ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.11 by root, Tue May 13 18:50:27 2008 UTC vs.
Revision 1.30 by root, Wed Jun 3 12:24:49 2009 UTC

		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008 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>
…		…
70	/* buffer size for various temporary buffers */	113	/* buffer size for various temporary buffers */
71	#define EIO_BUFSIZE 65536	114	#define EIO_BUFSIZE 65536
72		115
73	#define dBUF \	116	#define dBUF \
74	char *eio_buf; \	117	char *eio_buf; \
75	X_LOCK (wrklock); \	118	ETP_WORKER_LOCK (self); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	119	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (wrklock); \	120	ETP_WORKER_UNLOCK (self); \
78	errno = ENOMEM; \	121	errno = ENOMEM; \
79	if (!eio_buf) \	122	if (!eio_buf) \
80	return -1;	123	return -1;
81		124
82	#define EIO_TICKS ((1000000 + 1023) >> 10)	125	#define EIO_TICKS ((1000000 + 1023) >> 10)
83		126
		127	/*****************************************************************************/
		128
		129	#if __GNUC__ >= 3
		130	# define expect(expr,value) __builtin_expect ((expr),(value))
		131	#else
		132	# define expect(expr,value) (expr)
		133	#endif
		134
		135	#define expect_false(expr) expect ((expr) != 0, 0)
		136	#define expect_true(expr) expect ((expr) != 0, 1)
		137
		138	/*****************************************************************************/
		139
		140	#define ETP_PRI_MIN EIO_PRI_MIN
		141	#define ETP_PRI_MAX EIO_PRI_MAX
		142
		143	struct etp_worker;
		144
		145	#define ETP_REQ eio_req
		146	#define ETP_DESTROY(req) eio_destroy (req)
		147	static int eio_finish (eio_req *req);
		148	#define ETP_FINISH(req) eio_finish (req)
		149	static void eio_execute (struct etp_worker *self, eio_req *req);
		150	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		151
		152	#define ETP_WORKER_CLEAR(req) \
		153	if (wrk->dbuf) \
		154	{ \
		155	free (wrk->dbuf); \
		156	wrk->dbuf = 0; \
		157	} \
		158	\
		159	if (wrk->dirp) \
		160	{ \
		161	closedir (wrk->dirp); \
		162	wrk->dirp = 0; \
		163	}
		164	#define ETP_WORKER_COMMON \
		165	void *dbuf; \
		166	DIR *dirp;
		167
		168	/*****************************************************************************/
		169
		170	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		171
		172	/* calculcate time difference in ~1/EIO_TICKS of a second */
		173	static int tvdiff (struct timeval *tv1, struct timeval *tv2)
		174	{
		175	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
		176	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
		177	}
		178
		179	static unsigned int started, idle, wanted = 4;
		180
84	static void (*want_poll_cb) (void);	181	static void (*want_poll_cb) (void);
85	static void (*done_poll_cb) (void);	182	static void (*done_poll_cb) (void);
86		183
87	static unsigned int max_poll_time = 0;	184	static unsigned int max_poll_time; /* reslock */
88	static unsigned int max_poll_reqs = 0;	185	static unsigned int max_poll_reqs; /* reslock */
89		186
90	/* calculcate time difference in ~1/EIO_TICKS of a second */	187	static volatile unsigned int nreqs; /* reqlock */
91	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	188	static volatile unsigned int nready; /* reqlock */
92	{	189	static volatile unsigned int npending; /* reqlock */
93	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	190	static volatile unsigned int max_idle = 4;
94	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
95	}
96		191
97	static unsigned int started, idle, wanted = 4;
98
99	/* worker threads management */
100	static mutex_t wrklock = X_MUTEX_INIT;	192	static mutex_t wrklock = X_MUTEX_INIT;
101
102	typedef struct worker
103	{
104	/* locked by wrklock */
105	struct worker *prev, *next;
106
107	thread_t tid;
108
109	/* locked by reslock, reqlock or wrklock */
110	eio_req *req; /* currently processed request */
111	void *dbuf;
112	DIR *dirp;
113	} worker;
114
115	static worker wrk_first = { &wrk_first, &wrk_first, 0 };
116
117	static void worker_clear (worker *wrk)
118	{
119	if (wrk->dirp)
120	{
121	closedir (wrk->dirp);
122	wrk->dirp = 0;
123	}
124
125	if (wrk->dbuf)
126	{
127	free (wrk->dbuf);
128	wrk->dbuf = 0;
129	}
130	}
131
132	static void worker_free (worker *wrk)
133	{
134	wrk->next->prev = wrk->prev;
135	wrk->prev->next = wrk->next;
136
137	free (wrk);
138	}
139
140	static volatile unsigned int nreqs, nready, npending;
141	static volatile unsigned int max_idle = 4;
142
143	static mutex_t reslock = X_MUTEX_INIT;	193	static mutex_t reslock = X_MUTEX_INIT;
144	static mutex_t reqlock = X_MUTEX_INIT;	194	static mutex_t reqlock = X_MUTEX_INIT;
145	static cond_t reqwait = X_COND_INIT;	195	static cond_t reqwait = X_COND_INIT;
146		196
		197	#if !HAVE_PREADWRITE
		198	/*
		199	* make our pread/pwrite emulation safe against themselves, but not against
		200	* normal read/write by using a mutex. slows down execution a lot,
		201	* but that's your problem, not mine.
