ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/cvsroot/libeio/eio.c

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.5 by root, Sun May 11 01:08:05 2008 UTC vs.
Revision 1.32 by root, Sat Jun 6 18:06:55 2009 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>
36	# include <unistd.h>	78	# include <unistd.h>
37	# include <utime.h>	79	# include <utime.h>
38	# include <signal.h>	80	# include <signal.h>
39	# include <dirent.h>	81	# include <dirent.h>
40		82
		83	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		84	# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
		85	# define D_INO(de) (de)->d_fileno
		86	# define _DIRENT_HAVE_D_TYPE /* sigh */
		87	# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
		88	# define D_INO(de) (de)->d_ino
		89	# endif
		90
		91	# ifdef _DIRENT_HAVE_D_TYPE
		92	# define D_TYPE(de) (de)->d_type
		93	# endif
		94
41	# ifndef EIO_STRUCT_DIRENT	95	# ifndef EIO_STRUCT_DIRENT
42	# define EIO_STRUCT_DIRENT struct dirent	96	# define EIO_STRUCT_DIRENT struct dirent
43	# endif	97	# endif
44		98
45	#endif	99	#endif
46
47	# ifndef EIO_STRUCT_STAT
48	# define EIO_STRUCT_STAT struct stat
49	# endif
50		100
51	#if HAVE_SENDFILE	101	#if HAVE_SENDFILE
52	# if __linux	102	# if __linux
53	# include <sys/sendfile.h>	103	# include <sys/sendfile.h>
54	# elif __freebsd	104	# elif __freebsd
…		…
61	# else	111	# else
62	# error sendfile support requested but not available	112	# error sendfile support requested but not available
63	# endif	113	# endif
64	#endif	114	#endif
65		115
		116	#ifndef D_TYPE
		117	# define D_TYPE(de) 0
		118	#endif
		119
		120	#ifndef D_INO
		121	# define D_INO(de) 0
		122	#endif
		123
66	/* number of seconds after which an idle threads exit */	124	/* number of seconds after which an idle threads exit */
67	#define IDLE_TIMEOUT 10	125	#define IDLE_TIMEOUT 10
68		126
69	/* used for struct dirent, AIX doesn't provide it */	127	/* used for struct dirent, AIX doesn't provide it */
70	#ifndef NAME_MAX	128	#ifndef NAME_MAX
…		…
74	/* buffer size for various temporary buffers */	132	/* buffer size for various temporary buffers */
75	#define EIO_BUFSIZE 65536	133	#define EIO_BUFSIZE 65536
76		134
77	#define dBUF \	135	#define dBUF \
78	char *eio_buf; \	136	char *eio_buf; \
79	X_LOCK (wrklock); \	137	ETP_WORKER_LOCK (self); \
80	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	138	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81	X_UNLOCK (wrklock); \	139	ETP_WORKER_UNLOCK (self); \
		140	errno = ENOMEM; \
82	if (!eio_buf) \	141	if (!eio_buf) \
83	return -1;	142	return -1;
84		143
85	#define EIO_TICKS ((1000000 + 1023) >> 10)	144	#define EIO_TICKS ((1000000 + 1023) >> 10)
86		145
		146	/*****************************************************************************/
		147
		148	#if __GNUC__ >= 3
		149	# define expect(expr,value) __builtin_expect ((expr),(value))
		150	#else
		151	# define expect(expr,value) (expr)
		152	#endif
		153
		154	#define expect_false(expr) expect ((expr) != 0, 0)
		155	#define expect_true(expr) expect ((expr) != 0, 1)
		156
		157	/*****************************************************************************/
		158
		159	#define ETP_PRI_MIN EIO_PRI_MIN
		160	#define ETP_PRI_MAX EIO_PRI_MAX
		161
		162	struct etp_worker;
		163
		164	#define ETP_REQ eio_req
		165	#define ETP_DESTROY(req) eio_destroy (req)
		166	static int eio_finish (eio_req *req);
		167	#define ETP_FINISH(req) eio_finish (req)
		168	static void eio_execute (struct etp_worker *self, eio_req *req);
		169	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		170
		171	#define ETP_WORKER_CLEAR(req) \
		172	if (wrk->dbuf) \
		173	{ \
		174	free (wrk->dbuf); \
		175	wrk->dbuf = 0; \
		176	} \
		177	\
		178	if (wrk->dirp) \
		179	{ \
		180	closedir (wrk->dirp); \
		181	wrk->dirp = 0; \
		182	}
		183
		184	#define ETP_WORKER_COMMON \
		185	void *dbuf; \
		186	DIR *dirp;
		187
		188	/*****************************************************************************/
		189
		190	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		191
		192	/* calculcate time difference in ~1/EIO_TICKS of a second */
		193	static int tvdiff (struct timeval *tv1, struct timeval *tv2)
		194	{
		195	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
		196	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
		197	}
		198
		199	static unsigned int started, idle, wanted = 4;
		200
87	static void (*want_poll_cb) (void);	201	static void (*want_poll_cb) (void);
88	static void (*done_poll_cb) (void);	202	static void (*done_poll_cb) (void);
89		203
90	static unsigned int max_poll_time = 0;	204	static unsigned int max_poll_time; /* reslock */
91	static unsigned int max_poll_reqs = 0;	205	static unsigned int max_poll_reqs; /* reslock */
92		206
93	/* calculcate time difference in ~1/EIO_TICKS of a second */	207	static volatile unsigned int nreqs; /* reqlock */
94	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	208	static volatile unsigned int nready; /* reqlock */
95	{	209	static volatile unsigned int npending; /* reqlock */
96	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	210	static volatile unsigned int max_idle = 4;
97	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
98	}
99		211
100	static unsigned int started, idle, wanted;
101
102	/* worker threads management */
103	static mutex_t wrklock = X_MUTEX_INIT;	212	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;	213	static mutex_t reslock = X_MUTEX_INIT;
146	static mutex_t reqlock = X_MUTEX_INIT;	214	static mutex_t reqlock = X_MUTEX_INIT;
147	static cond_t reqwait = X_COND_INIT;	215	static cond_t reqwait = X_COND_INIT;
148		216
		217	#if !HAVE_PREADWRITE
		218	/*
		219	* make our pread/pwrite emulation safe against themselves, but not against
		220	* normal read/write by using a mutex. slows down execution a lot,
		221	* but that's your problem, not mine.
