ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.13 by root, Tue May 13 19:34:11 2008 UTC vs.
Revision 1.49 by root, Sat Jan 2 14:24:32 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>
…		…
25	#endif	68	#endif
26		69
27	#ifdef _WIN32	70	#ifdef _WIN32
28		71
29	/*doh*/	72	/*doh*/
30
31	#else	73	#else
32		74
33	# include "config.h"	75	# include "config.h"
34	# include <sys/time.h>	76	# include <sys/time.h>
35	# include <sys/select.h>	77	# include <sys/select.h>
		78	# include <sys/mman.h>
36	# include <unistd.h>	79	# include <unistd.h>
37	# include <utime.h>	80	# include <utime.h>
38	# include <signal.h>	81	# include <signal.h>
39	# include <dirent.h>	82	# include <dirent.h>
40		83
		84	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		85	# if __freebsd || defined __NetBSD__ || defined __OpenBSD__
		86	# define _DIRENT_HAVE_D_TYPE /* sigh */
		87	# define D_INO(de) (de)->d_fileno
		88	# define D_NAMLEN(de) (de)->d_namlen
		89	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		90	# define D_INO(de) (de)->d_ino
		91	# endif
		92
		93	#ifdef _D_EXACT_NAMLEN
		94	# undef D_NAMLEN
		95	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		96	#endif
		97
		98	# ifdef _DIRENT_HAVE_D_TYPE
		99	# define D_TYPE(de) (de)->d_type
		100	# endif
		101
41	# ifndef EIO_STRUCT_DIRENT	102	# ifndef EIO_STRUCT_DIRENT
42	# define EIO_STRUCT_DIRENT struct dirent	103	# define EIO_STRUCT_DIRENT struct dirent
43	# endif	104	# endif
44		105
45	#endif	106	#endif
46		107
47	#if HAVE_SENDFILE	108	#if HAVE_SENDFILE
48	# if __linux	109	# if __linux
49	# include <sys/sendfile.h>	110	# include <sys/sendfile.h>
50	# elif __freebsd	111	# elif __freebsd || defined __APPLE__
51	# include <sys/socket.h>	112	# include <sys/socket.h>
52	# include <sys/uio.h>	113	# include <sys/uio.h>
53	# elif __hpux	114	# elif __hpux
54	# include <sys/socket.h>	115	# include <sys/socket.h>
55	# elif __solaris /* not yet */	116	# elif __solaris
56	# include <sys/sendfile.h>	117	# include <sys/sendfile.h>
57	# else	118	# else
58	# error sendfile support requested but not available	119	# error sendfile support requested but not available
59	# endif	120	# endif
60	#endif	121	#endif
61		122
		123	#ifndef D_TYPE
		124	# define D_TYPE(de) 0
		125	#endif
		126	#ifndef D_INO
		127	# define D_INO(de) 0
		128	#endif
		129	#ifndef D_NAMLEN
		130	# define D_NAMLEN(de) strlen ((de)->d_name)
		131	#endif
		132
62	/* number of seconds after which an idle threads exit */	133	/* number of seconds after which an idle threads exit */
63	#define IDLE_TIMEOUT 10	134	#define IDLE_TIMEOUT 10
64		135
65	/* used for struct dirent, AIX doesn't provide it */	136	/* used for struct dirent, AIX doesn't provide it */
66	#ifndef NAME_MAX	137	#ifndef NAME_MAX
…		…
70	/* buffer size for various temporary buffers */	141	/* buffer size for various temporary buffers */
71	#define EIO_BUFSIZE 65536	142	#define EIO_BUFSIZE 65536
72		143
73	#define dBUF \	144	#define dBUF \
74	char *eio_buf; \	145	char *eio_buf; \
75	X_LOCK (etplock); \	146	ETP_WORKER_LOCK (self); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	147	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (etplock); \	148	ETP_WORKER_UNLOCK (self); \
78	errno = ENOMEM; \	149	errno = ENOMEM; \
79	if (!eio_buf) \	150	if (!eio_buf) \
80	return -1;	151	return -1;
81		152
82	#define EIO_TICKS ((1000000 + 1023) >> 10)	153	#define EIO_TICKS ((1000000 + 1023) >> 10)
83		154
84	/*****************************************************************************/	155	/*****************************************************************************/
85		156
		157	#if __GNUC__ >= 3
		158	# define expect(expr,value) __builtin_expect ((expr),(value))
		159	#else
		160	# define expect(expr,value) (expr)
		161	#endif
		162
		163	#define expect_false(expr) expect ((expr) != 0, 0)
		164	#define expect_true(expr) expect ((expr) != 0, 1)
		165
		166	/*****************************************************************************/
		167
		168	#define ETP_PRI_MIN EIO_PRI_MIN
		169	#define ETP_PRI_MAX EIO_PRI_MAX
		170
		171	struct etp_worker;
		172
		173	#define ETP_REQ eio_req
		174	#define ETP_DESTROY(req) eio_destroy (req)
		175	static int eio_finish (eio_req *req);
		176	#define ETP_FINISH(req) eio_finish (req)
		177	static void eio_execute (struct etp_worker *self, eio_req *req);
		178	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		179
		180	#define ETP_WORKER_CLEAR(req) \
		181	if (wrk->dbuf) \
		182	{ \
		183	free (wrk->dbuf); \
		184	wrk->dbuf = 0; \
		185	} \
		186	\
		187	if (wrk->dirp) \
		188	{ \
		189	closedir (wrk->dirp); \
		190	wrk->dirp = 0; \
		191	}
		192
		193	#define ETP_WORKER_COMMON \
		194	void *dbuf; \
		195	DIR *dirp;
		196
		197	/*****************************************************************************/
		198
		199	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		200
86	/* calculcate time difference in ~1/EIO_TICKS of a second */	201	/* calculate time difference in ~1/EIO_TICKS of a second */
87	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	202	static int tvdiff (struct timeval *tv1, struct timeval *tv2)
88	{	203	{
89	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	204	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
90	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	205	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
91	}	206	}
92		207
93	static unsigned int started, idle, wanted = 4;	208	static unsigned int started, idle, wanted = 4;
94		209
95	typedef struct etp_pool
96	{
97	void (*want_poll_cb) (void);	210	static void (*want_poll_cb) (void);
98	void (*done_poll_cb) (void);	211	static void (*done_poll_cb) (void);
99		212
100	unsigned int max_poll_time;	213	static unsigned int max_poll_time; /* reslock */
101	unsigned int max_poll_reqs;	214	static unsigned int max_poll_reqs; /* reslock */
102	} etp_pool;
103		215
104	static volatile unsigned int nreqs, nready, npending;	216	static volatile unsigned int nreqs; /* reqlock */
		217	static volatile unsigned int nready; /* reqlock */
		218	static volatile unsigned int npending; /* reqlock */
105	static volatile unsigned int max_idle = 4;	219	static volatile unsigned int max_idle = 4;
106		220
107	static mutex_t etplock = X_MUTEX_INIT;	221	static mutex_t wrklock = X_MUTEX_INIT;
108	static mutex_t reslock = X_MUTEX_INIT;	222	static mutex_t reslock = X_MUTEX_INIT;
109	static mutex_t reqlock = X_MUTEX_INIT;	223	static mutex_t reqlock = X_MUTEX_INIT;
110	static cond_t reqwait = X_COND_INIT;	224	static cond_t reqwait = X_COND_INIT;
111		225
		226	#if !HAVE_PREADWRITE
		227	/*
		228	* make our pread/pwrite emulation safe against themselves, but not against
		229	* normal read/write by using a mutex. slows down execution a lot,
		230	* but that's your problem, not mine.
