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 1.32 by root, Sat Jun 6 18:06:55 2009 UTC

		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libeio@schmorp.de>
		5	* All rights reserved.
		6	*
		7	* Redistribution and use in source and binary forms, with or without modifica-
		8	* tion, are permitted provided that the following conditions are met:
		9	*
		10	* 1. Redistributions of source code must retain the above copyright notice,
		11	* this list of conditions and the following disclaimer.
		12	*
		13	* 2. Redistributions in binary form must reproduce the above copyright
		14	* notice, this list of conditions and the following disclaimer in the
		15	* documentation and/or other materials provided with the distribution.
		16	*
		17	* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
		18	* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
		19	* CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
		20	* EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
		21	* CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
		22	* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
		23	* OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
		24	* WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
		25	* ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
		26	* OF THE POSSIBILITY OF SUCH DAMAGE.
		27	*
		28	* Alternatively, the contents of this file may be used under the terms of
		29	* the GNU General Public License ("GPL") version 2 or any later version,
		30	* in which case the provisions of the GPL are applicable instead of
		31	* the above. If you wish to allow the use of your version of this file
		32	* only under the terms of the GPL and not to allow others to use your
		33	* version of this file under the BSD license, indicate your decision
		34	* by deleting the provisions above and replace them with the notice
		35	* and other provisions required by the GPL. If you do not delete the
		36	* provisions above, a recipient may use your version of this file under
		37	* either the BSD or the GPL.
		38	*/
		39
1	#include "eio.h"	40	#include "eio.h"
		41
		42	#ifdef EIO_STACKSIZE
		43	# define XTHREAD_STACKSIZE EIO_STACKSIZE
		44	#endif
2	#include "xthread.h"	45	#include "xthread.h"
3		46
4	#include <errno.h>	47	#include <errno.h>
5	#include <stddef.h>	48	#include <stddef.h>
6	#include <stdlib.h>	49	#include <stdlib.h>
…		…
25	#endif	68	#endif
26		69
27	#ifdef _WIN32	70	#ifdef _WIN32
28		71
29	/*doh*/	72	/*doh*/
30
31	#else	73	#else
32		74
33	# include "config.h"	75	# include "config.h"
34	# include <sys/time.h>	76	# include <sys/time.h>
35	# include <sys/select.h>	77	# include <sys/select.h>
36	# include <unistd.h>	78	# include <unistd.h>
37	# include <utime.h>	79	# include <utime.h>
38	# include <signal.h>	80	# include <signal.h>
39	# include <dirent.h>	81	# include <dirent.h>
		82
		83	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		84	# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
		85	# define D_INO(de) (de)->d_fileno
		86	# define _DIRENT_HAVE_D_TYPE /* sigh */
		87	# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
		88	# define D_INO(de) (de)->d_ino
		89	# endif
		90
		91	# ifdef _DIRENT_HAVE_D_TYPE
		92	# define D_TYPE(de) (de)->d_type
		93	# endif
40		94
41	# ifndef EIO_STRUCT_DIRENT	95	# ifndef EIO_STRUCT_DIRENT
42	# define EIO_STRUCT_DIRENT struct dirent	96	# define EIO_STRUCT_DIRENT struct dirent
43	# endif	97	# endif
44		98
…		…
57	# else	111	# else
58	# error sendfile support requested but not available	112	# error sendfile support requested but not available
59	# endif	113	# endif
60	#endif	114	#endif
61		115
		116	#ifndef D_TYPE
		117	# define D_TYPE(de) 0
		118	#endif
		119
		120	#ifndef D_INO
		121	# define D_INO(de) 0
		122	#endif
		123
62	/* number of seconds after which an idle threads exit */	124	/* number of seconds after which an idle threads exit */
63	#define IDLE_TIMEOUT 10	125	#define IDLE_TIMEOUT 10
64		126
65	/* used for struct dirent, AIX doesn't provide it */	127	/* used for struct dirent, AIX doesn't provide it */
66	#ifndef NAME_MAX	128	#ifndef NAME_MAX
…		…
70	/* buffer size for various temporary buffers */	132	/* buffer size for various temporary buffers */
71	#define EIO_BUFSIZE 65536	133	#define EIO_BUFSIZE 65536
72		134
73	#define dBUF \	135	#define dBUF \
74	char *eio_buf; \	136	char *eio_buf; \
75	X_LOCK (wrklock); \	137	ETP_WORKER_LOCK (self); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	138	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (wrklock); \	139	ETP_WORKER_UNLOCK (self); \
78	errno = ENOMEM; \	140	errno = ENOMEM; \
79	if (!eio_buf) \	141	if (!eio_buf) \
80	return -1;	142	return -1;
81		143
82	#define EIO_TICKS ((1000000 + 1023) >> 10)	144	#define EIO_TICKS ((1000000 + 1023) >> 10)
83		145
		146	/*****************************************************************************/
		147
		148	#if __GNUC__ >= 3
		149	# define expect(expr,value) __builtin_expect ((expr),(value))
		150	#else
		151	# define expect(expr,value) (expr)
		152	#endif
		153
		154	#define expect_false(expr) expect ((expr) != 0, 0)
		155	#define expect_true(expr) expect ((expr) != 0, 1)
		156
		157	/*****************************************************************************/
		158
		159	#define ETP_PRI_MIN EIO_PRI_MIN
		160	#define ETP_PRI_MAX EIO_PRI_MAX
		161
		162	struct etp_worker;
		163