		202	*/
		203	static mutex_t preadwritelock = X_MUTEX_INIT;
		204	#endif
		205
		206	typedef struct etp_worker
		207	{
		208	/* locked by wrklock */
		209	struct etp_worker *prev, *next;
		210
		211	thread_t tid;
		212
		213	/* locked by reslock, reqlock or wrklock */
		214	ETP_REQ *req; /* currently processed request */
		215
		216	ETP_WORKER_COMMON
		217	} etp_worker;
		218
		219	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
		220
		221	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		222	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
		223
		224	/* worker threads management */
		225
		226	static void etp_worker_clear (etp_worker *wrk)
		227	{
		228	ETP_WORKER_CLEAR (wrk);
		229	}
		230
		231	static void etp_worker_free (etp_worker *wrk)
		232	{
		233	wrk->next->prev = wrk->prev;
		234	wrk->prev->next = wrk->next;
		235
		236	free (wrk);
		237	}
		238
147	unsigned int eio_nreqs (void)	239	static unsigned int etp_nreqs (void)
148	{	240	{
		241	int retval;
		242	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		243	retval = nreqs;
		244	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
149	return nreqs;	245	return retval;
150	}	246	}
151		247
152	unsigned int eio_nready (void)	248	static unsigned int etp_nready (void)
153	{	249	{
154	unsigned int retval;	250	unsigned int retval;
155		251
156	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	252	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157	retval = nready;	253	retval = nready;
158	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	254	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159		255
160	return retval;	256	return retval;
161	}	257	}
162		258
163	unsigned int eio_npending (void)	259	static unsigned int etp_npending (void)
164	{	260	{
165	unsigned int retval;	261	unsigned int retval;
166		262
167	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	263	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168	retval = npending;	264	retval = npending;
169	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	265	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170		266
171	return retval;	267	return retval;
172	}	268	}
173		269
174	unsigned int eio_nthreads (void)	270	static unsigned int etp_nthreads (void)
175	{	271	{
176	unsigned int retval;	272	unsigned int retval;
177		273
178	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	274	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179	retval = started;	275	retval = started;
…		…
186	* a somewhat faster data structure might be nice, but	282	* a somewhat faster data structure might be nice, but
187	* with 8 priorities this actually needs <20 insns	283	* with 8 priorities this actually needs <20 insns
188	* per shift, the most expensive operation.	284	* per shift, the most expensive operation.
189	*/	285	*/
190	typedef struct {	286	typedef struct {
191	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	287	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
192	int size;	288	int size;
193	} reqq;	289	} etp_reqq;
194		290
195	static reqq req_queue;	291	static etp_reqq req_queue;
196	static reqq res_queue;	292	static etp_reqq res_queue;
197		293
198	static int reqq_push (reqq *q, eio_req *req)	294	static int reqq_push (etp_reqq *q, ETP_REQ *req)
199	{	295	{
200	int pri = req->pri;	296	int pri = req->pri;
201	req->next = 0;	297	req->next = 0;
202		298
203	if (q->qe[pri])	299	if (q->qe[pri])
…		…
209	q->qe[pri] = q->qs[pri] = req;	305	q->qe[pri] = q->qs[pri] = req;
210		306
211	return q->size++;	307	return q->size++;
212	}	308	}
213		309
214	static eio_req *reqq_shift (reqq *q)	310	static ETP_REQ *reqq_shift (etp_reqq *q)
215	{	311	{
216	int pri;	312	int pri;
217		313
218	if (!q->size)	314	if (!q->size)
219	return 0;	315	return 0;
220		316
221	--q->size;	317	--q->size;
222		318
223	for (pri = EIO_NUM_PRI; pri--; )	319	for (pri = ETP_NUM_PRI; pri--; )
224	{	320	{
225	eio_req *req = q->qs[pri];	321	eio_req *req = q->qs[pri];
226		322
227	if (req)	323	if (req)
228	{	324	{
…		…
234	}	330	}
235		331
236	abort ();	332	abort ();
237	}	333	}
238		334
239	static void grp_try_feed (eio_req *grp)	335	static void etp_atfork_prepare (void)
240	{	336	{
241	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	337	X_LOCK (wrklock);
		338	X_LOCK (reqlock);
		339	X_LOCK (reslock);
		340	#if !HAVE_PREADWRITE
		341	X_LOCK (preadwritelock);
		342	#endif
		343	}
		344
		345	static void etp_atfork_parent (void)
		346	{
		347	#if !HAVE_PREADWRITE
		348	X_UNLOCK (preadwritelock);
		349	#endif
		350	X_UNLOCK (reslock);
		351	X_UNLOCK (reqlock);
		352	X_UNLOCK (wrklock);
		353	}
		354
		355	static void etp_atfork_child (void)
		356	{
		357	ETP_REQ *prv;
		358
		359	while ((prv = reqq_shift (&req_queue)))
		360	ETP_DESTROY (prv);
		361
		362	while ((prv = reqq_shift (&res_queue)))
		363	ETP_DESTROY (prv);
		364
		365	while (wrk_first.next != &wrk_first)
242	{	366	{
243	int old_len = grp->size;	367	etp_worker *wrk = wrk_first.next;
244		368
245	EIO_FEED (grp);	369	if (wrk->req)
		370	ETP_DESTROY (wrk->req);
246		371
247	/* stop if no progress has been made */	372	etp_worker_clear (wrk);
248	if (old_len == grp->size)	373	etp_worker_free (wrk);
249	{
250	grp->feed = 0;
251	break;
252	}
253	}	374	}
254	}
255		375
256	static int eio_finish (eio_req *req);	376	started = 0;
		377	idle = 0;
		378	nreqs = 0;
		379	nready = 0;
		380	npending = 0;
257		381
258	static int grp_dec (eio_req *grp)	382	etp_atfork_parent ();
259	{	383	}
260	--grp->size;
261		384
262	/* call feeder, if applicable */	385	static void
263	grp_try_feed (grp);	386	etp_once_init (void)
		387	{
		388	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
		389	}
264		390
265	/* finish, if done */	391	static int
266	if (!grp->size && grp->int1)	392	etp_init (void (*want_poll)(void), void (*done_poll)(void))