		231	*/
		232	static mutex_t preadwritelock = X_MUTEX_INIT;
		233	#endif
		234
112	typedef struct worker	235	typedef struct etp_worker
113	{	236	{
114	/* locked by etplock */	237	/* locked by wrklock */
115	struct worker *prev, *next;	238	struct etp_worker *prev, *next;
116		239
117	thread_t tid;	240	thread_t tid;
118		241
119	/* locked by reslock, reqlock or etplock */	242	/* locked by reslock, reqlock or wrklock */
120	eio_req *req; /* currently processed request */	243	ETP_REQ *req; /* currently processed request */
121	void *dbuf;	244
122	DIR *dirp;	245	ETP_WORKER_COMMON
123	} worker;	246	} etp_worker;
124		247
125	static worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	248	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
		249
		250	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		251	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
126		252
127	/* worker threads management */	253	/* worker threads management */
128		254
129	static void worker_clear (worker *wrk)	255	static void etp_worker_clear (etp_worker *wrk)
130	{	256	{
131	if (wrk->dirp)	257	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	}	258	}
143		259
144	static void worker_free (worker *wrk)	260	static void etp_worker_free (etp_worker *wrk)
145	{	261	{
146	wrk->next->prev = wrk->prev;	262	wrk->next->prev = wrk->prev;
147	wrk->prev->next = wrk->next;	263	wrk->prev->next = wrk->next;
148		264
149	free (wrk);	265	free (wrk);
150	}	266	}
151		267
152	unsigned int eio_nreqs (void)	268	static unsigned int etp_nreqs (void)
153	{	269	{
		270	int retval;
		271	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		272	retval = nreqs;
		273	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154	return nreqs;	274	return retval;
155	}	275	}
156		276
157	unsigned int eio_nready (void)	277	static unsigned int etp_nready (void)
158	{	278	{
159	unsigned int retval;	279	unsigned int retval;
160		280
161	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	281	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162	retval = nready;	282	retval = nready;
163	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	283	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164		284
165	return retval;	285	return retval;
166	}	286	}
167		287
168	unsigned int eio_npending (void)	288	static unsigned int etp_npending (void)
169	{	289	{
170	unsigned int retval;	290	unsigned int retval;
171		291
172	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	292	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173	retval = npending;	293	retval = npending;
174	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	294	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175		295
176	return retval;	296	return retval;
177	}	297	}
178		298
179	unsigned int eio_nthreads (void)	299	static unsigned int etp_nthreads (void)
180	{	300	{
181	unsigned int retval;	301	unsigned int retval;
182		302
183	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	303	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184	retval = started;	304	retval = started;
…		…
191	* a somewhat faster data structure might be nice, but	311	* a somewhat faster data structure might be nice, but
192	* with 8 priorities this actually needs <20 insns	312	* with 8 priorities this actually needs <20 insns
193	* per shift, the most expensive operation.	313	* per shift, the most expensive operation.
194	*/	314	*/
195	typedef struct {	315	typedef struct {
196	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	316	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197	int size;	317	int size;
198	} reqq;	318	} etp_reqq;
199		319
200	static reqq req_queue;	320	static etp_reqq req_queue;
201	static reqq res_queue;	321	static etp_reqq res_queue;
202		322
203	static int reqq_push (reqq *q, eio_req *req)	323	static int reqq_push (etp_reqq *q, ETP_REQ *req)
204	{	324	{
205	int pri = req->pri;	325	int pri = req->pri;
206	req->next = 0;	326	req->next = 0;
207		327
208	if (q->qe[pri])	328	if (q->qe[pri])
…		…
214	q->qe[pri] = q->qs[pri] = req;	334	q->qe[pri] = q->qs[pri] = req;
215		335
216	return q->size++;	336	return q->size++;
217	}	337	}
218		338
219	static eio_req *reqq_shift (reqq *q)	339	static ETP_REQ *reqq_shift (etp_reqq *q)
220	{	340	{
221	int pri;	341	int pri;
222		342
223	if (!q->size)	343	if (!q->size)
224	return 0;	344	return 0;
225		345
226	--q->size;	346	--q->size;
227		347
228	for (pri = EIO_NUM_PRI; pri--; )	348	for (pri = ETP_NUM_PRI; pri--; )
229	{	349	{
230	eio_req *req = q->qs[pri];	350	eio_req *req = q->qs[pri];
231		351
232	if (req)	352	if (req)
233	{	353	{
…		…
241	abort ();	361	abort ();
242	}	362	}
243		363
244	static void etp_atfork_prepare (void)	364	static void etp_atfork_prepare (void)
245	{	365	{
246	X_LOCK (etplock);	366	X_LOCK (wrklock);
247	X_LOCK (reqlock);	367	X_LOCK (reqlock);
248	X_LOCK (reslock);	368	X_LOCK (reslock);
249	#if !HAVE_PREADWRITE	369	#if !HAVE_PREADWRITE
250	X_LOCK (preadwritelock);	370	X_LOCK (preadwritelock);
251	#endif	371	#endif
252	#if !HAVE_READDIR_R
253	X_LOCK (readdirlock);
254	#endif
255	}	372	}
256		373
257	static void etp_atfork_parent (void)	374	static void etp_atfork_parent (void)
258	{	375	{
259	#if !HAVE_READDIR_R
260	X_UNLOCK (readdirlock);
261	#endif
262	#if !HAVE_PREADWRITE	376	#if !HAVE_PREADWRITE
263	X_UNLOCK (preadwritelock);	377	X_UNLOCK (preadwritelock);
264	#endif	378	#endif
265	X_UNLOCK (reslock);	379	X_UNLOCK (reslock);
266	X_UNLOCK (reqlock);	380	X_UNLOCK (reqlock);
267	X_UNLOCK (etplock);	381	X_UNLOCK (wrklock);
268	}	382	}
269		383
270	static void etp_atfork_child (void)	384	static void etp_atfork_child (void)
271	{	385	{
272	eio_req *prv;	386	ETP_REQ *prv;
273		387
274	while (prv = reqq_shift (&req_queue))	388	while ((prv = reqq_shift (&req_queue)))
275	eio_destroy (prv);	389	ETP_DESTROY (prv);
276		390
277	while (prv = reqq_shift (&res_queue))	391	while ((prv = reqq_shift (&res_queue)))
278	eio_destroy (prv);	392	ETP_DESTROY (prv);
279		393
280	while (wrk_first.next != &wrk_first)	394	while (wrk_first.next != &wrk_first)
281	{	395	{
282	worker *wrk = wrk_first.next;	396	etp_worker *wrk = wrk_first.next;
283		397
284	if (wrk->req)	398	if (wrk->req)
285	eio_destroy (wrk->req);	399	ETP_DESTROY (wrk->req);
286		400
287	worker_clear (wrk);	401	etp_worker_clear (wrk);
288	worker_free (wrk);	402	etp_worker_free (wrk);
289	}	403	}
290		404
291	started = 0;	405	started = 0;
292	idle = 0;	406	idle = 0;
293	nreqs = 0;	407	nreqs = 0;
…		…
302	{	416	{
303	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);	417	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
304	}	418	}
305		419
306	static int	420	static int
