ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.13 by root, Tue May 13 19:34:11 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 (etplock); \	118	ETP_WORKER_LOCK (self); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	119	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (etplock); \	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
84	/*****************************************************************************/	127	/*****************************************************************************/
85		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
86	/* calculcate time difference in ~1/EIO_TICKS of a second */	172	/* calculcate time difference in ~1/EIO_TICKS of a second */
87	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	173	static int tvdiff (struct timeval *tv1, struct timeval *tv2)
88	{	174	{
89	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	175	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
90	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	176	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
91	}	177	}
92		178
93	static unsigned int started, idle, wanted = 4;	179	static unsigned int started, idle, wanted = 4;
94		180
95	typedef struct etp_pool
96	{
97	void (*want_poll_cb) (void);	181	static void (*want_poll_cb) (void);
98	void (*done_poll_cb) (void);	182	static void (*done_poll_cb) (void);
99		183
100	unsigned int max_poll_time;	184	static unsigned int max_poll_time; /* reslock */
101	unsigned int max_poll_reqs;	185	static unsigned int max_poll_reqs; /* reslock */
102	} etp_pool;
103		186
104	static volatile unsigned int nreqs, nready, npending;	187	static volatile unsigned int nreqs; /* reqlock */
		188	static volatile unsigned int nready; /* reqlock */
		189	static volatile unsigned int npending; /* reqlock */
105	static volatile unsigned int max_idle = 4;	190	static volatile unsigned int max_idle = 4;
106		191
107	static mutex_t etplock = X_MUTEX_INIT;	192	static mutex_t wrklock = X_MUTEX_INIT;
108	static mutex_t reslock = X_MUTEX_INIT;	193	static mutex_t reslock = X_MUTEX_INIT;
109	static mutex_t reqlock = X_MUTEX_INIT;	194	static mutex_t reqlock = X_MUTEX_INIT;
110	static cond_t reqwait = X_COND_INIT;	195	static cond_t reqwait = X_COND_INIT;
111		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
112	typedef struct worker	206	typedef struct etp_worker
113	{	207	{
114	/* locked by etplock */	208	/* locked by wrklock */
115	struct worker *prev, *next;	209	struct etp_worker *prev, *next;
116		210
117	thread_t tid;	211	thread_t tid;
118		212
119	/* locked by reslock, reqlock or etplock */	213	/* locked by reslock, reqlock or wrklock */
120	eio_req *req; /* currently processed request */	214	ETP_REQ *req; /* currently processed request */
121	void *dbuf;	215
122	DIR *dirp;	216	ETP_WORKER_COMMON
123	} worker;	217	} etp_worker;
124		218
125	static worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	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)
126		223
127	/* worker threads management */	224	/* worker threads management */
128		225
129	static void worker_clear (worker *wrk)	226	static void etp_worker_clear (etp_worker *wrk)
130	{	227	{
131	if (wrk->dirp)	228	ETP_WORKER_CLEAR (wrk);
132	{
133	closedir (wrk->dirp);
134	wrk->dirp = 0;
135	}
136
137	if (wrk->dbuf)
138	{
139	free (wrk->dbuf);
140	wrk->dbuf = 0;
141	}
142	}	229	}
143		230
144	static void worker_free (worker *wrk)	231	static void etp_worker_free (etp_worker *wrk)
145	{	232	{
146	wrk->next->prev = wrk->prev;	233	wrk->next->prev = wrk->prev;
147	wrk->prev->next = wrk->next;	234	wrk->prev->next = wrk->next;
148		235
149	free (wrk);	236	free (wrk);
150	}	237	}
151		238
152	unsigned int eio_nreqs (void)	239	static unsigned int etp_nreqs (void)
153	{	240	{
		241	int retval;
		242	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		243	retval = nreqs;
		244	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154	return nreqs;	245	return retval;
155	}	246	}
156		247
157	unsigned int eio_nready (void)	248	static unsigned int etp_nready (void)
158	{	249	{
159	unsigned int retval;	250	unsigned int retval;
160		251
161	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	252	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162	retval = nready;	253	retval = nready;
163	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	254	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164		255
165	return retval;	256	return retval;
166	}	257	}
167		258
168	unsigned int eio_npending (void)	259	static unsigned int etp_npending (void)
169	{	260	{
170	unsigned int retval;	261	unsigned int retval;
171		262
172	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	263	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173	retval = npending;	264	retval = npending;
174	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	265	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175		266
176	return retval;	267	return retval;
177	}	268	}
178		269
179	unsigned int eio_nthreads (void)	270	static unsigned int etp_nthreads (void)
180	{	271	{
181	unsigned int retval;	272	unsigned int retval;
182		273
183	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	274	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184	retval = started;	275	retval = started;
…		…
191	* a somewhat faster data structure might be nice, but	282	* a somewhat faster data structure might be nice, but
192	* with 8 priorities this actually needs <20 insns	283	* with 8 priorities this actually needs <20 insns
193	* per shift, the most expensive operation.	284	* per shift, the most expensive operation.
