ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.8 by root, Sun May 11 20:08:37 2008 UTC vs.
Revision 1.85 by root, Wed Jul 13 00:53:03 2011 UTC

		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
		5	* All rights reserved.
		6	*
		7	* Redistribution and use in source and binary forms, with or without modifica-
		8	* tion, are permitted provided that the following conditions are met:
		9	*
		10	* 1. Redistributions of source code must retain the above copyright notice,
		11	* this list of conditions and the following disclaimer.
		12	*
		13	* 2. Redistributions in binary form must reproduce the above copyright
		14	* notice, this list of conditions and the following disclaimer in the
		15	* documentation and/or other materials provided with the distribution.
		16	*
		17	* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
		18	* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
		19	* CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
		20	* EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
		21	* CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
		22	* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
		23	* OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
		24	* WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
		25	* ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
		26	* OF THE POSSIBILITY OF SUCH DAMAGE.
		27	*
		28	* Alternatively, the contents of this file may be used under the terms of
		29	* the GNU General Public License ("GPL") version 2 or any later version,
		30	* in which case the provisions of the GPL are applicable instead of
		31	* the above. If you wish to allow the use of your version of this file
		32	* only under the terms of the GPL and not to allow others to use your
		33	* version of this file under the BSD license, indicate your decision
		34	* by deleting the provisions above and replace them with the notice
		35	* and other provisions required by the GPL. If you do not delete the
		36	* provisions above, a recipient may use your version of this file under
		37	* either the BSD or the GPL.
		38	*/
		39
		40	#ifndef _WIN32
		41	# include "config.h"
		42	#endif
		43
1	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
		46
		47	#ifdef EIO_STACKSIZE
		48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
		49	#endif
2	#include "xthread.h"	50	#include "xthread.h"
3		51
4	#include <errno.h>	52	#include <errno.h>
5	#include <stddef.h>	53	#include <stddef.h>
6	#include <stdlib.h>	54	#include <stdlib.h>
7	#include <string.h>	55	#include <string.h>
8	#include <errno.h>	56	#include <errno.h>
9	#include <sys/types.h>	57	#include <sys/types.h>
10	#include <sys/stat.h>	58	#include <sys/stat.h>
		59	#include <sys/statvfs.h>
11	#include <limits.h>	60	#include <limits.h>
12	#include <fcntl.h>	61	#include <fcntl.h>
13	#include <assert.h>	62	#include <assert.h>
14		63
		64	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
		65	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
		66	#if HAVE_STDINT_H
		67	# include <stdint.h>
		68	#endif
		69
		70	#ifndef ECANCELED
		71	# define ECANCELED EDOM
		72	#endif
		73
		74	static void eio_destroy (eio_req *req);
		75
15	#ifndef EIO_FINISH	76	#ifndef EIO_FINISH
16	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	77	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17	#endif	78	#endif
18		79
19	#ifndef EIO_DESTROY	80	#ifndef EIO_DESTROY
…		…
25	#endif	86	#endif
26		87
27	#ifdef _WIN32	88	#ifdef _WIN32
28		89
29	/*doh*/	90	/*doh*/
30
31	#else	91	#else
32		92
33	# include "config.h"
34	# include <sys/time.h>	93	# include <sys/time.h>
35	# include <sys/select.h>	94	# include <sys/select.h>
36	# include <unistd.h>	95	# include <unistd.h>
37	# include <utime.h>	96	# include <utime.h>
38	# include <signal.h>	97	# include <signal.h>
39	# include <dirent.h>	98	# include <dirent.h>
40		99
		100	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		101	# include <sys/mman.h>
		102	#endif
		103
		104	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		105	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		106	# define _DIRENT_HAVE_D_TYPE /* sigh */
		107	# define D_INO(de) (de)->d_fileno
		108	# define D_NAMLEN(de) (de)->d_namlen
		109	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		110	# define D_INO(de) (de)->d_ino
		111	# endif
		112
		113	#ifdef _D_EXACT_NAMLEN
		114	# undef D_NAMLEN
		115	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		116	#endif
		117
		118	# ifdef _DIRENT_HAVE_D_TYPE
		119	# define D_TYPE(de) (de)->d_type
		120	# endif
		121
41	# ifndef EIO_STRUCT_DIRENT	122	# ifndef EIO_STRUCT_DIRENT
42	# define EIO_STRUCT_DIRENT struct dirent	123	# define EIO_STRUCT_DIRENT struct dirent
43	# endif	124	# endif
44		125
45	#endif	126	#endif
46		127
47	# ifndef EIO_STRUCT_STAT
48	# define EIO_STRUCT_STAT struct stat
49	# endif
50
51	#if HAVE_SENDFILE	128	#if HAVE_SENDFILE
52	# if __linux	129	# if __linux
53	# include <sys/sendfile.h>	130	# include <sys/sendfile.h>
54	# elif __freebsd	131	# elif __FreeBSD__ || defined __APPLE__
55	# include <sys/socket.h>	132	# include <sys/socket.h>
56	# include <sys/uio.h>	133	# include <sys/uio.h>
57	# elif __hpux	134	# elif __hpux
58	# include <sys/socket.h>	135	# include <sys/socket.h>
59	# elif __solaris /* not yet */	136	# elif __solaris
60	# include <sys/sendfile.h>	137	# include <sys/sendfile.h>
61	# else	138	# else
62	# error sendfile support requested but not available	139	# error sendfile support requested but not available
63	# endif	140	# endif
64	#endif	141	#endif
65		142
66	/* number of seconds after which an idle threads exit */	143	#ifndef D_TYPE
67	#define IDLE_TIMEOUT 10	144	# define D_TYPE(de) 0
		145	#endif
		146	#ifndef D_INO
		147	# define D_INO(de) 0
		148	#endif
		149	#ifndef D_NAMLEN
		150	# define D_NAMLEN(de) strlen ((de)->d_name)
		151	#endif
68		152
69	/* used for struct dirent, AIX doesn't provide it */	153	/* used for struct dirent, AIX doesn't provide it */
70	#ifndef NAME_MAX	154	#ifndef NAME_MAX
71	# define NAME_MAX 4096	155	# define NAME_MAX 4096
72	#endif	156	#endif
73		157
		158	/* used for readlink etc. */
		159	#ifndef PATH_MAX
		160	# define PATH_MAX 4096
		161	#endif
		162
74	/* buffer size for various temporary buffers */	163	/* buffer size for various temporary buffers */
75	#define EIO_BUFSIZE 65536	164	#define EIO_BUFSIZE 65536
76		165
77	#define dBUF \	166	#define dBUF \
78	char *eio_buf; \	167	char *eio_buf; \
79	X_LOCK (wrklock); \	168	ETP_WORKER_LOCK (self); \
80	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	169	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81	X_UNLOCK (wrklock); \	170	ETP_WORKER_UNLOCK (self); \
		171	errno = ENOMEM; \
82	if (!eio_buf) \	172	if (!eio_buf) \
83	return -1;	173	return -1;
84		174
85	#define EIO_TICKS ((1000000 + 1023) >> 10)	175	#define EIO_TICKS ((1000000 + 1023) >> 10)
86		176
		177	#define ETP_PRI_MIN EIO_PRI_MIN
		178	#define ETP_PRI_MAX EIO_PRI_MAX
		179
		180	struct etp_worker;
		181
		182	#define ETP_REQ eio_req
		183	#define ETP_DESTROY(req) eio_destroy (req)
		184	static int eio_finish (eio_req *req);
		185	#define ETP_FINISH(req) eio_finish (req)
		186	static void eio_execute (struct etp_worker *self, eio_req *req);
		187	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		188
		189	#define ETP_WORKER_CLEAR(req) \
		190	if (wrk->dbuf) \
		191	{ \
		192	free (wrk->dbuf); \
		193	wrk->dbuf = 0; \
		194	} \
		195	\
		196	if (wrk->dirp) \
		197	{ \
		198	closedir (wrk->dirp); \
		199	wrk->dirp = 0; \
		200	}
		201
		202	#define ETP_WORKER_COMMON \
		203	void *dbuf; \
		204	DIR *dirp;
		205
		206	/*****************************************************************************/
		207
		208	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		209
		210	/* calculate time difference in ~1/EIO_TICKS of a second */
		211	ecb_inline int
		212	tvdiff (struct timeval *tv1, struct timeval *tv2)
		213	{
		214	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
		215	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
		216	}
		217
		218	static unsigned int started, idle, wanted = 4;
		219
87	static void (*want_poll_cb) (void);	220	static void (*want_poll_cb) (void);
88	static void (*done_poll_cb) (void);	221	static void (*done_poll_cb) (void);
89		222
90	static unsigned int max_poll_time = 0;	223	static unsigned int max_poll_time; /* reslock */
91	static unsigned int max_poll_reqs = 0;	224	static unsigned int max_poll_reqs; /* reslock */
92		225
93	/* calculcate time difference in ~1/EIO_TICKS of a second */	226	static volatile unsigned int nreqs; /* reqlock */
94	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	227	static volatile unsigned int nready; /* reqlock */
95	{	228	static volatile unsigned int npending; /* reqlock */
96	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	229	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	230	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98	}
99		231
100	static unsigned int started, idle, wanted;	232	static xmutex_t wrklock;
		233	static xmutex_t reslock;
		234	static xmutex_t reqlock;
		235	static xcond_t reqwait;
		236
		237	#if !HAVE_PREADWRITE
		238	/*
		239	* make our pread/pwrite emulation safe against themselves, but not against
		240	* normal read/write by using a mutex. slows down execution a lot,
		241	* but that's your problem, not mine.
		242	*/
		243	static xmutex_t preadwritelock = X_MUTEX_INIT;
		244	#endif
		245
		246	typedef struct etp_worker
		247	{
		248	/* locked by wrklock */
		249	struct etp_worker *prev, *next;
		250
		251	xthread_t tid;
		252
		253	/* locked by reslock, reqlock or wrklock */
		254	ETP_REQ *req; /* currently processed request */
		255
		256	ETP_WORKER_COMMON
		257	} etp_worker;
		258
		259	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
		260
		261	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		262	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
101		263
102	/* worker threads management */	264	/* worker threads management */
103	static mutex_t wrklock = X_MUTEX_INIT;
104		265
105	typedef struct worker {	266	static void ecb_cold
106	/* locked by wrklock */
107	struct worker *prev, *next;
108
109	thread_t tid;
110
111	/* locked by reslock, reqlock or wrklock */
112	eio_req *req; /* currently processed request */
113	void *dbuf;
114	DIR *dirp;
115	} worker;
116
117	static worker wrk_first = { &wrk_first, &wrk_first, 0 };
118
119	static void worker_clear (worker *wrk)	267	etp_worker_clear (etp_worker *wrk)
120	{	268	{
121	if (wrk->dirp)	269	ETP_WORKER_CLEAR (wrk);
122	{
123	closedir (wrk->dirp);
124	wrk->dirp = 0;
125	}
126
127	if (wrk->dbuf)
128	{
129	free (wrk->dbuf);
130	wrk->dbuf = 0;
131	}
132	}	270	}
133		271
134	static void worker_free (worker *wrk)	272	static void ecb_cold
		273	etp_worker_free (etp_worker *wrk)
135	{	274	{
136	wrk->next->prev = wrk->prev;	275	wrk->next->prev = wrk->prev;
137	wrk->prev->next = wrk->next;	276	wrk->prev->next = wrk->next;
138		277
139	free (wrk);	278	free (wrk);
140	}	279	}
141		280
142	static volatile unsigned int nreqs, nready, npending;	281	static unsigned int
143	static volatile unsigned int max_idle = 4;	282	etp_nreqs (void)
144
145	static mutex_t reslock = X_MUTEX_INIT;
146	static mutex_t reqlock = X_MUTEX_INIT;
147	static cond_t reqwait = X_COND_INIT;
148
149	unsigned int eio_nreqs (void)
150	{	283	{
		284	int retval;
		285	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		286	retval = nreqs;
		287	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151	return nreqs;	288	return retval;
152	}	289	}
153		290
154	unsigned int eio_nready (void)	291	static unsigned int
		292	etp_nready (void)
155	{	293	{
156	unsigned int retval;	294	unsigned int retval;
157		295
158	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	296	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159	retval = nready;	297	retval = nready;
160	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	298	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161		299
162	return retval;	300	return retval;
163	}	301	}
164		302
165	unsigned int eio_npending (void)	303	static unsigned int
		304	etp_npending (void)
166	{	305	{
167	unsigned int retval;	306	unsigned int retval;
168		307
169	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	308	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170	retval = npending;	309	retval = npending;
171	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	310	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172		311
173	return retval;	312	return retval;
174	}	313	}
175		314
176	unsigned int eio_nthreads (void)	315	static unsigned int
		316	etp_nthreads (void)
177	{	317	{
178	unsigned int retval;	318	unsigned int retval;
179		319
180	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	320	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181	retval = started;	321	retval = started;
…		…
188	* a somewhat faster data structure might be nice, but	328	* a somewhat faster data structure might be nice, but
189	* with 8 priorities this actually needs <20 insns	329	* with 8 priorities this actually needs <20 insns
190	* per shift, the most expensive operation.	330	* per shift, the most expensive operation.