		222	*/
		223	static mutex_t preadwritelock = X_MUTEX_INIT;
		224	#endif
		225
		226	typedef struct etp_worker
		227	{
		228	/* locked by wrklock */
		229	struct etp_worker *prev, *next;
		230
		231	thread_t tid;
		232
		233	/* locked by reslock, reqlock or wrklock */
		234	ETP_REQ *req; /* currently processed request */
		235
		236	ETP_WORKER_COMMON
		237	} etp_worker;
		238
		239	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
		240
		241	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		242	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
		243
		244	/* worker threads management */
		245
		246	static void etp_worker_clear (etp_worker *wrk)
		247	{
		248	ETP_WORKER_CLEAR (wrk);
		249	}
		250
		251	static void etp_worker_free (etp_worker *wrk)
		252	{
		253	wrk->next->prev = wrk->prev;
		254	wrk->prev->next = wrk->next;
		255
		256	free (wrk);
		257	}
		258
149	unsigned int eio_nreqs (void)	259	static unsigned int etp_nreqs (void)
150	{	260	{
		261	int retval;
		262	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		263	retval = nreqs;
		264	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151	return nreqs;	265	return retval;
152	}	266	}
153		267
154	unsigned int eio_nready (void)	268	static unsigned int etp_nready (void)
155	{	269	{
156	unsigned int retval;	270	unsigned int retval;
157		271
158	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	272	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159	retval = nready;	273	retval = nready;
160	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	274	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161		275
162	return retval;	276	return retval;
163	}	277	}
164		278
165	unsigned int eio_npending (void)	279	static unsigned int etp_npending (void)
166	{	280	{
167	unsigned int retval;	281	unsigned int retval;
168		282
169	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	283	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170	retval = npending;	284	retval = npending;
171	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	285	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172		286
173	return retval;	287	return retval;
174	}	288	}
175		289
176	unsigned int eio_nthreads (void)	290	static unsigned int etp_nthreads (void)
177	{	291	{
178	unsigned int retval;	292	unsigned int retval;
179		293
180	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	294	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181	retval = started;	295	retval = started;
…		…
188	* a somewhat faster data structure might be nice, but	302	* a somewhat faster data structure might be nice, but
189	* with 8 priorities this actually needs <20 insns	303	* with 8 priorities this actually needs <20 insns
190	* per shift, the most expensive operation.	304	* per shift, the most expensive operation.
191	*/	305	*/
192	typedef struct {	306	typedef struct {
193	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	307	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194	int size;	308	int size;
195	} reqq;	309	} etp_reqq;
196		310
197	static reqq req_queue;	311	static etp_reqq req_queue;
198	static reqq res_queue;	312	static etp_reqq res_queue;
199		313
200	static int reqq_push (reqq *q, eio_req *req)	314	static int reqq_push (etp_reqq *q, ETP_REQ *req)
201	{	315	{
202	int pri = req->pri;	316	int pri = req->pri;
203	req->next = 0;	317	req->next = 0;
204		318
205	if (q->qe[pri])	319	if (q->qe[pri])
…		…
211	q->qe[pri] = q->qs[pri] = req;	325	q->qe[pri] = q->qs[pri] = req;
212		326
213	return q->size++;	327	return q->size++;
214	}	328	}
215		329
216	static eio_req *reqq_shift (reqq *q)	330	static ETP_REQ *reqq_shift (etp_reqq *q)
217	{	331	{
218	int pri;	332	int pri;
219		333
220	if (!q->size)	334	if (!q->size)
221	return 0;	335	return 0;
222		336
223	--q->size;	337	--q->size;
224		338
225	for (pri = EIO_NUM_PRI; pri--; )	339	for (pri = ETP_NUM_PRI; pri--; )
226	{	340	{
227	eio_req *req = q->qs[pri];	341	eio_req *req = q->qs[pri];
228		342
229	if (req)	343	if (req)
230	{	344	{
…		…
236	}	350	}
237		351
238	abort ();	352	abort ();
239	}	353	}
240		354
241	static void grp_try_feed (eio_req *grp)	355	static void etp_atfork_prepare (void)
242	{	356	{
243	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	357	X_LOCK (wrklock);
		358	X_LOCK (reqlock);
		359	X_LOCK (reslock);
		360	#if !HAVE_PREADWRITE
		361	X_LOCK (preadwritelock);
		362	#endif
		363	}
		364
		365	static void etp_atfork_parent (void)
		366	{
		367	#if !HAVE_PREADWRITE
		368	X_UNLOCK (preadwritelock);
		369	#endif
		370	X_UNLOCK (reslock);
		371	X_UNLOCK (reqlock);
		372	X_UNLOCK (wrklock);
		373	}
		374
		375	static void etp_atfork_child (void)
		376	{
		377	ETP_REQ *prv;
		378
		379	while ((prv = reqq_shift (&req_queue)))
		380	ETP_DESTROY (prv);
		381
		382	while ((prv = reqq_shift (&res_queue)))
		383	ETP_DESTROY (prv);
		384
		385	while (wrk_first.next != &wrk_first)
244	{	386	{
245	int old_len = grp->size;	387	etp_worker *wrk = wrk_first.next;
246		388
247	EIO_FEED (grp);	389	if (wrk->req)
		390	ETP_DESTROY (wrk->req);
248		391
249	/* stop if no progress has been made */	392	etp_worker_clear (wrk);
250	if (old_len == grp->size)	393	etp_worker_free (wrk);
251	{
252	grp->feed = 0;
253	break;
254	}
255	}	394	}
256	}
257		395
258	static int eio_finish (eio_req *req);	396	started = 0;
		397	idle = 0;
		398	nreqs = 0;
		399	nready = 0;
		400	npending = 0;
259		401
260	static int grp_dec (eio_req *grp)	402	etp_atfork_parent ();
261	{	403	}
262	--grp->size;
263		404
264	/* call feeder, if applicable */	405	static void
265	grp_try_feed (grp);	406	etp_once_init (void)
		407	{
		408	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
		409	}
266		410
267	/* finish, if done */	411	static int
268	if (!grp->size && grp->int1)	412	etp_init (void (*want_poll)(void), void (*done_poll)(void))
269	return eio_finish (grp);	413	{
270	else	414	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
		415
		416	pthread_once (&doinit, etp_once_init);
		417
		418	want_poll_cb = want_poll;
		419	done_poll_cb = done_poll;
		420
271	return 0;	421	return 0;
272	}	422	}
273		423
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);	424	X_THREAD_PROC (etp_proc);
323		425
324	static void start_thread (void)	426	static void etp_start_thread (void)
325	{	427	{
326	worker *wrk = calloc (1, sizeof (worker));	428	etp_worker *wrk = calloc (1, sizeof (etp_worker));
327		429
328	if (!wrk)	430	/*TODO*/
