ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.2 by root, Sat May 10 23:33:40 2008 UTC vs.
Revision 1.84 by root, Thu Jul 7 22:39:56 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
6	#include "EXTERN.h"
7	#include "perl.h"
8	#include "XSUB.h"
9
10	#include <stddef.h>	53	#include <stddef.h>
11	#include <stdlib.h>	54	#include <stdlib.h>
		55	#include <string.h>
12	#include <errno.h>	56	#include <errno.h>
13	#include <sys/types.h>	57	#include <sys/types.h>
14	#include <sys/stat.h>	58	#include <sys/stat.h>
		59	#include <sys/statvfs.h>
15	#include <limits.h>	60	#include <limits.h>
16	#include <fcntl.h>	61	#include <fcntl.h>
17	#include <sched.h>	62	#include <assert.h>
		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);
18		75
19	#ifndef EIO_FINISH	76	#ifndef EIO_FINISH
20	# 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
21	#endif	78	#endif
22		79
…		…
29	#endif	86	#endif
30		87
31	#ifdef _WIN32	88	#ifdef _WIN32
32		89
33	/*doh*/	90	/*doh*/
34
35	#else	91	#else
36		92
37	# include "config.h"
38	# include <sys/time.h>	93	# include <sys/time.h>
39	# include <sys/select.h>	94	# include <sys/select.h>
40	# include <unistd.h>	95	# include <unistd.h>
41	# include <utime.h>	96	# include <utime.h>
42	# include <signal.h>	97	# include <signal.h>
		98	# include <dirent.h>
		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
43		121
44	# ifndef EIO_STRUCT_DIRENT	122	# ifndef EIO_STRUCT_DIRENT
45	# define EIO_STRUCT_DIRENT struct dirent	123	# define EIO_STRUCT_DIRENT struct dirent
46	# endif	124	# endif
47		125
48	#endif	126	#endif
49		127
50	# ifndef EIO_STRUCT_STAT
51	# define EIO_STRUCT_STAT struct stat
52	# endif
53
54	#if HAVE_SENDFILE	128	#if HAVE_SENDFILE
55	# if __linux	129	# if __linux
56	# include <sys/sendfile.h>	130	# include <sys/sendfile.h>
57	# elif __freebsd	131	# elif __FreeBSD__ || defined __APPLE__
58	# include <sys/socket.h>	132	# include <sys/socket.h>
59	# include <sys/uio.h>	133	# include <sys/uio.h>
60	# elif __hpux	134	# elif __hpux
61	# include <sys/socket.h>	135	# include <sys/socket.h>
62	# elif __solaris /* not yet */	136	# elif __solaris
63	# include <sys/sendfile.h>	137	# include <sys/sendfile.h>
64	# else	138	# else
65	# error sendfile support requested but not available	139	# error sendfile support requested but not available
66	# endif	140	# endif
67	#endif	141	#endif
68		142
69	/* number of seconds after which an idle threads exit */	143	#ifndef D_TYPE
70	#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
71		152
72	/* used for struct dirent, AIX doesn't provide it */	153	/* used for struct dirent, AIX doesn't provide it */
73	#ifndef NAME_MAX	154	#ifndef NAME_MAX
74	# define NAME_MAX 4096	155	# define NAME_MAX 4096
75	#endif	156	#endif
76		157
		158	/* used for readlink etc. */
		159	#ifndef PATH_MAX
		160	# define PATH_MAX 4096
		161	#endif
		162
77	/* buffer size for various temporary buffers */	163	/* buffer size for various temporary buffers */
78	#define EIO_BUFSIZE 65536	164	#define EIO_BUFSIZE 65536
79		165
80	#define dBUF \	166	#define dBUF \
81	char *eio_buf; \	167	char *eio_buf; \
82	X_LOCK (wrklock); \	168	ETP_WORKER_LOCK (self); \
83	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	169	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
84	X_UNLOCK (wrklock); \	170	ETP_WORKER_UNLOCK (self); \
		171	errno = ENOMEM; \
85	if (!eio_buf) \	172	if (!eio_buf) \
86	return -1;	173	return -1;
87		174
88	#define EIO_TICKS ((1000000 + 1023) >> 10)	175	#define EIO_TICKS ((1000000 + 1023) >> 10)
89		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
90	static void (*want_poll_cb) (void);	220	static void (*want_poll_cb) (void);
91	static void (*done_poll_cb) (void);	221	static void (*done_poll_cb) (void);
92		222
93	static unsigned int max_poll_time = 0;	223	static unsigned int max_poll_time; /* reslock */
94	static unsigned int max_poll_reqs = 0;	224	static unsigned int max_poll_reqs; /* reslock */
95		225
96	/* calculcate time difference in ~1/EIO_TICKS of a second */	226	static volatile unsigned int nreqs; /* reqlock */
97	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	227	static volatile unsigned int nready; /* reqlock */
98	{	228	static volatile unsigned int npending; /* reqlock */
99	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 */
100	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	230	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
101	}
102		231
103	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)
104		263
105	/* worker threads management */	264	/* worker threads management */
106	static mutex_t wrklock = X_MUTEX_INIT;
107		265
108	typedef struct worker {	266	static void ecb_cold
109	/* locked by wrklock */
110	struct worker *prev, *next;
111
112	thread_t tid;
113
114	/* locked by reslock, reqlock or wrklock */
115	eio_req *req; /* currently processed request */
116	void *dbuf;
117	DIR *dirp;
118	} worker;
119
120	static worker wrk_first = { &wrk_first, &wrk_first, 0 };
121
122	static void worker_clear (worker *wrk)	267	etp_worker_clear (etp_worker *wrk)
123	{	268	{
124	if (wrk->dirp)	269	ETP_WORKER_CLEAR (wrk);
125	{
126	closedir (wrk->dirp);
127	wrk->dirp = 0;
128	}
129
130	if (wrk->dbuf)
131	{
132	free (wrk->dbuf);
133	wrk->dbuf = 0;
134	}
135	}	270	}
136		271
137	static void worker_free (worker *wrk)	272	static void ecb_cold
		273	etp_worker_free (etp_worker *wrk)
138	{	274	{
139	wrk->next->prev = wrk->prev;	275	wrk->next->prev = wrk->prev;
140	wrk->prev->next = wrk->next;	276	wrk->prev->next = wrk->next;
141		277
142	free (wrk);	278	free (wrk);
143	}	279	}
144		280
145	static volatile unsigned int nreqs, nready, npending;	281	static unsigned int
146	static volatile unsigned int max_idle = 4;	282	etp_nreqs (void)
147
148	static mutex_t reslock = X_MUTEX_INIT;
149	static mutex_t reqlock = X_MUTEX_INIT;
150	static cond_t reqwait = X_COND_INIT;
151
152	unsigned int eio_nreqs (void)
153	{	283	{
		284	int retval;
		285	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		286	retval = nreqs;
		287	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154	return nreqs;	288	return retval;
155	}	289	}
156		290
157	unsigned int eio_nready (void)	291	static unsigned int
		292	etp_nready (void)
158	{	293	{
159	unsigned int retval;	294	unsigned int retval;
160		295
161	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	296	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162	retval = nready;	297	retval = nready;
163	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	298	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164		299
165	return retval;	300	return retval;
166	}	301	}
167		302
168	unsigned int eio_npending (void)	303	static unsigned int
		304	etp_npending (void)
169	{	305	{
170	unsigned int retval;	306	unsigned int retval;
171		307
172	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	308	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173	retval = npending;	309	retval = npending;
174	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	310	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175		311
176	return retval;	312	return retval;
177	}	313	}
178		314
179	unsigned int eio_nthreads (void)	315	static unsigned int
		316	etp_nthreads (void)
180	{	317	{
181	unsigned int retval;	318	unsigned int retval;
182		319
183	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	320	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184	retval = started;	321	retval = started;
…		…
191	* a somewhat faster data structure might be nice, but	328	* a somewhat faster data structure might be nice, but
192	* with 8 priorities this actually needs <20 insns	329	* with 8 priorities this actually needs <20 insns
193	* per shift, the most expensive operation.	330	* per shift, the most expensive operation.
