ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/cvsroot/libeio/eio.c

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.13 by root, Tue May 13 19:34:11 2008 UTC vs.
Revision 1.71 by root, Fri Jun 10 12:35:18 2011 UTC

		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008,2009,2010,2011 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
		40	#ifndef _WIN32
		41	# include "config.h"
		42	#endif
		43
1	#include "eio.h"	44	#include "eio.h"
		45
		46	#ifdef EIO_STACKSIZE
		47	# define XTHREAD_STACKSIZE EIO_STACKSIZE
		48	#endif
2	#include "xthread.h"	49	#include "xthread.h"
3		50
4	#include <errno.h>	51	#include <errno.h>
5	#include <stddef.h>	52	#include <stddef.h>
6	#include <stdlib.h>	53	#include <stdlib.h>
7	#include <string.h>	54	#include <string.h>
8	#include <errno.h>	55	#include <errno.h>
9	#include <sys/types.h>	56	#include <sys/types.h>
10	#include <sys/stat.h>	57	#include <sys/stat.h>
		58	#include <sys/statvfs.h>
11	#include <limits.h>	59	#include <limits.h>
12	#include <fcntl.h>	60	#include <fcntl.h>
13	#include <assert.h>	61	#include <assert.h>
14		62
		63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
		64	/* intptr_t only comes form stdint.h, says idiot openbsd coder */
		65	#if HAVE_STDINT_H
		66	# include <stdint.h>
		67	#endif
		68
15	#ifndef EIO_FINISH	69	#ifndef EIO_FINISH
16	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	70	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17	#endif	71	#endif
18		72
19	#ifndef EIO_DESTROY	73	#ifndef EIO_DESTROY
…		…
25	#endif	79	#endif
26		80
27	#ifdef _WIN32	81	#ifdef _WIN32
28		82
29	/*doh*/	83	/*doh*/
30
31	#else	84	#else
32		85
33	# include "config.h"
34	# include <sys/time.h>	86	# include <sys/time.h>
35	# include <sys/select.h>	87	# include <sys/select.h>
36	# include <unistd.h>	88	# include <unistd.h>
37	# include <utime.h>	89	# include <utime.h>
38	# include <signal.h>	90	# include <signal.h>
39	# include <dirent.h>	91	# include <dirent.h>
40		92
		93	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		94	# include <sys/mman.h>
		95	#endif
		96
		97	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		98	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		99	# define _DIRENT_HAVE_D_TYPE /* sigh */
		100	# define D_INO(de) (de)->d_fileno
		101	# define D_NAMLEN(de) (de)->d_namlen
		102	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		103	# define D_INO(de) (de)->d_ino
		104	# endif
		105
		106	#ifdef _D_EXACT_NAMLEN
		107	# undef D_NAMLEN
		108	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		109	#endif
		110
		111	# ifdef _DIRENT_HAVE_D_TYPE
		112	# define D_TYPE(de) (de)->d_type
		113	# endif
		114
41	# ifndef EIO_STRUCT_DIRENT	115	# ifndef EIO_STRUCT_DIRENT
42	# define EIO_STRUCT_DIRENT struct dirent	116	# define EIO_STRUCT_DIRENT struct dirent
43	# endif	117	# endif
44		118
45	#endif	119	#endif
46		120
47	#if HAVE_SENDFILE	121	#if HAVE_SENDFILE
48	# if __linux	122	# if __linux
49	# include <sys/sendfile.h>	123	# include <sys/sendfile.h>
50	# elif __freebsd	124	# elif __FreeBSD__ || defined __APPLE__
51	# include <sys/socket.h>	125	# include <sys/socket.h>
52	# include <sys/uio.h>	126	# include <sys/uio.h>
53	# elif __hpux	127	# elif __hpux
54	# include <sys/socket.h>	128	# include <sys/socket.h>
55	# elif __solaris /* not yet */	129	# elif __solaris
56	# include <sys/sendfile.h>	130	# include <sys/sendfile.h>
57	# else	131	# else
58	# error sendfile support requested but not available	132	# error sendfile support requested but not available
59	# endif	133	# endif
60	#endif	134	#endif
61		135
62	/* number of seconds after which an idle threads exit */	136	#ifndef D_TYPE
63	#define IDLE_TIMEOUT 10	137	# define D_TYPE(de) 0
		138	#endif
		139	#ifndef D_INO
		140	# define D_INO(de) 0
		141	#endif
		142	#ifndef D_NAMLEN
		143	# define D_NAMLEN(de) strlen ((de)->d_name)
		144	#endif
64		145
65	/* used for struct dirent, AIX doesn't provide it */	146	/* used for struct dirent, AIX doesn't provide it */
66	#ifndef NAME_MAX	147	#ifndef NAME_MAX
67	# define NAME_MAX 4096	148	# define NAME_MAX 4096
68	#endif	149	#endif
69		150
		151	/* used for readlink etc. */
		152	#ifndef PATH_MAX
		153	# define PATH_MAX 4096
		154	#endif
		155
70	/* buffer size for various temporary buffers */	156	/* buffer size for various temporary buffers */
71	#define EIO_BUFSIZE 65536	157	#define EIO_BUFSIZE 65536
72		158
73	#define dBUF \	159	#define dBUF \
74	char *eio_buf; \	160	char *eio_buf; \
75	X_LOCK (etplock); \	161	ETP_WORKER_LOCK (self); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	162	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (etplock); \	163	ETP_WORKER_UNLOCK (self); \
78	errno = ENOMEM; \	164	errno = ENOMEM; \
79	if (!eio_buf) \	165	if (!eio_buf) \
80	return -1;	166	return -1;
81		167
82	#define EIO_TICKS ((1000000 + 1023) >> 10)	168	#define EIO_TICKS ((1000000 + 1023) >> 10)
83		169
84	/*****************************************************************************/	170	/*****************************************************************************/
85		171
		172	#if __GNUC__ >= 3
		173	# define expect(expr,value) __builtin_expect ((expr),(value))
		174	#else
		175	# define expect(expr,value) (expr)
		176	#endif
		177
		178	#define expect_false(expr) expect ((expr) != 0, 0)
		179	#define expect_true(expr) expect ((expr) != 0, 1)
		180
		181	/*****************************************************************************/
		182
		183	#define ETP_PRI_MIN EIO_PRI_MIN
		184	#define ETP_PRI_MAX EIO_PRI_MAX
		185
		186	struct etp_worker;
		187
		188	#define ETP_REQ eio_req
		189	#define ETP_DESTROY(req) eio_destroy (req)
		190	static int eio_finish (eio_req *req);
		191	#define ETP_FINISH(req) eio_finish (req)
		192	static void eio_execute (struct etp_worker *self, eio_req *req);
		193	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		194
		195	#define ETP_WORKER_CLEAR(req) \
		196	if (wrk->dbuf) \
		197	{ \
		198	free (wrk->dbuf); \
		199	wrk->dbuf = 0; \
		200	} \
		201	\
		202	if (wrk->dirp) \
		203	{ \
		204	closedir (wrk->dirp); \
		205	wrk->dirp = 0; \
		206	}
		207
		208	#define ETP_WORKER_COMMON \
		209	void *dbuf; \
		210	DIR *dirp;
		211
		212	/*****************************************************************************/
		213
		214	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		215
86	/* calculcate time difference in ~1/EIO_TICKS of a second */	216	/* calculate time difference in ~1/EIO_TICKS of a second */
87	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	217	static int tvdiff (struct timeval *tv1, struct timeval *tv2)
88	{	218	{
89	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	219	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
90	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	220	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
91	}	221	}
92		222
93	static unsigned int started, idle, wanted = 4;	223	static unsigned int started, idle, wanted = 4;
94		224
95	typedef struct etp_pool
96	{
97	void (*want_poll_cb) (void);	225	static void (*want_poll_cb) (void);
98	void (*done_poll_cb) (void);	226	static void (*done_poll_cb) (void);
99		227
100	unsigned int max_poll_time;	228	static unsigned int max_poll_time; /* reslock */
101	unsigned int max_poll_reqs;	229	static unsigned int max_poll_reqs; /* reslock */
102	} etp_pool;
103		230
104	static volatile unsigned int nreqs, nready, npending;	231	static volatile unsigned int nreqs; /* reqlock */
105	static volatile unsigned int max_idle = 4;	232	static volatile unsigned int nready; /* reqlock */
		233	static volatile unsigned int npending; /* reqlock */
		234	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		235	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
106		236
		237	static xmutex_t wrklock;
		238	static xmutex_t reslock;
		239	static xmutex_t reqlock;
		240	static xcond_t reqwait;
		241
		242	#if !HAVE_PREADWRITE
		243	/*
		244	* make our pread/pwrite emulation safe against themselves, but not against
		245	* normal read/write by using a mutex. slows down execution a lot,
		246	* but that's your problem, not mine.