329	croak ("unable to allocate worker thread data");	431	assert (("unable to allocate worker thread data", wrk));
330		432
331	X_LOCK (wrklock);	433	X_LOCK (wrklock);
332		434
333	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	435	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334	{	436	{
335	wrk->prev = &wrk_first;	437	wrk->prev = &wrk_first;
336	wrk->next = wrk_first.next;	438	wrk->next = wrk_first.next;
337	wrk_first.next->prev = wrk;	439	wrk_first.next->prev = wrk;
338	wrk_first.next = wrk;	440	wrk_first.next = wrk;
…		…
342	free (wrk);	444	free (wrk);
343		445
344	X_UNLOCK (wrklock);	446	X_UNLOCK (wrklock);
345	}	447	}
346		448
347	static void maybe_start_thread (void)	449	static void etp_maybe_start_thread (void)
348	{	450	{
349	if (eio_nthreads () >= wanted)	451	if (expect_true (etp_nthreads () >= wanted))
350	return;	452	return;
351		453
352	/* todo: maybe use idle here, but might be less exact */	454	/* todo: maybe use idle here, but might be less exact */
353	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	455	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354	return;	456	return;
355		457
356	start_thread ();	458	etp_start_thread ();
357	}	459	}
358		460
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)	461	static void etp_end_thread (void)
373	{	462	{
374	eio_req *req = calloc (1, sizeof (eio_req));	463	eio_req *req = calloc (1, sizeof (eio_req));
375		464
376	req->type = EIO_QUIT;	465	req->type = -1;
377	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	466	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
378		467
379	X_LOCK (reqlock);	468	X_LOCK (reqlock);
380	reqq_push (&req_queue, req);	469	reqq_push (&req_queue, req);
381	X_COND_SIGNAL (reqwait);	470	X_COND_SIGNAL (reqwait);
382	X_UNLOCK (reqlock);	471	X_UNLOCK (reqlock);
…		…
384	X_LOCK (wrklock);	473	X_LOCK (wrklock);
385	--started;	474	--started;
386	X_UNLOCK (wrklock);	475	X_UNLOCK (wrklock);
387	}	476	}
388		477
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)	478	static int etp_poll (void)
426	{	479	{
427	int maxreqs = max_poll_reqs;	480	unsigned int maxreqs;
		481	unsigned int maxtime;
428	struct timeval tv_start, tv_now;	482	struct timeval tv_start, tv_now;
429	eio_req *req;
430		483
		484	X_LOCK (reslock);
		485	maxreqs = max_poll_reqs;
		486	maxtime = max_poll_time;
		487	X_UNLOCK (reslock);
		488
431	if (max_poll_time)	489	if (maxtime)
432	gettimeofday (&tv_start, 0);	490	gettimeofday (&tv_start, 0);
433		491
434	for (;;)	492	for (;;)
435	{	493	{
		494	ETP_REQ *req;
		495
436	maybe_start_thread ();	496	etp_maybe_start_thread ();
437		497
438	X_LOCK (reslock);	498	X_LOCK (reslock);
439	req = reqq_shift (&res_queue);	499	req = reqq_shift (&res_queue);
440		500
441	if (req)	501	if (req)
…		…
449	X_UNLOCK (reslock);	509	X_UNLOCK (reslock);
450		510
451	if (!req)	511	if (!req)
452	return 0;	512	return 0;
453		513
		514	X_LOCK (reqlock);
454	--nreqs;	515	--nreqs;
		516	X_UNLOCK (reqlock);
455		517
456	if (req->type == EIO_GROUP && req->size)	518	if (expect_false (req->type == EIO_GROUP && req->size))
457	{	519	{
458	req->int1 = 1; /* mark request as delayed */	520	req->int1 = 1; /* mark request as delayed */
459	continue;	521	continue;
460	}	522	}
461	else	523	else
462	{	524	{
463	int res = eio_finish (req);	525	int res = ETP_FINISH (req);
464	if (res)	526	if (expect_false (res))
465	return res;	527	return res;
466	}	528	}
467		529
468	if (maxreqs && !--maxreqs)	530	if (expect_false (maxreqs && !--maxreqs))
469	break;	531	break;
470		532
471	if (max_poll_time)	533	if (maxtime)
472	{	534	{
473	gettimeofday (&tv_now, 0);	535	gettimeofday (&tv_now, 0);
474		536
475	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	537	if (tvdiff (&tv_start, &tv_now) >= maxtime)
476	break;	538	break;
477	}	539	}
478	}	540	}
479		541
480	errno = EAGAIN;	542	errno = EAGAIN;
481	return -1;	543	return -1;
482	}	544	}
483		545
		546	static void etp_cancel (ETP_REQ *req)
		547	{
		548	X_LOCK (wrklock);
		549	req->flags |= EIO_FLAG_CANCELLED;
		550	X_UNLOCK (wrklock);
		551
		552	eio_grp_cancel (req);
		553	}
		554
		555	static void etp_submit (ETP_REQ *req)
		556	{
		557	req->pri -= ETP_PRI_MIN;
		558
		559	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		560	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		561
		562	if (expect_false (req->type == EIO_GROUP))
		563	{
		564	/* I hope this is worth it :/ */
		565	X_LOCK (reqlock);
		566	++nreqs;
		567	X_UNLOCK (reqlock);
		568
		569	X_LOCK (reslock);
		570
		571	++npending;
		572
		573	if (!reqq_push (&res_queue, req) && want_poll_cb)
		574	want_poll_cb ();
		575
		576	X_UNLOCK (reslock);
		577	}
		578	else
		579	{
		580	X_LOCK (reqlock);
		581	++nreqs;
		582	++nready;
		583	reqq_push (&req_queue, req);
		584	X_COND_SIGNAL (reqwait);
		585	X_UNLOCK (reqlock);
		586
		587	etp_maybe_start_thread ();
		588	}
		589	}
		590
		591	static void etp_set_max_poll_time (double nseconds)
		592	{
		593	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		594	max_poll_time = nseconds;
		595	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		596	}
		597
		598	static void etp_set_max_poll_reqs (unsigned int maxreqs)
		599	{
		600	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		601	max_poll_reqs = maxreqs;
		602	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		603	}
		604
		605	static void etp_set_max_idle (unsigned int nthreads)
		606	{
		607	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		608	max_idle = nthreads <= 0 ? 1 : nthreads;
		609	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		610	}
		611
		612	static void etp_set_min_parallel (unsigned int nthreads)
		613	{
		614	if (wanted < nthreads)
		615	wanted = nthreads;
		616	}
		617
		618	static void etp_set_max_parallel (unsigned int nthreads)
		619	{
		620	if (wanted > nthreads)
		621	wanted = nthreads;
		622
		623	while (started > wanted)
		624	etp_end_thread ();
		625	}
		626
		627	/*****************************************************************************/
		628
		629	static void grp_try_feed (eio_req *grp)
		630	{
		631	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		632	{
		633	grp->flags &= ~EIO_FLAG_GROUPADD;
		634
		635	EIO_FEED (grp);
		636
		637	/* stop if no progress has been made */
		638	if (!(grp->flags & EIO_FLAG_GROUPADD))
		639	{
		640	grp->feed = 0;
		641	break;
		642	}
		643	}
		644	}
		645
		646	static int grp_dec (eio_req *grp)
		647	{
		648	--grp->size;
		649
		650	/* call feeder, if applicable */
		651	grp_try_feed (grp);
		652
