[ViewVC] Diff of: cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.10 by root, Tue May 13 17:08:15 2008 UTC vs.
Revision 1.46 by root, Sat Jan 2 12:58:37 2010 UTC

…		…
		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008,2009 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 (wrklock); \	146	ETP_WORKER_LOCK (self); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	147	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (wrklock); \	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
		155	/*****************************************************************************/
		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
		201	/* calculate time difference in ~1/EIO_TICKS of a second */
		202	static int tvdiff (struct timeval tv1, struct timeval tv2)
		203	{
		204	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
		205	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
		206	}
		207
		208	static unsigned int started, idle, wanted = 4;
		209
84	static void (*want_poll_cb) (void);	210	static void (*want_poll_cb) (void);
85	static void (*done_poll_cb) (void);	211	static void (*done_poll_cb) (void);
86		212
87	static unsigned int max_poll_time = 0;	213	static unsigned int max_poll_time; /* reslock */
88	static unsigned int max_poll_reqs = 0;	214	static unsigned int max_poll_reqs; /* reslock */
89		215
90	/* calculcate time difference in ~1/EIO_TICKS of a second */	216	static volatile unsigned int nreqs; /* reqlock */
91	static int tvdiff (struct timeval tv1, struct timeval tv2)	217	static volatile unsigned int nready; /* reqlock */
92	{	218	static volatile unsigned int npending; /* reqlock */
93	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	219	static volatile unsigned int max_idle = 4;
94	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
95	}
96		220
97	static unsigned int started, idle, wanted = 4;
98
99	/* worker threads management */
100	static mutex_t wrklock = X_MUTEX_INIT;	221	static mutex_t wrklock = X_MUTEX_INIT;
101
102	typedef struct worker
103	{
104	/* locked by wrklock */
105	struct worker prev, next;
106
107	thread_t tid;
108
109	/* locked by reslock, reqlock or wrklock */
110	eio_req req; / currently processed request */
111	void *dbuf;
112	DIR *dirp;
113	} worker;
114
115	static worker wrk_first = { &wrk_first, &wrk_first, 0 };
116
117	static void worker_clear (worker *wrk)
118	{
119	if (wrk->dirp)
120	{
121	closedir (wrk->dirp);
122	wrk->dirp = 0;
123	}
124
125	if (wrk->dbuf)
126	{
127	free (wrk->dbuf);
128	wrk->dbuf = 0;
129	}
130	}
131
132	static void worker_free (worker *wrk)
133	{
134	wrk->next->prev = wrk->prev;
135	wrk->prev->next = wrk->next;
136
137	free (wrk);
138	}
139
140	static volatile unsigned int nreqs, nready, npending;
141	static volatile unsigned int max_idle = 4;
142
143	static mutex_t reslock = X_MUTEX_INIT;	222	static mutex_t reslock = X_MUTEX_INIT;
144	static mutex_t reqlock = X_MUTEX_INIT;	223	static mutex_t reqlock = X_MUTEX_INIT;
145	static cond_t reqwait = X_COND_INIT;	224	static cond_t reqwait = X_COND_INIT;
146		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
		235	typedef struct etp_worker
		236	{
		237	/* locked by wrklock */
		238	struct etp_worker prev, next;
		239
		240	thread_t tid;
		241
		242	/* locked by reslock, reqlock or wrklock */
		243	ETP_REQ req; / currently processed request */
		244
		245	ETP_WORKER_COMMON
		246	} etp_worker;
		247
		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)
		252
		253	/* worker threads management */
		254
		255	static void etp_worker_clear (etp_worker *wrk)
		256	{
		257	ETP_WORKER_CLEAR (wrk);
		258	}
		259
		260	static void etp_worker_free (etp_worker *wrk)
		261	{
		262	wrk->next->prev = wrk->prev;
		263	wrk->prev->next = wrk->next;
		264
		265	free (wrk);
		266	}
		267
147	unsigned int eio_nreqs (void)	268	static unsigned int etp_nreqs (void)
148	{	269	{
		270	int retval;
		271	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		272	retval = nreqs;
		273	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
149	return nreqs;	274	return retval;
150	}	275	}
151		276
152	unsigned int eio_nready (void)	277	static unsigned int etp_nready (void)
153	{	278	{
154	unsigned int retval;	279	unsigned int retval;
155		280
156	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	281	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157	retval = nready;	282	retval = nready;
158	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	283	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159		284
160	return retval;	285	return retval;
161	}	286	}
162		287
163	unsigned int eio_npending (void)	288	static unsigned int etp_npending (void)
164	{	289	{
165	unsigned int retval;	290	unsigned int retval;
166		291
167	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	292	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168	retval = npending;	293	retval = npending;
169	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	294	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170		295
171	return retval;	296	return retval;
172	}	297	}
173		298
174	unsigned int eio_nthreads (void)	299	static unsigned int etp_nthreads (void)
175	{	300	{
176	unsigned int retval;	301	unsigned int retval;
177		302
178	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	303	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179	retval = started;	304	retval = started;
…		…
186	* a somewhat faster data structure might be nice, but	311	* a somewhat faster data structure might be nice, but
187	* with 8 priorities this actually needs <20 insns	312	* with 8 priorities this actually needs <20 insns
188	* per shift, the most expensive operation.	313	* per shift, the most expensive operation.