194	*/	331	*/
195	typedef struct {	332	typedef struct {
196	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 */
197	int size;	334	int size;
198	} reqq;	335	} etp_reqq;
199		336
200	static reqq req_queue;	337	static etp_reqq req_queue;
201	static reqq res_queue;	338	static etp_reqq res_queue;
202		339
203	static int reqq_push (reqq *q, eio_req *req)	340	static int ecb_noinline
		341	reqq_push (etp_reqq *q, ETP_REQ *req)
204	{	342	{
205	int pri = req->pri;	343	int pri = req->pri;
206	req->next = 0;	344	req->next = 0;
207		345
208	if (q->qe[pri])	346	if (q->qe[pri])
…		…
214	q->qe[pri] = q->qs[pri] = req;	352	q->qe[pri] = q->qs[pri] = req;
215		353
216	return q->size++;	354	return q->size++;
217	}	355	}
218		356
219	static eio_req *reqq_shift (reqq *q)	357	static ETP_REQ * ecb_noinline
		358	reqq_shift (etp_reqq *q)
220	{	359	{
221	int pri;	360	int pri;
222		361
223	if (!q->size)	362	if (!q->size)
224	return 0;	363	return 0;
225		364
226	--q->size;	365	--q->size;
227		366
228	for (pri = EIO_NUM_PRI; pri--; )	367	for (pri = ETP_NUM_PRI; pri--; )
229	{	368	{
230	eio_req *req = q->qs[pri];	369	eio_req *req = q->qs[pri];
231		370
232	if (req)	371	if (req)
233	{	372	{
…		…
239	}	378	}
240		379
241	abort ();	380	abort ();
242	}	381	}
243		382
244	static void grp_try_feed (eio_req *grp)	383	static void ecb_cold
		384	etp_thread_init (void)
245	{	385	{
246	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	386	X_MUTEX_CREATE (wrklock);
		387	X_MUTEX_CREATE (reslock);
		388	X_MUTEX_CREATE (reqlock);
		389	X_COND_CREATE (reqwait);
		390	}
		391
		392	static void ecb_cold
		393	etp_atfork_prepare (void)
		394	{
		395	X_LOCK (wrklock);
		396	X_LOCK (reqlock);
		397	X_LOCK (reslock);
		398	#if !HAVE_PREADWRITE
		399	X_LOCK (preadwritelock);
		400	#endif
		401	}
		402
		403	static void ecb_cold
		404	etp_atfork_parent (void)
		405	{
		406	#if !HAVE_PREADWRITE
		407	X_UNLOCK (preadwritelock);
		408	#endif
		409	X_UNLOCK (reslock);
		410	X_UNLOCK (reqlock);
		411	X_UNLOCK (wrklock);
		412	}
		413
		414	static void ecb_cold
		415	etp_atfork_child (void)
		416	{
		417	ETP_REQ *prv;
		418
		419	while ((prv = reqq_shift (&req_queue)))
		420	ETP_DESTROY (prv);
		421
		422	while ((prv = reqq_shift (&res_queue)))
		423	ETP_DESTROY (prv);
		424
		425	while (wrk_first.next != &wrk_first)
247	{	426	{
248	int old_len = grp->size;	427	etp_worker *wrk = wrk_first.next;
249		428
250	EIO_FEED (grp);	429	if (wrk->req)
		430	ETP_DESTROY (wrk->req);
251		431
252	/* stop if no progress has been made */	432	etp_worker_clear (wrk);
253	if (old_len == grp->size)	433	etp_worker_free (wrk);
254	{
255	grp->feed = 0;
256	break;
257	}
258	}	434	}
259	}
260		435
261	static int eio_finish (eio_req *req);	436	started = 0;
		437	idle = 0;
		438	nreqs = 0;
		439	nready = 0;
		440	npending = 0;
262		441
263	static int grp_dec (eio_req *grp)	442	etp_thread_init ();
264	{	443	}
265	--grp->size;
266		444
267	/* call feeder, if applicable */	445	static void ecb_cold
268	grp_try_feed (grp);	446	etp_once_init (void)
		447	{
		448	etp_thread_init ();
		449	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
		450	}
269		451
270	/* finish, if done */	452	static int ecb_cold
271	if (!grp->size && grp->int1)	453	etp_init (void (*want_poll)(void), void (*done_poll)(void))
272	return eio_finish (grp);	454	{
273	else	455	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
		456
		457	pthread_once (&doinit, etp_once_init);
		458
		459	want_poll_cb = want_poll;
		460	done_poll_cb = done_poll;
		461
274	return 0;	462	return 0;
275	}	463	}
276		464
277	void eio_destroy (eio_req *req)
278	{
279	if ((req)->flags & EIO_FLAG_PTR2_FREE)
280	free (req->ptr2);
281
282	EIO_DESTROY (req);
283	}
284
285	static int eio_finish (eio_req *req)
286	{
287	int res = EIO_FINISH (req);
288
289	if (req->grp)
290	{
291	int res2;
292	eio_req *grp = req->grp;
293
294	/* unlink request */
295	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
296	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
297
298	if (grp->grp_first == req)
299	grp->grp_first = req->grp_next;
300
301	res2 = grp_dec (grp);
302
303	if (!res && res2)
304	res = res2;
305	}
306
307	eio_destroy (req);
308
309	return res;
310	}
311
312	void eio_grp_cancel (eio_req *grp)
313	{
314	for (grp = grp->grp_first; grp; grp = grp->grp_next)
315	eio_cancel (grp);
316	}
317
318	void eio_cancel (eio_req *req)
319	{
320	req->flags |= EIO_FLAG_CANCELLED;
321
322	eio_grp_cancel (req);
323	}
324
325	X_THREAD_PROC (eio_proc);	465	X_THREAD_PROC (etp_proc);
326		466
327	static void start_thread (void)	467	static void ecb_cold
		468	etp_start_thread (void)
328	{	469	{
329	worker *wrk = calloc (1, sizeof (worker));	470	etp_worker *wrk = calloc (1, sizeof (etp_worker));
330		471
331	if (!wrk)	472	/*TODO*/
332	croak ("unable to allocate worker thread data");	473	assert (("unable to allocate worker thread data", wrk));
333		474
334	X_LOCK (wrklock);	475	X_LOCK (wrklock);
335		476
336	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	477	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
337	{	478	{
338	wrk->prev = &wrk_first;	479	wrk->prev = &wrk_first;
339	wrk->next = wrk_first.next;	480	wrk->next = wrk_first.next;
340	wrk_first.next->prev = wrk;	481	wrk_first.next->prev = wrk;
341	wrk_first.next = wrk;	482	wrk_first.next = wrk;
…		…
345	free (wrk);	486	free (wrk);
346		487
347	X_UNLOCK (wrklock);	488	X_UNLOCK (wrklock);
348	}	489	}
349		490
		491	static void
350	static void maybe_start_thread (void)	492	etp_maybe_start_thread (void)
351	{	493	{
352	if (eio_nthreads () >= wanted)	494	if (ecb_expect_true (etp_nthreads () >= wanted))
353	return;	495	return;
354		496
355	/* todo: maybe use idle here, but might be less exact */	497	/* todo: maybe use idle here, but might be less exact */
356	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	498	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
357	return;	499	return;
358		500
359	start_thread ();	501	etp_start_thread ();
360	}	502	}
361		503
362	void eio_submit (eio_req *req)	504	static void ecb_cold
363	{	505	etp_end_thread (void)
364	++nreqs;
365
366	X_LOCK (reqlock);
367	++nready;
368	reqq_push (&req_queue, req);
369	X_COND_SIGNAL (reqwait);
370	X_UNLOCK (reqlock);
371
372	maybe_start_thread ();
373	}
374
375	static void end_thread (void)
376	{	506	{
377	eio_req *req = calloc (1, sizeof (eio_req));	507	eio_req *req = calloc (1, sizeof (eio_req));
378		508
379	req->type = EIO_QUIT;	509	req->type = -1;
380	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	510	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381		511
382	X_LOCK (reqlock);	512	X_LOCK (reqlock);
383	reqq_push (&req_queue, req);	513	reqq_push (&req_queue, req);
384	X_COND_SIGNAL (reqwait);	514	X_COND_SIGNAL (reqwait);
385	X_UNLOCK (reqlock);	515	X_UNLOCK (reqlock);
…		…
387	X_LOCK (wrklock);	517	X_LOCK (wrklock);
388	--started;	518	--started;
389	X_UNLOCK (wrklock);	519	X_UNLOCK (wrklock);
390	}	520	}
391		521
392	void eio_set_max_poll_time (double nseconds)	522	static int
393	{
394	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
395	max_poll_time = nseconds;
396	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397	}
398
399	void eio_set_max_poll_reqs (unsigned int maxreqs)
400	{
401	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
402	max_poll_reqs = maxreqs;
403	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404	}
405
406	void eio_set_max_idle (unsigned int nthreads)
407	{
408	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
409	max_idle = nthreads <= 0 ? 1 : nthreads;
410	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411	}
412
413	void eio_set_min_parallel (unsigned int nthreads)
414	{
415	if (wanted < nthreads)
416	wanted = nthreads;
417	}
418
419	void eio_set_max_parallel (unsigned int nthreads)
420	{
421	if (wanted > nthreads)
422	wanted = nthreads;
423
424	while (started > wanted)
425	end_thread ();
426	}
427
428	int eio_poll (void)	523	etp_poll (void)
429	{	524	{
430	int maxreqs = max_poll_reqs;	525	unsigned int maxreqs;
		526	unsigned int maxtime;
431	struct timeval tv_start, tv_now;	527	struct timeval tv_start, tv_now;
432	eio_req *req;
433		528
		529	X_LOCK (reslock);
		530	maxreqs = max_poll_reqs;
		531	maxtime = max_poll_time;
		532	X_UNLOCK (reslock);
		533
434	if (max_poll_time)	534	if (maxtime)
435	gettimeofday (&tv_start, 0);	535	gettimeofday (&tv_start, 0);
436		536
437	for (;;)	537	for (;;)
438	{	538	{
		539	ETP_REQ *req;
		540
439	maybe_start_thread ();	541	etp_maybe_start_thread ();
440		542
441	X_LOCK (reslock);	543	X_LOCK (reslock);
442	req = reqq_shift (&res_queue);	544	req = reqq_shift (&res_queue);
443		545
444	if (req)	546	if (req)
…		…
452	X_UNLOCK (reslock);	554	X_UNLOCK (reslock);
453		555
454	if (!req)	556	if (!req)
455	return 0;	557	return 0;
456		558
		559	X_LOCK (reqlock);
457	--nreqs;	560	--nreqs;
		561	X_UNLOCK (reqlock);
458		562
459	if (req->type == EIO_GROUP && req->size)	563	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
460	{	564	{
461	req->int1 = 1; /* mark request as delayed */	565	req->int1 = 1; /* mark request as delayed */
462	continue;	566	continue;
463	}	567	}
464	else	568	else
465	{	569	{
466	int res = eio_finish (req);	570	int res = ETP_FINISH (req);
467	if (res)	571	if (ecb_expect_false (res))
468	return res;	572	return res;
469	}	573	}
470		574
471	if (maxreqs && !--maxreqs)	575	if (ecb_expect_false (maxreqs && !--maxreqs))
472	break;	576	break;
473		577
474	if (max_poll_time)	578	if (maxtime)
475	{	579	{
476	gettimeofday (&tv_now, 0);	580	gettimeofday (&tv_now, 0);
477		581
478	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	582	if (tvdiff (&tv_start, &tv_now) >= maxtime)
479	break;	583	break;
480	}	584	}
481	}	585	}
482		586
483	errno = EAGAIN;	587	errno = EAGAIN;
484	return -1;	588	return -1;
485	}	589	}
486		590
		591	static void
		592	etp_cancel (ETP_REQ *req)
		593	{
		594	req->cancelled = 1;
		595
		596	eio_grp_cancel (req);
		597	}
		598
		599	static void
		600	etp_submit (ETP_REQ *req)
		601	{
		602	req->pri -= ETP_PRI_MIN;
		603
		604	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		605	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		606
		607	if (ecb_expect_false (req->type == EIO_GROUP))
		608	{
		609	/* I hope this is worth it :/ */
		610	X_LOCK (reqlock);
		611	++nreqs;
		612	X_UNLOCK (reqlock);
		613
		614	X_LOCK (reslock);
		615
		616	++npending;
		617
		618	if (!reqq_push (&res_queue, req) && want_poll_cb)
		619	want_poll_cb ();
		620
		621	X_UNLOCK (reslock);
		622	}
		623	else
		624	{
		625	X_LOCK (reqlock);
		626	++nreqs;
		627	++nready;
		628	reqq_push (&req_queue, req);
		629	X_COND_SIGNAL (reqwait);
		630	X_UNLOCK (reqlock);
		631
		632	etp_maybe_start_thread ();
		633	}
		634	}
		635
		636	static void ecb_cold
		637	etp_set_max_poll_time (double nseconds)
		638	{
		639	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		640	max_poll_time = nseconds * EIO_TICKS;
		641	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		642	}