194	*/	285	*/
195	typedef struct {	286	typedef struct {
196	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 */
197	int size;	288	int size;
198	} reqq;	289	} etp_reqq;
199		290
200	static reqq req_queue;	291	static etp_reqq req_queue;
201	static reqq res_queue;	292	static etp_reqq res_queue;
202		293
203	static int reqq_push (reqq *q, eio_req *req)	294	static int reqq_push (etp_reqq *q, ETP_REQ *req)
204	{	295	{
205	int pri = req->pri;	296	int pri = req->pri;
206	req->next = 0;	297	req->next = 0;
207		298
208	if (q->qe[pri])	299	if (q->qe[pri])
…		…
214	q->qe[pri] = q->qs[pri] = req;	305	q->qe[pri] = q->qs[pri] = req;
215		306
216	return q->size++;	307	return q->size++;
217	}	308	}
218		309
219	static eio_req *reqq_shift (reqq *q)	310	static ETP_REQ *reqq_shift (etp_reqq *q)
220	{	311	{
221	int pri;	312	int pri;
222		313
223	if (!q->size)	314	if (!q->size)
224	return 0;	315	return 0;
225		316
226	--q->size;	317	--q->size;
227		318
228	for (pri = EIO_NUM_PRI; pri--; )	319	for (pri = ETP_NUM_PRI; pri--; )
229	{	320	{
230	eio_req *req = q->qs[pri];	321	eio_req *req = q->qs[pri];
231		322
232	if (req)	323	if (req)
233	{	324	{
…		…
241	abort ();	332	abort ();
242	}	333	}
243		334
244	static void etp_atfork_prepare (void)	335	static void etp_atfork_prepare (void)
245	{	336	{
246	X_LOCK (etplock);	337	X_LOCK (wrklock);
247	X_LOCK (reqlock);	338	X_LOCK (reqlock);
248	X_LOCK (reslock);	339	X_LOCK (reslock);
249	#if !HAVE_PREADWRITE	340	#if !HAVE_PREADWRITE
250	X_LOCK (preadwritelock);	341	X_LOCK (preadwritelock);
251	#endif	342	#endif
252	#if !HAVE_READDIR_R
253	X_LOCK (readdirlock);
254	#endif
255	}	343	}
256		344
257	static void etp_atfork_parent (void)	345	static void etp_atfork_parent (void)
258	{	346	{
259	#if !HAVE_READDIR_R
260	X_UNLOCK (readdirlock);
261	#endif
262	#if !HAVE_PREADWRITE	347	#if !HAVE_PREADWRITE
263	X_UNLOCK (preadwritelock);	348	X_UNLOCK (preadwritelock);
264	#endif	349	#endif
265	X_UNLOCK (reslock);	350	X_UNLOCK (reslock);
266	X_UNLOCK (reqlock);	351	X_UNLOCK (reqlock);
267	X_UNLOCK (etplock);	352	X_UNLOCK (wrklock);
268	}	353	}
269		354
270	static void etp_atfork_child (void)	355	static void etp_atfork_child (void)
271	{	356	{
272	eio_req *prv;	357	ETP_REQ *prv;
273		358
274	while (prv = reqq_shift (&req_queue))	359	while ((prv = reqq_shift (&req_queue)))
275	eio_destroy (prv);	360	ETP_DESTROY (prv);
276		361
277	while (prv = reqq_shift (&res_queue))	362	while ((prv = reqq_shift (&res_queue)))
278	eio_destroy (prv);	363	ETP_DESTROY (prv);
279		364
280	while (wrk_first.next != &wrk_first)	365	while (wrk_first.next != &wrk_first)
281	{	366	{
282	worker *wrk = wrk_first.next;	367	etp_worker *wrk = wrk_first.next;
283		368
284	if (wrk->req)	369	if (wrk->req)
285	eio_destroy (wrk->req);	370	ETP_DESTROY (wrk->req);
286		371
287	worker_clear (wrk);	372	etp_worker_clear (wrk);
288	worker_free (wrk);	373	etp_worker_free (wrk);
289	}	374	}
290		375
291	started = 0;	376	started = 0;
292	idle = 0;	377	idle = 0;
293	nreqs = 0;	378	nreqs = 0;
…		…
302	{	387	{
303	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);	388	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
304	}	389	}
305		390
306	static int	391	static int
307	etp_init (etp_pool *etp, void (*want_poll)(void), void (*done_poll)(void))	392	etp_init (void (*want_poll)(void), void (*done_poll)(void))
308	{	393	{
309	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	394	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
310		395
311	pthread_once (&doinit, etp_once_init);	396	pthread_once (&doinit, etp_once_init);
312		397
313	memset (etp, 0, sizeof *etp);
314
315	etp->want_poll_cb = want_poll;	398	want_poll_cb = want_poll;