307	etp_init (etp_pool *etp, void (*want_poll)(void), void (*done_poll)(void))	421	etp_init (void (*want_poll)(void), void (*done_poll)(void))
308	{	422	{
309	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	423	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
310		424
311	pthread_once (&doinit, etp_once_init);	425	pthread_once (&doinit, etp_once_init);
312		426
313	memset (etp, 0, sizeof *etp);
314
315	etp->want_poll_cb = want_poll;	427	want_poll_cb = want_poll;
316	etp->done_poll_cb = done_poll;	428	done_poll_cb = done_poll;
317	}
318		429
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;	430	return 0;
354	}	431	}
355		432
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);	433	X_THREAD_PROC (etp_proc);
407		434
408	static void start_thread (void)	435	static void etp_start_thread (void)
409	{	436	{
410	worker *wrk = calloc (1, sizeof (worker));	437	etp_worker *wrk = calloc (1, sizeof (etp_worker));
411		438
412	/*TODO*/	439	/*TODO*/
413	assert (("unable to allocate worker thread data", wrk));	440	assert (("unable to allocate worker thread data", wrk));
414		441
415	X_LOCK (etplock);	442	X_LOCK (wrklock);
416		443
417	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	444	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
418	{	445	{
419	wrk->prev = &wrk_first;	446	wrk->prev = &wrk_first;
420	wrk->next = wrk_first.next;	447	wrk->next = wrk_first.next;
421	wrk_first.next->prev = wrk;	448	wrk_first.next->prev = wrk;
422	wrk_first.next = wrk;	449	wrk_first.next = wrk;
423	++started;	450	++started;
424	}	451	}
425	else	452	else
426	free (wrk);	453	free (wrk);
427		454
428	X_UNLOCK (etplock);	455	X_UNLOCK (wrklock);
429	}	456	}
430		457
431	static void maybe_start_thread (void)	458	static void etp_maybe_start_thread (void)
432	{	459	{
433	if (eio_nthreads () >= wanted)	460	if (expect_true (etp_nthreads () >= wanted))
434	return;	461	return;
435		462
436	/* todo: maybe use idle here, but might be less exact */	463	/* todo: maybe use idle here, but might be less exact */
437	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	464	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
438	return;	465	return;
439		466
440	start_thread ();	467	etp_start_thread ();
441	}	468	}
442		469
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)	470	static void etp_end_thread (void)
462	{	471	{
463	eio_req *req = calloc (1, sizeof (eio_req));	472	eio_req *req = calloc (1, sizeof (eio_req));
464		473
465	req->type = EIO_QUIT;	474	req->type = -1;
466	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	475	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
467		476
468	X_LOCK (reqlock);	477	X_LOCK (reqlock);
469	reqq_push (&req_queue, req);	478	reqq_push (&req_queue, req);
470	X_COND_SIGNAL (reqwait);	479	X_COND_SIGNAL (reqwait);
471	X_UNLOCK (reqlock);	480	X_UNLOCK (reqlock);
472		481
473	X_LOCK (etplock);	482	X_LOCK (wrklock);
474	--started;	483	--started;
475	X_UNLOCK (etplock);	484	X_UNLOCK (wrklock);
476	}	485	}
477		486
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)	487	static int etp_poll (void)
515	{	488	{
516	int maxreqs = etp.max_poll_reqs;	489	unsigned int maxreqs;
		490	unsigned int maxtime;
517	struct timeval tv_start, tv_now;	491	struct timeval tv_start, tv_now;
518	eio_req *req;
519		492
520	if (etp.max_poll_time)	493	X_LOCK (reslock);
		494	maxreqs = max_poll_reqs;
		495	maxtime = max_poll_time;
		496	X_UNLOCK (reslock);
		497
		498	if (maxtime)
521	gettimeofday (&tv_start, 0);	499	gettimeofday (&tv_start, 0);
522		500
523	for (;;)	501	for (;;)
524	{	502	{
		503	ETP_REQ *req;
		504
525	maybe_start_thread ();	505	etp_maybe_start_thread ();
526		506
527	X_LOCK (reslock);	507	X_LOCK (reslock);
528	req = reqq_shift (&res_queue);	508	req = reqq_shift (&res_queue);
529		509
530	if (req)	510	if (req)
531	{	511	{
532	--npending;	512	--npending;
533		513
534	if (!res_queue.size && etp.done_poll_cb)	514	if (!res_queue.size && done_poll_cb)
535	etp.done_poll_cb ();	515	done_poll_cb ();
536	}	516	}
537		517
538	X_UNLOCK (reslock);	518	X_UNLOCK (reslock);
539		519
540	if (!req)	520	if (!req)
541	return 0;	521	return 0;
542		522
		523	X_LOCK (reqlock);
543	--nreqs;	524	--nreqs;
		525	X_UNLOCK (reqlock);
544		526
545	if (req->type == EIO_GROUP && req->size)	527	if (expect_false (req->type == EIO_GROUP && req->size))
546	{	528	{
547	req->int1 = 1; /* mark request as delayed */	529	req->int1 = 1; /* mark request as delayed */
548	continue;	530	continue;
549	}	531	}
550	else	532	else
551	{	533	{
552	int res = eio_finish (req);	534	int res = ETP_FINISH (req);
553	if (res)	535	if (expect_false (res))
554	return res;	536	return res;
555	}	537	}
556		538
557	if (maxreqs && !--maxreqs)	539	if (expect_false (maxreqs && !--maxreqs))
558	break;	540	break;
559		541
560	if (etp.max_poll_time)	542	if (maxtime)
561	{	543	{
562	gettimeofday (&tv_now, 0);	544	gettimeofday (&tv_now, 0);
563		545
564	if (tvdiff (&tv_start, &tv_now) >= etp.max_poll_time)	546	if (tvdiff (&tv_start, &tv_now) >= maxtime)
565	break;	547	break;
566	}	548	}
567	}	549	}
568		550
569	errno = EAGAIN;	551	errno = EAGAIN;
570	return -1;	552	return -1;
571	}	553	}
572		554
		555	static void etp_cancel (ETP_REQ *req)
		556	{
		557	X_LOCK (wrklock);
		558	req->flags |= EIO_FLAG_CANCELLED;
		559	X_UNLOCK (wrklock);
		560
		561	eio_grp_cancel (req);
		562	}
		563
		564	static void etp_submit (ETP_REQ *req)
		565	{
		566	req->pri -= ETP_PRI_MIN;
		567
		568	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		569	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		570
		571	if (expect_false (req->type == EIO_GROUP))
		572	{
		573	/* I hope this is worth it :/ */
		574	X_LOCK (reqlock);
		575	++nreqs;
		576	X_UNLOCK (reqlock);
		577
		578	X_LOCK (reslock);
		579
		580	++npending;
		581
		582	if (!reqq_push (&res_queue, req) && want_poll_cb)
		583	want_poll_cb ();
		584
		585	X_UNLOCK (reslock);
		586	}
		587	else
		588	{
		589	X_LOCK (reqlock);
		590	++nreqs;
		591	++nready;
		592	reqq_push (&req_queue, req);
		593	X_COND_SIGNAL (reqwait);
		594	X_UNLOCK (reqlock);
		595
		596	etp_maybe_start_thread ();
		597	}
		598	}
		599
		600	static void etp_set_max_poll_time (double nseconds)
		601	{
		602	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		603	max_poll_time = nseconds * EIO_TICKS;
		604	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		605	}
		606
		607	static void etp_set_max_poll_reqs (unsigned int maxreqs)
		608	{
		609	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		610	max_poll_reqs = maxreqs;
		611	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		612	}
		613
		614	static void etp_set_max_idle (unsigned int nthreads)
		615	{
		616	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		617	max_idle = nthreads <= 0 ? 1 : nthreads;
