ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.13 by root, Tue May 13 19:34:11 2008 UTC vs.
Revision 1.53 by root, Sat Jan 9 10:03:09 2010 UTC

		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008,2009,2010 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>
7	#include <string.h>	50	#include <string.h>
8	#include <errno.h>	51	#include <errno.h>
9	#include <sys/types.h>	52	#include <sys/types.h>
10	#include <sys/stat.h>	53	#include <sys/stat.h>
		54	#include <sys/statvfs.h>
11	#include <limits.h>	55	#include <limits.h>
12	#include <fcntl.h>	56	#include <fcntl.h>
13	#include <assert.h>	57	#include <assert.h>
14		58
15	#ifndef EIO_FINISH	59	#ifndef EIO_FINISH
…		…
25	#endif	69	#endif
26		70
27	#ifdef _WIN32	71	#ifdef _WIN32
28		72
29	/*doh*/	73	/*doh*/
30
31	#else	74	#else
32		75
33	# include "config.h"	76	# include "config.h"
34	# include <sys/time.h>	77	# include <sys/time.h>
35	# include <sys/select.h>	78	# include <sys/select.h>
		79	# include <sys/mman.h>
36	# include <unistd.h>	80	# include <unistd.h>
37	# include <utime.h>	81	# include <utime.h>
38	# include <signal.h>	82	# include <signal.h>
39	# include <dirent.h>	83	# include <dirent.h>
40		84
		85	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		86	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		87	# define _DIRENT_HAVE_D_TYPE /* sigh */
		88	# define D_INO(de) (de)->d_fileno
		89	# define D_NAMLEN(de) (de)->d_namlen
		90	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		91	# define D_INO(de) (de)->d_ino
		92	# endif
		93
		94	#ifdef _D_EXACT_NAMLEN
		95	# undef D_NAMLEN
		96	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		97	#endif
		98
		99	# ifdef _DIRENT_HAVE_D_TYPE
		100	# define D_TYPE(de) (de)->d_type
		101	# endif
		102
41	# ifndef EIO_STRUCT_DIRENT	103	# ifndef EIO_STRUCT_DIRENT
42	# define EIO_STRUCT_DIRENT struct dirent	104	# define EIO_STRUCT_DIRENT struct dirent
43	# endif	105	# endif
44		106
45	#endif	107	#endif
46		108
47	#if HAVE_SENDFILE	109	#if HAVE_SENDFILE
48	# if __linux	110	# if __linux
49	# include <sys/sendfile.h>	111	# include <sys/sendfile.h>
50	# elif __freebsd	112	# elif __FreeBSD__ || defined __APPLE__
51	# include <sys/socket.h>	113	# include <sys/socket.h>
52	# include <sys/uio.h>	114	# include <sys/uio.h>
53	# elif __hpux	115	# elif __hpux
54	# include <sys/socket.h>	116	# include <sys/socket.h>
55	# elif __solaris /* not yet */	117	# elif __solaris
56	# include <sys/sendfile.h>	118	# include <sys/sendfile.h>
57	# else	119	# else
58	# error sendfile support requested but not available	120	# error sendfile support requested but not available
59	# endif	121	# endif
60	#endif	122	#endif
61		123
		124	#ifndef D_TYPE
		125	# define D_TYPE(de) 0
		126	#endif
		127	#ifndef D_INO
		128	# define D_INO(de) 0
		129	#endif
		130	#ifndef D_NAMLEN
		131	# define D_NAMLEN(de) strlen ((de)->d_name)
		132	#endif
		133
62	/* number of seconds after which an idle threads exit */	134	/* number of seconds after which an idle threads exit */
63	#define IDLE_TIMEOUT 10	135	#define IDLE_TIMEOUT 10
64		136
65	/* used for struct dirent, AIX doesn't provide it */	137	/* used for struct dirent, AIX doesn't provide it */
66	#ifndef NAME_MAX	138	#ifndef NAME_MAX
…		…
70	/* buffer size for various temporary buffers */	142	/* buffer size for various temporary buffers */
71	#define EIO_BUFSIZE 65536	143	#define EIO_BUFSIZE 65536
72		144
73	#define dBUF \	145	#define dBUF \
74	char *eio_buf; \	146	char *eio_buf; \
75	X_LOCK (etplock); \	147	ETP_WORKER_LOCK (self); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	148	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (etplock); \	149	ETP_WORKER_UNLOCK (self); \
78	errno = ENOMEM; \	150	errno = ENOMEM; \
79	if (!eio_buf) \	151	if (!eio_buf) \
80	return -1;	152	return -1;
81		153
82	#define EIO_TICKS ((1000000 + 1023) >> 10)	154	#define EIO_TICKS ((1000000 + 1023) >> 10)
83		155
84	/*****************************************************************************/	156	/*****************************************************************************/
85		157
		158	#if __GNUC__ >= 3
		159	# define expect(expr,value) __builtin_expect ((expr),(value))
		160	#else
		161	# define expect(expr,value) (expr)
		162	#endif
		163
		164	#define expect_false(expr) expect ((expr) != 0, 0)
		165	#define expect_true(expr) expect ((expr) != 0, 1)
		166
		167	/*****************************************************************************/
		168
		169	#define ETP_PRI_MIN EIO_PRI_MIN
		170	#define ETP_PRI_MAX EIO_PRI_MAX
		171
		172	struct etp_worker;
		173
		174	#define ETP_REQ eio_req
		175	#define ETP_DESTROY(req) eio_destroy (req)
		176	static int eio_finish (eio_req *req);
		177	#define ETP_FINISH(req) eio_finish (req)
		178	static void eio_execute (struct etp_worker *self, eio_req *req);
		179	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		180
		181	#define ETP_WORKER_CLEAR(req) \
		182	if (wrk->dbuf) \
		183	{ \
		184	free (wrk->dbuf); \
		185	wrk->dbuf = 0; \
		186	} \
		187	\
		188	if (wrk->dirp) \
		189	{ \
		190	closedir (wrk->dirp); \
		191	wrk->dirp = 0; \
		192	}
		193
		194	#define ETP_WORKER_COMMON \
		195	void *dbuf; \
		196	DIR *dirp;
		197
		198	/*****************************************************************************/
		199
		200	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		201
86	/* calculcate time difference in ~1/EIO_TICKS of a second */	202	/* calculate time difference in ~1/EIO_TICKS of a second */
87	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	203	static int tvdiff (struct timeval *tv1, struct timeval *tv2)
88	{	204	{
89	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	205	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
90	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	206	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
91	}	207	}
92		208
93	static unsigned int started, idle, wanted = 4;	209	static unsigned int started, idle, wanted = 4;
94		210
95	typedef struct etp_pool
96	{
97	void (*want_poll_cb) (void);	211	static void (*want_poll_cb) (void);
98	void (*done_poll_cb) (void);	212	static void (*done_poll_cb) (void);
99		213
100	unsigned int max_poll_time;	214	static unsigned int max_poll_time; /* reslock */
101	unsigned int max_poll_reqs;	215	static unsigned int max_poll_reqs; /* reslock */
102	} etp_pool;
103		216
104	static volatile unsigned int nreqs, nready, npending;	217	static volatile unsigned int nreqs; /* reqlock */
		218	static volatile unsigned int nready; /* reqlock */
		219	static volatile unsigned int npending; /* reqlock */
105	static volatile unsigned int max_idle = 4;	220	static volatile unsigned int max_idle = 4;
106		221
107	static mutex_t etplock = X_MUTEX_INIT;	222	static mutex_t wrklock = X_MUTEX_INIT;
108	static mutex_t reslock = X_MUTEX_INIT;	223	static mutex_t reslock = X_MUTEX_INIT;
109	static mutex_t reqlock = X_MUTEX_INIT;	224	static mutex_t reqlock = X_MUTEX_INIT;
110	static cond_t reqwait = X_COND_INIT;	225	static cond_t reqwait = X_COND_INIT;
111		226
		227	#if !HAVE_PREADWRITE
		228	/*
		229	* make our pread/pwrite emulation safe against themselves, but not against
		230	* normal read/write by using a mutex. slows down execution a lot,
		231	* but that's your problem, not mine.