316	etp->done_poll_cb = done_poll;	399	done_poll_cb = done_poll;
317	}
318		400
319	static etp_pool etp;
320
321	/*****************************************************************************/
322
323	static void grp_try_feed (eio_req *grp)
324	{
325	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
326	{
327	int old_len = grp->size;
328
329	EIO_FEED (grp);
330
331	/* stop if no progress has been made */
332	if (old_len == grp->size)
333	{
334	grp->feed = 0;
335	break;
336	}
337	}
338	}
339
340	static int eio_finish (eio_req *req);
341
342	static int grp_dec (eio_req *grp)
343	{
344	--grp->size;
345
346	/* call feeder, if applicable */
347	grp_try_feed (grp);
348
349	/* finish, if done */
350	if (!grp->size && grp->int1)
351	return eio_finish (grp);
352	else
353	return 0;	401	return 0;
354	}	402	}
355		403
356	void eio_destroy (eio_req *req)
357	{
358	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
359	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
360
361	EIO_DESTROY (req);
362	}
363
364	static int eio_finish (eio_req *req)
365	{
366	int res = EIO_FINISH (req);
367
368	if (req->grp)
369	{
370	int res2;
371	eio_req *grp = req->grp;
372
373	/* unlink request */
374	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
375	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
376
377	if (grp->grp_first == req)
378	grp->grp_first = req->grp_next;
379
380	res2 = grp_dec (grp);
381
382	if (!res && res2)
383	res = res2;
384	}
385
386	eio_destroy (req);
387
388	return res;
389	}
390
391	void eio_grp_cancel (eio_req *grp)
392	{
393	for (grp = grp->grp_first; grp; grp = grp->grp_next)
394	eio_cancel (grp);
395	}
396
397	void eio_cancel (eio_req *req)
398	{
399	X_LOCK (etplock);
400	req->flags |= EIO_FLAG_CANCELLED;
401	X_UNLOCK (etplock);
402
403	eio_grp_cancel (req);
404	}
405
406	X_THREAD_PROC (eio_proc);	404	X_THREAD_PROC (etp_proc);
407		405
408	static void start_thread (void)	406	static void etp_start_thread (void)
409	{	407	{
410	worker *wrk = calloc (1, sizeof (worker));	408	etp_worker *wrk = calloc (1, sizeof (etp_worker));
411		409
412	/*TODO*/	410	/*TODO*/
413	assert (("unable to allocate worker thread data", wrk));	411	assert (("unable to allocate worker thread data", wrk));
414		412
415	X_LOCK (etplock);	413	X_LOCK (wrklock);
416		414
417	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	415	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
418	{	416	{
419	wrk->prev = &wrk_first;	417	wrk->prev = &wrk_first;
420	wrk->next = wrk_first.next;	418	wrk->next = wrk_first.next;
421	wrk_first.next->prev = wrk;	419	wrk_first.next->prev = wrk;
422	wrk_first.next = wrk;	420	wrk_first.next = wrk;
423	++started;	421	++started;
424	}	422	}
425	else	423	else
426	free (wrk);	424	free (wrk);
427		425
428	X_UNLOCK (etplock);	426	X_UNLOCK (wrklock);
429	}	427	}
430		428
431	static void maybe_start_thread (void)	429	static void etp_maybe_start_thread (void)
432	{	430	{
433	if (eio_nthreads () >= wanted)	431	if (expect_true (etp_nthreads () >= wanted))
434	return;	432	return;
435		433
436	/* todo: maybe use idle here, but might be less exact */	434	/* todo: maybe use idle here, but might be less exact */
437	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 ()))
438	return;	436	return;
439		437
440	start_thread ();	438	etp_start_thread ();
441	}	439	}
442		440
443	void eio_submit (eio_req *req)
444	{
445	req->pri += EIO_PRI_BIAS;
446
447	if (req->pri < EIO_PRI_MIN + EIO_PRI_BIAS) req->pri = EIO_PRI_MIN + EIO_PRI_BIAS;
448	if (req->pri > EIO_PRI_MAX + EIO_PRI_BIAS) req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
449
450	++nreqs;
451
452	X_LOCK (reqlock);
453	++nready;
454	reqq_push (&req_queue, req);