		618	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		619	}
		620
		621	static void etp_set_min_parallel (unsigned int nthreads)
		622	{
		623	if (wanted < nthreads)
		624	wanted = nthreads;
		625	}
		626
		627	static void etp_set_max_parallel (unsigned int nthreads)
		628	{
		629	if (wanted > nthreads)
		630	wanted = nthreads;
		631
		632	while (started > wanted)
		633	etp_end_thread ();
		634	}
		635
		636	/*****************************************************************************/
		637
		638	static void grp_try_feed (eio_req *grp)
		639	{
		640	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		641	{
		642	grp->flags &= ~EIO_FLAG_GROUPADD;
		643
		644	EIO_FEED (grp);
		645
		646	/* stop if no progress has been made */
		647	if (!(grp->flags & EIO_FLAG_GROUPADD))
		648	{
		649	grp->feed = 0;
		650	break;
		651	}
		652	}
		653	}
		654
		655	static int grp_dec (eio_req *grp)
		656	{
		657	--grp->size;
		658
		659	/* call feeder, if applicable */
		660	grp_try_feed (grp);
		661
		662	/* finish, if done */
		663	if (!grp->size && grp->int1)
		664	return eio_finish (grp);
		665	else
		666	return 0;
		667	}
		668
		669	void eio_destroy (eio_req *req)
		670	{
		671	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		672	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		673
		674	EIO_DESTROY (req);
		675	}
		676
		677	static int eio_finish (eio_req *req)
		678	{
		679	int res = EIO_FINISH (req);
		680
		681	if (req->grp)
		682	{
		683	int res2;
		684	eio_req *grp = req->grp;
		685
		686	/* unlink request */
		687	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		688	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		689
		690	if (grp->grp_first == req)
		691	grp->grp_first = req->grp_next;
		692
		693	res2 = grp_dec (grp);
		694
		695	if (!res && res2)
		696	res = res2;
		697	}
		698
		699	eio_destroy (req);
		700
		701	return res;
		702	}
		703
		704	void eio_grp_cancel (eio_req *grp)
		705	{
		706	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		707	eio_cancel (grp);
		708	}
		709
		710	void eio_cancel (eio_req *req)
		711	{
		712	etp_cancel (req);
		713	}
		714
		715	void eio_submit (eio_req *req)
		716	{
		717	etp_submit (req);
		718	}
		719
		720	unsigned int eio_nreqs (void)
		721	{
		722	return etp_nreqs ();
		723	}
		724
		725	unsigned int eio_nready (void)
		726	{
		727	return etp_nready ();
		728	}
		729
		730	unsigned int eio_npending (void)
		731	{
		732	return etp_npending ();
		733	}
		734
		735	unsigned int eio_nthreads (void)
		736	{
		737	return etp_nthreads ();
		738	}
		739
		740	void eio_set_max_poll_time (double nseconds)
		741	{
		742	etp_set_max_poll_time (nseconds);
		743	}
		744
		745	void eio_set_max_poll_reqs (unsigned int maxreqs)
		746	{
		747	etp_set_max_poll_reqs (maxreqs);
		748	}
		749
		750	void eio_set_max_idle (unsigned int nthreads)
		751	{
		752	etp_set_max_idle (nthreads);
		753	}
		754
		755	void eio_set_min_parallel (unsigned int nthreads)
		756	{
		757	etp_set_min_parallel (nthreads);
		758	}
		759
		760	void eio_set_max_parallel (unsigned int nthreads)
		761	{
		762	etp_set_max_parallel (nthreads);
		763	}
		764
		765	int eio_poll (void)
		766	{
		767	return etp_poll ();
		768	}
		769
573	/*****************************************************************************/	770	/*****************************************************************************/
574	/* work around various missing functions */	771	/* work around various missing functions */
575		772
576	#if !HAVE_PREADWRITE	773	#if !HAVE_PREADWRITE
		774	# undef pread
		775	# undef pwrite
577	# define pread eio__pread	776	# define pread eio__pread
578	# define pwrite eio__pwrite	777	# 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		778
587	static ssize_t	779	static ssize_t
588	eio__pread (int fd, void *buf, size_t count, off_t offset)	780	eio__pread (int fd, void *buf, size_t count, off_t offset)
589	{	781	{
590	ssize_t res;	782	ssize_t res;
…		…
608		800
609	X_LOCK (preadwritelock);	801	X_LOCK (preadwritelock);
610	ooffset = lseek (fd, 0, SEEK_CUR);	802	ooffset = lseek (fd, 0, SEEK_CUR);
611	lseek (fd, offset, SEEK_SET);	803	lseek (fd, offset, SEEK_SET);
612	res = write (fd, buf, count);	804	res = write (fd, buf, count);
613	lseek (fd, offset, SEEK_SET);	805	lseek (fd, ooffset, SEEK_SET);
614	X_UNLOCK (preadwritelock);	806	X_UNLOCK (preadwritelock);
615		807
616	return res;	808	return res;
617	}	809	}
618	#endif	810	#endif
619		811
620	#ifndef HAVE_FUTIMES	812	#ifndef HAVE_FUTIMES
621		813
		814	# undef utimes
		815	# undef futimes
622	# define utimes(path,times) eio__utimes (path, times)	816	# define utimes(path,times) eio__utimes (path, times)
623	# define futimes(fd,times) eio__futimes (fd, times)	817	# define futimes(fd,times) eio__futimes (fd, times)
624		818
625	static int	819	static int
626	eio__utimes (const char *filename, const struct timeval times[2])	820	eio__utimes (const char *filename, const struct timeval times[2])
…		…
645	}	839	}
646		840
647	#endif	841	#endif
648		842
649	#if !HAVE_FDATASYNC	843	#if !HAVE_FDATASYNC
		844	# undef fdatasync
650	# define fdatasync fsync	845	# define fdatasync(fd) fsync (fd)
651	#endif	846	#endif
		847
		848	/* sync_file_range always needs emulation */
		849	int
		850	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		851	{
		852	#if HAVE_SYNC_FILE_RANGE
		853	int res;
		854
		855	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		856	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		857	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		858	{
		859	flags = 0
		860	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		861	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		862	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		863	}
		864
		865	res = sync_file_range (fd, offset, nbytes, flags);
		866
		867	if (!res || errno != ENOSYS)
		868	return res;
		869	#endif
		870
		871	/* even though we could play tricks with the flags, it's better to always
		872	* call fdatasync, as that matches the expectation of its users best */
		873	return fdatasync (fd);
		874	}
652		875
653	#if !HAVE_READAHEAD	876	#if !HAVE_READAHEAD
		877	# undef readahead
654	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	878	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
655		879
656	static ssize_t	880	static ssize_t
657	eio__readahead (int fd, off_t offset, size_t count, worker *self)	881	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
658	{	882	{
659	size_t todo = count;	883	size_t todo = count;
660	dBUF;	884	dBUF;
661		885
662	while (todo > 0)	886	while (todo > 0)
…		…
672	return count;	896	return count;
673	}	897	}
674		898
675	#endif	899	#endif
