ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.10 by root, Tue May 13 17:08:15 2008 UTC vs.
Revision 1.27 by root, Wed Oct 22 18:15:36 2008 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"
2	#include "xthread.h"	41	#include "xthread.h"
3		42
4	#include <errno.h>	43	#include <errno.h>
5	#include <stddef.h>	44	#include <stddef.h>
…		…
70	/* buffer size for various temporary buffers */	109	/* buffer size for various temporary buffers */
71	#define EIO_BUFSIZE 65536	110	#define EIO_BUFSIZE 65536
72		111
73	#define dBUF \	112	#define dBUF \
74	char *eio_buf; \	113	char *eio_buf; \
75	X_LOCK (wrklock); \	114	ETP_WORKER_LOCK (self); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	115	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (wrklock); \	116	ETP_WORKER_UNLOCK (self); \
78	errno = ENOMEM; \	117	errno = ENOMEM; \
79	if (!eio_buf) \	118	if (!eio_buf) \
80	return -1;	119	return -1;
81		120
82	#define EIO_TICKS ((1000000 + 1023) >> 10)	121	#define EIO_TICKS ((1000000 + 1023) >> 10)
83		122
		123	/*****************************************************************************/
		124
		125	#if __GNUC__ >= 3
		126	# define expect(expr,value) __builtin_expect ((expr),(value))
		127	#else
		128	# define expect(expr,value) (expr)
		129	#endif
		130
		131	#define expect_false(expr) expect ((expr) != 0, 0)
		132	#define expect_true(expr) expect ((expr) != 0, 1)
		133
		134	/*****************************************************************************/
		135
		136	#define ETP_PRI_MIN EIO_PRI_MIN
		137	#define ETP_PRI_MAX EIO_PRI_MAX
		138
		139	struct etp_worker;
		140
		141	#define ETP_REQ eio_req
		142	#define ETP_DESTROY(req) eio_destroy (req)
		143	static int eio_finish (eio_req *req);
		144	#define ETP_FINISH(req) eio_finish (req)
		145	static void eio_execute (struct etp_worker *self, eio_req *req);
		146	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		147
		148	#define ETP_WORKER_CLEAR(req) \
		149	if (wrk->dbuf) \
		150	{ \
		151	free (wrk->dbuf); \
		152	wrk->dbuf = 0; \
		153	} \
		154	\
		155	if (wrk->dirp) \
		156	{ \
		157	closedir (wrk->dirp); \
		158	wrk->dirp = 0; \
		159	}
		160	#define ETP_WORKER_COMMON \
		161	void *dbuf; \
		162	DIR *dirp;
		163
		164	/*****************************************************************************/
		165
		166	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		167
		168	/* calculcate time difference in ~1/EIO_TICKS of a second */
		169	static int tvdiff (struct timeval *tv1, struct timeval *tv2)
		170	{
		171	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
		172	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
		173	}
		174
		175	static unsigned int started, idle, wanted = 4;
		176
84	static void (*want_poll_cb) (void);	177	static void (*want_poll_cb) (void);
85	static void (*done_poll_cb) (void);	178	static void (*done_poll_cb) (void);
86		179
87	static unsigned int max_poll_time = 0;	180	static unsigned int max_poll_time; /* reslock */
88	static unsigned int max_poll_reqs = 0;	181	static unsigned int max_poll_reqs; /* reslock */
89		182
90	/* calculcate time difference in ~1/EIO_TICKS of a second */	183	static volatile unsigned int nreqs; /* reqlock */
91	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	184	static volatile unsigned int nready; /* reqlock */
92	{	185	static volatile unsigned int npending; /* reqlock */
93	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	186	static volatile unsigned int max_idle = 4;
94	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
95	}
96		187
97	static unsigned int started, idle, wanted = 4;
98
99	/* worker threads management */
100	static mutex_t wrklock = X_MUTEX_INIT;	188	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;	189	static mutex_t reslock = X_MUTEX_INIT;
144	static mutex_t reqlock = X_MUTEX_INIT;	190	static mutex_t reqlock = X_MUTEX_INIT;
145	static cond_t reqwait = X_COND_INIT;	191	static cond_t reqwait = X_COND_INIT;
146		192
		193	#if !HAVE_PREADWRITE
		194	/*
		195	* make our pread/pwrite emulation safe against themselves, but not against
		196	* normal read/write by using a mutex. slows down execution a lot,
		197	* but that's your problem, not mine.