267	return eio_finish (grp);	393	{
268	else	394	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
		395
		396	pthread_once (&doinit, etp_once_init);
		397
		398	want_poll_cb = want_poll;
		399	done_poll_cb = done_poll;
		400
269	return 0;	401	return 0;
270	}	402	}
271		403
272	void eio_destroy (eio_req *req)
273	{
274	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
275	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
276
277	EIO_DESTROY (req);
278	}
279
280	static int eio_finish (eio_req *req)
281	{
282	int res = EIO_FINISH (req);
283
284	if (req->grp)
285	{
286	int res2;
287	eio_req *grp = req->grp;
288
289	/* unlink request */
290	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
291	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
292
293	if (grp->grp_first == req)
294	grp->grp_first = req->grp_next;
295
296	res2 = grp_dec (grp);
297
298	if (!res && res2)
299	res = res2;
300	}
301
302	eio_destroy (req);
303
304	return res;
305	}
306
307	void eio_grp_cancel (eio_req *grp)
308	{
309	for (grp = grp->grp_first; grp; grp = grp->grp_next)
310	eio_cancel (grp);
311	}
312
313	void eio_cancel (eio_req *req)
314	{
315	X_LOCK (wrklock);
316	req->flags |= EIO_FLAG_CANCELLED;
317	X_UNLOCK (wrklock);
318
319	eio_grp_cancel (req);
320	}
321
322	X_THREAD_PROC (eio_proc);	404	X_THREAD_PROC (etp_proc);
323		405
324	static void start_thread (void)	406	static void etp_start_thread (void)
325	{	407	{
326	worker *wrk = calloc (1, sizeof (worker));	408	etp_worker *wrk = calloc (1, sizeof (etp_worker));
327		409
328	/*TODO*/	410	/*TODO*/
329	assert (("unable to allocate worker thread data", wrk));	411	assert (("unable to allocate worker thread data", wrk));
330		412
331	X_LOCK (wrklock);	413	X_LOCK (wrklock);
332		414
333	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	415	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334	{	416	{
335	wrk->prev = &wrk_first;	417	wrk->prev = &wrk_first;
336	wrk->next = wrk_first.next;	418	wrk->next = wrk_first.next;
337	wrk_first.next->prev = wrk;	419	wrk_first.next->prev = wrk;
338	wrk_first.next = wrk;	420	wrk_first.next = wrk;
…		…
342	free (wrk);	424	free (wrk);
343		425
344	X_UNLOCK (wrklock);	426	X_UNLOCK (wrklock);
345	}	427	}
346		428
347	static void maybe_start_thread (void)	429	static void etp_maybe_start_thread (void)
348	{	430	{
349	if (eio_nthreads () >= wanted)	431	if (expect_true (etp_nthreads () >= wanted))
350	return;	432	return;
351		433
352	/* todo: maybe use idle here, but might be less exact */	434	/* todo: maybe use idle here, but might be less exact */
353	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	435	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354	return;	436	return;
355		437
356	start_thread ();	438	etp_start_thread ();
357	}	439	}
358		440
359	void eio_submit (eio_req *req)
360	{
361	req->pri += EIO_PRI_BIAS;
362
363	if (req->pri < EIO_PRI_MIN + EIO_PRI_BIAS) req->pri = EIO_PRI_MIN + EIO_PRI_BIAS;
364	if (req->pri > EIO_PRI_MAX + EIO_PRI_BIAS) req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
365
366	++nreqs;
367
368	X_LOCK (reqlock);
369	++nready;
370	reqq_push (&req_queue, req);
371	X_COND_SIGNAL (reqwait);
372	X_UNLOCK (reqlock);
373
374	maybe_start_thread ();
375	}
376
377	static void end_thread (void)	441	static void etp_end_thread (void)
378	{	442	{
379	eio_req *req = calloc (1, sizeof (eio_req));	443	eio_req *req = calloc (1, sizeof (eio_req));
380		444
381	req->type = EIO_QUIT;	445	req->type = -1;
382	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	446	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
383		447
384	X_LOCK (reqlock);	448	X_LOCK (reqlock);
385	reqq_push (&req_queue, req);	449	reqq_push (&req_queue, req);
386	X_COND_SIGNAL (reqwait);	450	X_COND_SIGNAL (reqwait);
387	X_UNLOCK (reqlock);	451	X_UNLOCK (reqlock);
…		…
389	X_LOCK (wrklock);	453	X_LOCK (wrklock);
390	--started;	454	--started;
391	X_UNLOCK (wrklock);	455	X_UNLOCK (wrklock);
392	}	456	}
393		457
394	void eio_set_max_poll_time (double nseconds)
395	{
396	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
397	max_poll_time = nseconds;
398	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
399	}
400
401	void eio_set_max_poll_reqs (unsigned int maxreqs)
402	{
403	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
404	max_poll_reqs = maxreqs;
405	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
406	}
407
408	void eio_set_max_idle (unsigned int nthreads)
409	{
410	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
411	max_idle = nthreads <= 0 ? 1 : nthreads;
412	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
413	}
414
415	void eio_set_min_parallel (unsigned int nthreads)
416	{
417	if (wanted < nthreads)
418	wanted = nthreads;
419	}
420
421	void eio_set_max_parallel (unsigned int nthreads)
422	{
423	if (wanted > nthreads)
424	wanted = nthreads;
425
426	while (started > wanted)
427	end_thread ();
428	}
429
430	int eio_poll (void)	458	static int etp_poll (void)
431	{	459	{
432	int maxreqs = max_poll_reqs;	460	unsigned int maxreqs;
		461	unsigned int maxtime;
433	struct timeval tv_start, tv_now;	462	struct timeval tv_start, tv_now;
434	eio_req *req;
435		463
		464	X_LOCK (reslock);
		465	maxreqs = max_poll_reqs;
		466	maxtime = max_poll_time;
		467	X_UNLOCK (reslock);
		468
436	if (max_poll_time)	469	if (maxtime)
437	gettimeofday (&tv_start, 0);	470	gettimeofday (&tv_start, 0);
438		471
439	for (;;)	472	for (;;)
440	{	473	{
		474	ETP_REQ *req;
		475
441	maybe_start_thread ();	476	etp_maybe_start_thread ();
442		477
443	X_LOCK (reslock);	478	X_LOCK (reslock);
444	req = reqq_shift (&res_queue);	479	req = reqq_shift (&res_queue);
445		480
446	if (req)	481	if (req)
…		…
454	X_UNLOCK (reslock);	489	X_UNLOCK (reslock);
455		490
456	if (!req)	491	if (!req)
457	return 0;	492	return 0;
458		493