		247	*/
107	static mutex_t etplock = X_MUTEX_INIT;	248	static xmutex_t preadwritelock = X_MUTEX_INIT;
108	static mutex_t reslock = X_MUTEX_INIT;	249	#endif
109	static mutex_t reqlock = X_MUTEX_INIT;
110	static cond_t reqwait = X_COND_INIT;
111		250
112	typedef struct worker	251	typedef struct etp_worker
113	{	252	{
114	/* locked by etplock */	253	/* locked by wrklock */
115	struct worker *prev, *next;	254	struct etp_worker *prev, *next;
116		255
117	thread_t tid;	256	xthread_t tid;
118		257
119	/* locked by reslock, reqlock or etplock */	258	/* locked by reslock, reqlock or wrklock */
120	eio_req *req; /* currently processed request */	259	ETP_REQ *req; /* currently processed request */
121	void *dbuf;	260
122	DIR *dirp;	261	ETP_WORKER_COMMON
123	} worker;	262	} etp_worker;
124		263
125	static worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	264	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
		265
		266	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		267	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
126		268
127	/* worker threads management */	269	/* worker threads management */
128		270
129	static void worker_clear (worker *wrk)	271	static void etp_worker_clear (etp_worker *wrk)
130	{	272	{
131	if (wrk->dirp)	273	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	}	274	}
143		275
144	static void worker_free (worker *wrk)	276	static void etp_worker_free (etp_worker *wrk)
145	{	277	{
146	wrk->next->prev = wrk->prev;	278	wrk->next->prev = wrk->prev;
147	wrk->prev->next = wrk->next;	279	wrk->prev->next = wrk->next;
148		280
149	free (wrk);	281	free (wrk);
150	}	282	}
151		283
152	unsigned int eio_nreqs (void)	284	static unsigned int etp_nreqs (void)
153	{	285	{
		286	int retval;
		287	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		288	retval = nreqs;
		289	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154	return nreqs;	290	return retval;
155	}	291	}
156		292
157	unsigned int eio_nready (void)	293	static unsigned int etp_nready (void)
158	{	294	{
159	unsigned int retval;	295	unsigned int retval;
160		296
161	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	297	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162	retval = nready;	298	retval = nready;
163	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	299	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164		300
165	return retval;	301	return retval;
166	}	302	}
167		303
168	unsigned int eio_npending (void)	304	static unsigned int etp_npending (void)
169	{	305	{
170	unsigned int retval;	306	unsigned int retval;
171		307
172	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	308	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173	retval = npending;	309	retval = npending;
174	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	310	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175		311
176	return retval;	312	return retval;
177	}	313	}
178		314
179	unsigned int eio_nthreads (void)	315	static unsigned int etp_nthreads (void)
180	{	316	{
181	unsigned int retval;	317	unsigned int retval;
182		318
183	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	319	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184	retval = started;	320	retval = started;
…		…
191	* a somewhat faster data structure might be nice, but	327	* a somewhat faster data structure might be nice, but
192	* with 8 priorities this actually needs <20 insns	328	* with 8 priorities this actually needs <20 insns
193	* per shift, the most expensive operation.	329	* per shift, the most expensive operation.
194	*/	330	*/
195	typedef struct {	331	typedef struct {
196	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	332	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197	int size;	333	int size;
198	} reqq;	334	} etp_reqq;
199		335
200	static reqq req_queue;	336	static etp_reqq req_queue;
201	static reqq res_queue;	337	static etp_reqq res_queue;
202		338
203	static int reqq_push (reqq *q, eio_req *req)	339	static int reqq_push (etp_reqq *q, ETP_REQ *req)
204	{	340	{
205	int pri = req->pri;	341	int pri = req->pri;
206	req->next = 0;	342	req->next = 0;
207		343
208	if (q->qe[pri])	344	if (q->qe[pri])
…		…
214	q->qe[pri] = q->qs[pri] = req;	350	q->qe[pri] = q->qs[pri] = req;
215		351
216	return q->size++;	352	return q->size++;
217	}	353	}
218		354
219	static eio_req *reqq_shift (reqq *q)	355	static ETP_REQ *reqq_shift (etp_reqq *q)
220	{	356	{
221	int pri;	357	int pri;
222		358
223	if (!q->size)	359	if (!q->size)
224	return 0;	360	return 0;
225		361
226	--q->size;	362	--q->size;
227		363
228	for (pri = EIO_NUM_PRI; pri--; )	364	for (pri = ETP_NUM_PRI; pri--; )
229	{	365	{
230	eio_req *req = q->qs[pri];	366	eio_req *req = q->qs[pri];
231		367
232	if (req)	368	if (req)
233	{	369	{
…		…
239	}	375	}
240		376
241	abort ();	377	abort ();
242	}	378	}
243		379
		380	static void etp_thread_init (void)
		381	{
		382	X_MUTEX_CREATE (wrklock);
		383	X_MUTEX_CREATE (reslock);
		384	X_MUTEX_CREATE (reqlock);
		385	X_COND_CREATE (reqwait);
		386	}
		387
244	static void etp_atfork_prepare (void)	388	static void etp_atfork_prepare (void)
245	{	389	{
246	X_LOCK (etplock);	390	X_LOCK (wrklock);
247	X_LOCK (reqlock);	391	X_LOCK (reqlock);
248	X_LOCK (reslock);	392	X_LOCK (reslock);
249	#if !HAVE_PREADWRITE	393	#if !HAVE_PREADWRITE
250	X_LOCK (preadwritelock);	394	X_LOCK (preadwritelock);
251	#endif	395	#endif
252	#if !HAVE_READDIR_R
253	X_LOCK (readdirlock);
254	#endif
255	}	396	}
256		397
257	static void etp_atfork_parent (void)	398	static void etp_atfork_parent (void)
258	{	399	{
259	#if !HAVE_READDIR_R
260	X_UNLOCK (readdirlock);
261	#endif
262	#if !HAVE_PREADWRITE	400	#if !HAVE_PREADWRITE
263	X_UNLOCK (preadwritelock);	401	X_UNLOCK (preadwritelock);
264	#endif	402	#endif
265	X_UNLOCK (reslock);	403	X_UNLOCK (reslock);
266	X_UNLOCK (reqlock);	404	X_UNLOCK (reqlock);
267	X_UNLOCK (etplock);	405	X_UNLOCK (wrklock);
268	}	406	}
269		407
270	static void etp_atfork_child (void)	408	static void etp_atfork_child (void)
271	{	409	{
272	eio_req *prv;	410	ETP_REQ *prv;
273		411
274	while (prv = reqq_shift (&req_queue))	412	while ((prv = reqq_shift (&req_queue)))
275	eio_destroy (prv);	413	ETP_DESTROY (prv);
276		414
277	while (prv = reqq_shift (&res_queue))	415	while ((prv = reqq_shift (&res_queue)))
278	eio_destroy (prv);	416	ETP_DESTROY (prv);
279		417
280	while (wrk_first.next != &wrk_first)	418	while (wrk_first.next != &wrk_first)
281	{	419	{
282	worker *wrk = wrk_first.next;	420	etp_worker *wrk = wrk_first.next;
283		421
284	if (wrk->req)	422	if (wrk->req)
285	eio_destroy (wrk->req);	423	ETP_DESTROY (wrk->req);
286		424
287	worker_clear (wrk);	425	etp_worker_clear (wrk);
288	worker_free (wrk);	426	etp_worker_free (wrk);
289	}	427	}
290		428
291	started = 0;	429	started = 0;
292	idle = 0;	430	idle = 0;
293	nreqs = 0;	431	nreqs = 0;
294	nready = 0;	432	nready = 0;
295	npending = 0;	433	npending = 0;
296		434
297	etp_atfork_parent ();	435	etp_thread_init ();
298	}	436	}
299		437
300	static void	438	static void
301	etp_once_init (void)	439	etp_once_init (void)
302	{	440	{
		441	etp_thread_init ();
303	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);	442	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
304	}	443	}
305		444
306	static int	445	static int
307	etp_init (etp_pool *etp, void (*want_poll)(void), void (*done_poll)(void))	446	etp_init (void (*want_poll)(void), void (*done_poll)(void))
308	{	447	{