455	X_COND_SIGNAL (reqwait);
456	X_UNLOCK (reqlock);
457
458	maybe_start_thread ();
459	}
460
461	static void end_thread (void)	441	static void etp_end_thread (void)
462	{	442	{
463	eio_req *req = calloc (1, sizeof (eio_req));	443	eio_req *req = calloc (1, sizeof (eio_req));
464		444
465	req->type = EIO_QUIT;	445	req->type = -1;
466	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	446	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
467		447
468	X_LOCK (reqlock);	448	X_LOCK (reqlock);
469	reqq_push (&req_queue, req);	449	reqq_push (&req_queue, req);
470	X_COND_SIGNAL (reqwait);	450	X_COND_SIGNAL (reqwait);
471	X_UNLOCK (reqlock);	451	X_UNLOCK (reqlock);
472		452
473	X_LOCK (etplock);	453	X_LOCK (wrklock);
474	--started;	454	--started;
475	X_UNLOCK (etplock);	455	X_UNLOCK (wrklock);
476	}	456	}
477		457
478	void eio_set_max_poll_time (double nseconds)
479	{
480	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
481	etp.max_poll_time = nseconds;
482	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
483	}
484
485	void eio_set_max_poll_reqs (unsigned int maxreqs)
486	{
487	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
488	etp.max_poll_reqs = maxreqs;
489	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
490	}
491
492	void eio_set_max_idle (unsigned int nthreads)
493	{
494	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
495	max_idle = nthreads <= 0 ? 1 : nthreads;
496	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
497	}
498
499	void eio_set_min_parallel (unsigned int nthreads)
500	{
501	if (wanted < nthreads)
502	wanted = nthreads;
503	}
504
505	void eio_set_max_parallel (unsigned int nthreads)
506	{
507	if (wanted > nthreads)
508	wanted = nthreads;
509
510	while (started > wanted)
511	end_thread ();
512	}
513
514	int eio_poll (void)	458	static int etp_poll (void)
515	{	459	{
516	int maxreqs = etp.max_poll_reqs;	460	unsigned int maxreqs;
		461	unsigned int maxtime;
517	struct timeval tv_start, tv_now;	462	struct timeval tv_start, tv_now;
518	eio_req *req;
519		463
520	if (etp.max_poll_time)	464	X_LOCK (reslock);
		465	maxreqs = max_poll_reqs;
		466	maxtime = max_poll_time;
		467	X_UNLOCK (reslock);
		468
		469	if (maxtime)
521	gettimeofday (&tv_start, 0);	470	gettimeofday (&tv_start, 0);
522		471
523	for (;;)	472	for (;;)
524	{	473	{
		474	ETP_REQ *req;
		475
525	maybe_start_thread ();	476	etp_maybe_start_thread ();
526		477
527	X_LOCK (reslock);	478	X_LOCK (reslock);
528	req = reqq_shift (&res_queue);	479	req = reqq_shift (&res_queue);
529		480
530	if (req)	481	if (req)
531	{	482	{
532	--npending;	483	--npending;
533		484
534	if (!res_queue.size && etp.done_poll_cb)	485	if (!res_queue.size && done_poll_cb)
535	etp.done_poll_cb ();	486	done_poll_cb ();
536	}	487	}
537		488
538	X_UNLOCK (reslock);	489	X_UNLOCK (reslock);
539		490
540	if (!req)	491	if (!req)
541	return 0;	492	return 0;
542		493
		494	X_LOCK (reqlock);
543	--nreqs;	495	--nreqs;
		496	X_UNLOCK (reqlock);
544		497
545	if (req->type == EIO_GROUP && req->size)	498	if (expect_false (req->type == EIO_GROUP && req->size))
546	{	499	{
547	req->int1 = 1; /* mark request as delayed */	500	req->int1 = 1; /* mark request as delayed */
548	continue;	501	continue;
549	}	502	}
550	else	503	else
551	{	504	{
552	int res = eio_finish (req);	505	int res = ETP_FINISH (req);
553	if (res)	506	if (expect_false (res))
554	return res;	507	return res;
555	}	508	}
556		509
557	if (maxreqs && !--maxreqs)	510	if (expect_false (maxreqs && !--maxreqs))
558	break;	511	break;
559		512
560	if (etp.max_poll_time)	513	if (maxtime)
561	{	514	{
562	gettimeofday (&tv_now, 0);	515	gettimeofday (&tv_now, 0);
563		516
