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 1.54 by root, Tue Feb 23 14:32:45 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
67	# define NAME_MAX 4096	139	# define NAME_MAX 4096
68	#endif	140	#endif
69		141
		142	/* used for readlink etc. */
		143	#ifndef PATH_MAX
		144	# define PATH_MAX 4096
		145	#endif
		146
70	/* buffer size for various temporary buffers */	147	/* buffer size for various temporary buffers */
71	#define EIO_BUFSIZE 65536	148	#define EIO_BUFSIZE 65536
72		149
73	#define dBUF \	150	#define dBUF \
74	char *eio_buf; \	151	char *eio_buf; \
75	X_LOCK (etplock); \	152	ETP_WORKER_LOCK (self); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	153	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (etplock); \	154	ETP_WORKER_UNLOCK (self); \
78	errno = ENOMEM; \	155	errno = ENOMEM; \
79	if (!eio_buf) \	156	if (!eio_buf) \
80	return -1;	157	return -1;
81		158
82	#define EIO_TICKS ((1000000 + 1023) >> 10)	159	#define EIO_TICKS ((1000000 + 1023) >> 10)
83		160
84	/*****************************************************************************/	161	/*****************************************************************************/
85		162
		163	#if __GNUC__ >= 3
		164	# define expect(expr,value) __builtin_expect ((expr),(value))
		165	#else
		166	# define expect(expr,value) (expr)
		167	#endif
		168
		169	#define expect_false(expr) expect ((expr) != 0, 0)
		170	#define expect_true(expr) expect ((expr) != 0, 1)
		171
		172	/*****************************************************************************/
		173
		174	#define ETP_PRI_MIN EIO_PRI_MIN
		175	#define ETP_PRI_MAX EIO_PRI_MAX
		176
		177	struct etp_worker;
		178
		179	#define ETP_REQ eio_req
		180	#define ETP_DESTROY(req) eio_destroy (req)
		181	static int eio_finish (eio_req *req);
		182	#define ETP_FINISH(req) eio_finish (req)
		183	static void eio_execute (struct etp_worker *self, eio_req *req);
		184	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		185
		186	#define ETP_WORKER_CLEAR(req) \
		187	if (wrk->dbuf) \
		188	{ \
		189	free (wrk->dbuf); \
		190	wrk->dbuf = 0; \
		191	} \
		192	\
		193	if (wrk->dirp) \
		194	{ \
		195	closedir (wrk->dirp); \
		196	wrk->dirp = 0; \
		197	}
		198
		199	#define ETP_WORKER_COMMON \
		200	void *dbuf; \
		201	DIR *dirp;
		202
		203	/*****************************************************************************/
		204
		205	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		206
86	/* calculcate time difference in ~1/EIO_TICKS of a second */	207	/* calculate time difference in ~1/EIO_TICKS of a second */
87	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	208	static int tvdiff (struct timeval *tv1, struct timeval *tv2)
88	{	209	{
89	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	210	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
90	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	211	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
91	}	212	}
92		213
93	static unsigned int started, idle, wanted = 4;	214	static unsigned int started, idle, wanted = 4;
94		215
95	typedef struct etp_pool
96	{
97	void (*want_poll_cb) (void);	216	static void (*want_poll_cb) (void);
98	void (*done_poll_cb) (void);	217	static void (*done_poll_cb) (void);
99		218
100	unsigned int max_poll_time;	219	static unsigned int max_poll_time; /* reslock */
101	unsigned int max_poll_reqs;	220	static unsigned int max_poll_reqs; /* reslock */
102	} etp_pool;
103		221
104	static volatile unsigned int nreqs, nready, npending;	222	static volatile unsigned int nreqs; /* reqlock */
		223	static volatile unsigned int nready; /* reqlock */
		224	static volatile unsigned int npending; /* reqlock */
105	static volatile unsigned int max_idle = 4;	225	static volatile unsigned int max_idle = 4;
106		226
107	static mutex_t etplock = X_MUTEX_INIT;	227	static mutex_t wrklock = X_MUTEX_INIT;
108	static mutex_t reslock = X_MUTEX_INIT;	228	static mutex_t reslock = X_MUTEX_INIT;
109	static mutex_t reqlock = X_MUTEX_INIT;	229	static mutex_t reqlock = X_MUTEX_INIT;
110	static cond_t reqwait = X_COND_INIT;	230	static cond_t reqwait = X_COND_INIT;
111		231
		232	#if !HAVE_PREADWRITE
		233	/*
		234	* make our pread/pwrite emulation safe against themselves, but not against
		235	* normal read/write by using a mutex. slows down execution a lot,
		236	* but that's your problem, not mine.
		237	*/
		238	static mutex_t preadwritelock = X_MUTEX_INIT;
		239	#endif
		240
112	typedef struct worker	241	typedef struct etp_worker
113	{	242	{
114	/* locked by etplock */	243	/* locked by wrklock */
115	struct worker *prev, *next;	244	struct etp_worker *prev, *next;
116		245
117	thread_t tid;	246	thread_t tid;
118		247
119	/* locked by reslock, reqlock or etplock */	248	/* locked by reslock, reqlock or wrklock */
120	eio_req *req; /* currently processed request */	249	ETP_REQ *req; /* currently processed request */
121	void *dbuf;	250
122	DIR *dirp;	251	ETP_WORKER_COMMON
123	} worker;	252	} etp_worker;
124		253
125	static worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	254	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
		255
		256	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		257	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
126		258
127	/* worker threads management */	259	/* worker threads management */
128		260
129	static void worker_clear (worker *wrk)	261	static void etp_worker_clear (etp_worker *wrk)
130	{	262	{
131	if (wrk->dirp)	263	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	}	264	}
143		265
144	static void worker_free (worker *wrk)	266	static void etp_worker_free (etp_worker *wrk)
145	{	267	{
146	wrk->next->prev = wrk->prev;	268	wrk->next->prev = wrk->prev;
147	wrk->prev->next = wrk->next;	269	wrk->prev->next = wrk->next;
148		270
149	free (wrk);	271	free (wrk);
150	}	272	}
151		273
152	unsigned int eio_nreqs (void)	274	static unsigned int etp_nreqs (void)
153	{	275	{
		276	int retval;
		277	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		278	retval = nreqs;
		279	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154	return nreqs;	280	return retval;
155	}	281	}
156		282
157	unsigned int eio_nready (void)	283	static unsigned int etp_nready (void)
158	{	284	{
159	unsigned int retval;	285	unsigned int retval;
160		286
161	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	287	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162	retval = nready;	288	retval = nready;
163	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	289	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164		290
165	return retval;	291	return retval;
166	}	292	}
167		293
168	unsigned int eio_npending (void)	294	static unsigned int etp_npending (void)
169	{	295	{
170	unsigned int retval;	296	unsigned int retval;
171		297
172	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	298	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173	retval = npending;	299	retval = npending;
174	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	300	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175		301
176	return retval;	302	return retval;
177	}	303	}
178		304
179	unsigned int eio_nthreads (void)	305	static unsigned int etp_nthreads (void)
180	{	306	{
181	unsigned int retval;	307	unsigned int retval;
182		308
183	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	309	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184	retval = started;	310	retval = started;
…		…
191	* a somewhat faster data structure might be nice, but	317	* a somewhat faster data structure might be nice, but
192	* with 8 priorities this actually needs <20 insns	318	* with 8 priorities this actually needs <20 insns
193	* per shift, the most expensive operation.	319	* per shift, the most expensive operation.