		232	*/
		233	static mutex_t preadwritelock = X_MUTEX_INIT;
		234	#endif
		235
112	typedef struct worker	236	typedef struct etp_worker
113	{	237	{
114	/* locked by etplock */	238	/* locked by wrklock */
115	struct worker *prev, *next;	239	struct etp_worker *prev, *next;
116		240
117	thread_t tid;	241	thread_t tid;
118		242
119	/* locked by reslock, reqlock or etplock */	243	/* locked by reslock, reqlock or wrklock */
120	eio_req *req; /* currently processed request */	244	ETP_REQ *req; /* currently processed request */
121	void *dbuf;	245
122	DIR *dirp;	246	ETP_WORKER_COMMON
123	} worker;	247	} etp_worker;
124		248
125	static worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	249	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
		250
		251	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		252	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
126		253
127	/* worker threads management */	254	/* worker threads management */
128		255
129	static void worker_clear (worker *wrk)	256	static void etp_worker_clear (etp_worker *wrk)
130	{	257	{
131	if (wrk->dirp)	258	ETP_WORKER_CLEAR (wrk);
132	{
133	closedir (wrk->dirp);
134	wrk->dirp = 0;
135	}
136
137	if (wrk->dbuf)
138	{
139	free (wrk->dbuf);
140	wrk->dbuf = 0;
141	}
142	}	259	}
143		260
144	static void worker_free (worker *wrk)	261	static void etp_worker_free (etp_worker *wrk)
145	{	262	{
146	wrk->next->prev = wrk->prev;	263	wrk->next->prev = wrk->prev;
147	wrk->prev->next = wrk->next;	264	wrk->prev->next = wrk->next;
148		265
149	free (wrk);	266	free (wrk);
150	}	267	}
151		268
152	unsigned int eio_nreqs (void)	269	static unsigned int etp_nreqs (void)
153	{	270	{
		271	int retval;
		272	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		273	retval = nreqs;
		274	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154	return nreqs;	275	return retval;
155	}	276	}
156		277
157	unsigned int eio_nready (void)	278	static unsigned int etp_nready (void)
158	{	279	{
159	unsigned int retval;	280	unsigned int retval;
160		281
161	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	282	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162	retval = nready;	283	retval = nready;
163	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	284	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164		285
165	return retval;	286	return retval;
166	}	287	}
167		288
168	unsigned int eio_npending (void)	289	static unsigned int etp_npending (void)
169	{	290	{
170	unsigned int retval;	291	unsigned int retval;
171		292
172	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	293	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173	retval = npending;	294	retval = npending;
174	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	295	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175		296
176	return retval;	297	return retval;
177	}	298	}
178		299
179	unsigned int eio_nthreads (void)	300	static unsigned int etp_nthreads (void)
180	{	301	{
181	unsigned int retval;	302	unsigned int retval;
182		303
183	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	304	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184	retval = started;	305	retval = started;
…		…
191	* a somewhat faster data structure might be nice, but	312	* a somewhat faster data structure might be nice, but
192	* with 8 priorities this actually needs <20 insns	313	* with 8 priorities this actually needs <20 insns
193	* per shift, the most expensive operation.	314	* per shift, the most expensive operation.
194	*/	315	*/
195	typedef struct {	316	typedef struct {
196	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	317	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197	int size;	318	int size;
198	} reqq;	319	} etp_reqq;
199		320
200	static reqq req_queue;	321	static etp_reqq req_queue;
201	static reqq res_queue;	322	static etp_reqq res_queue;
202		323
203	static int reqq_push (reqq *q, eio_req *req)	324	static int reqq_push (etp_reqq *q, ETP_REQ *req)
204	{	325	{
205	int pri = req->pri;	326	int pri = req->pri;
206	req->next = 0;	327	req->next = 0;
207		328
208	if (q->qe[pri])	329	if (q->qe[pri])
…		…
214	q->qe[pri] = q->qs[pri] = req;	335	q->qe[pri] = q->qs[pri] = req;
215		336
216	return q->size++;	337	return q->size++;
217	}	338	}
218		339
219	static eio_req *reqq_shift (reqq *q)	340	static ETP_REQ *reqq_shift (etp_reqq *q)
220	{	341	{
221	int pri;	342	int pri;
222		343
223	if (!q->size)	344	if (!q->size)
224	return 0;	345	return 0;
225		346
226	--q->size;	347	--q->size;
227		348
228	for (pri = EIO_NUM_PRI; pri--; )	349	for (pri = ETP_NUM_PRI; pri--; )
229	{	350	{
230	eio_req *req = q->qs[pri];	351	eio_req *req = q->qs[pri];
231		352
232	if (req)	353	if (req)
233	{	354	{
…		…
241	abort ();	362	abort ();
242	}	363	}
243		364
244	static void etp_atfork_prepare (void)	365	static void etp_atfork_prepare (void)
245	{	366	{
246	X_LOCK (etplock);	367	X_LOCK (wrklock);
247	X_LOCK (reqlock);	368	X_LOCK (reqlock);
248	X_LOCK (reslock);	369	X_LOCK (reslock);
249	#if !HAVE_PREADWRITE	370	#if !HAVE_PREADWRITE
250	X_LOCK (preadwritelock);	371	X_LOCK (preadwritelock);
251	#endif	372	#endif
252	#if !HAVE_READDIR_R
253	X_LOCK (readdirlock);
254	#endif
255	}	373	}
256		374
257	static void etp_atfork_parent (void)	375	static void etp_atfork_parent (void)
258	{	376	{
259	#if !HAVE_READDIR_R
260	X_UNLOCK (readdirlock);
261	#endif
262	#if !HAVE_PREADWRITE	377	#if !HAVE_PREADWRITE
263	X_UNLOCK (preadwritelock);	378	X_UNLOCK (preadwritelock);
264	#endif	379	#endif
265	X_UNLOCK (reslock);	380	X_UNLOCK (reslock);
266	X_UNLOCK (reqlock);	381	X_UNLOCK (reqlock);
267	X_UNLOCK (etplock);	382	X_UNLOCK (wrklock);
268	}	383	}
269		384
270	static void etp_atfork_child (void)	385	static void etp_atfork_child (void)
271	{	386	{
272	eio_req *prv;	387	ETP_REQ *prv;
273		388
274	while (prv = reqq_shift (&req_queue))	389	while ((prv = reqq_shift (&req_queue)))
275	eio_destroy (prv);	390	ETP_DESTROY (prv);
276		391
277	while (prv = reqq_shift (&res_queue))	392	while ((prv = reqq_shift (&res_queue)))
278	eio_destroy (prv);	393	ETP_DESTROY (prv);
279		394
280	while (wrk_first.next != &wrk_first)	395	while (wrk_first.next != &wrk_first)
281	{	396	{
282	worker *wrk = wrk_first.next;	397	etp_worker *wrk = wrk_first.next;
283		398
284	if (wrk->req)	399	if (wrk->req)
285	eio_destroy (wrk->req);	400	ETP_DESTROY (wrk->req);
286		401
287	worker_clear (wrk);	402	etp_worker_clear (wrk);
288	worker_free (wrk);	403	etp_worker_free (wrk);
289	}	404	}
290		405
291	started = 0;	406	started = 0;
292	idle = 0;	407	idle = 0;
293	nreqs = 0;	408	nreqs = 0;
…		…
302	{	417	{
303	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);	418	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
304	}	419	}
305		420
306	static int	421	static int
