ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/cvsroot/libeio/eio.c

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.11 by root, Tue May 13 18:50:27 2008 UTC vs.
Revision 1.52 by root, Fri Jan 8 09:29:56 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 (wrklock); \	147	ETP_WORKER_LOCK (self); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	148	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (wrklock); \	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
		156	/*****************************************************************************/
		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
		202	/* calculate time difference in ~1/EIO_TICKS of a second */
		203	static int tvdiff (struct timeval *tv1, struct timeval *tv2)
		204	{
		205	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
		206	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
		207	}
		208
		209	static unsigned int started, idle, wanted = 4;
		210
84	static void (*want_poll_cb) (void);	211	static void (*want_poll_cb) (void);
85	static void (*done_poll_cb) (void);	212	static void (*done_poll_cb) (void);
86		213
87	static unsigned int max_poll_time = 0;	214	static unsigned int max_poll_time; /* reslock */
88	static unsigned int max_poll_reqs = 0;	215	static unsigned int max_poll_reqs; /* reslock */
89		216
90	/* calculcate time difference in ~1/EIO_TICKS of a second */	217	static volatile unsigned int nreqs; /* reqlock */
91	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	218	static volatile unsigned int nready; /* reqlock */
92	{	219	static volatile unsigned int npending; /* reqlock */
93	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	220	static volatile unsigned int max_idle = 4;
94	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
95	}
96		221
97	static unsigned int started, idle, wanted = 4;
98
99	/* worker threads management */
100	static mutex_t wrklock = X_MUTEX_INIT;	222	static mutex_t wrklock = X_MUTEX_INIT;
101
102	typedef struct worker
103	{
104	/* locked by wrklock */
105	struct worker *prev, *next;
106
107	thread_t tid;
108
109	/* locked by reslock, reqlock or wrklock */
110	eio_req *req; /* currently processed request */
111	void *dbuf;
112	DIR *dirp;
113	} worker;
114
115	static worker wrk_first = { &wrk_first, &wrk_first, 0 };
116
117	static void worker_clear (worker *wrk)
118	{
119	if (wrk->dirp)
120	{
121	closedir (wrk->dirp);
122	wrk->dirp = 0;
123	}
124
125	if (wrk->dbuf)
126	{
127	free (wrk->dbuf);
128	wrk->dbuf = 0;
129	}
130	}
131
132	static void worker_free (worker *wrk)
133	{
134	wrk->next->prev = wrk->prev;
135	wrk->prev->next = wrk->next;
136
137	free (wrk);
138	}
139
140	static volatile unsigned int nreqs, nready, npending;
141	static volatile unsigned int max_idle = 4;
142
143	static mutex_t reslock = X_MUTEX_INIT;	223	static mutex_t reslock = X_MUTEX_INIT;
144	static mutex_t reqlock = X_MUTEX_INIT;	224	static mutex_t reqlock = X_MUTEX_INIT;
145	static cond_t reqwait = X_COND_INIT;	225	static cond_t reqwait = X_COND_INIT;
146		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
		236	typedef struct etp_worker
		237	{
		238	/* locked by wrklock */
		239	struct etp_worker *prev, *next;
		240
		241	thread_t tid;
		242
		243	/* locked by reslock, reqlock or wrklock */
		244	ETP_REQ *req; /* currently processed request */
		245
		246	ETP_WORKER_COMMON
		247	} etp_worker;
		248
		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)
		253
		254	/* worker threads management */
		255
		256	static void etp_worker_clear (etp_worker *wrk)
		257	{
		258	ETP_WORKER_CLEAR (wrk);
		259	}
		260
		261	static void etp_worker_free (etp_worker *wrk)
		262	{
		263	wrk->next->prev = wrk->prev;
		264	wrk->prev->next = wrk->next;
		265
		266	free (wrk);
		267	}
		268
147	unsigned int eio_nreqs (void)	269	static unsigned int etp_nreqs (void)
148	{	270	{
		271	int retval;
		272	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		273	retval = nreqs;
		274	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
149	return nreqs;	275	return retval;
150	}	276	}
151		277
152	unsigned int eio_nready (void)	278	static unsigned int etp_nready (void)
153	{	279	{
154	unsigned int retval;	280	unsigned int retval;
155		281
156	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	282	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157	retval = nready;	283	retval = nready;
158	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	284	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159		285
160	return retval;	286	return retval;
161	}	287	}
162		288
163	unsigned int eio_npending (void)	289	static unsigned int etp_npending (void)
164	{	290	{
165	unsigned int retval;	291	unsigned int retval;
166		292
167	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	293	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168	retval = npending;	294	retval = npending;
169	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	295	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170		296
171	return retval;	297	return retval;
172	}	298	}
173		299
174	unsigned int eio_nthreads (void)	300	static unsigned int etp_nthreads (void)
175	{	301	{
176	unsigned int retval;	302	unsigned int retval;
177		303
178	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	304	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179	retval = started;	305	retval = started;
…		…
186	* a somewhat faster data structure might be nice, but	312	* a somewhat faster data structure might be nice, but
187	* with 8 priorities this actually needs <20 insns	313	* with 8 priorities this actually needs <20 insns
188	* per shift, the most expensive operation.	314	* per shift, the most expensive operation.