		643
		644	static void ecb_cold
		645	etp_set_max_poll_reqs (unsigned int maxreqs)
		646	{
		647	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		648	max_poll_reqs = maxreqs;
		649	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		650	}
		651
		652	static void ecb_cold
		653	etp_set_max_idle (unsigned int nthreads)
		654	{
		655	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		656	max_idle = nthreads;
		657	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		658	}
		659
		660	static void ecb_cold
		661	etp_set_idle_timeout (unsigned int seconds)
		662	{
		663	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		664	idle_timeout = seconds;
		665	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		666	}
		667
		668	static void ecb_cold
		669	etp_set_min_parallel (unsigned int nthreads)
		670	{
		671	if (wanted < nthreads)
		672	wanted = nthreads;
		673	}
		674
		675	static void ecb_cold
		676	etp_set_max_parallel (unsigned int nthreads)
		677	{
		678	if (wanted > nthreads)
		679	wanted = nthreads;
		680
		681	while (started > wanted)
		682	etp_end_thread ();
		683	}
		684
		685	/*****************************************************************************/
		686
		687	static void
		688	grp_try_feed (eio_req *grp)
		689	{
		690	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		691	{
		692	grp->flags &= ~EIO_FLAG_GROUPADD;
		693
		694	EIO_FEED (grp);
		695
		696	/* stop if no progress has been made */
		697	if (!(grp->flags & EIO_FLAG_GROUPADD))
		698	{
		699	grp->feed = 0;
		700	break;
		701	}
		702	}
		703	}
		704
		705	static int
		706	grp_dec (eio_req *grp)
		707	{
		708	--grp->size;
		709
		710	/* call feeder, if applicable */
		711	grp_try_feed (grp);
		712
		713	/* finish, if done */
		714	if (!grp->size && grp->int1)
		715	return eio_finish (grp);
		716	else
		717	return 0;
		718	}
		719
		720	static void
		721	eio_destroy (eio_req *req)
		722	{
		723	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		724	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		725
		726	EIO_DESTROY (req);
		727	}
		728
		729	static int
		730	eio_finish (eio_req *req)
		731	{
		732	int res = EIO_FINISH (req);
		733
		734	if (req->grp)
		735	{
		736	int res2;
		737	eio_req *grp = req->grp;
		738
		739	/* unlink request */
		740	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		741	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		742
		743	if (grp->grp_first == req)
		744	grp->grp_first = req->grp_next;
		745
		746	res2 = grp_dec (grp);
		747
		748	if (!res)
		749	res = res2;
		750	}
		751
		752	eio_destroy (req);
		753
		754	return res;
		755	}
		756
		757	void
		758	eio_grp_cancel (eio_req *grp)
		759	{
		760	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		761	eio_cancel (grp);
		762	}
		763
		764	void
		765	eio_cancel (eio_req *req)
		766	{
		767	etp_cancel (req);
		768	}
		769
		770	void
		771	eio_submit (eio_req *req)
		772	{
		773	etp_submit (req);
		774	}
		775
		776	unsigned int
		777	eio_nreqs (void)
		778	{
		779	return etp_nreqs ();
		780	}
		781
		782	unsigned int
		783	eio_nready (void)
		784	{
		785	return etp_nready ();
		786	}
		787
		788	unsigned int
		789	eio_npending (void)
		790	{
		791	return etp_npending ();
		792	}
		793
		794	unsigned int ecb_cold
		795	eio_nthreads (void)
		796	{
		797	return etp_nthreads ();
		798	}
		799
		800	void ecb_cold
		801	eio_set_max_poll_time (double nseconds)
		802	{
		803	etp_set_max_poll_time (nseconds);
		804	}
		805
		806	void ecb_cold
		807	eio_set_max_poll_reqs (unsigned int maxreqs)
		808	{
		809	etp_set_max_poll_reqs (maxreqs);
		810	}
		811
		812	void ecb_cold
		813	eio_set_max_idle (unsigned int nthreads)
		814	{
		815	etp_set_max_idle (nthreads);
		816	}
		817
		818	void ecb_cold
		819	eio_set_idle_timeout (unsigned int seconds)
		820	{
		821	etp_set_idle_timeout (seconds);
		822	}
		823
		824	void ecb_cold
		825	eio_set_min_parallel (unsigned int nthreads)
		826	{
		827	etp_set_min_parallel (nthreads);
		828	}
		829
		830	void ecb_cold
		831	eio_set_max_parallel (unsigned int nthreads)
		832	{
		833	etp_set_max_parallel (nthreads);
		834	}
		835
		836	int eio_poll (void)
		837	{
		838	return etp_poll ();
		839	}
		840
487	/*****************************************************************************/	841	/*****************************************************************************/
488	/* work around various missing functions */	842	/* work around various missing functions */
489		843
490	#if !HAVE_PREADWRITE	844	#if !HAVE_PREADWRITE
		845	# undef pread
		846	# undef pwrite
491	# define pread aio_pread	847	# define pread eio__pread
492	# define pwrite aio_pwrite	848	# define pwrite eio__pwrite
493		849
494	/*	850	static ssize_t
495	* make our pread/pwrite safe against themselves, but not against	851	eio__pread (int fd, void *buf, size_t count, off_t offset)
496	* normal read/write by using a mutex. slows down execution a lot,
497	* but that's your problem, not mine.
498	*/
499	static mutex_t preadwritelock = X_MUTEX_INIT;
500
501	static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
502	{	852	{
503	ssize_t res;	853	ssize_t res;
504	off_t ooffset;	854	off_t ooffset;
505		855
506	X_LOCK (preadwritelock);	856	X_LOCK (preadwritelock);
…		…
511	X_UNLOCK (preadwritelock);	861	X_UNLOCK (preadwritelock);
512		862
513	return res;	863	return res;
514	}	864	}
515		865
516	static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset)	866	static ssize_t
		867	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
517	{	868	{
518	ssize_t res;	869	ssize_t res;
519	off_t ooffset;	870	off_t ooffset;
520		871
521	X_LOCK (preadwritelock);	872	X_LOCK (preadwritelock);
522	ooffset = lseek (fd, 0, SEEK_CUR);	873	ooffset = lseek (fd, 0, SEEK_CUR);
523	lseek (fd, offset, SEEK_SET);	874	lseek (fd, offset, SEEK_SET);
524	res = write (fd, buf, count);	875	res = write (fd, buf, count);
525	lseek (fd, offset, SEEK_SET);	876	lseek (fd, ooffset, SEEK_SET);
526	X_UNLOCK (preadwritelock);	877	X_UNLOCK (preadwritelock);
527		878
528	return res;	879	return res;
529	}	880	}
530	#endif	881	#endif
531		882
532	#ifndef HAVE_FUTIMES	883	#ifndef HAVE_UTIMES
533		884
		885	# undef utimes
534	# define utimes(path,times) aio_utimes (path, times)	886	# define utimes(path,times) eio__utimes (path, times)
535	# define futimes(fd,times) aio_futimes (fd, times)
536		887
		888	static int
537	static int aio_utimes (const char *filename, const struct timeval times[2])	889	eio__utimes (const char *filename, const struct timeval times[2])
538	{	890	{
539	if (times)	891	if (times)
540	{	892	{
541	struct utimbuf buf;	893	struct utimbuf buf;
542		894
…		…
547	}	899	}
548	else	900	else
549	return utime (filename, 0);	901	return utime (filename, 0);
550	}	902	}
551		903
		904	#endif
		905
		906	#ifndef HAVE_FUTIMES
		907
		908	# undef futimes
		909	# define futimes(fd,times) eio__futimes (fd, times)
		910
		911	static int
552	static int aio_futimes (int fd, const struct timeval tv[2])	912	eio__futimes (int fd, const struct timeval tv[2])
553	{	913	{
554	errno = ENOSYS;	914	errno = ENOSYS;
555	return -1;	915	return -1;
556	}	916	}
557		917
558	#endif	918	#endif
559		919
560	#if !HAVE_FDATASYNC	920	#if !HAVE_FDATASYNC
		921	# undef fdatasync
561	# define fdatasync fsync	922	# define fdatasync(fd) fsync (fd)
562	#endif	923	#endif
		924
		925	/* sync_file_range always needs emulation */
		926	static int
		927	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		928	{
		929	#if HAVE_SYNC_FILE_RANGE
		930	int res;
		931
		932	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		933	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		934	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		935	{
		936	flags = 0
		937	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		938	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		939	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		940	}
		941
		942	res = sync_file_range (fd, offset, nbytes, flags);
		943
		944	if (!res || errno != ENOSYS)
		945	return res;
		946	#endif
		947
		948	/* even though we could play tricks with the flags, it's better to always
		949	* call fdatasync, as that matches the expectation of its users best */
		950	return fdatasync (fd);
		951	}
		952
		953	static int
		954	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		955	{
		956	#if HAVE_FALLOCATE
		957	return fallocate (fd, mode, offset, len);
		958	#else
		959	errno = ENOSYS;
		960	return -1;
		961	#endif
		962	}
563		963
564	#if !HAVE_READAHEAD	964	#if !HAVE_READAHEAD
		965	# undef readahead
565	# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self)	966	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566		967
		968	static ssize_t
567	static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self)	969	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568	{	970	{
569	size_t todo = count;	971	size_t todo = count;
570	dBUF;	972	dBUF;
571		973
572	while (todo > 0)	974	while (todo > 0)
573	{	975	{
574	size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;	976	size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575		977
576	pread (fd, aio_buf, len, offset);	978	pread (fd, eio_buf, len, offset);
577	offset += len;	979	offset += len;
578	todo -= len;	980	todo -= len;
579	}	981	}
580		982
581	errno = 0;	983	errno = 0;
582	return count;	984	return count;
583	}	985	}
584		986
585	#endif	987	#endif
586		988
587	#if !HAVE_READDIR_R
588	# define readdir_r aio_readdir_r
589
590	static mutex_t readdirlock = X_MUTEX_INIT;
591
592	static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
593	{
594	X_DIRENT *e;
595	int errorno;
596
597	X_LOCK (readdirlock);
598
599	e = readdir (dirp);
600	errorno = errno;
601
602	if (e)
603	{
604	*res = ent;
605	strcpy (ent->d_name, e->d_name);
606	}
607	else
608	*res = 0;
609
610	X_UNLOCK (readdirlock);
611
612	errno = errorno;
613	return e ? 0 : -1;
614	}
615	#endif
616
617	/* sendfile always needs emulation */	989	/* sendfile always needs emulation */
		990	static ssize_t
618	static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self)	991	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
619	{	992	{
		993	ssize_t written = 0;
620	ssize_t res;	994	ssize_t res;
621		995
622	if (!count)	996	if (!count)
623	return 0;	997	return 0;
624		998
		999	for (;;)
		1000	{
		1001	#ifdef __APPLE__
		1002	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1003	#endif
625	#if HAVE_SENDFILE	1004	#if HAVE_SENDFILE
626	# if __linux	1005	# if __linux
		1006	off_t soffset = offset;
627	res = sendfile (ofd, ifd, &offset, count);	1007	res = sendfile (ofd, ifd, &soffset, count);
628		1008
629	# elif __freebsd	1009	# elif __FreeBSD__
630	/*	1010	/*
631	* Of course, the freebsd sendfile is a dire hack with no thoughts	1011	* Of course, the freebsd sendfile is a dire hack with no thoughts
632	* wasted on making it similar to other I/O functions.	1012	* wasted on making it similar to other I/O functions.