307	etp_init (etp_pool *etp, void (*want_poll)(void), void (*done_poll)(void))	422	etp_init (void (*want_poll)(void), void (*done_poll)(void))
308	{	423	{
309	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	424	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
310		425
311	pthread_once (&doinit, etp_once_init);	426	pthread_once (&doinit, etp_once_init);
312		427
313	memset (etp, 0, sizeof *etp);
314
315	etp->want_poll_cb = want_poll;	428	want_poll_cb = want_poll;
316	etp->done_poll_cb = done_poll;	429	done_poll_cb = done_poll;
317	}
318		430
319	static etp_pool etp;
320
321	/*****************************************************************************/
322
323	static void grp_try_feed (eio_req *grp)
324	{
325	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
326	{
327	int old_len = grp->size;
328
329	EIO_FEED (grp);
330
331	/* stop if no progress has been made */
332	if (old_len == grp->size)
333	{
334	grp->feed = 0;
335	break;
336	}
337	}
338	}
339
340	static int eio_finish (eio_req *req);
341
342	static int grp_dec (eio_req *grp)
343	{
344	--grp->size;
345
346	/* call feeder, if applicable */
347	grp_try_feed (grp);
348
349	/* finish, if done */
350	if (!grp->size && grp->int1)
351	return eio_finish (grp);
352	else
353	return 0;	431	return 0;
354	}	432	}
355		433
356	void eio_destroy (eio_req *req)
357	{
358	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
359	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
360
361	EIO_DESTROY (req);
362	}
363
364	static int eio_finish (eio_req *req)
365	{
366	int res = EIO_FINISH (req);
367
368	if (req->grp)
369	{
370	int res2;
371	eio_req *grp = req->grp;
372
373	/* unlink request */
374	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
375	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
376
377	if (grp->grp_first == req)
378	grp->grp_first = req->grp_next;
379
380	res2 = grp_dec (grp);
381
382	if (!res && res2)
383	res = res2;
384	}
385
386	eio_destroy (req);
387
388	return res;
389	}
390
391	void eio_grp_cancel (eio_req *grp)
392	{
393	for (grp = grp->grp_first; grp; grp = grp->grp_next)
394	eio_cancel (grp);
395	}
396
397	void eio_cancel (eio_req *req)
398	{
399	X_LOCK (etplock);
400	req->flags |= EIO_FLAG_CANCELLED;
401	X_UNLOCK (etplock);
402
403	eio_grp_cancel (req);
404	}
405
406	X_THREAD_PROC (eio_proc);	434	X_THREAD_PROC (etp_proc);
407		435
408	static void start_thread (void)	436	static void etp_start_thread (void)
409	{	437	{
410	worker *wrk = calloc (1, sizeof (worker));	438	etp_worker *wrk = calloc (1, sizeof (etp_worker));
411		439
412	/*TODO*/	440	/*TODO*/
413	assert (("unable to allocate worker thread data", wrk));	441	assert (("unable to allocate worker thread data", wrk));
414		442
415	X_LOCK (etplock);	443	X_LOCK (wrklock);
416		444
417	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	445	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
418	{	446	{
419	wrk->prev = &wrk_first;	447	wrk->prev = &wrk_first;
420	wrk->next = wrk_first.next;	448	wrk->next = wrk_first.next;
421	wrk_first.next->prev = wrk;	449	wrk_first.next->prev = wrk;
422	wrk_first.next = wrk;	450	wrk_first.next = wrk;
423	++started;	451	++started;
424	}	452	}
425	else	453	else
426	free (wrk);	454	free (wrk);
427		455
428	X_UNLOCK (etplock);	456	X_UNLOCK (wrklock);
429	}	457	}
430		458
431	static void maybe_start_thread (void)	459	static void etp_maybe_start_thread (void)
432	{	460	{
433	if (eio_nthreads () >= wanted)	461	if (expect_true (etp_nthreads () >= wanted))
434	return;	462	return;
435		463
436	/* todo: maybe use idle here, but might be less exact */	464	/* todo: maybe use idle here, but might be less exact */
437	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	465	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
438	return;	466	return;
439		467
440	start_thread ();	468	etp_start_thread ();
441	}	469	}
442		470
443	void eio_submit (eio_req *req)
444	{
445	req->pri += EIO_PRI_BIAS;
446
447	if (req->pri < EIO_PRI_MIN + EIO_PRI_BIAS) req->pri = EIO_PRI_MIN + EIO_PRI_BIAS;
448	if (req->pri > EIO_PRI_MAX + EIO_PRI_BIAS) req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
449
450	++nreqs;
451
452	X_LOCK (reqlock);
453	++nready;
454	reqq_push (&req_queue, req);
455	X_COND_SIGNAL (reqwait);
456	X_UNLOCK (reqlock);
457
458	maybe_start_thread ();
459	}
460
461	static void end_thread (void)	471	static void etp_end_thread (void)
462	{	472	{
463	eio_req *req = calloc (1, sizeof (eio_req));	473	eio_req *req = calloc (1, sizeof (eio_req));
464		474
465	req->type = EIO_QUIT;	475	req->type = -1;
466	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	476	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
467		477
468	X_LOCK (reqlock);	478	X_LOCK (reqlock);
469	reqq_push (&req_queue, req);	479	reqq_push (&req_queue, req);
470	X_COND_SIGNAL (reqwait);	480	X_COND_SIGNAL (reqwait);
471	X_UNLOCK (reqlock);	481	X_UNLOCK (reqlock);
472		482
473	X_LOCK (etplock);	483	X_LOCK (wrklock);
474	--started;	484	--started;
475	X_UNLOCK (etplock);	485	X_UNLOCK (wrklock);
476	}	486	}
477		487
478	void eio_set_max_poll_time (double nseconds)
479	{
480	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
481	etp.max_poll_time = nseconds;
482	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
483	}
484
485	void eio_set_max_poll_reqs (unsigned int maxreqs)
486	{
487	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
488	etp.max_poll_reqs = maxreqs;
489	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
490	}
491
492	void eio_set_max_idle (unsigned int nthreads)
493	{
494	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
495	max_idle = nthreads <= 0 ? 1 : nthreads;
496	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
497	}
498
499	void eio_set_min_parallel (unsigned int nthreads)
500	{
501	if (wanted < nthreads)
502	wanted = nthreads;
503	}
504
505	void eio_set_max_parallel (unsigned int nthreads)
506	{
507	if (wanted > nthreads)
508	wanted = nthreads;
509
510	while (started > wanted)
511	end_thread ();
512	}
513
514	int eio_poll (void)	488	static int etp_poll (void)
515	{	489	{
516	int maxreqs = etp.max_poll_reqs;	490	unsigned int maxreqs;
		491	unsigned int maxtime;
517	struct timeval tv_start, tv_now;	492	struct timeval tv_start, tv_now;
518	eio_req *req;
519		493
520	if (etp.max_poll_time)	494	X_LOCK (reslock);
		495	maxreqs = max_poll_reqs;
		496	maxtime = max_poll_time;
		497	X_UNLOCK (reslock);
		498
		499	if (maxtime)
521	gettimeofday (&tv_start, 0);	500	gettimeofday (&tv_start, 0);
522		501
523	for (;;)	502	for (;;)
524	{	503	{
		504	ETP_REQ *req;
		505
525	maybe_start_thread ();	506	etp_maybe_start_thread ();
526		507
527	X_LOCK (reslock);	508	X_LOCK (reslock);
528	req = reqq_shift (&res_queue);	509	req = reqq_shift (&res_queue);
529		510
530	if (req)	511	if (req)
531	{	512	{
532	--npending;	513	--npending;
533		514
