)(void))
939	{	1501	{
940	X_LOCK (wrklock);	1502	return etp_init (want_poll, done_poll);
941	X_LOCK (reqlock);
942	X_LOCK (reslock);
943	#if !HAVE_PREADWRITE
944	X_LOCK (preadwritelock);
945	#endif
946	#if !HAVE_READDIR_R
947	X_LOCK (readdirlock);
948	#endif
949	}	1503	}
950		1504
951	static void eio_atfork_parent (void)	1505	static void eio_api_destroy (eio_req *req)
952	{	1506	{
953	#if !HAVE_READDIR_R	1507	free (req);
954	X_UNLOCK (readdirlock);
955	#endif
956	#if !HAVE_PREADWRITE
957	X_UNLOCK (preadwritelock);
958	#endif
959	X_UNLOCK (reslock);
960	X_UNLOCK (reqlock);
961	X_UNLOCK (wrklock);
962	}	1508	}
963		1509
964	static void eio_atfork_child (void)	1510	#define REQ(rtype) \
965	{	1511	eio_req *req; \
966	eio_req *prv;	1512	\
		1513	req = (eio_req *)calloc (1, sizeof *req); \
		1514	if (!req) \
		1515	return 0; \
		1516	\
		1517	req->type = rtype; \
		1518	req->pri = pri; \
		1519	req->finish = cb; \
		1520	req->data = data; \
		1521	req->destroy = eio_api_destroy;
967		1522
968	while (prv = reqq_shift (&req_queue))	1523	#define SEND eio_submit (req); return req
969	eio_destroy (prv);
970		1524
971	while (prv = reqq_shift (&res_queue))	1525	#define PATH \
972	eio_destroy (prv);	1526	req->flags |= EIO_FLAG_PTR1_FREE; \
		1527	req->ptr1 = strdup (path); \
		1528	if (!req->ptr1) \
		1529	{ \
		1530	eio_api_destroy (req); \
		1531	return 0; \
		1532	}
973		1533
974	while (wrk_first.next != &wrk_first)	1534	static void eio_execute (etp_worker *self, eio_req *req)
		1535	{
		1536	errno = 0;
		1537
		1538	switch (req->type)
975	{	1539	{
976	worker *wrk = wrk_first.next;	1540	case EIO_READ: ALLOC (req->size);
		1541	req->result = req->offs >= 0
		1542	? pread (req->int1, req->ptr2, req->size, req->offs)
		1543	: read (req->int1, req->ptr2, req->size); break;
		1544	case EIO_WRITE: req->result = req->offs >= 0
		1545	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1546	: write (req->int1, req->ptr2, req->size); break;
977		1547
978	if (wrk->req)	1548	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
979	eio_destroy (wrk->req);	1549	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
980		1550
981	worker_clear (wrk);	1551	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
982	worker_free (wrk);	1552	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1553	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1554	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1555	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1556	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1557
		1558	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1559	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1560	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1561	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1562	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1563	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1564
		1565	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1566	case EIO_CLOSE: req->result = close (req->int1); break;
		1567	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1568	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1569	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1570	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1571	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1572	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1573	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1574	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
		1575
		1576	case EIO_READLINK: ALLOC (NAME_MAX);
		1577	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
		1578
		1579	case EIO_SYNC: req->result = 0; sync (); break;
		1580	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1581	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1582	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
		1583	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1584	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1585
		1586	case EIO_READDIR: eio__scandir (req, self); break;
		1587
		1588	case EIO_BUSY:
		1589	#ifdef _WIN32
		1590	Sleep (req->nv1 * 1000.);
		1591	#else
		1592	{
		1593	struct timeval tv;
		1594
		1595	tv.tv_sec = req->nv1;
		1596	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
		1597
		1598	req->result = select (0, 0, 0, 0, &tv);
		1599	}
		1600	#endif
		1601	break;
		1602
		1603	case EIO_UTIME:
		1604	case EIO_FUTIME:
		1605	{
		1606	struct timeval tv[2];
		1607	struct timeval *times;
		1608
		1609	if (req->nv1 != -1. || req->nv2 != -1.)