		198	*/
		199	static mutex_t preadwritelock = X_MUTEX_INIT;
		200	#endif
		201
		202	typedef struct etp_worker
		203	{
		204	/* locked by wrklock */
		205	struct etp_worker *prev, *next;
		206
		207	thread_t tid;
		208
		209	/* locked by reslock, reqlock or wrklock */
		210	ETP_REQ *req; /* currently processed request */
		211
		212	ETP_WORKER_COMMON
		213	} etp_worker;
		214
		215	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
		216
		217	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		218	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
		219
		220	/* worker threads management */
		221
		222	static void etp_worker_clear (etp_worker *wrk)
		223	{
		224	ETP_WORKER_CLEAR (wrk);
		225	}
		226
		227	static void etp_worker_free (etp_worker *wrk)
		228	{
		229	wrk->next->prev = wrk->prev;
		230	wrk->prev->next = wrk->next;
		231
		232	free (wrk);
		233	}
		234
147	unsigned int eio_nreqs (void)	235	static unsigned int etp_nreqs (void)
148	{	236	{
		237	int retval;
		238	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		239	retval = nreqs;
		240	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
149	return nreqs;	241	return retval;
150	}	242	}
151		243
152	unsigned int eio_nready (void)	244	static unsigned int etp_nready (void)
153	{	245	{
154	unsigned int retval;	246	unsigned int retval;
155		247
156	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	248	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157	retval = nready;	249	retval = nready;
158	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	250	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159		251
160	return retval;	252	return retval;
161	}	253	}
162		254
163	unsigned int eio_npending (void)	255	static unsigned int etp_npending (void)
164	{	256	{
165	unsigned int retval;	257	unsigned int retval;
166		258
167	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	259	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168	retval = npending;	260	retval = npending;
169	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	261	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170		262
171	return retval;	263	return retval;
172	}	264	}
173		265
174	unsigned int eio_nthreads (void)	266	static unsigned int etp_nthreads (void)
175	{	267	{
176	unsigned int retval;	268	unsigned int retval;
177		269
178	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	270	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179	retval = started;	271	retval = started;
…		…
186	* a somewhat faster data structure might be nice, but	278	* a somewhat faster data structure might be nice, but
187	* with 8 priorities this actually needs <20 insns	279	* with 8 priorities this actually needs <20 insns
188	* per shift, the most expensive operation.	280	* per shift, the most expensive operation.
189	*/	281	*/
190	typedef struct {	282	typedef struct {
191	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	283	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
192	int size;	284	int size;
193	} reqq;	285	} etp_reqq;
194		286
195	static reqq req_queue;	287	static etp_reqq req_queue;
196	static reqq res_queue;	288	static etp_reqq res_queue;
197		289
198	static int reqq_push (reqq *q, eio_req *req)	290	static int reqq_push (etp_reqq *q, ETP_REQ *req)
199	{	291	{
200	int pri = req->pri;	292	int pri = req->pri;
201	req->next = 0;	293	req->next = 0;
202		294
203	if (q->qe[pri])	295	if (q->qe[pri])
…		…
209	q->qe[pri] = q->qs[pri] = req;	301	q->qe[pri] = q->qs[pri] = req;
210		302
211	return q->size++;	303	return q->size++;
212	}	304	}
213		305
214	static eio_req *reqq_shift (reqq *q)	306	static ETP_REQ *reqq_shift (etp_reqq *q)
215	{	307	{
216	int pri;	308	int pri;
217		309
218	if (!q->size)	310	if (!q->size)
219	return 0;	311	return 0;
220		312
221	--q->size;	313	--q->size;
222		314
223	for (pri = EIO_NUM_PRI; pri--; )	315	for (pri = ETP_NUM_PRI; pri--; )
224	{	316	{
225	eio_req *req = q->qs[pri];	317	eio_req *req = q->qs[pri];
226		318
227	if (req)	319	if (req)
228	{	320	{
…		…
234	}	326	}
235		327
236	abort ();	328	abort ();
237	}	329	}
238		330
239	static void grp_try_feed (eio_req *grp)	331	static void etp_atfork_prepare (void)
240	{	332	{
241	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	333	X_LOCK (wrklock);
		334	X_LOCK (reqlock);
		335	X_LOCK (reslock);
		336	#if !HAVE_PREADWRITE
		337	X_LOCK (preadwritelock);
		338	#endif
		339	}
		340
		341	static void etp_atfork_parent (void)
		342	{
		343	#if !HAVE_PREADWRITE
		344	X_UNLOCK (preadwritelock);
		345	#endif
		346	X_UNLOCK (reslock);
		347	X_UNLOCK (reqlock);
		348	X_UNLOCK (wrklock);
		349	}
		350
		351	static void etp_atfork_child (void)
		352	{
		353	ETP_REQ *prv;
		354
		355	while ((prv = reqq_shift (&req_queue)))
		356	ETP_DESTROY (prv);
		357
		358	while ((prv = reqq_shift (&res_queue)))
		359	ETP_DESTROY (prv);
		360
		361	while (wrk_first.next != &wrk_first)
242	{	362	{
243	int old_len = grp->size;	363	etp_worker *wrk = wrk_first.next;
244		364
245	EIO_FEED (grp);	365	if (wrk->req)
		366	ETP_DESTROY (wrk->req);
246		367
247	/* stop if no progress has been made */	368	etp_worker_clear (wrk);
248	if (old_len == grp->size)	369	etp_worker_free (wrk);
249	{
250	grp->feed = 0;
251	break;
252	}
253	}	370	}
254	}
255		371
256	static int eio_finish (eio_req *req);	372	started = 0;
		373	idle = 0;
		374	nreqs = 0;
		375	nready = 0;
		376	npending = 0;
257		377
258	static int grp_dec (eio_req *grp)	378	etp_atfork_parent ();
259	{	379	}
260	--grp->size;
261		380
262	/* call feeder, if applicable */	381	static void
263	grp_try_feed (grp);	382	etp_once_init (void)
		383	{
		384	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