633	*/	1013	*/
634	{
635	off_t sbytes;	1014	off_t sbytes;
636	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1015	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637		1016
638	if (res < 0 && sbytes)	1017	#if 0 /* according to the manpage, this is correct, but broken behaviour */
639	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1018	/* freebsd' sendfile will return 0 on success */
		1019	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1020	/* not on e.g. EIO or EPIPE - sounds broken */
		1021	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
640	res = sbytes;	1022	res = sbytes;
641	}	1023	#endif
		1024
		1025	/* according to source inspection, this is correct, and useful behaviour */
		1026	if (sbytes)
		1027	res = sbytes;
		1028
		1029	# elif defined (__APPLE__)
		1030	off_t sbytes = count;
		1031	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1032
		1033	/* according to the manpage, sbytes is always valid */
		1034	if (sbytes)
		1035	res = sbytes;
642		1036
643	# elif __hpux	1037	# elif __hpux
644	res = sendfile (ofd, ifd, offset, count, 0, 0);	1038	res = sendfile (ofd, ifd, offset, count, 0, 0);
645		1039
646	# elif __solaris	1040	# elif __solaris
647	{
648	struct sendfilevec vec;	1041	struct sendfilevec vec;
649	size_t sbytes;	1042	size_t sbytes;
650		1043
651	vec.sfv_fd = ifd;	1044	vec.sfv_fd = ifd;
652	vec.sfv_flag = 0;	1045	vec.sfv_flag = 0;
653	vec.sfv_off = offset;	1046	vec.sfv_off = offset;
654	vec.sfv_len = count;	1047	vec.sfv_len = count;
655		1048
656	res = sendfilev (ofd, &vec, 1, &sbytes);	1049	res = sendfilev (ofd, &vec, 1, &sbytes);
657		1050
658	if (res < 0 && sbytes)	1051	if (res < 0 && sbytes)
659	res = sbytes;	1052	res = sbytes;
660	}
661		1053
662	# endif	1054	# endif
		1055
		1056	#elif defined (_WIN32)
		1057	/* does not work, just for documentation of what would need to be done */
		1058	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1059	/* libeio guarantees that the file offset does not change, and windows */
		1060	/* has no way to get an independent handle to the same file description */
		1061	HANDLE h = TO_SOCKET (ifd);
		1062	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1063	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1064
663	#else	1065	#else
664	res = -1;	1066	res = -1;
665	errno = ENOSYS;	1067	errno = ENOSYS;
666	#endif	1068	#endif
667		1069
		1070	/* we assume sendfile can copy at least 128mb in one go */
		1071	if (res <= 128 * 1024 * 1024)
		1072	{
		1073	if (res > 0)
		1074	written += res;
		1075
		1076	if (written)
		1077	return written;
		1078
		1079	break;
		1080	}
		1081	else
		1082	{
		1083	/* if we requested more, then probably the kernel was lazy */
		1084	written += res;
		1085	offset += res;
		1086	count -= res;
		1087
		1088	if (!count)
		1089	return written;
		1090	}
		1091	}
		1092
668	if (res < 0	1093	if (res < 0
669	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1094	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1095	/* BSDs */
		1096	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1097	|| errno == ENOTSUP
		1098	#endif
		1099	|| errno == EOPNOTSUPP /* BSDs */
670	#if __solaris	1100	#if __solaris
671	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1101	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
672	#endif	1102	#endif
673	)	1103	)
674	)	1104	)
…		…
705	}	1135	}
706		1136
707	return res;	1137	return res;
708	}	1138	}
709		1139
		1140	#ifdef PAGESIZE
		1141	# define eio_pagesize() PAGESIZE
		1142	#else
		1143	static intptr_t
		1144	eio_pagesize (void)
		1145	{
		1146	static intptr_t page;
		1147
		1148	if (!page)
		1149	page = sysconf (_SC_PAGESIZE);
		1150
		1151	return page;
		1152	}
		1153	#endif
		1154
		1155	static void
		1156	eio_page_align (void **addr, size_t *length)
		1157	{
		1158	intptr_t mask = eio_pagesize () - 1;
		1159
		1160	/* round down addr */
		1161	intptr_t adj = mask & (intptr_t)*addr;
		1162
		1163	*addr = (void *)((intptr_t)*addr - adj);
		1164	*length += adj;
		1165
		1166	/* round up length */
		1167	*length = (*length + mask) & ~mask;
		1168	}
		1169
		1170	#if !_POSIX_MEMLOCK
		1171	# define eio__mlockall(a) ((errno = ENOSYS), -1)
		1172	#else
		1173
		1174	static int
		1175	eio__mlockall (int flags)
		1176	{
		1177	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1178	extern int mallopt (int, int);
		1179	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1180	#endif
		1181
		1182	if (EIO_MCL_CURRENT != MCL_CURRENT
		1183	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1184	{
		1185	flags = 0
		1186	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1187	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1188	}
		1189
		1190	return mlockall (flags);
		1191	}
		1192	#endif
		1193
		1194	#if !_POSIX_MEMLOCK_RANGE
		1195	# define eio__mlock(a,b) ((errno = ENOSYS), -1)
		1196	#else
		1197
		1198	static int
		1199	eio__mlock (void *addr, size_t length)
		1200	{
		1201	eio_page_align (&addr, &length);
		1202
		1203	return mlock (addr, length);
		1204	}
		1205
		1206	#endif
		1207
		1208	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1209	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
		1210	#else
		1211
		1212	static int
		1213	eio__msync (void *mem, size_t len, int flags)
		1214	{
		1215	eio_page_align (&mem, &len);
		1216
		1217	if (EIO_MS_ASYNC != MS_SYNC
		1218	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1219	|| EIO_MS_SYNC != MS_SYNC)
		1220	{
		1221	flags = 0
		1222	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1223	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1224	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1225	}
		1226
		1227	return msync (mem, len, flags);
		1228	}
		1229
		1230	#endif
		1231
		1232	static int
		1233	eio__mtouch (eio_req *req)
		1234	{
		1235	void *mem = req->ptr2;
		1236	size_t len = req->size;
		1237	int flags = req->int1;
		1238
		1239	eio_page_align (&mem, &len);
		1240
		1241	{
		1242	intptr_t addr = (intptr_t)mem;
		1243	intptr_t end = addr + len;
		1244	intptr_t page = eio_pagesize ();
		1245
		1246	if (addr < end)
		1247	if (flags & EIO_MT_MODIFY) /* modify */
		1248	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1249	else
		1250	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1251	}
		1252
		1253	return 0;
		1254	}
		1255
		1256	/*****************************************************************************/
		1257	/* requests implemented outside eio_execute, because they are so large */
		1258
		1259	static void
		1260	eio__realpath (eio_req *req, etp_worker *self)
		1261	{
		1262	char *rel = req->ptr1;
		1263	char *res;
		1264	char *tmp1, *tmp2;
		1265	#if SYMLOOP_MAX > 32
		1266	int symlinks = SYMLOOP_MAX;
		1267	#else
		1268	int symlinks = 32;
		1269	#endif
		1270
		1271	req->result = -1;
		1272
		1273	errno = EINVAL;
		1274	if (!rel)
		1275	return;
		1276
		1277	errno = ENOENT;
		1278	if (!*rel)
		1279	return;
		1280
		1281	if (!req->ptr2)
		1282	{
		1283	X_LOCK (wrklock);
		1284	req->flags |= EIO_FLAG_PTR2_FREE;
		1285	X_UNLOCK (wrklock);
		1286	req->ptr2 = malloc (PATH_MAX * 3);
		1287
		1288	errno = ENOMEM;
		1289	if (!req->ptr2)
		1290	return;
		1291	}
		1292
		1293	res = req->ptr2;
		1294	tmp1 = res + PATH_MAX;
		1295	tmp2 = tmp1 + PATH_MAX;
		1296
		1297	#if 0 /* disabled, the musl way to do things is just too racy */
		1298	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1299	/* on linux we may be able to ask the kernel */
		1300	{
		1301	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1302
		1303	if (fd >= 0)
		1304	{
		1305	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1306	req->result = readlink (tmp1, res, PATH_MAX);
		1307	close (fd);
		1308
		1309	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1310
		1311	if (req->result > 0)
		1312	goto done;
		1313	}
		1314	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1315	return;
		1316	}
		1317	#endif
		1318	#endif
		1319
		1320	if (*rel != '/')
		1321	{
		1322	if (!getcwd (res, PATH_MAX))
		1323	return;
		1324
		1325	if (res [1]) /* only use if not / */
		1326	res += strlen (res);
		1327	}
		1328
		1329	while (*rel)
		1330	{
		1331	ssize_t len, linklen;
		1332	char *beg = rel;
		1333
		1334	while (*rel && *rel != '/')
		1335	++rel;
		1336
		1337	len = rel - beg;
		1338
		1339	if (!len) /* skip slashes */
		1340	{
		1341	++rel;
		1342	continue;
		1343	}
		1344
		1345	if (beg [0] == '.')