189	*/	314	*/
190	typedef struct {	315	typedef struct {
191	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 */
192	int size;	317	int size;
193	} reqq;	318	} etp_reqq;
194		319
195	static reqq req_queue;	320	static etp_reqq req_queue;
196	static reqq res_queue;	321	static etp_reqq res_queue;
197		322
198	static int reqq_push (reqq q, eio_req req)	323	static int reqq_push (etp_reqq q, ETP_REQ req)
199	{	324	{
200	int pri = req->pri;	325	int pri = req->pri;
201	req->next = 0;	326	req->next = 0;
202		327
203	if (q->qe[pri])	328	if (q->qe[pri])
…		…
209	q->qe[pri] = q->qs[pri] = req;	334	q->qe[pri] = q->qs[pri] = req;
210		335
211	return q->size++;	336	return q->size++;
212	}	337	}
213		338
214	static eio_req reqq_shift (reqq q)	339	static ETP_REQ reqq_shift (etp_reqq q)
215	{	340	{
216	int pri;	341	int pri;
217		342
218	if (!q->size)	343	if (!q->size)
219	return 0;	344	return 0;
220		345
221	--q->size;	346	--q->size;
222		347
223	for (pri = EIO_NUM_PRI; pri--; )	348	for (pri = ETP_NUM_PRI; pri--; )
224	{	349	{
225	eio_req *req = q->qs[pri];	350	eio_req *req = q->qs[pri];
226		351
227	if (req)	352	if (req)
228	{	353	{
…		…
234	}	359	}
235		360
236	abort ();	361	abort ();
237	}	362	}
238		363
239	static void grp_try_feed (eio_req *grp)	364	static void etp_atfork_prepare (void)
240	{	365	{
241	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	366	X_LOCK (wrklock);
		367	X_LOCK (reqlock);
		368	X_LOCK (reslock);
		369	#if !HAVE_PREADWRITE
		370	X_LOCK (preadwritelock);
		371	#endif
		372	}
		373
		374	static void etp_atfork_parent (void)
		375	{
		376	#if !HAVE_PREADWRITE
		377	X_UNLOCK (preadwritelock);
		378	#endif
		379	X_UNLOCK (reslock);
		380	X_UNLOCK (reqlock);
		381	X_UNLOCK (wrklock);
		382	}
		383
		384	static void etp_atfork_child (void)
		385	{
		386	ETP_REQ *prv;
		387
		388	while ((prv = reqq_shift (&req_queue)))
		389	ETP_DESTROY (prv);
		390
		391	while ((prv = reqq_shift (&res_queue)))
		392	ETP_DESTROY (prv);
		393
		394	while (wrk_first.next != &wrk_first)
242	{	395	{
243	int old_len = grp->size;	396	etp_worker *wrk = wrk_first.next;
244		397
245	EIO_FEED (grp);	398	if (wrk->req)
		399	ETP_DESTROY (wrk->req);
246		400
247	/* stop if no progress has been made */	401	etp_worker_clear (wrk);
248	if (old_len == grp->size)	402	etp_worker_free (wrk);
249	{
250	grp->feed = 0;
251	break;
252	}
253	}	403	}
254	}
255		404
256	static int eio_finish (eio_req *req);	405	started = 0;
		406	idle = 0;
		407	nreqs = 0;
		408	nready = 0;
		409	npending = 0;
257		410
258	static int grp_dec (eio_req *grp)	411	etp_atfork_parent ();
259	{	412	}
260	--grp->size;
261		413
262	/* call feeder, if applicable */	414	static void
263	grp_try_feed (grp);	415	etp_once_init (void)
		416	{
		417	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
		418	}
264		419
265	/* finish, if done */	420	static int
266	if (!grp->size && grp->int1)	421	etp_init (void (want_poll)(void), void (done_poll)(void))
267	return eio_finish (grp);	422	{
268	else	423	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
		424
		425	pthread_once (&doinit, etp_once_init);
		426
		427	want_poll_cb = want_poll;
		428	done_poll_cb = done_poll;
		429
269	return 0;	430	return 0;
270	}	431	}
271		432
272	void eio_destroy (eio_req *req)
273	{
274	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
275	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
276
277	EIO_DESTROY (req);
278	}
279
280	static int eio_finish (eio_req *req)
281	{
282	int res = EIO_FINISH (req);
283
284	if (req->grp)
285	{
286	int res2;
287	eio_req *grp = req->grp;
288
289	/* unlink request */
290	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
291	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
292
293	if (grp->grp_first == req)
294	grp->grp_first = req->grp_next;
295
296	res2 = grp_dec (grp);
297
298	if (!res && res2)
299	res = res2;
300	}
301
302	eio_destroy (req);
303
304	return res;
305	}
306
307	void eio_grp_cancel (eio_req *grp)
308	{
309	for (grp = grp->grp_first; grp; grp = grp->grp_next)
310	eio_cancel (grp);
311	}
312
313	void eio_cancel (eio_req *req)
314	{
315	X_LOCK (wrklock);
316	req->flags \|= EIO_FLAG_CANCELLED;
317	X_UNLOCK (wrklock);
318
319	eio_grp_cancel (req);
320	}
321
322	X_THREAD_PROC (eio_proc);	433	X_THREAD_PROC (etp_proc);
323		434
324	static void start_thread (void)	435	static void etp_start_thread (void)
325	{	436	{
326	worker *wrk = calloc (1, sizeof (worker));	437	etp_worker *wrk = calloc (1, sizeof (etp_worker));
327		438
328	/TODO/	439	/TODO/
329	assert (("unable to allocate worker thread data", wrk));	440	assert (("unable to allocate worker thread data", wrk));
330		441
331	X_LOCK (wrklock);	442	X_LOCK (wrklock);
332		443
333	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	444	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334	{	445	{
335	wrk->prev = &wrk_first;	446	wrk->prev = &wrk_first;
336	wrk->next = wrk_first.next;	447	wrk->next = wrk_first.next;
337	wrk_first.next->prev = wrk;	448	wrk_first.next->prev = wrk;
338	wrk_first.next = wrk;	449	wrk_first.next = wrk;
…		…
342	free (wrk);	453	free (wrk);
343		454
344	X_UNLOCK (wrklock);	455	X_UNLOCK (wrklock);
345	}	456	}
346		457
347	static void maybe_start_thread (void)	458	static void etp_maybe_start_thread (void)
348	{	459	{
349	if (eio_nthreads () >= wanted)	460	if (expect_true (etp_nthreads () >= wanted))
350	return;	461	return;
351		462
352	/* todo: maybe use idle here, but might be less exact */	463	/* todo: maybe use idle here, but might be less exact */
353	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 ()))
354	return;	465	return;
355		466
356	start_thread ();	467	etp_start_thread ();
357	}	468	}
358		469
359	void eio_submit (eio_req *req)
360	{
361	++nreqs;
362
363	X_LOCK (reqlock);
364	++nready;
365	reqq_push (&req_queue, req);
366	X_COND_SIGNAL (reqwait);
367	X_UNLOCK (reqlock);
368
369	maybe_start_thread ();
370	}
371
372	static void end_thread (void)	470	static void etp_end_thread (void)
373	{	471	{
374	eio_req *req = calloc (1, sizeof (eio_req));	472	eio_req *req = calloc (1, sizeof (eio_req));
375		473
376	req->type = EIO_QUIT;	474	req->type = -1;