		385	}
264		386
265	/* finish, if done */	387	static int
266	if (!grp->size && grp->int1)	388	etp_init (void (*want_poll)(void), void (*done_poll)(void))
267	return eio_finish (grp);	389	{
268	else	390	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
		391
		392	pthread_once (&doinit, etp_once_init);
		393
		394	want_poll_cb = want_poll;
		395	done_poll_cb = done_poll;
		396
269	return 0;	397	return 0;
270	}	398	}
271		399
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);	400	X_THREAD_PROC (etp_proc);
323		401
324	static void start_thread (void)	402	static void etp_start_thread (void)
325	{	403	{
326	worker *wrk = calloc (1, sizeof (worker));	404	etp_worker *wrk = calloc (1, sizeof (etp_worker));
327		405
328	/*TODO*/	406	/*TODO*/
329	assert (("unable to allocate worker thread data", wrk));	407	assert (("unable to allocate worker thread data", wrk));
330		408
331	X_LOCK (wrklock);	409	X_LOCK (wrklock);
332		410
333	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	411	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334	{	412	{
335	wrk->prev = &wrk_first;	413	wrk->prev = &wrk_first;
336	wrk->next = wrk_first.next;	414	wrk->next = wrk_first.next;
337	wrk_first.next->prev = wrk;	415	wrk_first.next->prev = wrk;
338	wrk_first.next = wrk;	416	wrk_first.next = wrk;
…		…
342	free (wrk);	420	free (wrk);
343		421
344	X_UNLOCK (wrklock);	422	X_UNLOCK (wrklock);
345	}	423	}
346		424
347	static void maybe_start_thread (void)	425	static void etp_maybe_start_thread (void)
348	{	426	{
349	if (eio_nthreads () >= wanted)	427	if (expect_true (etp_nthreads () >= wanted))
350	return;	428	return;
351		429
352	/* todo: maybe use idle here, but might be less exact */	430	/* todo: maybe use idle here, but might be less exact */
353	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	431	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354	return;	432	return;
355		433
356	start_thread ();	434	etp_start_thread ();
357	}	435	}
358		436
359	void eio_submit (eio_req *req)
360	{
361	++nreqs;
362
363	X_LOCK (reqlock);
364	++nready;
365	reqq_push (&req_queue, req);
366	X_COND_SIGNAL (reqwait);
367	X_UNLOCK (reqlock);
368
369	maybe_start_thread ();
370	}
371
372	static void end_thread (void)	437	static void etp_end_thread (void)
373	{	438	{
374	eio_req *req = calloc (1, sizeof (eio_req));	439	eio_req *req = calloc (1, sizeof (eio_req));
375		440
376	req->type = EIO_QUIT;	441	req->type = -1;
377	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	442	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
378		443
379	X_LOCK (reqlock);	444	X_LOCK (reqlock);
380	reqq_push (&req_queue, req);	445	reqq_push (&req_queue, req);
381	X_COND_SIGNAL (reqwait);	446	X_COND_SIGNAL (reqwait);
382	X_UNLOCK (reqlock);	447	X_UNLOCK (reqlock);
…		…
384	X_LOCK (wrklock);	449	X_LOCK (wrklock);
385	--started;	450	--started;
386	X_UNLOCK (wrklock);	451	X_UNLOCK (wrklock);
387	}	452	}
388		453
389	void eio_set_max_poll_time (double nseconds)
390	{
391	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
392	max_poll_time = nseconds;
393	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
394	}
395
396	void eio_set_max_poll_reqs (unsigned int maxreqs)
397	{
398	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
399	max_poll_reqs = maxreqs;
400	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
401	}
402
403	void eio_set_max_idle (unsigned int nthreads)
404	{
405	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
406	max_idle = nthreads <= 0 ? 1 : nthreads;
407	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
408	}
409
410	void eio_set_min_parallel (unsigned int nthreads)
411	{
412	if (wanted < nthreads)
413	wanted = nthreads;
414	}
415
416	void eio_set_max_parallel (unsigned int nthreads)
417	{
418	if (wanted > nthreads)
419	wanted = nthreads;
420
421	while (started > wanted)
422	end_thread ();
423	}
424
425	int eio_poll (void)	454	static int etp_poll (void)
426	{	455	{
427	int maxreqs = max_poll_reqs;	456	unsigned int maxreqs;
		457	unsigned int maxtime;
428	struct timeval tv_start, tv_now;	458	struct timeval tv_start, tv_now;
429	eio_req *req;
430		459
		460	X_LOCK (reslock);
		461	maxreqs = max_poll_reqs;
		462	maxtime = max_poll_time;
		463	X_UNLOCK (reslock);
		464
431	if (max_poll_time)	465	if (maxtime)
432	gettimeofday (&tv_start, 0);	466	gettimeofday (&tv_start, 0);
433		467
434	for (;;)	468	for (;;)
435	{	469	{
		470	ETP_REQ *req;
		471
436	maybe_start_thread ();	472	etp_maybe_start_thread ();
437		473
438	X_LOCK (reslock);	474	X_LOCK (reslock);
439	req = reqq_shift (&res_queue);	475	req = reqq_shift (&res_queue);
440		476
441	if (req)	477	if (req)
…		…
449	X_UNLOCK (reslock);	485	X_UNLOCK (reslock);
450		486
451	if (!req)	487	if (!req)
452	return 0;	488	return 0;
453		489
		490	X_LOCK (reqlock);
454	--nreqs;	491	--nreqs;
		492	X_UNLOCK (reqlock);
455		493
456	if (req->type == EIO_GROUP && req->size)	494	if (expect_false (req->type == EIO_GROUP && req->size))
457	{	495	{
458	req->int1 = 1; /* mark request as delayed */	496	req->int1 = 1; /* mark request as delayed */
459	continue;	497	continue;
460	}	498	}
461	else	499	else
462	{	500	{
463	int res = eio_finish (req);	501	int res = ETP_FINISH (req);
464	if (res)	502	if (expect_false (res))
465	return res;	503	return res;
466	}	504	}
467		505
468	if (maxreqs && !--maxreqs)	506	if (expect_false (maxreqs && !--maxreqs))
469	break;	507	break;
470		508
471	if (max_poll_time)	509	if (maxtime)
472	{	510	{
473	gettimeofday (&tv_now, 0);	511	gettimeofday (&tv_now, 0);
474		512
475	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	513	if (tvdiff (&tv_start, &tv_now) >= maxtime)