		1610	{
		1611	tv[0].tv_sec = req->nv1;
		1612	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		1613	tv[1].tv_sec = req->nv2;
		1614	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		1615
		1616	times = tv;
		1617	}
		1618	else
		1619	times = 0;
		1620
		1621
		1622	req->result = req->type == EIO_FUTIME
		1623	? futimes (req->int1, times)
		1624	: utimes (req->ptr1, times);
		1625	}
		1626	break;
		1627
		1628	case EIO_GROUP:
		1629	abort (); /* handled in eio_request */
		1630
		1631	case EIO_NOP:
		1632	req->result = 0;
		1633	break;
		1634
		1635	case EIO_CUSTOM:
		1636	((void (*)(eio_req *))req->feed) (req);
		1637	break;
		1638
		1639	default:
		1640	req->result = -1;
		1641	break;
983	}	1642	}
984		1643
985	started = 0;	1644	req->errorno = errno;
986	idle = 0;
987	nreqs = 0;
988	nready = 0;
989	npending = 0;
990
991	eio_atfork_parent ();
992	}	1645	}
993		1646
994	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1647	#ifndef EIO_NO_WRAPPERS
995	{
996	want_poll_cb = want_poll;
997	done_poll_cb = done_poll;
998		1648
999	#ifdef _WIN32	1649	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1000	X_MUTEX_CHECK (wrklock);	1650	{
1001	X_MUTEX_CHECK (reslock);	1651	REQ (EIO_NOP); SEND;
1002	X_MUTEX_CHECK (reqlock);
1003	X_MUTEX_CHECK (reqwait);
1004	X_MUTEX_CHECK (preadwritelock);
1005	X_MUTEX_CHECK (readdirlock);
1006
1007	X_COND_CHECK (reqwait);
1008	#endif
1009
1010	X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1011	}	1652	}
1012		1653
1013	#if 0	1654	eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
		1655	{
		1656	REQ (EIO_BUSY); req->nv1 = delay; SEND;
		1657	}
1014		1658
1015	eio_req *eio_fsync (int fd, eio_cb cb);	1659	eio_req *eio_sync (int pri, eio_cb cb, void *data)
		1660	{
		1661	REQ (EIO_SYNC); SEND;
		1662	}
		1663
		1664	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
		1665	{
		1666	REQ (EIO_FSYNC); req->int1 = fd; SEND;
		1667	}
		1668
		1669	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1670	{
		1671	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1672	}
		1673
		1674	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1675	{
		1676	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1677	}
		1678
		1679	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		1680	{
		1681	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		1682	}
		1683
1016	eio_req *eio_fdatasync (int fd, eio_cb cb);	1684	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1017	eio_req *eio_dupclose (int fd, eio_cb cb);	1685	{
		1686	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		1687	}
		1688
		1689	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
		1690	{
		1691	REQ (EIO_CLOSE); req->int1 = fd; SEND;
		1692	}
		1693
1018	eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb);	1694	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1019	eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb);	1695	{
1020	eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb);	1696	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1021	eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */	1697	}
		1698
		1699	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
		1700	{
		1701	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
		1702	}
		1703
		1704	eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
		1705	{
		1706	REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
		1707	}
		1708
		1709	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
		1710	{
		1711	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1712	}
		1713
1022	eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb);	1714	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
		1715	{
		1716	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
		1717	}
		1718
1023	eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb);	1719	eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
		1720	{
		1721	REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
		1722	}
		1723
1024	eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb);	1724	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
		1725	{
		1726	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
		1727	}
		1728
1025	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb);	1729	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
		1730	{
		1731	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
		1732	}
		1733
1026	eio_req *eio_dup2 (int fd, int fd2, eio_cb cb);	1734	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
		1735	{