		164	#define ETP_REQ eio_req
		165	#define ETP_DESTROY(req) eio_destroy (req)
		166	static int eio_finish (eio_req *req);
		167	#define ETP_FINISH(req) eio_finish (req)
		168	static void eio_execute (struct etp_worker *self, eio_req *req);
		169	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		170
		171	#define ETP_WORKER_CLEAR(req) \
		172	if (wrk->dbuf) \
		173	{ \
		174	free (wrk->dbuf); \
		175	wrk->dbuf = 0; \
		176	} \
		177	\
		178	if (wrk->dirp) \
		179	{ \
		180	closedir (wrk->dirp); \
		181	wrk->dirp = 0; \
		182	}
		183
		184	#define ETP_WORKER_COMMON \
		185	void *dbuf; \
		186	DIR *dirp;
		187
		188	/*****************************************************************************/
		189
		190	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		191
		192	/* calculcate time difference in ~1/EIO_TICKS of a second */
		193	static int tvdiff (struct timeval *tv1, struct timeval *tv2)
		194	{
		195	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
		196	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
		197	}
		198
		199	static unsigned int started, idle, wanted = 4;
		200
84	static void (*want_poll_cb) (void);	201	static void (*want_poll_cb) (void);
85	static void (*done_poll_cb) (void);	202	static void (*done_poll_cb) (void);
86		203
87	static unsigned int max_poll_time = 0;	204	static unsigned int max_poll_time; /* reslock */
88	static unsigned int max_poll_reqs = 0;	205	static unsigned int max_poll_reqs; /* reslock */
89		206
90	/* calculcate time difference in ~1/EIO_TICKS of a second */	207	static volatile unsigned int nreqs; /* reqlock */
91	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	208	static volatile unsigned int nready; /* reqlock */
92	{	209	static volatile unsigned int npending; /* reqlock */
93	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	210	static volatile unsigned int max_idle = 4;
94	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
95	}
96		211
97	static unsigned int started, idle, wanted = 4;
98
99	/* worker threads management */
100	static mutex_t wrklock = X_MUTEX_INIT;	212	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;	213	static mutex_t reslock = X_MUTEX_INIT;
144	static mutex_t reqlock = X_MUTEX_INIT;	214	static mutex_t reqlock = X_MUTEX_INIT;
145	static cond_t reqwait = X_COND_INIT;	215	static cond_t reqwait = X_COND_INIT;
146		216
		217	#if !HAVE_PREADWRITE
		218	/*
		219	* make our pread/pwrite emulation safe against themselves, but not against
		220	* normal read/write by using a mutex. slows down execution a lot,
		221	* but that's your problem, not mine.
		222	*/
		223	static mutex_t preadwritelock = X_MUTEX_INIT;
		224	#endif
		225
		226	typedef struct etp_worker
		227	{
		228	/* locked by wrklock */
		229	struct etp_worker *prev, *next;
		230
		231	thread_t tid;
		232
		233	/* locked by reslock, reqlock or wrklock */
		234	ETP_REQ *req; /* currently processed request */
		235
		236	ETP_WORKER_COMMON
		237	} etp_worker;
		238
		239	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
		240
		241	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		242	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
		243
		244	/* worker threads management */
		245
		246	static void etp_worker_clear (etp_worker *wrk)
		247	{
		248	ETP_WORKER_CLEAR (wrk);
		249	}
		250
		251	static void etp_worker_free (etp_worker *wrk)
		252	{
		253	wrk->next->prev = wrk->prev;
		254	wrk->prev->next = wrk->next;
		255
		256	free (wrk);
		257	}
		258
147	unsigned int eio_nreqs (void)	259	static unsigned int etp_nreqs (void)
148	{	260	{
		261	int retval;
		262	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		263	retval = nreqs;
		264	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
149	return nreqs;	265	return retval;
150	}	266	}
151		267
152	unsigned int eio_nready (void)	268	static unsigned int etp_nready (void)
153	{	269	{
154	unsigned int retval;	270	unsigned int retval;
155		271
156	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	272	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157	retval = nready;	273	retval = nready;
158	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	274	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159		275
160	return retval;	276	return retval;
161	}	277	}
162		278
163	unsigned int eio_npending (void)	279	static unsigned int etp_npending (void)
164	{	280	{
165	unsigned int retval;	281	unsigned int retval;
166		282
167	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	283	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168	retval = npending;	284	retval = npending;
169	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	285	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170		286
171	return retval;	287	return retval;
172	}	288	}
173		289
174	unsigned int eio_nthreads (void)	290	static unsigned int etp_nthreads (void)
175	{	291	{
176	unsigned int retval;	292	unsigned int retval;
177		293
178	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	294	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179	retval = started;	295	retval = started;
…		…
186	* a somewhat faster data structure might be nice, but	302	* a somewhat faster data structure might be nice, but
187	* with 8 priorities this actually needs <20 insns	303	* with 8 priorities this actually needs <20 insns
188	* per shift, the most expensive operation.	304	* per shift, the most expensive operation.