476	break;	514	break;
477	}	515	}
478	}	516	}
479		517
480	errno = EAGAIN;	518	errno = EAGAIN;
481	return -1;	519	return -1;
482	}	520	}
483		521
		522	static void etp_cancel (ETP_REQ *req)
		523	{
		524	X_LOCK (wrklock);
		525	req->flags |= EIO_FLAG_CANCELLED;
		526	X_UNLOCK (wrklock);
		527
		528	eio_grp_cancel (req);
		529	}
		530
		531	static void etp_submit (ETP_REQ *req)
		532	{
		533	req->pri -= ETP_PRI_MIN;
		534
		535	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		536	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		537
		538	if (expect_false (req->type == EIO_GROUP))
		539	{
		540	/* I hope this is worth it :/ */
		541	X_LOCK (reqlock);
		542	++nreqs;
		543	X_UNLOCK (reqlock);
		544
		545	X_LOCK (reslock);
		546
		547	++npending;
		548
		549	if (!reqq_push (&res_queue, req) && want_poll_cb)
		550	want_poll_cb ();
		551
		552	X_UNLOCK (reslock);
		553	}
		554	else
		555	{
		556	X_LOCK (reqlock);
		557	++nreqs;
		558	++nready;
		559	reqq_push (&req_queue, req);
		560	X_COND_SIGNAL (reqwait);
		561	X_UNLOCK (reqlock);
		562
		563	etp_maybe_start_thread ();
		564	}
		565	}
		566
		567	static void etp_set_max_poll_time (double nseconds)
		568	{
		569	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		570	max_poll_time = nseconds;
		571	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		572	}
		573
		574	static void etp_set_max_poll_reqs (unsigned int maxreqs)
		575	{
		576	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		577	max_poll_reqs = maxreqs;
		578	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		579	}
		580
		581	static void etp_set_max_idle (unsigned int nthreads)
		582	{
		583	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		584	max_idle = nthreads <= 0 ? 1 : nthreads;
		585	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		586	}
		587
		588	static void etp_set_min_parallel (unsigned int nthreads)
		589	{
		590	if (wanted < nthreads)
		591	wanted = nthreads;
		592	}
		593
		594	static void etp_set_max_parallel (unsigned int nthreads)
		595	{
		596	if (wanted > nthreads)
		597	wanted = nthreads;
		598
		599	while (started > wanted)
		600	etp_end_thread ();
		601	}
		602
		603	/*****************************************************************************/
		604
		605	static void grp_try_feed (eio_req *grp)
		606	{
		607	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		608	{
		609	grp->flags &= ~EIO_FLAG_GROUPADD;
		610
		611	EIO_FEED (grp);
		612
		613	/* stop if no progress has been made */
		614	if (!(grp->flags & EIO_FLAG_GROUPADD))
		615	{
		616	grp->feed = 0;
		617	break;
		618	}
		619	}
		620	}
		621
		622	static int grp_dec (eio_req *grp)
		623	{
		624	--grp->size;
		625
		626	/* call feeder, if applicable */
		627	grp_try_feed (grp);
		628
		629	/* finish, if done */
		630	if (!grp->size && grp->int1)
		631	return eio_finish (grp);
		632	else
		633	return 0;
		634	}
		635
		636	void eio_destroy (eio_req *req)
		637	{
		638	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		639	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		640
		641	EIO_DESTROY (req);
		642	}
		643
		644	static int eio_finish (eio_req *req)
		645	{
		646	int res = EIO_FINISH (req);
		647
		648	if (req->grp)
		649	{
		650	int res2;
		651	eio_req *grp = req->grp;
		652
		653	/* unlink request */
		654	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		655	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		656
		657	if (grp->grp_first == req)
		658	grp->grp_first = req->grp_next;
		659
		660	res2 = grp_dec (grp);
		661
		662	if (!res && res2)
		663	res = res2;
		664	}
		665
		666	eio_destroy (req);
		667
		668	return res;
		669	}
		670
		671	void eio_grp_cancel (eio_req *grp)
		672	{
		673	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		674	eio_cancel (grp);
		675	}
		676
		677	void eio_cancel (eio_req *req)
		678	{
		679	etp_cancel (req);
		680	}
		681
		682	void eio_submit (eio_req *req)
		683	{
		684	etp_submit (req);
		685	}
		686
		687	unsigned int eio_nreqs (void)
		688	{
		689	return etp_nreqs ();
		690	}
		691
		692	unsigned int eio_nready (void)
		693	{
		694	return etp_nready ();
		695	}
		696
		697	unsigned int eio_npending (void)
		698	{
		699	return etp_npending ();
		700	}
		701
		702	unsigned int eio_nthreads (void)
		703	{
		704	return etp_nthreads ();
		705	}
		706
		707	void eio_set_max_poll_time (double nseconds)
		708	{
		709	etp_set_max_poll_time (nseconds);
		710	}
		711
		712	void eio_set_max_poll_reqs (unsigned int maxreqs)
		713	{
		714	etp_set_max_poll_reqs (maxreqs);
		715	}
		716
		717	void eio_set_max_idle (unsigned int nthreads)
		718	{
		719	etp_set_max_idle (nthreads);
		720	}
		721
		722	void eio_set_min_parallel (unsigned int nthreads)
		723	{
		724	etp_set_min_parallel (nthreads);
		725	}
		726
		727	void eio_set_max_parallel (unsigned int nthreads)
		728	{
		729	etp_set_max_parallel (nthreads);
		730	}
		731
		732	int eio_poll (void)
		733	{
		734	return etp_poll ();
		735	}
		736
484	/*****************************************************************************/	737	/*****************************************************************************/
485	/* work around various missing functions */	738	/* work around various missing functions */
486		739
487	#if !HAVE_PREADWRITE	740	#if !HAVE_PREADWRITE
		741	# undef pread
		742	# undef pwrite
488	# define pread eio__pread	743	# define pread eio__pread
489	# define pwrite eio__pwrite	744	# define pwrite eio__pwrite
490
491	/*
492	* make our pread/pwrite safe against themselves, but not against
493	* normal read/write by using a mutex. slows down execution a lot,
494	* but that's your problem, not mine.