		494	X_LOCK (reqlock);
459	--nreqs;	495	--nreqs;
		496	X_UNLOCK (reqlock);
460		497
461	if (req->type == EIO_GROUP && req->size)	498	if (expect_false (req->type == EIO_GROUP && req->size))
462	{	499	{
463	req->int1 = 1; /* mark request as delayed */	500	req->int1 = 1; /* mark request as delayed */
464	continue;	501	continue;
465	}	502	}
466	else	503	else
467	{	504	{
468	int res = eio_finish (req);	505	int res = ETP_FINISH (req);
469	if (res)	506	if (expect_false (res))
470	return res;	507	return res;
471	}	508	}
472		509
473	if (maxreqs && !--maxreqs)	510	if (expect_false (maxreqs && !--maxreqs))
474	break;	511	break;
475		512
476	if (max_poll_time)	513	if (maxtime)
477	{	514	{
478	gettimeofday (&tv_now, 0);	515	gettimeofday (&tv_now, 0);
479		516
480	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	517	if (tvdiff (&tv_start, &tv_now) >= maxtime)
481	break;	518	break;
482	}	519	}
483	}	520	}
484		521
485	errno = EAGAIN;	522	errno = EAGAIN;
486	return -1;	523	return -1;
487	}	524	}
488		525
		526	static void etp_cancel (ETP_REQ *req)
		527	{
		528	X_LOCK (wrklock);
		529	req->flags |= EIO_FLAG_CANCELLED;
		530	X_UNLOCK (wrklock);
		531
		532	eio_grp_cancel (req);
		533	}
		534
		535	static void etp_submit (ETP_REQ *req)
		536	{
		537	req->pri -= ETP_PRI_MIN;
		538
		539	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		540	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		541
		542	if (expect_false (req->type == EIO_GROUP))
		543	{
		544	/* I hope this is worth it :/ */
		545	X_LOCK (reqlock);
		546	++nreqs;
		547	X_UNLOCK (reqlock);
		548
		549	X_LOCK (reslock);
		550
		551	++npending;
		552
		553	if (!reqq_push (&res_queue, req) && want_poll_cb)
		554	want_poll_cb ();
		555
		556	X_UNLOCK (reslock);
		557	}
		558	else
		559	{
		560	X_LOCK (reqlock);
		561	++nreqs;
		562	++nready;
		563	reqq_push (&req_queue, req);
		564	X_COND_SIGNAL (reqwait);
		565	X_UNLOCK (reqlock);
		566
		567	etp_maybe_start_thread ();
		568	}
		569	}
		570
		571	static void etp_set_max_poll_time (double nseconds)
		572	{
		573	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		574	max_poll_time = nseconds;
		575	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		576	}
		577
		578	static void etp_set_max_poll_reqs (unsigned int maxreqs)
		579	{
		580	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		581	max_poll_reqs = maxreqs;
		582	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		583	}
		584
		585	static void etp_set_max_idle (unsigned int nthreads)
		586	{
		587	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		588	max_idle = nthreads <= 0 ? 1 : nthreads;
		589	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		590	}
		591
		592	static void etp_set_min_parallel (unsigned int nthreads)
		593	{
		594	if (wanted < nthreads)
		595	wanted = nthreads;
		596	}
		597
		598	static void etp_set_max_parallel (unsigned int nthreads)
		599	{
		600	if (wanted > nthreads)
		601	wanted = nthreads;
		602
		603	while (started > wanted)
		604	etp_end_thread ();
		605	}
		606
		607	/*****************************************************************************/
		608
		609	static void grp_try_feed (eio_req *grp)
		610	{
		611	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		612	{
		613	grp->flags &= ~EIO_FLAG_GROUPADD;
		614
		615	EIO_FEED (grp);
		616
		617	/* stop if no progress has been made */
		618	if (!(grp->flags & EIO_FLAG_GROUPADD))
		619	{
		620	grp->feed = 0;
		621	break;
		622	}
		623	}
		624	}
		625
		626	static int grp_dec (eio_req *grp)
		627	{
		628	--grp->size;
		629
		630	/* call feeder, if applicable */
		631	grp_try_feed (grp);
		632
		633	/* finish, if done */
		634	if (!grp->size && grp->int1)
		635	return eio_finish (grp);
		636	else
		637	return 0;
		638	}
		639
		640	void eio_destroy (eio_req *req)
		641	{
		642	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		643	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		644
		645	EIO_DESTROY (req);
		646	}
		647
		648	static int eio_finish (eio_req *req)
		649	{
		650	int res = EIO_FINISH (req);
		651
		652	if (req->grp)
		653	{
		654	int res2;
		655	eio_req *grp = req->grp;
		656
		657	/* unlink request */
		658	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		659	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		660
		661	if (grp->grp_first == req)
		662	grp->grp_first = req->grp_next;
		663
		664	res2 = grp_dec (grp);
		665
		666	if (!res && res2)
		667	res = res2;
		668	}
		669
		670	eio_destroy (req);
		671
		672	return res;
		673	}
		674
		675	void eio_grp_cancel (eio_req *grp)
		676	{
		677	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		678	eio_cancel (grp);
		679	}
		680
		681	void eio_cancel (eio_req *req)
		682	{
		683	etp_cancel (req);
		684	}
		685
		686	void eio_submit (eio_req *req)
		687	{
		688	etp_submit (req);
		689	}
		690
		691	unsigned int eio_nreqs (void)
		692	{
		693	return etp_nreqs ();
		694	}
		695
		696	unsigned int eio_nready (void)
		697	{
		698	return etp_nready ();
		699	}
		700
		701	unsigned int eio_npending (void)
		702	{
		703	return etp_npending ();
		704	}
		705
		706	unsigned int eio_nthreads (void)
		707	{
		708	return etp_nthreads ();
		709	}
		710
		711	void eio_set_max_poll_time (double nseconds)
		712	{
		713	etp_set_max_poll_time (nseconds);
		714	}
		715
		716	void eio_set_max_poll_reqs (unsigned int maxreqs)
		717	{
		718	etp_set_max_poll_reqs (maxreqs);
		719	}
		720
		721	void eio_set_max_idle (unsigned int nthreads)
		722	{
		723	etp_set_max_idle (nthreads);
		724	}
		725
		726	void eio_set_min_parallel (unsigned int nthreads)
		727	{
		728	etp_set_min_parallel (nthreads);
		729	}
		730
		731	void eio_set_max_parallel (unsigned int nthreads)
		732	{
		733	etp_set_max_parallel (nthreads);
		734	}
		735
		736	int eio_poll (void)
		737	{
		738	return etp_poll ();