189	*/	305	*/
190	typedef struct {	306	typedef struct {
191	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	307	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
192	int size;	308	int size;
193	} reqq;	309	} etp_reqq;
194		310
195	static reqq req_queue;	311	static etp_reqq req_queue;
196	static reqq res_queue;	312	static etp_reqq res_queue;
197		313
198	static int reqq_push (reqq *q, eio_req *req)	314	static int reqq_push (etp_reqq *q, ETP_REQ *req)
199	{	315	{
200	int pri = req->pri;	316	int pri = req->pri;
201	req->next = 0;	317	req->next = 0;
202		318
203	if (q->qe[pri])	319	if (q->qe[pri])
…		…
209	q->qe[pri] = q->qs[pri] = req;	325	q->qe[pri] = q->qs[pri] = req;
210		326
211	return q->size++;	327	return q->size++;
212	}	328	}
213		329
214	static eio_req *reqq_shift (reqq *q)	330	static ETP_REQ *reqq_shift (etp_reqq *q)
215	{	331	{
216	int pri;	332	int pri;
217		333
218	if (!q->size)	334	if (!q->size)
219	return 0;	335	return 0;
220		336
221	--q->size;	337	--q->size;
222		338
223	for (pri = EIO_NUM_PRI; pri--; )	339	for (pri = ETP_NUM_PRI; pri--; )
224	{	340	{
225	eio_req *req = q->qs[pri];	341	eio_req *req = q->qs[pri];
226		342
227	if (req)	343	if (req)
228	{	344	{
…		…
234	}	350	}
235		351
236	abort ();	352	abort ();
237	}	353	}
238		354
239	static void grp_try_feed (eio_req *grp)	355	static void etp_atfork_prepare (void)
240	{	356	{
241	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	357	X_LOCK (wrklock);
		358	X_LOCK (reqlock);
		359	X_LOCK (reslock);
		360	#if !HAVE_PREADWRITE
		361	X_LOCK (preadwritelock);
		362	#endif
		363	}
		364
		365	static void etp_atfork_parent (void)
		366	{
		367	#if !HAVE_PREADWRITE
		368	X_UNLOCK (preadwritelock);
		369	#endif
		370	X_UNLOCK (reslock);
		371	X_UNLOCK (reqlock);
		372	X_UNLOCK (wrklock);
		373	}
		374
		375	static void etp_atfork_child (void)
		376	{
		377	ETP_REQ *prv;
		378
		379	while ((prv = reqq_shift (&req_queue)))
		380	ETP_DESTROY (prv);
		381
		382	while ((prv = reqq_shift (&res_queue)))
		383	ETP_DESTROY (prv);
		384
		385	while (wrk_first.next != &wrk_first)
242	{	386	{
243	int old_len = grp->size;	387	etp_worker *wrk = wrk_first.next;
244		388
245	EIO_FEED (grp);	389	if (wrk->req)
		390	ETP_DESTROY (wrk->req);
246		391
247	/* stop if no progress has been made */	392	etp_worker_clear (wrk);
248	if (old_len == grp->size)	393	etp_worker_free (wrk);
249	{
250	grp->feed = 0;
251	break;
252	}
253	}	394	}
254	}
255		395
256	static int eio_finish (eio_req *req);	396	started = 0;
		397	idle = 0;
		398	nreqs = 0;
		399	nready = 0;
		400	npending = 0;
257		401
258	static int grp_dec (eio_req *grp)	402	etp_atfork_parent ();
259	{	403	}
260	--grp->size;
261		404
262	/* call feeder, if applicable */	405	static void
263	grp_try_feed (grp);	406	etp_once_init (void)
		407	{
		408	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
		409	}
264		410
265	/* finish, if done */	411	static int
266	if (!grp->size && grp->int1)	412	etp_init (void (*want_poll)(void), void (*done_poll)(void))
267	return eio_finish (grp);	413	{
268	else	414	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
		415
		416	pthread_once (&doinit, etp_once_init);
		417
		418	want_poll_cb = want_poll;
		419	done_poll_cb = done_poll;
		420
269	return 0;	421	return 0;
270	}	422	}
271		423
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);	424	X_THREAD_PROC (etp_proc);
323		425
324	static void start_thread (void)	426	static void etp_start_thread (void)
325	{	427	{
326	worker *wrk = calloc (1, sizeof (worker));	428	etp_worker *wrk = calloc (1, sizeof (etp_worker));
327		429
328	/*TODO*/	430	/*TODO*/
329	assert (("unable to allocate worker thread data", wrk));	431	assert (("unable to allocate worker thread data", wrk));
330		432
331	X_LOCK (wrklock);	433	X_LOCK (wrklock);
332		434
333	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	435	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334	{	436	{
335	wrk->prev = &wrk_first;	437	wrk->prev = &wrk_first;
336	wrk->next = wrk_first.next;	438	wrk->next = wrk_first.next;
337	wrk_first.next->prev = wrk;	439	wrk_first.next->prev = wrk;
338	wrk_first.next = wrk;	440	wrk_first.next = wrk;
…		…
342	free (wrk);	444	free (wrk);
343		445
344	X_UNLOCK (wrklock);	446	X_UNLOCK (wrklock);
345	}	447	}
346		448
347	static void maybe_start_thread (void)	449	static void etp_maybe_start_thread (void)
348	{	450	{
349	if (eio_nthreads () >= wanted)	451	if (expect_true (etp_nthreads () >= wanted))
350	return;	452	return;
351		453
352	/* todo: maybe use idle here, but might be less exact */	454	/* todo: maybe use idle here, but might be less exact */
353	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	455	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354	return;	456	return;
355		457
356	start_thread ();	458	etp_start_thread ();
357	}	459	}
358		460
359	void eio_submit (eio_req *req)
360	{
361	++nreqs;
362
363	X_LOCK (reqlock);
364	++nready;
365	reqq_push (&req_queue, req);
366	X_COND_SIGNAL (reqwait);
367	X_UNLOCK (reqlock);
368
369	maybe_start_thread ();
370	}
371
372	static void end_thread (void)	461	static void etp_end_thread (void)
373	{	462	{
374	eio_req *req = calloc (1, sizeof (eio_req));	463	eio_req *req = calloc (1, sizeof (eio_req));
375		464
376	req->type = EIO_QUIT;	465	req->type = -1;
377	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	466	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
378		467
379	X_LOCK (reqlock);	468	X_LOCK (reqlock);
380	reqq_push (&req_queue, req);	469	reqq_push (&req_queue, req);
381	X_COND_SIGNAL (reqwait);	470	X_COND_SIGNAL (reqwait);
382	X_UNLOCK (reqlock);	471	X_UNLOCK (reqlock);
…		…
384	X_LOCK (wrklock);	473	X_LOCK (wrklock);
385	--started;	474	--started;
386	X_UNLOCK (wrklock);	475	X_UNLOCK (wrklock);
387	}	476	}
388		477
389	void eio_set_max_poll_time (double nseconds)
390	{
391	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
392	max_poll_time = nseconds;
393	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
394	}
395
396	void eio_set_max_poll_reqs (unsigned int maxreqs)
397	{
398	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