		1346	{
		1347	if (len == 1)
		1348	continue; /* . - nop */
		1349
		1350	if (beg [1] == '.' && len == 2)
		1351	{
		1352	/* .. - back up one component, if possible */
		1353
		1354	while (res != req->ptr2)
		1355	if (*--res == '/')
		1356	break;
		1357
		1358	continue;
		1359	}
		1360	}
		1361
		1362	errno = ENAMETOOLONG;
		1363	if (res + 1 + len + 1 >= tmp1)
		1364	return;
		1365
		1366	/* copy one component */
		1367	*res = '/';
		1368	memcpy (res + 1, beg, len);
		1369
		1370	/* zero-terminate, for readlink */
		1371	res [len + 1] = 0;
		1372
		1373	/* now check if it's a symlink */
		1374	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1375
		1376	if (linklen < 0)
		1377	{
		1378	if (errno != EINVAL)
		1379	return;
		1380
		1381	/* it's a normal directory. hopefully */
		1382	res += len + 1;
		1383	}
		1384	else
		1385	{
		1386	/* yay, it was a symlink - build new path in tmp2 */
		1387	int rellen = strlen (rel);
		1388
		1389	errno = ENAMETOOLONG;
		1390	if (linklen + 1 + rellen >= PATH_MAX)
		1391	return;
		1392
		1393	errno = ELOOP;
		1394	if (!--symlinks)
		1395	return;
		1396
		1397	if (*tmp1 == '/')
		1398	res = req->ptr2; /* symlink resolves to an absolute path */
		1399
		1400	/* we need to be careful, as rel might point into tmp2 already */
		1401	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1402	tmp2 [linklen] = '/';
		1403	memcpy (tmp2, tmp1, linklen);
		1404
		1405	rel = tmp2;
		1406	}
		1407	}
		1408
		1409	/* special case for the lone root path */
		1410	if (res == req->ptr2)
		1411	*res++ = '/';
		1412
		1413	req->result = res - (char *)req->ptr2;
		1414
		1415	done:
		1416	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1417	}
		1418
		1419	static signed char
		1420	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1421	{
		1422	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1423	: a->inode < b->inode ? -1
		1424	: a->inode > b->inode ? 1
		1425	: 0;
		1426	}
		1427
		1428	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1429
		1430	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1431	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1432
		1433	static void
		1434	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1435	{
		1436	unsigned char bits [9 + sizeof (ino_t) * 8];
		1437	unsigned char *bit = bits;
		1438
		1439	assert (CHAR_BIT == 8);
		1440	assert (sizeof (eio_dirent) * 8 < 256);
		1441	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1442	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1443
		1444	if (size <= EIO_SORT_FAST)
		1445	return;
		1446
		1447	/* first prepare an array of bits to test in our radix sort */
		1448	/* try to take endianness into account, as well as differences in ino_t sizes */
		1449	/* inode_bits must contain all inodes ORed together */
		1450	/* which is used to skip bits that are 0 everywhere, which is very common */
		1451	{
		1452	ino_t endianness;
		1453	int i, j;
		1454
		1455	/* we store the byte offset of byte n into byte n of "endianness" */
		1456	for (i = 0; i < sizeof (ino_t); ++i)
		1457	((unsigned char *)&endianness)[i] = i;
		1458
		1459	*bit++ = 0;
		1460
		1461	for (i = 0; i < sizeof (ino_t); ++i)
		1462	{
		1463	/* shifting off the byte offsets out of "endianness" */
		1464	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1465	endianness >>= 8;
		1466
		1467	for (j = 0; j < 8; ++j)
		1468	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1469	*bit++ = offs + j;
		1470	}
		1471
		1472	for (j = 0; j < 8; ++j)
		1473	if (score_bits & (1 << j))
		1474	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1475	}
		1476
		1477	/* now actually do the sorting (a variant of MSD radix sort) */
		1478	{
		1479	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1480	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1481	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1482	int stk_idx = 0;
		1483
		1484	base_stk [stk_idx] = dents;
		1485	end_stk [stk_idx] = dents + size;
		1486	bit_stk [stk_idx] = bit - 1;
		1487
		1488	do
		1489	{
		1490	base = base_stk [stk_idx];
		1491	end = end_stk [stk_idx];
		1492	bit = bit_stk [stk_idx];
		1493
		1494	for (;;)
		1495	{
		1496	unsigned char O = *bit >> 3;
		1497	unsigned char M = 1 << (*bit & 7);
		1498
		1499	eio_dirent *a = base;
		1500	eio_dirent *b = end;
		1501
		1502	if (b - a < EIO_SORT_CUTOFF)
		1503	break;
		1504
		1505	/* now bit-partition the array on the bit */
		1506	/* this ugly asymmetric loop seems to perform much better than typical */
		1507	/* partition algos found in the literature */
		1508	do
		1509	if (!(((unsigned char *)a)[O] & M))
		1510	++a;
		1511	else if (!(((unsigned char *)--b)[O] & M))
		1512	{
		1513	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1514	++a;
		1515	}
		1516	while (b > a);
		1517
		1518	/* next bit, or stop, if no bits left in this path */
		1519	if (!*--bit)
		1520	break;
		1521
		1522	base_stk [stk_idx] = a;
		1523	end_stk [stk_idx] = end;
		1524	bit_stk [stk_idx] = bit;
		1525	++stk_idx;
		1526
		1527	end = a;
		1528	}
		1529	}
		1530	while (stk_idx--);
		1531	}
		1532	}
		1533
		1534	static void
		1535	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1536	{
		1537	/* first move the smallest element to the front, to act as a sentinel */
		1538	{
		1539	int i;
		1540	eio_dirent *min = dents;
		1541
		1542	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1543	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1544	if (EIO_DENT_CMP (dents [i], <, *min))
		1545	min = &dents [i];
		1546
		1547	/* swap elements 0 and j (minimum) */
		1548	{
		1549	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1550	}
		1551	}
		1552
		1553	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1554	{
		1555	eio_dirent *i, *j;
		1556
		1557	for (i = dents + 1; i < dents + size; ++i)
		1558	{
		1559	eio_dirent value = *i;
		1560
		1561	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1562	j [1] = j [0];
		1563
		1564	j [1] = value;
		1565	}
		1566	}
		1567	}
		1568
		1569	static void
		1570	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1571	{
		1572	if (size <= 1)
		1573	return; /* our insertion sort relies on size > 0 */
		1574
		1575	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1576	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1577	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1578
		1579	/* use an insertion sort at the end, or for small arrays, */
		1580	/* as insertion sort is more efficient for small partitions */
		1581	eio_dent_insertion_sort (dents, size);
		1582	}
		1583
710	/* read a full directory */	1584	/* read a full directory */
		1585	static void
711	static void scandir_ (eio_req *req, worker *self)	1586	eio__scandir (eio_req *req, etp_worker *self)
712	{	1587	{
713	DIR *dirp;	1588	DIR *dirp;
714	union
715	{
716	EIO_STRUCT_DIRENT d;
717	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
718	} *u;
719	EIO_STRUCT_DIRENT *entp;	1589	EIO_STRUCT_DIRENT *entp;
720	char *name, *names;	1590	char *name, *names;
721	int memlen = 4096;	1591	int namesalloc = 4096;
722	int memofs = 0;	1592	int namesoffs = 0;
		1593	int flags = req->int1;
		1594	eio_dirent *dents = 0;
		1595	int dentalloc = 128;
723	int res = 0;	1596	int dentoffs = 0;
		1597	ino_t inode_bits = 0;
		1598
		1599	req->result = -1;
		1600
		1601	if (!(flags & EIO_READDIR_DENTS))
		1602	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
724		1603
725	X_LOCK (wrklock);	1604	X_LOCK (wrklock);
		1605	/* the corresponding closedir is in ETP_WORKER_CLEAR */
726	self->dirp = dirp = opendir (req->ptr1);	1606	self->dirp = dirp = opendir (req->ptr1);
727	self->dbuf = u = malloc (sizeof (*u));	1607
728	req->flags |= EIO_FLAG_PTR2_FREE;	1608	if (req->flags & EIO_FLAG_PTR1_FREE)
		1609	free (req->ptr1);
		1610
		1611	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1612	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
729	req->ptr2 = names = malloc (memlen);	1613	req->ptr2 = names = malloc (namesalloc);
730	X_UNLOCK (wrklock);	1614	X_UNLOCK (wrklock);
731		1615
732	if (dirp && u && names)	1616	if (dirp && names && (!flags || dents))
733	for (;;)	1617	for (;;)
734	{	1618	{
735	errno = 0;	1619	errno = 0;
736	readdir_r (dirp, &u->d, &entp);	1620	entp = readdir (dirp);
737		1621
738	if (!entp)	1622	if (!entp)
		1623	{
		1624	if (errno)
		1625	break;
		1626
		1627	/* sort etc. */
		1628	req->int1 = flags;
		1629	req->result = dentoffs;
		1630
		1631	if (flags & EIO_READDIR_STAT_ORDER)
		1632	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1633	else if (flags & EIO_READDIR_DIRS_FIRST)
		1634	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1635	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1636	else
		1637	{
		1638	/* in this case, all is known, and we just put dirs first and sort them */
		1639	eio_dirent *oth = dents + dentoffs;
		1640	eio_dirent *dir = dents;
		1641
		1642	/* now partition dirs to the front, and non-dirs to the back */
		1643	/* by walking from both sides and swapping if necessary */
		1644	while (oth > dir)
		1645	{
		1646	if (dir->type == EIO_DT_DIR)
		1647	++dir;
		1648	else if ((--oth)->type == EIO_DT_DIR)
		1649	{
		1650	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1651
		1652	++dir;
		1653	}
		1654	}
		1655
		1656	/* now sort the dirs only (dirs all have the same score) */
		1657	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1658	}
		1659
739	break;	1660	break;
		1661	}
740		1662
		1663	/* now add the entry to our list(s) */
741	name = entp->d_name;	1664	name = entp->d_name;
742		1665
		1666	/* skip . and .. entries */
743	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1667	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
744	{	1668	{
745	int len = strlen (name) + 1;	1669	int len = D_NAMLEN (entp) + 1;
746		1670
747	res++;	1671	while (ecb_expect_false (namesoffs + len > namesalloc))
748
749	while (memofs + len > memlen)
750	{	1672	{
751	memlen *= 2;	1673	namesalloc *= 2;
752	X_LOCK (wrklock);	1674	X_LOCK (wrklock);
753	req->ptr2 = names = realloc (names, memlen);	1675	req->ptr2 = names = realloc (names, namesalloc);
754	X_UNLOCK (wrklock);	1676	X_UNLOCK (wrklock);
755		1677
756	if (!names)	1678	if (!names)
757	break;	1679	break;
758	}	1680	}
759		1681
760	memcpy (names + memofs, name, len);	1682	memcpy (names + namesoffs, name, len);
		1683
		1684	if (dents)
		1685	{
		1686	struct eio_dirent *ent;
		1687
		1688	if (ecb_expect_false (dentoffs == dentalloc))
		1689	{
		1690	dentalloc *= 2;
		1691	X_LOCK (wrklock);
		1692	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1693	X_UNLOCK (wrklock);
		1694
		1695	if (!dents)
		1696	break;
		1697	}
		1698
		1699	ent = dents + dentoffs;
		1700
		1701	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1702	ent->namelen = len - 1;
		1703	ent->inode = D_INO (entp);
		1704
		1705	inode_bits |= ent->inode;
		1706
		1707	switch (D_TYPE (entp))
		1708	{