564	if (tvdiff (&tv_start, &tv_now) >= etp.max_poll_time)	517	if (tvdiff (&tv_start, &tv_now) >= maxtime)
565	break;	518	break;
566	}	519	}
567	}	520	}
568		521
569	errno = EAGAIN;	522	errno = EAGAIN;
570	return -1;	523	return -1;
571	}	524	}
572		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
573	/*****************************************************************************/	741	/*****************************************************************************/
574	/* work around various missing functions */	742	/* work around various missing functions */
575		743
576	#if !HAVE_PREADWRITE	744	#if !HAVE_PREADWRITE
		745	# undef pread
		746	# undef pwrite
577	# define pread eio__pread	747	# define pread eio__pread
578	# define pwrite eio__pwrite	748	# define pwrite eio__pwrite
579
580	/*
581	* make our pread/pwrite safe against themselves, but not against
582	* normal read/write by using a mutex. slows down execution a lot,
583	* but that's your problem, not mine.
584	*/
585	static mutex_t preadwritelock = X_MUTEX_INIT;
586		749
587	static ssize_t	750	static ssize_t
588	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)
589	{	752	{
590	ssize_t res;	753	ssize_t res;
…		…
608		771
609	X_LOCK (preadwritelock);	772	X_LOCK (preadwritelock);
610	ooffset = lseek (fd, 0, SEEK_CUR);	773	ooffset = lseek (fd, 0, SEEK_CUR);
611	lseek (fd, offset, SEEK_SET);	774	lseek (fd, offset, SEEK_SET);
612	res = write (fd, buf, count);	775	res = write (fd, buf, count);
613	lseek (fd, offset, SEEK_SET);	776	lseek (fd, ooffset, SEEK_SET);
614	X_UNLOCK (preadwritelock);	777	X_UNLOCK (preadwritelock);
615		778
616	return res;	779	return res;
617	}	780	}
618	#endif	781	#endif
619		782
620	#ifndef HAVE_FUTIMES	783	#ifndef HAVE_FUTIMES
621		784
		785	# undef utimes
		786	# undef futimes
622	# define utimes(path,times) eio__utimes (path, times)	787	# define utimes(path,times) eio__utimes (path, times)
623	# define futimes(fd,times) eio__futimes (fd, times)	788	# define futimes(fd,times) eio__futimes (fd, times)
624		789
625	static int	790	static int
626	eio__utimes (const char *filename, const struct timeval times[2])	791	eio__utimes (const char *filename, const struct timeval times[2])
…		…
645	}	810	}
646		811
647	#endif	812	#endif
648		813
649	#if !HAVE_FDATASYNC	814	#if !HAVE_FDATASYNC
		815	# undef fdatasync
650	# define fdatasync fsync	816	# define fdatasync(fd) fsync (fd)
651	#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	}
652		846
653	#if !HAVE_READAHEAD	847	#if !HAVE_READAHEAD
		848	# undef readahead
654	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	849	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
655		850
656	static ssize_t	851	static ssize_t
657	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)
658	{	853	{
659	size_t todo = count;	854	size_t todo = count;
660	dBUF;	855	dBUF;
661		856
662	while (todo > 0)	857	while (todo > 0)
…		…
672	return count;	867	return count;
673	}	868	}
674		869
675	#endif	870	#endif
676		871
677	#if !HAVE_READDIR_R
678	# define readdir_r eio__readdir_r
679
680	static mutex_t readdirlock = X_MUTEX_INIT;
681
682	static int
683	eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
684	{
685	EIO_STRUCT_DIRENT *e;
686	int errorno;
687
688	X_LOCK (readdirlock);
689
690	e = readdir (dirp);
691	errorno = errno;
692
693	if (e)
694	{
695	*res = ent;
696	strcpy (ent->d_name, e->d_name);
697	}
698	else
699	*res = 0;
700
701	X_UNLOCK (readdirlock);
702
703	errno = errorno;
704	return e ? 0 : -1;
705	}
706	#endif
707
708	/* sendfile always needs emulation */	872	/* sendfile always needs emulation */
709	static ssize_t	873	static ssize_t
710	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)
711	{	875	{