377	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	475	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
378		476
379	X_LOCK (reqlock);	477	X_LOCK (reqlock);
380	reqq_push (&req_queue, req);	478	reqq_push (&req_queue, req);
381	X_COND_SIGNAL (reqwait);	479	X_COND_SIGNAL (reqwait);
382	X_UNLOCK (reqlock);	480	X_UNLOCK (reqlock);
…		…
384	X_LOCK (wrklock);	482	X_LOCK (wrklock);
385	--started;	483	--started;
386	X_UNLOCK (wrklock);	484	X_UNLOCK (wrklock);
387	}	485	}
388		486
389	void eio_set_max_poll_time (double nseconds)
390	{
391	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
392	max_poll_time = nseconds;
393	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
394	}
395
396	void eio_set_max_poll_reqs (unsigned int maxreqs)
397	{
398	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
399	max_poll_reqs = maxreqs;
400	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
401	}
402
403	void eio_set_max_idle (unsigned int nthreads)
404	{
405	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
406	max_idle = nthreads <= 0 ? 1 : nthreads;
407	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
408	}
409
410	void eio_set_min_parallel (unsigned int nthreads)
411	{
412	if (wanted < nthreads)
413	wanted = nthreads;
414	}
415
416	void eio_set_max_parallel (unsigned int nthreads)
417	{
418	if (wanted > nthreads)
419	wanted = nthreads;
420
421	while (started > wanted)
422	end_thread ();
423	}
424
425	int eio_poll (void)	487	static int etp_poll (void)
426	{	488	{
427	int maxreqs = max_poll_reqs;	489	unsigned int maxreqs;
		490	unsigned int maxtime;
428	struct timeval tv_start, tv_now;	491	struct timeval tv_start, tv_now;
429	eio_req *req;
430		492
		493	X_LOCK (reslock);
		494	maxreqs = max_poll_reqs;
		495	maxtime = max_poll_time;
		496	X_UNLOCK (reslock);
		497
431	if (max_poll_time)	498	if (maxtime)
432	gettimeofday (&tv_start, 0);	499	gettimeofday (&tv_start, 0);
433		500
434	for (;;)	501	for (;;)
435	{	502	{
		503	ETP_REQ *req;
		504
436	maybe_start_thread ();	505	etp_maybe_start_thread ();
437		506
438	X_LOCK (reslock);	507	X_LOCK (reslock);
439	req = reqq_shift (&res_queue);	508	req = reqq_shift (&res_queue);
440		509
441	if (req)	510	if (req)
…		…
449	X_UNLOCK (reslock);	518	X_UNLOCK (reslock);
450		519
451	if (!req)	520	if (!req)
452	return 0;	521	return 0;
453		522
		523	X_LOCK (reqlock);
454	--nreqs;	524	--nreqs;
		525	X_UNLOCK (reqlock);
455		526
456	if (req->type == EIO_GROUP && req->size)	527	if (expect_false (req->type == EIO_GROUP && req->size))
457	{	528	{
458	req->int1 = 1; /* mark request as delayed */	529	req->int1 = 1; /* mark request as delayed */
459	continue;	530	continue;
460	}	531	}
461	else	532	else
462	{	533	{
463	int res = eio_finish (req);	534	int res = ETP_FINISH (req);
464	if (res)	535	if (expect_false (res))
465	return res;	536	return res;
466	}	537	}
467		538
468	if (maxreqs && !--maxreqs)	539	if (expect_false (maxreqs && !--maxreqs))
469	break;	540	break;
470		541
471	if (max_poll_time)	542	if (maxtime)
472	{	543	{
473	gettimeofday (&tv_now, 0);	544	gettimeofday (&tv_now, 0);
474		545
475	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	546	if (tvdiff (&tv_start, &tv_now) >= maxtime)
476	break;	547	break;
477	}	548	}
478	}	549	}
479		550
480	errno = EAGAIN;	551	errno = EAGAIN;
481	return -1;	552	return -1;
482	}	553	}
483		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
484	/*****************************************************************************/	770	/*****************************************************************************/
485	/* work around various missing functions */	771	/* work around various missing functions */
486		772
487	#if !HAVE_PREADWRITE	773	#if !HAVE_PREADWRITE
		774	# undef pread
		775	# undef pwrite
488	# define pread eio__pread	776	# define pread eio__pread
489	# define pwrite eio__pwrite	777	# define pwrite eio__pwrite
490
491	/*
492	* make our pread/pwrite safe against themselves, but not against
493	* normal read/write by using a mutex. slows down execution a lot,
494	* but that's your problem, not mine.
495	*/
496	static mutex_t preadwritelock = X_MUTEX_INIT;
497		778
498	static ssize_t	779	static ssize_t
499	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)
500	{	781	{
501	ssize_t res;	782	ssize_t res;
…		…
519		800
520	X_LOCK (preadwritelock);	801	X_LOCK (preadwritelock);
521	ooffset = lseek (fd, 0, SEEK_CUR);	802	ooffset = lseek (fd, 0, SEEK_CUR);
522	lseek (fd, offset, SEEK_SET);	803	lseek (fd, offset, SEEK_SET);
523	res = write (fd, buf, count);	804	res = write (fd, buf, count);
524	lseek (fd, offset, SEEK_SET);	805	lseek (fd, ooffset, SEEK_SET);
525	X_UNLOCK (preadwritelock);	806	X_UNLOCK (preadwritelock);
526		807
527	return res;	808	return res;
528	}	809	}
529	#endif	810	#endif
530		811
531	#ifndef HAVE_FUTIMES	812	#ifndef HAVE_FUTIMES
532		813
		814	# undef utimes
		815	# undef futimes
533	# define utimes(path,times) eio__utimes (path, times)	816	# define utimes(path,times) eio__utimes (path, times)
534	# define futimes(fd,times) eio__futimes (fd, times)	817	# define futimes(fd,times) eio__futimes (fd, times)
535		818
536	static int	819	static int
537	eio__utimes (const char *filename, const struct timeval times[2])	820	eio__utimes (const char *filename, const struct timeval times[2])
…		…
556	}	839	}
557		840
558	#endif	841	#endif
559		842
560	#if !HAVE_FDATASYNC	843	#if !HAVE_FDATASYNC
		844	# undef fdatasync
561	# define fdatasync fsync	845	# define fdatasync(fd) fsync (fd)
562	#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	}
563		875
564	#if !HAVE_READAHEAD	876	#if !HAVE_READAHEAD
		877	# undef readahead
565	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	878	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566		879
567	static ssize_t	880	static ssize_t
568	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)
569	{	882	{
570	size_t todo = count;	883	size_t todo = count;
571	dBUF;	884	dBUF;
572		885
573	while (todo > 0)	886	while (todo > 0)
…		…
583	return count;	896	return count;
584	}	897	}
585		898
586	#endif	899	#endif
587		900
588	#if !HAVE_READDIR_R
589	# define readdir_r eio__readdir_r
590
591	static mutex_t readdirlock = X_MUTEX_INIT;