676		900
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 */	901	/* sendfile always needs emulation */
709	static ssize_t	902	static ssize_t
710	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self)	903	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
711	{	904	{
712	ssize_t res;	905	ssize_t res;
713		906
714	if (!count)	907	if (!count)
715	return 0;	908	return 0;
…		…
730	if (res < 0 && sbytes)	923	if (res < 0 && sbytes)
731	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	924	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */
732	res = sbytes;	925	res = sbytes;
733	}	926	}
734		927
		928	# elif defined (__APPLE__)
		929
		930	{
		931	off_t sbytes = count;
		932	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		933
		934	if (res < 0 && errno == EAGAIN && sbytes)
		935	res = sbytes;
		936	}
		937
735	# elif __hpux	938	# elif __hpux
736	res = sendfile (ofd, ifd, offset, count, 0, 0);	939	res = sendfile (ofd, ifd, offset, count, 0, 0);
737		940
738	# elif __solaris	941	# elif __solaris
739	{	942	{
…		…
750	if (res < 0 && sbytes)	953	if (res < 0 && sbytes)
751	res = sbytes;	954	res = sbytes;
752	}	955	}
753		956
754	# endif	957	# endif
		958
		959	#elif defined (_WIN32)
		960
		961	/* does not work, just for documentation of what would need to be done */
		962	{
		963	HANDLE h = TO_SOCKET (ifd);
		964	SetFilePointer (h, offset, 0, FILE_BEGIN);
		965	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		966	}
		967
755	#else	968	#else
756	res = -1;	969	res = -1;
757	errno = ENOSYS;	970	errno = ENOSYS;
758	#endif	971	#endif
759		972
760	if (res < 0	973	if (res < 0
761	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	974	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		975	/* BSDs */
		976	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		977	|| errno == ENOTSUP
		978	#endif
		979	|| errno == EOPNOTSUPP /* BSDs */
762	#if __solaris	980	#if __solaris
763	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	981	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
764	#endif	982	#endif
765	)	983	)
766	)	984	)
…		…
797	}	1015	}
798		1016
799	return res;	1017	return res;
800	}	1018	}
801		1019
		1020	static signed char
		1021	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1022	{
		1023	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1024	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
		1025	}
		1026
		1027	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1028
		1029	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1030	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1031
		1032	static void
		1033	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1034	{
		1035	unsigned char bits [9 + sizeof (ino_t) * 8];
		1036	unsigned char *bit = bits;
		1037
		1038	assert (CHAR_BIT == 8);
		1039	assert (sizeof (eio_dirent) * 8 < 256);
		1040	assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
		1041	assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
		1042
		1043	if (size <= EIO_SORT_FAST)
		1044	return;
		1045
		1046	/* first prepare an array of bits to test in our radix sort */
		1047	/* try to take endianness into account, as well as differences in ino_t sizes */
		1048	/* inode_bits must contain all inodes ORed together */
		1049	/* which is used to skip bits that are 0 everywhere, which is very common */
		1050	{
		1051	ino_t endianness;
		1052	int i, j;
		1053
		1054	/* we store the byte offset of byte n into byte n of "endianness" */
		1055	for (i = 0; i < sizeof (ino_t); ++i)
		1056	((unsigned char *)&endianness)[i] = i;
		1057
		1058	*bit++ = 0;
		1059
		1060	for (i = 0; i < sizeof (ino_t); ++i)
		1061	{
		1062	/* shifting off the byte offsets out of "endianness" */
		1063	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1064	endianness >>= 8;
		1065
		1066	for (j = 0; j < 8; ++j)
		1067	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1068	*bit++ = offs + j;
		1069	}
		1070
		1071	for (j = 0; j < 8; ++j)
		1072	if (score_bits & (1 << j))
		1073	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1074	}
		1075
		1076	/* now actually do the sorting (a variant of MSD radix sort) */
		1077	{
		1078	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1079	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1080	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1081	int stk_idx = 0;
		1082
		1083	base_stk [stk_idx] = dents;
		1084	end_stk [stk_idx] = dents + size;
		1085	bit_stk [stk_idx] = bit - 1;
		1086
		1087	do
		1088	{
		1089	base = base_stk [stk_idx];
		1090	end = end_stk [stk_idx];
		1091	bit = bit_stk [stk_idx];
		1092
		1093	for (;;)
		1094	{
		1095	unsigned char O = *bit >> 3;
		1096	unsigned char M = 1 << (*bit & 7);
		1097
		1098	eio_dirent *a = base;
		1099	eio_dirent *b = end;
		1100
		1101	if (b - a < EIO_SORT_CUTOFF)
		1102	break;
		1103
		1104	/* now bit-partition the array on the bit */
		1105	/* this ugly asymmetric loop seems to perform much better than typical */
		1106	/* partition algos found in the literature */
		1107	do
		1108	if (!(((unsigned char *)a)[O] & M))
		1109	++a;
		1110	else if (!(((unsigned char *)--b)[O] & M))
		1111	{
		1112	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1113	++a;
		1114	}
		1115	while (b > a);
		1116
		1117	/* next bit, or stop, if no bits left in this path */
		1118	if (!*--bit)
		1119	break;
		1120
		1121	base_stk [stk_idx] = a;
		1122	end_stk [stk_idx] = end;
		1123	bit_stk [stk_idx] = bit;
		1124	++stk_idx;
		1125
		1126	end = a;
		1127	}
		1128	}
		1129	while (stk_idx--);
		1130	}
		1131	}
		1132
		1133	static void
		1134	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1135	{
		1136	/* first move the smallest element to the front, to act as a sentinel */
		1137	{
		1138	int i;
		1139	eio_dirent *min = dents;
		1140
		1141	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1142	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1143	if (EIO_DENT_CMP (dents [i], <, *min))
		1144	min = &dents [i];
		1145
		1146	/* swap elements 0 and j (minimum) */
		1147	{
		1148	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1149	}
		1150	}
		1151
		1152	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1153	{
		1154	eio_dirent *i, *j;
		1155
		1156	for (i = dents + 1; i < dents + size; ++i)
		1157	{
		1158	eio_dirent value = *i;
		1159
		1160	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1161	j [1] = j [0];
		1162
		1163	j [1] = value;
		1164	}
		1165	}
		1166	}
		1167
		1168	static void
		1169	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1170	{
		1171	if (size <= 1)
		1172	return; /* our insertion sort relies on size > 0 */
		1173