		739	}
		740
489	/*****************************************************************************/	741	/*****************************************************************************/
490	/* work around various missing functions */	742	/* work around various missing functions */
491		743
492	#if !HAVE_PREADWRITE	744	#if !HAVE_PREADWRITE
		745	# undef pread
		746	# undef pwrite
493	# define pread eio__pread	747	# define pread eio__pread
494	# define pwrite eio__pwrite	748	# define pwrite eio__pwrite
495
496	/*
497	* make our pread/pwrite safe against themselves, but not against
498	* normal read/write by using a mutex. slows down execution a lot,
499	* but that's your problem, not mine.
500	*/
501	static mutex_t preadwritelock = X_MUTEX_INIT;
502		749
503	static ssize_t	750	static ssize_t
504	eio__pread (int fd, void *buf, size_t count, off_t offset)	751	eio__pread (int fd, void *buf, size_t count, off_t offset)
505	{	752	{
506	ssize_t res;	753	ssize_t res;
…		…
524		771
525	X_LOCK (preadwritelock);	772	X_LOCK (preadwritelock);
526	ooffset = lseek (fd, 0, SEEK_CUR);	773	ooffset = lseek (fd, 0, SEEK_CUR);
527	lseek (fd, offset, SEEK_SET);	774	lseek (fd, offset, SEEK_SET);
528	res = write (fd, buf, count);	775	res = write (fd, buf, count);
529	lseek (fd, offset, SEEK_SET);	776	lseek (fd, ooffset, SEEK_SET);
530	X_UNLOCK (preadwritelock);	777	X_UNLOCK (preadwritelock);
531		778
532	return res;	779	return res;
533	}	780	}
534	#endif	781	#endif
535		782
536	#ifndef HAVE_FUTIMES	783	#ifndef HAVE_FUTIMES
537		784
		785	# undef utimes
		786	# undef futimes
538	# define utimes(path,times) eio__utimes (path, times)	787	# define utimes(path,times) eio__utimes (path, times)
539	# define futimes(fd,times) eio__futimes (fd, times)	788	# define futimes(fd,times) eio__futimes (fd, times)
540		789
541	static int	790	static int
542	eio__utimes (const char *filename, const struct timeval times[2])	791	eio__utimes (const char *filename, const struct timeval times[2])
…		…
561	}	810	}
562		811
563	#endif	812	#endif
564		813
565	#if !HAVE_FDATASYNC	814	#if !HAVE_FDATASYNC
		815	# undef fdatasync
566	# define fdatasync fsync	816	# define fdatasync(fd) fsync (fd)
567	#endif	817	#endif
		818
		819	/* sync_file_range always needs emulation */
		820	int
		821	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		822	{
		823	#if HAVE_SYNC_FILE_RANGE
		824	int res;
		825
		826	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		827	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		828	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		829	{
		830	flags = 0
		831	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		832	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		833	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		834	}
		835
		836	res = sync_file_range (fd, offset, nbytes, flags);
		837
		838	if (!res || errno != ENOSYS)
		839	return res;
		840	#endif
		841
		842	/* even though we could play tricks with the flags, it's better to always
		843	* call fdatasync, as thta matches the expectation of it's users best */
		844	return fdatasync (fd);
		845	}
568		846
569	#if !HAVE_READAHEAD	847	#if !HAVE_READAHEAD
		848	# undef readahead
570	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	849	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
571		850
572	static ssize_t	851	static ssize_t
573	eio__readahead (int fd, off_t offset, size_t count, worker *self)	852	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
574	{	853	{
575	size_t todo = count;	854	size_t todo = count;
576	dBUF;	855	dBUF;
577		856
578	while (todo > 0)	857	while (todo > 0)
…		…
588	return count;	867	return count;
589	}	868	}
590		869
591	#endif	870	#endif
592		871
593	#if !HAVE_READDIR_R
594	# define readdir_r eio__readdir_r
595
596	static mutex_t readdirlock = X_MUTEX_INIT;
597
598	static int
599	eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
600	{
601	EIO_STRUCT_DIRENT *e;
602	int errorno;
603
604	X_LOCK (readdirlock);
605
606	e = readdir (dirp);
607	errorno = errno;
608
609	if (e)
610	{
611	*res = ent;
612	strcpy (ent->d_name, e->d_name);
613	}
614	else
615	*res = 0;
616
617	X_UNLOCK (readdirlock);
618
619	errno = errorno;
620	return e ? 0 : -1;
621	}
622	#endif
623
624	/* sendfile always needs emulation */	872	/* sendfile always needs emulation */
625	static ssize_t	873	static ssize_t
626	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self)	874	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
627	{	875	{
628	ssize_t res;	876	ssize_t res;
629		877
630	if (!count)	878	if (!count)
631	return 0;	879	return 0;
…		…
715	return res;	963	return res;
716	}	964	}
717		965
718	/* read a full directory */	966	/* read a full directory */
719	static void	967	static void
720	eio__scandir (eio_req *req, worker *self)	968	eio__scandir (eio_req *req, etp_worker *self)
721	{	969	{
722	DIR *dirp;	970	DIR *dirp;
723	union
724	{
725	EIO_STRUCT_DIRENT d;
726	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
727	} *u;
728	EIO_STRUCT_DIRENT *entp;	971	EIO_STRUCT_DIRENT *entp;
729	char *name, *names;	972	char *name, *names;
730	int memlen = 4096;	973	int memlen = 4096;
731	int memofs = 0;	974	int memofs = 0;
732	int res = 0;	975	int res = 0;
733		976
734	X_LOCK (wrklock);	977	X_LOCK (wrklock);
		978	/* the corresponding closedir is in ETP_WORKER_CLEAR */
735	self->dirp = dirp = opendir (req->ptr1);	979	self->dirp = dirp = opendir (req->ptr1);
736	self->dbuf = u = malloc (sizeof (*u));
737	req->flags |= EIO_FLAG_PTR2_FREE;	980	req->flags |= EIO_FLAG_PTR2_FREE;
738	req->ptr2 = names = malloc (memlen);	981	req->ptr2 = names = malloc (memlen);
739	X_UNLOCK (wrklock);	982	X_UNLOCK (wrklock);
740		983
741	if (dirp && u && names)	984	if (dirp && names)
742	for (;;)	985	for (;;)
743	{	986	{
744	errno = 0;	987	errno = 0;