534	if (!res_queue.size && etp.done_poll_cb)	515	if (!res_queue.size && done_poll_cb)
535	etp.done_poll_cb ();	516	done_poll_cb ();
536	}	517	}
537		518
538	X_UNLOCK (reslock);	519	X_UNLOCK (reslock);
539		520
540	if (!req)	521	if (!req)
541	return 0;	522	return 0;
542		523
		524	X_LOCK (reqlock);
543	--nreqs;	525	--nreqs;
		526	X_UNLOCK (reqlock);
544		527
545	if (req->type == EIO_GROUP && req->size)	528	if (expect_false (req->type == EIO_GROUP && req->size))
546	{	529	{
547	req->int1 = 1; /* mark request as delayed */	530	req->int1 = 1; /* mark request as delayed */
548	continue;	531	continue;
549	}	532	}
550	else	533	else
551	{	534	{
552	int res = eio_finish (req);	535	int res = ETP_FINISH (req);
553	if (res)	536	if (expect_false (res))
554	return res;	537	return res;
555	}	538	}
556		539
557	if (maxreqs && !--maxreqs)	540	if (expect_false (maxreqs && !--maxreqs))
558	break;	541	break;
559		542
560	if (etp.max_poll_time)	543	if (maxtime)
561	{	544	{
562	gettimeofday (&tv_now, 0);	545	gettimeofday (&tv_now, 0);
563		546
564	if (tvdiff (&tv_start, &tv_now) >= etp.max_poll_time)	547	if (tvdiff (&tv_start, &tv_now) >= maxtime)
565	break;	548	break;
566	}	549	}
567	}	550	}
568		551
569	errno = EAGAIN;	552	errno = EAGAIN;
570	return -1;	553	return -1;
571	}	554	}
572		555
		556	static void etp_cancel (ETP_REQ *req)
		557	{
		558	X_LOCK (wrklock);
		559	req->flags |= EIO_FLAG_CANCELLED;
		560	X_UNLOCK (wrklock);
		561
		562	eio_grp_cancel (req);
		563	}
		564
		565	static void etp_submit (ETP_REQ *req)
		566	{
		567	req->pri -= ETP_PRI_MIN;
		568
		569	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		570	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		571
		572	if (expect_false (req->type == EIO_GROUP))
		573	{
		574	/* I hope this is worth it :/ */
		575	X_LOCK (reqlock);
		576	++nreqs;
		577	X_UNLOCK (reqlock);
		578
		579	X_LOCK (reslock);
		580
		581	++npending;
		582
		583	if (!reqq_push (&res_queue, req) && want_poll_cb)
		584	want_poll_cb ();
		585
		586	X_UNLOCK (reslock);
		587	}
		588	else
		589	{
		590	X_LOCK (reqlock);
		591	++nreqs;
		592	++nready;
		593	reqq_push (&req_queue, req);
		594	X_COND_SIGNAL (reqwait);
		595	X_UNLOCK (reqlock);
		596
		597	etp_maybe_start_thread ();
		598	}
		599	}
		600
		601	static void etp_set_max_poll_time (double nseconds)
		602	{
		603	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		604	max_poll_time = nseconds * EIO_TICKS;
		605	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		606	}
		607
		608	static void etp_set_max_poll_reqs (unsigned int maxreqs)
		609	{
		610	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		611	max_poll_reqs = maxreqs;
		612	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		613	}
		614
		615	static void etp_set_max_idle (unsigned int nthreads)
		616	{
		617	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		618	max_idle = nthreads <= 0 ? 1 : nthreads;
		619	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		620	}
		621
		622	static void etp_set_min_parallel (unsigned int nthreads)
		623	{
		624	if (wanted < nthreads)
		625	wanted = nthreads;
		626	}
		627
		628	static void etp_set_max_parallel (unsigned int nthreads)
		629	{
		630	if (wanted > nthreads)
		631	wanted = nthreads;
		632
		633	while (started > wanted)
		634	etp_end_thread ();
		635	}
		636
		637	/*****************************************************************************/
		638
		639	static void grp_try_feed (eio_req *grp)
		640	{
		641	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		642	{
		643	grp->flags &= ~EIO_FLAG_GROUPADD;
		644
		645	EIO_FEED (grp);
		646
		647	/* stop if no progress has been made */
		648	if (!(grp->flags & EIO_FLAG_GROUPADD))
		649	{
		650	grp->feed = 0;
		651	break;
		652	}
		653	}
		654	}
		655
		656	static int grp_dec (eio_req *grp)
		657	{
		658	--grp->size;
		659
		660	/* call feeder, if applicable */
		661	grp_try_feed (grp);
		662
		663	/* finish, if done */
		664	if (!grp->size && grp->int1)
		665	return eio_finish (grp);
		666	else
		667	return 0;
		668	}
		669
		670	void eio_destroy (eio_req *req)
		671	{
		672	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		673	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		674
		675	EIO_DESTROY (req);
		676	}
		677
		678	static int eio_finish (eio_req *req)
		679	{
		680	int res = EIO_FINISH (req);
		681
		682	if (req->grp)
		683	{
		684	int res2;
		685	eio_req *grp = req->grp;
		686
		687	/* unlink request */
		688	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		689	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		690
		691	if (grp->grp_first == req)
		692	grp->grp_first = req->grp_next;
		693
		694	res2 = grp_dec (grp);
		695
		696	if (!res && res2)
		697	res = res2;
		698	}
		699
		700	eio_destroy (req);
		701
		702	return res;
		703	}
		704
		705	void eio_grp_cancel (eio_req *grp)
		706	{
		707	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		708	eio_cancel (grp);
		709	}
		710
		711	void eio_cancel (eio_req *req)
		712	{
		713	etp_cancel (req);
		714	}
		715
		716	void eio_submit (eio_req *req)
		717	{
		718	etp_submit (req);
		719	}
		720
		721	unsigned int eio_nreqs (void)
		722	{
		723	return etp_nreqs ();
		724	}
		725
		726	unsigned int eio_nready (void)
		727	{
		728	return etp_nready ();
		729	}
		730
		731	unsigned int eio_npending (void)
		732	{
		733	return etp_npending ();
		734	}
		735
		736	unsigned int eio_nthreads (void)
		737	{
		738	return etp_nthreads ();
		739	}
		740
		741	void eio_set_max_poll_time (double nseconds)
		742	{
		743	etp_set_max_poll_time (nseconds);
		744	}
		745
		746	void eio_set_max_poll_reqs (unsigned int maxreqs)
		747	{
		748	etp_set_max_poll_reqs (maxreqs);
		749	}
		750
		751	void eio_set_max_idle (unsigned int nthreads)
		752	{
		753	etp_set_max_idle (nthreads);
		754	}
		755
		756	void eio_set_min_parallel (unsigned int nthreads)
		757	{
		758	etp_set_min_parallel (nthreads);
		759	}
		760
		761	void eio_set_max_parallel (unsigned int nthreads)
		762	{
		763	etp_set_max_parallel (nthreads);
		764	}
		765
		766	int eio_poll (void)
		767	{
		768	return etp_poll ();
		769	}
		770
573	/*****************************************************************************/	771	/*****************************************************************************/
574	/* work around various missing functions */	772	/* work around various missing functions */
575		773
576	#if !HAVE_PREADWRITE	774	#if !HAVE_PREADWRITE
		775	# undef pread
		776	# undef pwrite
577	# define pread eio__pread	777	# define pread eio__pread
578	# define pwrite eio__pwrite	778	# define pwrite eio__pwrite
579
580	/*
581	* make our pread/pwrite safe against themselves, but not against
582	* normal read/write by using a mutex. slows down execution a lot,
583	* but that's your problem, not mine.