		653	/* finish, if done */
		654	if (!grp->size && grp->int1)
		655	return eio_finish (grp);
		656	else
		657	return 0;
		658	}
		659
		660	void eio_destroy (eio_req *req)
		661	{
		662	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		663	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		664
		665	EIO_DESTROY (req);
		666	}
		667
		668	static int eio_finish (eio_req *req)
		669	{
		670	int res = EIO_FINISH (req);
		671
		672	if (req->grp)
		673	{
		674	int res2;
		675	eio_req *grp = req->grp;
		676
		677	/* unlink request */
		678	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		679	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		680
		681	if (grp->grp_first == req)
		682	grp->grp_first = req->grp_next;
		683
		684	res2 = grp_dec (grp);
		685
		686	if (!res && res2)
		687	res = res2;
		688	}
		689
		690	eio_destroy (req);
		691
		692	return res;
		693	}
		694
		695	void eio_grp_cancel (eio_req *grp)
		696	{
		697	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		698	eio_cancel (grp);
		699	}
		700
		701	void eio_cancel (eio_req *req)
		702	{
		703	etp_cancel (req);
		704	}
		705
		706	void eio_submit (eio_req *req)
		707	{
		708	etp_submit (req);
		709	}
		710
		711	unsigned int eio_nreqs (void)
		712	{
		713	return etp_nreqs ();
		714	}
		715
		716	unsigned int eio_nready (void)
		717	{
		718	return etp_nready ();
		719	}
		720
		721	unsigned int eio_npending (void)
		722	{
		723	return etp_npending ();
		724	}
		725
		726	unsigned int eio_nthreads (void)
		727	{
		728	return etp_nthreads ();
		729	}
		730
		731	void eio_set_max_poll_time (double nseconds)
		732	{
		733	etp_set_max_poll_time (nseconds);
		734	}
		735
		736	void eio_set_max_poll_reqs (unsigned int maxreqs)
		737	{
		738	etp_set_max_poll_reqs (maxreqs);
		739	}
		740
		741	void eio_set_max_idle (unsigned int nthreads)
		742	{
		743	etp_set_max_idle (nthreads);
		744	}
		745
		746	void eio_set_min_parallel (unsigned int nthreads)
		747	{
		748	etp_set_min_parallel (nthreads);
		749	}
		750
		751	void eio_set_max_parallel (unsigned int nthreads)
		752	{
		753	etp_set_max_parallel (nthreads);
		754	}
		755
		756	int eio_poll (void)
		757	{
		758	return etp_poll ();
		759	}
		760
484	/*****************************************************************************/	761	/*****************************************************************************/
485	/* work around various missing functions */	762	/* work around various missing functions */
486		763
487	#if !HAVE_PREADWRITE	764	#if !HAVE_PREADWRITE
		765	# undef pread
		766	# undef pwrite
488	# define pread aio_pread	767	# define pread eio__pread
489	# define pwrite aio_pwrite	768	# define pwrite eio__pwrite
490		769
491	/*	770	static ssize_t
492	* make our pread/pwrite safe against themselves, but not against	771	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	{	772	{
500	ssize_t res;	773	ssize_t res;
501	off_t ooffset;	774	off_t ooffset;
502		775
503	X_LOCK (preadwritelock);	776	X_LOCK (preadwritelock);
…		…
508	X_UNLOCK (preadwritelock);	781	X_UNLOCK (preadwritelock);
509		782
510	return res;	783	return res;
511	}	784	}
512		785
513	static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset)	786	static ssize_t
		787	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
514	{	788	{
515	ssize_t res;	789	ssize_t res;
516	off_t ooffset;	790	off_t ooffset;
517		791
518	X_LOCK (preadwritelock);	792	X_LOCK (preadwritelock);
519	ooffset = lseek (fd, 0, SEEK_CUR);	793	ooffset = lseek (fd, 0, SEEK_CUR);
520	lseek (fd, offset, SEEK_SET);	794	lseek (fd, offset, SEEK_SET);
521	res = write (fd, buf, count);	795	res = write (fd, buf, count);
522	lseek (fd, offset, SEEK_SET);	796	lseek (fd, ooffset, SEEK_SET);
523	X_UNLOCK (preadwritelock);	797	X_UNLOCK (preadwritelock);
524		798
525	return res;	799	return res;
526	}	800	}
527	#endif	801	#endif
528		802
529	#ifndef HAVE_FUTIMES	803	#ifndef HAVE_FUTIMES
530		804
		805	# undef utimes
		806	# undef futimes
531	# define utimes(path,times) aio_utimes (path, times)	807	# define utimes(path,times) eio__utimes (path, times)
532	# define futimes(fd,times) aio_futimes (fd, times)	808	# define futimes(fd,times) eio__futimes (fd, times)
533		809
		810	static int
534	static int aio_utimes (const char *filename, const struct timeval times[2])	811	eio__utimes (const char *filename, const struct timeval times[2])
535	{	812	{
536	if (times)	813	if (times)
537	{	814	{
538	struct utimbuf buf;	815	struct utimbuf buf;
539		816
…		…
544	}	821	}
545	else	822	else
546	return utime (filename, 0);	823	return utime (filename, 0);
547	}	824	}
548		825
549	static int aio_futimes (int fd, const struct timeval tv[2])	826	static int eio__futimes (int fd, const struct timeval tv[2])
550	{	827	{
551	errno = ENOSYS;	828	errno = ENOSYS;
552	return -1;	829	return -1;
553	}	830	}
554		831
555	#endif	832	#endif
556		833
557	#if !HAVE_FDATASYNC	834	#if !HAVE_FDATASYNC
		835	# undef fdatasync
558	# define fdatasync fsync	836	# define fdatasync(fd) fsync (fd)
559	#endif	837	#endif
		838
		839	/* sync_file_range always needs emulation */
		840	int
		841	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		842	{
		843	#if HAVE_SYNC_FILE_RANGE
		844	int res;
		845
		846	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		847	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		848	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		849	{
		850	flags = 0
		851	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		852	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		853	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		854	}
		855
		856	res = sync_file_range (fd, offset, nbytes, flags);
		857
		858	if (!res || errno != ENOSYS)
		859	return res;
		860	#endif
		861
		862	/* even though we could play tricks with the flags, it's better to always
		863	* call fdatasync, as thta matches the expectation of it's users best */
		864	return fdatasync (fd);
		865	}
560		866
561	#if !HAVE_READAHEAD	867	#if !HAVE_READAHEAD
		868	# undef readahead
562	# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self)	869	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
563		870
		871	static ssize_t
564	static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self)	872	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
565	{	873	{
566	size_t todo = count;	874	size_t todo = count;
567	dBUF;	875	dBUF;
568		876
569	while (todo > 0)	877	while (todo > 0)
…		…
579	return count;	887	return count;
580	}	888	}