		1736	REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
		1737	}
		1738
1027	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb);	1739	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
		1740	{
		1741	REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
		1742	}
		1743
1028	eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb);	1744	eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1029	eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */	1745	{
1030	eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */	1746	REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1031	eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */	1747	}
		1748
1032	eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb);	1749	eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
		1750	{
		1751	REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
		1752	}
		1753
1033	eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb);	1754	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1034	eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb);	1755	{
1035	eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb);	1756	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
		1757	}
		1758
1036	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb);	1759	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
		1760	{
		1761	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
		1762	}
		1763
		1764	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
		1765	{
		1766	REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
		1767	}
		1768
		1769	eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
		1770	{
		1771	REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
		1772	}
		1773
		1774	static eio_req *
		1775	eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
		1776	{
		1777	REQ (type); PATH; SEND;
		1778	}
		1779
		1780	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
		1781	{
		1782	return eio__1path (EIO_READLINK, path, pri, cb, data);
		1783	}
		1784
		1785	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
		1786	{
		1787	return eio__1path (EIO_STAT, path, pri, cb, data);
		1788	}
		1789
		1790	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
		1791	{
		1792	return eio__1path (EIO_LSTAT, path, pri, cb, data);
		1793	}
		1794
1037	eio_req *eio_unlink (const char *path, eio_cb cb);	1795	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
		1796	{
		1797	return eio__1path (EIO_UNLINK, path, pri, cb, data);
		1798	}
		1799
1038	eio_req *eio_rmdir (const char *path, eio_cb cb);	1800	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1039	eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */	1801	{
		1802	return eio__1path (EIO_RMDIR, path, pri, cb, data);
		1803	}
		1804
		1805	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
		1806	{
		1807	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
		1808	}
		1809
1040	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb);	1810	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1041	eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */
1042	eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */
1043	void
1044	aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1045	PROTOTYPE: $$$;$
1046	PPCODE:
1047	{	1811	{
1048	dREQ;	1812	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1049
1050	req->type = EIO_OPEN;
1051	req->sv1 = newSVsv (pathname);
1052	req->ptr1 = SvPVbyte_nolen (req->sv1);
1053	req->int1 = flags;
1054	req->int2 = mode;
1055
1056	EIO_SEND;
1057	}	1813	}
1058		1814
1059	void	1815	static eio_req *
1060	aio_fsync (SV *fh, SV *callback=&PL_sv_undef)	1816	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1061	PROTOTYPE: $;$
1062	ALIAS:
1063	aio_fsync = EIO_FSYNC
1064	aio_fdatasync = EIO_FDATASYNC
1065	PPCODE:
1066	{	1817	{
1067	dREQ;	1818	REQ (type); PATH;
1068		1819
1069	req->type = ix;
1070	req->sv1 = newSVsv (fh);
1071	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1072
1073	EIO_SEND (req);
1074	}
1075
1076	void
1077	aio_close (SV *fh, SV *callback=&PL_sv_undef)
1078	PROTOTYPE: $;$
1079	PPCODE:
1080	{
1081	dREQ;
1082
1083	req->type = EIO_CLOSE;
1084	req->sv1 = newSVsv (fh);
1085	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1086
1087	EIO_SEND (req);
1088	}
1089
1090	void
1091	aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1092	ALIAS:
1093	aio_read = EIO_READ
1094	aio_write = EIO_WRITE
1095	PROTOTYPE: $$$$$;$
1096	PPCODE:
1097	{
1098	STRLEN svlen;
1099	char *svptr = SvPVbyte (data, svlen);
1100	UV len = SvUV (length);
1101
1102	SvUPGRADE (data, SVt_PV);