745	readdir_r (dirp, &u->d, &entp);	988	entp = readdir (dirp);
746		989
747	if (!entp)	990	if (!entp)
748	break;	991	break;
749		992
750	name = entp->d_name;	993	name = entp->d_name;
…		…
773		1016
774	if (errno)	1017	if (errno)
775	res = -1;	1018	res = -1;
776		1019
777	req->result = res;	1020	req->result = res;
		1021	}
		1022
		1023	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1024	# undef msync
		1025	# define msync(a,b,c) ((errno = ENOSYS), -1)
		1026	#endif
		1027
		1028	int
		1029	eio__mtouch (void *mem, size_t len, int flags)
		1030	{
		1031	intptr_t addr = (intptr_t)mem;
		1032	intptr_t end = addr + len;
		1033	#ifdef PAGESIZE
		1034	const intptr_t page = PAGESIZE;
		1035	#else
		1036	static intptr_t page;
		1037
		1038	if (!page)
		1039	page = sysconf (_SC_PAGESIZE);
		1040	#endif
		1041
		1042	addr &= ~(page - 1); /* assume page size is always a power of two */
		1043
		1044	if (addr < end)
		1045	if (flags) /* modify */
		1046	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
		1047	else
		1048	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
		1049
		1050	return 0;
778	}	1051	}
779		1052
780	/*****************************************************************************/	1053	/*****************************************************************************/
781		1054
782	#define ALLOC(len) \	1055	#define ALLOC(len) \
…		…
792	req->result = -1; \	1065	req->result = -1; \
793	break; \	1066	break; \
794	} \	1067	} \
795	}	1068	}
796		1069
797	X_THREAD_PROC (eio_proc)	1070	X_THREAD_PROC (etp_proc)
798	{	1071	{
799	eio_req *req;	1072	ETP_REQ *req;
800	struct timespec ts;	1073	struct timespec ts;
801	worker *self = (worker *)thr_arg;	1074	etp_worker *self = (etp_worker *)thr_arg;
802		1075
803	/* try to distribute timeouts somewhat randomly */	1076	/* try to distribute timeouts somewhat randomly */
804	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1077	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
805		1078
806	for (;;)	1079	for (;;)
807	{	1080	{
808	ts.tv_sec = time (0) + IDLE_TIMEOUT;
809
810	X_LOCK (reqlock);	1081	X_LOCK (reqlock);
811		1082
812	for (;;)	1083	for (;;)
813	{	1084	{
814	self->req = req = reqq_shift (&req_queue);	1085	self->req = req = reqq_shift (&req_queue);
…		…
816	if (req)	1087	if (req)
817	break;	1088	break;
818		1089
819	++idle;	1090	++idle;
820		1091
		1092	ts.tv_sec = time (0) + IDLE_TIMEOUT;
821	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1093	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
822	{	1094	{
823	if (idle > max_idle)	1095	if (idle > max_idle)
824	{	1096	{
825	--idle;	1097	--idle;
…		…
830	goto quit;	1102	goto quit;
831	}	1103	}
832		1104
833	/* we are allowed to idle, so do so without any timeout */	1105	/* we are allowed to idle, so do so without any timeout */
834	X_COND_WAIT (reqwait, reqlock);	1106	X_COND_WAIT (reqwait, reqlock);
835	ts.tv_sec = time (0) + IDLE_TIMEOUT;
836	}	1107	}
837		1108
838	--idle;	1109	--idle;
839	}	1110	}
840		1111
841	--nready;	1112	--nready;
842		1113
843	X_UNLOCK (reqlock);	1114	X_UNLOCK (reqlock);
844		1115
845	errno = 0; /* strictly unnecessary */	1116	if (req->type < 0)
		1117	goto quit;
846		1118
847	if (!EIO_CANCELLED (req))	1119	if (!EIO_CANCELLED (req))
848	switch (req->type)	1120	ETP_EXECUTE (self, req);
849	{
850	case EIO_READ: ALLOC (req->size);
851	req->result = req->offs >= 0
852	? pread (req->int1, req->ptr2, req->size, req->offs)
853	: read (req->int1, req->ptr2, req->size); break;
854	case EIO_WRITE: req->result = req->offs >= 0
855	? pwrite (req->int1, req->ptr2, req->size, req->offs)
856	: write (req->int1, req->ptr2, req->size); break;
857
858	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
859	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
860
861	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
862	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
863	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
864	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
865	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
866	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
867
868	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
869	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
870	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
871	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
872	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
873	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
874
875	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
876	case EIO_CLOSE: req->result = close (req->int1); break;
877	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
878	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
879	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
880	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
881	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
882	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
883	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
884	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
885
886	case EIO_READLINK: ALLOC (NAME_MAX);
887	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
888
889	case EIO_SYNC: req->result = 0; sync (); break;
890	case EIO_FSYNC: req->result = fsync (req->int1); break;
891	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
892
893	case EIO_READDIR: eio__scandir (req, self); break;
894
895	case EIO_BUSY:
896	#ifdef _WIN32
897	Sleep (req->nv1 * 1000.);
898	#else
899	{
900	struct timeval tv;
901
902	tv.tv_sec = req->nv1;
903	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
904
905	req->result = select (0, 0, 0, 0, &tv);
906	}
907	#endif
908	break;
909
910	case EIO_UTIME:
911	case EIO_FUTIME:
912	{
913	struct timeval tv[2];
914	struct timeval *times;
915
916	if (req->nv1 != -1. || req->nv2 != -1.)