189	*/	315	*/
190	typedef struct {	316	typedef struct {
191	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 */
192	int size;	318	int size;
193	} reqq;	319	} etp_reqq;
194		320
195	static reqq req_queue;	321	static etp_reqq req_queue;
196	static reqq res_queue;	322	static etp_reqq res_queue;
197		323
198	static int reqq_push (reqq *q, eio_req *req)	324	static int reqq_push (etp_reqq *q, ETP_REQ *req)
199	{	325	{
200	int pri = req->pri;	326	int pri = req->pri;
201	req->next = 0;	327	req->next = 0;
202		328
203	if (q->qe[pri])	329	if (q->qe[pri])
…		…
209	q->qe[pri] = q->qs[pri] = req;	335	q->qe[pri] = q->qs[pri] = req;
210		336
211	return q->size++;	337	return q->size++;
212	}	338	}
213		339
214	static eio_req *reqq_shift (reqq *q)	340	static ETP_REQ *reqq_shift (etp_reqq *q)
215	{	341	{
216	int pri;	342	int pri;
217		343
218	if (!q->size)	344	if (!q->size)
219	return 0;	345	return 0;
220		346
221	--q->size;	347	--q->size;
222		348
223	for (pri = EIO_NUM_PRI; pri--; )	349	for (pri = ETP_NUM_PRI; pri--; )
224	{	350	{
225	eio_req *req = q->qs[pri];	351	eio_req *req = q->qs[pri];
226		352
227	if (req)	353	if (req)
228	{	354	{
…		…
234	}	360	}
235		361
236	abort ();	362	abort ();
237	}	363	}
238		364
239	static void grp_try_feed (eio_req *grp)	365	static void etp_atfork_prepare (void)
240	{	366	{
241	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	367	X_LOCK (wrklock);
		368	X_LOCK (reqlock);
		369	X_LOCK (reslock);
		370	#if !HAVE_PREADWRITE
		371	X_LOCK (preadwritelock);
		372	#endif
		373	}
		374
		375	static void etp_atfork_parent (void)
		376	{
		377	#if !HAVE_PREADWRITE
		378	X_UNLOCK (preadwritelock);
		379	#endif
		380	X_UNLOCK (reslock);
		381	X_UNLOCK (reqlock);
		382	X_UNLOCK (wrklock);
		383	}
		384
		385	static void etp_atfork_child (void)
		386	{
		387	ETP_REQ *prv;
		388
		389	while ((prv = reqq_shift (&req_queue)))
		390	ETP_DESTROY (prv);
		391
		392	while ((prv = reqq_shift (&res_queue)))
		393	ETP_DESTROY (prv);
		394
		395	while (wrk_first.next != &wrk_first)
242	{	396	{
243	int old_len = grp->size;	397	etp_worker *wrk = wrk_first.next;
244		398
245	EIO_FEED (grp);	399	if (wrk->req)
		400	ETP_DESTROY (wrk->req);
246		401
247	/* stop if no progress has been made */	402	etp_worker_clear (wrk);
248	if (old_len == grp->size)	403	etp_worker_free (wrk);
249	{
250	grp->feed = 0;
251	break;
252	}
253	}	404	}
254	}
255		405
256	static int eio_finish (eio_req *req);	406	started = 0;
		407	idle = 0;
		408	nreqs = 0;
		409	nready = 0;
		410	npending = 0;
257		411
258	static int grp_dec (eio_req *grp)	412	etp_atfork_parent ();
259	{	413	}
260	--grp->size;
261		414
262	/* call feeder, if applicable */	415	static void
263	grp_try_feed (grp);	416	etp_once_init (void)
		417	{
		418	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
		419	}
264		420
265	/* finish, if done */	421	static int
266	if (!grp->size && grp->int1)	422	etp_init (void (*want_poll)(void), void (*done_poll)(void))
267	return eio_finish (grp);	423	{
268	else	424	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
		425
		426	pthread_once (&doinit, etp_once_init);
		427
		428	want_poll_cb = want_poll;
		429	done_poll_cb = done_poll;
		430
269	return 0;	431	return 0;
270	}	432	}
271		433
272	void eio_destroy (eio_req *req)
273	{
274	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
275	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
276
277	EIO_DESTROY (req);
278	}
279
280	static int eio_finish (eio_req *req)
281	{
282	int res = EIO_FINISH (req);
283
284	if (req->grp)
285	{
286	int res2;
287	eio_req *grp = req->grp;
288
289	/* unlink request */
290	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
291	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
292
293	if (grp->grp_first == req)
294	grp->grp_first = req->grp_next;
295
296	res2 = grp_dec (grp);
297
298	if (!res && res2)
299	res = res2;
300	}
301
302	eio_destroy (req);
303
304	return res;
305	}
306
307	void eio_grp_cancel (eio_req *grp)
308	{
309	for (grp = grp->grp_first; grp; grp = grp->grp_next)
310	eio_cancel (grp);
311	}
312
313	void eio_cancel (eio_req *req)
314	{
315	X_LOCK (wrklock);
316	req->flags |= EIO_FLAG_CANCELLED;
317	X_UNLOCK (wrklock);
318
319	eio_grp_cancel (req);
320	}
321
322	X_THREAD_PROC (eio_proc);	434	X_THREAD_PROC (etp_proc);
323		435
324	static void start_thread (void)	436	static void etp_start_thread (void)
325	{	437	{
326	worker *wrk = calloc (1, sizeof (worker));	438	etp_worker *wrk = calloc (1, sizeof (etp_worker));
327		439
328	/*TODO*/	440	/*TODO*/
329	assert (("unable to allocate worker thread data", wrk));	441	assert (("unable to allocate worker thread data", wrk));
330		442
331	X_LOCK (wrklock);	443	X_LOCK (wrklock);
332		444
333	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	445	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334	{	446	{
335	wrk->prev = &wrk_first;	447	wrk->prev = &wrk_first;
336	wrk->next = wrk_first.next;	448	wrk->next = wrk_first.next;
337	wrk_first.next->prev = wrk;	449	wrk_first.next->prev = wrk;
338	wrk_first.next = wrk;	450	wrk_first.next = wrk;
…		…
342	free (wrk);	454	free (wrk);
343		455
344	X_UNLOCK (wrklock);	456	X_UNLOCK (wrklock);
345	}	457	}
346		458