712	ssize_t res;	876	ssize_t res;
713		877
714	if (!count)	878	if (!count)
715	return 0;	879	return 0;
…		…
799	return res;	963	return res;
800	}	964	}
801		965
802	/* read a full directory */	966	/* read a full directory */
803	static void	967	static void
804	eio__scandir (eio_req *req, worker *self)	968	eio__scandir (eio_req *req, etp_worker *self)
805	{	969	{
806	DIR *dirp;	970	DIR *dirp;
807	union
808	{
809	EIO_STRUCT_DIRENT d;
810	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
811	} *u;
812	EIO_STRUCT_DIRENT *entp;	971	EIO_STRUCT_DIRENT *entp;
813	char *name, *names;	972	char *name, *names;
814	int memlen = 4096;	973	int memlen = 4096;
815	int memofs = 0;	974	int memofs = 0;
816	int res = 0;	975	int res = 0;
817		976
818	X_LOCK (etplock);	977	X_LOCK (wrklock);
		978	/* the corresponding closedir is in ETP_WORKER_CLEAR */
819	self->dirp = dirp = opendir (req->ptr1);	979	self->dirp = dirp = opendir (req->ptr1);
820	self->dbuf = u = malloc (sizeof (*u));
821	req->flags |= EIO_FLAG_PTR2_FREE;	980	req->flags |= EIO_FLAG_PTR2_FREE;
822	req->ptr2 = names = malloc (memlen);	981	req->ptr2 = names = malloc (memlen);
823	X_UNLOCK (etplock);	982	X_UNLOCK (wrklock);
824		983
825	if (dirp && u && names)	984	if (dirp && names)
826	for (;;)	985	for (;;)
827	{	986	{
828	errno = 0;	987	errno = 0;
829	readdir_r (dirp, &u->d, &entp);	988	entp = readdir (dirp);
830		989
831	if (!entp)	990	if (!entp)
832	break;	991	break;
833		992
834	name = entp->d_name;	993	name = entp->d_name;
…		…
840	res++;	999	res++;
841		1000
842	while (memofs + len > memlen)	1001	while (memofs + len > memlen)
843	{	1002	{
844	memlen *= 2;	1003	memlen *= 2;
845	X_LOCK (etplock);	1004	X_LOCK (wrklock);
846	req->ptr2 = names = realloc (names, memlen);	1005	req->ptr2 = names = realloc (names, memlen);
847	X_UNLOCK (etplock);	1006	X_UNLOCK (wrklock);
848		1007
849	if (!names)	1008	if (!names)
850	break;	1009	break;
851	}	1010	}
852		1011
…		…
859	res = -1;	1018	res = -1;
860		1019
861	req->result = res;	1020	req->result = res;
862	}	1021	}
863		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;
		1051	}
		1052
864	/*****************************************************************************/	1053	/*****************************************************************************/
865		1054
866	#define ALLOC(len) \	1055	#define ALLOC(len) \
867	if (!req->ptr2) \	1056	if (!req->ptr2) \
868	{ \	1057	{ \
869	X_LOCK (etplock); \	1058	X_LOCK (wrklock); \
870	req->flags |= EIO_FLAG_PTR2_FREE; \	1059	req->flags |= EIO_FLAG_PTR2_FREE; \
871	X_UNLOCK (etplock); \	1060	X_UNLOCK (wrklock); \
872	req->ptr2 = malloc (len); \	1061	req->ptr2 = malloc (len); \
873	if (!req->ptr2) \	1062	if (!req->ptr2) \
874	{ \	1063	{ \
875	errno = ENOMEM; \	1064	errno = ENOMEM; \
876	req->result = -1; \	1065	req->result = -1; \
877	break; \	1066	break; \
878	} \	1067	} \
879	}	1068	}
880		1069
881	X_THREAD_PROC (eio_proc)	1070	X_THREAD_PROC (etp_proc)
882	{	1071	{
883	eio_req *req;	1072	ETP_REQ *req;
884	struct timespec ts;	1073	struct timespec ts;
885	worker *self = (worker *)thr_arg;	1074	etp_worker *self = (etp_worker *)thr_arg;
886		1075
887	/* try to distribute timeouts somewhat randomly */	1076	/* try to distribute timeouts somewhat randomly */
888	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1077	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
889		1078
890	for (;;)	1079	for (;;)
891	{	1080	{
892	ts.tv_sec = time (0) + IDLE_TIMEOUT;
893
894	X_LOCK (reqlock);	1081	X_LOCK (reqlock);
895		1082
896	for (;;)	1083	for (;;)
897	{	1084	{
898	self->req = req = reqq_shift (&req_queue);	1085	self->req = req = reqq_shift (&req_queue);
…		…
900	if (req)	1087	if (req)
901	break;	1088	break;