191	*/	331	*/
192	typedef struct {	332	typedef struct {
193	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	333	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194	int size;	334	int size;
195	} reqq;	335	} etp_reqq;
196		336
197	static reqq req_queue;	337	static etp_reqq req_queue;
198	static reqq res_queue;	338	static etp_reqq res_queue;
199		339
200	static int reqq_push (reqq *q, eio_req *req)	340	static int ecb_noinline
		341	reqq_push (etp_reqq *q, ETP_REQ *req)
201	{	342	{
202	int pri = req->pri;	343	int pri = req->pri;
203	req->next = 0;	344	req->next = 0;
204		345
205	if (q->qe[pri])	346	if (q->qe[pri])
…		…
211	q->qe[pri] = q->qs[pri] = req;	352	q->qe[pri] = q->qs[pri] = req;
212		353
213	return q->size++;	354	return q->size++;
214	}	355	}
215		356
216	static eio_req *reqq_shift (reqq *q)	357	static ETP_REQ * ecb_noinline
		358	reqq_shift (etp_reqq *q)
217	{	359	{
218	int pri;	360	int pri;
219		361
220	if (!q->size)	362	if (!q->size)
221	return 0;	363	return 0;
222		364
223	--q->size;	365	--q->size;
224		366
225	for (pri = EIO_NUM_PRI; pri--; )	367	for (pri = ETP_NUM_PRI; pri--; )
226	{	368	{
227	eio_req *req = q->qs[pri];	369	eio_req *req = q->qs[pri];
228		370
229	if (req)	371	if (req)
230	{	372	{
…		…
236	}	378	}
237		379
238	abort ();	380	abort ();
239	}	381	}
240		382
241	static void grp_try_feed (eio_req *grp)	383	static void ecb_cold
		384	etp_thread_init (void)
242	{	385	{
243	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	386	#if !HAVE_PREADWRITE
		387	X_MUTEX_CREATE (preadwritelock);
		388	#endif
		389	X_MUTEX_CREATE (wrklock);
		390	X_MUTEX_CREATE (reslock);
		391	X_MUTEX_CREATE (reqlock);
		392	X_COND_CREATE (reqwait);
		393	}
		394
		395	static void ecb_cold
		396	etp_atfork_prepare (void)
		397	{
		398	}
		399
		400	static void ecb_cold
		401	etp_atfork_parent (void)
		402	{
		403	}
		404
		405	static void ecb_cold
		406	etp_atfork_child (void)
		407	{
		408	ETP_REQ *prv;
		409
		410	while ((prv = reqq_shift (&req_queue)))
		411	ETP_DESTROY (prv);
		412
		413	while ((prv = reqq_shift (&res_queue)))
		414	ETP_DESTROY (prv);
		415
		416	while (wrk_first.next != &wrk_first)
244	{	417	{
245	int old_len = grp->size;	418	etp_worker *wrk = wrk_first.next;
246		419
247	EIO_FEED (grp);	420	if (wrk->req)
		421	ETP_DESTROY (wrk->req);
248		422
249	/* stop if no progress has been made */	423	etp_worker_clear (wrk);
250	if (old_len == grp->size)	424	etp_worker_free (wrk);
251	{
252	grp->feed = 0;
253	break;
254	}
255	}	425	}
256	}
257		426
258	static int eio_finish (eio_req *req);	427	started = 0;
		428	idle = 0;
		429	nreqs = 0;
		430	nready = 0;
		431	npending = 0;
259		432
260	static int grp_dec (eio_req *grp)	433	etp_thread_init ();
261	{	434	}
262	--grp->size;
263		435
264	/* call feeder, if applicable */	436	static void ecb_cold
265	grp_try_feed (grp);	437	etp_once_init (void)
		438	{
		439	etp_thread_init ();
		440	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
		441	}
266		442
267	/* finish, if done */	443	static int ecb_cold
268	if (!grp->size && grp->int1)	444	etp_init (void (*want_poll)(void), void (*done_poll)(void))
269	return eio_finish (grp);	445	{
270	else	446	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
		447
		448	pthread_once (&doinit, etp_once_init);
		449
		450	want_poll_cb = want_poll;
		451	done_poll_cb = done_poll;
		452
271	return 0;	453	return 0;
272	}	454	}
273		455
274	void eio_destroy (eio_req *req)
275	{
276	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
277	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
278
279	EIO_DESTROY (req);
280	}
281
282	static int eio_finish (eio_req *req)
283	{
284	int res = EIO_FINISH (req);
285
286	if (req->grp)
287	{
288	int res2;
289	eio_req *grp = req->grp;
290
291	/* unlink request */
292	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
293	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
294
295	if (grp->grp_first == req)
296	grp->grp_first = req->grp_next;
297
298	res2 = grp_dec (grp);
299
300	if (!res && res2)
301	res = res2;
302	}
303
304	eio_destroy (req);
305
306	return res;
307	}
308
309	void eio_grp_cancel (eio_req *grp)
310	{
311	for (grp = grp->grp_first; grp; grp = grp->grp_next)
312	eio_cancel (grp);
313	}
314
315	void eio_cancel (eio_req *req)
316	{
317	X_LOCK (wrklock);
318	req->flags |= EIO_FLAG_CANCELLED;
319	X_UNLOCK (wrklock);
320
321	eio_grp_cancel (req);
322	}
323
324	X_THREAD_PROC (eio_proc);	456	X_THREAD_PROC (etp_proc);
325		457
326	static void start_thread (void)	458	static void ecb_cold
		459	etp_start_thread (void)
327	{	460	{
328	worker *wrk = calloc (1, sizeof (worker));	461	etp_worker *wrk = calloc (1, sizeof (etp_worker));
329		462
330	/*TODO*/	463	/*TODO*/
331	assert (("unable to allocate worker thread data", !wrk));	464	assert (("unable to allocate worker thread data", wrk));
332		465
333	X_LOCK (wrklock);	466	X_LOCK (wrklock);
334		467
335	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	468	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
336	{	469	{
337	wrk->prev = &wrk_first;	470	wrk->prev = &wrk_first;
338	wrk->next = wrk_first.next;	471	wrk->next = wrk_first.next;
339	wrk_first.next->prev = wrk;	472	wrk_first.next->prev = wrk;
340	wrk_first.next = wrk;	473	wrk_first.next = wrk;
…		…
344	free (wrk);	477	free (wrk);
345		478
346	X_UNLOCK (wrklock);	479	X_UNLOCK (wrklock);
347	}	480	}
348		481
		482	static void
349	static void maybe_start_thread (void)	483	etp_maybe_start_thread (void)
350	{	484	{
351	if (eio_nthreads () >= wanted)	485	if (ecb_expect_true (etp_nthreads () >= wanted))
352	return;	486	return;
353		487
354	/* todo: maybe use idle here, but might be less exact */	488	/* todo: maybe use idle here, but might be less exact */
355	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	489	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
356	return;	490	return;
357		491
358	start_thread ();	492	etp_start_thread ();
359	}	493	}
360		494
361	void eio_submit (eio_req *req)	495	static void ecb_cold
362	{	496	etp_end_thread (void)
363	++nreqs;
364
365	X_LOCK (reqlock);
366	++nready;
367	reqq_push (&req_queue, req);
368	X_COND_SIGNAL (reqwait);
369	X_UNLOCK (reqlock);
370
371	maybe_start_thread ();
372	}
373
374	static void end_thread (void)
375	{	497	{
376	eio_req *req = calloc (1, sizeof (eio_req));	498	eio_req *req = calloc (1, sizeof (eio_req));
377		499
378	req->type = EIO_QUIT;	500	req->type = -1;
379	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	501	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
380		502
381	X_LOCK (reqlock);	503	X_LOCK (reqlock);
382	reqq_push (&req_queue, req);	504	reqq_push (&req_queue, req);
383	X_COND_SIGNAL (reqwait);	505	X_COND_SIGNAL (reqwait);
384	X_UNLOCK (reqlock);	506	X_UNLOCK (reqlock);
…		…
386	X_LOCK (wrklock);	508	X_LOCK (wrklock);
387	--started;	509	--started;
388	X_UNLOCK (wrklock);	510	X_UNLOCK (wrklock);
389	}	511	}
390		512
391	void eio_set_max_poll_time (double nseconds)	513	static int
392	{
393	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
394	max_poll_time = nseconds;
395	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
396	}
397
398	void eio_set_max_poll_reqs (unsigned int maxreqs)
399	{
400	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
401	max_poll_reqs = maxreqs;
402	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
403	}
404
405	void eio_set_max_idle (unsigned int nthreads)
406	{
407	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
408	max_idle = nthreads <= 0 ? 1 : nthreads;
409	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
410	}
411
412	void eio_set_min_parallel (unsigned int nthreads)
413	{
414	if (wanted < nthreads)
415	wanted = nthreads;
416	}
417
418	void eio_set_max_parallel (unsigned int nthreads)
419	{
420	if (wanted > nthreads)
421	wanted = nthreads;
422
423	while (started > wanted)
424	end_thread ();
425	}
426
427	int eio_poll (void)	514	etp_poll (void)
428	{	515	{
429	int maxreqs = max_poll_reqs;	516	unsigned int maxreqs;