495	*/
496	static mutex_t preadwritelock = X_MUTEX_INIT;
497		745
498	static ssize_t	746	static ssize_t
499	eio__pread (int fd, void *buf, size_t count, off_t offset)	747	eio__pread (int fd, void *buf, size_t count, off_t offset)
500	{	748	{
501	ssize_t res;	749	ssize_t res;
…		…
528	}	776	}
529	#endif	777	#endif
530		778
531	#ifndef HAVE_FUTIMES	779	#ifndef HAVE_FUTIMES
532		780
		781	# undef utimes
		782	# undef futimes
533	# define utimes(path,times) eio__utimes (path, times)	783	# define utimes(path,times) eio__utimes (path, times)
534	# define futimes(fd,times) eio__futimes (fd, times)	784	# define futimes(fd,times) eio__futimes (fd, times)
535		785
536	static int	786	static int
537	eio__utimes (const char *filename, const struct timeval times[2])	787	eio__utimes (const char *filename, const struct timeval times[2])
…		…
556	}	806	}
557		807
558	#endif	808	#endif
559		809
560	#if !HAVE_FDATASYNC	810	#if !HAVE_FDATASYNC
		811	# undef fdatasync
561	# define fdatasync fsync	812	# define fdatasync(fd) fsync (fd)
562	#endif	813	#endif
		814
		815	/* sync_file_range always needs emulation */
		816	int
		817	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		818	{
		819	#if HAVE_SYNC_FILE_RANGE
		820	int res;
		821
		822	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		823	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		824	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		825	{
		826	flags = 0
		827	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		828	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		829	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		830	}
		831
		832	res = sync_file_range (fd, offset, nbytes, flags);
		833
		834	if (res != ENOSYS)
		835	return res;
		836	#endif
		837
		838	/* even though we could play tricks with the flags, it's better to always
		839	* call fdatasync, as thta matches the expectation of it's users best */
		840	return fdatasync (fd);
		841	}
563		842
564	#if !HAVE_READAHEAD	843	#if !HAVE_READAHEAD
		844	# undef readahead
565	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	845	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566		846
567	static ssize_t	847	static ssize_t
568	eio__readahead (int fd, off_t offset, size_t count, worker *self)	848	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
569	{	849	{
570	size_t todo = count;	850	size_t todo = count;
571	dBUF;	851	dBUF;
572		852
573	while (todo > 0)	853	while (todo > 0)
…		…
583	return count;	863	return count;
584	}	864	}
585		865
586	#endif	866	#endif
587		867
588	#if !HAVE_READDIR_R
589	# define readdir_r eio__readdir_r
590
591	static mutex_t readdirlock = X_MUTEX_INIT;
592
593	static int
594	eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
595	{
596	EIO_STRUCT_DIRENT *e;
597	int errorno;
598
599	X_LOCK (readdirlock);
600
601	e = readdir (dirp);
602	errorno = errno;
603
604	if (e)
605	{
606	*res = ent;
607	strcpy (ent->d_name, e->d_name);
608	}
609	else
610	*res = 0;
611
612	X_UNLOCK (readdirlock);
613
614	errno = errorno;
615	return e ? 0 : -1;
616	}
617	#endif
618
619	/* sendfile always needs emulation */	868	/* sendfile always needs emulation */
620	static ssize_t	869	static ssize_t
621	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self)	870	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
622	{	871	{
623	ssize_t res;	872	ssize_t res;
624		873
625	if (!count)	874	if (!count)
626	return 0;	875	return 0;
…		…
710	return res;	959	return res;
711	}	960	}
712		961
713	/* read a full directory */	962	/* read a full directory */
714	static void	963	static void
715	eio__scandir (eio_req *req, worker *self)	964	eio__scandir (eio_req *req, etp_worker *self)
716	{	965	{
717	DIR *dirp;	966	DIR *dirp;
718	union
719	{
720	EIO_STRUCT_DIRENT d;
721	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
722	} *u;
723	EIO_STRUCT_DIRENT *entp;	967	EIO_STRUCT_DIRENT *entp;
724	char *name, *names;	968	char *name, *names;
725	int memlen = 4096;	969	int memlen = 4096;
726	int memofs = 0;	970	int memofs = 0;
727	int res = 0;	971	int res = 0;
728		972
729	X_LOCK (wrklock);	973	X_LOCK (wrklock);
		974	/* the corresponding closedir is in ETP_WORKER_CLEAR */
730	self->dirp = dirp = opendir (req->ptr1);	975	self->dirp = dirp = opendir (req->ptr1);
731	self->dbuf = u = malloc (sizeof (*u));
732	req->flags |= EIO_FLAG_PTR2_FREE;	976	req->flags |= EIO_FLAG_PTR2_FREE;
733	req->ptr2 = names = malloc (memlen);	977	req->ptr2 = names = malloc (memlen);
734	X_UNLOCK (wrklock);	978	X_UNLOCK (wrklock);
735		979
736	if (dirp && u && names)	980	if (dirp && names)
737	for (;;)	981	for (;;)
738	{	982	{
739	errno = 0;	983	errno = 0;
740	readdir_r (dirp, &u->d, &entp);	984	entp = readdir (dirp);
741		985
742	if (!entp)	986	if (!entp)
743	break;	987	break;
744		988
745	name = entp->d_name;	989	name = entp->d_name;
…		…
768		1012
769	if (errno)	1013	if (errno)
770	res = -1;	1014	res = -1;
771		1015
772	req->result = res;	1016	req->result = res;
		1017	}
		1018
		1019	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1020	# undef msync
		1021	# define msync(a,b,c) ENOSYS
		1022	#endif
		1023
		1024	int
		1025	eio__mtouch (void *mem, size_t len, int flags)
		1026	{
		1027	intptr_t addr = (intptr_t)mem;