194	*/	320	*/
195	typedef struct {	321	typedef struct {
196	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	322	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197	int size;	323	int size;
198	} reqq;	324	} etp_reqq;
199		325
200	static reqq req_queue;	326	static etp_reqq req_queue;
201	static reqq res_queue;	327	static etp_reqq res_queue;
202		328
203	static int reqq_push (reqq *q, eio_req *req)	329	static int reqq_push (etp_reqq *q, ETP_REQ *req)
204	{	330	{
205	int pri = req->pri;	331	int pri = req->pri;
206	req->next = 0;	332	req->next = 0;
207		333
208	if (q->qe[pri])	334	if (q->qe[pri])
…		…
214	q->qe[pri] = q->qs[pri] = req;	340	q->qe[pri] = q->qs[pri] = req;
215		341
216	return q->size++;	342	return q->size++;
217	}	343	}
218		344
219	static eio_req *reqq_shift (reqq *q)	345	static ETP_REQ *reqq_shift (etp_reqq *q)
220	{	346	{
221	int pri;	347	int pri;
222		348
223	if (!q->size)	349	if (!q->size)
224	return 0;	350	return 0;
225		351
226	--q->size;	352	--q->size;
227		353
228	for (pri = EIO_NUM_PRI; pri--; )	354	for (pri = ETP_NUM_PRI; pri--; )
229	{	355	{
230	eio_req *req = q->qs[pri];	356	eio_req *req = q->qs[pri];
231		357
232	if (req)	358	if (req)
233	{	359	{
…		…
241	abort ();	367	abort ();
242	}	368	}
243		369
244	static void etp_atfork_prepare (void)	370	static void etp_atfork_prepare (void)
245	{	371	{
246	X_LOCK (etplock);	372	X_LOCK (wrklock);
247	X_LOCK (reqlock);	373	X_LOCK (reqlock);
248	X_LOCK (reslock);	374	X_LOCK (reslock);
249	#if !HAVE_PREADWRITE	375	#if !HAVE_PREADWRITE
250	X_LOCK (preadwritelock);	376	X_LOCK (preadwritelock);
251	#endif	377	#endif
252	#if !HAVE_READDIR_R
253	X_LOCK (readdirlock);
254	#endif
255	}	378	}
256		379
257	static void etp_atfork_parent (void)	380	static void etp_atfork_parent (void)
258	{	381	{
259	#if !HAVE_READDIR_R
260	X_UNLOCK (readdirlock);
261	#endif
262	#if !HAVE_PREADWRITE	382	#if !HAVE_PREADWRITE
263	X_UNLOCK (preadwritelock);	383	X_UNLOCK (preadwritelock);
264	#endif	384	#endif
265	X_UNLOCK (reslock);	385	X_UNLOCK (reslock);
266	X_UNLOCK (reqlock);	386	X_UNLOCK (reqlock);
267	X_UNLOCK (etplock);	387	X_UNLOCK (wrklock);
268	}	388	}
269		389
270	static void etp_atfork_child (void)	390	static void etp_atfork_child (void)
271	{	391	{
272	eio_req *prv;	392	ETP_REQ *prv;
273		393
274	while (prv = reqq_shift (&req_queue))	394	while ((prv = reqq_shift (&req_queue)))
275	eio_destroy (prv);	395	ETP_DESTROY (prv);
276		396
277	while (prv = reqq_shift (&res_queue))	397	while ((prv = reqq_shift (&res_queue)))
278	eio_destroy (prv);	398	ETP_DESTROY (prv);
279		399
280	while (wrk_first.next != &wrk_first)	400	while (wrk_first.next != &wrk_first)
281	{	401	{
282	worker *wrk = wrk_first.next;	402	etp_worker *wrk = wrk_first.next;
283		403
284	if (wrk->req)	404	if (wrk->req)
285	eio_destroy (wrk->req);	405	ETP_DESTROY (wrk->req);
286		406
287	worker_clear (wrk);	407	etp_worker_clear (wrk);
288	worker_free (wrk);	408	etp_worker_free (wrk);
289	}	409	}
290		410
291	started = 0;	411	started = 0;
292	idle = 0;	412	idle = 0;
293	nreqs = 0;	413	nreqs = 0;
…		…
302	{	422	{
303	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);	423	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
304	}	424	}
305		425
306	static int	426	static int
307	etp_init (etp_pool *etp, void (*want_poll)(void), void (*done_poll)(void))	427	etp_init (void (*want_poll)(void), void (*done_poll)(void))
308	{	428	{
309	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	429	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
310		430
311	pthread_once (&doinit, etp_once_init);	431	pthread_once (&doinit, etp_once_init);
312		432
313	memset (etp, 0, sizeof *etp);
314
315	etp->want_poll_cb = want_poll;	433	want_poll_cb = want_poll;
316	etp->done_poll_cb = done_poll;	434	done_poll_cb = done_poll;
317	}
318		435
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;	436	return 0;
354	}	437	}
355		438
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);	439	X_THREAD_PROC (etp_proc);
407		440
408	static void start_thread (void)	441	static void etp_start_thread (void)
409	{	442	{
410	worker *wrk = calloc (1, sizeof (worker));	443	etp_worker *wrk = calloc (1, sizeof (etp_worker));
411		444
412	/*TODO*/	445	/*TODO*/
413	assert (("unable to allocate worker thread data", wrk));	446	assert (("unable to allocate worker thread data", wrk));
414		447
415	X_LOCK (etplock);	448	X_LOCK (wrklock);
416		449
417	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	450	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
418	{	451	{
419	wrk->prev = &wrk_first;	452	wrk->prev = &wrk_first;
420	wrk->next = wrk_first.next;	453	wrk->next = wrk_first.next;
421	wrk_first.next->prev = wrk;	454	wrk_first.next->prev = wrk;
422	wrk_first.next = wrk;	455	wrk_first.next = wrk;
423	++started;	456	++started;
424	}	457	}
425	else	458	else
426	free (wrk);	459	free (wrk);
427		460
428	X_UNLOCK (etplock);	461	X_UNLOCK (wrklock);
429	}	462	}
430		463
431	static void maybe_start_thread (void)	464	static void etp_maybe_start_thread (void)
432	{	465	{
433	if (eio_nthreads () >= wanted)	466	if (expect_true (etp_nthreads () >= wanted))
434	return;	467	return;
435		468
436	/* todo: maybe use idle here, but might be less exact */	469	/* todo: maybe use idle here, but might be less exact */
437	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	470	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
438	return;	471	return;
439		472
440	start_thread ();	473	etp_start_thread ();
441	}	474	}
442		475
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)	476	static void etp_end_thread (void)
462	{	477	{
463	eio_req *req = calloc (1, sizeof (eio_req));	478	eio_req *req = calloc (1, sizeof (eio_req));
464		479
465	req->type = EIO_QUIT;	480	req->type = -1;
466	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	481	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