		1709	default:
		1710	ent->type = EIO_DT_UNKNOWN;
		1711	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1712	break;
		1713
		1714	#ifdef DT_FIFO
		1715	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1716	#endif
		1717	#ifdef DT_CHR
		1718	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1719	#endif
		1720	#ifdef DT_MPC
		1721	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1722	#endif
		1723	#ifdef DT_DIR
		1724	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1725	#endif
		1726	#ifdef DT_NAM
		1727	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1728	#endif
		1729	#ifdef DT_BLK
		1730	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1731	#endif
		1732	#ifdef DT_MPB
		1733	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1734	#endif
		1735	#ifdef DT_REG
		1736	case DT_REG: ent->type = EIO_DT_REG; break;
		1737	#endif
		1738	#ifdef DT_NWK
		1739	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1740	#endif
		1741	#ifdef DT_CMP
		1742	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1743	#endif
		1744	#ifdef DT_LNK
		1745	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1746	#endif
		1747	#ifdef DT_SOCK
		1748	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1749	#endif
		1750	#ifdef DT_DOOR
		1751	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1752	#endif
		1753	#ifdef DT_WHT
		1754	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1755	#endif
		1756	}
		1757
		1758	ent->score = 7;
		1759
		1760	if (flags & EIO_READDIR_DIRS_FIRST)
		1761	{
		1762	if (ent->type == EIO_DT_UNKNOWN)
		1763	{
		1764	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1765	ent->score = 1;
		1766	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1767	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1768	}
		1769	else if (ent->type == EIO_DT_DIR)
		1770	ent->score = 0;
		1771	}
		1772	}
		1773
761	memofs += len;	1774	namesoffs += len;
		1775	++dentoffs;
		1776	}
		1777
		1778	if (EIO_CANCELLED (req))
		1779	{
		1780	errno = ECANCELED;
		1781	break;
762	}	1782	}
763	}	1783	}
764
765	if (errno)
766	res = -1;
767
768	req->result = res;
769	}	1784	}
770		1785
771	/*****************************************************************************/	1786	/*****************************************************************************/
772		1787
		1788	#define ALLOC(len) \
		1789	if (!req->ptr2) \
		1790	{ \
		1791	X_LOCK (wrklock); \
		1792	req->flags |= EIO_FLAG_PTR2_FREE; \
		1793	X_UNLOCK (wrklock); \
		1794	req->ptr2 = malloc (len); \
		1795	if (!req->ptr2) \
		1796	{ \
		1797	errno = ENOMEM; \
		1798	req->result = -1; \
		1799	break; \
		1800	} \
		1801	}
		1802
773	X_THREAD_PROC (eio_proc)	1803	X_THREAD_PROC (etp_proc)
774	{	1804	{
775	eio_req *req;	1805	ETP_REQ *req;
776	struct timespec ts;	1806	struct timespec ts;
777	worker *self = (worker *)thr_arg;	1807	etp_worker *self = (etp_worker *)thr_arg;
778		1808
779	/* try to distribute timeouts somewhat randomly */	1809	/* try to distribute timeouts somewhat randomly */
780	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1810	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781		1811
782	for (;;)	1812	for (;;)
783	{	1813	{
784	ts.tv_sec = time (0) + IDLE_TIMEOUT;
785
786	X_LOCK (reqlock);	1814	X_LOCK (reqlock);
787		1815
788	for (;;)	1816	for (;;)
789	{	1817	{
790	self->req = req = reqq_shift (&req_queue);	1818	self->req = req = reqq_shift (&req_queue);
…		…
792	if (req)	1820	if (req)
793	break;	1821	break;
794		1822
795	++idle;	1823	++idle;
796		1824
		1825	ts.tv_sec = time (0) + idle_timeout;
797	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1826	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
798	{	1827	{
799	if (idle > max_idle)	1828	if (idle > max_idle)
800	{	1829	{
801	--idle;	1830	--idle;
…		…
806	goto quit;	1835	goto quit;
807	}	1836	}
808		1837
809	/* we are allowed to idle, so do so without any timeout */	1838	/* we are allowed to idle, so do so without any timeout */
810	X_COND_WAIT (reqwait, reqlock);	1839	X_COND_WAIT (reqwait, reqlock);
811	ts.tv_sec = time (0) + IDLE_TIMEOUT;
812	}	1840	}
813		1841
814	--idle;	1842	--idle;
815	}	1843	}
816		1844
817	--nready;	1845	--nready;
818		1846
819	X_UNLOCK (reqlock);	1847	X_UNLOCK (reqlock);
820		1848
821	errno = 0; /* strictly unnecessary */
822
823	if (!EIO_CANCELLED (req))
824	switch (req->type)	1849	if (req->type < 0)
825	{
826	case EIO_READ: req->result = req->offs >= 0
827	? pread (req->int1, req->ptr2, req->size, req->offs)
828	: read (req->int1, req->ptr2, req->size); break;
829	case EIO_WRITE: req->result = req->offs >= 0
830	? pwrite (req->int1, req->ptr2, req->size, req->offs)
831	: write (req->int1, req->ptr2, req->size); break;
832
833	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
834	case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
835
836	case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
837	case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
838	case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
839
840	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
841	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
842	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
843	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
844	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
845	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
846
847	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
848	case EIO_CLOSE: req->result = close (req->int1); break;
849	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
850	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
851	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
852	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
853	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
854	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
855	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
856	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
857	case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
858
859	case EIO_SYNC: req->result = 0; sync (); break;
860	case EIO_FSYNC: req->result = fsync (req->int1); break;
861	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
862
863	case EIO_READDIR: scandir_ (req, self); break;
864
865	case EIO_BUSY:
866	#ifdef _WIN32
867	Sleep (req->nv1 * 1000.);
868	#else
869	{
870	struct timeval tv;
871
872	tv.tv_sec = req->nv1;
873	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
874
875	req->result = select (0, 0, 0, 0, &tv);
876	}
877	#endif
878	break;
879
880	case EIO_UTIME:
881	case EIO_FUTIME:
882	{
883	struct timeval tv[2];
884	struct timeval *times;
885
886	if (req->nv1 != -1. || req->nv2 != -1.)
887	{
888	tv[0].tv_sec = req->nv1;
889	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
890	tv[1].tv_sec = req->nv2;
891	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
892
893	times = tv;
894	}
895	else
896	times = 0;
897
898
899	req->result = req->type == EIO_FUTIME
900	? futimes (req->int1, times)
901	: utimes (req->ptr1, times);
902	}
903
904	case EIO_GROUP:
905	case EIO_NOP:
906	break;
907
908	case EIO_QUIT:
909	goto quit;	1850	goto quit;
910		1851
911	default:	1852	ETP_EXECUTE (self, req);
912	req->result = -1;
913	break;
914	}
915
916	req->errorno = errno;
917		1853
918	X_LOCK (reslock);	1854	X_LOCK (reslock);
919		1855
920	++npending;	1856	++npending;
921		1857
922	if (!reqq_push (&res_queue, req) && want_poll_cb)	1858	if (!reqq_push (&res_queue, req) && want_poll_cb)
923	want_poll_cb ();	1859	want_poll_cb ();
924		1860
925	self->req = 0;	1861	self->req = 0;
926	worker_clear (self);	1862	etp_worker_clear (self);
927		1863
928	X_UNLOCK (reslock);	1864	X_UNLOCK (reslock);
929	}	1865	}
930		1866
931	quit:	1867	quit:
932	X_LOCK (wrklock);	1868	X_LOCK (wrklock);
933	worker_free (self);	1869	etp_worker_free (self);
934	X_UNLOCK (wrklock);	1870	X_UNLOCK (wrklock);
935		1871
936	return 0;	1872	return 0;
937	}	1873	}
938		1874
939	/*****************************************************************************/	1875	/*****************************************************************************/
940		1876
941	static void eio_atfork_prepare (void)	1877	int ecb_cold
		1878	eio_init (void (*want_poll)(void), void (*done_poll)(void))
942	{	1879	{
943	X_LOCK (wrklock);	1880	return etp_init (want_poll, done_poll);
944	X_LOCK (reqlock);
945	X_LOCK (reslock);
946	#if !HAVE_PREADWRITE
947	X_LOCK (preadwritelock);
948	#endif
949	#if !HAVE_READDIR_R
950	X_LOCK (readdirlock);
951	#endif
952	}	1881	}
953		1882
954	static void eio_atfork_parent (void)	1883	ecb_inline void
		1884	eio_api_destroy (eio_req *req)
955	{	1885	{
956	#if !HAVE_READDIR_R	1886	free (req);
957	X_UNLOCK (readdirlock);
958	#endif
959	#if !HAVE_PREADWRITE
960	X_UNLOCK (preadwritelock);
961	#endif
962	X_UNLOCK (reslock);
963	X_UNLOCK (reqlock);
964	X_UNLOCK (wrklock);
965	}	1887	}
966		1888
967	static void eio_atfork_child (void)	1889	#define REQ(rtype) \
968	{	1890	eio_req *req; \
969	eio_req *prv;	1891	\
		1892	req = (eio_req *)calloc (1, sizeof *req); \
		1893	if (!req) \
		1894	return 0; \
		1895	\
		1896	req->type = rtype; \
		1897	req->pri = pri; \
		1898	req->finish = cb; \
		1899	req->data = data; \
		1900	req->destroy = eio_api_destroy;
970		1901
971	while (prv = reqq_shift (&req_queue))	1902	#define SEND eio_submit (req); return req
972	eio_destroy (prv);
973		1903
974	while (prv = reqq_shift (&res_queue))	1904	#define PATH \
975	eio_destroy (prv);	1905	req->flags |= EIO_FLAG_PTR1_FREE; \
		1906	req->ptr1 = strdup (path); \
		1907	if (!req->ptr1) \
		1908	{ \
		1909	eio_api_destroy (req); \
		1910	return 0; \
		1911	}
976		1912
977	while (wrk_first.next != &wrk_first)	1913	static void
		1914	eio_execute (etp_worker *self, eio_req *req)
		1915	{
		1916	if (ecb_expect_false (EIO_CANCELLED (req)))
978	{	1917	{
979	worker *wrk = wrk_first.next;	1918	req->result = -1;
		1919	req->errorno = ECANCELED;
		1920	return;
		1921	}
980		1922
981	if (wrk->req)	1923	switch (req->type)
982	eio_destroy (wrk->req);
983
984	worker_clear (wrk);
985	worker_free (wrk);
986	}	1924	{
		1925	case EIO_READ: ALLOC (req->size);
		1926	req->result = req->offs >= 0
		1927	? pread (req->int1, req->ptr2, req->size, req->offs)
		1928	: read (req->int1, req->ptr2, req->size); break;
		1929	case EIO_WRITE: req->result = req->offs >= 0
		1930	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1931	: write (req->int1, req->ptr2, req->size); break;
987		1932
988	started = 0;	1933	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
989	idle = 0;	1934	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
990	nreqs = 0;
991	nready = 0;
992	npending = 0;
993		1935
994	eio_atfork_parent ();	1936	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
995	}	1937	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1938	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1939	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1940	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1941	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