581		889
582	#endif	890	#endif
583		891
584	#if !HAVE_READDIR_R
585	# define readdir_r aio_readdir_r
586
587	static mutex_t readdirlock = X_MUTEX_INIT;
588
589	static int readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
590	{
591	EIO_STRUCT_DIRENT *e;
592	int errorno;
593
594	X_LOCK (readdirlock);
595
596	e = readdir (dirp);
597	errorno = errno;
598
599	if (e)
600	{
601	*res = ent;
602	strcpy (ent->d_name, e->d_name);
603	}
604	else
605	*res = 0;
606
607	X_UNLOCK (readdirlock);
608
609	errno = errorno;
610	return e ? 0 : -1;
611	}
612	#endif
613
614	/* sendfile always needs emulation */	892	/* sendfile always needs emulation */
		893	static ssize_t
615	static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self)	894	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
616	{	895	{
617	ssize_t res;	896	ssize_t res;
618		897
619	if (!count)	898	if (!count)
620	return 0;	899	return 0;
…		…
702	}	981	}
703		982
704	return res;	983	return res;
705	}	984	}
706		985
		986	static int
		987	eio_dent_cmp (const void *a_, const void *b_)
		988	{
		989	const eio_dirent *a = (const eio_dirent *)a_;
		990	const eio_dirent *b = (const eio_dirent *)b_;
		991
		992	return (int)b->score - (int)a->score ? (int)b->score - (int)a->score
		993	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */
		994	}
		995
707	/* read a full directory */	996	/* read a full directory */
		997	static void
708	static void scandir_ (eio_req *req, worker *self)	998	eio__scandir (eio_req *req, etp_worker *self)
709	{	999	{
710	DIR *dirp;	1000	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;	1001	EIO_STRUCT_DIRENT *entp;
717	char *name, *names;	1002	unsigned char *name, *names;
718	int memlen = 4096;	1003	int namesalloc = 4096;
719	int memofs = 0;	1004	int namesoffs = 0;
		1005	int flags = req->int1;
		1006	eio_dirent *dents = 0;
		1007	int dentalloc = 128;
720	int res = 0;	1008	int dentoffs = 0;
		1009
		1010	req->result = -1;
		1011
		1012	if (!(flags & EIO_READDIR_DENTS))
		1013	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
721		1014
722	X_LOCK (wrklock);	1015	X_LOCK (wrklock);
		1016	/* the corresponding closedir is in ETP_WORKER_CLEAR */
723	self->dirp = dirp = opendir (req->ptr1);	1017	self->dirp = dirp = opendir (req->ptr1);
724	self->dbuf = u = malloc (sizeof (*u));
725	req->flags |= EIO_FLAG_PTR2_FREE;	1018	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
726	req->ptr2 = names = malloc (memlen);	1019	req->ptr1 = names = malloc (namesalloc);
		1020	req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
727	X_UNLOCK (wrklock);	1021	X_UNLOCK (wrklock);
728		1022
729	if (dirp && u && names)	1023	if (dirp && names && (!flags || dents))
730	for (;;)	1024	for (;;)
731	{	1025	{
732	errno = 0;	1026	errno = 0;
733	readdir_r (dirp, &u->d, &entp);	1027	entp = readdir (dirp);
734		1028
735	if (!entp)	1029	if (!entp)
		1030	{
		1031	if (errno)
		1032	break;
		1033
		1034	/* sort etc. */
		1035	req->int1 = flags;
		1036	req->result = dentoffs;
		1037
		1038	if (dents)
		1039	{
		1040	eio_dirent *ent = dents + dentoffs;
		1041
		1042	while (ent > dents)
		1043	(--ent)->name = names + (size_t)ent->name;
		1044	}
		1045
		1046	if (flags & EIO_READDIR_STAT_ORDER
		1047	|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))
		1048	{
		1049	/* pray your qsort doesn't use quicksort */
		1050	qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
		1051	}
		1052	else if (flags & EIO_READDIR_DIRS_FIRST)
		1053	{
		1054	/* in this case, all is known, and we just put dirs first and sort them */
		1055	eio_dirent *ent = dents + dentoffs;
		1056	eio_dirent *dir = dents;
		1057
		1058	while (ent > dir)
		1059	{
		1060	if (dir->type == DT_DIR)
		1061	++dir;
		1062	else
		1063	{
		1064	--ent;
		1065
		1066	if (ent->type == DT_DIR)
		1067	{
		1068	eio_dirent tmp = *dir;
		1069	*dir = *ent;
		1070	*ent = tmp;
		1071
		1072	++dir;
		1073	}
		1074	}
		1075	}
		1076
		1077	/* now sort the dirs only */
		1078	qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
		1079	}
		1080
		1081	/* only provide the names array unless DENTS is specified */
		1082	if (!(flags & EIO_READDIR_DENTS))
		1083	{
		1084	X_LOCK (wrklock);
		1085	assert (!dents);
		1086	req->ptr1 = 0;
		1087	req->ptr2 = names;
		1088	X_UNLOCK (wrklock);
		1089	}
		1090
736	break;	1091	break;
		1092	}
737		1093
		1094	/* now add the entry to our list(s) */
738	name = entp->d_name;	1095	name = entp->d_name;
739		1096
		1097	/* skip . and .. entries */
740	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1098	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
741	{	1099	{
742	int len = strlen (name) + 1;	1100	int len = strlen (name) + 1;
743		1101
744	res++;	1102	while (expect_false (namesoffs + len > namesalloc))
745
746	while (memofs + len > memlen)
747	{	1103	{
748	memlen *= 2;	1104	namesalloc *= 2;
749	X_LOCK (wrklock);	1105	X_LOCK (wrklock);
750	req->ptr2 = names = realloc (names, memlen);	1106	req->ptr1 = names = realloc (names, namesalloc);
751	X_UNLOCK (wrklock);	1107	X_UNLOCK (wrklock);
752		1108
753	if (!names)	1109	if (!names)
754	break;	1110	break;
755	}	1111	}
756		1112
757	memcpy (names + memofs, name, len);	1113	memcpy (names + namesoffs, name, len);
		1114
		1115	if (dents)
		1116	{
		1117	struct eio_dirent *ent;
		1118
		1119	if (expect_false (dentoffs == dentalloc))
		1120	{
		1121	dentalloc *= 2;
		1122	X_LOCK (wrklock);
		1123	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1124	X_UNLOCK (wrklock);
		1125
		1126	if (!dents)
		1127	break;
		1128	}
		1129
		1130	ent = dents + dentoffs;
		1131
		1132	ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */
		1133	ent->namelen = len - 1;
		1134	ent->inode = D_INO (entp);
		1135
		1136	switch (D_TYPE (entp))
		1137	{
		1138	default:
		1139	ent->type = EIO_DT_UNKNOWN;
		1140	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1141	break;
		1142
		1143	#ifdef DT_FIFO
		1144	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1145	#endif
		1146	#ifdef DT_CHR
		1147	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1148	#endif
		1149	#ifdef DT_DIR
		1150	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1151	#endif
		1152	#ifdef DT_BLK
		1153	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1154	#endif
		1155	#ifdef DT_REG
		1156	case DT_REG: ent->type = EIO_DT_REG; break;
		1157	#endif
		1158	#ifdef DT_LNK
		1159	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1160	#endif
		1161	#ifdef DT_SOCK