309	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	448	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
310		449
311	pthread_once (&doinit, etp_once_init);	450	pthread_once (&doinit, etp_once_init);
312		451
313	memset (etp, 0, sizeof *etp);
314
315	etp->want_poll_cb = want_poll;	452	want_poll_cb = want_poll;
316	etp->done_poll_cb = done_poll;	453	done_poll_cb = done_poll;
317	}
318		454
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;	455	return 0;
354	}	456	}
355		457
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);	458	X_THREAD_PROC (etp_proc);
407		459
408	static void start_thread (void)	460	static void etp_start_thread (void)
409	{	461	{
410	worker *wrk = calloc (1, sizeof (worker));	462	etp_worker *wrk = calloc (1, sizeof (etp_worker));
411		463
412	/*TODO*/	464	/*TODO*/
413	assert (("unable to allocate worker thread data", wrk));	465	assert (("unable to allocate worker thread data", wrk));
414		466
415	X_LOCK (etplock);	467	X_LOCK (wrklock);
416		468
417	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	469	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
418	{	470	{
419	wrk->prev = &wrk_first;	471	wrk->prev = &wrk_first;
420	wrk->next = wrk_first.next;	472	wrk->next = wrk_first.next;
421	wrk_first.next->prev = wrk;	473	wrk_first.next->prev = wrk;
422	wrk_first.next = wrk;	474	wrk_first.next = wrk;
423	++started;	475	++started;
424	}	476	}
425	else	477	else
426	free (wrk);	478	free (wrk);
427		479
428	X_UNLOCK (etplock);	480	X_UNLOCK (wrklock);
429	}	481	}
430		482
431	static void maybe_start_thread (void)	483	static void etp_maybe_start_thread (void)
432	{	484	{
433	if (eio_nthreads () >= wanted)	485	if (expect_true (etp_nthreads () >= wanted))
434	return;	486	return;
435		487
436	/* todo: maybe use idle here, but might be less exact */	488	/* todo: maybe use idle here, but might be less exact */
437	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	489	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
438	return;	490	return;
439		491
440	start_thread ();	492	etp_start_thread ();
441	}	493	}
442		494
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)	495	static void etp_end_thread (void)
462	{	496	{
463	eio_req *req = calloc (1, sizeof (eio_req));	497	eio_req *req = calloc (1, sizeof (eio_req));
464		498
465	req->type = EIO_QUIT;	499	req->type = -1;
466	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	500	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
467		501
468	X_LOCK (reqlock);	502	X_LOCK (reqlock);
469	reqq_push (&req_queue, req);	503	reqq_push (&req_queue, req);
470	X_COND_SIGNAL (reqwait);	504	X_COND_SIGNAL (reqwait);
471	X_UNLOCK (reqlock);	505	X_UNLOCK (reqlock);
472		506
473	X_LOCK (etplock);	507	X_LOCK (wrklock);
474	--started;	508	--started;
475	X_UNLOCK (etplock);	509	X_UNLOCK (wrklock);
476	}	510	}
477		511
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)	512	static int etp_poll (void)
515	{	513	{
516	int maxreqs = etp.max_poll_reqs;	514	unsigned int maxreqs;
		515	unsigned int maxtime;
517	struct timeval tv_start, tv_now;	516	struct timeval tv_start, tv_now;
518	eio_req *req;
519		517
520	if (etp.max_poll_time)	518	X_LOCK (reslock);
		519	maxreqs = max_poll_reqs;
		520	maxtime = max_poll_time;
		521	X_UNLOCK (reslock);
		522
		523	if (maxtime)
521	gettimeofday (&tv_start, 0);	524	gettimeofday (&tv_start, 0);
522		525
523	for (;;)	526	for (;;)
524	{	527	{
		528	ETP_REQ *req;
		529
525	maybe_start_thread ();	530	etp_maybe_start_thread ();
526		531
527	X_LOCK (reslock);	532	X_LOCK (reslock);
528	req = reqq_shift (&res_queue);	533	req = reqq_shift (&res_queue);
529		534
530	if (req)	535	if (req)
531	{	536	{
532	--npending;	537	--npending;
533		538
534	if (!res_queue.size && etp.done_poll_cb)	539	if (!res_queue.size && done_poll_cb)
535	etp.done_poll_cb ();	540	done_poll_cb ();
536	}	541	}
537		542
538	X_UNLOCK (reslock);	543	X_UNLOCK (reslock);
539		544
540	if (!req)	545	if (!req)
541	return 0;	546	return 0;
542		547
		548	X_LOCK (reqlock);
543	--nreqs;	549	--nreqs;
		550	X_UNLOCK (reqlock);
544		551
545	if (req->type == EIO_GROUP && req->size)	552	if (expect_false (req->type == EIO_GROUP && req->size))
546	{	553	{
547	req->int1 = 1; /* mark request as delayed */	554	req->int1 = 1; /* mark request as delayed */
548	continue;	555	continue;
549	}	556	}
550	else	557	else
551	{	558	{
552	int res = eio_finish (req);	559	int res = ETP_FINISH (req);
553	if (res)	560	if (expect_false (res))
554	return res;	561	return res;
555	}	562	}
556		563
557	if (maxreqs && !--maxreqs)	564	if (expect_false (maxreqs && !--maxreqs))
558	break;	565	break;
559		566
560	if (etp.max_poll_time)	567	if (maxtime)
561	{	568	{
562	gettimeofday (&tv_now, 0);	569	gettimeofday (&tv_now, 0);
563		570
564	if (tvdiff (&tv_start, &tv_now) >= etp.max_poll_time)	571	if (tvdiff (&tv_start, &tv_now) >= maxtime)
565	break;	572	break;
566	}	573	}
567	}	574	}
568		575
569	errno = EAGAIN;	576	errno = EAGAIN;
570	return -1;	577	return -1;
571	}	578	}
572		579
		580	static void etp_cancel (ETP_REQ *req)
		581	{
		582	X_LOCK (wrklock);
		583	req->flags |= EIO_FLAG_CANCELLED;
		584	X_UNLOCK (wrklock);
		585
		586	eio_grp_cancel (req);
		587	}
		588
		589	static void etp_submit (ETP_REQ *req)
		590	{
		591	req->pri -= ETP_PRI_MIN;
		592
		593	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		594	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		595
		596	if (expect_false (req->type == EIO_GROUP))
		597	{
		598	/* I hope this is worth it :/ */
		599	X_LOCK (reqlock);
		600	++nreqs;
		601	X_UNLOCK (reqlock);
		602
		603	X_LOCK (reslock);
		604
		605	++npending;
		606
		607	if (!reqq_push (&res_queue, req) && want_poll_cb)
		608	want_poll_cb ();
		609
		610	X_UNLOCK (reslock);
		611	}
		612	else
		613	{
		614	X_LOCK (reqlock);
		615	++nreqs;
		616	++nready;
		617	reqq_push (&req_queue, req);
		618	X_COND_SIGNAL (reqwait);
		619	X_UNLOCK (reqlock);
		620
		621	etp_maybe_start_thread ();
		622	}
		623	}
		624
		625	static void etp_set_max_poll_time (double nseconds)
		626	{
		627	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		628	max_poll_time = nseconds * EIO_TICKS;
		629	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		630	}
		631
		632	static void etp_set_max_poll_reqs (unsigned int maxreqs)
		633	{
		634	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		635	max_poll_reqs = maxreqs;
		636	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		637	}
		638
		639	static void etp_set_max_idle (unsigned int nthreads)
		640	{
		641	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		642	max_idle = nthreads;
		643	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		644	}
		645
		646	static void etp_set_idle_timeout (unsigned int seconds)
		647	{
		648	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		649	idle_timeout = seconds;
		650	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		651	}
		652
		653	static void etp_set_min_parallel (unsigned int nthreads)
		654	{
		655	if (wanted < nthreads)
		656	wanted = nthreads;
		657	}
		658
		659	static void etp_set_max_parallel (unsigned int nthreads)
		660	{
		661	if (wanted > nthreads)
		662	wanted = nthreads;
		663
		664	while (started > wanted)
		665	etp_end_thread ();
		666	}
		667
		668	/*****************************************************************************/
		669
		670	static void grp_try_feed (eio_req *grp)
		671	{
		672	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		673	{
		674	grp->flags &= ~EIO_FLAG_GROUPADD;
		675
		676	EIO_FEED (grp);
		677