347	static void maybe_start_thread (void)	459	static void etp_maybe_start_thread (void)
348	{	460	{
349	if (eio_nthreads () >= wanted)	461	if (expect_true (etp_nthreads () >= wanted))
350	return;	462	return;
351		463
352	/* todo: maybe use idle here, but might be less exact */	464	/* todo: maybe use idle here, but might be less exact */
353	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 ()))
354	return;	466	return;
355		467
356	start_thread ();	468	etp_start_thread ();
357	}	469	}
358		470
359	void eio_submit (eio_req *req)
360	{
361	req->pri += EIO_PRI_BIAS;
362
363	if (req->pri < EIO_PRI_MIN + EIO_PRI_BIAS) req->pri = EIO_PRI_MIN + EIO_PRI_BIAS;
364	if (req->pri > EIO_PRI_MAX + EIO_PRI_BIAS) req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
365
366	++nreqs;
367
368	X_LOCK (reqlock);
369	++nready;
370	reqq_push (&req_queue, req);
371	X_COND_SIGNAL (reqwait);
372	X_UNLOCK (reqlock);
373
374	maybe_start_thread ();
375	}
376
377	static void end_thread (void)	471	static void etp_end_thread (void)
378	{	472	{
379	eio_req *req = calloc (1, sizeof (eio_req));	473	eio_req *req = calloc (1, sizeof (eio_req));
380		474
381	req->type = EIO_QUIT;	475	req->type = -1;
382	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	476	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
383		477
384	X_LOCK (reqlock);	478	X_LOCK (reqlock);
385	reqq_push (&req_queue, req);	479	reqq_push (&req_queue, req);
386	X_COND_SIGNAL (reqwait);	480	X_COND_SIGNAL (reqwait);
387	X_UNLOCK (reqlock);	481	X_UNLOCK (reqlock);
…		…
389	X_LOCK (wrklock);	483	X_LOCK (wrklock);
390	--started;	484	--started;
391	X_UNLOCK (wrklock);	485	X_UNLOCK (wrklock);
392	}	486	}
393		487
394	void eio_set_max_poll_time (double nseconds)
395	{
396	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
397	max_poll_time = nseconds;
398	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
399	}
400
401	void eio_set_max_poll_reqs (unsigned int maxreqs)
402	{
403	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
404	max_poll_reqs = maxreqs;
405	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
406	}
407
408	void eio_set_max_idle (unsigned int nthreads)
409	{
410	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
411	max_idle = nthreads <= 0 ? 1 : nthreads;
412	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
413	}
414
415	void eio_set_min_parallel (unsigned int nthreads)
416	{
417	if (wanted < nthreads)
418	wanted = nthreads;
419	}
420
421	void eio_set_max_parallel (unsigned int nthreads)
422	{
423	if (wanted > nthreads)
424	wanted = nthreads;
425
426	while (started > wanted)
427	end_thread ();
428	}
429
430	int eio_poll (void)	488	static int etp_poll (void)
431	{	489	{
432	int maxreqs = max_poll_reqs;	490	unsigned int maxreqs;
		491	unsigned int maxtime;
433	struct timeval tv_start, tv_now;	492	struct timeval tv_start, tv_now;
434	eio_req *req;
435		493
		494	X_LOCK (reslock);
		495	maxreqs = max_poll_reqs;
		496	maxtime = max_poll_time;
		497	X_UNLOCK (reslock);
		498
436	if (max_poll_time)	499	if (maxtime)
437	gettimeofday (&tv_start, 0);	500	gettimeofday (&tv_start, 0);
438		501
439	for (;;)	502	for (;;)
440	{	503	{
		504	ETP_REQ *req;
		505
441	maybe_start_thread ();	506	etp_maybe_start_thread ();
442		507
443	X_LOCK (reslock);	508	X_LOCK (reslock);
444	req = reqq_shift (&res_queue);	509	req = reqq_shift (&res_queue);
445		510
446	if (req)	511	if (req)
…		…
454	X_UNLOCK (reslock);	519	X_UNLOCK (reslock);
455		520
456	if (!req)	521	if (!req)
457	return 0;	522	return 0;
458		523
		524	X_LOCK (reqlock);
459	--nreqs;	525	--nreqs;
		526	X_UNLOCK (reqlock);
460		527
461	if (req->type == EIO_GROUP && req->size)	528	if (expect_false (req->type == EIO_GROUP && req->size))
462	{	529	{
463	req->int1 = 1; /* mark request as delayed */	530	req->int1 = 1; /* mark request as delayed */
464	continue;	531	continue;
465	}	532	}
466	else	533	else
467	{	534	{
468	int res = eio_finish (req);	535	int res = ETP_FINISH (req);
469	if (res)	536	if (expect_false (res))
470	return res;	537	return res;
471	}	538	}
472		539
473	if (maxreqs && !--maxreqs)	540	if (expect_false (maxreqs && !--maxreqs))
474	break;	541	break;
475		542
476	if (max_poll_time)	543	if (maxtime)
477	{	544	{
478	gettimeofday (&tv_now, 0);	545	gettimeofday (&tv_now, 0);
479		546
480	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	547	if (tvdiff (&tv_start, &tv_now) >= maxtime)
481	break;	548	break;
482	}	549	}
483	}	550	}
484		551
485	errno = EAGAIN;	552	errno = EAGAIN;
486	return -1;	553	return -1;
487	}	554	}
488		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
489	/*****************************************************************************/	771	/*****************************************************************************/
490	/* work around various missing functions */	772	/* work around various missing functions */
491		773
492	#if !HAVE_PREADWRITE	774	#if !HAVE_PREADWRITE
		775	# undef pread
		776	# undef pwrite
493	# define pread eio__pread	777	# define pread eio__pread
494	# define pwrite eio__pwrite	778	# define pwrite eio__pwrite
495
496	/*
497	* make our pread/pwrite safe against themselves, but not against
498	* normal read/write by using a mutex. slows down execution a lot,
499	* but that's your problem, not mine.