		1162	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1163	#endif
		1164	#ifdef DT_WHT
		1165	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1166	#endif
		1167	}
		1168
		1169	ent->score = 0;
		1170
		1171	if (flags & EIO_READDIR_DIRS_FIRST)
		1172	{
		1173	if (ent->type == EIO_DT_UNKNOWN)
		1174	{
		1175	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1176	ent->score = 98;
		1177	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1178	ent->score = len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */
		1179	}
		1180	else if (ent->type == DT_DIR)
		1181	ent->score = 100;
		1182	}
		1183	}
		1184
758	memofs += len;	1185	namesoffs += len;
		1186	++dentoffs;
759	}	1187	}
760	}	1188	}
761		1189	else
762	if (errno)
763	res = -1;
764
765	req->result = res;	1190	req->result = -1;
		1191	}
		1192
		1193	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1194	# undef msync
		1195	# define msync(a,b,c) ((errno = ENOSYS), -1)
		1196	#endif
		1197
		1198	int
		1199	eio__mtouch (void *mem, size_t len, int flags)
		1200	{
		1201	intptr_t addr = (intptr_t)mem;
		1202	intptr_t end = addr + len;
		1203	#ifdef PAGESIZE
		1204	const intptr_t page = PAGESIZE;
		1205	#else
		1206	static intptr_t page;
		1207
		1208	if (!page)
		1209	page = sysconf (_SC_PAGESIZE);
		1210	#endif
		1211
		1212	addr &= ~(page - 1); /* assume page size is always a power of two */
		1213
		1214	if (addr < end)
		1215	if (flags) /* modify */
		1216	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
		1217	else
		1218	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
		1219
		1220	return 0;
766	}	1221	}
767		1222
768	/*****************************************************************************/	1223	/*****************************************************************************/
769		1224
		1225	#define ALLOC(len) \
		1226	if (!req->ptr2) \
		1227	{ \
		1228	X_LOCK (wrklock); \
		1229	req->flags |= EIO_FLAG_PTR2_FREE; \
		1230	X_UNLOCK (wrklock); \
		1231	req->ptr2 = malloc (len); \
		1232	if (!req->ptr2) \
		1233	{ \
		1234	errno = ENOMEM; \
		1235	req->result = -1; \
		1236	break; \
		1237	} \
		1238	}
		1239
770	X_THREAD_PROC (eio_proc)	1240	X_THREAD_PROC (etp_proc)
771	{	1241	{
772	eio_req *req;	1242	ETP_REQ *req;
773	struct timespec ts;	1243	struct timespec ts;
774	worker *self = (worker *)thr_arg;	1244	etp_worker *self = (etp_worker *)thr_arg;
775		1245
776	/* try to distribute timeouts somewhat randomly */	1246	/* try to distribute timeouts somewhat randomly */
777	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1247	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
778		1248
779	for (;;)	1249	for (;;)
780	{	1250	{
781	ts.tv_sec = time (0) + IDLE_TIMEOUT;
782
783	X_LOCK (reqlock);	1251	X_LOCK (reqlock);
784		1252
785	for (;;)	1253	for (;;)
786	{	1254	{
787	self->req = req = reqq_shift (&req_queue);	1255	self->req = req = reqq_shift (&req_queue);
…		…
789	if (req)	1257	if (req)
790	break;	1258	break;
791		1259
792	++idle;	1260	++idle;
793		1261
		1262	ts.tv_sec = time (0) + IDLE_TIMEOUT;
794	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1263	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
795	{	1264	{
796	if (idle > max_idle)	1265	if (idle > max_idle)
797	{	1266	{
798	--idle;	1267	--idle;
…		…
803	goto quit;	1272	goto quit;
804	}	1273	}
805		1274
806	/* we are allowed to idle, so do so without any timeout */	1275	/* we are allowed to idle, so do so without any timeout */
807	X_COND_WAIT (reqwait, reqlock);	1276	X_COND_WAIT (reqwait, reqlock);
808	ts.tv_sec = time (0) + IDLE_TIMEOUT;
809	}	1277	}
810		1278
811	--idle;	1279	--idle;
812	}	1280	}
813		1281
814	--nready;	1282	--nready;
815		1283
816	X_UNLOCK (reqlock);	1284	X_UNLOCK (reqlock);
817		1285
818	errno = 0; /* strictly unnecessary */	1286	if (req->type < 0)
		1287	goto quit;
819		1288
820	if (!EIO_CANCELLED (req))	1289	if (!EIO_CANCELLED (req))
821	switch (req->type)	1290	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		1291
915	X_LOCK (reslock);	1292	X_LOCK (reslock);
916		1293
917	++npending;	1294	++npending;
918		1295
919	if (!reqq_push (&res_queue, req) && want_poll_cb)	1296	if (!reqq_push (&res_queue, req) && want_poll_cb)
920	want_poll_cb ();	1297	want_poll_cb ();
921		1298
922	self->req = 0;	1299	self->req = 0;
923	worker_clear (self);	1300	etp_worker_clear (self);
924		1301
925	X_UNLOCK (reslock);	1302	X_UNLOCK (reslock);
926	}	1303	}
927		1304
928	quit:	1305	quit:
929	X_LOCK (wrklock);	1306	X_LOCK (wrklock);
930	worker_free (self);	1307	etp_worker_free (self);
931	X_UNLOCK (wrklock);	1308	X_UNLOCK (wrklock);
932		1309
933	return 0;	1310	return 0;
934	}	1311	}
935		1312
936	/*****************************************************************************/	1313	/*****************************************************************************/
937		1314
938	static void eio_atfork_prepare (void)	1315	int eio_init (void (*want_poll)(void), void (*done_poll)(void))
939	{	1316	{
940	X_LOCK (wrklock);	1317	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	}	1318	}
950		1319
951	static void eio_atfork_parent (void)	1320	static void eio_api_destroy (eio_req *req)
952	{	1321	{
953	#if !HAVE_READDIR_R	1322	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	}	1323	}
963		1324
964	static void eio_atfork_child (void)	1325	#define REQ(rtype) \
965	{	1326	eio_req *req; \
966	eio_req *prv;	1327	\
		1328	req = (eio_req *)calloc (1, sizeof *req); \
		1329	if (!req) \
		1330	return 0; \
		1331	\
		1332	req->type = rtype; \
		1333	req->pri = pri; \
		1334	req->finish = cb; \
		1335	req->data = data; \
		1336	req->destroy = eio_api_destroy;
967		1337
968	while (prv = reqq_shift (&req_queue))	1338	#define SEND eio_submit (req); return req
969	eio_destroy (prv);
970		1339
971	while (prv = reqq_shift (&res_queue))	1340	#define PATH \
972	eio_destroy (prv);	1341	req->flags |= EIO_FLAG_PTR1_FREE; \
		1342	req->ptr1 = strdup (path); \
		1343	if (!req->ptr1) \
		1344	{ \
		1345	eio_api_destroy (req); \
		1346	return 0; \
		1347	}
973		1348
974	while (wrk_first.next != &wrk_first)	1349	static void eio_execute (etp_worker *self, eio_req *req)
		1350	{
		1351	errno = 0;
		1352
		1353	switch (req->type)
975	{	1354	{
976	worker *wrk = wrk_first.next;	1355	case EIO_READ: ALLOC (req->size);
		1356	req->result = req->offs >= 0
		1357	? pread (req->int1, req->ptr2, req->size, req->offs)
		1358	: read (req->int1, req->ptr2, req->size); break;
		1359	case EIO_WRITE: req->result = req->offs >= 0