		1174	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1175	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1176	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1177
		1178	/* use an insertion sort at the end, or for small arrays, */
		1179	/* as insertion sort is more efficient for small partitions */
		1180	eio_dent_insertion_sort (dents, size);
		1181	}
		1182
802	/* read a full directory */	1183	/* read a full directory */
803	static void	1184	static void
804	eio__scandir (eio_req *req, worker *self)	1185	eio__scandir (eio_req *req, etp_worker *self)
805	{	1186	{
806	DIR *dirp;	1187	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;	1188	EIO_STRUCT_DIRENT *entp;
813	char *name, *names;	1189	char *name, *names;
814	int memlen = 4096;	1190	int namesalloc = 4096;
815	int memofs = 0;	1191	int namesoffs = 0;
		1192	int flags = req->int1;
		1193	eio_dirent *dents = 0;
		1194	int dentalloc = 128;
816	int res = 0;	1195	int dentoffs = 0;
		1196	ino_t inode_bits = 0;
817		1197
		1198	req->result = -1;
		1199
		1200	if (!(flags & EIO_READDIR_DENTS))
		1201	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
		1202
818	X_LOCK (etplock);	1203	X_LOCK (wrklock);
		1204	/* the corresponding closedir is in ETP_WORKER_CLEAR */
819	self->dirp = dirp = opendir (req->ptr1);	1205	self->dirp = dirp = opendir (req->ptr1);
820	self->dbuf = u = malloc (sizeof (*u));
821	req->flags |= EIO_FLAG_PTR2_FREE;	1206	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1207	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
822	req->ptr2 = names = malloc (memlen);	1208	req->ptr2 = names = malloc (namesalloc);
823	X_UNLOCK (etplock);	1209	X_UNLOCK (wrklock);
824		1210
825	if (dirp && u && names)	1211	if (dirp && names && (!flags || dents))
826	for (;;)	1212	for (;;)
827	{	1213	{
828	errno = 0;	1214	errno = 0;
829	readdir_r (dirp, &u->d, &entp);	1215	entp = readdir (dirp);
830		1216
831	if (!entp)	1217	if (!entp)
		1218	{
		1219	if (errno)
		1220	break;
		1221
		1222	/* sort etc. */
		1223	req->int1 = flags;
		1224	req->result = dentoffs;
		1225
		1226	if (flags & EIO_READDIR_STAT_ORDER)
		1227	eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
		1228	else if (flags & EIO_READDIR_DIRS_FIRST)
		1229	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1230	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1231	else
		1232	{
		1233	/* in this case, all is known, and we just put dirs first and sort them */
		1234	eio_dirent *oth = dents + dentoffs;
		1235	eio_dirent *dir = dents;
		1236
		1237	/* now partition dirs to the front, and non-dirs to the back */
		1238	/* by walking from both sides and swapping if necessary */
		1239	/* also clear score, so it doesn't influence sorting */
		1240	while (oth > dir)
		1241	{
		1242	if (dir->type == EIO_DT_DIR)
		1243	++dir;
		1244	else if ((--oth)->type == EIO_DT_DIR)
		1245	{
		1246	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1247
		1248	++dir;
		1249	}
		1250	}
		1251
		1252	/* now sort the dirs only */
		1253	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1254	}
		1255
832	break;	1256	break;
		1257	}
833		1258
		1259	/* now add the entry to our list(s) */
834	name = entp->d_name;	1260	name = entp->d_name;
835		1261
		1262	/* skip . and .. entries */
836	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1263	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
837	{	1264	{
838	int len = strlen (name) + 1;	1265	int len = D_NAMLEN (entp) + 1;
839		1266
840	res++;	1267	while (expect_false (namesoffs + len > namesalloc))
841
842	while (memofs + len > memlen)
843	{	1268	{
844	memlen *= 2;	1269	namesalloc *= 2;
845	X_LOCK (etplock);	1270	X_LOCK (wrklock);
846	req->ptr2 = names = realloc (names, memlen);	1271	req->ptr2 = names = realloc (names, namesalloc);
847	X_UNLOCK (etplock);	1272	X_UNLOCK (wrklock);
848		1273
849	if (!names)	1274	if (!names)
850	break;	1275	break;
851	}	1276	}
852		1277
853	memcpy (names + memofs, name, len);	1278	memcpy (names + namesoffs, name, len);
		1279
		1280	if (dents)
		1281	{
		1282	struct eio_dirent *ent;
		1283
		1284	if (expect_false (dentoffs == dentalloc))
		1285	{
		1286	dentalloc *= 2;
		1287	X_LOCK (wrklock);
		1288	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1289	X_UNLOCK (wrklock);
		1290
		1291	if (!dents)
		1292	break;
		1293	}
		1294
		1295	ent = dents + dentoffs;
		1296
		1297	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1298	ent->namelen = len - 1;
		1299	ent->inode = D_INO (entp);
		1300
		1301	inode_bits |= ent->inode;
		1302
		1303	switch (D_TYPE (entp))
		1304	{
		1305	default:
		1306	ent->type = EIO_DT_UNKNOWN;
		1307	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1308	break;
		1309
		1310	#ifdef DT_FIFO
		1311	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1312	#endif
		1313	#ifdef DT_CHR
		1314	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1315	#endif
		1316	#ifdef DT_MPC
		1317	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1318	#endif
		1319	#ifdef DT_DIR
		1320	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1321	#endif
		1322	#ifdef DT_NAM
		1323	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1324	#endif
		1325	#ifdef DT_BLK
		1326	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1327	#endif
		1328	#ifdef DT_MPB
		1329	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1330	#endif
		1331	#ifdef DT_REG
		1332	case DT_REG: ent->type = EIO_DT_REG; break;
		1333	#endif
		1334	#ifdef DT_NWK
		1335	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1336	#endif
		1337	#ifdef DT_CMP
		1338	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1339	#endif
		1340	#ifdef DT_LNK
		1341	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1342	#endif
		1343	#ifdef DT_SOCK
		1344	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1345	#endif
		1346	#ifdef DT_DOOR
		1347	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1348	#endif
		1349	#ifdef DT_WHT
		1350	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1351	#endif
		1352	}
		1353
		1354	ent->score = 7;
		1355
		1356	if (flags & EIO_READDIR_DIRS_FIRST)
		1357	{
		1358	if (ent->type == EIO_DT_UNKNOWN)
		1359	{
		1360	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1361	ent->score = 1;
		1362	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1363	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1364	}
		1365	else if (ent->type == EIO_DT_DIR)
		1366	ent->score = 0;
		1367	}
		1368	}
		1369
854	memofs += len;	1370	namesoffs += len;
		1371	++dentoffs;
		1372	}
		1373
		1374	if (EIO_CANCELLED (req))
		1375	{
		1376	errno = ECANCELED;
		1377	break;
855	}	1378	}
856	}	1379	}
		1380	}
857		1381
858	if (errno)	1382	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
859	res = -1;	1383	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