		678	/* stop if no progress has been made */
		679	if (!(grp->flags & EIO_FLAG_GROUPADD))
		680	{
		681	grp->feed = 0;
		682	break;
		683	}
		684	}
		685	}
		686
		687	static int grp_dec (eio_req *grp)
		688	{
		689	--grp->size;
		690
		691	/* call feeder, if applicable */
		692	grp_try_feed (grp);
		693
		694	/* finish, if done */
		695	if (!grp->size && grp->int1)
		696	return eio_finish (grp);
		697	else
		698	return 0;
		699	}
		700
		701	void eio_destroy (eio_req *req)
		702	{
		703	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		704	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		705
		706	EIO_DESTROY (req);
		707	}
		708
		709	static int eio_finish (eio_req *req)
		710	{
		711	int res = EIO_FINISH (req);
		712
		713	if (req->grp)
		714	{
		715	int res2;
		716	eio_req *grp = req->grp;
		717
		718	/* unlink request */
		719	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		720	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		721
		722	if (grp->grp_first == req)
		723	grp->grp_first = req->grp_next;
		724
		725	res2 = grp_dec (grp);
		726
		727	if (!res && res2)
		728	res = res2;
		729	}
		730
		731	eio_destroy (req);
		732
		733	return res;
		734	}
		735
		736	void eio_grp_cancel (eio_req *grp)
		737	{
		738	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		739	eio_cancel (grp);
		740	}
		741
		742	void eio_cancel (eio_req *req)
		743	{
		744	etp_cancel (req);
		745	}
		746
		747	void eio_submit (eio_req *req)
		748	{
		749	etp_submit (req);
		750	}
		751
		752	unsigned int eio_nreqs (void)
		753	{
		754	return etp_nreqs ();
		755	}
		756
		757	unsigned int eio_nready (void)
		758	{
		759	return etp_nready ();
		760	}
		761
		762	unsigned int eio_npending (void)
		763	{
		764	return etp_npending ();
		765	}
		766
		767	unsigned int eio_nthreads (void)
		768	{
		769	return etp_nthreads ();
		770	}
		771
		772	void eio_set_max_poll_time (double nseconds)
		773	{
		774	etp_set_max_poll_time (nseconds);
		775	}
		776
		777	void eio_set_max_poll_reqs (unsigned int maxreqs)
		778	{
		779	etp_set_max_poll_reqs (maxreqs);
		780	}
		781
		782	void eio_set_max_idle (unsigned int nthreads)
		783	{
		784	etp_set_max_idle (nthreads);
		785	}
		786
		787	void eio_set_idle_timeout (unsigned int seconds)
		788	{
		789	etp_set_idle_timeout (seconds);
		790	}
		791
		792	void eio_set_min_parallel (unsigned int nthreads)
		793	{
		794	etp_set_min_parallel (nthreads);
		795	}
		796
		797	void eio_set_max_parallel (unsigned int nthreads)
		798	{
		799	etp_set_max_parallel (nthreads);
		800	}
		801
		802	int eio_poll (void)
		803	{
		804	return etp_poll ();
		805	}
		806
573	/*****************************************************************************/	807	/*****************************************************************************/
574	/* work around various missing functions */	808	/* work around various missing functions */
575		809
576	#if !HAVE_PREADWRITE	810	#if !HAVE_PREADWRITE
		811	# undef pread
		812	# undef pwrite
577	# define pread eio__pread	813	# define pread eio__pread
578	# define pwrite eio__pwrite	814	# 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		815
587	static ssize_t	816	static ssize_t
588	eio__pread (int fd, void *buf, size_t count, off_t offset)	817	eio__pread (int fd, void *buf, size_t count, off_t offset)
589	{	818	{
590	ssize_t res;	819	ssize_t res;
…		…
608		837
609	X_LOCK (preadwritelock);	838	X_LOCK (preadwritelock);
610	ooffset = lseek (fd, 0, SEEK_CUR);	839	ooffset = lseek (fd, 0, SEEK_CUR);
611	lseek (fd, offset, SEEK_SET);	840	lseek (fd, offset, SEEK_SET);
612	res = write (fd, buf, count);	841	res = write (fd, buf, count);
613	lseek (fd, offset, SEEK_SET);	842	lseek (fd, ooffset, SEEK_SET);
614	X_UNLOCK (preadwritelock);	843	X_UNLOCK (preadwritelock);
615		844
616	return res;	845	return res;
617	}	846	}
618	#endif	847	#endif
619		848
620	#ifndef HAVE_FUTIMES	849	#ifndef HAVE_UTIMES
621		850
		851	# undef utimes
622	# define utimes(path,times) eio__utimes (path, times)	852	# define utimes(path,times) eio__utimes (path, times)
623	# define futimes(fd,times) eio__futimes (fd, times)
624		853
625	static int	854	static int
626	eio__utimes (const char *filename, const struct timeval times[2])	855	eio__utimes (const char *filename, const struct timeval times[2])
627	{	856	{
628	if (times)	857	if (times)
…		…
636	}	865	}
637	else	866	else
638	return utime (filename, 0);	867	return utime (filename, 0);
639	}	868	}
640		869
		870	#endif
		871
		872	#ifndef HAVE_FUTIMES
		873
		874	# undef futimes
		875	# define futimes(fd,times) eio__futimes (fd, times)
		876
641	static int eio__futimes (int fd, const struct timeval tv[2])	877	static int eio__futimes (int fd, const struct timeval tv[2])
642	{	878	{
643	errno = ENOSYS;	879	errno = ENOSYS;
644	return -1;	880	return -1;
645	}	881	}
646		882
647	#endif	883	#endif
648		884
649	#if !HAVE_FDATASYNC	885	#if !HAVE_FDATASYNC
		886	# undef fdatasync
650	# define fdatasync fsync	887	# define fdatasync(fd) fsync (fd)
651	#endif	888	#endif
		889
		890	/* sync_file_range always needs emulation */
		891	int
		892	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		893	{
		894	#if HAVE_SYNC_FILE_RANGE
		895	int res;
		896
		897	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		898	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		899	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		900	{
		901	flags = 0
		902	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		903	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		904	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		905	}
		906
		907	res = sync_file_range (fd, offset, nbytes, flags);
		908
		909	if (!res || errno != ENOSYS)
		910	return res;
		911	#endif
		912
		913	/* even though we could play tricks with the flags, it's better to always
		914	* call fdatasync, as that matches the expectation of its users best */
		915	return fdatasync (fd);
		916	}
652		917
653	#if !HAVE_READAHEAD	918	#if !HAVE_READAHEAD
		919	# undef readahead
654	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	920	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
655		921
656	static ssize_t	922	static ssize_t
657	eio__readahead (int fd, off_t offset, size_t count, worker *self)	923	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
658	{	924	{
659	size_t todo = count;	925	size_t todo = count;
660	dBUF;	926	dBUF;
661		927
662	while (todo > 0)	928	while (todo > 0)
…		…
672	return count;	938	return count;
673	}	939	}
674		940
675	#endif	941	#endif
676		942
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 */	943	/* sendfile always needs emulation */
709	static ssize_t	944	static ssize_t
710	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self)	945	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
711	{	946	{
		947	ssize_t written = 0;
712	ssize_t res;	948	ssize_t res;
713		949
714	if (!count)	950	if (!count)
715	return 0;	951	return 0;
716		952
		953	for (;;)
		954	{
717	#if HAVE_SENDFILE	955	#if HAVE_SENDFILE
718	# if __linux	956	# if __linux
		957	off_t soffset = offset;
719	res = sendfile (ofd, ifd, &offset, count);	958	res = sendfile (ofd, ifd, &soffset, count);
720		959
721	# elif __freebsd	960	# elif __FreeBSD__
722	/*	961	/*
723	* Of course, the freebsd sendfile is a dire hack with no thoughts	962	* Of course, the freebsd sendfile is a dire hack with no thoughts
724	* wasted on making it similar to other I/O functions.	963	* wasted on making it similar to other I/O functions.