500	*/
501	static mutex_t preadwritelock = X_MUTEX_INIT;
502		779
503	static ssize_t	780	static ssize_t
504	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)
505	{	782	{
506	ssize_t res;	783	ssize_t res;
…		…
524		801
525	X_LOCK (preadwritelock);	802	X_LOCK (preadwritelock);
526	ooffset = lseek (fd, 0, SEEK_CUR);	803	ooffset = lseek (fd, 0, SEEK_CUR);
527	lseek (fd, offset, SEEK_SET);	804	lseek (fd, offset, SEEK_SET);
528	res = write (fd, buf, count);	805	res = write (fd, buf, count);
529	lseek (fd, offset, SEEK_SET);	806	lseek (fd, ooffset, SEEK_SET);
530	X_UNLOCK (preadwritelock);	807	X_UNLOCK (preadwritelock);
531		808
532	return res;	809	return res;
533	}	810	}
534	#endif	811	#endif
535		812
536	#ifndef HAVE_FUTIMES	813	#ifndef HAVE_FUTIMES
537		814
		815	# undef utimes
		816	# undef futimes
538	# define utimes(path,times) eio__utimes (path, times)	817	# define utimes(path,times) eio__utimes (path, times)
539	# define futimes(fd,times) eio__futimes (fd, times)	818	# define futimes(fd,times) eio__futimes (fd, times)
540		819
541	static int	820	static int
542	eio__utimes (const char *filename, const struct timeval times[2])	821	eio__utimes (const char *filename, const struct timeval times[2])
…		…
561	}	840	}
562		841
563	#endif	842	#endif
564		843
565	#if !HAVE_FDATASYNC	844	#if !HAVE_FDATASYNC
		845	# undef fdatasync
566	# define fdatasync fsync	846	# define fdatasync(fd) fsync (fd)
567	#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	}
568		876
569	#if !HAVE_READAHEAD	877	#if !HAVE_READAHEAD
		878	# undef readahead
570	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	879	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
571		880
572	static ssize_t	881	static ssize_t
573	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)
574	{	883	{
575	size_t todo = count;	884	size_t todo = count;
576	dBUF;	885	dBUF;
577		886
578	while (todo > 0)	887	while (todo > 0)
…		…
588	return count;	897	return count;
589	}	898	}
590		899
591	#endif	900	#endif
592		901
593	#if !HAVE_READDIR_R
594	# define readdir_r eio__readdir_r
595
596	static mutex_t readdirlock = X_MUTEX_INIT;
597
598	static int
599	eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
600	{
601	EIO_STRUCT_DIRENT *e;
602	int errorno;
603
604	X_LOCK (readdirlock);
605
606	e = readdir (dirp);
607	errorno = errno;
608
609	if (e)
610	{
611	*res = ent;
612	strcpy (ent->d_name, e->d_name);
613	}
614	else
615	*res = 0;
616
617	X_UNLOCK (readdirlock);
618
619	errno = errorno;
620	return e ? 0 : -1;
621	}
622	#endif
623
624	/* sendfile always needs emulation */	902	/* sendfile always needs emulation */
625	static ssize_t	903	static ssize_t
626	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)
627	{	905	{
628	ssize_t res;	906	ssize_t res;
629		907
630	if (!count)	908	if (!count)
631	return 0;	909	return 0;
632		910
633	#if HAVE_SENDFILE	911	#if HAVE_SENDFILE
634	# if __linux	912	# if __linux
635	res = sendfile (ofd, ifd, &offset, count);	913	res = sendfile (ofd, ifd, &offset, count);
636		914
637	# elif __freebsd	915	# elif __FreeBSD__
638	/*	916	/*
639	* 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
640	* wasted on making it similar to other I/O functions.	918	* wasted on making it similar to other I/O functions.
641	*/	919	*/
642	{	920	{
643	off_t sbytes;	921	off_t sbytes;
644	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	922	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
645		923
646	if (res < 0 && sbytes)	924	/* freebsd' sendfile will return 0 on success */
647	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	925	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		926	/* not on e.g. EIO or EPIPE - sounds broken */
		927	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
		928	res = sbytes;
		929	}
		930
		931	# elif defined (__APPLE__)
		932
		933	{
		934	off_t sbytes = count;
		935	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		936
		937	/* according to the manpage, sbytes is always valid */
		938	if (sbytes)
648	res = sbytes;	939	res = sbytes;
649	}	940	}
650		941
651	# elif __hpux	942	# elif __hpux
652	res = sendfile (ofd, ifd, offset, count, 0, 0);	943	res = sendfile (ofd, ifd, offset, count, 0, 0);
…		…
666	if (res < 0 && sbytes)	957	if (res < 0 && sbytes)
667	res = sbytes;	958	res = sbytes;
668	}	959	}
669		960
670	# endif	961	# endif
		962
		963	#elif defined (_WIN32)
		964
		965	/* does not work, just for documentation of what would need to be done */
		966	{
		967	HANDLE h = TO_SOCKET (ifd);
		968	SetFilePointer (h, offset, 0, FILE_BEGIN);
		969	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		970	}
		971
671	#else	972	#else
672	res = -1;	973	res = -1;
673	errno = ENOSYS;	974	errno = ENOSYS;
674	#endif	975	#endif
675		976
676	if (res < 0	977	if (res < 0
677	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	978	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		979	/* BSDs */
		980	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		981	|| errno == ENOTSUP
		982	#endif
		983	|| errno == EOPNOTSUPP /* BSDs */
678	#if __solaris	984	#if __solaris
679	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	985	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
680	#endif	986	#endif
681	)	987	)
682	)	988	)
…		…
713	}	1019	}
714		1020
715	return res;	1021	return res;
716	}	1022	}
717		1023
		1024	static signed char
		1025	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1026	{
		1027	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1028	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
		1029	}
		1030
		1031	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1032
		1033	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1034	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1035
		1036	static void
		1037	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1038	{
		1039	unsigned char bits [9 + sizeof (ino_t) * 8];
		1040	unsigned char *bit = bits;
		1041
		1042	assert (CHAR_BIT == 8);
		1043	assert (sizeof (eio_dirent) * 8 < 256);
		1044	assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
		1045	assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
		1046
		1047	if (size <= EIO_SORT_FAST)
		1048	return;
		1049
		1050	/* first prepare an array of bits to test in our radix sort */
		1051	/* try to take endianness into account, as well as differences in ino_t sizes */
		1052	/* inode_bits must contain all inodes ORed together */
		1053	/* which is used to skip bits that are 0 everywhere, which is very common */
		1054	{
		1055	ino_t endianness;
		1056	int i, j;
		1057
		1058	/* we store the byte offset of byte n into byte n of "endianness" */
		1059	for (i = 0; i < sizeof (ino_t); ++i)
		1060	((unsigned char *)&endianness)[i] = i;
		1061
		1062	*bit++ = 0;
		1063
		1064	for (i = 0; i < sizeof (ino_t); ++i)
		1065	{
		1066	/* shifting off the byte offsets out of "endianness" */
		1067	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1068	endianness >>= 8;
		1069
		1070	for (j = 0; j < 8; ++j)
		1071	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1072	*bit++ = offs + j;
		1073	}
		1074
		1075	for (j = 0; j < 8; ++j)
		1076	if (score_bits & (1 << j))
		1077	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1078	}
		1079
		1080	/* now actually do the sorting (a variant of MSD radix sort) */
		1081	{
		1082	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1083	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1084	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1085	int stk_idx = 0;
		1086
		1087	base_stk [stk_idx] = dents;