467		482
468	X_LOCK (reqlock);	483	X_LOCK (reqlock);
469	reqq_push (&req_queue, req);	484	reqq_push (&req_queue, req);
470	X_COND_SIGNAL (reqwait);	485	X_COND_SIGNAL (reqwait);
471	X_UNLOCK (reqlock);	486	X_UNLOCK (reqlock);
472		487
473	X_LOCK (etplock);	488	X_LOCK (wrklock);
474	--started;	489	--started;
475	X_UNLOCK (etplock);	490	X_UNLOCK (wrklock);
476	}	491	}
477		492
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)	493	static int etp_poll (void)
515	{	494	{
516	int maxreqs = etp.max_poll_reqs;	495	unsigned int maxreqs;
		496	unsigned int maxtime;
517	struct timeval tv_start, tv_now;	497	struct timeval tv_start, tv_now;
518	eio_req *req;
519		498
520	if (etp.max_poll_time)	499	X_LOCK (reslock);
		500	maxreqs = max_poll_reqs;
		501	maxtime = max_poll_time;
		502	X_UNLOCK (reslock);
		503
		504	if (maxtime)
521	gettimeofday (&tv_start, 0);	505	gettimeofday (&tv_start, 0);
522		506
523	for (;;)	507	for (;;)
524	{	508	{
		509	ETP_REQ *req;
		510
525	maybe_start_thread ();	511	etp_maybe_start_thread ();
526		512
527	X_LOCK (reslock);	513	X_LOCK (reslock);
528	req = reqq_shift (&res_queue);	514	req = reqq_shift (&res_queue);
529		515
530	if (req)	516	if (req)
531	{	517	{
532	--npending;	518	--npending;
533		519
534	if (!res_queue.size && etp.done_poll_cb)	520	if (!res_queue.size && done_poll_cb)
535	etp.done_poll_cb ();	521	done_poll_cb ();
536	}	522	}
537		523
538	X_UNLOCK (reslock);	524	X_UNLOCK (reslock);
539		525
540	if (!req)	526	if (!req)
541	return 0;	527	return 0;
542		528
		529	X_LOCK (reqlock);
543	--nreqs;	530	--nreqs;
		531	X_UNLOCK (reqlock);
544		532
545	if (req->type == EIO_GROUP && req->size)	533	if (expect_false (req->type == EIO_GROUP && req->size))
546	{	534	{
547	req->int1 = 1; /* mark request as delayed */	535	req->int1 = 1; /* mark request as delayed */
548	continue;	536	continue;
549	}	537	}
550	else	538	else
551	{	539	{
552	int res = eio_finish (req);	540	int res = ETP_FINISH (req);
553	if (res)	541	if (expect_false (res))
554	return res;	542	return res;
555	}	543	}
556		544
557	if (maxreqs && !--maxreqs)	545	if (expect_false (maxreqs && !--maxreqs))
558	break;	546	break;
559		547
560	if (etp.max_poll_time)	548	if (maxtime)
561	{	549	{
562	gettimeofday (&tv_now, 0);	550	gettimeofday (&tv_now, 0);
563		551
564	if (tvdiff (&tv_start, &tv_now) >= etp.max_poll_time)	552	if (tvdiff (&tv_start, &tv_now) >= maxtime)
565	break;	553	break;
566	}	554	}
567	}	555	}
568		556
569	errno = EAGAIN;	557	errno = EAGAIN;
570	return -1;	558	return -1;
571	}	559	}
572		560
		561	static void etp_cancel (ETP_REQ *req)
		562	{
		563	X_LOCK (wrklock);
		564	req->flags |= EIO_FLAG_CANCELLED;
		565	X_UNLOCK (wrklock);
		566
		567	eio_grp_cancel (req);
		568	}
		569
		570	static void etp_submit (ETP_REQ *req)
		571	{
		572	req->pri -= ETP_PRI_MIN;
		573
		574	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		575	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		576
		577	if (expect_false (req->type == EIO_GROUP))
		578	{
		579	/* I hope this is worth it :/ */
		580	X_LOCK (reqlock);
		581	++nreqs;
		582	X_UNLOCK (reqlock);
		583
		584	X_LOCK (reslock);
		585
		586	++npending;
		587
		588	if (!reqq_push (&res_queue, req) && want_poll_cb)
		589	want_poll_cb ();
		590
		591	X_UNLOCK (reslock);
		592	}
		593	else
		594	{
		595	X_LOCK (reqlock);
		596	++nreqs;
		597	++nready;
		598	reqq_push (&req_queue, req);
		599	X_COND_SIGNAL (reqwait);
		600	X_UNLOCK (reqlock);
		601
		602	etp_maybe_start_thread ();
		603	}
		604	}
		605
		606	static void etp_set_max_poll_time (double nseconds)
		607	{
		608	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		609	max_poll_time = nseconds * EIO_TICKS;
		610	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		611	}
		612
		613	static void etp_set_max_poll_reqs (unsigned int maxreqs)
		614	{
		615	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		616	max_poll_reqs = maxreqs;
		617	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		618	}
		619
		620	static void etp_set_max_idle (unsigned int nthreads)
		621	{
		622	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		623	max_idle = nthreads <= 0 ? 1 : nthreads;
		624	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		625	}
		626
		627	static void etp_set_min_parallel (unsigned int nthreads)
		628	{
		629	if (wanted < nthreads)
		630	wanted = nthreads;
		631	}
		632
		633	static void etp_set_max_parallel (unsigned int nthreads)
		634	{
		635	if (wanted > nthreads)
		636	wanted = nthreads;
		637
		638	while (started > wanted)
		639	etp_end_thread ();
		640	}
		641
		642	/*****************************************************************************/
		643
		644	static void grp_try_feed (eio_req *grp)
		645	{
		646	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		647	{
		648	grp->flags &= ~EIO_FLAG_GROUPADD;
		649
		650	EIO_FEED (grp);
		651
		652	/* stop if no progress has been made */
		653	if (!(grp->flags & EIO_FLAG_GROUPADD))
		654	{
		655	grp->feed = 0;
		656	break;
		657	}
		658	}
		659	}
		660
		661	static int grp_dec (eio_req *grp)
		662	{
		663	--grp->size;
		664
		665	/* call feeder, if applicable */
		666	grp_try_feed (grp);
		667
		668	/* finish, if done */
		669	if (!grp->size && grp->int1)
		670	return eio_finish (grp);
		671	else
		672	return 0;
		673	}
		674
		675	void eio_destroy (eio_req *req)
		676	{
		677	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		678	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		679
		680	EIO_DESTROY (req);
		681	}
		682
		683	static int eio_finish (eio_req *req)
		684	{
		685	int res = EIO_FINISH (req);
		686
		687	if (req->grp)
		688	{
		689	int res2;
		690	eio_req *grp = req->grp;
		691
		692	/* unlink request */
		693	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		694	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		695
		696	if (grp->grp_first == req)