		517	unsigned int maxtime;
430	struct timeval tv_start, tv_now;	518	struct timeval tv_start, tv_now;
431	eio_req *req;
432		519
		520	X_LOCK (reslock);
		521	maxreqs = max_poll_reqs;
		522	maxtime = max_poll_time;
		523	X_UNLOCK (reslock);
		524
433	if (max_poll_time)	525	if (maxtime)
434	gettimeofday (&tv_start, 0);	526	gettimeofday (&tv_start, 0);
435		527
436	for (;;)	528	for (;;)
437	{	529	{
		530	ETP_REQ *req;
		531
438	maybe_start_thread ();	532	etp_maybe_start_thread ();
439		533
440	X_LOCK (reslock);	534	X_LOCK (reslock);
441	req = reqq_shift (&res_queue);	535	req = reqq_shift (&res_queue);
442		536
443	if (req)	537	if (req)
…		…
451	X_UNLOCK (reslock);	545	X_UNLOCK (reslock);
452		546
453	if (!req)	547	if (!req)
454	return 0;	548	return 0;
455		549
		550	X_LOCK (reqlock);
456	--nreqs;	551	--nreqs;
		552	X_UNLOCK (reqlock);
457		553
458	if (req->type == EIO_GROUP && req->size)	554	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
459	{	555	{
460	req->int1 = 1; /* mark request as delayed */	556	req->int1 = 1; /* mark request as delayed */
461	continue;	557	continue;
462	}	558	}
463	else	559	else
464	{	560	{
465	int res = eio_finish (req);	561	int res = ETP_FINISH (req);
466	if (res)	562	if (ecb_expect_false (res))
467	return res;	563	return res;
468	}	564	}
469		565
470	if (maxreqs && !--maxreqs)	566	if (ecb_expect_false (maxreqs && !--maxreqs))
471	break;	567	break;
472		568
473	if (max_poll_time)	569	if (maxtime)
474	{	570	{
475	gettimeofday (&tv_now, 0);	571	gettimeofday (&tv_now, 0);
476		572
477	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	573	if (tvdiff (&tv_start, &tv_now) >= maxtime)
478	break;	574	break;
479	}	575	}
480	}	576	}
481		577
482	errno = EAGAIN;	578	errno = EAGAIN;
483	return -1;	579	return -1;
484	}	580	}
485		581
		582	static void
		583	etp_cancel (ETP_REQ *req)
		584	{
		585	req->cancelled = 1;
		586
		587	eio_grp_cancel (req);
		588	}
		589
		590	static void
		591	etp_submit (ETP_REQ *req)
		592	{
		593	req->pri -= ETP_PRI_MIN;
		594
		595	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		596	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		597
		598	if (ecb_expect_false (req->type == EIO_GROUP))
		599	{
		600	/* I hope this is worth it :/ */
		601	X_LOCK (reqlock);
		602	++nreqs;
		603	X_UNLOCK (reqlock);
		604
		605	X_LOCK (reslock);
		606
		607	++npending;
		608
		609	if (!reqq_push (&res_queue, req) && want_poll_cb)
		610	want_poll_cb ();
		611
		612	X_UNLOCK (reslock);
		613	}
		614	else
		615	{
		616	X_LOCK (reqlock);
		617	++nreqs;
		618	++nready;
		619	reqq_push (&req_queue, req);
		620	X_COND_SIGNAL (reqwait);
		621	X_UNLOCK (reqlock);
		622
		623	etp_maybe_start_thread ();
		624	}
		625	}
		626
		627	static void ecb_cold
		628	etp_set_max_poll_time (double nseconds)
		629	{
		630	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		631	max_poll_time = nseconds * EIO_TICKS;
		632	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		633	}
		634
		635	static void ecb_cold
		636	etp_set_max_poll_reqs (unsigned int maxreqs)
		637	{
		638	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		639	max_poll_reqs = maxreqs;
		640	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		641	}
		642
		643	static void ecb_cold
		644	etp_set_max_idle (unsigned int nthreads)
		645	{
		646	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		647	max_idle = nthreads;
		648	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		649	}
		650
		651	static void ecb_cold
		652	etp_set_idle_timeout (unsigned int seconds)
		653	{
		654	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		655	idle_timeout = seconds;
		656	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		657	}
		658
		659	static void ecb_cold
		660	etp_set_min_parallel (unsigned int nthreads)
		661	{
		662	if (wanted < nthreads)
		663	wanted = nthreads;
		664	}
		665
		666	static void ecb_cold
		667	etp_set_max_parallel (unsigned int nthreads)
		668	{
		669	if (wanted > nthreads)
		670	wanted = nthreads;
		671
		672	while (started > wanted)
		673	etp_end_thread ();
		674	}
		675
		676	/*****************************************************************************/
		677
		678	static void
		679	grp_try_feed (eio_req *grp)
		680	{
		681	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		682	{
		683	grp->flags &= ~EIO_FLAG_GROUPADD;
		684
		685	EIO_FEED (grp);
		686
		687	/* stop if no progress has been made */
		688	if (!(grp->flags & EIO_FLAG_GROUPADD))
		689	{
		690	grp->feed = 0;
		691	break;
		692	}
		693	}
		694	}
		695
		696	static int
		697	grp_dec (eio_req *grp)
		698	{
		699	--grp->size;
		700
		701	/* call feeder, if applicable */
		702	grp_try_feed (grp);
		703
		704	/* finish, if done */
		705	if (!grp->size && grp->int1)
		706	return eio_finish (grp);
		707	else
		708	return 0;
		709	}
		710
		711	static void
		712	eio_destroy (eio_req *req)
		713	{
		714	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		715	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		716
		717	EIO_DESTROY (req);
		718	}
		719
		720	static int
		721	eio_finish (eio_req *req)
		722	{
		723	int res = EIO_FINISH (req);
		724
		725	if (req->grp)
		726	{
		727	int res2;
		728	eio_req *grp = req->grp;
		729
		730	/* unlink request */
		731	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		732	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		733
		734	if (grp->grp_first == req)
		735	grp->grp_first = req->grp_next;
		736
		737	res2 = grp_dec (grp);
		738
		739	if (!res)
		740	res = res2;
		741	}
		742
		743	eio_destroy (req);
		744
		745	return res;
		746	}
		747
		748	void
		749	eio_grp_cancel (eio_req *grp)
		750	{
		751	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		752	eio_cancel (grp);
		753	}
		754
		755	void
		756	eio_cancel (eio_req *req)
		757	{
		758	etp_cancel (req);
		759	}
		760
		761	void
		762	eio_submit (eio_req *req)
		763	{
		764	etp_submit (req);
		765	}
		766
		767	unsigned int
		768	eio_nreqs (void)
		769	{
		770	return etp_nreqs ();
		771	}
		772
		773	unsigned int
		774	eio_nready (void)
		775	{
		776	return etp_nready ();
		777	}
		778
		779	unsigned int
		780	eio_npending (void)
		781	{
		782	return etp_npending ();
		783	}
		784
		785	unsigned int ecb_cold
		786	eio_nthreads (void)
		787	{
		788	return etp_nthreads ();
		789	}
		790
		791	void ecb_cold
		792	eio_set_max_poll_time (double nseconds)
		793	{
		794	etp_set_max_poll_time (nseconds);
		795	}
		796
		797	void ecb_cold
		798	eio_set_max_poll_reqs (unsigned int maxreqs)
		799	{
		800	etp_set_max_poll_reqs (maxreqs);
		801	}
		802
		803	void ecb_cold
		804	eio_set_max_idle (unsigned int nthreads)
		805	{
		806	etp_set_max_idle (nthreads);
		807	}
		808
		809	void ecb_cold
		810	eio_set_idle_timeout (unsigned int seconds)
		811	{
		812	etp_set_idle_timeout (seconds);
		813	}
		814
		815	void ecb_cold
		816	eio_set_min_parallel (unsigned int nthreads)
		817	{
		818	etp_set_min_parallel (nthreads);
		819	}
		820
		821	void ecb_cold
		822	eio_set_max_parallel (unsigned int nthreads)
		823	{
		824	etp_set_max_parallel (nthreads);
		825	}
		826
		827	int eio_poll (void)
		828	{
		829	return etp_poll ();
		830	}
		831
486	/*****************************************************************************/	832	/*****************************************************************************/
487	/* work around various missing functions */	833	/* work around various missing functions */
488		834
489	#if !HAVE_PREADWRITE	835	#if !HAVE_PREADWRITE
		836	# undef pread
		837	# undef pwrite
490	# define pread aio_pread	838	# define pread eio__pread
491	# define pwrite aio_pwrite	839	# define pwrite eio__pwrite
492		840
493	/*	841	static ssize_t
494	* make our pread/pwrite safe against themselves, but not against	842	eio__pread (int fd, void *buf, size_t count, off_t offset)
495	* normal read/write by using a mutex. slows down execution a lot,
496	* but that's your problem, not mine.