996		1942
997	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1943	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
998	{	1944	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
999	want_poll_cb = want_poll;	1945	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1000	done_poll_cb = done_poll;	1946	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1001		1947
		1948	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1949	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1950	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1951	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1952	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1953	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1954
		1955	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1956	case EIO_CLOSE: req->result = close (req->int1); break;
		1957	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1958	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1959	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1960	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1961	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1962	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1963	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1964	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		1965
		1966	case EIO_REALPATH: eio__realpath (req, self); break;
		1967
		1968	case EIO_READLINK: ALLOC (PATH_MAX);
		1969	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
		1970
		1971	case EIO_SYNC: req->result = 0; sync (); break;
		1972	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1973	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1974	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1975	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		1976	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		1977	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		1978	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1979	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
		1980
		1981	case EIO_READDIR: eio__scandir (req, self); break;
		1982
		1983	case EIO_BUSY:
1002	#ifdef _WIN32	1984	#ifdef _WIN32
1003	X_MUTEX_CHECK (wrklock);	1985	Sleep (req->nv1 * 1e3);
1004	X_MUTEX_CHECK (reslock);	1986	#else
1005	X_MUTEX_CHECK (reqlock);	1987	{
1006	X_MUTEX_CHECK (reqwait);	1988	struct timeval tv;
1007	X_MUTEX_CHECK (preadwritelock);
1008	X_MUTEX_CHECK (readdirlock);
1009		1989
1010	X_COND_CHECK (reqwait);	1990	tv.tv_sec = req->nv1;
1011	#endif	1991	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1012		1992
1013	X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);	1993	req->result = select (0, 0, 0, 0, &tv);
1014	}	1994	}
		1995	#endif
		1996	break;
1015		1997
1016	#if 0	1998	case EIO_UTIME:
		1999	case EIO_FUTIME:
		2000	{
		2001	struct timeval tv[2];
		2002	struct timeval *times;
1017		2003
		2004	if (req->nv1 != -1. || req->nv2 != -1.)
		2005	{
		2006	tv[0].tv_sec = req->nv1;
		2007	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		2008	tv[1].tv_sec = req->nv2;
		2009	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		2010
		2011	times = tv;
		2012	}
		2013	else
		2014	times = 0;
		2015
		2016	req->result = req->type == EIO_FUTIME
		2017	? futimes (req->int1, times)
		2018	: utimes (req->ptr1, times);
		2019	}
		2020	break;
		2021
		2022	case EIO_GROUP:
		2023	abort (); /* handled in eio_request */
		2024
		2025	case EIO_NOP:
		2026	req->result = 0;
		2027	break;
		2028
		2029	case EIO_CUSTOM:
		2030	req->feed (req);
		2031	break;
		2032
		2033	default:
		2034	errno = ENOSYS;
		2035	req->result = -1;
		2036	break;
		2037	}
		2038
		2039	req->errorno = errno;
		2040	}
		2041
		2042	#ifndef EIO_NO_WRAPPERS
		2043
		2044	eio_req *eio_nop (int pri, eio_cb cb, void *data)
		2045	{
		2046	REQ (EIO_NOP); SEND;
		2047	}
		2048
		2049	eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
		2050	{
		2051	REQ (EIO_BUSY); req->nv1 = delay; SEND;
		2052	}
		2053
1018	eio_req *eio_fsync (int fd, eio_cb cb);	2054	eio_req *eio_sync (int pri, eio_cb cb, void *data)
		2055	{
		2056	REQ (EIO_SYNC); SEND;
		2057	}
		2058
		2059	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
		2060	{
		2061	REQ (EIO_FSYNC); req->int1 = fd; SEND;
		2062	}
		2063
		2064	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2065	{
		2066	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2067	}
		2068
		2069	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2070	{
		2071	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2072	}
		2073
		2074	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2075	{
		2076	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2077	}
		2078
		2079	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2080	{
		2081	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2082	}
		2083
		2084	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2085	{
		2086	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2087	}
		2088
		2089	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2090	{
		2091	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2092	}
		2093
1019	eio_req *eio_fdatasync (int fd, eio_cb cb);	2094	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1020	eio_req *eio_dupclose (int fd, eio_cb cb);	2095	{
		2096	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2097	}
		2098
		2099	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
		2100	{
		2101	REQ (EIO_CLOSE); req->int1 = fd; SEND;
		2102	}
		2103
1021	eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb);	2104	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1022	eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb);	2105	{
1023	eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb);	2106	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1024	eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */	2107	}
		2108
		2109	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
		2110	{
		2111	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
		2112	}
		2113
		2114	eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
		2115	{
		2116	REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
		2117	}
		2118
		2119	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
		2120	{
		2121	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		2122	}
		2123
		2124	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2125	{
		2126	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2127	}
		2128
1025	eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb);	2129	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
		2130	{
		2131	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
		2132	}
		2133
1026	eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb);	2134	eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
		2135	{
		2136	REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
		2137	}
		2138
1027	eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb);	2139	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
		2140	{
		2141	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
		2142	}
		2143
1028	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb);	2144	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
		2145	{
		2146	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
		2147	}
		2148
1029	eio_req *eio_dup2 (int fd, int fd2, eio_cb cb);	2149	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
		2150	{
		2151	REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
		2152	}
		2153
1030	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb);	2154	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
		2155	{
		2156	REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
		2157	}
		2158
1031	eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb);	2159	eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1032	eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */	2160	{
1033	eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */	2161	REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1034	eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */	2162	}
		2163
1035	eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb);	2164	eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
		2165	{
		2166	REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
		2167	}
		2168
1036	eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb);	2169	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1037	eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb);	2170	{
1038	eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb);	2171	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
		2172	}
		2173
1039	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb);	2174	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
		2175	{
		2176	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
		2177	}
		2178
		2179	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
		2180	{
		2181	REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
		2182	}
		2183
		2184	eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
		2185	{
		2186	REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
		2187	}
		2188
		2189	static eio_req *
		2190	eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
		2191	{
		2192	REQ (type); PATH; SEND;
		2193	}
		2194
		2195	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
		2196	{
		2197	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2198	}
		2199
		2200	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2201	{
		2202	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2203	}
		2204
		2205	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
		2206	{
		2207	return eio__1path (EIO_STAT, path, pri, cb, data);
		2208	}
		2209
		2210	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
		2211	{
		2212	return eio__1path (EIO_LSTAT, path, pri, cb, data);
		2213	}
		2214
		2215	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2216	{
		2217	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2218	}
		2219
1040	eio_req *eio_unlink (const char *path, eio_cb cb);	2220	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
		2221	{
		2222	return eio__1path (EIO_UNLINK, path, pri, cb, data);
		2223	}
		2224
1041	eio_req *eio_rmdir (const char *path, eio_cb cb);	2225	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1042	eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */	2226	{
		2227	return eio__1path (EIO_RMDIR, path, pri, cb, data);
		2228	}
		2229
		2230	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
		2231	{
		2232	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
		2233	}
		2234
1043	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb);	2235	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1044	eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */	2236	{