592
593	static int
594	eio__readdir_r (DIR dirp, EIO_STRUCT_DIRENT ent, EIO_STRUCT_DIRENT **res)
595	{
596	EIO_STRUCT_DIRENT *e;
597	int errorno;
598
599	X_LOCK (readdirlock);
600
601	e = readdir (dirp);
602	errorno = errno;
603
604	if (e)
605	{
606	*res = ent;
607	strcpy (ent->d_name, e->d_name);
608	}
609	else
610	*res = 0;
611
612	X_UNLOCK (readdirlock);
613
614	errno = errorno;
615	return e ? 0 : -1;
616	}
617	#endif
618
619	/* sendfile always needs emulation */	901	/* sendfile always needs emulation */
620	static ssize_t	902	static ssize_t
621	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)
622	{	904	{
623	ssize_t res;	905	ssize_t res;
624		906
625	if (!count)	907	if (!count)
626	return 0;	908	return 0;
…		…
641	if (res < 0 && sbytes)	923	if (res < 0 && sbytes)
642	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	924	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */
643	res = sbytes;	925	res = sbytes;
644	}	926	}
645		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
646	# elif __hpux	938	# elif __hpux
647	res = sendfile (ofd, ifd, offset, count, 0, 0);	939	res = sendfile (ofd, ifd, offset, count, 0, 0);
648		940
649	# elif __solaris	941	# elif __solaris
650	{	942	{
…		…
661	if (res < 0 && sbytes)	953	if (res < 0 && sbytes)
662	res = sbytes;	954	res = sbytes;
663	}	955	}
664		956
665	# 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
666	#else	968	#else
667	res = -1;	969	res = -1;
668	errno = ENOSYS;	970	errno = ENOSYS;
669	#endif	971	#endif
670		972
…		…
708	}	1010	}
709		1011
710	return res;	1012	return res;
711	}	1013	}
712		1014
		1015	static signed char
		1016	eio_dent_cmp (const eio_dirent a, const eio_dirent b)
		1017	{
		1018	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1019	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
		1020	}
		1021
		1022	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1023
		1024	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1025	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1026
		1027	static void
		1028	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1029	{
		1030	unsigned char bits [9 + sizeof (ino_t) * 8];
		1031	unsigned char *bit = bits;
		1032
		1033	assert (CHAR_BIT == 8);
		1034	assert (sizeof (eio_dirent) * 8 < 256);
		1035	assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
		1036	assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
		1037
		1038	if (size <= EIO_SORT_FAST)
		1039	return;
		1040
		1041	/* first prepare an array of bits to test in our radix sort */
		1042	/* try to take endianness into account, as well as differences in ino_t sizes */
		1043	/* inode_bits must contain all inodes ORed together */
		1044	/* which is used to skip bits that are 0 everywhere, which is very common */
		1045	{
		1046	ino_t endianness;
		1047	int i, j;
		1048
		1049	/* we store the byte offset of byte n into byte n of "endianness" */
		1050	for (i = 0; i < sizeof (ino_t); ++i)
		1051	((unsigned char *)&endianness)[i] = i;
		1052
		1053	*bit++ = 0;
		1054
		1055	for (i = 0; i < sizeof (ino_t); ++i)
		1056	{
		1057	/* shifting off the byte offsets out of "endianness" */
		1058	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1059	endianness >>= 8;
		1060
		1061	for (j = 0; j < 8; ++j)
		1062	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1063	*bit++ = offs + j;
		1064	}
		1065
		1066	for (j = 0; j < 8; ++j)
		1067	if (score_bits & (1 << j))
		1068	bit++ = offsetof (eio_dirent, score) 8 + j;
		1069	}
		1070
		1071	/* now actually do the sorting (a variant of MSD radix sort) */
		1072	{
		1073	eio_dirent base_stk [9 + sizeof (ino_t) 8], *base;
		1074	eio_dirent end_stk [9 + sizeof (ino_t) 8], *end;
		1075	unsigned char bit_stk [9 + sizeof (ino_t) 8];
		1076	int stk_idx = 0;
		1077
		1078	base_stk [stk_idx] = dents;
		1079	end_stk [stk_idx] = dents + size;
		1080	bit_stk [stk_idx] = bit - 1;
		1081
		1082	do
		1083	{
		1084	base = base_stk [stk_idx];
		1085	end = end_stk [stk_idx];
		1086	bit = bit_stk [stk_idx];
		1087
		1088	for (;;)
		1089	{
		1090	unsigned char O = *bit >> 3;
		1091	unsigned char M = 1 << (*bit & 7);
		1092
		1093	eio_dirent *a = base;
		1094	eio_dirent *b = end;
		1095
		1096	if (b - a < EIO_SORT_CUTOFF)
		1097	break;
		1098
		1099	/* now bit-partition the array on the bit */
		1100	/* this ugly asymmetric loop seems to perform much better than typical */
		1101	/* partition algos found in the literature */
		1102	do
		1103	if (!(((unsigned char *)a)[O] & M))
		1104	++a;
		1105	else if (!(((unsigned char *)--b)[O] & M))
		1106	{
		1107	eio_dirent tmp = a; a = b; b = tmp;
		1108	++a;
		1109	}
		1110	while (b > a);
		1111
		1112	/* next bit, or stop, if no bits left in this path */
		1113	if (!*--bit)
		1114	break;
		1115
		1116	base_stk [stk_idx] = a;
		1117	end_stk [stk_idx] = end;
		1118	bit_stk [stk_idx] = bit;
		1119	++stk_idx;
		1120
		1121	end = a;
		1122	}
		1123	}
		1124	while (stk_idx--);
		1125	}
		1126	}
		1127
		1128	static void
		1129	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1130	{
		1131	/* first move the smallest element to the front, to act as a sentinel */
		1132	{
		1133	int i;
		1134	eio_dirent *min = dents;
		1135
		1136	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1137	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1138	if (EIO_DENT_CMP (dents [i], <, *min))
		1139	min = &dents [i];
		1140
		1141	/* swap elements 0 and j (minimum) */
		1142	{
		1143	eio_dirent tmp = dents; dents = min; min = tmp;
		1144	}
		1145	}
		1146
		1147	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1148	{
		1149	eio_dirent i, j;
		1150
		1151	for (i = dents + 1; i < dents + size; ++i)
		1152	{
		1153	eio_dirent value = *i;
		1154
		1155	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1156	j [1] = j [0];
		1157
		1158	j [1] = value;
		1159	}
		1160	}
		1161	}
		1162
		1163	static void
		1164	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1165	{
		1166	if (size <= 1)
		1167	return; /* our insertion sort relies on size > 0 */
		1168
		1169	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1170	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1171	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1172