		1028	intptr_t end = addr + len;
		1029	#ifdef PAGESIZE
		1030	const intptr_t page = PAGESIZE;
		1031	#else
		1032	static intptr_t page;
		1033
		1034	if (!page)
		1035	page = sysconf (_SC_PAGESIZE);
		1036	#endif
		1037
		1038	addr &= ~(page - 1); /* assume page size is always a power of two */
		1039
		1040	if (addr < end)
		1041	if (flags) /* modify */
		1042	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
		1043	else
		1044	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
		1045
		1046	return 0;
773	}	1047	}
774		1048
775	/*****************************************************************************/	1049	/*****************************************************************************/
776		1050
777	#define ALLOC(len) \	1051	#define ALLOC(len) \
…		…
787	req->result = -1; \	1061	req->result = -1; \
788	break; \	1062	break; \
789	} \	1063	} \
790	}	1064	}
791		1065
792	X_THREAD_PROC (eio_proc)	1066	X_THREAD_PROC (etp_proc)
793	{	1067	{
794	eio_req *req;	1068	ETP_REQ *req;
795	struct timespec ts;	1069	struct timespec ts;
796	worker *self = (worker *)thr_arg;	1070	etp_worker *self = (etp_worker *)thr_arg;
797		1071
798	/* try to distribute timeouts somewhat randomly */	1072	/* try to distribute timeouts somewhat randomly */
799	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1073	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
800		1074
801	for (;;)	1075	for (;;)
802	{	1076	{
803	ts.tv_sec = time (0) + IDLE_TIMEOUT;
804
805	X_LOCK (reqlock);	1077	X_LOCK (reqlock);
806		1078
807	for (;;)	1079	for (;;)
808	{	1080	{
809	self->req = req = reqq_shift (&req_queue);	1081	self->req = req = reqq_shift (&req_queue);
…		…
811	if (req)	1083	if (req)
812	break;	1084	break;
813		1085
814	++idle;	1086	++idle;
815		1087
		1088	ts.tv_sec = time (0) + IDLE_TIMEOUT;
816	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1089	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
817	{	1090	{
818	if (idle > max_idle)	1091	if (idle > max_idle)
819	{	1092	{
820	--idle;	1093	--idle;
…		…
825	goto quit;	1098	goto quit;
826	}	1099	}
827		1100
828	/* we are allowed to idle, so do so without any timeout */	1101	/* we are allowed to idle, so do so without any timeout */
829	X_COND_WAIT (reqwait, reqlock);	1102	X_COND_WAIT (reqwait, reqlock);
830	ts.tv_sec = time (0) + IDLE_TIMEOUT;
831	}	1103	}
832		1104
833	--idle;	1105	--idle;
834	}	1106	}
835		1107
836	--nready;	1108	--nready;
837		1109
838	X_UNLOCK (reqlock);	1110	X_UNLOCK (reqlock);
839		1111
840	errno = 0; /* strictly unnecessary */	1112	if (req->type < 0)
		1113	goto quit;
841		1114
842	if (!EIO_CANCELLED (req))	1115	if (!EIO_CANCELLED (req))
843	switch (req->type)	1116	ETP_EXECUTE (self, req);
844	{
845	case EIO_READ: ALLOC (req->size);
846	req->result = req->offs >= 0
847	? pread (req->int1, req->ptr2, req->size, req->offs)
848	: read (req->int1, req->ptr2, req->size); break;
849	case EIO_WRITE: req->result = req->offs >= 0
850	? pwrite (req->int1, req->ptr2, req->size, req->offs)
851	: write (req->int1, req->ptr2, req->size); break;
852
853	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
854	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
855
856	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
857	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
858	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
859	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
860	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
861	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
862
863	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
864	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
865	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
866	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
867	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
868	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
869
870	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
871	case EIO_CLOSE: req->result = close (req->int1); break;
872	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
873	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
874	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
875	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
876	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
877	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
878	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
879	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
880
881	case EIO_READLINK: ALLOC (NAME_MAX);
882	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
883
884	case EIO_SYNC: req->result = 0; sync (); break;
885	case EIO_FSYNC: req->result = fsync (req->int1); break;
886	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
887
888	case EIO_READDIR: eio__scandir (req, self); break;
889
890	case EIO_BUSY:
891	#ifdef _WIN32
892	Sleep (req->nv1 * 1000.);
893	#else
894	{
895	struct timeval tv;
896
897	tv.tv_sec = req->nv1;
898	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
899
900	req->result = select (0, 0, 0, 0, &tv);
901	}
902	#endif
903	break;
904
905	case EIO_UTIME:
906	case EIO_FUTIME:
907	{
908	struct timeval tv[2];
909	struct timeval *times;
910
911	if (req->nv1 != -1. || req->nv2 != -1.)