725	*/	964	*/
726	{
727	off_t sbytes;	965	off_t sbytes;
728	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	966	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
729		967
730	if (res < 0 && sbytes)	968	#if 0 /* according to the manpage, this is correct, but broken behaviour */
731	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	969	/* freebsd' sendfile will return 0 on success */
		970	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		971	/* not on e.g. EIO or EPIPE - sounds broken */
		972	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
732	res = sbytes;	973	res = sbytes;
733	}	974	#endif
		975
		976	/* according to source inspection, this is correct, and useful behaviour */
		977	if (sbytes)
		978	res = sbytes;
		979
		980	# elif defined (__APPLE__)
		981	off_t sbytes = count;
		982	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		983
		984	/* according to the manpage, sbytes is always valid */
		985	if (sbytes)
		986	res = sbytes;
734		987
735	# elif __hpux	988	# elif __hpux
736	res = sendfile (ofd, ifd, offset, count, 0, 0);	989	res = sendfile (ofd, ifd, offset, count, 0, 0);
737		990
738	# elif __solaris	991	# elif __solaris
739	{
740	struct sendfilevec vec;	992	struct sendfilevec vec;
741	size_t sbytes;	993	size_t sbytes;
742		994
743	vec.sfv_fd = ifd;	995	vec.sfv_fd = ifd;
744	vec.sfv_flag = 0;	996	vec.sfv_flag = 0;
745	vec.sfv_off = offset;	997	vec.sfv_off = offset;
746	vec.sfv_len = count;	998	vec.sfv_len = count;
747		999
748	res = sendfilev (ofd, &vec, 1, &sbytes);	1000	res = sendfilev (ofd, &vec, 1, &sbytes);
749		1001
750	if (res < 0 && sbytes)	1002	if (res < 0 && sbytes)
751	res = sbytes;	1003	res = sbytes;
752	}
753		1004
754	# endif	1005	# endif
		1006
		1007	#elif defined (_WIN32)
		1008	/* does not work, just for documentation of what would need to be done */
		1009	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1010	/* libeio guarantees that the file offset does not change, and windows */
		1011	/* has no way to get an independent handle to the same file description */
		1012	HANDLE h = TO_SOCKET (ifd);
		1013	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1014	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1015
755	#else	1016	#else
756	res = -1;	1017	res = -1;
757	errno = ENOSYS;	1018	errno = ENOSYS;
758	#endif	1019	#endif
759		1020
		1021	/* we assume sendfile can copy at least 128mb in one go */
		1022	if (res <= 128 * 1024 * 1024)
		1023	{
		1024	if (res > 0)
		1025	written += res;
		1026
		1027	if (written)
		1028	return written;
		1029
		1030	break;
		1031	}
		1032	else
		1033	{
		1034	/* if we requested more, then probably the kernel was lazy */
		1035	written += res;
		1036	offset += res;
		1037	count -= res;
		1038
		1039	if (!count)
		1040	return written;
		1041	}
		1042	}
		1043
760	if (res < 0	1044	if (res < 0
761	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1045	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1046	/* BSDs */
		1047	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1048	|| errno == ENOTSUP
		1049	#endif
		1050	|| errno == EOPNOTSUPP /* BSDs */
762	#if __solaris	1051	#if __solaris
763	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1052	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
764	#endif	1053	#endif
765	)	1054	)
766	)	1055	)
…		…
797	}	1086	}
798		1087
799	return res;	1088	return res;
800	}	1089	}
801		1090
		1091	static signed char
		1092	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1093	{
		1094	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1095	: a->inode < b->inode ? -1
		1096	: a->inode > b->inode ? 1
		1097	: 0;
		1098	}
		1099
		1100	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1101
		1102	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1103	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1104
		1105	static void
		1106	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1107	{
		1108	unsigned char bits [9 + sizeof (ino_t) * 8];
		1109	unsigned char *bit = bits;
		1110
		1111	assert (CHAR_BIT == 8);
		1112	assert (sizeof (eio_dirent) * 8 < 256);
		1113	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1114	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1115
		1116	if (size <= EIO_SORT_FAST)
		1117	return;
		1118
		1119	/* first prepare an array of bits to test in our radix sort */
		1120	/* try to take endianness into account, as well as differences in ino_t sizes */
		1121	/* inode_bits must contain all inodes ORed together */
		1122	/* which is used to skip bits that are 0 everywhere, which is very common */
		1123	{
		1124	ino_t endianness;
		1125	int i, j;
		1126
		1127	/* we store the byte offset of byte n into byte n of "endianness" */
		1128	for (i = 0; i < sizeof (ino_t); ++i)
		1129	((unsigned char *)&endianness)[i] = i;
		1130
		1131	*bit++ = 0;
		1132
		1133	for (i = 0; i < sizeof (ino_t); ++i)
		1134	{
		1135	/* shifting off the byte offsets out of "endianness" */
		1136	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1137	endianness >>= 8;
		1138
		1139	for (j = 0; j < 8; ++j)
		1140	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1141	*bit++ = offs + j;
		1142	}
		1143
		1144	for (j = 0; j < 8; ++j)
		1145	if (score_bits & (1 << j))
		1146	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1147	}
		1148
		1149	/* now actually do the sorting (a variant of MSD radix sort) */
		1150	{
		1151	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1152	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1153	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1154	int stk_idx = 0;
		1155
		1156	base_stk [stk_idx] = dents;
		1157	end_stk [stk_idx] = dents + size;
		1158	bit_stk [stk_idx] = bit - 1;
		1159
		1160	do
		1161	{
		1162	base = base_stk [stk_idx];
		1163	end = end_stk [stk_idx];
		1164	bit = bit_stk [stk_idx];
		1165
		1166	for (;;)
		1167	{
		1168	unsigned char O = *bit >> 3;
		1169	unsigned char M = 1 << (*bit & 7);
		1170
		1171	eio_dirent *a = base;
		1172	eio_dirent *b = end;
		1173
		1174	if (b - a < EIO_SORT_CUTOFF)
		1175	break;
		1176
		1177	/* now bit-partition the array on the bit */
		1178	/* this ugly asymmetric loop seems to perform much better than typical */
		1179	/* partition algos found in the literature */
		1180	do
		1181	if (!(((unsigned char *)a)[O] & M))
		1182	++a;
		1183	else if (!(((unsigned char *)--b)[O] & M))
		1184	{
		1185	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1186	++a;
		1187	}
		1188	while (b > a);
		1189
		1190	/* next bit, or stop, if no bits left in this path */
		1191	if (!*--bit)
		1192	break;
		1193
		1194	base_stk [stk_idx] = a;
		1195	end_stk [stk_idx] = end;
		1196	bit_stk [stk_idx] = bit;
		1197	++stk_idx;
		1198
		1199	end = a;
		1200	}
		1201	}
		1202	while (stk_idx--);
		1203	}
		1204	}
		1205
		1206	static void
		1207	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1208	{
		1209	/* first move the smallest element to the front, to act as a sentinel */
		1210	{
		1211	int i;
		1212	eio_dirent *min = dents;
		1213
		1214	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1215	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1216	if (EIO_DENT_CMP (dents [i], <, *min))
		1217	min = &dents [i];
		1218
		1219	/* swap elements 0 and j (minimum) */
		1220	{
		1221	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1222	}
		1223	}
		1224
		1225	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1226	{
		1227	eio_dirent *i, *j;
		1228
		1229	for (i = dents + 1; i < dents + size; ++i)
		1230	{
		1231	eio_dirent value = *i;
		1232
		1233	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1234	j [1] = j [0];
		1235
		1236	j [1] = value;
		1237	}
		1238	}
		1239	}
		1240
		1241	static void
		1242	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1243	{
		1244	if (size <= 1)
		1245	return; /* our insertion sort relies on size > 0 */