		1173	/* use an insertion sort at the end, or for small arrays, */
		1174	/* as insertion sort is more efficient for small partitions */
		1175	eio_dent_insertion_sort (dents, size);
		1176	}
		1177
713	/* read a full directory */	1178	/* read a full directory */
714	static void	1179	static void
715	eio__scandir (eio_req req, worker self)	1180	eio__scandir (eio_req req, etp_worker self)
716	{	1181	{
717	DIR *dirp;	1182	DIR *dirp;
718	union
719	{
720	EIO_STRUCT_DIRENT d;
721	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
722	} *u;
723	EIO_STRUCT_DIRENT *entp;	1183	EIO_STRUCT_DIRENT *entp;
724	char name, names;	1184	char name, names;
725	int memlen = 4096;	1185	int namesalloc = 4096;
726	int memofs = 0;	1186	int namesoffs = 0;
		1187	int flags = req->int1;
		1188	eio_dirent *dents = 0;
		1189	int dentalloc = 128;
727	int res = 0;	1190	int dentoffs = 0;
		1191	ino_t inode_bits = 0;
		1192
		1193	req->result = -1;
		1194
		1195	if (!(flags & EIO_READDIR_DENTS))
		1196	flags &= ~(EIO_READDIR_DIRS_FIRST \| EIO_READDIR_STAT_ORDER);
728		1197
729	X_LOCK (wrklock);	1198	X_LOCK (wrklock);
		1199	/* the corresponding closedir is in ETP_WORKER_CLEAR */
730	self->dirp = dirp = opendir (req->ptr1);	1200	self->dirp = dirp = opendir (req->ptr1);
731	self->dbuf = u = malloc (sizeof (*u));
732	req->flags \|= EIO_FLAG_PTR2_FREE;	1201	req->flags \|= EIO_FLAG_PTR1_FREE \| EIO_FLAG_PTR2_FREE;
		1202	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
733	req->ptr2 = names = malloc (memlen);	1203	req->ptr2 = names = malloc (namesalloc);
734	X_UNLOCK (wrklock);	1204	X_UNLOCK (wrklock);
735		1205
736	if (dirp && u && names)	1206	if (dirp && names && (!flags \|\| dents))
737	for (;;)	1207	for (;;)
738	{	1208	{
739	errno = 0;	1209	errno = 0;
740	readdir_r (dirp, &u->d, &entp);	1210	entp = readdir (dirp);
741		1211
742	if (!entp)	1212	if (!entp)
		1213	{
		1214	if (errno)
		1215	break;
		1216
		1217	/* sort etc. */
		1218	req->int1 = flags;
		1219	req->result = dentoffs;
		1220
		1221	if (flags & EIO_READDIR_STAT_ORDER)
		1222	eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
		1223	else if (flags & EIO_READDIR_DIRS_FIRST)
		1224	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1225	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1226	else
		1227	{
		1228	/* in this case, all is known, and we just put dirs first and sort them */
		1229	eio_dirent *oth = dents + dentoffs;
		1230	eio_dirent *dir = dents;
		1231
		1232	/* now partition dirs to the front, and non-dirs to the back */
		1233	/* by walking from both sides and swapping if necessary */
		1234	/* also clear score, so it doesn't influence sorting */
		1235	while (oth > dir)
		1236	{
		1237	if (dir->type == EIO_DT_DIR)
		1238	++dir;
		1239	else if ((--oth)->type == EIO_DT_DIR)
		1240	{
		1241	eio_dirent tmp = dir; dir = oth; oth = tmp;
		1242
		1243	++dir;
		1244	}
		1245	}
		1246
		1247	/* now sort the dirs only */
		1248	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1249	}
		1250
743	break;	1251	break;
		1252	}
744		1253
		1254	/* now add the entry to our list(s) */
745	name = entp->d_name;	1255	name = entp->d_name;
746		1256
		1257	/* skip . and .. entries */
747	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))	1258	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))
748	{	1259	{
749	int len = strlen (name) + 1;	1260	int len = D_NAMLEN (entp) + 1;
750		1261
751	res++;	1262	while (expect_false (namesoffs + len > namesalloc))
752
753	while (memofs + len > memlen)
754	{	1263	{
755	memlen *= 2;	1264	namesalloc *= 2;
756	X_LOCK (wrklock);	1265	X_LOCK (wrklock);
757	req->ptr2 = names = realloc (names, memlen);	1266	req->ptr2 = names = realloc (names, namesalloc);
758	X_UNLOCK (wrklock);	1267	X_UNLOCK (wrklock);
759		1268
760	if (!names)	1269	if (!names)
761	break;	1270	break;
762	}	1271	}
763		1272
764	memcpy (names + memofs, name, len);	1273	memcpy (names + namesoffs, name, len);
		1274
		1275	if (dents)
		1276	{
		1277	struct eio_dirent *ent;
		1278
		1279	if (expect_false (dentoffs == dentalloc))
		1280	{
		1281	dentalloc *= 2;
		1282	X_LOCK (wrklock);
		1283	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1284	X_UNLOCK (wrklock);
		1285
		1286	if (!dents)
		1287	break;
		1288	}
		1289
		1290	ent = dents + dentoffs;
		1291
		1292	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1293	ent->namelen = len - 1;
		1294	ent->inode = D_INO (entp);
		1295
		1296	inode_bits \|= ent->inode;
		1297
		1298	switch (D_TYPE (entp))
		1299	{
		1300	default:
		1301	ent->type = EIO_DT_UNKNOWN;
		1302	flags \|= EIO_READDIR_FOUND_UNKNOWN;
		1303	break;
		1304
		1305	#ifdef DT_FIFO
		1306	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1307	#endif
		1308	#ifdef DT_CHR
		1309	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1310	#endif
		1311	#ifdef DT_MPC
		1312	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1313	#endif
		1314	#ifdef DT_DIR
		1315	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1316	#endif
		1317	#ifdef DT_NAM
		1318	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1319	#endif
		1320	#ifdef DT_BLK
		1321	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1322	#endif
		1323	#ifdef DT_MPB
		1324	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1325	#endif
		1326	#ifdef DT_REG
		1327	case DT_REG: ent->type = EIO_DT_REG; break;
		1328	#endif
		1329	#ifdef DT_NWK
		1330	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1331	#endif
		1332	#ifdef DT_CMP
		1333	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1334	#endif
		1335	#ifdef DT_LNK
		1336	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1337	#endif
		1338	#ifdef DT_SOCK
		1339	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1340	#endif
		1341	#ifdef DT_DOOR
		1342	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1343	#endif
		1344	#ifdef DT_WHT
		1345	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1346	#endif
		1347	}
		1348
		1349	ent->score = 7;
		1350
		1351	if (flags & EIO_READDIR_DIRS_FIRST)
		1352	{
		1353	if (ent->type == EIO_DT_UNKNOWN)
		1354	{