860		1384	#else
861	req->result = res;	1385
		1386	int
		1387	eio__msync (void *mem, size_t len, int flags)
		1388	{
		1389	if (EIO_MS_ASYNC != MS_SYNC
		1390	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1391	|| EIO_MS_SYNC != MS_SYNC)
		1392	{
		1393	flags = 0
		1394	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1395	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1396	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1397	}
		1398
		1399	return msync (mem, len, flags);
		1400	}
		1401
		1402	#endif
		1403
		1404	int
		1405	eio__mtouch (void *mem, size_t len, int flags)
		1406	{
		1407	intptr_t addr = (intptr_t)mem;
		1408	intptr_t end = addr + len;
		1409	#ifdef PAGESIZE
		1410	const intptr_t page = PAGESIZE;
		1411	#else
		1412	static intptr_t page;
		1413
		1414	if (!page)
		1415	page = sysconf (_SC_PAGESIZE);
		1416	#endif
		1417
		1418	/* round down to start of page, although this is probably useless */
		1419	addr &= ~(page - 1); /* assume page size is always a power of two */
		1420
		1421	if (addr < end)
		1422	if (flags & EIO_MT_MODIFY) /* modify */
		1423	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
		1424	else
		1425	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
		1426
		1427	return 0;
862	}	1428	}
863		1429
864	/*****************************************************************************/	1430	/*****************************************************************************/
865		1431
866	#define ALLOC(len) \	1432	#define ALLOC(len) \
867	if (!req->ptr2) \	1433	if (!req->ptr2) \
868	{ \	1434	{ \
869	X_LOCK (etplock); \	1435	X_LOCK (wrklock); \
870	req->flags |= EIO_FLAG_PTR2_FREE; \	1436	req->flags |= EIO_FLAG_PTR2_FREE; \
871	X_UNLOCK (etplock); \	1437	X_UNLOCK (wrklock); \
872	req->ptr2 = malloc (len); \	1438	req->ptr2 = malloc (len); \
873	if (!req->ptr2) \	1439	if (!req->ptr2) \
874	{ \	1440	{ \
875	errno = ENOMEM; \	1441	errno = ENOMEM; \
876	req->result = -1; \	1442	req->result = -1; \
877	break; \	1443	break; \
878	} \	1444	} \
879	}	1445	}
880		1446
881	X_THREAD_PROC (eio_proc)	1447	X_THREAD_PROC (etp_proc)
882	{	1448	{
883	eio_req *req;	1449	ETP_REQ *req;
884	struct timespec ts;	1450	struct timespec ts;
885	worker *self = (worker *)thr_arg;	1451	etp_worker *self = (etp_worker *)thr_arg;
886		1452
887	/* try to distribute timeouts somewhat randomly */	1453	/* try to distribute timeouts somewhat randomly */
888	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1454	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
889		1455
890	for (;;)	1456	for (;;)
891	{	1457	{
892	ts.tv_sec = time (0) + IDLE_TIMEOUT;
893
894	X_LOCK (reqlock);	1458	X_LOCK (reqlock);
895		1459
896	for (;;)	1460	for (;;)
897	{	1461	{
898	self->req = req = reqq_shift (&req_queue);	1462	self->req = req = reqq_shift (&req_queue);
…		…
900	if (req)	1464	if (req)
901	break;	1465	break;
902		1466
903	++idle;	1467	++idle;
904		1468
		1469	ts.tv_sec = time (0) + IDLE_TIMEOUT;
905	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1470	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
906	{	1471	{
907	if (idle > max_idle)	1472	if (idle > max_idle)
908	{	1473	{
909	--idle;	1474	--idle;
910	X_UNLOCK (reqlock);	1475	X_UNLOCK (reqlock);
911	X_LOCK (etplock);	1476	X_LOCK (wrklock);
912	--started;	1477	--started;
913	X_UNLOCK (etplock);	1478	X_UNLOCK (wrklock);
914	goto quit;	1479	goto quit;
915	}	1480	}
916		1481
917	/* we are allowed to idle, so do so without any timeout */	1482	/* we are allowed to idle, so do so without any timeout */
918	X_COND_WAIT (reqwait, reqlock);	1483	X_COND_WAIT (reqwait, reqlock);
919	ts.tv_sec = time (0) + IDLE_TIMEOUT;
920	}	1484	}
921		1485
922	--idle;	1486	--idle;
923	}	1487	}
924		1488
925	--nready;	1489	--nready;
926		1490
927	X_UNLOCK (reqlock);	1491	X_UNLOCK (reqlock);
928		1492
929	errno = 0; /* strictly unnecessary */	1493	if (req->type < 0)
		1494	goto quit;
930		1495
931	if (!EIO_CANCELLED (req))	1496	if (!EIO_CANCELLED (req))
932	switch (req->type)	1497	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		1498
1033	X_LOCK (reslock);	1499	X_LOCK (reslock);
1034		1500
1035	++npending;	1501	++npending;
1036		1502
1037	if (!reqq_push (&res_queue, req) && etp.want_poll_cb)	1503	if (!reqq_push (&res_queue, req) && want_poll_cb)
1038	etp.want_poll_cb ();	1504	want_poll_cb ();
1039		1505
1040	self->req = 0;	1506	self->req = 0;
1041	worker_clear (self);	1507	etp_worker_clear (self);
1042		1508
1043	X_UNLOCK (reslock);	1509	X_UNLOCK (reslock);
1044	}	1510	}
1045		1511
1046	quit:	1512	quit:
1047	X_LOCK (etplock);	1513	X_LOCK (wrklock);
1048	worker_free (self);	1514	etp_worker_free (self);
1049	X_UNLOCK (etplock);	1515	X_UNLOCK (wrklock);
1050		1516
1051	return 0;	1517	return 0;
1052	}	1518	}
1053		1519
1054	/*****************************************************************************/	1520	/*****************************************************************************/
1055		1521
1056	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1522	int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1057	{	1523	{
1058	etp_init (&etp, want_poll, done_poll);	1524	return etp_init (want_poll, done_poll);
1059	}	1525	}
1060		1526
1061	static void eio_api_destroy (eio_req *req)	1527	static void eio_api_destroy (eio_req *req)
1062	{	1528	{
1063	free (req);	1529	free (req);
…		…
1085	{ \	1551	{ \
1086	eio_api_destroy (req); \	1552	eio_api_destroy (req); \
1087	return 0; \	1553	return 0; \
1088	}	1554	}
1089		1555
		1556	static void eio_execute (etp_worker *self, eio_req *req)
		1557	{
		1558	errno = 0;
		1559
		1560	switch (req->type)
		1561	{
		1562	case EIO_READ: ALLOC (req->size);
		1563	req->result = req->offs >= 0
		1564	? pread (req->int1, req->ptr2, req->size, req->offs)
		1565	: read (req->int1, req->ptr2, req->size); break;
		1566	case EIO_WRITE: req->result = req->offs >= 0
		1567	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1568	: write (req->int1, req->ptr2, req->size); break;
		1569
		1570	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		1571	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
		1572
		1573	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1574	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1575	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1576	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1577	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1578	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1579
		1580	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1581	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1582	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1583	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1584	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1585	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1586