584	*/
585	static mutex_t preadwritelock = X_MUTEX_INIT;
586		779
587	static ssize_t	780	static ssize_t
588	eio__pread (int fd, void *buf, size_t count, off_t offset)	781	eio__pread (int fd, void *buf, size_t count, off_t offset)
589	{	782	{
590	ssize_t res;	783	ssize_t res;
…		…
608		801
609	X_LOCK (preadwritelock);	802	X_LOCK (preadwritelock);
610	ooffset = lseek (fd, 0, SEEK_CUR);	803	ooffset = lseek (fd, 0, SEEK_CUR);
611	lseek (fd, offset, SEEK_SET);	804	lseek (fd, offset, SEEK_SET);
612	res = write (fd, buf, count);	805	res = write (fd, buf, count);
613	lseek (fd, offset, SEEK_SET);	806	lseek (fd, ooffset, SEEK_SET);
614	X_UNLOCK (preadwritelock);	807	X_UNLOCK (preadwritelock);
615		808
616	return res;	809	return res;
617	}	810	}
618	#endif	811	#endif
619		812
620	#ifndef HAVE_FUTIMES	813	#ifndef HAVE_FUTIMES
621		814
		815	# undef utimes
		816	# undef futimes
622	# define utimes(path,times) eio__utimes (path, times)	817	# define utimes(path,times) eio__utimes (path, times)
623	# define futimes(fd,times) eio__futimes (fd, times)	818	# define futimes(fd,times) eio__futimes (fd, times)
624		819
625	static int	820	static int
626	eio__utimes (const char *filename, const struct timeval times[2])	821	eio__utimes (const char *filename, const struct timeval times[2])
…		…
645	}	840	}
646		841
647	#endif	842	#endif
648		843
649	#if !HAVE_FDATASYNC	844	#if !HAVE_FDATASYNC
		845	# undef fdatasync
650	# define fdatasync fsync	846	# define fdatasync(fd) fsync (fd)
651	#endif	847	#endif
		848
		849	/* sync_file_range always needs emulation */
		850	int
		851	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		852	{
		853	#if HAVE_SYNC_FILE_RANGE
		854	int res;
		855
		856	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		857	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		858	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		859	{
		860	flags = 0
		861	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		862	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		863	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		864	}
		865
		866	res = sync_file_range (fd, offset, nbytes, flags);
		867
		868	if (!res || errno != ENOSYS)
		869	return res;
		870	#endif
		871
		872	/* even though we could play tricks with the flags, it's better to always
		873	* call fdatasync, as that matches the expectation of its users best */
		874	return fdatasync (fd);
		875	}
652		876
653	#if !HAVE_READAHEAD	877	#if !HAVE_READAHEAD
		878	# undef readahead
654	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	879	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
655		880
656	static ssize_t	881	static ssize_t
657	eio__readahead (int fd, off_t offset, size_t count, worker *self)	882	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
658	{	883	{
659	size_t todo = count;	884	size_t todo = count;
660	dBUF;	885	dBUF;
661		886
662	while (todo > 0)	887	while (todo > 0)
…		…
672	return count;	897	return count;
673	}	898	}
674		899
675	#endif	900	#endif
676		901
677	#if !HAVE_READDIR_R
678	# define readdir_r eio__readdir_r
679
680	static mutex_t readdirlock = X_MUTEX_INIT;
681
682	static int
683	eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
684	{
685	EIO_STRUCT_DIRENT *e;
686	int errorno;
687
688	X_LOCK (readdirlock);
689
690	e = readdir (dirp);
691	errorno = errno;
692
693	if (e)
694	{
695	*res = ent;
696	strcpy (ent->d_name, e->d_name);
697	}
698	else
699	*res = 0;
700
701	X_UNLOCK (readdirlock);
702
703	errno = errorno;
704	return e ? 0 : -1;
705	}
706	#endif
707
708	/* sendfile always needs emulation */	902	/* sendfile always needs emulation */
709	static ssize_t	903	static ssize_t
710	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self)	904	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
711	{	905	{
712	ssize_t res;	906	ssize_t res;
713		907
714	if (!count)	908	if (!count)
715	return 0;	909	return 0;
716		910
717	#if HAVE_SENDFILE	911	#if HAVE_SENDFILE
718	# if __linux	912	# if __linux
719	res = sendfile (ofd, ifd, &offset, count);	913	res = sendfile (ofd, ifd, &offset, count);
720		914
721	# elif __freebsd	915	# elif __FreeBSD__
722	/*	916	/*
723	* Of course, the freebsd sendfile is a dire hack with no thoughts	917	* Of course, the freebsd sendfile is a dire hack with no thoughts
724	* wasted on making it similar to other I/O functions.	918	* wasted on making it similar to other I/O functions.
725	*/	919	*/
726	{	920	{
727	off_t sbytes;	921	off_t sbytes;
728	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	922	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
729		923
730	if (res < 0 && sbytes)	924	#if 0 /* according to the manpage, this is correct, but broken behaviour */
731	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	925	/* freebsd' sendfile will return 0 on success */
		926	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		927	/* not on e.g. EIO or EPIPE - sounds broken */
		928	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
		929	res = sbytes;
		930	#endif
		931
		932	/* according to source inspection, this is correct, and useful behaviour */
		933	if (sbytes)
		934	res = sbytes;
		935	}
		936
		937	# elif defined (__APPLE__)
		938
		939	{
		940	off_t sbytes = count;
		941	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		942
		943	/* according to the manpage, sbytes is always valid */
		944	if (sbytes)
732	res = sbytes;	945	res = sbytes;
733	}	946	}
734		947
735	# elif __hpux	948	# elif __hpux
736	res = sendfile (ofd, ifd, offset, count, 0, 0);	949	res = sendfile (ofd, ifd, offset, count, 0, 0);
…		…
750	if (res < 0 && sbytes)	963	if (res < 0 && sbytes)
751	res = sbytes;	964	res = sbytes;
752	}	965	}
753		966
754	# endif	967	# endif
		968
		969	#elif defined (_WIN32)
		970
		971	/* does not work, just for documentation of what would need to be done */
		972	{
		973	HANDLE h = TO_SOCKET (ifd);
		974	SetFilePointer (h, offset, 0, FILE_BEGIN);
		975	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		976	}
		977
755	#else	978	#else
756	res = -1;	979	res = -1;
757	errno = ENOSYS;	980	errno = ENOSYS;
758	#endif	981	#endif
759		982
760	if (res < 0	983	if (res < 0
761	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	984	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		985	/* BSDs */
		986	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		987	|| errno == ENOTSUP
		988	#endif
		989	|| errno == EOPNOTSUPP /* BSDs */
762	#if __solaris	990	#if __solaris
763	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	991	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
764	#endif	992	#endif
765	)	993	)
766	)	994	)
…		…
797	}	1025	}
798		1026
799	return res;	1027	return res;
800	}	1028	}
801		1029
		1030	static signed char
		1031	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1032	{
		1033	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1034	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
		1035	}
		1036
		1037	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1038
		1039	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1040	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1041
		1042	static void
		1043	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1044	{
		1045	unsigned char bits [9 + sizeof (ino_t) * 8];
		1046	unsigned char *bit = bits;
		1047
		1048	assert (CHAR_BIT == 8);
		1049	assert (sizeof (eio_dirent) * 8 < 256);
		1050	assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
		1051	assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
		1052
		1053	if (size <= EIO_SORT_FAST)
		1054	return;
		1055
		1056	/* first prepare an array of bits to test in our radix sort */
		1057	/* try to take endianness into account, as well as differences in ino_t sizes */
		1058	/* inode_bits must contain all inodes ORed together */
		1059	/* which is used to skip bits that are 0 everywhere, which is very common */
		1060	{
		1061	ino_t endianness;
		1062	int i, j;
		1063
		1064	/* we store the byte offset of byte n into byte n of "endianness" */
		1065	for (i = 0; i < sizeof (ino_t); ++i)
		1066	((unsigned char *)&endianness)[i] = i;
		1067
		1068	*bit++ = 0;
		1069
		1070	for (i = 0; i < sizeof (ino_t); ++i)
		1071	{