		1360	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1361	: write (req->int1, req->ptr2, req->size); break;
977		1362
978	if (wrk->req)	1363	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
979	eio_destroy (wrk->req);	1364	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
980		1365
981	worker_clear (wrk);	1366	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
982	worker_free (wrk);	1367	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
983	}	1368	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1369	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1370	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1371	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
984		1372
985	started = 0;	1373	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
986	idle = 0;	1374	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
987	nreqs = 0;	1375	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
988	nready = 0;	1376	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
989	npending = 0;	1377	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1378	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
990		1379
991	eio_atfork_parent ();	1380	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
992	}	1381	case EIO_CLOSE: req->result = close (req->int1); break;
		1382	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1383	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1384	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1385	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1386	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1387	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1388	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1389	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
993		1390
994	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1391	case EIO_READLINK: ALLOC (NAME_MAX);
995	{	1392	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
996	want_poll_cb = want_poll;
997	done_poll_cb = done_poll;
998		1393
		1394	case EIO_SYNC: req->result = 0; sync (); break;
		1395	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1396	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1397	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
		1398	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1399	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1400
		1401	case EIO_READDIR: eio__scandir (req, self); break;
		1402
		1403	case EIO_BUSY:
999	#ifdef _WIN32	1404	#ifdef _WIN32
1000	X_MUTEX_CHECK (wrklock);	1405	Sleep (req->nv1 * 1000.);
1001	X_MUTEX_CHECK (reslock);	1406	#else
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	}
1012
1013	#if 0
1014
1015	eio_req *eio_fsync (int fd, eio_cb cb);
1016	eio_req *eio_fdatasync (int fd, eio_cb cb);
1017	eio_req *eio_dupclose (int fd, eio_cb cb);
1018	eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb);
1019	eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1020	eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1021	eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1022	eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb);
1023	eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb);
1024	eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb);
1025	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb);
1026	eio_req *eio_dup2 (int fd, int fd2, eio_cb cb);
1027	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb);
1028	eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb);
1029	eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1030	eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1031	eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1032	eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb);
1033	eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb);
1034	eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb);
1035	eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb);
1036	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb);
1037	eio_req *eio_unlink (const char *path, eio_cb cb);
1038	eio_req *eio_rmdir (const char *path, eio_cb cb);
1039	eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1040	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb);
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: $$$;$
1046	PPCODE:
1047	{
1048	dREQ;
1049
1050	req->type = EIO_OPEN;
1051	req->sv1 = newSVsv (pathname);
1052	req->ptr1 = SvPVbyte_nolen (req->sv1);
1053	req->int1 = flags;
1054	req->int2 = mode;
1055
1056	EIO_SEND;
1057	}
1058
1059	void
1060	aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1061	PROTOTYPE: $;$
1062	ALIAS:
1063	aio_fsync = EIO_FSYNC
1064	aio_fdatasync = EIO_FDATASYNC
1065	PPCODE:
1066	{
1067	dREQ;
1068
1069	req->type = ix;
1070	req->sv1 = newSVsv (fh);
1071	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1072
1073	EIO_SEND (req);
1074	}
1075
1076	void
1077	aio_close (SV *fh, SV *callback=&PL_sv_undef)
1078	PROTOTYPE: $;$
1079	PPCODE:
1080	{
1081	dREQ;
1082
1083	req->type = EIO_CLOSE;
1084	req->sv1 = newSVsv (fh);
1085	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1086
1087	EIO_SEND (req);
1088	}
1089
1090	void
1091	aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1092	ALIAS:
1093	aio_read = EIO_READ
1094	aio_write = EIO_WRITE
1095	PROTOTYPE: $$$$$;$
1096	PPCODE:
1097	{
1098	STRLEN svlen;
1099	char *svptr = SvPVbyte (data, svlen);
1100	UV len = SvUV (length);
1101
1102	SvUPGRADE (data, SVt_PV);
1103	SvPOK_on (data);
1104
1105	if (dataoffset < 0)
1106	dataoffset += svlen;
1107
1108	if (dataoffset < 0 || dataoffset > svlen)
1109	croak ("dataoffset outside of data scalar");
1110
1111	if (ix == EIO_WRITE)
1112	{
1113	/* write: check length and adjust. */
1114	if (!SvOK (length) || len + dataoffset > svlen)
1115	len = svlen - dataoffset;
1116	}
1117	else
1118	{
1119	/* read: grow scalar as necessary */
1120	svptr = SvGROW (data, len + dataoffset + 1);
1121	}
1122
1123	if (len < 0)
1124	croak ("length must not be negative");
1125
1126	{	1407	{
1127	dREQ;	1408	struct timeval tv;
1128		1409
1129	req->type = ix;	1410	tv.tv_sec = req->nv1;
1130	req->sv1 = newSVsv (fh);	1411	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
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		1412
1139	if (!SvREADONLY (data))	1413	req->result = select (0, 0, 0, 0, &tv);
		1414	}
		1415	#endif
		1416	break;
		1417
		1418	case EIO_UTIME:
		1419	case EIO_FUTIME:
		1420	{
		1421	struct timeval tv[2];
		1422	struct timeval *times;
		1423
		1424	if (req->nv1 != -1. || req->nv2 != -1.)