912	{
913	tv[0].tv_sec = req->nv1;
914	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
915	tv[1].tv_sec = req->nv2;
916	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
917
918	times = tv;
919	}
920	else
921	times = 0;
922
923
924	req->result = req->type == EIO_FUTIME
925	? futimes (req->int1, times)
926	: utimes (req->ptr1, times);
927	}
928
929	case EIO_GROUP:
930	case EIO_NOP:
931	req->result = 0;
932	break;
933
934	case EIO_QUIT:
935	goto quit;
936
937	default:
938	req->result = -1;
939	break;
940	}
941
942	req->errorno = errno;
943		1117
944	X_LOCK (reslock);	1118	X_LOCK (reslock);
945		1119
946	++npending;	1120	++npending;
947		1121
948	if (!reqq_push (&res_queue, req) && want_poll_cb)	1122	if (!reqq_push (&res_queue, req) && want_poll_cb)
949	want_poll_cb ();	1123	want_poll_cb ();
950		1124
951	self->req = 0;	1125	self->req = 0;
952	worker_clear (self);	1126	etp_worker_clear (self);
953		1127
954	X_UNLOCK (reslock);	1128	X_UNLOCK (reslock);
955	}	1129	}
956		1130
957	quit:	1131	quit:
958	X_LOCK (wrklock);	1132	X_LOCK (wrklock);
959	worker_free (self);	1133	etp_worker_free (self);
960	X_UNLOCK (wrklock);	1134	X_UNLOCK (wrklock);
961		1135
962	return 0;	1136	return 0;
963	}	1137	}
964		1138
965	/*****************************************************************************/	1139	/*****************************************************************************/
966		1140
967	static void eio_atfork_prepare (void)
968	{
969	X_LOCK (wrklock);
970	X_LOCK (reqlock);
971	X_LOCK (reslock);
972	#if !HAVE_PREADWRITE
973	X_LOCK (preadwritelock);
974	#endif
975	#if !HAVE_READDIR_R
976	X_LOCK (readdirlock);
977	#endif
978	}
979
980	static void eio_atfork_parent (void)
981	{
982	#if !HAVE_READDIR_R
983	X_UNLOCK (readdirlock);
984	#endif
985	#if !HAVE_PREADWRITE
986	X_UNLOCK (preadwritelock);
987	#endif
988	X_UNLOCK (reslock);
989	X_UNLOCK (reqlock);
990	X_UNLOCK (wrklock);
991	}
992
993	static void eio_atfork_child (void)
994	{
995	eio_req *prv;
996
997	while (prv = reqq_shift (&req_queue))
998	eio_destroy (prv);
999
1000	while (prv = reqq_shift (&res_queue))
1001	eio_destroy (prv);
1002
1003	while (wrk_first.next != &wrk_first)
1004	{
1005	worker *wrk = wrk_first.next;
1006
1007	if (wrk->req)
1008	eio_destroy (wrk->req);
1009
1010	worker_clear (wrk);
1011	worker_free (wrk);
1012	}
1013
1014	started = 0;
1015	idle = 0;
1016	nreqs = 0;
1017	nready = 0;
1018	npending = 0;
1019
1020	eio_atfork_parent ();
1021	}
1022
1023	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1141	int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1024	{	1142	{
1025	want_poll_cb = want_poll;	1143	return etp_init (want_poll, done_poll);
1026	done_poll_cb = done_poll;
1027
1028	#ifdef _WIN32
1029	X_MUTEX_CHECK (wrklock);
1030	X_MUTEX_CHECK (reslock);
1031	X_MUTEX_CHECK (reqlock);
1032	X_MUTEX_CHECK (reqwait);
1033	X_MUTEX_CHECK (preadwritelock);
1034	X_MUTEX_CHECK (readdirlock);
1035
1036	X_COND_CHECK (reqwait);
1037	#endif
1038
1039	X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1040	}	1144	}
1041		1145
1042	static void eio_api_destroy (eio_req *req)	1146	static void eio_api_destroy (eio_req *req)
1043	{	1147	{
1044	free (req);	1148	free (req);
…		…
1049	\	1153	\
1050	req = (eio_req *)calloc (1, sizeof *req); \	1154	req = (eio_req *)calloc (1, sizeof *req); \
1051	if (!req) \	1155	if (!req) \
1052	return 0; \	1156	return 0; \
1053	\	1157	\
1054	req->type = rtype; \	1158	req->type = rtype; \
1055	req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \	1159	req->pri = pri; \
1056	req->finish = cb; \	1160	req->finish = cb; \
		1161	req->data = data; \
1057	req->destroy = eio_api_destroy;	1162	req->destroy = eio_api_destroy;
1058		1163
1059	#define SEND eio_submit (req); return req	1164	#define SEND eio_submit (req); return req
1060		1165
1061	#define PATH \	1166	#define PATH \
…		…
1065	{ \	1170	{ \
1066	eio_api_destroy (req); \	1171	eio_api_destroy (req); \
1067	return 0; \	1172	return 0; \
1068	}	1173	}
1069		1174
		1175	static void eio_execute (etp_worker *self, eio_req *req)
		1176	{
		1177	errno = 0;
		1178
		1179	switch (req->type)
		1180	{
		1181	case EIO_READ: ALLOC (req->size);
		1182	req->result = req->offs >= 0
		1183	? pread (req->int1, req->ptr2, req->size, req->offs)
		1184	: read (req->int1, req->ptr2, req->size); break;
		1185	case EIO_WRITE: req->result = req->offs >= 0
		1186	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1187	: write (req->int1, req->ptr2, req->size); break;
		1188
		1189	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		1190	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
		1191
		1192	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1193	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1194	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1195	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1196	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1197	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1198
		1199	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1200	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1201	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1202	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1203	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1204	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1205
		1206	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1207	case EIO_CLOSE: req->result = close (req->int1); break;
		1208	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1209	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1210	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1211	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1212	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1213	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1214	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1215	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
		1216
		1217	case EIO_READLINK: ALLOC (NAME_MAX);
		1218	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
		1219
		1220	case EIO_SYNC: req->result = 0; sync (); break;
		1221	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1222	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1223	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
		1224	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1225	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1226
		1227	case EIO_READDIR: eio__scandir (req, self); break;
		1228
		1229	case EIO_BUSY:
		1230	#ifdef _WIN32
		1231	Sleep (req->nv1 * 1000.);
		1232	#else
		1233	{
		1234	struct timeval tv;
		1235
		1236	tv.tv_sec = req->nv1;
		1237	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
		1238
		1239	req->result = select (0, 0, 0, 0, &tv);
		1240	}
		1241	#endif
		1242	break;
		1243
		1244	case EIO_UTIME:
		1245	case EIO_FUTIME:
		1246	{
		1247	struct timeval tv[2];
		1248	struct timeval *times;
		1249
		1250	if (req->nv1 != -1. || req->nv2 != -1.)