917	{
918	tv[0].tv_sec = req->nv1;
919	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
920	tv[1].tv_sec = req->nv2;
921	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
922
923	times = tv;
924	}
925	else
926	times = 0;
927
928
929	req->result = req->type == EIO_FUTIME
930	? futimes (req->int1, times)
931	: utimes (req->ptr1, times);
932	}
933
934	case EIO_GROUP:
935	case EIO_NOP:
936	req->result = 0;
937	break;
938
939	case EIO_QUIT:
940	goto quit;
941
942	default:
943	req->result = -1;
944	break;
945	}
946
947	req->errorno = errno;
948		1121
949	X_LOCK (reslock);	1122	X_LOCK (reslock);
950		1123
951	++npending;	1124	++npending;
952		1125
953	if (!reqq_push (&res_queue, req) && want_poll_cb)	1126	if (!reqq_push (&res_queue, req) && want_poll_cb)
954	want_poll_cb ();	1127	want_poll_cb ();
955		1128
956	self->req = 0;	1129	self->req = 0;
957	worker_clear (self);	1130	etp_worker_clear (self);
958		1131
959	X_UNLOCK (reslock);	1132	X_UNLOCK (reslock);
960	}	1133	}
961		1134
962	quit:	1135	quit:
963	X_LOCK (wrklock);	1136	X_LOCK (wrklock);
964	worker_free (self);	1137	etp_worker_free (self);
965	X_UNLOCK (wrklock);	1138	X_UNLOCK (wrklock);
966		1139
967	return 0;	1140	return 0;
968	}	1141	}
969		1142
970	/*****************************************************************************/	1143	/*****************************************************************************/
971		1144
972	static void eio_atfork_prepare (void)
973	{
974	X_LOCK (wrklock);
975	X_LOCK (reqlock);
976	X_LOCK (reslock);
977	#if !HAVE_PREADWRITE
978	X_LOCK (preadwritelock);
979	#endif
980	#if !HAVE_READDIR_R
981	X_LOCK (readdirlock);
982	#endif
983	}
984
985	static void eio_atfork_parent (void)
986	{
987	#if !HAVE_READDIR_R
988	X_UNLOCK (readdirlock);
989	#endif
990	#if !HAVE_PREADWRITE
991	X_UNLOCK (preadwritelock);
992	#endif
993	X_UNLOCK (reslock);
994	X_UNLOCK (reqlock);
995	X_UNLOCK (wrklock);
996	}
997
998	static void eio_atfork_child (void)
999	{
1000	eio_req *prv;
1001
1002	while (prv = reqq_shift (&req_queue))
1003	eio_destroy (prv);
1004
1005	while (prv = reqq_shift (&res_queue))
1006	eio_destroy (prv);
1007
1008	while (wrk_first.next != &wrk_first)
1009	{
1010	worker *wrk = wrk_first.next;
1011
1012	if (wrk->req)
1013	eio_destroy (wrk->req);
1014
1015	worker_clear (wrk);
1016	worker_free (wrk);
1017	}
1018
1019	started = 0;
1020	idle = 0;
1021	nreqs = 0;
1022	nready = 0;
1023	npending = 0;
1024
1025	eio_atfork_parent ();
1026	}
1027
1028	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1145	int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1029	{	1146	{
1030	want_poll_cb = want_poll;	1147	return etp_init (want_poll, done_poll);
1031	done_poll_cb = done_poll;
1032
1033	#ifdef _WIN32
1034	X_MUTEX_CHECK (wrklock);
1035	X_MUTEX_CHECK (reslock);
1036	X_MUTEX_CHECK (reqlock);
1037	X_MUTEX_CHECK (reqwait);
1038	X_MUTEX_CHECK (preadwritelock);
1039	X_MUTEX_CHECK (readdirlock);
1040
1041	X_COND_CHECK (reqwait);
1042	#endif
1043
1044	X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1045	}	1148	}
1046		1149
1047	static void eio_api_destroy (eio_req *req)	1150	static void eio_api_destroy (eio_req *req)
1048	{	1151	{
1049	free (req);	1152	free (req);
…		…
1071	{ \	1174	{ \
1072	eio_api_destroy (req); \	1175	eio_api_destroy (req); \
1073	return 0; \	1176	return 0; \
1074	}	1177	}
1075		1178
		1179	static void eio_execute (etp_worker *self, eio_req *req)
		1180	{
		1181	errno = 0;
		1182
		1183	switch (req->type)
		1184	{
		1185	case EIO_READ: ALLOC (req->size);
		1186	req->result = req->offs >= 0
		1187	? pread (req->int1, req->ptr2, req->size, req->offs)
		1188	: read (req->int1, req->ptr2, req->size); break;
		1189	case EIO_WRITE: req->result = req->offs >= 0
		1190	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1191	: write (req->int1, req->ptr2, req->size); break;
		1192
		1193	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		1194	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
		1195
		1196	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1197	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1198	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1199	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1200	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1201	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1202
		1203	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1204	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1205	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1206	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1207	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1208	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1209
		1210	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1211	case EIO_CLOSE: req->result = close (req->int1); break;
		1212	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1213	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1214	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1215	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1216	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1217	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1218	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1219	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
		1220
		1221	case EIO_READLINK: ALLOC (NAME_MAX);
		1222	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