		1088	end_stk [stk_idx] = dents + size;
		1089	bit_stk [stk_idx] = bit - 1;
		1090
		1091	do
		1092	{
		1093	base = base_stk [stk_idx];
		1094	end = end_stk [stk_idx];
		1095	bit = bit_stk [stk_idx];
		1096
		1097	for (;;)
		1098	{
		1099	unsigned char O = *bit >> 3;
		1100	unsigned char M = 1 << (*bit & 7);
		1101
		1102	eio_dirent *a = base;
		1103	eio_dirent *b = end;
		1104
		1105	if (b - a < EIO_SORT_CUTOFF)
		1106	break;
		1107
		1108	/* now bit-partition the array on the bit */
		1109	/* this ugly asymmetric loop seems to perform much better than typical */
		1110	/* partition algos found in the literature */
		1111	do
		1112	if (!(((unsigned char *)a)[O] & M))
		1113	++a;
		1114	else if (!(((unsigned char *)--b)[O] & M))
		1115	{
		1116	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1117	++a;
		1118	}
		1119	while (b > a);
		1120
		1121	/* next bit, or stop, if no bits left in this path */
		1122	if (!*--bit)
		1123	break;
		1124
		1125	base_stk [stk_idx] = a;
		1126	end_stk [stk_idx] = end;
		1127	bit_stk [stk_idx] = bit;
		1128	++stk_idx;
		1129
		1130	end = a;
		1131	}
		1132	}
		1133	while (stk_idx--);
		1134	}
		1135	}
		1136
		1137	static void
		1138	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1139	{
		1140	/* first move the smallest element to the front, to act as a sentinel */
		1141	{
		1142	int i;
		1143	eio_dirent *min = dents;
		1144
		1145	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1146	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1147	if (EIO_DENT_CMP (dents [i], <, *min))
		1148	min = &dents [i];
		1149
		1150	/* swap elements 0 and j (minimum) */
		1151	{
		1152	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1153	}
		1154	}
		1155
		1156	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1157	{
		1158	eio_dirent *i, *j;
		1159
		1160	for (i = dents + 1; i < dents + size; ++i)
		1161	{
		1162	eio_dirent value = *i;
		1163
		1164	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1165	j [1] = j [0];
		1166
		1167	j [1] = value;
		1168	}
		1169	}
		1170	}
		1171
		1172	static void
		1173	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1174	{
		1175	if (size <= 1)
		1176	return; /* our insertion sort relies on size > 0 */
		1177
		1178	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1179	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1180	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1181
		1182	/* use an insertion sort at the end, or for small arrays, */
		1183	/* as insertion sort is more efficient for small partitions */
		1184	eio_dent_insertion_sort (dents, size);
		1185	}
		1186
718	/* read a full directory */	1187	/* read a full directory */
719	static void	1188	static void
720	eio__scandir (eio_req *req, worker *self)	1189	eio__scandir (eio_req *req, etp_worker *self)
721	{	1190	{
722	DIR *dirp;	1191	DIR *dirp;
723	union
724	{
725	EIO_STRUCT_DIRENT d;
726	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
727	} *u;
728	EIO_STRUCT_DIRENT *entp;	1192	EIO_STRUCT_DIRENT *entp;
729	char *name, *names;	1193	char *name, *names;
730	int memlen = 4096;	1194	int namesalloc = 4096;
731	int memofs = 0;	1195	int namesoffs = 0;
		1196	int flags = req->int1;
		1197	eio_dirent *dents = 0;
		1198	int dentalloc = 128;
732	int res = 0;	1199	int dentoffs = 0;
		1200	ino_t inode_bits = 0;
		1201
		1202	req->result = -1;
		1203
		1204	if (!(flags & EIO_READDIR_DENTS))
		1205	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
733		1206
734	X_LOCK (wrklock);	1207	X_LOCK (wrklock);
		1208	/* the corresponding closedir is in ETP_WORKER_CLEAR */
735	self->dirp = dirp = opendir (req->ptr1);	1209	self->dirp = dirp = opendir (req->ptr1);
736	self->dbuf = u = malloc (sizeof (*u));
737	req->flags |= EIO_FLAG_PTR2_FREE;	1210	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1211	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
738	req->ptr2 = names = malloc (memlen);	1212	req->ptr2 = names = malloc (namesalloc);
739	X_UNLOCK (wrklock);	1213	X_UNLOCK (wrklock);
740		1214
741	if (dirp && u && names)	1215	if (dirp && names && (!flags || dents))
742	for (;;)	1216	for (;;)
743	{	1217	{
744	errno = 0;	1218	errno = 0;
745	readdir_r (dirp, &u->d, &entp);	1219	entp = readdir (dirp);
746		1220
747	if (!entp)	1221	if (!entp)
		1222	{
		1223	if (errno)
		1224	break;
		1225
		1226	/* sort etc. */
		1227	req->int1 = flags;
		1228	req->result = dentoffs;
		1229
		1230	if (flags & EIO_READDIR_STAT_ORDER)
		1231	eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
		1232	else if (flags & EIO_READDIR_DIRS_FIRST)
		1233	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1234	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1235	else
		1236	{
		1237	/* in this case, all is known, and we just put dirs first and sort them */
		1238	eio_dirent *oth = dents + dentoffs;
		1239	eio_dirent *dir = dents;
		1240
		1241	/* now partition dirs to the front, and non-dirs to the back */
		1242	/* by walking from both sides and swapping if necessary */
		1243	/* also clear score, so it doesn't influence sorting */
		1244	while (oth > dir)
		1245	{
		1246	if (dir->type == EIO_DT_DIR)
		1247	++dir;
		1248	else if ((--oth)->type == EIO_DT_DIR)
		1249	{
		1250	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1251
		1252	++dir;
		1253	}
		1254	}
		1255
		1256	/* now sort the dirs only */
		1257	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1258	}
		1259
748	break;	1260	break;
		1261	}
749		1262
		1263	/* now add the entry to our list(s) */
750	name = entp->d_name;	1264	name = entp->d_name;
751		1265
		1266	/* skip . and .. entries */
752	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1267	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
753	{	1268	{
754	int len = strlen (name) + 1;	1269	int len = D_NAMLEN (entp) + 1;
755		1270
756	res++;	1271	while (expect_false (namesoffs + len > namesalloc))
757
758	while (memofs + len > memlen)
759	{	1272	{
760	memlen *= 2;	1273	namesalloc *= 2;
761	X_LOCK (wrklock);	1274	X_LOCK (wrklock);
762	req->ptr2 = names = realloc (names, memlen);	1275	req->ptr2 = names = realloc (names, namesalloc);
763	X_UNLOCK (wrklock);	1276	X_UNLOCK (wrklock);
764		1277
765	if (!names)	1278	if (!names)
766	break;	1279	break;
767	}	1280	}
768		1281
769	memcpy (names + memofs, name, len);	1282	memcpy (names + namesoffs, name, len);
		1283
		1284	if (dents)
		1285	{
		1286	struct eio_dirent *ent;
		1287
		1288	if (expect_false (dentoffs == dentalloc))
		1289	{
		1290	dentalloc *= 2;
		1291	X_LOCK (wrklock);
		1292	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1293	X_UNLOCK (wrklock);
		1294
		1295	if (!dents)
		1296	break;
		1297	}
		1298
		1299	ent = dents + dentoffs;
		1300
		1301	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1302	ent->namelen = len - 1;
		1303	ent->inode = D_INO (entp);
		1304
		1305	inode_bits |= ent->inode;
		1306
		1307	switch (D_TYPE (entp))
		1308	{
		1309	default:
		1310	ent->type = EIO_DT_UNKNOWN;
		1311	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1312	break;
		1313
		1314	#ifdef DT_FIFO
		1315	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1316	#endif
		1317	#ifdef DT_CHR
		1318	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1319	#endif
		1320	#ifdef DT_MPC
		1321	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1322	#endif
		1323	#ifdef DT_DIR
		1324	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1325	#endif
		1326	#ifdef DT_NAM
		1327	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1328	#endif
		1329	#ifdef DT_BLK