1103	SvPOK_on (data);
1104
1105	if (dataoffset < 0)
1106	dataoffset += svlen;
1107
1108	if (dataoffset < 0 || dataoffset > svlen)
1109	croak ("dataoffset outside of data scalar");
1110
1111	if (ix == EIO_WRITE)
1112	{
1113	/* write: check length and adjust. */
1114	if (!SvOK (length) || len + dataoffset > svlen)
1115	len = svlen - dataoffset;
1116	}
1117	else
1118	{
1119	/* read: grow scalar as necessary */
1120	svptr = SvGROW (data, len + dataoffset + 1);
1121	}
1122
1123	if (len < 0)
1124	croak ("length must not be negative");
1125
1126	{
1127	dREQ;
1128
1129	req->type = ix;
1130	req->sv1 = newSVsv (fh);
1131	req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1132	: IoOFP (sv_2io (fh)));
1133	req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1134	req->size = len;
1135	req->sv2 = SvREFCNT_inc (data);
1136	req->ptr2 = (char *)svptr + dataoffset;
1137	req->stroffset = dataoffset;
1138
1139	if (!SvREADONLY (data))
1140	{
1141	SvREADONLY_on (data);
1142	req->flags |= FLAG_SV2_RO_OFF;
1143	}
1144
1145	EIO_SEND;
1146	}
1147	}
1148
1149	void
1150	aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1151	PROTOTYPE: $$;$
1152	PPCODE:
1153	{
1154	SV *data;
1155	dREQ;
1156
1157	data = newSV (NAME_MAX);
1158	SvPOK_on (data);
1159
1160	req->type = EIO_READLINK;
1161	req->sv1 = newSVsv (path);
1162	req->ptr1 = SvPVbyte_nolen (req->sv1);
1163	req->sv2 = data;
1164	req->ptr2 = SvPVbyte_nolen (data);
1165
1166	EIO_SEND;
1167	}
1168
1169	void
1170	aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1171	PROTOTYPE: $$$$;$
1172	PPCODE:
1173	{
1174	dREQ;
1175
1176	req->type = EIO_SENDFILE;
1177	req->sv1 = newSVsv (out_fh);
1178	req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1179	req->sv2 = newSVsv (in_fh);
1180	req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1181	req->offs = SvVAL64 (in_offset);
1182	req->size = length;
1183
1184	EIO_SEND;
1185	}
1186
1187	void
1188	aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1189	PROTOTYPE: $$$;$
1190	PPCODE:
1191	{
1192	dREQ;
1193
1194	req->type = EIO_READAHEAD;
1195	req->sv1 = newSVsv (fh);
1196	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1197	req->offs = SvVAL64 (offset);
1198	req->size = length;
1199
1200	EIO_SEND;
1201	}
1202
1203	void
1204	aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1205	ALIAS:
1206	aio_stat = EIO_STAT
1207	aio_lstat = EIO_LSTAT
1208	PPCODE:
1209	{
1210	dREQ;
1211
1212	req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1213	if (!req->ptr2)
1214	{
1215	req_destroy (req);
1216	croak ("out of memory during aio_stat statdata allocation");
1217	}
1218
1219	req->flags |= FLAG_PTR2_FREE;	1820	req->flags |= EIO_FLAG_PTR2_FREE;
1220	req->sv1 = newSVsv (fh_or_path);	1821	req->ptr2 = strdup (new_path);
		1822	if (!req->ptr2)
		1823	{
		1824	eio_api_destroy (req);
		1825	return 0;
		1826	}
1221		1827
1222	if (SvPOK (fh_or_path))	1828	SEND;
1223	{
1224	req->type = ix;
1225	req->ptr1 = SvPVbyte_nolen (req->sv1);
1226	}
1227	else
1228	{
1229	req->type = EIO_FSTAT;
1230	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1231	}
1232
1233	EIO_SEND;
1234	}	1829	}
1235		1830
1236	void	1831	eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1237	aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1238	PPCODE:
1239	{	1832	{
1240	dREQ;	1833	return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
1241
1242	req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1243	req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1244	req->sv1 = newSVsv (fh_or_path);
1245
1246	if (SvPOK (fh_or_path))
1247	{
1248	req->type = EIO_UTIME;
1249	req->ptr1 = SvPVbyte_nolen (req->sv1);
1250	}
1251	else
1252	{
1253	req->type = EIO_FUTIME;
1254	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1255	}
1256
1257	EIO_SEND;
1258	}	1834	}
1259		1835
1260	void	1836	eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1261	aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1262	PPCODE:
1263	{	1837	{
1264	dREQ;	1838	return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
1265
1266	req->sv1 = newSVsv (fh_or_path);
1267	req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1268
1269	if (SvPOK (fh_or_path))
1270	{
1271	req->type = EIO_TRUNCATE;
1272	req->ptr1 = SvPVbyte_nolen (req->sv1);
1273	}
1274	else
1275	{
1276	req->type = EIO_FTRUNCATE;
1277	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1278	}
1279
1280	EIO_SEND;
1281	}	1839	}
1282		1840
1283	void	1841	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1284	aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1285	ALIAS:
1286	aio_chmod = EIO_CHMOD
1287	aio_fchmod = EIO_FCHMOD
1288	aio_mkdir = EIO_MKDIR
1289	PPCODE:
1290	{	1842	{
1291	dREQ;	1843	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1292
1293	req->type = type;
1294	req->int2 = mode;
1295	req->sv1 = newSVsv (fh_or_path);
1296
1297	if (ix == EIO_FCHMOD)
1298	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1299	else
1300	req->ptr1 = SvPVbyte_nolen (req->sv1);
1301
1302	EIO_SEND;
1303	}	1844	}
1304		1845
1305	void	1846	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1306	aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1307	PPCODE:
1308	{	1847	{
1309	dREQ;	1848	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1310
1311	req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1312	req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1313	req->sv1 = newSVsv (fh_or_path);
1314
1315	if (SvPOK (fh_or_path))
1316	{
1317	req->type = EIO_CHOWN;
1318	req->ptr1 = SvPVbyte_nolen (req->sv1);
1319	}
1320	else
1321	{
1322	req->type = EIO_FCHOWN;
1323	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1324	}
1325
1326	EIO_SEND;
1327	}	1849	}
1328		1850
1329	void
1330	aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1331	ALIAS:
1332	aio_unlink = EIO_UNLINK
1333	aio_rmdir = EIO_RMDIR
1334	aio_readdir = EIO_READDIR
1335	PPCODE:
1336	{
1337	dREQ;
1338
1339	req->type = ix;
1340	req->sv1 = newSVsv (pathname);
1341	req->ptr1 = SvPVbyte_nolen (req->sv1);
1342
1343	EIO_SEND;
1344	}
1345
1346	void
1347	aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1348	ALIAS:
1349	aio_link = EIO_LINK
1350	aio_symlink = EIO_SYMLINK
1351	aio_rename = EIO_RENAME
1352	PPCODE:
1353	{
1354	dREQ;
1355
1356	req->type = ix;
1357	req->sv1 = newSVsv (oldpath);
1358	req->ptr1 = SvPVbyte_nolen (req->sv1);
1359	req->sv2 = newSVsv (newpath);
1360	req->ptr2 = SvPVbyte_nolen (req->sv2);
1361
1362	EIO_SEND;
1363	}
1364
1365	void
1366	aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1367	PPCODE:
1368	{
1369	dREQ;
1370
1371	req->type = EIO_MKNOD;
1372	req->sv1 = newSVsv (pathname);
1373	req->ptr1 = SvPVbyte_nolen (req->sv1);
1374	req->int2 = (mode_t)mode;
1375	req->offs = dev;
1376
1377	EIO_SEND;
1378	}
1379
1380	void
1381	aio_busy (double delay, SV *callback=&PL_sv_undef)
1382	PPCODE:
1383	{
1384	dREQ;
1385
1386	req->type = EIO_BUSY;
1387	req->nv1 = delay < 0. ? 0. : delay;
1388
1389	EIO_SEND;
1390	}
1391
1392	void
1393	aio_group (SV *callback=&PL_sv_undef)
1394	PROTOTYPE: ;$
1395	PPCODE:
1396	{
1397	dREQ;
1398
1399	req->type = EIO_GROUP;
1400
1401	req_send (req);
1402	XPUSHs (req_sv (req, AIO_GRP_KLASS));
1403	}
1404
1405	void
1406	aio_nop (SV *callback=&PL_sv_undef)
1407	ALIAS:
1408	aio_nop = EIO_NOP
1409	aio_sync = EIO_SYNC
1410	PPCODE:
1411	{
1412	dREQ;
1413
1414	#endif	1851	#endif
		1852
		1853	eio_req *eio_grp (eio_cb cb, void *data)
		1854	{
		1855	const int pri = EIO_PRI_MAX;
		1856
		1857	REQ (EIO_GROUP); SEND;
		1858	}
		1859
		1860	#undef REQ
		1861	#undef PATH
		1862	#undef SEND
		1863
		1864	/*****************************************************************************/
		1865	/* grp functions */
1415		1866
1416	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	1867	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1417	{	1868	{
1418	grp->int2 = limit;	1869	grp->int2 = limit;
1419	grp->feed = feed;	1870	grp->feed = feed;
…		…
1430		1881
1431	void eio_grp_add (eio_req *grp, eio_req *req)	1882	void eio_grp_add (eio_req *grp, eio_req *req)
1432	{	1883	{
1433	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	1884	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1434		1885
		1886	grp->flags |= EIO_FLAG_GROUPADD;
		1887
1435	++grp->size;	1888	++grp->size;
1436	req->grp = grp;	1889	req->grp = grp;
1437		1890
1438	req->grp_prev = 0;	1891	req->grp_prev = 0;
1439	req->grp_next = grp->grp_first;	1892	req->grp_next = grp->grp_first;
…		…
1442	grp->grp_first->grp_prev = req;	1895	grp->grp_first->grp_prev = req;
1443		1896
1444	grp->grp_first = req;	1897	grp->grp_first = req;
1445	}	1898	}
1446		1899
		1900	/*****************************************************************************/
		1901	/* misc garbage */
1447		1902
		1903	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
		1904	{
		1905	etp_worker wrk;
		1906	ssize_t ret;
		1907
		1908	wrk.dbuf = 0;
		1909
		1910	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
		1911
		1912	if (wrk.dbuf)
		1913	free (wrk.dbuf);
		1914
		1915	return ret;
		1916	}
		1917

Diff Legend

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