		1223
		1224	case EIO_SYNC: req->result = 0; sync (); break;
		1225	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1226	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1227	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
		1228	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1229	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1230
		1231	case EIO_READDIR: eio__scandir (req, self); break;
		1232
		1233	case EIO_BUSY:
		1234	#ifdef _WIN32
		1235	Sleep (req->nv1 * 1000.);
		1236	#else
		1237	{
		1238	struct timeval tv;
		1239
		1240	tv.tv_sec = req->nv1;
		1241	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
		1242
		1243	req->result = select (0, 0, 0, 0, &tv);
		1244	}
		1245	#endif
		1246	break;
		1247
		1248	case EIO_UTIME:
		1249	case EIO_FUTIME:
		1250	{
		1251	struct timeval tv[2];
		1252	struct timeval *times;
		1253
		1254	if (req->nv1 != -1. || req->nv2 != -1.)
		1255	{
		1256	tv[0].tv_sec = req->nv1;
		1257	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		1258	tv[1].tv_sec = req->nv2;
		1259	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		1260
		1261	times = tv;
		1262	}
		1263	else
		1264	times = 0;
		1265
		1266
		1267	req->result = req->type == EIO_FUTIME
		1268	? futimes (req->int1, times)
		1269	: utimes (req->ptr1, times);
		1270	}
		1271	break;
		1272
		1273	case EIO_GROUP:
		1274	abort (); /* handled in eio_request */
		1275
		1276	case EIO_NOP:
		1277	req->result = 0;
		1278	break;
		1279
		1280	case EIO_CUSTOM:
		1281	((void (*)(eio_req *))req->feed) (req);
		1282	break;
		1283
		1284	default:
		1285	req->result = -1;
		1286	break;
		1287	}
		1288
		1289	req->errorno = errno;
		1290	}
		1291
		1292	#ifndef EIO_NO_WRAPPERS
		1293
1076	eio_req *eio_nop (int pri, eio_cb cb, void *data)	1294	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1077	{	1295	{
1078	REQ (EIO_NOP); SEND;	1296	REQ (EIO_NOP); SEND;
1079	}	1297	}
1080		1298
…		…
1089	}	1307	}
1090		1308
1091	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	1309	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1092	{	1310	{
1093	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1311	REQ (EIO_FSYNC); req->int1 = fd; SEND;
		1312	}
		1313
		1314	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1315	{
		1316	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1317	}
		1318
		1319	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1320	{
		1321	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1322	}
		1323
		1324	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		1325	{
		1326	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1094	}	1327	}
1095		1328
1096	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	1329	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1097	{	1330	{
1098	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1331	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
…		…
1219	return eio__1path (EIO_READDIR, path, pri, cb, data);	1452	return eio__1path (EIO_READDIR, path, pri, cb, data);
1220	}	1453	}
1221		1454
1222	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	1455	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1223	{	1456	{
1224	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	1457	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1225	}	1458	}
1226		1459
1227	static eio_req *	1460	static eio_req *
1228	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1461	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1229	{	1462	{
…		…
1253	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1486	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1254	{	1487	{
1255	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	1488	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1256	}	1489	}
1257		1490
		1491	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
		1492	{
		1493	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
		1494	}
		1495
		1496	#endif
		1497
1258	eio_req *eio_grp (eio_cb cb, void *data)	1498	eio_req *eio_grp (eio_cb cb, void *data)
1259	{	1499	{
1260	const int pri = EIO_PRI_MAX;	1500	const int pri = EIO_PRI_MAX;
1261		1501
1262	REQ (EIO_GROUP); SEND;	1502	REQ (EIO_GROUP); SEND;
…		…
1286		1526
1287	void eio_grp_add (eio_req *grp, eio_req *req)	1527	void eio_grp_add (eio_req *grp, eio_req *req)
1288	{	1528	{
1289	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	1529	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1290		1530
		1531	grp->flags |= EIO_FLAG_GROUPADD;
		1532
1291	++grp->size;	1533	++grp->size;
1292	req->grp = grp;	1534	req->grp = grp;
1293		1535
1294	req->grp_prev = 0;	1536	req->grp_prev = 0;
1295	req->grp_next = grp->grp_first;	1537	req->grp_next = grp->grp_first;
…		…
1303	/*****************************************************************************/	1545	/*****************************************************************************/
1304	/* misc garbage */	1546	/* misc garbage */
1305		1547
1306	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	1548	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1307	{	1549	{
1308	worker wrk;	1550	etp_worker wrk;
1309		1551
1310	wrk.dbuf = 0;	1552	wrk.dbuf = 0;
1311		1553
1312	eio__sendfile (ofd, ifd, offset, count, &wrk);	1554	eio__sendfile (ofd, ifd, offset, count, &wrk);
1313		1555

Diff Legend

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