		1330	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1331	#endif
		1332	#ifdef DT_MPB
		1333	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1334	#endif
		1335	#ifdef DT_REG
		1336	case DT_REG: ent->type = EIO_DT_REG; break;
		1337	#endif
		1338	#ifdef DT_NWK
		1339	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1340	#endif
		1341	#ifdef DT_CMP
		1342	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1343	#endif
		1344	#ifdef DT_LNK
		1345	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1346	#endif
		1347	#ifdef DT_SOCK
		1348	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1349	#endif
		1350	#ifdef DT_DOOR
		1351	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1352	#endif
		1353	#ifdef DT_WHT
		1354	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1355	#endif
		1356	}
		1357
		1358	ent->score = 7;
		1359
		1360	if (flags & EIO_READDIR_DIRS_FIRST)
		1361	{
		1362	if (ent->type == EIO_DT_UNKNOWN)
		1363	{
		1364	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1365	ent->score = 1;
		1366	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1367	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1368	}
		1369	else if (ent->type == EIO_DT_DIR)
		1370	ent->score = 0;
		1371	}
		1372	}
		1373
770	memofs += len;	1374	namesoffs += len;
		1375	++dentoffs;
		1376	}
		1377
		1378	if (EIO_CANCELLED (req))
		1379	{
		1380	errno = ECANCELED;
		1381	break;
771	}	1382	}
772	}	1383	}
		1384	}
773		1385
774	if (errno)	1386	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
775	res = -1;	1387	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
776		1388	#else
777	req->result = res;	1389
		1390	int
		1391	eio__msync (void *mem, size_t len, int flags)
		1392	{
		1393	if (EIO_MS_ASYNC != MS_SYNC
		1394	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1395	|| EIO_MS_SYNC != MS_SYNC)
		1396	{
		1397	flags = 0
		1398	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1399	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1400	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1401	}
		1402
		1403	return msync (mem, len, flags);
		1404	}
		1405
		1406	#endif
		1407
		1408	int
		1409	eio__mtouch (void *mem, size_t len, int flags)
		1410	{
		1411	intptr_t addr = (intptr_t)mem;
		1412	intptr_t end = addr + len;
		1413	#ifdef PAGESIZE
		1414	const intptr_t page = PAGESIZE;
		1415	#else
		1416	static intptr_t page;
		1417
		1418	if (!page)
		1419	page = sysconf (_SC_PAGESIZE);
		1420	#endif
		1421
		1422	/* round down to start of page, although this is probably useless */
		1423	addr &= ~(page - 1); /* assume page size is always a power of two */
		1424
		1425	if (addr < end)
		1426	if (flags & EIO_MT_MODIFY) /* modify */
		1427	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
		1428	else
		1429	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
		1430
		1431	return 0;
778	}	1432	}
779		1433
780	/*****************************************************************************/	1434	/*****************************************************************************/
781		1435
782	#define ALLOC(len) \	1436	#define ALLOC(len) \
…		…
792	req->result = -1; \	1446	req->result = -1; \
793	break; \	1447	break; \
794	} \	1448	} \
795	}	1449	}
796		1450
797	X_THREAD_PROC (eio_proc)	1451	X_THREAD_PROC (etp_proc)
798	{	1452	{
799	eio_req *req;	1453	ETP_REQ *req;
800	struct timespec ts;	1454	struct timespec ts;
801	worker *self = (worker *)thr_arg;	1455	etp_worker *self = (etp_worker *)thr_arg;
802		1456
803	/* try to distribute timeouts somewhat randomly */	1457	/* try to distribute timeouts somewhat randomly */
804	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1458	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
805		1459
806	for (;;)	1460	for (;;)
807	{	1461	{
808	ts.tv_sec = time (0) + IDLE_TIMEOUT;
809
810	X_LOCK (reqlock);	1462	X_LOCK (reqlock);
811		1463
812	for (;;)	1464	for (;;)
813	{	1465	{
814	self->req = req = reqq_shift (&req_queue);	1466	self->req = req = reqq_shift (&req_queue);
…		…
816	if (req)	1468	if (req)
817	break;	1469	break;
818		1470
819	++idle;	1471	++idle;
820		1472
		1473	ts.tv_sec = time (0) + IDLE_TIMEOUT;
821	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1474	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
822	{	1475	{
823	if (idle > max_idle)	1476	if (idle > max_idle)
824	{	1477	{
825	--idle;	1478	--idle;
…		…
830	goto quit;	1483	goto quit;
831	}	1484	}
832		1485
833	/* we are allowed to idle, so do so without any timeout */	1486	/* we are allowed to idle, so do so without any timeout */
834	X_COND_WAIT (reqwait, reqlock);	1487	X_COND_WAIT (reqwait, reqlock);
835	ts.tv_sec = time (0) + IDLE_TIMEOUT;
836	}	1488	}
837		1489
838	--idle;	1490	--idle;
839	}	1491	}
840		1492
841	--nready;	1493	--nready;
842		1494
843	X_UNLOCK (reqlock);	1495	X_UNLOCK (reqlock);
844		1496
845	errno = 0; /* strictly unnecessary */	1497	if (req->type < 0)
		1498	goto quit;
846		1499
847	if (!EIO_CANCELLED (req))	1500	if (!EIO_CANCELLED (req))
848	switch (req->type)	1501	ETP_EXECUTE (self, req);
849	{
850	case EIO_READ: ALLOC (req->size);
851	req->result = req->offs >= 0
852	? pread (req->int1, req->ptr2, req->size, req->offs)
853	: read (req->int1, req->ptr2, req->size); break;
854	case EIO_WRITE: req->result = req->offs >= 0
855	? pwrite (req->int1, req->ptr2, req->size, req->offs)
856	: write (req->int1, req->ptr2, req->size); break;
857
858	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
859	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
860
861	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
862	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
863	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
864	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
865	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
866	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
867
868	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
869	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
870	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
871	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
872	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
873	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
874
875	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
876	case EIO_CLOSE: req->result = close (req->int1); break;
877	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
878	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
879	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
880	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
881	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
882	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
883	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
884	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
885
886	case EIO_READLINK: ALLOC (NAME_MAX);
887	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
888
889	case EIO_SYNC: req->result = 0; sync (); break;
890	case EIO_FSYNC: req->result = fsync (req->int1); break;
891	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
892
893	case EIO_READDIR: eio__scandir (req, self); break;
894
895	case EIO_BUSY:
896	#ifdef _WIN32
897	Sleep (req->nv1 * 1000.);
898	#else
899	{
900	struct timeval tv;
901
902	tv.tv_sec = req->nv1;
903	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
904
905	req->result = select (0, 0, 0, 0, &tv);
906	}
907	#endif
908	break;
909
910	case EIO_UTIME:
911	case EIO_FUTIME:
912	{
913	struct timeval tv[2];
914	struct timeval *times;
915
916	if (req->nv1 != -1. || req->nv2 != -1.)