399	max_poll_reqs = maxreqs;
400	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
401	}
402
403	void eio_set_max_idle (unsigned int nthreads)
404	{
405	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
406	max_idle = nthreads <= 0 ? 1 : nthreads;
407	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
408	}
409
410	void eio_set_min_parallel (unsigned int nthreads)
411	{
412	if (wanted < nthreads)
413	wanted = nthreads;
414	}
415
416	void eio_set_max_parallel (unsigned int nthreads)
417	{
418	if (wanted > nthreads)
419	wanted = nthreads;
420
421	while (started > wanted)
422	end_thread ();
423	}
424
425	int eio_poll (void)	478	static int etp_poll (void)
426	{	479	{
427	int maxreqs = max_poll_reqs;	480	unsigned int maxreqs;
		481	unsigned int maxtime;
428	struct timeval tv_start, tv_now;	482	struct timeval tv_start, tv_now;
429	eio_req *req;
430		483
		484	X_LOCK (reslock);
		485	maxreqs = max_poll_reqs;
		486	maxtime = max_poll_time;
		487	X_UNLOCK (reslock);
		488
431	if (max_poll_time)	489	if (maxtime)
432	gettimeofday (&tv_start, 0);	490	gettimeofday (&tv_start, 0);
433		491
434	for (;;)	492	for (;;)
435	{	493	{
		494	ETP_REQ *req;
		495
436	maybe_start_thread ();	496	etp_maybe_start_thread ();
437		497
438	X_LOCK (reslock);	498	X_LOCK (reslock);
439	req = reqq_shift (&res_queue);	499	req = reqq_shift (&res_queue);
440		500
441	if (req)	501	if (req)
…		…
449	X_UNLOCK (reslock);	509	X_UNLOCK (reslock);
450		510
451	if (!req)	511	if (!req)
452	return 0;	512	return 0;
453		513
		514	X_LOCK (reqlock);
454	--nreqs;	515	--nreqs;
		516	X_UNLOCK (reqlock);
455		517
456	if (req->type == EIO_GROUP && req->size)	518	if (expect_false (req->type == EIO_GROUP && req->size))
457	{	519	{
458	req->int1 = 1; /* mark request as delayed */	520	req->int1 = 1; /* mark request as delayed */
459	continue;	521	continue;
460	}	522	}
461	else	523	else
462	{	524	{
463	int res = eio_finish (req);	525	int res = ETP_FINISH (req);
464	if (res)	526	if (expect_false (res))
465	return res;	527	return res;
466	}	528	}
467		529
468	if (maxreqs && !--maxreqs)	530	if (expect_false (maxreqs && !--maxreqs))
469	break;	531	break;
470		532
471	if (max_poll_time)	533	if (maxtime)
472	{	534	{
473	gettimeofday (&tv_now, 0);	535	gettimeofday (&tv_now, 0);
474		536
475	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	537	if (tvdiff (&tv_start, &tv_now) >= maxtime)
476	break;	538	break;
477	}	539	}
478	}	540	}
479		541
480	errno = EAGAIN;	542	errno = EAGAIN;
481	return -1;	543	return -1;
482	}	544	}
483		545
		546	static void etp_cancel (ETP_REQ *req)
		547	{
		548	X_LOCK (wrklock);
		549	req->flags |= EIO_FLAG_CANCELLED;
		550	X_UNLOCK (wrklock);
		551
		552	eio_grp_cancel (req);
		553	}
		554
		555	static void etp_submit (ETP_REQ *req)
		556	{
		557	req->pri -= ETP_PRI_MIN;
		558
		559	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		560	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		561
		562	if (expect_false (req->type == EIO_GROUP))
		563	{
		564	/* I hope this is worth it :/ */
		565	X_LOCK (reqlock);
		566	++nreqs;
		567	X_UNLOCK (reqlock);
		568
		569	X_LOCK (reslock);
		570
		571	++npending;
		572
		573	if (!reqq_push (&res_queue, req) && want_poll_cb)
		574	want_poll_cb ();
		575
		576	X_UNLOCK (reslock);
		577	}
		578	else
		579	{
		580	X_LOCK (reqlock);
		581	++nreqs;
		582	++nready;
		583	reqq_push (&req_queue, req);
		584	X_COND_SIGNAL (reqwait);
		585	X_UNLOCK (reqlock);
		586
		587	etp_maybe_start_thread ();
		588	}
		589	}
		590
		591	static void etp_set_max_poll_time (double nseconds)
		592	{
		593	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		594	max_poll_time = nseconds;
		595	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		596	}
		597
		598	static void etp_set_max_poll_reqs (unsigned int maxreqs)
		599	{
		600	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		601	max_poll_reqs = maxreqs;
		602	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		603	}
		604
		605	static void etp_set_max_idle (unsigned int nthreads)
		606	{
		607	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		608	max_idle = nthreads <= 0 ? 1 : nthreads;
		609	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		610	}
		611
		612	static void etp_set_min_parallel (unsigned int nthreads)
		613	{
		614	if (wanted < nthreads)
		615	wanted = nthreads;
		616	}
		617
		618	static void etp_set_max_parallel (unsigned int nthreads)
		619	{
		620	if (wanted > nthreads)
		621	wanted = nthreads;
		622
		623	while (started > wanted)
		624	etp_end_thread ();
		625	}
		626
		627	/*****************************************************************************/
		628
		629	static void grp_try_feed (eio_req *grp)
		630	{
		631	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		632	{
		633	grp->flags &= ~EIO_FLAG_GROUPADD;
		634
		635	EIO_FEED (grp);
		636
		637	/* stop if no progress has been made */
		638	if (!(grp->flags & EIO_FLAG_GROUPADD))
		639	{
		640	grp->feed = 0;
		641	break;
		642	}
		643	}
		644	}
		645
		646	static int grp_dec (eio_req *grp)
		647	{
		648	--grp->size;
		649
		650	/* call feeder, if applicable */
		651	grp_try_feed (grp);
		652
		653	/* finish, if done */
		654	if (!grp->size && grp->int1)
		655	return eio_finish (grp);
		656	else
		657	return 0;
		658	}
		659
		660	void eio_destroy (eio_req *req)
		661	{
		662	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		663	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		664
		665	EIO_DESTROY (req);
		666	}
		667
		668	static int eio_finish (eio_req *req)
		669	{
		670	int res = EIO_FINISH (req);
		671
		672	if (req->grp)
		673	{
		674	int res2;
		675	eio_req *grp = req->grp;
		676
		677	/* unlink request */
		678	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		679	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		680
		681	if (grp->grp_first == req)
		682	grp->grp_first = req->grp_next;
		683
		684	res2 = grp_dec (grp);
		685
		686	if (!res && res2)
		687	res = res2;
		688	}
		689
		690	eio_destroy (req);
		691
		692	return res;
		693	}
		694