497	*/
498	static mutex_t preadwritelock = X_MUTEX_INIT;
499
500	static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
501	{	843	{
502	ssize_t res;	844	ssize_t res;
503	off_t ooffset;	845	off_t ooffset;
504		846
505	X_LOCK (preadwritelock);	847	X_LOCK (preadwritelock);
…		…
510	X_UNLOCK (preadwritelock);	852	X_UNLOCK (preadwritelock);
511		853
512	return res;	854	return res;
513	}	855	}
514		856
515	static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset)	857	static ssize_t
		858	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
516	{	859	{
517	ssize_t res;	860	ssize_t res;
518	off_t ooffset;	861	off_t ooffset;
519		862
520	X_LOCK (preadwritelock);	863	X_LOCK (preadwritelock);
521	ooffset = lseek (fd, 0, SEEK_CUR);	864	ooffset = lseek (fd, 0, SEEK_CUR);
522	lseek (fd, offset, SEEK_SET);	865	lseek (fd, offset, SEEK_SET);
523	res = write (fd, buf, count);	866	res = write (fd, buf, count);
524	lseek (fd, offset, SEEK_SET);	867	lseek (fd, ooffset, SEEK_SET);
525	X_UNLOCK (preadwritelock);	868	X_UNLOCK (preadwritelock);
526		869
527	return res;	870	return res;
528	}	871	}
529	#endif	872	#endif
530		873
531	#ifndef HAVE_FUTIMES	874	#ifndef HAVE_UTIMES
532		875
		876	# undef utimes
533	# define utimes(path,times) aio_utimes (path, times)	877	# define utimes(path,times) eio__utimes (path, times)
534	# define futimes(fd,times) aio_futimes (fd, times)
535		878
		879	static int
536	static int aio_utimes (const char *filename, const struct timeval times[2])	880	eio__utimes (const char *filename, const struct timeval times[2])
537	{	881	{
538	if (times)	882	if (times)
539	{	883	{
540	struct utimbuf buf;	884	struct utimbuf buf;
541		885
…		…
546	}	890	}
547	else	891	else
548	return utime (filename, 0);	892	return utime (filename, 0);
549	}	893	}
550		894
		895	#endif
		896
		897	#ifndef HAVE_FUTIMES
		898
		899	# undef futimes
		900	# define futimes(fd,times) eio__futimes (fd, times)
		901
		902	static int
551	static int aio_futimes (int fd, const struct timeval tv[2])	903	eio__futimes (int fd, const struct timeval tv[2])
552	{	904	{
553	errno = ENOSYS;	905	errno = ENOSYS;
554	return -1;	906	return -1;
555	}	907	}
556		908
557	#endif	909	#endif
558		910
559	#if !HAVE_FDATASYNC	911	#if !HAVE_FDATASYNC
		912	# undef fdatasync
560	# define fdatasync fsync	913	# define fdatasync(fd) fsync (fd)
561	#endif	914	#endif
		915
		916	/* sync_file_range always needs emulation */
		917	static int
		918	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		919	{
		920	#if HAVE_SYNC_FILE_RANGE
		921	int res;
		922
		923	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		924	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		925	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		926	{
		927	flags = 0
		928	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		929	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		930	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		931	}
		932
		933	res = sync_file_range (fd, offset, nbytes, flags);
		934
		935	if (!res || errno != ENOSYS)
		936	return res;
		937	#endif
		938
		939	/* even though we could play tricks with the flags, it's better to always
		940	* call fdatasync, as that matches the expectation of its users best */
		941	return fdatasync (fd);
		942	}
		943
		944	static int
		945	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		946	{
		947	#if HAVE_FALLOCATE
		948	return fallocate (fd, mode, offset, len);
		949	#else
		950	errno = ENOSYS;
		951	return -1;
		952	#endif
		953	}
562		954
563	#if !HAVE_READAHEAD	955	#if !HAVE_READAHEAD
		956	# undef readahead
564	# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self)	957	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565		958
		959	static ssize_t
566	static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self)	960	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
567	{	961	{
568	size_t todo = count;	962	size_t todo = count;
569	dBUF;	963	dBUF;
570		964
571	while (todo > 0)	965	while (todo > 0)
…		…
581	return count;	975	return count;
582	}	976	}
583		977
584	#endif	978	#endif
585		979
586	#if !HAVE_READDIR_R
587	# define readdir_r aio_readdir_r
588
589	static mutex_t readdirlock = X_MUTEX_INIT;
590
591	static int readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
592	{
593	EIO_STRUCT_DIRENT *e;
594	int errorno;
595
596	X_LOCK (readdirlock);
597
598	e = readdir (dirp);
599	errorno = errno;
600
601	if (e)
602	{
603	*res = ent;
604	strcpy (ent->d_name, e->d_name);
605	}
606	else
607	*res = 0;
608
609	X_UNLOCK (readdirlock);
610
611	errno = errorno;
612	return e ? 0 : -1;
613	}
614	#endif
615
616	/* sendfile always needs emulation */	980	/* sendfile always needs emulation */
		981	static ssize_t
617	static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self)	982	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
618	{	983	{
		984	ssize_t written = 0;
619	ssize_t res;	985	ssize_t res;
620		986
621	if (!count)	987	if (!count)
622	return 0;	988	return 0;
623		989
		990	for (;;)
		991	{
		992	#ifdef __APPLE__
		993	# undef HAVE_SENDFILE /* broken, as everything on os x */
		994	#endif
624	#if HAVE_SENDFILE	995	#if HAVE_SENDFILE
625	# if __linux	996	# if __linux
		997	off_t soffset = offset;
626	res = sendfile (ofd, ifd, &offset, count);	998	res = sendfile (ofd, ifd, &soffset, count);
627		999
628	# elif __freebsd	1000	# elif __FreeBSD__
629	/*	1001	/*
630	* Of course, the freebsd sendfile is a dire hack with no thoughts	1002	* Of course, the freebsd sendfile is a dire hack with no thoughts
631	* wasted on making it similar to other I/O functions.	1003	* wasted on making it similar to other I/O functions.
632	*/	1004	*/
633	{
634	off_t sbytes;	1005	off_t sbytes;
635	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1006	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636		1007
637	if (res < 0 && sbytes)	1008	#if 0 /* according to the manpage, this is correct, but broken behaviour */
638	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1009	/* freebsd' sendfile will return 0 on success */
		1010	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1011	/* not on e.g. EIO or EPIPE - sounds broken */
		1012	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
639	res = sbytes;	1013	res = sbytes;
640	}	1014	#endif
		1015
		1016	/* according to source inspection, this is correct, and useful behaviour */
		1017	if (sbytes)
		1018	res = sbytes;
		1019
		1020	# elif defined (__APPLE__)
		1021	off_t sbytes = count;
		1022	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1023
		1024	/* according to the manpage, sbytes is always valid */
		1025	if (sbytes)
		1026	res = sbytes;
641		1027
642	# elif __hpux	1028	# elif __hpux
643	res = sendfile (ofd, ifd, offset, count, 0, 0);	1029	res = sendfile (ofd, ifd, offset, count, 0, 0);
644		1030
645	# elif __solaris	1031	# elif __solaris
646	{
647	struct sendfilevec vec;	1032	struct sendfilevec vec;
648	size_t sbytes;	1033	size_t sbytes;
649		1034
650	vec.sfv_fd = ifd;	1035	vec.sfv_fd = ifd;
651	vec.sfv_flag = 0;	1036	vec.sfv_flag = 0;
652	vec.sfv_off = offset;	1037	vec.sfv_off = offset;
653	vec.sfv_len = count;	1038	vec.sfv_len = count;
654		1039
655	res = sendfilev (ofd, &vec, 1, &sbytes);	1040	res = sendfilev (ofd, &vec, 1, &sbytes);
656		1041
657	if (res < 0 && sbytes)	1042	if (res < 0 && sbytes)
658	res = sbytes;	1043	res = sbytes;
659	}
660		1044
661	# endif	1045	# endif
		1046
		1047	#elif defined (_WIN32)
		1048	/* does not work, just for documentation of what would need to be done */
		1049	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1050	/* libeio guarantees that the file offset does not change, and windows */
		1051	/* has no way to get an independent handle to the same file description */
		1052	HANDLE h = TO_SOCKET (ifd);
		1053	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1054	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1055
662	#else	1056	#else
663	res = -1;	1057	res = -1;
664	errno = ENOSYS;	1058	errno = ENOSYS;
665	#endif	1059	#endif
666		1060
		1061	/* we assume sendfile can copy at least 128mb in one go */
		1062	if (res <= 128 * 1024 * 1024)
		1063	{
		1064	if (res > 0)
		1065	written += res;
		1066
		1067	if (written)
		1068	return written;
		1069
		1070	break;
		1071	}
		1072	else
		1073	{
		1074	/* if we requested more, then probably the kernel was lazy */
		1075	written += res;
		1076	offset += res;
		1077	count -= res;
		1078
		1079	if (!count)
		1080	return written;
		1081	}
		1082	}
		1083
667	if (res < 0	1084	if (res < 0
668	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1085	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1086	/* BSDs */
		1087	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1088	|| errno == ENOTSUP
		1089	#endif
		1090	|| errno == EOPNOTSUPP /* BSDs */
669	#if __solaris	1091	#if __solaris
670	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1092	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
671	#endif	1093	#endif
672	)	1094	)
673	)	1095	)
…		…
704	}	1126	}
705		1127
706	return res;	1128	return res;
707	}	1129	}
708		1130
		1131	#ifdef PAGESIZE
		1132	# define eio_pagesize() PAGESIZE
		1133	#else
		1134	static intptr_t
		1135	eio_pagesize (void)
		1136	{
		1137	static intptr_t page;
		1138
		1139	if (!page)
		1140	page = sysconf (_SC_PAGESIZE);
		1141
		1142	return page;
		1143	}
		1144	#endif
		1145
		1146	static void
		1147	eio_page_align (void **addr, size_t *length)
		1148	{
		1149	intptr_t mask = eio_pagesize () - 1;
		1150
		1151	/* round down addr */
		1152	intptr_t adj = mask & (intptr_t)*addr;
		1153
		1154	*addr = (void *)((intptr_t)*addr - adj);
		1155	*length += adj;
		1156
		1157	/* round up length */
		1158	*length = (*length + mask) & ~mask;
		1159	}
		1160
		1161	#if !_POSIX_MEMLOCK
		1162	# define eio__mlockall(a) ((errno = ENOSYS), -1)
		1163	#else
		1164
		1165	static int
		1166	eio__mlockall (int flags)
		1167	{
		1168	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1169	extern int mallopt (int, int);
		1170	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1171	#endif
		1172
		1173	if (EIO_MCL_CURRENT != MCL_CURRENT
		1174	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1175	{
		1176	flags = 0
		1177	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1178	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1179	}
		1180
		1181	return mlockall (flags);
		1182	}
		1183	#endif
		1184
		1185	#if !_POSIX_MEMLOCK_RANGE
		1186	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
		1187	#else
		1188
		1189	static int
		1190	eio__mlock (void *addr, size_t length)
		1191	{
		1192	eio_page_align (&addr, &length);
		1193
		1194	return mlock (addr, length);
		1195	}
		1196
		1197	#endif
		1198
		1199	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1200	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
		1201	#else
		1202
		1203	static int
		1204	eio__msync (void *mem, size_t len, int flags)
		1205	{
		1206	eio_page_align (&mem, &len);
		1207
		1208	if (EIO_MS_ASYNC != MS_SYNC
		1209	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1210	|| EIO_MS_SYNC != MS_SYNC)
		1211	{
		1212	flags = 0
		1213	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1214	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1215	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1216	}
		1217
		1218	return msync (mem, len, flags);
		1219	}
		1220
		1221	#endif
		1222
		1223	static int
		1224	eio__mtouch (eio_req *req)
		1225	{
		1226	void *mem = req->ptr2;
		1227	size_t len = req->size;
		1228	int flags = req->int1;
		1229
		1230	eio_page_align (&mem, &len);
		1231
		1232	{
		1233	intptr_t addr = (intptr_t)mem;
		1234	intptr_t end = addr + len;
		1235	intptr_t page = eio_pagesize ();
		1236
		1237	if (addr < end)
		1238	if (flags & EIO_MT_MODIFY) /* modify */
		1239	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1240	else
		1241	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1242	}
		1243
		1244	return 0;
		1245	}
		1246
		1247	/*****************************************************************************/
		1248	/* requests implemented outside eio_execute, because they are so large */
		1249
		1250	static void
		1251	eio__realpath (eio_req *req, etp_worker *self)
		1252	{
		1253	char *rel = req->ptr1;
		1254	char *res;
		1255	char *tmp1, *tmp2;
		1256	#if SYMLOOP_MAX > 32
		1257	int symlinks = SYMLOOP_MAX;
		1258	#else
		1259	int symlinks = 32;
		1260	#endif
		1261
		1262	req->result = -1;
		1263
		1264	errno = EINVAL;
		1265	if (!rel)
		1266	return;
		1267
		1268	errno = ENOENT;
		1269	if (!*rel)
		1270	return;
		1271
		1272	if (!req->ptr2)
		1273	{
		1274	X_LOCK (wrklock);
		1275	req->flags |= EIO_FLAG_PTR2_FREE;
		1276	X_UNLOCK (wrklock);
		1277	req->ptr2 = malloc (PATH_MAX * 3);
		1278
		1279	errno = ENOMEM;
		1280	if (!req->ptr2)
		1281	return;
		1282	}
		1283
		1284	res = req->ptr2;
		1285	tmp1 = res + PATH_MAX;
		1286	tmp2 = tmp1 + PATH_MAX;
		1287
		1288	#if 0 /* disabled, the musl way to do things is just too racy */
		1289	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1290	/* on linux we may be able to ask the kernel */
		1291	{
		1292	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1293
		1294	if (fd >= 0)
		1295	{
		1296	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1297	req->result = readlink (tmp1, res, PATH_MAX);
		1298	close (fd);
		1299
		1300	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1301
		1302	if (req->result > 0)
		1303	goto done;
		1304	}
		1305	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1306	return;
		1307	}
		1308	#endif
		1309	#endif
		1310
		1311	if (*rel != '/')
		1312	{
		1313	if (!getcwd (res, PATH_MAX))
		1314	return;
		1315
		1316	if (res [1]) /* only use if not / */
		1317	res += strlen (res);
		1318	}
		1319
		1320	while (*rel)
		1321	{
		1322	ssize_t len, linklen;
		1323	char *beg = rel;
		1324
		1325	while (*rel && *rel != '/')
		1326	++rel;
		1327
		1328	len = rel - beg;
		1329
		1330	if (!len) /* skip slashes */
		1331	{
		1332	++rel;
		1333	continue;
		1334	}
		1335
		1336	if (beg [0] == '.')