917	{
918	tv[0].tv_sec = req->nv1;
919	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
920	tv[1].tv_sec = req->nv2;
921	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
922
923	times = tv;
924	}
925	else
926	times = 0;
927
928
929	req->result = req->type == EIO_FUTIME
930	? futimes (req->int1, times)
931	: utimes (req->ptr1, times);
932	}
933
934	case EIO_GROUP:
935	case EIO_NOP:
936	req->result = 0;
937	break;
938
939	case EIO_QUIT:
940	goto quit;
941
942	default:
943	req->result = -1;
944	break;
945	}
946
947	req->errorno = errno;
948		1502
949	X_LOCK (reslock);	1503	X_LOCK (reslock);
950		1504
951	++npending;	1505	++npending;
952		1506
953	if (!reqq_push (&res_queue, req) && want_poll_cb)	1507	if (!reqq_push (&res_queue, req) && want_poll_cb)
954	want_poll_cb ();	1508	want_poll_cb ();
955		1509
956	self->req = 0;	1510	self->req = 0;
957	worker_clear (self);	1511	etp_worker_clear (self);
958		1512
959	X_UNLOCK (reslock);	1513	X_UNLOCK (reslock);
960	}	1514	}
961		1515
962	quit:	1516	quit:
963	X_LOCK (wrklock);	1517	X_LOCK (wrklock);
964	worker_free (self);	1518	etp_worker_free (self);
965	X_UNLOCK (wrklock);	1519	X_UNLOCK (wrklock);
966		1520
967	return 0;	1521	return 0;
968	}	1522	}
969		1523
970	/*****************************************************************************/	1524	/*****************************************************************************/
971		1525
972	static void eio_atfork_prepare (void)
973	{
974	X_LOCK (wrklock);
975	X_LOCK (reqlock);
976	X_LOCK (reslock);
977	#if !HAVE_PREADWRITE
978	X_LOCK (preadwritelock);
979	#endif
980	#if !HAVE_READDIR_R
981	X_LOCK (readdirlock);
982	#endif
983	}
984
985	static void eio_atfork_parent (void)
986	{
987	#if !HAVE_READDIR_R
988	X_UNLOCK (readdirlock);
989	#endif
990	#if !HAVE_PREADWRITE
991	X_UNLOCK (preadwritelock);
992	#endif
993	X_UNLOCK (reslock);
994	X_UNLOCK (reqlock);
995	X_UNLOCK (wrklock);
996	}
997
998	static void eio_atfork_child (void)
999	{
1000	eio_req *prv;
1001
1002	while (prv = reqq_shift (&req_queue))
1003	eio_destroy (prv);
1004
1005	while (prv = reqq_shift (&res_queue))
1006	eio_destroy (prv);
1007
1008	while (wrk_first.next != &wrk_first)
1009	{
1010	worker *wrk = wrk_first.next;
1011
1012	if (wrk->req)
1013	eio_destroy (wrk->req);
1014
1015	worker_clear (wrk);
1016	worker_free (wrk);
1017	}
1018
1019	started = 0;
1020	idle = 0;
1021	nreqs = 0;
1022	nready = 0;
1023	npending = 0;
1024
1025	eio_atfork_parent ();
1026	}
1027
1028	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1526	int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1029	{	1527	{
1030	want_poll_cb = want_poll;	1528	return etp_init (want_poll, done_poll);
1031	done_poll_cb = done_poll;
1032
1033	#ifdef _WIN32
1034	X_MUTEX_CHECK (wrklock);
1035	X_MUTEX_CHECK (reslock);
1036	X_MUTEX_CHECK (reqlock);
1037	X_MUTEX_CHECK (reqwait);
1038	X_MUTEX_CHECK (preadwritelock);
1039	X_MUTEX_CHECK (readdirlock);
1040
1041	X_COND_CHECK (reqwait);
1042	#endif
1043
1044	X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1045	}	1529	}
1046		1530
1047	static void eio_api_destroy (eio_req *req)	1531	static void eio_api_destroy (eio_req *req)
1048	{	1532	{
1049	free (req);	1533	free (req);
…		…
1071	{ \	1555	{ \
1072	eio_api_destroy (req); \	1556	eio_api_destroy (req); \
1073	return 0; \	1557	return 0; \
1074	}	1558	}
1075		1559
		1560	static void eio_execute (etp_worker *self, eio_req *req)
		1561	{
		1562	errno = 0;
		1563
		1564	switch (req->type)
		1565	{
		1566	case EIO_READ: ALLOC (req->size);
		1567	req->result = req->offs >= 0
		1568	? pread (req->int1, req->ptr2, req->size, req->offs)
		1569	: read (req->int1, req->ptr2, req->size); break;
		1570	case EIO_WRITE: req->result = req->offs >= 0
		1571	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1572	: write (req->int1, req->ptr2, req->size); break;
		1573
		1574	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		1575	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
		1576
		1577	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1578	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1579	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1580	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1581	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1582	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1583
		1584	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1585	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1586	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1587	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1588
		1589	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1590	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1591	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1592	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1593	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1594	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1595
		1596	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1597	case EIO_CLOSE: req->result = close (req->int1); break;
		1598	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1599	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1600	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1601	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1602	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1603	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1604	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1605	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
		1606
		1607	case EIO_READLINK: ALLOC (NAME_MAX);
		1608	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
		1609
		1610	case EIO_SYNC: req->result = 0; sync (); break;
		1611	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1612	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1613	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1614	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1615	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1616
		1617	case EIO_READDIR: eio__scandir (req, self); break;
		1618
		1619	case EIO_BUSY:
		1620	#ifdef _WIN32
		1621	Sleep (req->nv1 * 1000.);
		1622	#else
		1623	{
		1624	struct timeval tv;
		1625
		1626	tv.tv_sec = req->nv1;
		1627	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
		1628
		1629	req->result = select (0, 0, 0, 0, &tv);
		1630	}
		1631	#endif
		1632	break;
		1633
		1634	case EIO_UTIME:
		1635	case EIO_FUTIME:
		1636	{
		1637	struct timeval tv[2];
		1638	struct timeval *times;
		1639
		1640	if (req->nv1 != -1. || req->nv2 != -1.)