		695	void eio_grp_cancel (eio_req *grp)
		696	{
		697	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		698	eio_cancel (grp);
		699	}
		700
		701	void eio_cancel (eio_req *req)
		702	{
		703	etp_cancel (req);
		704	}
		705
		706	void eio_submit (eio_req *req)
		707	{
		708	etp_submit (req);
		709	}
		710
		711	unsigned int eio_nreqs (void)
		712	{
		713	return etp_nreqs ();
		714	}
		715
		716	unsigned int eio_nready (void)
		717	{
		718	return etp_nready ();
		719	}
		720
		721	unsigned int eio_npending (void)
		722	{
		723	return etp_npending ();
		724	}
		725
		726	unsigned int eio_nthreads (void)
		727	{
		728	return etp_nthreads ();
		729	}
		730
		731	void eio_set_max_poll_time (double nseconds)
		732	{
		733	etp_set_max_poll_time (nseconds);
		734	}
		735
		736	void eio_set_max_poll_reqs (unsigned int maxreqs)
		737	{
		738	etp_set_max_poll_reqs (maxreqs);
		739	}
		740
		741	void eio_set_max_idle (unsigned int nthreads)
		742	{
		743	etp_set_max_idle (nthreads);
		744	}
		745
		746	void eio_set_min_parallel (unsigned int nthreads)
		747	{
		748	etp_set_min_parallel (nthreads);
		749	}
		750
		751	void eio_set_max_parallel (unsigned int nthreads)
		752	{
		753	etp_set_max_parallel (nthreads);
		754	}
		755
		756	int eio_poll (void)
		757	{
		758	return etp_poll ();
		759	}
		760
484	/*****************************************************************************/	761	/*****************************************************************************/
485	/* work around various missing functions */	762	/* work around various missing functions */
486		763
487	#if !HAVE_PREADWRITE	764	#if !HAVE_PREADWRITE
		765	# undef pread
		766	# undef pwrite
488	# define pread eio__pread	767	# define pread eio__pread
489	# define pwrite eio__pwrite	768	# 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		769
498	static ssize_t	770	static ssize_t
499	eio__pread (int fd, void *buf, size_t count, off_t offset)	771	eio__pread (int fd, void *buf, size_t count, off_t offset)
500	{	772	{
501	ssize_t res;	773	ssize_t res;
…		…
519		791
520	X_LOCK (preadwritelock);	792	X_LOCK (preadwritelock);
521	ooffset = lseek (fd, 0, SEEK_CUR);	793	ooffset = lseek (fd, 0, SEEK_CUR);
522	lseek (fd, offset, SEEK_SET);	794	lseek (fd, offset, SEEK_SET);
523	res = write (fd, buf, count);	795	res = write (fd, buf, count);
524	lseek (fd, offset, SEEK_SET);	796	lseek (fd, ooffset, SEEK_SET);
525	X_UNLOCK (preadwritelock);	797	X_UNLOCK (preadwritelock);
526		798
527	return res;	799	return res;
528	}	800	}
529	#endif	801	#endif
530		802
531	#ifndef HAVE_FUTIMES	803	#ifndef HAVE_FUTIMES
532		804
		805	# undef utimes
		806	# undef futimes
533	# define utimes(path,times) eio__utimes (path, times)	807	# define utimes(path,times) eio__utimes (path, times)
534	# define futimes(fd,times) eio__futimes (fd, times)	808	# define futimes(fd,times) eio__futimes (fd, times)
535		809
536	static int	810	static int
537	eio__utimes (const char *filename, const struct timeval times[2])	811	eio__utimes (const char *filename, const struct timeval times[2])
…		…
556	}	830	}
557		831
558	#endif	832	#endif
559		833
560	#if !HAVE_FDATASYNC	834	#if !HAVE_FDATASYNC
		835	# undef fdatasync
561	# define fdatasync fsync	836	# define fdatasync(fd) fsync (fd)
562	#endif	837	#endif
		838
		839	/* sync_file_range always needs emulation */
		840	int
		841	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		842	{
		843	#if HAVE_SYNC_FILE_RANGE
		844	int res;
		845
		846	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		847	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		848	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		849	{
		850	flags = 0
		851	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		852	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		853	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		854	}
		855
		856	res = sync_file_range (fd, offset, nbytes, flags);
		857
		858	if (!res || errno != ENOSYS)
		859	return res;
		860	#endif
		861
		862	/* even though we could play tricks with the flags, it's better to always
		863	* call fdatasync, as thta matches the expectation of it's users best */
		864	return fdatasync (fd);
		865	}
563		866
564	#if !HAVE_READAHEAD	867	#if !HAVE_READAHEAD
		868	# undef readahead
565	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	869	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566		870
567	static ssize_t	871	static ssize_t
568	eio__readahead (int fd, off_t offset, size_t count, worker *self)	872	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
569	{	873	{
570	size_t todo = count;	874	size_t todo = count;
571	dBUF;	875	dBUF;
572		876
573	while (todo > 0)	877	while (todo > 0)
…		…
583	return count;	887	return count;
584	}	888	}
585		889
586	#endif	890	#endif
587		891
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 */	892	/* sendfile always needs emulation */
620	static ssize_t	893	static ssize_t
621	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self)	894	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
622	{	895	{
623	ssize_t res;	896	ssize_t res;
624		897
625	if (!count)	898	if (!count)
626	return 0;	899	return 0;
…		…
708	}	981	}
709		982
710	return res;	983	return res;
711	}	984	}
712		985
		986	static int
		987	eio_dent_cmp (const void *a_, const void *b_)
		988	{
		989	const eio_dirent *a = (const eio_dirent *)a_;
		990	const eio_dirent *b = (const eio_dirent *)b_;
		991
		992	return (int)b->score - (int)a->score ? (int)b->score - (int)a->score
		993	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */
		994	}
		995
713	/* read a full directory */	996	/* read a full directory */
714	static void	997	static void
715	eio__scandir (eio_req *req, worker *self)	998	eio__scandir (eio_req *req, etp_worker *self)
716	{	999	{
717	DIR *dirp;	1000	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;	1001	EIO_STRUCT_DIRENT *entp;
724	char *name, *names;	1002	unsigned char *name, *names;
725	int memlen = 4096;	1003	int namesalloc = 4096;
726	int memofs = 0;	1004	int namesoffs = 0;
		1005	int flags = req->int1;
		1006	eio_dirent *dents = 0;