902		1089
903	++idle;	1090	++idle;
904		1091
		1092	ts.tv_sec = time (0) + IDLE_TIMEOUT;
905	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1093	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
906	{	1094	{
907	if (idle > max_idle)	1095	if (idle > max_idle)
908	{	1096	{
909	--idle;	1097	--idle;
910	X_UNLOCK (reqlock);	1098	X_UNLOCK (reqlock);
911	X_LOCK (etplock);	1099	X_LOCK (wrklock);
912	--started;	1100	--started;
913	X_UNLOCK (etplock);	1101	X_UNLOCK (wrklock);
914	goto quit;	1102	goto quit;
915	}	1103	}
916		1104
917	/* we are allowed to idle, so do so without any timeout */	1105	/* we are allowed to idle, so do so without any timeout */
918	X_COND_WAIT (reqwait, reqlock);	1106	X_COND_WAIT (reqwait, reqlock);
919	ts.tv_sec = time (0) + IDLE_TIMEOUT;
920	}	1107	}
921		1108
922	--idle;	1109	--idle;
923	}	1110	}
924		1111
925	--nready;	1112	--nready;
926		1113
927	X_UNLOCK (reqlock);	1114	X_UNLOCK (reqlock);
928		1115
929	errno = 0; /* strictly unnecessary */	1116	if (req->type < 0)
		1117	goto quit;
930		1118
931	if (!EIO_CANCELLED (req))	1119	if (!EIO_CANCELLED (req))
932	switch (req->type)	1120	ETP_EXECUTE (self, req);
933	{
934	case EIO_READ: ALLOC (req->size);
935	req->result = req->offs >= 0
936	? pread (req->int1, req->ptr2, req->size, req->offs)
937	: read (req->int1, req->ptr2, req->size); break;
938	case EIO_WRITE: req->result = req->offs >= 0
939	? pwrite (req->int1, req->ptr2, req->size, req->offs)
940	: write (req->int1, req->ptr2, req->size); break;
941
942	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
943	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
944
945	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
946	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
947	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
948	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
949	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
950	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
951
952	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
953	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
954	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
955	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
956	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
957	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
958
959	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
960	case EIO_CLOSE: req->result = close (req->int1); break;
961	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
962	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
963	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
964	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
965	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
966	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
967	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
968	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
969
970	case EIO_READLINK: ALLOC (NAME_MAX);
971	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
972
973	case EIO_SYNC: req->result = 0; sync (); break;
974	case EIO_FSYNC: req->result = fsync (req->int1); break;
975	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
976
977	case EIO_READDIR: eio__scandir (req, self); break;
978
979	case EIO_BUSY:
980	#ifdef _WIN32
981	Sleep (req->nv1 * 1000.);
982	#else
983	{
984	struct timeval tv;
985
986	tv.tv_sec = req->nv1;
987	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
988
989	req->result = select (0, 0, 0, 0, &tv);
990	}
991	#endif
992	break;
993
994	case EIO_UTIME:
995	case EIO_FUTIME:
996	{
997	struct timeval tv[2];
998	struct timeval *times;
999
1000	if (req->nv1 != -1. || req->nv2 != -1.)