		1641	{
		1642	tv[0].tv_sec = req->nv1;
		1643	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		1644	tv[1].tv_sec = req->nv2;
		1645	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		1646
		1647	times = tv;
		1648	}
		1649	else
		1650	times = 0;
		1651
		1652
		1653	req->result = req->type == EIO_FUTIME
		1654	? futimes (req->int1, times)
		1655	: utimes (req->ptr1, times);
		1656	}
		1657	break;
		1658
		1659	case EIO_GROUP:
		1660	abort (); /* handled in eio_request */
		1661
		1662	case EIO_NOP:
		1663	req->result = 0;
		1664	break;
		1665
		1666	case EIO_CUSTOM:
		1667	((void (*)(eio_req *))req->feed) (req);
		1668	break;
		1669
		1670	default:
		1671	req->result = -1;
		1672	break;
		1673	}
		1674
		1675	req->errorno = errno;
		1676	}
		1677
		1678	#ifndef EIO_NO_WRAPPERS
		1679
1076	eio_req *eio_nop (int pri, eio_cb cb, void *data)	1680	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1077	{	1681	{
1078	REQ (EIO_NOP); SEND;	1682	REQ (EIO_NOP); SEND;
1079	}	1683	}
1080		1684
…		…
1091	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	1695	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1092	{	1696	{
1093	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1697	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1094	}	1698	}
1095		1699
		1700	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1701	{
		1702	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1703	}
		1704
		1705	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1706	{
		1707	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1708	}
		1709
		1710	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		1711	{
		1712	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		1713	}
		1714
1096	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	1715	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1097	{	1716	{
1098	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1717	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1099	}	1718	}
1100		1719
…		…
1119	}	1738	}
1120		1739
1121	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	1740	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1122	{	1741	{
1123	REQ (EIO_FSTAT); req->int1 = fd; SEND;	1742	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1743	}
		1744
		1745	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		1746	{
		1747	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1124	}	1748	}
1125		1749
1126	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	1750	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1127	{	1751	{
1128	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	1752	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
…		…
1202	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	1826	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1203	{	1827	{
1204	return eio__1path (EIO_LSTAT, path, pri, cb, data);	1828	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1205	}	1829	}
1206		1830
		1831	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		1832	{
		1833	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		1834	}
		1835
1207	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	1836	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1208	{	1837	{
1209	return eio__1path (EIO_UNLINK, path, pri, cb, data);	1838	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1210	}	1839	}
1211		1840
1212	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	1841	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1213	{	1842	{
1214	return eio__1path (EIO_RMDIR, path, pri, cb, data);	1843	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1215	}	1844	}
1216		1845
1217	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	1846	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1218	{	1847	{
1219	return eio__1path (EIO_READDIR, path, pri, cb, data);	1848	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1220	}	1849	}
1221		1850
1222	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	1851	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1223	{	1852	{
1224	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	1853	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1225	}	1854	}
1226		1855
1227	static eio_req *	1856	static eio_req *
1228	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1857	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1229	{	1858	{
…		…
1253	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1882	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1254	{	1883	{
1255	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	1884	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1256	}	1885	}
1257		1886
		1887	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
		1888	{
		1889	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
		1890	}
		1891
		1892	#endif
		1893
1258	eio_req *eio_grp (eio_cb cb, void *data)	1894	eio_req *eio_grp (eio_cb cb, void *data)
1259	{	1895	{
1260	const int pri = EIO_PRI_MAX;	1896	const int pri = EIO_PRI_MAX;
1261		1897
1262	REQ (EIO_GROUP); SEND;	1898	REQ (EIO_GROUP); SEND;
…		…
1286		1922
1287	void eio_grp_add (eio_req *grp, eio_req *req)	1923	void eio_grp_add (eio_req *grp, eio_req *req)
1288	{	1924	{
1289	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	1925	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1290		1926
		1927	grp->flags |= EIO_FLAG_GROUPADD;
		1928
1291	++grp->size;	1929	++grp->size;
1292	req->grp = grp;	1930	req->grp = grp;
1293		1931
1294	req->grp_prev = 0;	1932	req->grp_prev = 0;
1295	req->grp_next = grp->grp_first;	1933	req->grp_next = grp->grp_first;
…		…
1303	/*****************************************************************************/	1941	/*****************************************************************************/
1304	/* misc garbage */	1942	/* misc garbage */
1305		1943
1306	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	1944	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1307	{	1945	{
1308	worker wrk;	1946	etp_worker wrk;
		1947	ssize_t ret;
1309		1948
1310	wrk.dbuf = 0;	1949	wrk.dbuf = 0;
1311		1950
1312	eio__sendfile (ofd, ifd, offset, count, &wrk);	1951	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1313		1952
1314	if (wrk.dbuf)	1953	if (wrk.dbuf)
1315	free (wrk.dbuf);	1954	free (wrk.dbuf);
1316	}
1317		1955
		1956	return ret;
		1957	}
		1958

Diff Legend

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