		1337	{
		1338	if (len == 1)
		1339	continue; /* . - nop */
		1340
		1341	if (beg [1] == '.' && len == 2)
		1342	{
		1343	/* .. - back up one component, if possible */
		1344
		1345	while (res != req->ptr2)
		1346	if (*--res == '/')
		1347	break;
		1348
		1349	continue;
		1350	}
		1351	}
		1352
		1353	errno = ENAMETOOLONG;
		1354	if (res + 1 + len + 1 >= tmp1)
		1355	return;
		1356
		1357	/* copy one component */
		1358	*res = '/';
		1359	memcpy (res + 1, beg, len);
		1360
		1361	/* zero-terminate, for readlink */
		1362	res [len + 1] = 0;
		1363
		1364	/* now check if it's a symlink */
		1365	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1366
		1367	if (linklen < 0)
		1368	{
		1369	if (errno != EINVAL)
		1370	return;
		1371
		1372	/* it's a normal directory. hopefully */
		1373	res += len + 1;
		1374	}
		1375	else
		1376	{
		1377	/* yay, it was a symlink - build new path in tmp2 */
		1378	int rellen = strlen (rel);
		1379
		1380	errno = ENAMETOOLONG;
		1381	if (linklen + 1 + rellen >= PATH_MAX)
		1382	return;
		1383
		1384	errno = ELOOP;
		1385	if (!--symlinks)
		1386	return;
		1387
		1388	if (*tmp1 == '/')
		1389	res = req->ptr2; /* symlink resolves to an absolute path */
		1390
		1391	/* we need to be careful, as rel might point into tmp2 already */
		1392	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1393	tmp2 [linklen] = '/';
		1394	memcpy (tmp2, tmp1, linklen);
		1395
		1396	rel = tmp2;
		1397	}
		1398	}
		1399
		1400	/* special case for the lone root path */
		1401	if (res == req->ptr2)
		1402	*res++ = '/';
		1403
		1404	req->result = res - (char *)req->ptr2;
		1405
		1406	done:
		1407	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1408	}
		1409
		1410	static signed char
		1411	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1412	{
		1413	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1414	: a->inode < b->inode ? -1
		1415	: a->inode > b->inode ? 1
		1416	: 0;
		1417	}
		1418
		1419	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1420
		1421	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1422	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1423
		1424	static void
		1425	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1426	{
		1427	unsigned char bits [9 + sizeof (ino_t) * 8];
		1428	unsigned char *bit = bits;
		1429
		1430	assert (CHAR_BIT == 8);
		1431	assert (sizeof (eio_dirent) * 8 < 256);
		1432	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1433	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1434
		1435	if (size <= EIO_SORT_FAST)
		1436	return;
		1437
		1438	/* first prepare an array of bits to test in our radix sort */
		1439	/* try to take endianness into account, as well as differences in ino_t sizes */
		1440	/* inode_bits must contain all inodes ORed together */
		1441	/* which is used to skip bits that are 0 everywhere, which is very common */
		1442	{
		1443	ino_t endianness;
		1444	int i, j;
		1445
		1446	/* we store the byte offset of byte n into byte n of "endianness" */
		1447	for (i = 0; i < sizeof (ino_t); ++i)
		1448	((unsigned char *)&endianness)[i] = i;
		1449
		1450	*bit++ = 0;
		1451
		1452	for (i = 0; i < sizeof (ino_t); ++i)
		1453	{
		1454	/* shifting off the byte offsets out of "endianness" */
		1455	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1456	endianness >>= 8;
		1457
		1458	for (j = 0; j < 8; ++j)
		1459	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1460	*bit++ = offs + j;
		1461	}
		1462
		1463	for (j = 0; j < 8; ++j)
		1464	if (score_bits & (1 << j))
		1465	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1466	}
		1467
		1468	/* now actually do the sorting (a variant of MSD radix sort) */
		1469	{
		1470	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1471	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1472	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1473	int stk_idx = 0;
		1474
		1475	base_stk [stk_idx] = dents;
		1476	end_stk [stk_idx] = dents + size;
		1477	bit_stk [stk_idx] = bit - 1;
		1478
		1479	do
		1480	{
		1481	base = base_stk [stk_idx];
		1482	end = end_stk [stk_idx];
		1483	bit = bit_stk [stk_idx];
		1484
		1485	for (;;)
		1486	{
		1487	unsigned char O = *bit >> 3;
		1488	unsigned char M = 1 << (*bit & 7);
		1489
		1490	eio_dirent *a = base;
		1491	eio_dirent *b = end;
		1492
		1493	if (b - a < EIO_SORT_CUTOFF)
		1494	break;
		1495
		1496	/* now bit-partition the array on the bit */
		1497	/* this ugly asymmetric loop seems to perform much better than typical */
		1498	/* partition algos found in the literature */
		1499	do
		1500	if (!(((unsigned char *)a)[O] & M))
		1501	++a;
		1502	else if (!(((unsigned char *)--b)[O] & M))
		1503	{
		1504	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1505	++a;
		1506	}
		1507	while (b > a);
		1508
		1509	/* next bit, or stop, if no bits left in this path */
		1510	if (!*--bit)
		1511	break;
		1512
		1513	base_stk [stk_idx] = a;
		1514	end_stk [stk_idx] = end;
		1515	bit_stk [stk_idx] = bit;
		1516	++stk_idx;
		1517
		1518	end = a;
		1519	}
		1520	}
		1521	while (stk_idx--);
		1522	}
		1523	}
		1524
		1525	static void
		1526	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1527	{
		1528	/* first move the smallest element to the front, to act as a sentinel */
		1529	{
		1530	int i;
		1531	eio_dirent *min = dents;
		1532
		1533	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1534	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1535	if (EIO_DENT_CMP (dents [i], <, *min))
		1536	min = &dents [i];
		1537
		1538	/* swap elements 0 and j (minimum) */
		1539	{
		1540	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1541	}
		1542	}
		1543
		1544	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1545	{
		1546	eio_dirent *i, *j;
		1547
		1548	for (i = dents + 1; i < dents + size; ++i)
		1549	{
		1550	eio_dirent value = *i;
		1551
		1552	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1553	j [1] = j [0];
		1554
		1555	j [1] = value;
		1556	}
		1557	}
		1558	}
		1559
		1560	static void
		1561	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1562	{
		1563	if (size <= 1)
		1564	return; /* our insertion sort relies on size > 0 */
		1565
		1566	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1567	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1568	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1569
		1570	/* use an insertion sort at the end, or for small arrays, */
		1571	/* as insertion sort is more efficient for small partitions */
		1572	eio_dent_insertion_sort (dents, size);
		1573	}
		1574
709	/* read a full directory */	1575	/* read a full directory */
		1576	static void
710	static void scandir_ (eio_req *req, worker *self)	1577	eio__scandir (eio_req *req, etp_worker *self)
711	{	1578	{
712	DIR *dirp;	1579	DIR *dirp;
713	union
714	{
715	EIO_STRUCT_DIRENT d;
716	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
717	} *u;
718	EIO_STRUCT_DIRENT *entp;	1580	EIO_STRUCT_DIRENT *entp;
719	char *name, *names;	1581	char *name, *names;
720	int memlen = 4096;	1582	int namesalloc = 4096;
721	int memofs = 0;	1583	int namesoffs = 0;
		1584	int flags = req->int1;
		1585	eio_dirent *dents = 0;
		1586	int dentalloc = 128;
722	int res = 0;	1587	int dentoffs = 0;
		1588	ino_t inode_bits = 0;
		1589
		1590	req->result = -1;
		1591
		1592	if (!(flags & EIO_READDIR_DENTS))
		1593	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
723		1594
724	X_LOCK (wrklock);	1595	X_LOCK (wrklock);
		1596	/* the corresponding closedir is in ETP_WORKER_CLEAR */
725	self->dirp = dirp = opendir (req->ptr1);	1597	self->dirp = dirp = opendir (req->ptr1);
726	self->dbuf = u = malloc (sizeof (*u));	1598
727	req->flags |= EIO_FLAG_PTR2_FREE;	1599	if (req->flags & EIO_FLAG_PTR1_FREE)
		1600	free (req->ptr1);
		1601
		1602	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1603	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
728	req->ptr2 = names = malloc (memlen);	1604	req->ptr2 = names = malloc (namesalloc);
729	X_UNLOCK (wrklock);	1605	X_UNLOCK (wrklock);
730		1606
731	if (dirp && u && names)	1607	if (dirp && names && (!flags || dents))
732	for (;;)	1608	for (;;)
733	{	1609	{
734	errno = 0;	1610	errno = 0;
735	readdir_r (dirp, &u->d, &entp);	1611	entp = readdir (dirp);
736		1612
737	if (!entp)	1613	if (!entp)
		1614	{
		1615	if (errno)
		1616	break;
		1617
		1618	/* sort etc. */
		1619	req->int1 = flags;
		1620	req->result = dentoffs;
		1621
		1622	if (flags & EIO_READDIR_STAT_ORDER)
		1623	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1624	else if (flags & EIO_READDIR_DIRS_FIRST)
		1625	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1626	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1627	else
		1628	{
		1629	/* in this case, all is known, and we just put dirs first and sort them */
		1630	eio_dirent *oth = dents + dentoffs;
		1631	eio_dirent *dir = dents;
		1632
		1633	/* now partition dirs to the front, and non-dirs to the back */
		1634	/* by walking from both sides and swapping if necessary */
		1635	while (oth > dir)
		1636	{
		1637	if (dir->type == EIO_DT_DIR)
		1638	++dir;
		1639	else if ((--oth)->type == EIO_DT_DIR)
		1640	{
		1641	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1642
		1643	++dir;
		1644	}
		1645	}
		1646
		1647	/* now sort the dirs only (dirs all have the same score) */
		1648	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1649	}
		1650
738	break;	1651	break;
		1652	}
739		1653
		1654	/* now add the entry to our list(s) */
740	name = entp->d_name;	1655	name = entp->d_name;
741		1656
		1657	/* skip . and .. entries */
742	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1658	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
743	{	1659	{
744	int len = strlen (name) + 1;	1660	int len = D_NAMLEN (entp) + 1;
745		1661
746	res++;	1662	while (ecb_expect_false (namesoffs + len > namesalloc))
747
748	while (memofs + len > memlen)
749	{	1663	{
750	memlen *= 2;	1664	namesalloc *= 2;
751	X_LOCK (wrklock);	1665	X_LOCK (wrklock);
752	req->ptr2 = names = realloc (names, memlen);	1666	req->ptr2 = names = realloc (names, namesalloc);
753	X_UNLOCK (wrklock);	1667	X_UNLOCK (wrklock);
754		1668
755	if (!names)	1669	if (!names)
756	break;	1670	break;
757	}	1671	}
758		1672
759	memcpy (names + memofs, name, len);	1673	memcpy (names + namesoffs, name, len);
		1674
		1675	if (dents)
		1676	{
		1677	struct eio_dirent *ent;
		1678
		1679	if (ecb_expect_false (dentoffs == dentalloc))
		1680	{
		1681	dentalloc *= 2;
		1682	X_LOCK (wrklock);
		1683	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1684	X_UNLOCK (wrklock);
		1685
		1686	if (!dents)
		1687	break;
		1688	}
		1689
		1690	ent = dents + dentoffs;
		1691
		1692	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1693	ent->namelen = len - 1;
		1694	ent->inode = D_INO (entp);
		1695
		1696	inode_bits |= ent->inode;
		1697
		1698	switch (D_TYPE (entp))
		1699	{
		1700	default:
		1701	ent->type = EIO_DT_UNKNOWN;
		1702	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1703	break;
		1704
		1705	#ifdef DT_FIFO
		1706	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1707	#endif
		1708	#ifdef DT_CHR
		1709	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1710	#endif
		1711	#ifdef DT_MPC
		1712	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1713	#endif
		1714	#ifdef DT_DIR
		1715	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1716	#endif
		1717	#ifdef DT_NAM
		1718	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1719	#endif
		1720	#ifdef DT_BLK
		1721	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1722	#endif
		1723	#ifdef DT_MPB
		1724	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1725	#endif
		1726	#ifdef DT_REG
		1727	case DT_REG: ent->type = EIO_DT_REG; break;
		1728	#endif
		1729	#ifdef DT_NWK
		1730	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1731	#endif
		1732	#ifdef DT_CMP
		1733	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1734	#endif
		1735	#ifdef DT_LNK
		1736	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1737	#endif
		1738	#ifdef DT_SOCK
		1739	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1740	#endif
		1741	#ifdef DT_DOOR
		1742	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1743	#endif
		1744	#ifdef DT_WHT
		1745	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1746	#endif
		1747	}
		1748
		1749	ent->score = 7;
		1750
		1751	if (flags & EIO_READDIR_DIRS_FIRST)
		1752	{
		1753	if (ent->type == EIO_DT_UNKNOWN)
		1754	{
		1755	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1756	ent->score = 1;
		1757	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1758	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1759	}
		1760	else if (ent->type == EIO_DT_DIR)
		1761	ent->score = 0;
		1762	}
		1763	}
		1764
760	memofs += len;	1765	namesoffs += len;
		1766	++dentoffs;
		1767	}
		1768
		1769	if (EIO_CANCELLED (req))
		1770	{
		1771	errno = ECANCELED;