1140	{	1425	{
1141	SvREADONLY_on (data);	1426	tv[0].tv_sec = req->nv1;
1142	req->flags |= FLAG_SV2_RO_OFF;	1427	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		1428	tv[1].tv_sec = req->nv2;
		1429	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		1430
		1431	times = tv;
1143	}	1432	}
		1433	else
		1434	times = 0;
1144		1435
1145	EIO_SEND;	1436
		1437	req->result = req->type == EIO_FUTIME
		1438	? futimes (req->int1, times)
		1439	: utimes (req->ptr1, times);
1146	}	1440	}
1147	}	1441	break;
1148		1442
1149	void	1443	case EIO_GROUP:
1150	aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)	1444	abort (); /* handled in eio_request */
1151	PROTOTYPE: $$;$
1152	PPCODE:
1153	{
1154	SV *data;
1155	dREQ;
1156		1445
1157	data = newSV (NAME_MAX);	1446	case EIO_NOP:
1158	SvPOK_on (data);	1447	req->result = 0;
		1448	break;
1159		1449
1160	req->type = EIO_READLINK;	1450	case EIO_CUSTOM:
1161	req->sv1 = newSVsv (path);	1451	((void (*)(eio_req *))req->feed) (req);
1162	req->ptr1 = SvPVbyte_nolen (req->sv1);	1452	break;
1163	req->sv2 = data;
1164	req->ptr2 = SvPVbyte_nolen (data);
1165		1453
1166	EIO_SEND;	1454	default:
1167	}	1455	req->result = -1;
		1456	break;
		1457	}
1168		1458
1169	void	1459	req->errorno = errno;
1170	aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1171	PROTOTYPE: $$$$;$
1172	PPCODE:
1173	{
1174	dREQ;
1175
1176	req->type = EIO_SENDFILE;
1177	req->sv1 = newSVsv (out_fh);
1178	req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1179	req->sv2 = newSVsv (in_fh);
1180	req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1181	req->offs = SvVAL64 (in_offset);
1182	req->size = length;
1183
1184	EIO_SEND;
1185	}	1460	}
1186		1461
1187	void	1462	#ifndef EIO_NO_WRAPPERS
1188	aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1189	PROTOTYPE: $$$;$
1190	PPCODE:
1191	{
1192	dREQ;
1193		1463
1194	req->type = EIO_READAHEAD;	1464	eio_req *eio_nop (int pri, eio_cb cb, void *data)
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	{	1465	{
1210	dREQ;	1466	REQ (EIO_NOP); SEND;
		1467	}
1211		1468
1212	req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));	1469	eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1213	if (!req->ptr2)	1470	{
1214	{	1471	REQ (EIO_BUSY); req->nv1 = delay; SEND;
1215	req_destroy (req);	1472	}
1216	croak ("out of memory during aio_stat statdata allocation");
1217	}
1218		1473
		1474	eio_req *eio_sync (int pri, eio_cb cb, void *data)
		1475	{
		1476	REQ (EIO_SYNC); SEND;
		1477	}
		1478
		1479	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
		1480	{
		1481	REQ (EIO_FSYNC); req->int1 = fd; SEND;
		1482	}
		1483
		1484	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1485	{
		1486	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1487	}
		1488
		1489	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1490	{
		1491	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1492	}
		1493
		1494	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		1495	{
		1496	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		1497	}
		1498
		1499	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		1500	{
		1501	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		1502	}
		1503
		1504	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
		1505	{
		1506	REQ (EIO_CLOSE); req->int1 = fd; SEND;
		1507	}
		1508
		1509	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
		1510	{
		1511	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
		1512	}
		1513
		1514	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
		1515	{
		1516	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
		1517	}
		1518
		1519	eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
		1520	{
		1521	REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
		1522	}
		1523
		1524	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
		1525	{
		1526	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1527	}
		1528
		1529	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
		1530	{
		1531	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
		1532	}
		1533
		1534	eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
		1535	{
		1536	REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
		1537	}
		1538
		1539	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
		1540	{
		1541	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
		1542	}
		1543
		1544	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
		1545	{
		1546	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
		1547	}
		1548
		1549	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
		1550	{
		1551	REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
		1552	}
		1553
		1554	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
		1555	{
		1556	REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
		1557	}
		1558
		1559	eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
		1560	{
		1561	REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
		1562	}
		1563
		1564	eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
		1565	{
		1566	REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
		1567	}
		1568
		1569	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
		1570	{
		1571	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
		1572	}
		1573
		1574	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
		1575	{
		1576	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
		1577	}
		1578
		1579	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
		1580	{
		1581	REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
		1582	}
		1583
		1584	eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
		1585	{
		1586	REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
		1587	}
		1588
		1589	static eio_req *
		1590	eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
		1591	{
		1592	REQ (type); PATH; SEND;
		1593	}
		1594
		1595	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
		1596	{
		1597	return eio__1path (EIO_READLINK, path, pri, cb, data);
		1598	}
		1599
		1600	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
		1601	{
		1602	return eio__1path (EIO_STAT, path, pri, cb, data);
		1603	}
		1604
		1605	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
		1606	{
		1607	return eio__1path (EIO_LSTAT, path, pri, cb, data);
		1608	}
		1609
		1610	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
		1611	{
		1612	return eio__1path (EIO_UNLINK, path, pri, cb, data);
		1613	}
		1614
		1615	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
		1616	{
		1617	return eio__1path (EIO_RMDIR, path, pri, cb, data);
		1618	}
		1619
		1620	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
		1621	{
		1622	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
		1623	}
		1624
		1625	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
		1626	{
		1627	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
		1628	}
		1629
		1630	static eio_req *
		1631	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		1632	{
		1633	REQ (type); PATH;
		1634
1219	req->flags |= FLAG_PTR2_FREE;	1635	req->flags |= EIO_FLAG_PTR2_FREE;
1220	req->sv1 = newSVsv (fh_or_path);	1636	req->ptr2 = strdup (new_path);
		1637	if (!req->ptr2)
		1638	{
		1639	eio_api_destroy (req);
		1640	return 0;
		1641	}
1221		1642
1222	if (SvPOK (fh_or_path))	1643	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	}	1644	}
1235		1645
1236	void	1646	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	{	1647	{
1240	dREQ;	1648	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	}	1649	}
1259		1650
1260	void	1651	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	{	1652	{
1264	dREQ;	1653	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	}	1654	}
1282		1655
1283	void	1656	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	{	1657	{
1291	dREQ;	1658	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	}	1659	}
1304		1660
1305	void	1661	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	{	1662	{
1309	dREQ;	1663	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	}	1664	}
1328		1665
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	1666	#endif
		1667
		1668	eio_req *eio_grp (eio_cb cb, void *data)
		1669	{
		1670	const int pri = EIO_PRI_MAX;
		1671
		1672	REQ (EIO_GROUP); SEND;
		1673	}
		1674
		1675	#undef REQ
		1676	#undef PATH
		1677	#undef SEND
		1678
		1679	/*****************************************************************************/
		1680	/* grp functions */
1415		1681
1416	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	1682	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1417	{	1683	{
1418	grp->int2 = limit;	1684	grp->int2 = limit;
1419	grp->feed = feed;	1685	grp->feed = feed;
…		…
1430		1696
1431	void eio_grp_add (eio_req *grp, eio_req *req)	1697	void eio_grp_add (eio_req *grp, eio_req *req)
1432	{	1698	{
1433	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	1699	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1434		1700
		1701	grp->flags |= EIO_FLAG_GROUPADD;
		1702
1435	++grp->size;	1703	++grp->size;
1436	req->grp = grp;	1704	req->grp = grp;
1437		1705
1438	req->grp_prev = 0;	1706	req->grp_prev = 0;
1439	req->grp_next = grp->grp_first;	1707	req->grp_next = grp->grp_first;
…		…
1442	grp->grp_first->grp_prev = req;	1710	grp->grp_first->grp_prev = req;
1443		1711
1444	grp->grp_first = req;	1712	grp->grp_first = req;
1445	}	1713	}
1446		1714
		1715	/*****************************************************************************/
		1716	/* misc garbage */
1447		1717
		1718	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
		1719	{
		1720	etp_worker wrk;
		1721
		1722	wrk.dbuf = 0;
		1723
		1724	eio__sendfile (ofd, ifd, offset, count, &wrk);
		1725
		1726	if (wrk.dbuf)
		1727	free (wrk.dbuf);
		1728	}
		1729

Diff Legend

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