		1072	/* shifting off the byte offsets out of "endianness" */
		1073	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1074	endianness >>= 8;
		1075
		1076	for (j = 0; j < 8; ++j)
		1077	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1078	*bit++ = offs + j;
		1079	}
		1080
		1081	for (j = 0; j < 8; ++j)
		1082	if (score_bits & (1 << j))
		1083	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1084	}
		1085
		1086	/* now actually do the sorting (a variant of MSD radix sort) */
		1087	{
		1088	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1089	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1090	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1091	int stk_idx = 0;
		1092
		1093	base_stk [stk_idx] = dents;
		1094	end_stk [stk_idx] = dents + size;
		1095	bit_stk [stk_idx] = bit - 1;
		1096
		1097	do
		1098	{
		1099	base = base_stk [stk_idx];
		1100	end = end_stk [stk_idx];
		1101	bit = bit_stk [stk_idx];
		1102
		1103	for (;;)
		1104	{
		1105	unsigned char O = *bit >> 3;
		1106	unsigned char M = 1 << (*bit & 7);
		1107
		1108	eio_dirent *a = base;
		1109	eio_dirent *b = end;
		1110
		1111	if (b - a < EIO_SORT_CUTOFF)
		1112	break;
		1113
		1114	/* now bit-partition the array on the bit */
		1115	/* this ugly asymmetric loop seems to perform much better than typical */
		1116	/* partition algos found in the literature */
		1117	do
		1118	if (!(((unsigned char *)a)[O] & M))
		1119	++a;
		1120	else if (!(((unsigned char *)--b)[O] & M))
		1121	{
		1122	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1123	++a;
		1124	}
		1125	while (b > a);
		1126
		1127	/* next bit, or stop, if no bits left in this path */
		1128	if (!*--bit)
		1129	break;
		1130
		1131	base_stk [stk_idx] = a;
		1132	end_stk [stk_idx] = end;
		1133	bit_stk [stk_idx] = bit;
		1134	++stk_idx;
		1135
		1136	end = a;
		1137	}
		1138	}
		1139	while (stk_idx--);
		1140	}
		1141	}
		1142
		1143	static void
		1144	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1145	{
		1146	/* first move the smallest element to the front, to act as a sentinel */
		1147	{
		1148	int i;
		1149	eio_dirent *min = dents;
		1150
		1151	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1152	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1153	if (EIO_DENT_CMP (dents [i], <, *min))
		1154	min = &dents [i];
		1155
		1156	/* swap elements 0 and j (minimum) */
		1157	{
		1158	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1159	}
		1160	}
		1161
		1162	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1163	{
		1164	eio_dirent *i, *j;
		1165
		1166	for (i = dents + 1; i < dents + size; ++i)
		1167	{
		1168	eio_dirent value = *i;
		1169
		1170	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1171	j [1] = j [0];
		1172
		1173	j [1] = value;
		1174	}
		1175	}
		1176	}
		1177
		1178	static void
		1179	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1180	{
		1181	if (size <= 1)
		1182	return; /* our insertion sort relies on size > 0 */
		1183
		1184	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1185	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1186	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1187
		1188	/* use an insertion sort at the end, or for small arrays, */
		1189	/* as insertion sort is more efficient for small partitions */
		1190	eio_dent_insertion_sort (dents, size);
		1191	}
		1192
802	/* read a full directory */	1193	/* read a full directory */
803	static void	1194	static void
804	eio__scandir (eio_req *req, worker *self)	1195	eio__scandir (eio_req *req, etp_worker *self)
805	{	1196	{
806	DIR *dirp;	1197	DIR *dirp;
807	union
808	{
809	EIO_STRUCT_DIRENT d;
810	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
811	} *u;
812	EIO_STRUCT_DIRENT *entp;	1198	EIO_STRUCT_DIRENT *entp;
813	char *name, *names;	1199	char *name, *names;
814	int memlen = 4096;	1200	int namesalloc = 4096;
815	int memofs = 0;	1201	int namesoffs = 0;
		1202	int flags = req->int1;
		1203	eio_dirent *dents = 0;
		1204	int dentalloc = 128;
816	int res = 0;	1205	int dentoffs = 0;
		1206	ino_t inode_bits = 0;
817		1207
		1208	req->result = -1;
		1209
		1210	if (!(flags & EIO_READDIR_DENTS))
		1211	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
		1212
818	X_LOCK (etplock);	1213	X_LOCK (wrklock);
		1214	/* the corresponding closedir is in ETP_WORKER_CLEAR */
819	self->dirp = dirp = opendir (req->ptr1);	1215	self->dirp = dirp = opendir (req->ptr1);
820	self->dbuf = u = malloc (sizeof (*u));
821	req->flags |= EIO_FLAG_PTR2_FREE;	1216	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1217	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
822	req->ptr2 = names = malloc (memlen);	1218	req->ptr2 = names = malloc (namesalloc);
823	X_UNLOCK (etplock);	1219	X_UNLOCK (wrklock);
824		1220
825	if (dirp && u && names)	1221	if (dirp && names && (!flags || dents))
826	for (;;)	1222	for (;;)
827	{	1223	{
828	errno = 0;	1224	errno = 0;
829	readdir_r (dirp, &u->d, &entp);	1225	entp = readdir (dirp);
830		1226
831	if (!entp)	1227	if (!entp)
		1228	{
		1229	if (errno)
		1230	break;
		1231
		1232	/* sort etc. */
		1233	req->int1 = flags;
		1234	req->result = dentoffs;
		1235
		1236	if (flags & EIO_READDIR_STAT_ORDER)
		1237	eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
		1238	else if (flags & EIO_READDIR_DIRS_FIRST)
		1239	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1240	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1241	else
		1242	{
		1243	/* in this case, all is known, and we just put dirs first and sort them */
		1244	eio_dirent *oth = dents + dentoffs;
		1245	eio_dirent *dir = dents;
		1246
		1247	/* now partition dirs to the front, and non-dirs to the back */
		1248	/* by walking from both sides and swapping if necessary */
		1249	/* also clear score, so it doesn't influence sorting */
		1250	while (oth > dir)
		1251	{
		1252	if (dir->type == EIO_DT_DIR)
		1253	++dir;
		1254	else if ((--oth)->type == EIO_DT_DIR)
		1255	{
		1256	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1257
		1258	++dir;
		1259	}
		1260	}
		1261
		1262	/* now sort the dirs only */
		1263	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1264	}
		1265
832	break;	1266	break;
		1267	}
833		1268
		1269	/* now add the entry to our list(s) */
834	name = entp->d_name;	1270	name = entp->d_name;
835		1271
		1272	/* skip . and .. entries */
836	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1273	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
837	{	1274	{
838	int len = strlen (name) + 1;	1275	int len = D_NAMLEN (entp) + 1;
839		1276
840	res++;	1277	while (expect_false (namesoffs + len > namesalloc))
841
842	while (memofs + len > memlen)
843	{	1278	{
844	memlen *= 2;	1279	namesalloc *= 2;
845	X_LOCK (etplock);	1280	X_LOCK (wrklock);
846	req->ptr2 = names = realloc (names, memlen);	1281	req->ptr2 = names = realloc (names, namesalloc);
847	X_UNLOCK (etplock);	1282	X_UNLOCK (wrklock);
848		1283
849	if (!names)	1284	if (!names)
850	break;	1285	break;
851	}	1286	}
852		1287
853	memcpy (names + memofs, name, len);	1288	memcpy (names + namesoffs, name, len);
		1289
		1290	if (dents)
		1291	{
		1292	struct eio_dirent *ent;
		1293
		1294	if (expect_false (dentoffs == dentalloc))
		1295	{
		1296	dentalloc *= 2;
		1297	X_LOCK (wrklock);
		1298	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1299	X_UNLOCK (wrklock);
		1300
		1301	if (!dents)
		1302	break;
		1303	}
		1304
		1305	ent = dents + dentoffs;
		1306
		1307	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1308	ent->namelen = len - 1;
		1309	ent->inode = D_INO (entp);
		1310
		1311	inode_bits |= ent->inode;
		1312
		1313	switch (D_TYPE (entp))
		1314	{
		1315	default:
		1316	ent->type = EIO_DT_UNKNOWN;
		1317	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1318	break;
		1319
		1320	#ifdef DT_FIFO
		1321	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1322	#endif
		1323	#ifdef DT_CHR
		1324	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1325	#endif
		1326	#ifdef DT_MPC
		1327	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1328	#endif
		1329	#ifdef DT_DIR
		1330	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1331	#endif
		1332	#ifdef DT_NAM
		1333	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1334	#endif