		1007	int dentalloc = 128;
727	int res = 0;	1008	int dentoffs = 0;
		1009
		1010	req->result = -1;
		1011
		1012	if (!(flags & EIO_READDIR_DENTS))
		1013	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
728		1014
729	X_LOCK (wrklock);	1015	X_LOCK (wrklock);
		1016	/* the corresponding closedir is in ETP_WORKER_CLEAR */
730	self->dirp = dirp = opendir (req->ptr1);	1017	self->dirp = dirp = opendir (req->ptr1);
731	self->dbuf = u = malloc (sizeof (*u));
732	req->flags |= EIO_FLAG_PTR2_FREE;	1018	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
733	req->ptr2 = names = malloc (memlen);	1019	req->ptr1 = names = malloc (namesalloc);
		1020	req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
734	X_UNLOCK (wrklock);	1021	X_UNLOCK (wrklock);
735		1022
736	if (dirp && u && names)	1023	if (dirp && names && (!flags || dents))
737	for (;;)	1024	for (;;)
738	{	1025	{
739	errno = 0;	1026	errno = 0;
740	readdir_r (dirp, &u->d, &entp);	1027	entp = readdir (dirp);
741		1028
742	if (!entp)	1029	if (!entp)
		1030	{
		1031	if (errno)
		1032	break;
		1033
		1034	/* sort etc. */
		1035	req->int1 = flags;
		1036	req->result = dentoffs;
		1037
		1038	if (dents)
		1039	{
		1040	eio_dirent *ent = dents + dentoffs;
		1041
		1042	while (ent > dents)
		1043	(--ent)->name = names + (size_t)ent->name;
		1044	}
		1045
		1046	if (flags & EIO_READDIR_STAT_ORDER
		1047	|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))
		1048	{
		1049	/* pray your qsort doesn't use quicksort */
		1050	qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
		1051	}
		1052	else if (flags & EIO_READDIR_DIRS_FIRST)
		1053	{
		1054	/* in this case, all is known, and we just put dirs first and sort them */
		1055	eio_dirent *ent = dents + dentoffs;
		1056	eio_dirent *dir = dents;
		1057
		1058	while (ent > dir)
		1059	{
		1060	if (dir->type == DT_DIR)
		1061	++dir;
		1062	else
		1063	{
		1064	--ent;
		1065
		1066	if (ent->type == DT_DIR)
		1067	{
		1068	eio_dirent tmp = *dir;
		1069	*dir = *ent;
		1070	*ent = tmp;
		1071
		1072	++dir;
		1073	}
		1074	}
		1075	}
		1076
		1077	/* now sort the dirs only */
		1078	qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
		1079	}
		1080
		1081	/* only provide the names array unless DENTS is specified */
		1082	if (!(flags & EIO_READDIR_DENTS))
		1083	{
		1084	X_LOCK (wrklock);
		1085	assert (!dents);
		1086	req->ptr1 = 0;
		1087	req->ptr2 = names;
		1088	X_UNLOCK (wrklock);
		1089	}
		1090
743	break;	1091	break;
		1092	}
744		1093
		1094	/* now add the entry to our list(s) */
745	name = entp->d_name;	1095	name = entp->d_name;
746		1096
		1097	/* skip . and .. entries */
747	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1098	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
748	{	1099	{
749	int len = strlen (name) + 1;	1100	int len = strlen (name) + 1;
750		1101
751	res++;	1102	while (expect_false (namesoffs + len > namesalloc))
752
753	while (memofs + len > memlen)
754	{	1103	{
755	memlen *= 2;	1104	namesalloc *= 2;
756	X_LOCK (wrklock);	1105	X_LOCK (wrklock);
757	req->ptr2 = names = realloc (names, memlen);	1106	req->ptr1 = names = realloc (names, namesalloc);
758	X_UNLOCK (wrklock);	1107	X_UNLOCK (wrklock);
759		1108
760	if (!names)	1109	if (!names)
761	break;	1110	break;
762	}	1111	}
763		1112
764	memcpy (names + memofs, name, len);	1113	memcpy (names + namesoffs, name, len);
		1114
		1115	if (dents)
		1116	{
		1117	struct eio_dirent *ent;
		1118
		1119	if (expect_false (dentoffs == dentalloc))
		1120	{
		1121	dentalloc *= 2;
		1122	X_LOCK (wrklock);
		1123	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1124	X_UNLOCK (wrklock);
		1125
		1126	if (!dents)
		1127	break;
		1128	}
		1129
		1130	ent = dents + dentoffs;
		1131
		1132	ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */
		1133	ent->namelen = len - 1;
		1134	ent->inode = D_INO (entp);
		1135
		1136	switch (D_TYPE (entp))
		1137	{
		1138	default:
		1139	ent->type = EIO_DT_UNKNOWN;
		1140	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1141	break;
		1142
		1143	#ifdef DT_FIFO
		1144	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1145	#endif
		1146	#ifdef DT_CHR
		1147	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1148	#endif
		1149	#ifdef DT_DIR
		1150	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1151	#endif
		1152	#ifdef DT_BLK
		1153	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1154	#endif
		1155	#ifdef DT_REG
		1156	case DT_REG: ent->type = EIO_DT_REG; break;
		1157	#endif
		1158	#ifdef DT_LNK
		1159	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1160	#endif
		1161	#ifdef DT_SOCK
		1162	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1163	#endif
		1164	#ifdef DT_WHT
		1165	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1166	#endif
		1167	}
		1168
		1169	ent->score = 0;
		1170
		1171	if (flags & EIO_READDIR_DIRS_FIRST)
		1172	{
		1173	if (ent->type == EIO_DT_UNKNOWN)
		1174	{
		1175	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1176	ent->score = 98;
		1177	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1178	ent->score = len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */
		1179	}
		1180	else if (ent->type == DT_DIR)
		1181	ent->score = 100;
		1182	}
		1183	}
		1184
765	memofs += len;	1185	namesoffs += len;
		1186	++dentoffs;
766	}	1187	}
767	}	1188	}
768		1189	else
769	if (errno)
770	res = -1;
771
772	req->result = res;	1190	req->result = -1;
		1191	}
		1192
		1193	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1194	# undef msync
		1195	# define msync(a,b,c) ((errno = ENOSYS), -1)
		1196	#endif
		1197
		1198	int
		1199	eio__mtouch (void *mem, size_t len, int flags)
		1200	{
		1201	intptr_t addr = (intptr_t)mem;
		1202	intptr_t end = addr + len;
		1203	#ifdef PAGESIZE
		1204	const intptr_t page = PAGESIZE;
		1205	#else
		1206	static intptr_t page;
		1207
		1208	if (!page)
		1209	page = sysconf (_SC_PAGESIZE);
		1210	#endif
		1211
		1212	addr &= ~(page - 1); /* assume page size is always a power of two */
		1213
		1214	if (addr < end)
		1215	if (flags) /* modify */