		697	grp->grp_first = req->grp_next;
		698
		699	res2 = grp_dec (grp);
		700
		701	if (!res && res2)
		702	res = res2;
		703	}
		704
		705	eio_destroy (req);
		706
		707	return res;
		708	}
		709
		710	void eio_grp_cancel (eio_req *grp)
		711	{
		712	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		713	eio_cancel (grp);
		714	}
		715
		716	void eio_cancel (eio_req *req)
		717	{
		718	etp_cancel (req);
		719	}
		720
		721	void eio_submit (eio_req *req)
		722	{
		723	etp_submit (req);
		724	}
		725
		726	unsigned int eio_nreqs (void)
		727	{
		728	return etp_nreqs ();
		729	}
		730
		731	unsigned int eio_nready (void)
		732	{
		733	return etp_nready ();
		734	}
		735
		736	unsigned int eio_npending (void)
		737	{
		738	return etp_npending ();
		739	}
		740
		741	unsigned int eio_nthreads (void)
		742	{
		743	return etp_nthreads ();
		744	}
		745
		746	void eio_set_max_poll_time (double nseconds)
		747	{
		748	etp_set_max_poll_time (nseconds);
		749	}
		750
		751	void eio_set_max_poll_reqs (unsigned int maxreqs)
		752	{
		753	etp_set_max_poll_reqs (maxreqs);
		754	}
		755
		756	void eio_set_max_idle (unsigned int nthreads)
		757	{
		758	etp_set_max_idle (nthreads);
		759	}
		760
		761	void eio_set_min_parallel (unsigned int nthreads)
		762	{
		763	etp_set_min_parallel (nthreads);
		764	}
		765
		766	void eio_set_max_parallel (unsigned int nthreads)
		767	{
		768	etp_set_max_parallel (nthreads);
		769	}
		770
		771	int eio_poll (void)
		772	{
		773	return etp_poll ();
		774	}
		775
573	/*****************************************************************************/	776	/*****************************************************************************/
574	/* work around various missing functions */	777	/* work around various missing functions */
575		778
576	#if !HAVE_PREADWRITE	779	#if !HAVE_PREADWRITE
		780	# undef pread
		781	# undef pwrite
577	# define pread eio__pread	782	# define pread eio__pread
578	# define pwrite eio__pwrite	783	# 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		784
587	static ssize_t	785	static ssize_t
588	eio__pread (int fd, void *buf, size_t count, off_t offset)	786	eio__pread (int fd, void *buf, size_t count, off_t offset)
589	{	787	{
590	ssize_t res;	788	ssize_t res;
…		…
608		806
609	X_LOCK (preadwritelock);	807	X_LOCK (preadwritelock);
610	ooffset = lseek (fd, 0, SEEK_CUR);	808	ooffset = lseek (fd, 0, SEEK_CUR);
611	lseek (fd, offset, SEEK_SET);	809	lseek (fd, offset, SEEK_SET);
612	res = write (fd, buf, count);	810	res = write (fd, buf, count);
613	lseek (fd, offset, SEEK_SET);	811	lseek (fd, ooffset, SEEK_SET);
614	X_UNLOCK (preadwritelock);	812	X_UNLOCK (preadwritelock);
615		813
616	return res;	814	return res;
617	}	815	}
618	#endif	816	#endif
619		817
620	#ifndef HAVE_FUTIMES	818	#ifndef HAVE_FUTIMES
621		819
		820	# undef utimes
		821	# undef futimes
622	# define utimes(path,times) eio__utimes (path, times)	822	# define utimes(path,times) eio__utimes (path, times)
623	# define futimes(fd,times) eio__futimes (fd, times)	823	# define futimes(fd,times) eio__futimes (fd, times)
624		824
625	static int	825	static int
626	eio__utimes (const char *filename, const struct timeval times[2])	826	eio__utimes (const char *filename, const struct timeval times[2])
…		…
645	}	845	}
646		846
647	#endif	847	#endif
648		848
649	#if !HAVE_FDATASYNC	849	#if !HAVE_FDATASYNC
		850	# undef fdatasync
650	# define fdatasync fsync	851	# define fdatasync(fd) fsync (fd)
651	#endif	852	#endif
		853
		854	/* sync_file_range always needs emulation */
		855	int
		856	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		857	{
		858	#if HAVE_SYNC_FILE_RANGE
		859	int res;
		860
		861	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		862	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		863	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		864	{
		865	flags = 0
		866	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		867	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		868	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		869	}
		870
		871	res = sync_file_range (fd, offset, nbytes, flags);
		872
		873	if (!res || errno != ENOSYS)
		874	return res;
		875	#endif
		876
		877	/* even though we could play tricks with the flags, it's better to always
		878	* call fdatasync, as that matches the expectation of its users best */
		879	return fdatasync (fd);
		880	}
652		881
653	#if !HAVE_READAHEAD	882	#if !HAVE_READAHEAD
		883	# undef readahead
654	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	884	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
655		885
656	static ssize_t	886	static ssize_t
657	eio__readahead (int fd, off_t offset, size_t count, worker *self)	887	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
658	{	888	{
659	size_t todo = count;	889	size_t todo = count;
660	dBUF;	890	dBUF;
661		891
662	while (todo > 0)	892	while (todo > 0)
…		…
672	return count;	902	return count;
673	}	903	}
674		904
675	#endif	905	#endif
676		906
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 */	907	/* sendfile always needs emulation */
709	static ssize_t	908	static ssize_t
710	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self)	909	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
711	{	910	{
712	ssize_t res;	911	ssize_t res;
713		912
714	if (!count)	913	if (!count)
715	return 0;	914	return 0;
716		915
717	#if HAVE_SENDFILE	916	#if HAVE_SENDFILE
718	# if __linux	917	# if __linux
719	res = sendfile (ofd, ifd, &offset, count);	918	res = sendfile (ofd, ifd, &offset, count);
720		919
721	# elif __freebsd	920	# elif __FreeBSD__
722	/*	921	/*
723	* Of course, the freebsd sendfile is a dire hack with no thoughts	922	* Of course, the freebsd sendfile is a dire hack with no thoughts
724	* wasted on making it similar to other I/O functions.	923	* wasted on making it similar to other I/O functions.