		1246
		1247	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1248	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1249	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1250
		1251	/* use an insertion sort at the end, or for small arrays, */
		1252	/* as insertion sort is more efficient for small partitions */
		1253	eio_dent_insertion_sort (dents, size);
		1254	}
		1255
802	/* read a full directory */	1256	/* read a full directory */
803	static void	1257	static void
804	eio__scandir (eio_req *req, worker *self)	1258	eio__scandir (eio_req *req, etp_worker *self)
805	{	1259	{
806	DIR *dirp;	1260	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;	1261	EIO_STRUCT_DIRENT *entp;
813	char *name, *names;	1262	char *name, *names;
814	int memlen = 4096;	1263	int namesalloc = 4096;
815	int memofs = 0;	1264	int namesoffs = 0;
		1265	int flags = req->int1;
		1266	eio_dirent *dents = 0;
		1267	int dentalloc = 128;
816	int res = 0;	1268	int dentoffs = 0;
		1269	ino_t inode_bits = 0;
817		1270
		1271	req->result = -1;
		1272
		1273	if (!(flags & EIO_READDIR_DENTS))
		1274	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
		1275
818	X_LOCK (etplock);	1276	X_LOCK (wrklock);
		1277	/* the corresponding closedir is in ETP_WORKER_CLEAR */
819	self->dirp = dirp = opendir (req->ptr1);	1278	self->dirp = dirp = opendir (req->ptr1);
820	self->dbuf = u = malloc (sizeof (*u));	1279
821	req->flags |= EIO_FLAG_PTR2_FREE;	1280	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1281	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
822	req->ptr2 = names = malloc (memlen);	1282	req->ptr2 = names = malloc (namesalloc);
823	X_UNLOCK (etplock);	1283	X_UNLOCK (wrklock);
824		1284
825	if (dirp && u && names)	1285	if (dirp && names && (!flags || dents))
826	for (;;)	1286	for (;;)
827	{	1287	{
828	errno = 0;	1288	errno = 0;
829	readdir_r (dirp, &u->d, &entp);	1289	entp = readdir (dirp);
830		1290
831	if (!entp)	1291	if (!entp)
		1292	{
		1293	if (errno)
		1294	break;
		1295
		1296	/* sort etc. */
		1297	req->int1 = flags;
		1298	req->result = dentoffs;
		1299
		1300	if (flags & EIO_READDIR_STAT_ORDER)
		1301	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1302	else if (flags & EIO_READDIR_DIRS_FIRST)
		1303	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1304	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1305	else
		1306	{
		1307	/* in this case, all is known, and we just put dirs first and sort them */
		1308	eio_dirent *oth = dents + dentoffs;
		1309	eio_dirent *dir = dents;
		1310
		1311	/* now partition dirs to the front, and non-dirs to the back */
		1312	/* by walking from both sides and swapping if necessary */
		1313	while (oth > dir)
		1314	{
		1315	if (dir->type == EIO_DT_DIR)
		1316	++dir;
		1317	else if ((--oth)->type == EIO_DT_DIR)
		1318	{
		1319	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1320
		1321	++dir;
		1322	}
		1323	}
		1324
		1325	/* now sort the dirs only (dirs all have the same score) */
		1326	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1327	}
		1328
832	break;	1329	break;
		1330	}
833		1331
		1332	/* now add the entry to our list(s) */
834	name = entp->d_name;	1333	name = entp->d_name;
835		1334
		1335	/* skip . and .. entries */
836	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1336	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
837	{	1337	{
838	int len = strlen (name) + 1;	1338	int len = D_NAMLEN (entp) + 1;
839		1339
840	res++;	1340	while (expect_false (namesoffs + len > namesalloc))
841
842	while (memofs + len > memlen)
843	{	1341	{
844	memlen *= 2;	1342	namesalloc *= 2;
845	X_LOCK (etplock);	1343	X_LOCK (wrklock);
846	req->ptr2 = names = realloc (names, memlen);	1344	req->ptr2 = names = realloc (names, namesalloc);
847	X_UNLOCK (etplock);	1345	X_UNLOCK (wrklock);
848		1346
849	if (!names)	1347	if (!names)
850	break;	1348	break;
851	}	1349	}
852		1350
853	memcpy (names + memofs, name, len);	1351	memcpy (names + namesoffs, name, len);
		1352
		1353	if (dents)
		1354	{
		1355	struct eio_dirent *ent;
		1356
		1357	if (expect_false (dentoffs == dentalloc))
		1358	{
		1359	dentalloc *= 2;
		1360	X_LOCK (wrklock);
		1361	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1362	X_UNLOCK (wrklock);
		1363
		1364	if (!dents)
		1365	break;
		1366	}
		1367
		1368	ent = dents + dentoffs;
		1369
		1370	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1371	ent->namelen = len - 1;
		1372	ent->inode = D_INO (entp);
		1373
		1374	inode_bits |= ent->inode;
		1375
		1376	switch (D_TYPE (entp))
		1377	{
		1378	default:
		1379	ent->type = EIO_DT_UNKNOWN;
		1380	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1381	break;
		1382
		1383	#ifdef DT_FIFO
		1384	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1385	#endif
		1386	#ifdef DT_CHR
		1387	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1388	#endif
		1389	#ifdef DT_MPC
		1390	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1391	#endif
		1392	#ifdef DT_DIR
		1393	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1394	#endif
		1395	#ifdef DT_NAM
		1396	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1397	#endif
		1398	#ifdef DT_BLK
		1399	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1400	#endif
		1401	#ifdef DT_MPB
		1402	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1403	#endif
		1404	#ifdef DT_REG
		1405	case DT_REG: ent->type = EIO_DT_REG; break;
		1406	#endif
		1407	#ifdef DT_NWK
		1408	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1409	#endif
		1410	#ifdef DT_CMP
		1411	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1412	#endif
		1413	#ifdef DT_LNK
		1414	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1415	#endif
		1416	#ifdef DT_SOCK
		1417	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1418	#endif
		1419	#ifdef DT_DOOR
		1420	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1421	#endif
		1422	#ifdef DT_WHT
		1423	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1424	#endif
		1425	}
		1426
		1427	ent->score = 7;
		1428
		1429	if (flags & EIO_READDIR_DIRS_FIRST)
		1430	{
		1431	if (ent->type == EIO_DT_UNKNOWN)
		1432	{
		1433	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1434	ent->score = 1;
		1435	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1436	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1437	}
		1438	else if (ent->type == EIO_DT_DIR)
		1439	ent->score = 0;
		1440	}
		1441	}
		1442
854	memofs += len;	1443	namesoffs += len;
		1444	++dentoffs;
		1445	}
		1446
		1447	if (EIO_CANCELLED (req))
		1448	{
		1449	errno = ECANCELED;
		1450	break;
855	}	1451	}
856	}	1452	}
		1453	}
857		1454
858	if (errno)	1455	#ifdef PAGESIZE
859	res = -1;	1456	# define eio_pagesize() PAGESIZE
		1457	#else
		1458	static intptr_t
		1459	eio_pagesize (void)
		1460	{
		1461	static intptr_t page;
		1462
		1463	if (!page)
		1464	page = sysconf (_SC_PAGESIZE);
		1465
		1466	return page;
		1467	}
		1468	#endif
		1469
		1470	static void
		1471	eio_page_align (void **addr, size_t *length)
		1472	{
		1473	intptr_t mask = eio_pagesize () - 1;
		1474
		1475	/* round down addr */
		1476	intptr_t adj = mask & (intptr_t)*addr;
		1477
		1478	*addr = (void *)((intptr_t)*addr - adj);
		1479	*length += adj;
		1480
		1481	/* round up length */
		1482	*length = (*length + mask) & ~mask;
		1483	}
		1484
		1485	#if !_POSIX_MEMLOCK
		1486	# define eio__mlockall(a) ((errno = ENOSYS), -1)
		1487	#else
		1488
		1489	static int
		1490	eio__mlockall (int flags)
		1491	{
		1492	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1493	extern int mallopt (int, int);
		1494	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1495	#endif
		1496
		1497	if (EIO_MCL_CURRENT != MCL_CURRENT
		1498	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1499	{
		1500	flags = 0
		1501	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1502	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1503	}