		1335	#ifdef DT_BLK
		1336	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1337	#endif
		1338	#ifdef DT_MPB
		1339	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1340	#endif
		1341	#ifdef DT_REG
		1342	case DT_REG: ent->type = EIO_DT_REG; break;
		1343	#endif
		1344	#ifdef DT_NWK
		1345	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1346	#endif
		1347	#ifdef DT_CMP
		1348	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1349	#endif
		1350	#ifdef DT_LNK
		1351	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1352	#endif
		1353	#ifdef DT_SOCK
		1354	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1355	#endif
		1356	#ifdef DT_DOOR
		1357	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1358	#endif
		1359	#ifdef DT_WHT
		1360	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1361	#endif
		1362	}
		1363
		1364	ent->score = 7;
		1365
		1366	if (flags & EIO_READDIR_DIRS_FIRST)
		1367	{
		1368	if (ent->type == EIO_DT_UNKNOWN)
		1369	{
		1370	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1371	ent->score = 1;
		1372	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1373	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1374	}
		1375	else if (ent->type == EIO_DT_DIR)
		1376	ent->score = 0;
		1377	}
		1378	}
		1379
854	memofs += len;	1380	namesoffs += len;
		1381	++dentoffs;
		1382	}
		1383
		1384	if (EIO_CANCELLED (req))
		1385	{
		1386	errno = ECANCELED;
		1387	break;
855	}	1388	}
856	}	1389	}
		1390	}
857		1391
858	if (errno)	1392	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
859	res = -1;	1393	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
860		1394	#else
861	req->result = res;	1395
		1396	int
		1397	eio__msync (void *mem, size_t len, int flags)
		1398	{
		1399	if (EIO_MS_ASYNC != MS_SYNC
		1400	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1401	|| EIO_MS_SYNC != MS_SYNC)
		1402	{
		1403	flags = 0
		1404	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1405	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1406	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1407	}
		1408
		1409	return msync (mem, len, flags);
		1410	}
		1411
		1412	#endif
		1413
		1414	int
		1415	eio__mtouch (void *mem, size_t len, int flags)
		1416	{
		1417	intptr_t addr = (intptr_t)mem;
		1418	intptr_t end = addr + len;
		1419	#ifdef PAGESIZE
		1420	const intptr_t page = PAGESIZE;
		1421	#else
		1422	static intptr_t page;
		1423
		1424	if (!page)
		1425	page = sysconf (_SC_PAGESIZE);
		1426	#endif
		1427
		1428	/* round down to start of page, although this is probably useless */
		1429	addr &= ~(page - 1); /* assume page size is always a power of two */
		1430
		1431	if (addr < end)
		1432	if (flags & EIO_MT_MODIFY) /* modify */
		1433	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
		1434	else
		1435	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
		1436
		1437	return 0;
862	}	1438	}
863		1439
864	/*****************************************************************************/	1440	/*****************************************************************************/
865		1441
866	#define ALLOC(len) \	1442	#define ALLOC(len) \
867	if (!req->ptr2) \	1443	if (!req->ptr2) \
868	{ \	1444	{ \
869	X_LOCK (etplock); \	1445	X_LOCK (wrklock); \
870	req->flags |= EIO_FLAG_PTR2_FREE; \	1446	req->flags |= EIO_FLAG_PTR2_FREE; \
871	X_UNLOCK (etplock); \	1447	X_UNLOCK (wrklock); \
872	req->ptr2 = malloc (len); \	1448	req->ptr2 = malloc (len); \
873	if (!req->ptr2) \	1449	if (!req->ptr2) \
874	{ \	1450	{ \
875	errno = ENOMEM; \	1451	errno = ENOMEM; \
876	req->result = -1; \	1452	req->result = -1; \
877	break; \	1453	break; \
878	} \	1454	} \
879	}	1455	}
880		1456
881	X_THREAD_PROC (eio_proc)	1457	X_THREAD_PROC (etp_proc)
882	{	1458	{
883	eio_req *req;	1459	ETP_REQ *req;
884	struct timespec ts;	1460	struct timespec ts;
885	worker *self = (worker *)thr_arg;	1461	etp_worker *self = (etp_worker *)thr_arg;
886		1462
887	/* try to distribute timeouts somewhat randomly */	1463	/* try to distribute timeouts somewhat randomly */
888	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1464	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
889		1465
890	for (;;)	1466	for (;;)
891	{	1467	{
892	ts.tv_sec = time (0) + IDLE_TIMEOUT;
893
894	X_LOCK (reqlock);	1468	X_LOCK (reqlock);
895		1469
896	for (;;)	1470	for (;;)
897	{	1471	{
898	self->req = req = reqq_shift (&req_queue);	1472	self->req = req = reqq_shift (&req_queue);
…		…
900	if (req)	1474	if (req)
901	break;	1475	break;
902		1476
903	++idle;	1477	++idle;
904		1478
		1479	ts.tv_sec = time (0) + IDLE_TIMEOUT;
905	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1480	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
906	{	1481	{
907	if (idle > max_idle)	1482	if (idle > max_idle)
908	{	1483	{
909	--idle;	1484	--idle;
910	X_UNLOCK (reqlock);	1485	X_UNLOCK (reqlock);
911	X_LOCK (etplock);	1486	X_LOCK (wrklock);
912	--started;	1487	--started;
913	X_UNLOCK (etplock);	1488	X_UNLOCK (wrklock);
914	goto quit;	1489	goto quit;
915	}	1490	}
916		1491
917	/* we are allowed to idle, so do so without any timeout */	1492	/* we are allowed to idle, so do so without any timeout */
918	X_COND_WAIT (reqwait, reqlock);	1493	X_COND_WAIT (reqwait, reqlock);
919	ts.tv_sec = time (0) + IDLE_TIMEOUT;
920	}	1494	}
921		1495
922	--idle;	1496	--idle;
923	}	1497	}
924		1498
925	--nready;	1499	--nready;
926		1500
927	X_UNLOCK (reqlock);	1501	X_UNLOCK (reqlock);
928		1502
929	errno = 0; /* strictly unnecessary */	1503	if (req->type < 0)
		1504	goto quit;
930		1505
931	if (!EIO_CANCELLED (req))	1506	if (!EIO_CANCELLED (req))
932	switch (req->type)	1507	ETP_EXECUTE (self, req);
933	{
934	case EIO_READ: ALLOC (req->size);
935	req->result = req->offs >= 0
936	? pread (req->int1, req->ptr2, req->size, req->offs)
937	: read (req->int1, req->ptr2, req->size); break;
938	case EIO_WRITE: req->result = req->offs >= 0
939	? pwrite (req->int1, req->ptr2, req->size, req->offs)
940	: write (req->int1, req->ptr2, req->size); break;
941
942	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
943	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
944
945	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
946	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
947	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
948	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
949	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
950	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
951
952	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
953	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
954	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
955	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
956	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
957	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
958
959	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
960	case EIO_CLOSE: req->result = close (req->int1); break;
961	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
962	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
963	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
964	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
965	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
966	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
967	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
968	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
969
970	case EIO_READLINK: ALLOC (NAME_MAX);
971	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
972
973	case EIO_SYNC: req->result = 0; sync (); break;
974	case EIO_FSYNC: req->result = fsync (req->int1); break;
975	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
976
977	case EIO_READDIR: eio__scandir (req, self); break;
978
979	case EIO_BUSY:
980	#ifdef _WIN32
981	Sleep (req->nv1 * 1000.);
982	#else
983	{
984	struct timeval tv;
985
986	tv.tv_sec = req->nv1;
987	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
988
989	req->result = select (0, 0, 0, 0, &tv);
990	}
991	#endif
992	break;
993
994	case EIO_UTIME:
995	case EIO_FUTIME:
996	{
997	struct timeval tv[2];
998	struct timeval *times;
999
1000	if (req->nv1 != -1. || req->nv2 != -1.)