725	*/	924	*/
726	{	925	{
727	off_t sbytes;	926	off_t sbytes;
728	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	927	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
729		928
730	if (res < 0 && sbytes)	929	#if 0 /* according to the manpage, this is correct, but broken behaviour */
731	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	930	/* freebsd' sendfile will return 0 on success */
		931	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		932	/* not on e.g. EIO or EPIPE - sounds broken */
		933	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
		934	res = sbytes;
		935	#endif
		936
		937	/* according to source inspection, this is correct, and useful behaviour */
		938	if (sbytes)
		939	res = sbytes;
		940	}
		941
		942	# elif defined (__APPLE__)
		943
		944	{
		945	off_t sbytes = count;
		946	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		947
		948	/* according to the manpage, sbytes is always valid */
		949	if (sbytes)
732	res = sbytes;	950	res = sbytes;
733	}	951	}
734		952
735	# elif __hpux	953	# elif __hpux
736	res = sendfile (ofd, ifd, offset, count, 0, 0);	954	res = sendfile (ofd, ifd, offset, count, 0, 0);
…		…
750	if (res < 0 && sbytes)	968	if (res < 0 && sbytes)
751	res = sbytes;	969	res = sbytes;
752	}	970	}
753		971
754	# endif	972	# endif
		973
		974	#elif defined (_WIN32)
		975
		976	/* does not work, just for documentation of what would need to be done */
		977	{
		978	HANDLE h = TO_SOCKET (ifd);
		979	SetFilePointer (h, offset, 0, FILE_BEGIN);
		980	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		981	}
		982
755	#else	983	#else
756	res = -1;	984	res = -1;
757	errno = ENOSYS;	985	errno = ENOSYS;
758	#endif	986	#endif
759		987
760	if (res < 0	988	if (res < 0
761	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	989	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		990	/* BSDs */
		991	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		992	|| errno == ENOTSUP
		993	#endif
		994	|| errno == EOPNOTSUPP /* BSDs */
762	#if __solaris	995	#if __solaris
763	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	996	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
764	#endif	997	#endif
765	)	998	)
766	)	999	)
…		…
797	}	1030	}
798		1031
799	return res;	1032	return res;
800	}	1033	}
801		1034
		1035	static signed char
		1036	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1037	{
		1038	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1039	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
		1040	}
		1041
		1042	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1043
		1044	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1045	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1046
		1047	static void
		1048	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1049	{
		1050	unsigned char bits [9 + sizeof (ino_t) * 8];
		1051	unsigned char *bit = bits;
		1052
		1053	assert (CHAR_BIT == 8);
		1054	assert (sizeof (eio_dirent) * 8 < 256);
		1055	assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
		1056	assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
		1057
		1058	if (size <= EIO_SORT_FAST)
		1059	return;
		1060
		1061	/* first prepare an array of bits to test in our radix sort */
		1062	/* try to take endianness into account, as well as differences in ino_t sizes */
		1063	/* inode_bits must contain all inodes ORed together */
		1064	/* which is used to skip bits that are 0 everywhere, which is very common */
		1065	{
		1066	ino_t endianness;
		1067	int i, j;
		1068
		1069	/* we store the byte offset of byte n into byte n of "endianness" */
		1070	for (i = 0; i < sizeof (ino_t); ++i)
		1071	((unsigned char *)&endianness)[i] = i;
		1072
		1073	*bit++ = 0;
		1074
		1075	for (i = 0; i < sizeof (ino_t); ++i)
		1076	{
		1077	/* shifting off the byte offsets out of "endianness" */
		1078	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1079	endianness >>= 8;
		1080
		1081	for (j = 0; j < 8; ++j)
		1082	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1083	*bit++ = offs + j;
		1084	}
		1085
		1086	for (j = 0; j < 8; ++j)
		1087	if (score_bits & (1 << j))
		1088	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1089	}
		1090
		1091	/* now actually do the sorting (a variant of MSD radix sort) */
		1092	{
		1093	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1094	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1095	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1096	int stk_idx = 0;
		1097
		1098	base_stk [stk_idx] = dents;
		1099	end_stk [stk_idx] = dents + size;
		1100	bit_stk [stk_idx] = bit - 1;
		1101
		1102	do
		1103	{
		1104	base = base_stk [stk_idx];
		1105	end = end_stk [stk_idx];
		1106	bit = bit_stk [stk_idx];
		1107
		1108	for (;;)
		1109	{
		1110	unsigned char O = *bit >> 3;
		1111	unsigned char M = 1 << (*bit & 7);
		1112
		1113	eio_dirent *a = base;
		1114	eio_dirent *b = end;
		1115
		1116	if (b - a < EIO_SORT_CUTOFF)
		1117	break;
		1118
		1119	/* now bit-partition the array on the bit */
		1120	/* this ugly asymmetric loop seems to perform much better than typical */
		1121	/* partition algos found in the literature */
		1122	do
		1123	if (!(((unsigned char *)a)[O] & M))
		1124	++a;
		1125	else if (!(((unsigned char *)--b)[O] & M))
		1126	{
		1127	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1128	++a;
		1129	}
		1130	while (b > a);
		1131
		1132	/* next bit, or stop, if no bits left in this path */
		1133	if (!*--bit)
		1134	break;
		1135
		1136	base_stk [stk_idx] = a;
		1137	end_stk [stk_idx] = end;
		1138	bit_stk [stk_idx] = bit;
		1139	++stk_idx;
		1140
		1141	end = a;
		1142	}
		1143	}
		1144	while (stk_idx--);
		1145	}
		1146	}
		1147
		1148	static void
		1149	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1150	{
		1151	/* first move the smallest element to the front, to act as a sentinel */
		1152	{
		1153	int i;
		1154	eio_dirent *min = dents;
		1155
		1156	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1157	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1158	if (EIO_DENT_CMP (dents [i], <, *min))
		1159	min = &dents [i];
		1160
		1161	/* swap elements 0 and j (minimum) */
		1162	{
		1163	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1164	}
		1165	}
		1166
		1167	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1168	{
		1169	eio_dirent *i, *j;
		1170
		1171	for (i = dents + 1; i < dents + size; ++i)
		1172	{
		1173	eio_dirent value = *i;
		1174
		1175	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1176	j [1] = j [0];
		1177
		1178	j [1] = value;
		1179	}
		1180	}
		1181	}
		1182
		1183	static void
		1184	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1185	{
		1186	if (size <= 1)
		1187	return; /* our insertion sort relies on size > 0 */
		1188
		1189	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1190	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1191	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1192
		1193	/* use an insertion sort at the end, or for small arrays, */
		1194	/* as insertion sort is more efficient for small partitions */
		1195	eio_dent_insertion_sort (dents, size);
		1196	}
		1197
802	/* read a full directory */	1198	/* read a full directory */
803	static void	1199	static void
804	eio__scandir (eio_req *req, worker *self)	1200	eio__scandir (eio_req *req, etp_worker *self)
805	{	1201	{
806	DIR *dirp;	1202	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;	1203	EIO_STRUCT_DIRENT *entp;