		1251	{
		1252	tv[0].tv_sec = req->nv1;
		1253	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		1254	tv[1].tv_sec = req->nv2;
		1255	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		1256
		1257	times = tv;
		1258	}
		1259	else
		1260	times = 0;
		1261
		1262
		1263	req->result = req->type == EIO_FUTIME
		1264	? futimes (req->int1, times)
		1265	: utimes (req->ptr1, times);
		1266	}
		1267	break;
		1268
		1269	case EIO_GROUP:
		1270	abort (); /* handled in eio_request */
		1271
		1272	case EIO_NOP:
		1273	req->result = 0;
		1274	break;
		1275
		1276	case EIO_CUSTOM:
		1277	((void (*)(eio_req *))req->feed) (req);
		1278	break;
		1279
		1280	default:
		1281	req->result = -1;
		1282	break;
		1283	}
		1284
		1285	req->errorno = errno;
		1286	}
		1287
		1288	#ifndef EIO_NO_WRAPPERS
		1289
1070	eio_req *eio_nop (int pri, eio_cb cb, void *data)	1290	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1071	{	1291	{
1072	REQ (EIO_NOP); SEND;	1292	REQ (EIO_NOP); SEND;
1073	}	1293	}
1074		1294
…		…
1083	}	1303	}
1084		1304
1085	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	1305	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1086	{	1306	{
1087	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1307	REQ (EIO_FSYNC); req->int1 = fd; SEND;
		1308	}
		1309
		1310	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1311	{
		1312	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1313	}
		1314
		1315	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1316	{
		1317	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1318	}
		1319
		1320	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		1321	{
		1322	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1088	}	1323	}
1089		1324
1090	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	1325	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1091	{	1326	{
1092	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1327	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
…		…
1213	return eio__1path (EIO_READDIR, path, pri, cb, data);	1448	return eio__1path (EIO_READDIR, path, pri, cb, data);
1214	}	1449	}
1215		1450
1216	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	1451	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1217	{	1452	{
1218	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	1453	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1219	}	1454	}
1220		1455
1221	static eio_req *	1456	static eio_req *
1222	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1457	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1223	{	1458	{
…		…
1247	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1482	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1248	{	1483	{
1249	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	1484	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1250	}	1485	}
1251		1486
		1487	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
		1488	{
		1489	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
		1490	}
		1491
		1492	#endif
		1493
1252	eio_req *eio_grp (eio_cb cb, void *data)	1494	eio_req *eio_grp (eio_cb cb, void *data)
1253	{	1495	{
1254	const int pri = EIO_PRI_MAX;	1496	const int pri = EIO_PRI_MAX;
1255		1497
1256	REQ (EIO_GROUP); SEND;	1498	REQ (EIO_GROUP); SEND;
…		…
1280		1522
1281	void eio_grp_add (eio_req *grp, eio_req *req)	1523	void eio_grp_add (eio_req *grp, eio_req *req)
1282	{	1524	{
1283	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	1525	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1284		1526
		1527	grp->flags |= EIO_FLAG_GROUPADD;
		1528
1285	++grp->size;	1529	++grp->size;
1286	req->grp = grp;	1530	req->grp = grp;
1287		1531
1288	req->grp_prev = 0;	1532	req->grp_prev = 0;
1289	req->grp_next = grp->grp_first;	1533	req->grp_next = grp->grp_first;
…		…
1297	/*****************************************************************************/	1541	/*****************************************************************************/
1298	/* misc garbage */	1542	/* misc garbage */
1299		1543
1300	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	1544	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1301	{	1545	{
1302	worker wrk;	1546	etp_worker wrk;
1303		1547
1304	wrk.dbuf = 0;	1548	wrk.dbuf = 0;
1305		1549
1306	eio__sendfile (ofd, ifd, offset, count, &wrk);	1550	eio__sendfile (ofd, ifd, offset, count, &wrk);
1307		1551

Diff Legend

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