1001	{
1002	tv[0].tv_sec = req->nv1;
1003	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1004	tv[1].tv_sec = req->nv2;
1005	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1006
1007	times = tv;
1008	}
1009	else
1010	times = 0;
1011
1012
1013	req->result = req->type == EIO_FUTIME
1014	? futimes (req->int1, times)
1015	: utimes (req->ptr1, times);
1016	}
1017
1018	case EIO_GROUP:
1019	case EIO_NOP:
1020	req->result = 0;
1021	break;
1022
1023	case EIO_QUIT:
1024	goto quit;
1025
1026	default:
1027	req->result = -1;
1028	break;
1029	}
1030
1031	req->errorno = errno;
1032		1121
1033	X_LOCK (reslock);	1122	X_LOCK (reslock);
1034		1123
1035	++npending;	1124	++npending;
1036		1125
1037	if (!reqq_push (&res_queue, req) && etp.want_poll_cb)	1126	if (!reqq_push (&res_queue, req) && want_poll_cb)
1038	etp.want_poll_cb ();	1127	want_poll_cb ();
1039		1128
1040	self->req = 0;	1129	self->req = 0;
1041	worker_clear (self);	1130	etp_worker_clear (self);
1042		1131
1043	X_UNLOCK (reslock);	1132	X_UNLOCK (reslock);
1044	}	1133	}
1045		1134
1046	quit:	1135	quit:
1047	X_LOCK (etplock);	1136	X_LOCK (wrklock);
1048	worker_free (self);	1137	etp_worker_free (self);
1049	X_UNLOCK (etplock);	1138	X_UNLOCK (wrklock);
1050		1139
1051	return 0;	1140	return 0;
1052	}	1141	}
1053		1142
1054	/*****************************************************************************/	1143	/*****************************************************************************/
1055		1144
1056	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1145	int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1057	{	1146	{
1058	etp_init (&etp, want_poll, done_poll);	1147	return etp_init (want_poll, done_poll);
1059	}	1148	}
1060		1149
1061	static void eio_api_destroy (eio_req *req)	1150	static void eio_api_destroy (eio_req *req)
1062	{	1151	{
1063	free (req);	1152	free (req);
…		…
1085	{ \	1174	{ \
1086	eio_api_destroy (req); \	1175	eio_api_destroy (req); \
1087	return 0; \	1176	return 0; \
1088	}	1177	}
1089		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
1090	#ifndef EIO_NO_WRAPPERS	1292	#ifndef EIO_NO_WRAPPERS
1091		1293
1092	eio_req *eio_nop (int pri, eio_cb cb, void *data)	1294	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1093	{	1295	{
1094	REQ (EIO_NOP); SEND;	1296	REQ (EIO_NOP); SEND;
…		…
1105	}	1307	}
1106		1308
1107	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)
1108	{	1310	{
1109	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;
1110	}	1327	}
1111		1328
1112	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)
1113	{	1330	{
1114	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1331	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
…		…
1235	return eio__1path (EIO_READDIR, path, pri, cb, data);	1452	return eio__1path (EIO_READDIR, path, pri, cb, data);
1236	}	1453	}
1237		1454
1238	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)
1239	{	1456	{
1240	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;
1241	}	1458	}
1242		1459
1243	static eio_req *	1460	static eio_req *
1244	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)
1245	{	1462	{
…		…
1269	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)
1270	{	1487	{
1271	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	1488	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1272	}	1489	}
1273		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
1274	#endif	1496	#endif
1275		1497
1276	eio_req *eio_grp (eio_cb cb, void *data)	1498	eio_req *eio_grp (eio_cb cb, void *data)
1277	{	1499	{
1278	const int pri = EIO_PRI_MAX;	1500	const int pri = EIO_PRI_MAX;
…		…
1304		1526
1305	void eio_grp_add (eio_req *grp, eio_req *req)	1527	void eio_grp_add (eio_req *grp, eio_req *req)
1306	{	1528	{
1307	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));
1308		1530
		1531	grp->flags |= EIO_FLAG_GROUPADD;
		1532
1309	++grp->size;	1533	++grp->size;
1310	req->grp = grp;	1534	req->grp = grp;
1311		1535
1312	req->grp_prev = 0;	1536	req->grp_prev = 0;
1313	req->grp_next = grp->grp_first;	1537	req->grp_next = grp->grp_first;
…		…
1321	/*****************************************************************************/	1545	/*****************************************************************************/
1322	/* misc garbage */	1546	/* misc garbage */
1323		1547
1324	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)
1325	{	1549	{
1326	worker wrk;	1550	etp_worker wrk;
1327		1551
1328	wrk.dbuf = 0;	1552	wrk.dbuf = 0;
1329		1553
1330	eio__sendfile (ofd, ifd, offset, count, &wrk);	1554	eio__sendfile (ofd, ifd, offset, count, &wrk);
1331		1555

Diff Legend

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