813	char *name, *names;	1204	char *name, *names;
814	int memlen = 4096;	1205	int namesalloc = 4096;
815	int memofs = 0;	1206	int namesoffs = 0;
		1207	int flags = req->int1;
		1208	eio_dirent *dents = 0;
		1209	int dentalloc = 128;
816	int res = 0;	1210	int dentoffs = 0;
		1211	ino_t inode_bits = 0;
817		1212
		1213	req->result = -1;
		1214
		1215	if (!(flags & EIO_READDIR_DENTS))
		1216	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
		1217
818	X_LOCK (etplock);	1218	X_LOCK (wrklock);
		1219	/* the corresponding closedir is in ETP_WORKER_CLEAR */
819	self->dirp = dirp = opendir (req->ptr1);	1220	self->dirp = dirp = opendir (req->ptr1);
820	self->dbuf = u = malloc (sizeof (*u));
821	req->flags |= EIO_FLAG_PTR2_FREE;	1221	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1222	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
822	req->ptr2 = names = malloc (memlen);	1223	req->ptr2 = names = malloc (namesalloc);
823	X_UNLOCK (etplock);	1224	X_UNLOCK (wrklock);
824		1225
825	if (dirp && u && names)	1226	if (dirp && names && (!flags || dents))
826	for (;;)	1227	for (;;)
827	{	1228	{
828	errno = 0;	1229	errno = 0;
829	readdir_r (dirp, &u->d, &entp);	1230	entp = readdir (dirp);
830		1231
831	if (!entp)	1232	if (!entp)
		1233	{
		1234	if (errno)
		1235	break;
		1236
		1237	/* sort etc. */
		1238	req->int1 = flags;
		1239	req->result = dentoffs;
		1240
		1241	if (flags & EIO_READDIR_STAT_ORDER)
		1242	eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
		1243	else if (flags & EIO_READDIR_DIRS_FIRST)
		1244	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1245	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1246	else
		1247	{
		1248	/* in this case, all is known, and we just put dirs first and sort them */
		1249	eio_dirent *oth = dents + dentoffs;
		1250	eio_dirent *dir = dents;
		1251
		1252	/* now partition dirs to the front, and non-dirs to the back */
		1253	/* by walking from both sides and swapping if necessary */
		1254	/* also clear score, so it doesn't influence sorting */
		1255	while (oth > dir)
		1256	{
		1257	if (dir->type == EIO_DT_DIR)
		1258	++dir;
		1259	else if ((--oth)->type == EIO_DT_DIR)
		1260	{
		1261	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1262
		1263	++dir;
		1264	}
		1265	}
		1266
		1267	/* now sort the dirs only */
		1268	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1269	}
		1270
832	break;	1271	break;
		1272	}
833		1273
		1274	/* now add the entry to our list(s) */
834	name = entp->d_name;	1275	name = entp->d_name;
835		1276
		1277	/* skip . and .. entries */
836	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1278	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
837	{	1279	{
838	int len = strlen (name) + 1;	1280	int len = D_NAMLEN (entp) + 1;
839		1281
840	res++;	1282	while (expect_false (namesoffs + len > namesalloc))
841
842	while (memofs + len > memlen)
843	{	1283	{
844	memlen *= 2;	1284	namesalloc *= 2;
845	X_LOCK (etplock);	1285	X_LOCK (wrklock);
846	req->ptr2 = names = realloc (names, memlen);	1286	req->ptr2 = names = realloc (names, namesalloc);
847	X_UNLOCK (etplock);	1287	X_UNLOCK (wrklock);
848		1288
849	if (!names)	1289	if (!names)
850	break;	1290	break;
851	}	1291	}
852		1292
853	memcpy (names + memofs, name, len);	1293	memcpy (names + namesoffs, name, len);
		1294
		1295	if (dents)
		1296	{
		1297	struct eio_dirent *ent;
		1298
		1299	if (expect_false (dentoffs == dentalloc))
		1300	{
		1301	dentalloc *= 2;
		1302	X_LOCK (wrklock);
		1303	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1304	X_UNLOCK (wrklock);
		1305
		1306	if (!dents)
		1307	break;
		1308	}
		1309
		1310	ent = dents + dentoffs;
		1311
		1312	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1313	ent->namelen = len - 1;
		1314	ent->inode = D_INO (entp);
		1315
		1316	inode_bits |= ent->inode;
		1317
		1318	switch (D_TYPE (entp))
		1319	{
		1320	default:
		1321	ent->type = EIO_DT_UNKNOWN;
		1322	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1323	break;
		1324
		1325	#ifdef DT_FIFO
		1326	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1327	#endif
		1328	#ifdef DT_CHR
		1329	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1330	#endif
		1331	#ifdef DT_MPC
		1332	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1333	#endif
		1334	#ifdef DT_DIR
		1335	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1336	#endif
		1337	#ifdef DT_NAM
		1338	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1339	#endif
		1340	#ifdef DT_BLK
		1341	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1342	#endif
		1343	#ifdef DT_MPB
		1344	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1345	#endif
		1346	#ifdef DT_REG
		1347	case DT_REG: ent->type = EIO_DT_REG; break;
		1348	#endif
		1349	#ifdef DT_NWK
		1350	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1351	#endif
		1352	#ifdef DT_CMP
		1353	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1354	#endif
		1355	#ifdef DT_LNK
		1356	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1357	#endif
		1358	#ifdef DT_SOCK
		1359	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1360	#endif
		1361	#ifdef DT_DOOR
		1362	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1363	#endif
		1364	#ifdef DT_WHT
		1365	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1366	#endif
		1367	}
		1368
		1369	ent->score = 7;
		1370
		1371	if (flags & EIO_READDIR_DIRS_FIRST)
		1372	{
		1373	if (ent->type == EIO_DT_UNKNOWN)
		1374	{
		1375	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1376	ent->score = 1;
		1377	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1378	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1379	}
		1380	else if (ent->type == EIO_DT_DIR)
		1381	ent->score = 0;
		1382	}
		1383	}
		1384
854	memofs += len;	1385	namesoffs += len;
		1386	++dentoffs;
		1387	}
		1388
		1389	if (EIO_CANCELLED (req))
		1390	{
		1391	errno = ECANCELED;
		1392	break;
855	}	1393	}
856	}	1394	}
		1395	}
857		1396
858	if (errno)	1397	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
859	res = -1;	1398	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
860		1399	#else
861	req->result = res;	1400
		1401	int
		1402	eio__msync (void *mem, size_t len, int flags)
		1403	{
		1404	if (EIO_MS_ASYNC != MS_SYNC
		1405	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1406	|| EIO_MS_SYNC != MS_SYNC)
		1407	{
		1408	flags = 0
		1409	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1410	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1411	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1412	}
		1413
		1414	return msync (mem, len, flags);
		1415	}
		1416
		1417	#endif
		1418
		1419	int
		1420	eio__mtouch (void *mem, size_t len, int flags)
		1421	{
		1422	intptr_t addr = (intptr_t)mem;
		1423	intptr_t end = addr + len;
		1424	#ifdef PAGESIZE
		1425	const intptr_t page = PAGESIZE;
		1426	#else
		1427	static intptr_t page;
		1428
		1429	if (!page)
		1430	page = sysconf (_SC_PAGESIZE);
		1431	#endif
		1432
		1433	/* round down to start of page, although this is probably useless */
		1434	addr &= ~(page - 1); /* assume page size is always a power of two */
		1435
		1436	if (addr < end)
		1437	if (flags & EIO_MT_MODIFY) /* modify */
		1438	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
		1439	else
		1440	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
		1441
		1442	return 0;