		1355	if (name == '.') / leading dots are likely directories, and, in any case, rare */
		1356	ent->score = 1;
		1357	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1358	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1359	}
		1360	else if (ent->type == EIO_DT_DIR)
		1361	ent->score = 0;
		1362	}
		1363	}
		1364
765	memofs += len;	1365	namesoffs += len;
		1366	++dentoffs;
		1367	}
		1368
		1369	if (EIO_CANCELLED (req))
		1370	{
		1371	errno = ECANCELED;
		1372	break;
766	}	1373	}
767	}	1374	}
		1375	}
768		1376
769	if (errno)	1377	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
770	res = -1;	1378	# undef msync
771		1379	# define msync(a,b,c) ((errno = ENOSYS), -1)
772	req->result = res;	1380	#endif
		1381
		1382	int
		1383	eio__mtouch (void *mem, size_t len, int flags)
		1384	{
		1385	intptr_t addr = (intptr_t)mem;
		1386	intptr_t end = addr + len;
		1387	#ifdef PAGESIZE
		1388	const intptr_t page = PAGESIZE;
		1389	#else
		1390	static intptr_t page;
		1391
		1392	if (!page)
		1393	page = sysconf (_SC_PAGESIZE);
		1394	#endif
		1395
		1396	/* round down to start of page, although this is probably useless */
		1397	addr &= ~(page - 1); /* assume page size is always a power of two */
		1398
		1399	if (addr < end)
		1400	if (flags) /* modify */
		1401	do { ((volatile sig_atomic_t )addr) \|= 0; } while ((addr += page) < len);
		1402	else
		1403	do { ((volatile sig_atomic_t )addr) ; } while ((addr += page) < len);
		1404
		1405	return 0;
773	}	1406	}
774		1407
775	/*****************************************************************************/	1408	/*****************************************************************************/
776		1409
777	#define ALLOC(len) \	1410	#define ALLOC(len) \
…		…
787	req->result = -1; \	1420	req->result = -1; \
788	break; \	1421	break; \
789	} \	1422	} \
790	}	1423	}
791		1424
792	X_THREAD_PROC (eio_proc)	1425	X_THREAD_PROC (etp_proc)
793	{	1426	{
794	eio_req *req;	1427	ETP_REQ *req;
795	struct timespec ts;	1428	struct timespec ts;
796	worker self = (worker )thr_arg;	1429	etp_worker self = (etp_worker )thr_arg;
797		1430
798	/* try to distribute timeouts somewhat randomly */	1431	/* try to distribute timeouts somewhat randomly */
799	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1432	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
800		1433
801	for (;;)	1434	for (;;)
802	{	1435	{
803	ts.tv_sec = time (0) + IDLE_TIMEOUT;
804
805	X_LOCK (reqlock);	1436	X_LOCK (reqlock);
806		1437
807	for (;;)	1438	for (;;)
808	{	1439	{
809	self->req = req = reqq_shift (&req_queue);	1440	self->req = req = reqq_shift (&req_queue);
…		…
811	if (req)	1442	if (req)
812	break;	1443	break;
813		1444
814	++idle;	1445	++idle;
815		1446
		1447	ts.tv_sec = time (0) + IDLE_TIMEOUT;
816	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1448	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
817	{	1449	{
818	if (idle > max_idle)	1450	if (idle > max_idle)
819	{	1451	{
820	--idle;	1452	--idle;
…		…
825	goto quit;	1457	goto quit;
826	}	1458	}
827		1459
828	/* we are allowed to idle, so do so without any timeout */	1460	/* we are allowed to idle, so do so without any timeout */
829	X_COND_WAIT (reqwait, reqlock);	1461	X_COND_WAIT (reqwait, reqlock);
830	ts.tv_sec = time (0) + IDLE_TIMEOUT;
831	}	1462	}
832		1463
833	--idle;	1464	--idle;
834	}	1465	}
835		1466
836	--nready;	1467	--nready;
837		1468
838	X_UNLOCK (reqlock);	1469	X_UNLOCK (reqlock);
839		1470
840	errno = 0; /* strictly unnecessary */	1471	if (req->type < 0)
		1472	goto quit;
841		1473
842	if (!EIO_CANCELLED (req))	1474	if (!EIO_CANCELLED (req))
843	switch (req->type)	1475	ETP_EXECUTE (self, req);
844	{
845	case EIO_READ: ALLOC (req->size);
846	req->result = req->offs >= 0
847	? pread (req->int1, req->ptr2, req->size, req->offs)
848	: read (req->int1, req->ptr2, req->size); break;
849	case EIO_WRITE: req->result = req->offs >= 0
850	? pwrite (req->int1, req->ptr2, req->size, req->offs)
851	: write (req->int1, req->ptr2, req->size); break;
852
853	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
854	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
855
856	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
857	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
858	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
859	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
860	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
861	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
862
863	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
864	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
865	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
866	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
867	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
868	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
869
870	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
871	case EIO_CLOSE: req->result = close (req->int1); break;
872	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
873	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
874	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
875	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
876	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
877	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
878	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
879	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
880
881	case EIO_READLINK: ALLOC (NAME_MAX);
882	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
883
884	case EIO_SYNC: req->result = 0; sync (); break;
885	case EIO_FSYNC: req->result = fsync (req->int1); break;
886	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
887
888	case EIO_READDIR: eio__scandir (req, self); break;
889
890	case EIO_BUSY:
891	#ifdef _WIN32
892	Sleep (req->nv1 * 1000.);
893	#else
894	{
895	struct timeval tv;
896
897	tv.tv_sec = req->nv1;
898	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
899
900	req->result = select (0, 0, 0, 0, &tv);
901	}
902	#endif
903	break;
904
905	case EIO_UTIME:
906	case EIO_FUTIME:
907	{
908	struct timeval tv[2];
909	struct timeval *times;
910
911	if (req->nv1 != -1. \|\| req->nv2 != -1.)