1001	{
1002	tv[0].tv_sec = req->nv1;
1003	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1004	tv[1].tv_sec = req->nv2;
1005	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1006
1007	times = tv;
1008	}
1009	else
1010	times = 0;
1011
1012
1013	req->result = req->type == EIO_FUTIME
1014	? futimes (req->int1, times)
1015	: utimes (req->ptr1, times);
1016	}
1017
1018	case EIO_GROUP:
1019	case EIO_NOP:
1020	req->result = 0;
1021	break;
1022
1023	case EIO_QUIT:
1024	goto quit;
1025
1026	default:
1027	req->result = -1;
1028	break;
1029	}
1030
1031	req->errorno = errno;
1032		1508
1033	X_LOCK (reslock);	1509	X_LOCK (reslock);
1034		1510
1035	++npending;	1511	++npending;
1036		1512
1037	if (!reqq_push (&res_queue, req) && etp.want_poll_cb)	1513	if (!reqq_push (&res_queue, req) && want_poll_cb)
1038	etp.want_poll_cb ();	1514	want_poll_cb ();
1039		1515
1040	self->req = 0;	1516	self->req = 0;
1041	worker_clear (self);	1517	etp_worker_clear (self);
1042		1518
1043	X_UNLOCK (reslock);	1519	X_UNLOCK (reslock);
1044	}	1520	}
1045		1521
1046	quit:	1522	quit:
1047	X_LOCK (etplock);	1523	X_LOCK (wrklock);
1048	worker_free (self);	1524	etp_worker_free (self);
1049	X_UNLOCK (etplock);	1525	X_UNLOCK (wrklock);
1050		1526
1051	return 0;	1527	return 0;
1052	}	1528	}
1053		1529
1054	/*****************************************************************************/	1530	/*****************************************************************************/
1055		1531
1056	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1532	int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1057	{	1533	{
1058	etp_init (&etp, want_poll, done_poll);	1534	return etp_init (want_poll, done_poll);
1059	}	1535	}
1060		1536
1061	static void eio_api_destroy (eio_req *req)	1537	static void eio_api_destroy (eio_req *req)
1062	{	1538	{
1063	free (req);	1539	free (req);
…		…
1085	{ \	1561	{ \
1086	eio_api_destroy (req); \	1562	eio_api_destroy (req); \
1087	return 0; \	1563	return 0; \
1088	}	1564	}
1089		1565
		1566	static void eio_execute (etp_worker *self, eio_req *req)
		1567	{
		1568	errno = 0;
		1569
		1570	switch (req->type)
		1571	{
		1572	case EIO_READ: ALLOC (req->size);
		1573	req->result = req->offs >= 0
		1574	? pread (req->int1, req->ptr2, req->size, req->offs)
		1575	: read (req->int1, req->ptr2, req->size); break;
		1576	case EIO_WRITE: req->result = req->offs >= 0
		1577	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1578	: write (req->int1, req->ptr2, req->size); break;
		1579
		1580	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		1581	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
		1582
		1583	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1584	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1585	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1586	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1587	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1588	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1589
		1590	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1591	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1592	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1593	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1594
		1595	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1596	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1597	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1598	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1599	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1600	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1601
		1602	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1603	case EIO_CLOSE: req->result = close (req->int1); break;
		1604	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1605	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1606	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1607	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1608	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1609	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1610	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1611	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
		1612
		1613	case EIO_READLINK: ALLOC (NAME_MAX);
		1614	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
		1615
		1616	case EIO_SYNC: req->result = 0; sync (); break;
		1617	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1618	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1619	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1620	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1621	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1622
		1623	case EIO_READDIR: eio__scandir (req, self); break;
		1624
		1625	case EIO_BUSY:
		1626	#ifdef _WIN32
		1627	Sleep (req->nv1 * 1000.);
		1628	#else
		1629	{
		1630	struct timeval tv;
		1631
		1632	tv.tv_sec = req->nv1;
		1633	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
		1634
		1635	req->result = select (0, 0, 0, 0, &tv);
		1636	}
		1637	#endif
		1638	break;
		1639
		1640	case EIO_UTIME:
		1641	case EIO_FUTIME:
		1642	{
		1643	struct timeval tv[2];
		1644	struct timeval *times;
		1645
		1646	if (req->nv1 != -1. || req->nv2 != -1.)
		1647	{
		1648	tv[0].tv_sec = req->nv1;
		1649	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		1650	tv[1].tv_sec = req->nv2;
		1651	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		1652
		1653	times = tv;
		1654	}
		1655	else
		1656	times = 0;
		1657
		1658
		1659	req->result = req->type == EIO_FUTIME
		1660	? futimes (req->int1, times)
		1661	: utimes (req->ptr1, times);
		1662	}
		1663	break;
		1664
		1665	case EIO_GROUP:
		1666	abort (); /* handled in eio_request */
		1667
		1668	case EIO_NOP:
		1669	req->result = 0;
		1670	break;
		1671
		1672	case EIO_CUSTOM:
		1673	((void (*)(eio_req *))req->feed) (req);
		1674	break;
		1675
		1676	default:
		1677	req->result = -1;
		1678	break;
		1679	}
		1680
		1681	req->errorno = errno;
		1682	}
		1683
1090	#ifndef EIO_NO_WRAPPERS	1684	#ifndef EIO_NO_WRAPPERS
1091		1685
1092	eio_req *eio_nop (int pri, eio_cb cb, void *data)	1686	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1093	{	1687	{
1094	REQ (EIO_NOP); SEND;	1688	REQ (EIO_NOP); SEND;
…		…
1107	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	1701	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1108	{	1702	{
1109	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1703	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1110	}	1704	}
1111		1705
		1706	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1707	{
		1708	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1709	}
		1710
		1711	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1712	{
		1713	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1714	}
		1715
		1716	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		1717	{
		1718	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		1719	}
		1720
1112	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	1721	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1113	{	1722	{
1114	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1723	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1115	}	1724	}
1116		1725
…		…
1135	}	1744	}
1136		1745
1137	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	1746	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1138	{	1747	{
1139	REQ (EIO_FSTAT); req->int1 = fd; SEND;	1748	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1749	}
		1750
		1751	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		1752	{
		1753	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1140	}	1754	}
1141		1755
1142	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	1756	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1143	{	1757	{
1144	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	1758	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
…		…
1218	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	1832	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1219	{	1833	{
1220	return eio__1path (EIO_LSTAT, path, pri, cb, data);	1834	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1221	}	1835	}
1222		1836
		1837	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		1838	{
		1839	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		1840	}
		1841
1223	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	1842	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1224	{	1843	{
1225	return eio__1path (EIO_UNLINK, path, pri, cb, data);	1844	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1226	}	1845	}
1227		1846
1228	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	1847	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1229	{	1848	{
1230	return eio__1path (EIO_RMDIR, path, pri, cb, data);	1849	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1231	}	1850	}
1232		1851
1233	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	1852	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1234	{	1853	{
1235	return eio__1path (EIO_READDIR, path, pri, cb, data);	1854	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1236	}	1855	}
1237		1856
1238	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	1857	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1239	{	1858	{
1240	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	1859	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1241	}	1860	}
1242		1861
1243	static eio_req *	1862	static eio_req *
1244	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1863	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1245	{	1864	{
…		…
1269	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1888	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1270	{	1889	{
1271	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	1890	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1272	}	1891	}
1273		1892
		1893	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
		1894	{
		1895	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
		1896	}
		1897
1274	#endif	1898	#endif
1275		1899
1276	eio_req *eio_grp (eio_cb cb, void *data)	1900	eio_req *eio_grp (eio_cb cb, void *data)
1277	{	1901	{
1278	const int pri = EIO_PRI_MAX;	1902	const int pri = EIO_PRI_MAX;
…		…
1304		1928
1305	void eio_grp_add (eio_req *grp, eio_req *req)	1929	void eio_grp_add (eio_req *grp, eio_req *req)
1306	{	1930	{
1307	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	1931	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1308		1932
		1933	grp->flags |= EIO_FLAG_GROUPADD;
		1934
1309	++grp->size;	1935	++grp->size;
1310	req->grp = grp;	1936	req->grp = grp;
1311		1937
1312	req->grp_prev = 0;	1938	req->grp_prev = 0;
1313	req->grp_next = grp->grp_first;	1939	req->grp_next = grp->grp_first;
…		…
1321	/*****************************************************************************/	1947	/*****************************************************************************/
1322	/* misc garbage */	1948	/* misc garbage */
1323		1949
1324	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	1950	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1325	{	1951	{
1326	worker wrk;	1952	etp_worker wrk;
		1953	ssize_t ret;
1327		1954
1328	wrk.dbuf = 0;	1955	wrk.dbuf = 0;
1329		1956
1330	eio__sendfile (ofd, ifd, offset, count, &wrk);	1957	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1331		1958
1332	if (wrk.dbuf)	1959	if (wrk.dbuf)
1333	free (wrk.dbuf);	1960	free (wrk.dbuf);
1334	}
1335		1961
		1962	return ret;
		1963	}
		1964

Diff Legend

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