862	}	1443	}
863		1444
864	/*****************************************************************************/	1445	/*****************************************************************************/
865		1446
866	#define ALLOC(len) \	1447	#define ALLOC(len) \
867	if (!req->ptr2) \	1448	if (!req->ptr2) \
868	{ \	1449	{ \
869	X_LOCK (etplock); \	1450	X_LOCK (wrklock); \
870	req->flags |= EIO_FLAG_PTR2_FREE; \	1451	req->flags |= EIO_FLAG_PTR2_FREE; \
871	X_UNLOCK (etplock); \	1452	X_UNLOCK (wrklock); \
872	req->ptr2 = malloc (len); \	1453	req->ptr2 = malloc (len); \
873	if (!req->ptr2) \	1454	if (!req->ptr2) \
874	{ \	1455	{ \
875	errno = ENOMEM; \	1456	errno = ENOMEM; \
876	req->result = -1; \	1457	req->result = -1; \
877	break; \	1458	break; \
878	} \	1459	} \
879	}	1460	}
880		1461
881	X_THREAD_PROC (eio_proc)	1462	X_THREAD_PROC (etp_proc)
882	{	1463	{
883	eio_req *req;	1464	ETP_REQ *req;
884	struct timespec ts;	1465	struct timespec ts;
885	worker *self = (worker *)thr_arg;	1466	etp_worker *self = (etp_worker *)thr_arg;
886		1467
887	/* try to distribute timeouts somewhat randomly */	1468	/* try to distribute timeouts somewhat randomly */
888	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1469	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
889		1470
890	for (;;)	1471	for (;;)
891	{	1472	{
892	ts.tv_sec = time (0) + IDLE_TIMEOUT;
893
894	X_LOCK (reqlock);	1473	X_LOCK (reqlock);
895		1474
896	for (;;)	1475	for (;;)
897	{	1476	{
898	self->req = req = reqq_shift (&req_queue);	1477	self->req = req = reqq_shift (&req_queue);
…		…
900	if (req)	1479	if (req)
901	break;	1480	break;
902		1481
903	++idle;	1482	++idle;
904		1483
		1484	ts.tv_sec = time (0) + IDLE_TIMEOUT;
905	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1485	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
906	{	1486	{
907	if (idle > max_idle)	1487	if (idle > max_idle)
908	{	1488	{
909	--idle;	1489	--idle;
910	X_UNLOCK (reqlock);	1490	X_UNLOCK (reqlock);
911	X_LOCK (etplock);	1491	X_LOCK (wrklock);
912	--started;	1492	--started;
913	X_UNLOCK (etplock);	1493	X_UNLOCK (wrklock);
914	goto quit;	1494	goto quit;
915	}	1495	}
916		1496
917	/* we are allowed to idle, so do so without any timeout */	1497	/* we are allowed to idle, so do so without any timeout */
918	X_COND_WAIT (reqwait, reqlock);	1498	X_COND_WAIT (reqwait, reqlock);
919	ts.tv_sec = time (0) + IDLE_TIMEOUT;
920	}	1499	}
921		1500
922	--idle;	1501	--idle;
923	}	1502	}
924		1503
925	--nready;	1504	--nready;
926		1505
927	X_UNLOCK (reqlock);	1506	X_UNLOCK (reqlock);
928		1507
929	errno = 0; /* strictly unnecessary */	1508	if (req->type < 0)
		1509	goto quit;
930		1510
931	if (!EIO_CANCELLED (req))	1511	if (!EIO_CANCELLED (req))
932	switch (req->type)	1512	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		1513
1033	X_LOCK (reslock);	1514	X_LOCK (reslock);
1034		1515
1035	++npending;	1516	++npending;
1036		1517
1037	if (!reqq_push (&res_queue, req) && etp.want_poll_cb)	1518	if (!reqq_push (&res_queue, req) && want_poll_cb)
1038	etp.want_poll_cb ();	1519	want_poll_cb ();
1039		1520
1040	self->req = 0;	1521	self->req = 0;
1041	worker_clear (self);	1522	etp_worker_clear (self);
1042		1523
1043	X_UNLOCK (reslock);	1524	X_UNLOCK (reslock);
1044	}	1525	}
1045		1526
1046	quit:	1527	quit:
1047	X_LOCK (etplock);	1528	X_LOCK (wrklock);
1048	worker_free (self);	1529	etp_worker_free (self);
1049	X_UNLOCK (etplock);	1530	X_UNLOCK (wrklock);
1050		1531
1051	return 0;	1532	return 0;
1052	}	1533	}
1053		1534
1054	/*****************************************************************************/	1535	/*****************************************************************************/
1055		1536
1056	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1537	int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1057	{	1538	{
1058	etp_init (&etp, want_poll, done_poll);	1539	return etp_init (want_poll, done_poll);
1059	}	1540	}
1060		1541
1061	static void eio_api_destroy (eio_req *req)	1542	static void eio_api_destroy (eio_req *req)
1062	{	1543	{
1063	free (req);	1544	free (req);
…		…
1085	{ \	1566	{ \
1086	eio_api_destroy (req); \	1567	eio_api_destroy (req); \
1087	return 0; \	1568	return 0; \
1088	}	1569	}
1089		1570
		1571	static void eio_execute (etp_worker *self, eio_req *req)
		1572	{
		1573	errno = 0;
		1574
		1575	switch (req->type)
		1576	{
		1577	case EIO_READ: ALLOC (req->size);
		1578	req->result = req->offs >= 0
		1579	? pread (req->int1, req->ptr2, req->size, req->offs)
		1580	: read (req->int1, req->ptr2, req->size); break;
		1581	case EIO_WRITE: req->result = req->offs >= 0
		1582	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1583	: write (req->int1, req->ptr2, req->size); break;
		1584
		1585	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		1586	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
		1587
		1588	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1589	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1590	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1591	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1592	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1593	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1594
		1595	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1596	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1597	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1598	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1599
		1600	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1601	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1602	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1603	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1604	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1605	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1606
		1607	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1608	case EIO_CLOSE: req->result = close (req->int1); break;
		1609	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1610	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1611	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1612	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1613	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1614	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1615	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1616	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
		1617
		1618	case EIO_READLINK: ALLOC (PATH_MAX);
		1619	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
		1620
		1621	case EIO_SYNC: req->result = 0; sync (); break;
		1622	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1623	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1624	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1625	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1626	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1627
		1628	case EIO_READDIR: eio__scandir (req, self); break;
		1629
		1630	case EIO_BUSY:
		1631	#ifdef _WIN32
		1632	Sleep (req->nv1 * 1000.);
		1633	#else
		1634	{
		1635	struct timeval tv;
		1636
		1637	tv.tv_sec = req->nv1;
		1638	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
		1639
		1640	req->result = select (0, 0, 0, 0, &tv);
		1641	}
		1642	#endif
		1643	break;
		1644
		1645	case EIO_UTIME:
		1646	case EIO_FUTIME:
		1647	{
		1648	struct timeval tv[2];
		1649	struct timeval *times;
		1650
		1651	if (req->nv1 != -1. || req->nv2 != -1.)