912	{
913	tv[0].tv_sec = req->nv1;
914	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
915	tv[1].tv_sec = req->nv2;
916	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
917
918	times = tv;
919	}
920	else
921	times = 0;
922
923
924	req->result = req->type == EIO_FUTIME
925	? futimes (req->int1, times)
926	: utimes (req->ptr1, times);
927	}
928
929	case EIO_GROUP:
930	case EIO_NOP:
931	req->result = 0;
932	break;
933
934	case EIO_QUIT:
935	goto quit;
936
937	default:
938	req->result = -1;
939	break;
940	}
941
942	req->errorno = errno;
943		1476
944	X_LOCK (reslock);	1477	X_LOCK (reslock);
945		1478
946	++npending;	1479	++npending;
947		1480
948	if (!reqq_push (&res_queue, req) && want_poll_cb)	1481	if (!reqq_push (&res_queue, req) && want_poll_cb)
949	want_poll_cb ();	1482	want_poll_cb ();
950		1483
951	self->req = 0;	1484	self->req = 0;
952	worker_clear (self);	1485	etp_worker_clear (self);
953		1486
954	X_UNLOCK (reslock);	1487	X_UNLOCK (reslock);
955	}	1488	}
956		1489
957	quit:	1490	quit:
958	X_LOCK (wrklock);	1491	X_LOCK (wrklock);
959	worker_free (self);	1492	etp_worker_free (self);
960	X_UNLOCK (wrklock);	1493	X_UNLOCK (wrklock);
961		1494
962	return 0;	1495	return 0;
963	}	1496	}
964		1497
965	/*****************************************************************************/	1498	/*****************************************************************************/
966		1499
967	static void eio_atfork_prepare (void)
968	{
969	X_LOCK (wrklock);
970	X_LOCK (reqlock);
971	X_LOCK (reslock);
972	#if !HAVE_PREADWRITE
973	X_LOCK (preadwritelock);
974	#endif
975	#if !HAVE_READDIR_R
976	X_LOCK (readdirlock);
977	#endif
978	}
979
980	static void eio_atfork_parent (void)
981	{
982	#if !HAVE_READDIR_R
983	X_UNLOCK (readdirlock);
984	#endif
985	#if !HAVE_PREADWRITE
986	X_UNLOCK (preadwritelock);
987	#endif
988	X_UNLOCK (reslock);
989	X_UNLOCK (reqlock);
990	X_UNLOCK (wrklock);
991	}
992
993	static void eio_atfork_child (void)
994	{
995	eio_req *prv;
996
997	while (prv = reqq_shift (&req_queue))
998	eio_destroy (prv);
999
1000	while (prv = reqq_shift (&res_queue))
1001	eio_destroy (prv);
1002
1003	while (wrk_first.next != &wrk_first)
1004	{
1005	worker *wrk = wrk_first.next;
1006
1007	if (wrk->req)
1008	eio_destroy (wrk->req);
1009
1010	worker_clear (wrk);
1011	worker_free (wrk);
1012	}
1013
1014	started = 0;
1015	idle = 0;
1016	nreqs = 0;
1017	nready = 0;
1018	npending = 0;
1019
1020	eio_atfork_parent ();
1021	}
1022
1023	int eio_init (void (want_poll)(void), void (done_poll)(void))	1500	int eio_init (void (want_poll)(void), void (done_poll)(void))
1024	{	1501	{
1025	want_poll_cb = want_poll;	1502	return etp_init (want_poll, done_poll);
1026	done_poll_cb = done_poll;
1027
1028	#ifdef _WIN32
1029	X_MUTEX_CHECK (wrklock);
1030	X_MUTEX_CHECK (reslock);
1031	X_MUTEX_CHECK (reqlock);
1032	X_MUTEX_CHECK (reqwait);
1033	X_MUTEX_CHECK (preadwritelock);
1034	X_MUTEX_CHECK (readdirlock);
1035
1036	X_COND_CHECK (reqwait);
1037	#endif
1038
1039	X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1040	}	1503	}
1041		1504
1042	static void eio_api_destroy (eio_req *req)	1505	static void eio_api_destroy (eio_req *req)
1043	{	1506	{
1044	free (req);	1507	free (req);
…		…
1049	\	1512	\
1050	req = (eio_req )calloc (1, sizeof req); \	1513	req = (eio_req )calloc (1, sizeof req); \
1051	if (!req) \	1514	if (!req) \
1052	return 0; \	1515	return 0; \
1053	\	1516	\
1054	req->type = rtype; \	1517	req->type = rtype; \
1055	req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \	1518	req->pri = pri; \
1056	req->finish = cb; \	1519	req->finish = cb; \
		1520	req->data = data; \
1057	req->destroy = eio_api_destroy;	1521	req->destroy = eio_api_destroy;
1058		1522
1059	#define SEND eio_submit (req); return req	1523	#define SEND eio_submit (req); return req
1060		1524
1061	#define PATH \	1525	#define PATH \
…		…
1065	{ \	1529	{ \
1066	eio_api_destroy (req); \	1530	eio_api_destroy (req); \
1067	return 0; \	1531	return 0; \
1068	}	1532	}
1069		1533
		1534	static void eio_execute (etp_worker self, eio_req req)
		1535	{
		1536	errno = 0;
		1537
		1538	switch (req->type)
		1539	{
		1540	case EIO_READ: ALLOC (req->size);
		1541	req->result = req->offs >= 0
		1542	? pread (req->int1, req->ptr2, req->size, req->offs)
		1543	: read (req->int1, req->ptr2, req->size); break;
		1544	case EIO_WRITE: req->result = req->offs >= 0
		1545	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1546	: write (req->int1, req->ptr2, req->size); break;
		1547
		1548	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		1549	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
		1550
		1551	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1552	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1553	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1554	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1555	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1556	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1557
		1558	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1559	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1560	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1561	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1562	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1563	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1564
		1565	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1566	case EIO_CLOSE: req->result = close (req->int1); break;
		1567	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1568	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1569	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1570	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1571	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1572	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1573	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1574	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
		1575
		1576	case EIO_READLINK: ALLOC (NAME_MAX);
		1577	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
		1578
		1579	case EIO_SYNC: req->result = 0; sync (); break;
		1580	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1581	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1582	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
		1583	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1584	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1585
		1586	case EIO_READDIR: eio__scandir (req, self); break;
		1587
		1588	case EIO_BUSY:
		1589	#ifdef _WIN32
		1590	Sleep (req->nv1 * 1000.);
		1591	#else
		1592	{
		1593	struct timeval tv;
		1594
		1595	tv.tv_sec = req->nv1;
		1596	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
		1597
		1598	req->result = select (0, 0, 0, 0, &tv);
		1599	}
		1600	#endif
		1601	break;
		1602
		1603	case EIO_UTIME:
		1604	case EIO_FUTIME:
		1605	{
		1606	struct timeval tv[2];
		1607	struct timeval *times;
		1608
		1609	if (req->nv1 != -1. \|\| req->nv2 != -1.)