		1216	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
		1217	else
		1218	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
		1219
		1220	return 0;
773	}	1221	}
774		1222
775	/*****************************************************************************/	1223	/*****************************************************************************/
776		1224
777	#define ALLOC(len) \	1225	#define ALLOC(len) \
…		…
787	req->result = -1; \	1235	req->result = -1; \
788	break; \	1236	break; \
789	} \	1237	} \
790	}	1238	}
791		1239
792	X_THREAD_PROC (eio_proc)	1240	X_THREAD_PROC (etp_proc)
793	{	1241	{
794	eio_req *req;	1242	ETP_REQ *req;
795	struct timespec ts;	1243	struct timespec ts;
796	worker *self = (worker *)thr_arg;	1244	etp_worker *self = (etp_worker *)thr_arg;
797		1245
798	/* try to distribute timeouts somewhat randomly */	1246	/* try to distribute timeouts somewhat randomly */
799	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1247	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
800		1248
801	for (;;)	1249	for (;;)
802	{	1250	{
803	ts.tv_sec = time (0) + IDLE_TIMEOUT;
804
805	X_LOCK (reqlock);	1251	X_LOCK (reqlock);
806		1252
807	for (;;)	1253	for (;;)
808	{	1254	{
809	self->req = req = reqq_shift (&req_queue);	1255	self->req = req = reqq_shift (&req_queue);
…		…
811	if (req)	1257	if (req)
812	break;	1258	break;
813		1259
814	++idle;	1260	++idle;
815		1261
		1262	ts.tv_sec = time (0) + IDLE_TIMEOUT;
816	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1263	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
817	{	1264	{
818	if (idle > max_idle)	1265	if (idle > max_idle)
819	{	1266	{
820	--idle;	1267	--idle;
…		…
825	goto quit;	1272	goto quit;
826	}	1273	}
827		1274
828	/* we are allowed to idle, so do so without any timeout */	1275	/* we are allowed to idle, so do so without any timeout */
829	X_COND_WAIT (reqwait, reqlock);	1276	X_COND_WAIT (reqwait, reqlock);
830	ts.tv_sec = time (0) + IDLE_TIMEOUT;
831	}	1277	}
832		1278
833	--idle;	1279	--idle;
834	}	1280	}
835		1281
836	--nready;	1282	--nready;
837		1283
838	X_UNLOCK (reqlock);	1284	X_UNLOCK (reqlock);
839		1285
840	errno = 0; /* strictly unnecessary */	1286	if (req->type < 0)
		1287	goto quit;
841		1288
842	if (!EIO_CANCELLED (req))	1289	if (!EIO_CANCELLED (req))
843	switch (req->type)	1290	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		1291
944	X_LOCK (reslock);	1292	X_LOCK (reslock);
945		1293
946	++npending;	1294	++npending;
947		1295
948	if (!reqq_push (&res_queue, req) && want_poll_cb)	1296	if (!reqq_push (&res_queue, req) && want_poll_cb)
949	want_poll_cb ();	1297	want_poll_cb ();
950		1298
951	self->req = 0;	1299	self->req = 0;
952	worker_clear (self);	1300	etp_worker_clear (self);
953		1301
954	X_UNLOCK (reslock);	1302	X_UNLOCK (reslock);
955	}	1303	}
956		1304
957	quit:	1305	quit:
958	X_LOCK (wrklock);	1306	X_LOCK (wrklock);
959	worker_free (self);	1307	etp_worker_free (self);
960	X_UNLOCK (wrklock);	1308	X_UNLOCK (wrklock);
961		1309
962	return 0;	1310	return 0;
963	}	1311	}
964		1312
965	/*****************************************************************************/	1313	/*****************************************************************************/
966		1314
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))	1315	int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1024	{	1316	{
1025	want_poll_cb = want_poll;	1317	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	}	1318	}
1041		1319
1042	static void eio_api_destroy (eio_req *req)	1320	static void eio_api_destroy (eio_req *req)
1043	{	1321	{
1044	free (req);	1322	free (req);
…		…
1049	\	1327	\
1050	req = (eio_req *)calloc (1, sizeof *req); \	1328	req = (eio_req *)calloc (1, sizeof *req); \
1051	if (!req) \	1329	if (!req) \
1052	return 0; \	1330	return 0; \
1053	\	1331	\
1054	req->type = rtype; \	1332	req->type = rtype; \
1055	req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \	1333	req->pri = pri; \
1056	req->finish = cb; \	1334	req->finish = cb; \
		1335	req->data = data; \
1057	req->destroy = eio_api_destroy;	1336	req->destroy = eio_api_destroy;
1058		1337
1059	#define SEND eio_submit (req); return req	1338	#define SEND eio_submit (req); return req
1060		1339
1061	#define PATH \	1340	#define PATH \
…		…
1065	{ \	1344	{ \
1066	eio_api_destroy (req); \	1345	eio_api_destroy (req); \
1067	return 0; \	1346	return 0; \
1068	}	1347	}
1069		1348
		1349	static void eio_execute (etp_worker *self, eio_req *req)
		1350	{
		1351	errno = 0;
		1352
		1353	switch (req->type)
		1354	{
		1355	case EIO_READ: ALLOC (req->size);
		1356	req->result = req->offs >= 0
		1357	? pread (req->int1, req->ptr2, req->size, req->offs)
		1358	: read (req->int1, req->ptr2, req->size); break;
		1359	case EIO_WRITE: req->result = req->offs >= 0
		1360	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1361	: write (req->int1, req->ptr2, req->size); break;
		1362
		1363	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		1364	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
		1365
		1366	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1367	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1368	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1369	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1370	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1371	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1372
		1373	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1374	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1375	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1376	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1377	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1378	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1379
		1380	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1381	case EIO_CLOSE: req->result = close (req->int1); break;
		1382	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1383	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1384	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1385	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1386	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1387	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1388	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1389	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