		1652	{
		1653	tv[0].tv_sec = req->nv1;
		1654	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		1655	tv[1].tv_sec = req->nv2;
		1656	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		1657
		1658	times = tv;
		1659	}
		1660	else
		1661	times = 0;
		1662
		1663
		1664	req->result = req->type == EIO_FUTIME
		1665	? futimes (req->int1, times)
		1666	: utimes (req->ptr1, times);
		1667	}
		1668	break;
		1669
		1670	case EIO_GROUP:
		1671	abort (); /* handled in eio_request */
		1672
		1673	case EIO_NOP:
		1674	req->result = 0;
		1675	break;
		1676
		1677	case EIO_CUSTOM:
		1678	((void (*)(eio_req *))req->feed) (req);
		1679	break;
		1680
		1681	default:
		1682	req->result = -1;
		1683	break;
		1684	}
		1685
		1686	req->errorno = errno;
		1687	}
		1688
1090	#ifndef EIO_NO_WRAPPERS	1689	#ifndef EIO_NO_WRAPPERS
1091		1690
1092	eio_req *eio_nop (int pri, eio_cb cb, void *data)	1691	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1093	{	1692	{
1094	REQ (EIO_NOP); SEND;	1693	REQ (EIO_NOP); SEND;
…		…
1107	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	1706	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1108	{	1707	{
1109	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1708	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1110	}	1709	}
1111		1710
		1711	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1712	{
		1713	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1714	}
		1715
		1716	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1717	{
		1718	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1719	}
		1720
		1721	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		1722	{
		1723	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		1724	}
		1725
1112	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	1726	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1113	{	1727	{
1114	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1728	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1115	}	1729	}
1116		1730
…		…
1135	}	1749	}
1136		1750
1137	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	1751	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1138	{	1752	{
1139	REQ (EIO_FSTAT); req->int1 = fd; SEND;	1753	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1754	}
		1755
		1756	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		1757	{
		1758	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1140	}	1759	}
1141		1760
1142	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	1761	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1143	{	1762	{
1144	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	1763	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
…		…
1218	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	1837	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1219	{	1838	{
1220	return eio__1path (EIO_LSTAT, path, pri, cb, data);	1839	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1221	}	1840	}
1222		1841
		1842	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		1843	{
		1844	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		1845	}
		1846
1223	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	1847	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1224	{	1848	{
1225	return eio__1path (EIO_UNLINK, path, pri, cb, data);	1849	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1226	}	1850	}
1227		1851
1228	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	1852	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1229	{	1853	{
1230	return eio__1path (EIO_RMDIR, path, pri, cb, data);	1854	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1231	}	1855	}
1232		1856
1233	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	1857	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1234	{	1858	{
1235	return eio__1path (EIO_READDIR, path, pri, cb, data);	1859	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1236	}	1860	}
1237		1861
1238	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	1862	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1239	{	1863	{
1240	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	1864	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1241	}	1865	}
1242		1866
1243	static eio_req *	1867	static eio_req *
1244	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1868	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1245	{	1869	{
…		…
1269	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1893	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1270	{	1894	{
1271	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	1895	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1272	}	1896	}
1273		1897
		1898	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
		1899	{
		1900	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
		1901	}
		1902
1274	#endif	1903	#endif
1275		1904
1276	eio_req *eio_grp (eio_cb cb, void *data)	1905	eio_req *eio_grp (eio_cb cb, void *data)
1277	{	1906	{
1278	const int pri = EIO_PRI_MAX;	1907	const int pri = EIO_PRI_MAX;
…		…
1304		1933
1305	void eio_grp_add (eio_req *grp, eio_req *req)	1934	void eio_grp_add (eio_req *grp, eio_req *req)
1306	{	1935	{
1307	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	1936	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1308		1937
		1938	grp->flags |= EIO_FLAG_GROUPADD;
		1939
1309	++grp->size;	1940	++grp->size;
1310	req->grp = grp;	1941	req->grp = grp;
1311		1942
1312	req->grp_prev = 0;	1943	req->grp_prev = 0;
1313	req->grp_next = grp->grp_first;	1944	req->grp_next = grp->grp_first;
…		…
1321	/*****************************************************************************/	1952	/*****************************************************************************/
1322	/* misc garbage */	1953	/* misc garbage */
1323		1954
1324	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	1955	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1325	{	1956	{
1326	worker wrk;	1957	etp_worker wrk;
		1958	ssize_t ret;
1327		1959
1328	wrk.dbuf = 0;	1960	wrk.dbuf = 0;
1329		1961
1330	eio__sendfile (ofd, ifd, offset, count, &wrk);	1962	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1331		1963
1332	if (wrk.dbuf)	1964	if (wrk.dbuf)
1333	free (wrk.dbuf);	1965	free (wrk.dbuf);
1334	}
1335		1966
		1967	return ret;
		1968	}
		1969

Diff Legend

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