		1772	break;
761	}	1773	}
762	}	1774	}
763
764	if (errno)
765	res = -1;
766
767	req->result = res;
768	}	1775	}
769		1776
770	/*****************************************************************************/	1777	/*****************************************************************************/
771		1778
772	#define ALLOC(len) \	1779	#define ALLOC(len) \
…		…
782	req->result = -1; \	1789	req->result = -1; \
783	break; \	1790	break; \
784	} \	1791	} \
785	}	1792	}
786		1793
787	X_THREAD_PROC (eio_proc)	1794	X_THREAD_PROC (etp_proc)
788	{	1795	{
789	eio_req *req;	1796	ETP_REQ *req;
790	struct timespec ts;	1797	struct timespec ts;
791	worker *self = (worker *)thr_arg;	1798	etp_worker *self = (etp_worker *)thr_arg;
		1799	int timeout;
792		1800
793	/* try to distribute timeouts somewhat randomly */	1801	/* try to distribute timeouts somewhat evenly */
794	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1802	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
795		1803
796	for (;;)	1804	for (;;)
797	{	1805	{
798	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1806	ts.tv_sec = 0;
799		1807
800	X_LOCK (reqlock);	1808	X_LOCK (reqlock);
801		1809
802	for (;;)	1810	for (;;)
803	{	1811	{
804	self->req = req = reqq_shift (&req_queue);	1812	self->req = req = reqq_shift (&req_queue);
805		1813
806	if (req)	1814	if (req)
807	break;	1815	break;
808		1816
		1817	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1818	{
		1819	X_UNLOCK (reqlock);
		1820	X_LOCK (wrklock);
		1821	--started;
		1822	X_UNLOCK (wrklock);
		1823	goto quit;
		1824	}
		1825
809	++idle;	1826	++idle;
810		1827
811	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1828	if (idle <= max_idle)
		1829	/* we are allowed to idle, so do so without any timeout */
		1830	X_COND_WAIT (reqwait, reqlock);
		1831	else
812	{	1832	{
813	if (idle > max_idle)	1833	/* initialise timeout once */
814	{	1834	if (!ts.tv_sec)
815	--idle;	1835	ts.tv_sec = time (0) + idle_timeout;
816	X_UNLOCK (reqlock);
817	X_LOCK (wrklock);
818	--started;
819	X_UNLOCK (wrklock);
820	goto quit;
821	}
822		1836
823	/* we are allowed to idle, so do so without any timeout */
824	X_COND_WAIT (reqwait, reqlock);	1837	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
825	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1838	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
826	}	1839	}
827		1840
828	--idle;	1841	--idle;
829	}	1842	}
830		1843
831	--nready;	1844	--nready;
832		1845
833	X_UNLOCK (reqlock);	1846	X_UNLOCK (reqlock);
834		1847
835	errno = 0; /* strictly unnecessary */
836
837	if (!EIO_CANCELLED (req))
838	switch (req->type)	1848	if (req->type < 0)
839	{
840	case EIO_READ: ALLOC (req->size);
841	req->result = req->offs >= 0
842	? pread (req->int1, req->ptr2, req->size, req->offs)
843	: read (req->int1, req->ptr2, req->size); break;
844	case EIO_WRITE: req->result = req->offs >= 0
845	? pwrite (req->int1, req->ptr2, req->size, req->offs)
846	: write (req->int1, req->ptr2, req->size); break;
847
848	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
849	case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
850
851	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
852	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
853	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
854	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
855	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
856	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
857
858	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
859	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
860	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
861	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
862	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
863	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
864
865	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
866	case EIO_CLOSE: req->result = close (req->int1); break;
867	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
868	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
869	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
870	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
871	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
872	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
873	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
874	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
875
876	case EIO_READLINK: ALLOC (NAME_MAX);
877	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
878
879	case EIO_SYNC: req->result = 0; sync (); break;
880	case EIO_FSYNC: req->result = fsync (req->int1); break;
881	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
882
883	case EIO_READDIR: scandir_ (req, self); break;
884
885	case EIO_BUSY:
886	#ifdef _WIN32
887	Sleep (req->nv1 * 1000.);
888	#else
889	{
890	struct timeval tv;
891
892	tv.tv_sec = req->nv1;
893	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
894
895	req->result = select (0, 0, 0, 0, &tv);
896	}
897	#endif
898	break;
899
900	case EIO_UTIME:
901	case EIO_FUTIME:
902	{
903	struct timeval tv[2];
904	struct timeval *times;
905
906	if (req->nv1 != -1. || req->nv2 != -1.)
907	{
908	tv[0].tv_sec = req->nv1;
909	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
910	tv[1].tv_sec = req->nv2;
911	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
912
913	times = tv;
914	}
915	else
916	times = 0;
917
918
919	req->result = req->type == EIO_FUTIME
920	? futimes (req->int1, times)
921	: utimes (req->ptr1, times);
922	}
923
924	case EIO_GROUP:
925	case EIO_NOP:
926	break;
927
928	case EIO_QUIT:
929	goto quit;	1849	goto quit;
930		1850
931	default:	1851	ETP_EXECUTE (self, req);
932	req->result = -1;
933	break;
934	}
935
936	req->errorno = errno;
937		1852
938	X_LOCK (reslock);	1853	X_LOCK (reslock);
939		1854
940	++npending;	1855	++npending;
941		1856
942	if (!reqq_push (&res_queue, req) && want_poll_cb)	1857	if (!reqq_push (&res_queue, req) && want_poll_cb)
943	want_poll_cb ();	1858	want_poll_cb ();
944		1859
945	self->req = 0;	1860	self->req = 0;
946	worker_clear (self);	1861	etp_worker_clear (self);
947		1862
948	X_UNLOCK (reslock);	1863	X_UNLOCK (reslock);
949	}	1864	}
950		1865
951	quit:	1866	quit:
952	X_LOCK (wrklock);	1867	X_LOCK (wrklock);
953	worker_free (self);	1868	etp_worker_free (self);
954	X_UNLOCK (wrklock);	1869	X_UNLOCK (wrklock);
955		1870
956	return 0;	1871	return 0;
957	}	1872	}
958		1873
959	/*****************************************************************************/	1874	/*****************************************************************************/
960		1875
961	static void eio_atfork_prepare (void)	1876	int ecb_cold
962	{
963	X_LOCK (wrklock);
964	X_LOCK (reqlock);
965	X_LOCK (reslock);
966	#if !HAVE_PREADWRITE
967	X_LOCK (preadwritelock);
968	#endif
969	#if !HAVE_READDIR_R
970	X_LOCK (readdirlock);
971	#endif
972	}
973
974	static void eio_atfork_parent (void)
975	{
976	#if !HAVE_READDIR_R
977	X_UNLOCK (readdirlock);
978	#endif
979	#if !HAVE_PREADWRITE
980	X_UNLOCK (preadwritelock);
981	#endif
982	X_UNLOCK (reslock);
983	X_UNLOCK (reqlock);
984	X_UNLOCK (wrklock);
985	}
986
987	static void eio_atfork_child (void)
988	{
989	eio_req *prv;
990
991	while (prv = reqq_shift (&req_queue))
992	eio_destroy (prv);
993
994	while (prv = reqq_shift (&res_queue))
995	eio_destroy (prv);
996
997	while (wrk_first.next != &wrk_first)
998	{
999	worker *wrk = wrk_first.next;
1000
1001	if (wrk->req)
1002	eio_destroy (wrk->req);
1003
1004	worker_clear (wrk);
1005	worker_free (wrk);
1006	}
1007
1008	started = 0;
1009	idle = 0;
1010	nreqs = 0;
1011	nready = 0;
1012	npending = 0;
1013
1014	eio_atfork_parent ();
1015	}
1016
1017	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1877	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1018	{	1878	{
1019	want_poll_cb = want_poll;	1879	return etp_init (want_poll, done_poll);
1020	done_poll_cb = done_poll;
1021
1022	#ifdef _WIN32
1023	X_MUTEX_CHECK (wrklock);
1024	X_MUTEX_CHECK (reslock);
1025	X_MUTEX_CHECK (reqlock);
1026	X_MUTEX_CHECK (reqwait);
1027	X_MUTEX_CHECK (preadwritelock);
1028	X_MUTEX_CHECK (readdirlock);
1029
1030	X_COND_CHECK (reqwait);
1031	#endif
1032
1033	X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1034	}	1880	}
1035		1881
		1882	ecb_inline void
1036	static void eio_api_destroy (eio_req *req)	1883	eio_api_destroy (eio_req *req)
1037	{	1884	{
1038	free (req);	1885	free (req);
1039	}	1886	}
1040		1887
1041	#define REQ(rtype) \	1888	#define REQ(rtype) \
…		…
1043	\	1890	\
1044	req = (eio_req *)calloc (1, sizeof *req); \	1891	req = (eio_req *)calloc (1, sizeof *req); \
1045	if (!req) \	1892	if (!req) \
1046	return 0; \	1893	return 0; \
1047	\	1894	\
1048	req->type = EIO_ ## rtype; \	1895	req->type = rtype; \
1049	req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \	1896	req->pri = pri; \
1050	req->finish = cb; \	1897	req->finish = cb; \
		1898	req->data = data; \
1051	req->destroy = eio_api_destroy;	1899	req->destroy = eio_api_destroy;
1052		1900
1053	#define SEND eio_submit (req); return req	1901	#define SEND eio_submit (req); return req
1054		1902
1055	#define PATH (void)0	1903	#define PATH \
		1904	req->flags |= EIO_FLAG_PTR1_FREE; \
		1905	req->ptr1 = strdup (path); \
		1906	if (!req->ptr1) \
		1907	{ \
		1908	eio_api_destroy (req); \
		1909	return 0; \
		1910	}
1056		1911
		1912	static void
		1913	eio_execute (etp_worker *self, eio_req *req)
		1914	{
		1915	if (ecb_expect_false (EIO_CANCELLED (req)))
		1916	{
		1917	req->result = -1;
		1918	req->errorno = ECANCELED;
		1919	return;
		1920	}
		1921
		1922	switch (req->type)
		1923	{
		1924	case EIO_READ: ALLOC (req->size);
		1925	req->result = req->offs >= 0
		1926	? pread (req->int1, req->ptr2, req->size, req->offs)
		1927	: read (req->int1, req->ptr2, req->size); break;
		1928	case EIO_WRITE: req->result = req->offs >= 0
		1929	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1930	: write (req->int1, req->ptr2, req->size); break;
		1931
		1932	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		1933	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
		1934
		1935	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1936	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1937	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1938	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1939	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1940	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1941
		1942	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1943	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1944	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1945	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1946
		1947	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1948	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1949	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1950	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1951	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1952	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1953
		1954	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1955	case EIO_CLOSE: req->result = close (req->int1); break;
		1956	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1957	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1958	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1959	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1960	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1961	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1962	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1963	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		1964
		1965	case EIO_REALPATH: eio__realpath (req, self); break;
		1966
		1967	case EIO_READLINK: ALLOC (PATH_MAX);
		1968	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
		1969
		1970	case EIO_SYNC: req->result = 0; sync (); break;
		1971	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1972	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1973	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1974	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		1975	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		1976	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		1977	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1978	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
		1979
		1980	case EIO_READDIR: eio__scandir (req, self); break;
		1981
		1982	case EIO_BUSY:
		1983	#ifdef _WIN32
		1984	Sleep (req->nv1 * 1e3);
		1985	#else
		1986	{
		1987	struct timeval tv;
		1988
		1989	tv.tv_sec = req->nv1;
		1990	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
		1991
		1992	req->result = select (0, 0, 0, 0, &tv);
		1993	}
		1994	#endif
		1995	break;
		1996
		1997	case EIO_UTIME:
		1998	case EIO_FUTIME:
		1999	{
		2000	struct timeval tv[2];
		2001	struct timeval *times;
		2002
		2003	if (req->nv1 != -1. || req->nv2 != -1.)