		1390
		1391	case EIO_READLINK: ALLOC (NAME_MAX);
		1392	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
		1393
		1394	case EIO_SYNC: req->result = 0; sync (); break;
		1395	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1396	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1397	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
		1398	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1399	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1400
		1401	case EIO_READDIR: eio__scandir (req, self); break;
		1402
		1403	case EIO_BUSY:
		1404	#ifdef _WIN32
		1405	Sleep (req->nv1 * 1000.);
		1406	#else
		1407	{
		1408	struct timeval tv;
		1409
		1410	tv.tv_sec = req->nv1;
		1411	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
		1412
		1413	req->result = select (0, 0, 0, 0, &tv);
		1414	}
		1415	#endif
		1416	break;
		1417
		1418	case EIO_UTIME:
		1419	case EIO_FUTIME:
		1420	{
		1421	struct timeval tv[2];
		1422	struct timeval *times;
		1423
		1424	if (req->nv1 != -1. || req->nv2 != -1.)
		1425	{
		1426	tv[0].tv_sec = req->nv1;
		1427	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		1428	tv[1].tv_sec = req->nv2;
		1429	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		1430
		1431	times = tv;
		1432	}
		1433	else
		1434	times = 0;
		1435
		1436
		1437	req->result = req->type == EIO_FUTIME
		1438	? futimes (req->int1, times)
		1439	: utimes (req->ptr1, times);
		1440	}
		1441	break;
		1442
		1443	case EIO_GROUP:
		1444	abort (); /* handled in eio_request */
		1445
		1446	case EIO_NOP:
		1447	req->result = 0;
		1448	break;
		1449
		1450	case EIO_CUSTOM:
		1451	((void (*)(eio_req *))req->feed) (req);
		1452	break;
		1453
		1454	default:
		1455	req->result = -1;
		1456	break;
		1457	}
		1458
		1459	req->errorno = errno;
		1460	}
		1461
		1462	#ifndef EIO_NO_WRAPPERS
		1463
1070	eio_req *eio_nop (int pri, eio_cb cb, void *data)	1464	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1071	{	1465	{
1072	REQ (EIO_NOP); SEND;	1466	REQ (EIO_NOP); SEND;
1073	}	1467	}
1074		1468
…		…
1083	}	1477	}
1084		1478
1085	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	1479	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1086	{	1480	{
1087	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1481	REQ (EIO_FSYNC); req->int1 = fd; SEND;
		1482	}
		1483
		1484	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1485	{
		1486	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1487	}
		1488
		1489	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1490	{
		1491	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1492	}
		1493
		1494	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		1495	{
		1496	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1088	}	1497	}
1089		1498
1090	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	1499	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1091	{	1500	{
1092	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1501	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
…		…
1206	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	1615	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1207	{	1616	{
1208	return eio__1path (EIO_RMDIR, path, pri, cb, data);	1617	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1209	}	1618	}
1210		1619
1211	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	1620	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1212	{	1621	{
1213	return eio__1path (EIO_READDIR, path, pri, cb, data);	1622	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1214	}	1623	}
1215		1624
1216	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	1625	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1217	{	1626	{
1218	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	1627	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1219	}	1628	}
1220		1629
1221	static eio_req *	1630	static eio_req *
1222	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1631	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1223	{	1632	{
…		…
1247	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1656	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1248	{	1657	{
1249	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	1658	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1250	}	1659	}
1251		1660
		1661	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
		1662	{
		1663	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
		1664	}
		1665
		1666	#endif
		1667
1252	eio_req *eio_grp (eio_cb cb, void *data)	1668	eio_req *eio_grp (eio_cb cb, void *data)
1253	{	1669	{
1254	const int pri = EIO_PRI_MAX;	1670	const int pri = EIO_PRI_MAX;
1255		1671
1256	REQ (EIO_GROUP); SEND;	1672	REQ (EIO_GROUP); SEND;
…		…
1280		1696
1281	void eio_grp_add (eio_req *grp, eio_req *req)	1697	void eio_grp_add (eio_req *grp, eio_req *req)
1282	{	1698	{
1283	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	1699	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1284		1700
		1701	grp->flags |= EIO_FLAG_GROUPADD;
		1702
1285	++grp->size;	1703	++grp->size;
1286	req->grp = grp;	1704	req->grp = grp;
1287		1705
1288	req->grp_prev = 0;	1706	req->grp_prev = 0;
1289	req->grp_next = grp->grp_first;	1707	req->grp_next = grp->grp_first;
…		…
1297	/*****************************************************************************/	1715	/*****************************************************************************/
1298	/* misc garbage */	1716	/* misc garbage */
1299		1717
1300	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	1718	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1301	{	1719	{
1302	worker wrk;	1720	etp_worker wrk;
1303		1721
1304	wrk.dbuf = 0;	1722	wrk.dbuf = 0;
1305		1723
1306	eio__sendfile (ofd, ifd, offset, count, &wrk);	1724	eio__sendfile (ofd, ifd, offset, count, &wrk);
1307		1725

Diff Legend

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