		1587	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1588	case EIO_CLOSE: req->result = close (req->int1); break;
		1589	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1590	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1591	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1592	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1593	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1594	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1595	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1596	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
		1597
		1598	case EIO_READLINK: ALLOC (NAME_MAX);
		1599	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
		1600
		1601	case EIO_SYNC: req->result = 0; sync (); break;
		1602	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1603	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1604	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1605	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1606	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1607
		1608	case EIO_READDIR: eio__scandir (req, self); break;
		1609
		1610	case EIO_BUSY:
		1611	#ifdef _WIN32
		1612	Sleep (req->nv1 * 1000.);
		1613	#else
		1614	{
		1615	struct timeval tv;
		1616
		1617	tv.tv_sec = req->nv1;
		1618	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
		1619
		1620	req->result = select (0, 0, 0, 0, &tv);
		1621	}
		1622	#endif
		1623	break;
		1624
		1625	case EIO_UTIME:
		1626	case EIO_FUTIME:
		1627	{
		1628	struct timeval tv[2];
		1629	struct timeval *times;
		1630
		1631	if (req->nv1 != -1. || req->nv2 != -1.)
		1632	{
		1633	tv[0].tv_sec = req->nv1;
		1634	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		1635	tv[1].tv_sec = req->nv2;
		1636	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		1637
		1638	times = tv;
		1639	}
		1640	else
		1641	times = 0;
		1642
		1643
		1644	req->result = req->type == EIO_FUTIME
		1645	? futimes (req->int1, times)
		1646	: utimes (req->ptr1, times);
		1647	}
		1648	break;
		1649
		1650	case EIO_GROUP:
		1651	abort (); /* handled in eio_request */
		1652
		1653	case EIO_NOP:
		1654	req->result = 0;
		1655	break;
		1656
		1657	case EIO_CUSTOM:
		1658	((void (*)(eio_req *))req->feed) (req);
		1659	break;
		1660
		1661	default:
		1662	req->result = -1;
		1663	break;
		1664	}
		1665
		1666	req->errorno = errno;
		1667	}
		1668
1090	#ifndef EIO_NO_WRAPPERS	1669	#ifndef EIO_NO_WRAPPERS
1091		1670
1092	eio_req *eio_nop (int pri, eio_cb cb, void *data)	1671	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1093	{	1672	{
1094	REQ (EIO_NOP); SEND;	1673	REQ (EIO_NOP); SEND;
…		…
1105	}	1684	}
1106		1685
1107	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	1686	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1108	{	1687	{
1109	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1688	REQ (EIO_FSYNC); req->int1 = fd; SEND;
		1689	}
		1690
		1691	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1692	{
		1693	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1694	}
		1695
		1696	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1697	{
		1698	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1699	}
		1700
		1701	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		1702	{
		1703	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1110	}	1704	}
1111		1705
1112	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	1706	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1113	{	1707	{
1114	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1708	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
…		…
1228	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	1822	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1229	{	1823	{
1230	return eio__1path (EIO_RMDIR, path, pri, cb, data);	1824	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1231	}	1825	}
1232		1826
1233	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	1827	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1234	{	1828	{
1235	return eio__1path (EIO_READDIR, path, pri, cb, data);	1829	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1236	}	1830	}
1237		1831
1238	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	1832	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1239	{	1833	{
1240	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	1834	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1241	}	1835	}
1242		1836
1243	static eio_req *	1837	static eio_req *
1244	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1838	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1245	{	1839	{
…		…
1269	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1863	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1270	{	1864	{
1271	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	1865	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1272	}	1866	}
1273		1867
		1868	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
		1869	{
		1870	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
		1871	}
		1872
1274	#endif	1873	#endif
1275		1874
1276	eio_req *eio_grp (eio_cb cb, void *data)	1875	eio_req *eio_grp (eio_cb cb, void *data)
1277	{	1876	{
1278	const int pri = EIO_PRI_MAX;	1877	const int pri = EIO_PRI_MAX;
…		…
1304		1903
1305	void eio_grp_add (eio_req *grp, eio_req *req)	1904	void eio_grp_add (eio_req *grp, eio_req *req)
1306	{	1905	{
1307	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	1906	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1308		1907
		1908	grp->flags |= EIO_FLAG_GROUPADD;
		1909
1309	++grp->size;	1910	++grp->size;
1310	req->grp = grp;	1911	req->grp = grp;
1311		1912
1312	req->grp_prev = 0;	1913	req->grp_prev = 0;
1313	req->grp_next = grp->grp_first;	1914	req->grp_next = grp->grp_first;
…		…
1321	/*****************************************************************************/	1922	/*****************************************************************************/
1322	/* misc garbage */	1923	/* misc garbage */
1323		1924
1324	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	1925	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1325	{	1926	{
1326	worker wrk;	1927	etp_worker wrk;
		1928	ssize_t ret;
1327		1929
1328	wrk.dbuf = 0;	1930	wrk.dbuf = 0;
1329		1931
1330	eio__sendfile (ofd, ifd, offset, count, &wrk);	1932	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1331		1933
1332	if (wrk.dbuf)	1934	if (wrk.dbuf)
1333	free (wrk.dbuf);	1935	free (wrk.dbuf);
1334	}
1335		1936
		1937	return ret;
		1938	}
		1939

Diff Legend

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