		2004	{
		2005	tv[0].tv_sec = req->nv1;
		2006	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		2007	tv[1].tv_sec = req->nv2;
		2008	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		2009
		2010	times = tv;
		2011	}
		2012	else
		2013	times = 0;
		2014
		2015	req->result = req->type == EIO_FUTIME
		2016	? futimes (req->int1, times)
		2017	: utimes (req->ptr1, times);
		2018	}
		2019	break;
		2020
		2021	case EIO_GROUP:
		2022	abort (); /* handled in eio_request */
		2023
		2024	case EIO_NOP:
		2025	req->result = 0;
		2026	break;
		2027
		2028	case EIO_CUSTOM:
		2029	req->feed (req);
		2030	break;
		2031
		2032	default:
		2033	errno = ENOSYS;
		2034	req->result = -1;
		2035	break;
		2036	}
		2037
		2038	req->errorno = errno;
		2039	}
		2040
		2041	#ifndef EIO_NO_WRAPPERS
		2042
		2043	eio_req *eio_nop (int pri, eio_cb cb, void *data)
		2044	{
		2045	REQ (EIO_NOP); SEND;
		2046	}
		2047
		2048	eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
		2049	{
		2050	REQ (EIO_BUSY); req->nv1 = delay; SEND;
		2051	}
		2052
1057	eio_req *eio_fsync (int fd, eio_cb cb)	2053	eio_req *eio_sync (int pri, eio_cb cb, void *data)
1058	{	2054	{
		2055	REQ (EIO_SYNC); SEND;
		2056	}
		2057
		2058	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
		2059	{
1059	REQ (FSYNC); req->int1 = fd; SEND;	2060	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1060	}	2061	}
1061		2062
		2063	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2064	{
		2065	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2066	}
		2067
		2068	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2069	{
		2070	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2071	}
		2072
		2073	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2074	{
		2075	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2076	}
		2077
		2078	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2079	{
		2080	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2081	}
		2082
		2083	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2084	{
		2085	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2086	}
		2087
		2088	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2089	{
		2090	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2091	}
		2092
1062	eio_req *eio_fdatasync (int fd, eio_cb cb)	2093	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1063	{	2094	{
1064	REQ (FDATASYNC); req->int1 = fd; SEND;	2095	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1065	}	2096	}
1066		2097
		2098	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
		2099	{
		2100	REQ (EIO_CLOSE); req->int1 = fd; SEND;
		2101	}
		2102
1067	eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb)	2103	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1068	{	2104	{
1069	REQ (READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;	2105	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1070	}	2106	}
1071		2107
1072	eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb)	2108	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1073	{	2109	{
1074	REQ (READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND;	2110	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1075	}	2111	}
1076		2112
1077	eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb)	2113	eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1078	{	2114	{
1079	REQ (WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND;	2115	REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1080	}	2116	}
1081		2117
1082	eio_req *eio_fstat (int fd, eio_cb cb)	2118	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1083	{	2119	{
1084	REQ (FSTAT); req->int1 = fd; SEND;	2120	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1085	}	2121	}
1086		2122
		2123	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2124	{
		2125	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2126	}
		2127
1087	eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb)	2128	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1088	{	2129	{
1089	REQ (FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2130	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1090	}	2131	}
1091		2132
1092	eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb)	2133	eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1093	{	2134	{
1094	REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;	2135	REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1095	}	2136	}
1096		2137
1097	eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb)	2138	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1098	{	2139	{
1099	REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2140	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1100	}	2141	}
1101		2142
1102	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb)	2143	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1103	{	2144	{
1104	REQ (FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2145	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1105	}	2146	}
1106		2147
1107	eio_req *eio_dup2 (int fd, int fd2, eio_cb cb)	2148	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1108	{	2149	{
1109	REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND;	2150	REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1110	}	2151	}
1111		2152
1112	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb)	2153	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1113	{	2154	{
1114	REQ (SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;	2155	REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1115	}	2156	}
1116		2157
1117	eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb)	2158	eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1118	{	2159	{
1119	REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;	2160	REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1120	}	2161	}
1121		2162
1122	eio_req *eio_readlink (const char *path, eio_cb cb)
1123	{
1124	REQ (READLINK); PATH; SEND;
1125	}
1126
1127	eio_req *eio_stat (const char *path, eio_cb cb)
1128	{
1129	REQ (STAT); PATH; SEND;
1130	}
1131
1132	eio_req *eio_lstat (const char *path, eio_cb cb)
1133	{
1134	REQ (LSTAT); PATH; SEND;
1135	}
1136
1137	eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb)	2163	eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1138	{	2164	{
1139	REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;	2165	REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1140	}	2166	}
1141		2167
1142	eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb)	2168	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1143	{	2169	{
1144	REQ (TRUNCATE); PATH; req->offs = offset; SEND;	2170	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1145	}	2171	}
1146		2172
1147	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb)	2173	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1148	{	2174	{
1149	REQ (CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2175	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1150	}	2176	}
1151		2177
1152	eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb)	2178	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1153	{	2179	{
1154	REQ (CHMOD); PATH; req->int2 = (long)mode; SEND;	2180	REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1155	}	2181	}
1156		2182
1157	eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb)	2183	eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1158	{	2184	{
1159	REQ (MKDIR); PATH; req->int2 = (long)mode; SEND;	2185	REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1160	}	2186	}
1161		2187
1162	eio_req *eio_unlink (const char *path, eio_cb cb)	2188	static eio_req *
		2189	eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1163	{	2190	{
1164	REQ (UNLINK); PATH; SEND;
1165	}
1166
1167	eio_req *eio_rmdir (const char *path, eio_cb cb)
1168	{
1169	REQ (RMDIR); PATH; SEND;	2191	REQ (type); PATH; SEND;
1170	}	2192	}
1171		2193
		2194	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
		2195	{
		2196	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2197	}
		2198
		2199	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2200	{
		2201	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2202	}
		2203
		2204	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
		2205	{
		2206	return eio__1path (EIO_STAT, path, pri, cb, data);
		2207	}
		2208
		2209	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
		2210	{
		2211	return eio__1path (EIO_LSTAT, path, pri, cb, data);
		2212	}
		2213
		2214	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2215	{
		2216	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2217	}
		2218
		2219	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
		2220	{
		2221	return eio__1path (EIO_UNLINK, path, pri, cb, data);
		2222	}
		2223
1172	eio_req *eio_readdir (const char *path, eio_cb cb)	2224	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1173	{	2225	{
1174	REQ (READDIR); PATH; SEND;	2226	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1175	}	2227	}
1176		2228
		2229	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
		2230	{
		2231	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
		2232	}
		2233
1177	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb)	2234	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1178	{	2235	{
1179	REQ (MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	2236	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1180	}	2237	}
1181		2238
1182	eio_req *eio_busy (double delay, eio_cb cb)	2239	static eio_req *
		2240	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1183	{	2241	{
1184	REQ (BUSY); req->nv1 = delay; SEND;	2242	REQ (type); PATH;
1185	}
1186		2243
		2244	req->flags |= EIO_FLAG_PTR2_FREE;
		2245	req->ptr2 = strdup (new_path);
		2246	if (!req->ptr2)
		2247	{
		2248	eio_api_destroy (req);
		2249	return 0;
		2250	}
		2251
		2252	SEND;
		2253	}
		2254
		2255	eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2256	{
		2257	return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
		2258	}
		2259
		2260	eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2261	{
		2262	return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
		2263	}
		2264
		2265	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2266	{