		1504
		1505	return mlockall (flags);
		1506	}
		1507	#endif
		1508
		1509	#if !_POSIX_MEMLOCK_RANGE
		1510	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
		1511	#else
		1512
		1513	static int
		1514	eio__mlock (void *addr, size_t length)
		1515	{
		1516	eio_page_align (&addr, &length);
		1517
		1518	return mlock (addr, length);
		1519	}
		1520
		1521	#endif
		1522
		1523	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1524	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
		1525	#else
		1526
		1527	int
		1528	eio__msync (void *mem, size_t len, int flags)
		1529	{
		1530	eio_page_align (&mem, &len);
		1531
		1532	if (EIO_MS_ASYNC != MS_SYNC
		1533	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1534	|| EIO_MS_SYNC != MS_SYNC)
		1535	{
		1536	flags = 0
		1537	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1538	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1539	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1540	}
		1541
		1542	return msync (mem, len, flags);
		1543	}
		1544
		1545	#endif
		1546
		1547	int
		1548	eio__mtouch (eio_req *req)
		1549	{
		1550	void *mem = req->ptr2;
		1551	size_t len = req->size;
		1552	int flags = req->int1;
		1553
		1554	eio_page_align (&mem, &len);
		1555
860		1556	{
861	req->result = res;	1557	intptr_t addr = (intptr_t)mem;
		1558	intptr_t end = addr + len;
		1559	intptr_t page = eio_pagesize ();
		1560
		1561	if (addr < end)
		1562	if (flags & EIO_MT_MODIFY) /* modify */
		1563	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1564	else
		1565	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1566	}
		1567
		1568	return 0;
862	}	1569	}
863		1570
864	/*****************************************************************************/	1571	/*****************************************************************************/
865		1572
866	#define ALLOC(len) \	1573	#define ALLOC(len) \
867	if (!req->ptr2) \	1574	if (!req->ptr2) \
868	{ \	1575	{ \
869	X_LOCK (etplock); \	1576	X_LOCK (wrklock); \
870	req->flags |= EIO_FLAG_PTR2_FREE; \	1577	req->flags |= EIO_FLAG_PTR2_FREE; \
871	X_UNLOCK (etplock); \	1578	X_UNLOCK (wrklock); \
872	req->ptr2 = malloc (len); \	1579	req->ptr2 = malloc (len); \
873	if (!req->ptr2) \	1580	if (!req->ptr2) \
874	{ \	1581	{ \
875	errno = ENOMEM; \	1582	errno = ENOMEM; \
876	req->result = -1; \	1583	req->result = -1; \
877	break; \	1584	break; \
878	} \	1585	} \
879	}	1586	}
880		1587
881	X_THREAD_PROC (eio_proc)	1588	X_THREAD_PROC (etp_proc)
882	{	1589	{
883	eio_req *req;	1590	ETP_REQ *req;
884	struct timespec ts;	1591	struct timespec ts;
885	worker *self = (worker *)thr_arg;	1592	etp_worker *self = (etp_worker *)thr_arg;
886		1593
887	/* try to distribute timeouts somewhat randomly */	1594	/* try to distribute timeouts somewhat randomly */
888	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1595	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
889		1596
890	for (;;)	1597	for (;;)
891	{	1598	{
892	ts.tv_sec = time (0) + IDLE_TIMEOUT;
893
894	X_LOCK (reqlock);	1599	X_LOCK (reqlock);
895		1600
896	for (;;)	1601	for (;;)
897	{	1602	{
898	self->req = req = reqq_shift (&req_queue);	1603	self->req = req = reqq_shift (&req_queue);
…		…
900	if (req)	1605	if (req)
901	break;	1606	break;
902		1607
903	++idle;	1608	++idle;
904		1609
		1610	ts.tv_sec = time (0) + idle_timeout;
905	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1611	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
906	{	1612	{
907	if (idle > max_idle)	1613	if (idle > max_idle)
908	{	1614	{
909	--idle;	1615	--idle;
910	X_UNLOCK (reqlock);	1616	X_UNLOCK (reqlock);
911	X_LOCK (etplock);	1617	X_LOCK (wrklock);
912	--started;	1618	--started;
913	X_UNLOCK (etplock);	1619	X_UNLOCK (wrklock);
914	goto quit;	1620	goto quit;
915	}	1621	}
916		1622
917	/* we are allowed to idle, so do so without any timeout */	1623	/* we are allowed to idle, so do so without any timeout */
918	X_COND_WAIT (reqwait, reqlock);	1624	X_COND_WAIT (reqwait, reqlock);
919	ts.tv_sec = time (0) + IDLE_TIMEOUT;
920	}	1625	}
921		1626
922	--idle;	1627	--idle;
923	}	1628	}
924		1629
925	--nready;	1630	--nready;
926		1631
927	X_UNLOCK (reqlock);	1632	X_UNLOCK (reqlock);
928		1633
929	errno = 0; /* strictly unnecessary */	1634	if (req->type < 0)
		1635	goto quit;
930		1636
931	if (!EIO_CANCELLED (req))	1637	if (!EIO_CANCELLED (req))
932	switch (req->type)	1638	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		1639
1033	X_LOCK (reslock);	1640	X_LOCK (reslock);
1034		1641
1035	++npending;	1642	++npending;
1036		1643
1037	if (!reqq_push (&res_queue, req) && etp.want_poll_cb)	1644	if (!reqq_push (&res_queue, req) && want_poll_cb)
1038	etp.want_poll_cb ();	1645	want_poll_cb ();
1039		1646
1040	self->req = 0;	1647	self->req = 0;
1041	worker_clear (self);	1648	etp_worker_clear (self);
1042		1649
1043	X_UNLOCK (reslock);	1650	X_UNLOCK (reslock);
1044	}	1651	}
1045		1652
1046	quit:	1653	quit:
1047	X_LOCK (etplock);	1654	X_LOCK (wrklock);
1048	worker_free (self);	1655	etp_worker_free (self);
1049	X_UNLOCK (etplock);	1656	X_UNLOCK (wrklock);
1050		1657
1051	return 0;	1658	return 0;
1052	}	1659	}
1053		1660
1054	/*****************************************************************************/	1661	/*****************************************************************************/
1055		1662
1056	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1663	int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1057	{	1664	{
1058	etp_init (&etp, want_poll, done_poll);	1665	return etp_init (want_poll, done_poll);
1059	}	1666	}
1060		1667
1061	static void eio_api_destroy (eio_req *req)	1668	static void eio_api_destroy (eio_req *req)
1062	{	1669	{
1063	free (req);	1670	free (req);
…		…
1085	{ \	1692	{ \
1086	eio_api_destroy (req); \	1693	eio_api_destroy (req); \
1087	return 0; \	1694	return 0; \
1088	}	1695	}
1089		1696
		1697	static void eio_execute (etp_worker *self, eio_req *req)
		1698	{
		1699	switch (req->type)
		1700	{
		1701	case EIO_READ: ALLOC (req->size);
		1702	req->result = req->offs >= 0
		1703	? pread (req->int1, req->ptr2, req->size, req->offs)
		1704	: read (req->int1, req->ptr2, req->size); break;
		1705	case EIO_WRITE: req->result = req->offs >= 0
		1706	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1707	: write (req->int1, req->ptr2, req->size); break;
		1708
		1709	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		1710	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
		1711
		1712	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1713	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1714	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1715	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1716	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1717	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1718
		1719	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1720	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1721	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1722	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1723
		1724	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1725	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1726	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1727	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1728	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1729	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1730
		1731	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1732	case EIO_CLOSE: req->result = close (req->int1); break;
		1733	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1734	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1735	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1736	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1737	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1738	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1739	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1740	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		1741