		1610	{
		1611	tv[0].tv_sec = req->nv1;
		1612	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		1613	tv[1].tv_sec = req->nv2;
		1614	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		1615
		1616	times = tv;
		1617	}
		1618	else
		1619	times = 0;
		1620
		1621
		1622	req->result = req->type == EIO_FUTIME
		1623	? futimes (req->int1, times)
		1624	: utimes (req->ptr1, times);
		1625	}
		1626	break;
		1627
		1628	case EIO_GROUP:
		1629	abort (); /* handled in eio_request */
		1630
		1631	case EIO_NOP:
		1632	req->result = 0;
		1633	break;
		1634
		1635	case EIO_CUSTOM:
		1636	((void ()(eio_req ))req->feed) (req);
		1637	break;
		1638
		1639	default:
		1640	req->result = -1;
		1641	break;
		1642	}
		1643
		1644	req->errorno = errno;
		1645	}
		1646
		1647	#ifndef EIO_NO_WRAPPERS
		1648
1070	eio_req eio_nop (int pri, eio_cb cb, void data)	1649	eio_req eio_nop (int pri, eio_cb cb, void data)
1071	{	1650	{
1072	REQ (EIO_NOP); SEND;	1651	REQ (EIO_NOP); SEND;
1073	}	1652	}
1074		1653
…		…
1083	}	1662	}
1084		1663
1085	eio_req eio_fsync (int fd, int pri, eio_cb cb, void data)	1664	eio_req eio_fsync (int fd, int pri, eio_cb cb, void data)
1086	{	1665	{
1087	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1666	REQ (EIO_FSYNC); req->int1 = fd; SEND;
		1667	}
		1668
		1669	eio_req eio_msync (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1670	{
		1671	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1672	}
		1673
		1674	eio_req eio_mtouch (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1675	{
		1676	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1677	}
		1678
		1679	eio_req eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void data)
		1680	{
		1681	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1088	}	1682	}
1089		1683
1090	eio_req eio_fdatasync (int fd, int pri, eio_cb cb, void data)	1684	eio_req eio_fdatasync (int fd, int pri, eio_cb cb, void data)
1091	{	1685	{
1092	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1686	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
…		…
1206	eio_req eio_rmdir (const char path, int pri, eio_cb cb, void *data)	1800	eio_req eio_rmdir (const char path, int pri, eio_cb cb, void *data)
1207	{	1801	{
1208	return eio__1path (EIO_RMDIR, path, pri, cb, data);	1802	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1209	}	1803	}
1210		1804
1211	eio_req eio_readdir (const char path, int pri, eio_cb cb, void *data)	1805	eio_req eio_readdir (const char path, int flags, int pri, eio_cb cb, void *data)
1212	{	1806	{
1213	return eio__1path (EIO_READDIR, path, pri, cb, data);	1807	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1214	}	1808	}
1215		1809
1216	eio_req eio_mknod (const char path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	1810	eio_req eio_mknod (const char path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1217	{	1811	{
1218	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	1812	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1219	}	1813	}
1220		1814
1221	static eio_req *	1815	static eio_req *
1222	eio__2path (int type, const char path, const char new_path, int pri, eio_cb cb, void *data)	1816	eio__2path (int type, const char path, const char new_path, int pri, eio_cb cb, void *data)
1223	{	1817	{
…		…
1247	eio_req eio_rename (const char path, const char new_path, int pri, eio_cb cb, void data)	1841	eio_req eio_rename (const char path, const char new_path, int pri, eio_cb cb, void data)
1248	{	1842	{
1249	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	1843	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1250	}	1844	}
1251		1845
		1846	eio_req eio_custom (eio_cb execute, int pri, eio_cb cb, void data)
		1847	{
		1848	REQ (EIO_CUSTOM); req->feed = (void ()(eio_req ))execute; SEND;
		1849	}
		1850
		1851	#endif
		1852
1252	eio_req eio_grp (eio_cb cb, void data)	1853	eio_req eio_grp (eio_cb cb, void data)
1253	{	1854	{
1254	const int pri = EIO_PRI_MAX;	1855	const int pri = EIO_PRI_MAX;
1255		1856
1256	REQ (EIO_GROUP); SEND;	1857	REQ (EIO_GROUP); SEND;
…		…
1280		1881
1281	void eio_grp_add (eio_req grp, eio_req req)	1882	void eio_grp_add (eio_req grp, eio_req req)
1282	{	1883	{
1283	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	1884	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1284		1885
		1886	grp->flags \|= EIO_FLAG_GROUPADD;
		1887
1285	++grp->size;	1888	++grp->size;
1286	req->grp = grp;	1889	req->grp = grp;
1287		1890
1288	req->grp_prev = 0;	1891	req->grp_prev = 0;
1289	req->grp_next = grp->grp_first;	1892	req->grp_next = grp->grp_first;
…		…
1297	/*****************************************************************************/	1900	/*****************************************************************************/
1298	/* misc garbage */	1901	/* misc garbage */
1299		1902
1300	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	1903	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1301	{	1904	{
1302	worker wrk;	1905	etp_worker wrk;
		1906	ssize_t ret;
1303		1907
1304	wrk.dbuf = 0;	1908	wrk.dbuf = 0;
1305		1909
1306	eio__sendfile (ofd, ifd, offset, count, &wrk);	1910	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1307		1911
1308	if (wrk.dbuf)	1912	if (wrk.dbuf)
1309	free (wrk.dbuf);	1913	free (wrk.dbuf);
1310	}
1311		1914
		1915	return ret;
		1916	}
		1917

Diff Legend

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

Comparing libeio/eio.c (file contents): Revision 1.10 by root, Tue May 13 17:08:15 2008 UTC vs. Revision 1.46 by root, Sat Jan 2 12:58:37 2010 UTC

Diff Legend

Comparing libeio/eio.c (file contents):
Revision 1.10 by root, Tue May 13 17:08:15 2008 UTC vs.
Revision 1.46 by root, Sat Jan 2 12:58:37 2010 UTC