1045	eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */	2237	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
		2238	}
		2239
		2240	static eio_req *
		2241	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2242	{
		2243	REQ (type); PATH;
		2244
		2245	req->flags |= EIO_FLAG_PTR2_FREE;
		2246	req->ptr2 = strdup (new_path);
		2247	if (!req->ptr2)
		2248	{
		2249	eio_api_destroy (req);
		2250	return 0;
		2251	}
		2252
		2253	SEND;
		2254	}
		2255
		2256	eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2257	{
		2258	return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
		2259	}
		2260
		2261	eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2262	{
		2263	return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
		2264	}
		2265
		2266	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2267	{
		2268	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
		2269	}
		2270
		2271	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
		2272	{
		2273	REQ (EIO_CUSTOM); req->feed = execute; SEND;
		2274	}
		2275
		2276	#endif
		2277
		2278	eio_req *eio_grp (eio_cb cb, void *data)
		2279	{
		2280	const int pri = EIO_PRI_MAX;
		2281
		2282	REQ (EIO_GROUP); SEND;
		2283	}
		2284
		2285	#undef REQ
		2286	#undef PATH
		2287	#undef SEND
		2288
		2289	/*****************************************************************************/
		2290	/* grp functions */
		2291
1046	void	2292	void
1047	aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1048	PROTOTYPE: $$$;$
1049	PPCODE:
1050	{
1051	dREQ;
1052
1053	req->type = EIO_OPEN;
1054	req->sv1 = newSVsv (pathname);
1055	req->ptr1 = SvPVbyte_nolen (req->sv1);
1056	req->int1 = flags;
1057	req->int2 = mode;
1058
1059	EIO_SEND;
1060	}
1061
1062	void
1063	aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1064	PROTOTYPE: $;$
1065	ALIAS:
1066	aio_fsync = EIO_FSYNC
1067	aio_fdatasync = EIO_FDATASYNC
1068	PPCODE:
1069	{
1070	dREQ;
1071
1072	req->type = ix;
1073	req->sv1 = newSVsv (fh);
1074	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1075
1076	EIO_SEND (req);
1077	}
1078
1079	void
1080	aio_close (SV *fh, SV *callback=&PL_sv_undef)
1081	PROTOTYPE: $;$
1082	PPCODE:
1083	{
1084	dREQ;
1085
1086	req->type = EIO_CLOSE;
1087	req->sv1 = newSVsv (fh);
1088	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1089
1090	EIO_SEND (req);
1091	}
1092
1093	void
1094	aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1095	ALIAS:
1096	aio_read = EIO_READ
1097	aio_write = EIO_WRITE
1098	PROTOTYPE: $$$$$;$
1099	PPCODE:
1100	{
1101	STRLEN svlen;
1102	char *svptr = SvPVbyte (data, svlen);
1103	UV len = SvUV (length);
1104
1105	SvUPGRADE (data, SVt_PV);
1106	SvPOK_on (data);
1107
1108	if (dataoffset < 0)
1109	dataoffset += svlen;
1110
1111	if (dataoffset < 0 || dataoffset > svlen)
1112	croak ("dataoffset outside of data scalar");
1113
1114	if (ix == EIO_WRITE)
1115	{
1116	/* write: check length and adjust. */
1117	if (!SvOK (length) || len + dataoffset > svlen)
1118	len = svlen - dataoffset;
1119	}
1120	else
1121	{
1122	/* read: grow scalar as necessary */
1123	svptr = SvGROW (data, len + dataoffset + 1);
1124	}
1125
1126	if (len < 0)
1127	croak ("length must not be negative");
1128
1129	{
1130	dREQ;
1131
1132	req->type = ix;
1133	req->sv1 = newSVsv (fh);
1134	req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1135	: IoOFP (sv_2io (fh)));
1136	req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1137	req->size = len;
1138	req->sv2 = SvREFCNT_inc (data);
1139	req->ptr2 = (char *)svptr + dataoffset;
1140	req->stroffset = dataoffset;
1141
1142	if (!SvREADONLY (data))
1143	{
1144	SvREADONLY_on (data);
1145	req->flags |= FLAG_SV2_RO_OFF;
1146	}
1147
1148	EIO_SEND;
1149	}
1150	}
1151
1152	void
1153	aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1154	PROTOTYPE: $$;$
1155	PPCODE:
1156	{
1157	SV *data;
1158	dREQ;
1159
1160	data = newSV (NAME_MAX);
1161	SvPOK_on (data);
1162
1163	req->type = EIO_READLINK;
1164	req->sv1 = newSVsv (path);
1165	req->ptr1 = SvPVbyte_nolen (req->sv1);
1166	req->sv2 = data;
1167	req->ptr2 = SvPVbyte_nolen (data);
1168
1169	EIO_SEND;
1170	}
1171
1172	void
1173	aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174	PROTOTYPE: $$$$;$
1175	PPCODE:
1176	{
1177	dREQ;
1178
1179	req->type = EIO_SENDFILE;
1180	req->sv1 = newSVsv (out_fh);
1181	req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182	req->sv2 = newSVsv (in_fh);
1183	req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184	req->offs = SvVAL64 (in_offset);
1185	req->size = length;
1186
1187	EIO_SEND;
1188	}
1189
1190	void
1191	aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192	PROTOTYPE: $$$;$
1193	PPCODE:
1194	{
1195	dREQ;
1196
1197	req->type = EIO_READAHEAD;
1198	req->sv1 = newSVsv (fh);
1199	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1200	req->offs = SvVAL64 (offset);
1201	req->size = length;
1202
1203	EIO_SEND;
1204	}
1205
1206	void
1207	aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1208	ALIAS:
1209	aio_stat = EIO_STAT
1210	aio_lstat = EIO_LSTAT
1211	PPCODE:
1212	{
1213	dREQ;
1214
1215	req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1216	if (!req->ptr2)
1217	{
1218	req_destroy (req);
1219	croak ("out of memory during aio_stat statdata allocation");
1220	}
1221
1222	req->flags |= FLAG_PTR2_FREE;
1223	req->sv1 = newSVsv (fh_or_path);
1224
1225	if (SvPOK (fh_or_path))
1226	{
1227	req->type = ix;
1228	req->ptr1 = SvPVbyte_nolen (req->sv1);
1229	}
1230	else
1231	{
1232	req->type = EIO_FSTAT;
1233	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1234	}
1235
1236	EIO_SEND;
1237	}
1238
1239	void
1240	aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1241	PPCODE:
1242	{
1243	dREQ;
1244
1245	req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1246	req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1247	req->sv1 = newSVsv (fh_or_path);
1248
1249	if (SvPOK (fh_or_path))
1250	{
1251	req->type = EIO_UTIME;
1252	req->ptr1 = SvPVbyte_nolen (req->sv1);
1253	}
1254	else
1255	{
1256	req->type = EIO_FUTIME;
1257	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1258	}
1259
1260	EIO_SEND;
1261	}
1262
1263	void
1264	aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1265	PPCODE:
1266	{
1267	dREQ;
1268
1269	req->sv1 = newSVsv (fh_or_path);
1270	req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1271
1272	if (SvPOK (fh_or_path))
1273	{
1274	req->type = EIO_TRUNCATE;
1275	req->ptr1 = SvPVbyte_nolen (req->sv1);
1276	}
1277	else
1278	{
1279	req->type = EIO_FTRUNCATE;
1280	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1281	}
1282
1283	EIO_SEND;
1284	}
1285
1286	void
1287	aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1288	ALIAS:
1289	aio_chmod = EIO_CHMOD
1290	aio_fchmod = EIO_FCHMOD
1291	aio_mkdir = EIO_MKDIR
1292	PPCODE:
1293	{
1294	dREQ;
1295
1296	req->type = type;
1297	req->int2 = mode;
1298	req->sv1 = newSVsv (fh_or_path);
1299
1300	if (ix == EIO_FCHMOD)
1301	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1302	else
1303	req->ptr1 = SvPVbyte_nolen (req->sv1);
1304
1305	EIO_SEND;
1306	}
1307
1308	void
1309	aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1310	PPCODE:
1311	{
1312	dREQ;
1313
1314	req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1315	req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1316	req->sv1 = newSVsv (fh_or_path);
1317
1318	if (SvPOK (fh_or_path))
1319	{
1320	req->type = EIO_CHOWN;
1321	req->ptr1 = SvPVbyte_nolen (req->sv1);
1322	}
1323	else
1324	{
1325	req->type = EIO_FCHOWN;
1326	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1327	}
1328
1329	EIO_SEND;
1330	}
1331
1332	void
1333	aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1334	ALIAS:
1335	aio_unlink = EIO_UNLINK
1336	aio_rmdir = EIO_RMDIR
1337	aio_readdir = EIO_READDIR
1338	PPCODE:
1339	{
1340	dREQ;
1341
1342	req->type = ix;
1343	req->sv1 = newSVsv (pathname);
1344	req->ptr1 = SvPVbyte_nolen (req->sv1);
1345
1346	EIO_SEND;
1347	}
1348
1349	void
1350	aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1351	ALIAS:
1352	aio_link = EIO_LINK
1353	aio_symlink = EIO_SYMLINK
1354	aio_rename = EIO_RENAME
1355	PPCODE:
1356	{
1357	dREQ;
1358
1359	req->type = ix;
1360	req->sv1 = newSVsv (oldpath);
1361	req->ptr1 = SvPVbyte_nolen (req->sv1);
1362	req->sv2 = newSVsv (newpath);
1363	req->ptr2 = SvPVbyte_nolen (req->sv2);
1364
1365	EIO_SEND;
1366	}
1367
1368	void
1369	aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1370	PPCODE:
1371	{
1372	dREQ;
1373
1374	req->type = EIO_MKNOD;
1375	req->sv1 = newSVsv (pathname);
1376	req->ptr1 = SvPVbyte_nolen (req->sv1);
1377	req->int2 = (mode_t)mode;
1378	req->offs = dev;
1379
1380	EIO_SEND;
1381	}
1382
1383	void
1384	aio_busy (double delay, SV *callback=&PL_sv_undef)
1385	PPCODE:
1386	{
1387	dREQ;
1388
1389	req->type = EIO_BUSY;
1390	req->nv1 = delay < 0. ? 0. : delay;
1391
1392	EIO_SEND;
1393	}
1394
1395	void
1396	aio_group (SV *callback=&PL_sv_undef)
1397	PROTOTYPE: ;$
1398	PPCODE:
1399	{
1400	dREQ;
1401
1402	req->type = EIO_GROUP;
1403
1404	req_send (req);
1405	XPUSHs (req_sv (req, AIO_GRP_KLASS));
1406	}
1407
1408	void
1409	aio_nop (SV *callback=&PL_sv_undef)
1410	ALIAS:
1411	aio_nop = EIO_NOP
1412	aio_sync = EIO_SYNC
1413	PPCODE:
1414	{
1415	dREQ;
1416
1417	#endif
1418
1419	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2293	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420	{	2294	{
1421	grp->int2 = limit;	2295	grp->int2 = limit;
1422	grp->feed = feed;	2296	grp->feed = feed;
1423		2297
1424	grp_try_feed (grp);	2298	grp_try_feed (grp);
1425	}	2299	}
1426		2300
		2301	void
1427	void eio_grp_limit (eio_req *grp, int limit)	2302	eio_grp_limit (eio_req *grp, int limit)
1428	{	2303	{
1429	grp->int2 = limit;	2304	grp->int2 = limit;
1430		2305
1431	grp_try_feed (grp);	2306	grp_try_feed (grp);
1432	}	2307	}
1433		2308
		2309	void
1434	void eio_grp_add (eio_req *grp, eio_req *req)	2310	eio_grp_add (eio_req *grp, eio_req *req)
1435	{	2311	{
1436	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2312	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
		2313
		2314	grp->flags |= EIO_FLAG_GROUPADD;
1437		2315
1438	++grp->size;	2316	++grp->size;
1439	req->grp = grp;	2317	req->grp = grp;
1440		2318
1441	req->grp_prev = 0;	2319	req->grp_prev = 0;
…		…
1445	grp->grp_first->grp_prev = req;	2323	grp->grp_first->grp_prev = req;
1446		2324
1447	grp->grp_first = req;	2325	grp->grp_first = req;
1448	}	2326	}
1449		2327
		2328	/*****************************************************************************/
		2329	/* misc garbage */
1450		2330
		2331	ssize_t
		2332	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
		2333	{
		2334	etp_worker wrk;
		2335	ssize_t ret;
		2336
		2337	wrk.dbuf = 0;
		2338
		2339	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
		2340
		2341	if (wrk.dbuf)
		2342	free (wrk.dbuf);
		2343
		2344	return ret;
		2345	}
		2346

Diff Legend

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