		1742	case EIO_READLINK: ALLOC (PATH_MAX);
		1743	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
		1744
		1745	case EIO_SYNC: req->result = 0; sync (); break;
		1746	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1747	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1748	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1749	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		1750	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		1751	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		1752	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1753
		1754	case EIO_READDIR: eio__scandir (req, self); break;
		1755
		1756	case EIO_BUSY:
		1757	#ifdef _WIN32
		1758	Sleep (req->nv1 * 1e3);
		1759	#else
		1760	{
		1761	struct timeval tv;
		1762
		1763	tv.tv_sec = req->nv1;
		1764	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
		1765
		1766	req->result = select (0, 0, 0, 0, &tv);
		1767	}
		1768	#endif
		1769	break;
		1770
		1771	case EIO_UTIME:
		1772	case EIO_FUTIME:
		1773	{
		1774	struct timeval tv[2];
		1775	struct timeval *times;
		1776
		1777	if (req->nv1 != -1. || req->nv2 != -1.)
		1778	{
		1779	tv[0].tv_sec = req->nv1;
		1780	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		1781	tv[1].tv_sec = req->nv2;
		1782	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		1783
		1784	times = tv;
		1785	}
		1786	else
		1787	times = 0;
		1788
		1789	req->result = req->type == EIO_FUTIME
		1790	? futimes (req->int1, times)
		1791	: utimes (req->ptr1, times);
		1792	}
		1793	break;
		1794
		1795	case EIO_GROUP:
		1796	abort (); /* handled in eio_request */
		1797
		1798	case EIO_NOP:
		1799	req->result = 0;
		1800	break;
		1801
		1802	case EIO_CUSTOM:
		1803	req->feed (req);
		1804	break;
		1805
		1806	default:
		1807	errno = ENOSYS;
		1808	req->result = -1;
		1809	break;
		1810	}
		1811
		1812	req->errorno = errno;
		1813	}
		1814
1090	#ifndef EIO_NO_WRAPPERS	1815	#ifndef EIO_NO_WRAPPERS
1091		1816
1092	eio_req *eio_nop (int pri, eio_cb cb, void *data)	1817	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1093	{	1818	{
1094	REQ (EIO_NOP); SEND;	1819	REQ (EIO_NOP); SEND;
…		…
1107	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	1832	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1108	{	1833	{
1109	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1834	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1110	}	1835	}
1111		1836
		1837	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1838	{
		1839	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1840	}
		1841
		1842	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1843	{
		1844	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1845	}
		1846
		1847	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		1848	{
		1849	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		1850	}
		1851
		1852	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		1853	{
		1854	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		1855	}
		1856
		1857	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		1858	{
		1859	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		1860	}
		1861
1112	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	1862	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1113	{	1863	{
1114	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1864	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1115	}	1865	}
1116		1866
…		…
1135	}	1885	}
1136		1886
1137	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	1887	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1138	{	1888	{
1139	REQ (EIO_FSTAT); req->int1 = fd; SEND;	1889	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1890	}
		1891
		1892	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		1893	{
		1894	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1140	}	1895	}
1141		1896
1142	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	1897	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1143	{	1898	{
1144	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	1899	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)	1973	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1219	{	1974	{
1220	return eio__1path (EIO_LSTAT, path, pri, cb, data);	1975	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1221	}	1976	}
1222		1977
		1978	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		1979	{
		1980	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		1981	}
		1982
1223	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	1983	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1224	{	1984	{
1225	return eio__1path (EIO_UNLINK, path, pri, cb, data);	1985	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1226	}	1986	}
1227		1987
1228	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	1988	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1229	{	1989	{
1230	return eio__1path (EIO_RMDIR, path, pri, cb, data);	1990	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1231	}	1991	}
1232		1992
1233	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	1993	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1234	{	1994	{
1235	return eio__1path (EIO_READDIR, path, pri, cb, data);	1995	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1236	}	1996	}
1237		1997
1238	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	1998	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1239	{	1999	{
1240	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	2000	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1241	}	2001	}
1242		2002
1243	static eio_req *	2003	static eio_req *
1244	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2004	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1245	{	2005	{
…		…
1269	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2029	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1270	{	2030	{
1271	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2031	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1272	}	2032	}
1273		2033
		2034	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
		2035	{
		2036	REQ (EIO_CUSTOM); req->feed = execute; SEND;
		2037	}
		2038
1274	#endif	2039	#endif
1275		2040
1276	eio_req *eio_grp (eio_cb cb, void *data)	2041	eio_req *eio_grp (eio_cb cb, void *data)
1277	{	2042	{
1278	const int pri = EIO_PRI_MAX;	2043	const int pri = EIO_PRI_MAX;
…		…
1304		2069
1305	void eio_grp_add (eio_req *grp, eio_req *req)	2070	void eio_grp_add (eio_req *grp, eio_req *req)
1306	{	2071	{
1307	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2072	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1308		2073
		2074	grp->flags |= EIO_FLAG_GROUPADD;
		2075
1309	++grp->size;	2076	++grp->size;
1310	req->grp = grp;	2077	req->grp = grp;
1311		2078
1312	req->grp_prev = 0;	2079	req->grp_prev = 0;
1313	req->grp_next = grp->grp_first;	2080	req->grp_next = grp->grp_first;
…		…
1321	/*****************************************************************************/	2088	/*****************************************************************************/
1322	/* misc garbage */	2089	/* misc garbage */
1323		2090
1324	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2091	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1325	{	2092	{
1326	worker wrk;	2093	etp_worker wrk;
		2094	ssize_t ret;
1327		2095
1328	wrk.dbuf = 0;	2096	wrk.dbuf = 0;
1329		2097
1330	eio__sendfile (ofd, ifd, offset, count, &wrk);	2098	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1331		2099
1332	if (wrk.dbuf)	2100	if (wrk.dbuf)
1333	free (wrk.dbuf);	2101	free (wrk.dbuf);
1334	}
1335		2102
		2103	return ret;
		2104	}
		2105

Diff Legend

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