		2267	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
		2268	}
		2269
		2270	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
		2271	{
		2272	REQ (EIO_CUSTOM); req->feed = execute; SEND;
		2273	}
		2274
		2275	#endif
		2276
1187	eio_req *eio_nop (eio_cb cb)	2277	eio_req *eio_grp (eio_cb cb, void *data)
1188	{	2278	{
		2279	const int pri = EIO_PRI_MAX;
		2280
1189	REQ (NOP); SEND;	2281	REQ (EIO_GROUP); SEND;
1190	}	2282	}
1191		2283
1192	#undef REQ	2284	#undef REQ
1193	#undef PATH	2285	#undef PATH
1194	#undef SEND	2286	#undef SEND
1195		2287
1196	#if 0	2288	/*****************************************************************************/
		2289	/* grp functions */
1197		2290
1198	void	2291	void
1199	aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1200	PROTOTYPE: $$$;$
1201	PPCODE:
1202	{
1203	dREQ;
1204
1205	req->type = EIO_OPEN;
1206	req->sv1 = newSVsv (pathname);
1207	req->ptr1 = SvPVbyte_nolen (req->sv1);
1208	req->int1 = flags;
1209	req->int2 = mode;
1210
1211	EIO_SEND;
1212	}
1213
1214	void
1215	aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1216	PROTOTYPE: $;$
1217	ALIAS:
1218	aio_fsync = EIO_FSYNC
1219	aio_fdatasync = EIO_FDATASYNC
1220	PPCODE:
1221	{
1222	dREQ;
1223
1224	req->type = ix;
1225	req->sv1 = newSVsv (fh);
1226	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1227
1228	EIO_SEND (req);
1229	}
1230
1231	void
1232	aio_close (SV *fh, SV *callback=&PL_sv_undef)
1233	PROTOTYPE: $;$
1234	PPCODE:
1235	{
1236	dREQ;
1237
1238	req->type = EIO_CLOSE;
1239	req->sv1 = newSVsv (fh);
1240	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1241
1242	EIO_SEND (req);
1243	}
1244
1245	void
1246	aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1247	ALIAS:
1248	aio_read = EIO_READ
1249	aio_write = EIO_WRITE
1250	PROTOTYPE: $$$$$;$
1251	PPCODE:
1252	{
1253	STRLEN svlen;
1254	char *svptr = SvPVbyte (data, svlen);
1255	UV len = SvUV (length);
1256
1257	SvUPGRADE (data, SVt_PV);
1258	SvPOK_on (data);
1259
1260	if (dataoffset < 0)
1261	dataoffset += svlen;
1262
1263	if (dataoffset < 0 || dataoffset > svlen)
1264	croak ("dataoffset outside of data scalar");
1265
1266	if (ix == EIO_WRITE)
1267	{
1268	/* write: check length and adjust. */
1269	if (!SvOK (length) || len + dataoffset > svlen)
1270	len = svlen - dataoffset;
1271	}
1272	else
1273	{
1274	/* read: grow scalar as necessary */
1275	svptr = SvGROW (data, len + dataoffset + 1);
1276	}
1277
1278	if (len < 0)
1279	croak ("length must not be negative");
1280
1281	{
1282	dREQ;
1283
1284	req->type = ix;
1285	req->sv1 = newSVsv (fh);
1286	req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1287	: IoOFP (sv_2io (fh)));
1288	req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1289	req->size = len;
1290	req->sv2 = SvREFCNT_inc (data);
1291	req->ptr2 = (char *)svptr + dataoffset;
1292	req->stroffset = dataoffset;
1293
1294	if (!SvREADONLY (data))
1295	{
1296	SvREADONLY_on (data);
1297	req->flags |= FLAG_SV2_RO_OFF;
1298	}
1299
1300	EIO_SEND;
1301	}
1302	}
1303
1304	void
1305	aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1306	PROTOTYPE: $$;$
1307	PPCODE:
1308	{
1309	SV *data;
1310	dREQ;
1311
1312	data = newSV (NAME_MAX);
1313	SvPOK_on (data);
1314
1315	req->type = EIO_READLINK;
1316	req->sv1 = newSVsv (path);
1317	req->ptr1 = SvPVbyte_nolen (req->sv1);
1318	req->sv2 = data;
1319	req->ptr2 = SvPVbyte_nolen (data);
1320
1321	EIO_SEND;
1322	}
1323
1324	void
1325	aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1326	PROTOTYPE: $$$$;$
1327	PPCODE:
1328	{
1329	dREQ;
1330
1331	req->type = EIO_SENDFILE;
1332	req->sv1 = newSVsv (out_fh);
1333	req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1334	req->sv2 = newSVsv (in_fh);
1335	req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1336	req->offs = SvVAL64 (in_offset);
1337	req->size = length;
1338
1339	EIO_SEND;
1340	}
1341
1342	void
1343	aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1344	PROTOTYPE: $$$;$
1345	PPCODE:
1346	{
1347	dREQ;
1348
1349	req->type = EIO_READAHEAD;
1350	req->sv1 = newSVsv (fh);
1351	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1352	req->offs = SvVAL64 (offset);
1353	req->size = length;
1354
1355	EIO_SEND;
1356	}
1357
1358	void
1359	aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1360	ALIAS:
1361	aio_stat = EIO_STAT
1362	aio_lstat = EIO_LSTAT
1363	PPCODE:
1364	{
1365	dREQ;
1366
1367	req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1368	if (!req->ptr2)
1369	{
1370	req_destroy (req);
1371	croak ("out of memory during aio_stat statdata allocation");
1372	}
1373
1374	req->flags |= FLAG_PTR2_FREE;
1375	req->sv1 = newSVsv (fh_or_path);
1376
1377	if (SvPOK (fh_or_path))
1378	{
1379	req->type = ix;
1380	req->ptr1 = SvPVbyte_nolen (req->sv1);
1381	}
1382	else
1383	{
1384	req->type = EIO_FSTAT;
1385	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1386	}
1387
1388	EIO_SEND;
1389	}
1390
1391	void
1392	aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1393	PPCODE:
1394	{
1395	dREQ;
1396
1397	req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1398	req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1399	req->sv1 = newSVsv (fh_or_path);
1400
1401	if (SvPOK (fh_or_path))
1402	{
1403	req->type = EIO_UTIME;
1404	req->ptr1 = SvPVbyte_nolen (req->sv1);
1405	}
1406	else
1407	{
1408	req->type = EIO_FUTIME;
1409	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1410	}
1411
1412	EIO_SEND;
1413	}
1414
1415	void
1416	aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1417	PPCODE:
1418	{
1419	dREQ;
1420
1421	req->sv1 = newSVsv (fh_or_path);
1422	req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1423
1424	if (SvPOK (fh_or_path))
1425	{
1426	req->type = EIO_TRUNCATE;
1427	req->ptr1 = SvPVbyte_nolen (req->sv1);
1428	}
1429	else
1430	{
1431	req->type = EIO_FTRUNCATE;
1432	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1433	}
1434
1435	EIO_SEND;
1436	}
1437
1438	void
1439	aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1440	ALIAS:
1441	aio_chmod = EIO_CHMOD
1442	aio_fchmod = EIO_FCHMOD
1443	aio_mkdir = EIO_MKDIR
1444	PPCODE:
1445	{
1446	dREQ;
1447
1448	req->type = type;
1449	req->int2 = mode;
1450	req->sv1 = newSVsv (fh_or_path);
1451
1452	if (ix == EIO_FCHMOD)
1453	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1454	else
1455	req->ptr1 = SvPVbyte_nolen (req->sv1);
1456
1457	EIO_SEND;
1458	}
1459
1460	void
1461	aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1462	PPCODE:
1463	{
1464	dREQ;
1465
1466	req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1467	req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1468	req->sv1 = newSVsv (fh_or_path);
1469
1470	if (SvPOK (fh_or_path))
1471	{
1472	req->type = EIO_CHOWN;
1473	req->ptr1 = SvPVbyte_nolen (req->sv1);
1474	}
1475	else
1476	{
1477	req->type = EIO_FCHOWN;
1478	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1479	}
1480
1481	EIO_SEND;
1482	}
1483
1484	void
1485	aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1486	ALIAS:
1487	aio_unlink = EIO_UNLINK
1488	aio_rmdir = EIO_RMDIR
1489	aio_readdir = EIO_READDIR
1490	PPCODE:
1491	{
1492	dREQ;
1493
1494	req->type = ix;
1495	req->sv1 = newSVsv (pathname);
1496	req->ptr1 = SvPVbyte_nolen (req->sv1);
1497
1498	EIO_SEND;
1499	}
1500
1501	void
1502	aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1503	ALIAS:
1504	aio_link = EIO_LINK
1505	aio_symlink = EIO_SYMLINK
1506	aio_rename = EIO_RENAME
1507	PPCODE:
1508	{
1509	dREQ;
1510
1511	req->type = ix;
1512	req->sv1 = newSVsv (oldpath);
1513	req->ptr1 = SvPVbyte_nolen (req->sv1);
1514	req->sv2 = newSVsv (newpath);
1515	req->ptr2 = SvPVbyte_nolen (req->sv2);
1516
1517	EIO_SEND;
1518	}
1519
1520	void
1521	aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1522	PPCODE:
1523	{
1524	dREQ;
1525
1526	req->type = EIO_MKNOD;
1527	req->sv1 = newSVsv (pathname);
1528	req->ptr1 = SvPVbyte_nolen (req->sv1);
1529	req->int2 = (mode_t)mode;
1530	req->offs = dev;
1531
1532	EIO_SEND;
1533	}
1534
1535	void
1536	aio_busy (double delay, SV *callback=&PL_sv_undef)
1537	PPCODE:
1538	{
1539	dREQ;
1540
1541	req->type = EIO_BUSY;
1542	req->nv1 = delay < 0. ? 0. : delay;
1543
1544	EIO_SEND;
1545	}
1546
1547	void
1548	aio_group (SV *callback=&PL_sv_undef)
1549	PROTOTYPE: ;$
1550	PPCODE:
1551	{
1552	dREQ;
1553
1554	req->type = EIO_GROUP;
1555
1556	req_send (req);
1557	XPUSHs (req_sv (req, AIO_GRP_KLASS));
1558	}
1559
1560	void
1561	aio_nop (SV *callback=&PL_sv_undef)
1562	ALIAS:
1563	aio_nop = EIO_NOP
1564	aio_sync = EIO_SYNC
1565	PPCODE:
1566	{
1567	dREQ;
1568
1569	#endif
1570
1571	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2292	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1572	{	2293	{
1573	grp->int2 = limit;	2294	grp->int2 = limit;
1574	grp->feed = feed;	2295	grp->feed = feed;
1575		2296
1576	grp_try_feed (grp);	2297	grp_try_feed (grp);
1577	}	2298	}
1578		2299
		2300	void
1579	void eio_grp_limit (eio_req *grp, int limit)	2301	eio_grp_limit (eio_req *grp, int limit)
1580	{	2302	{
1581	grp->int2 = limit;	2303	grp->int2 = limit;
1582		2304
1583	grp_try_feed (grp);	2305	grp_try_feed (grp);
1584	}	2306	}
1585		2307
		2308	void
1586	void eio_grp_add (eio_req *grp, eio_req *req)	2309	eio_grp_add (eio_req *grp, eio_req *req)
1587	{	2310	{
1588	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2311	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
		2312
		2313	grp->flags |= EIO_FLAG_GROUPADD;
1589		2314
1590	++grp->size;	2315	++grp->size;
1591	req->grp = grp;	2316	req->grp = grp;
1592		2317
1593	req->grp_prev = 0;	2318	req->grp_prev = 0;
…		…
1597	grp->grp_first->grp_prev = req;	2322	grp->grp_first->grp_prev = req;
1598		2323
1599	grp->grp_first = req;	2324	grp->grp_first = req;
1600	}	2325	}
1601		2326
		2327	/*****************************************************************************/
		2328	/* misc garbage */
1602		2329
		2330	ssize_t
		2331	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
		2332	{
		2333	etp_worker wrk;
		2334	ssize_t ret;
		2335
		2336	wrk.dbuf = 0;
		2337
		2338	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
		2339
		2340	if (wrk.dbuf)
		2341	free (wrk.dbuf);
		2342
		2343	return ret;
		2344	}
		2345

Diff Legend

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