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 1.36 by root, Fri Jun 12 00:43:16 2009 UTC

		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libeio@schmorp.de>
		5	* All rights reserved.
		6	*
		7	* Redistribution and use in source and binary forms, with or without modifica-
		8	* tion, are permitted provided that the following conditions are met:
		9	*
		10	* 1. Redistributions of source code must retain the above copyright notice,
		11	* this list of conditions and the following disclaimer.
		12	*
		13	* 2. Redistributions in binary form must reproduce the above copyright
		14	* notice, this list of conditions and the following disclaimer in the
		15	* documentation and/or other materials provided with the distribution.
		16	*
		17	* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
		18	* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
		19	* CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
		20	* EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
		21	* CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
		22	* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
		23	* OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
		24	* WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
		25	* ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
		26	* OF THE POSSIBILITY OF SUCH DAMAGE.
		27	*
		28	* Alternatively, the contents of this file may be used under the terms of
		29	* the GNU General Public License ("GPL") version 2 or any later version,
		30	* in which case the provisions of the GPL are applicable instead of
		31	* the above. If you wish to allow the use of your version of this file
		32	* only under the terms of the GPL and not to allow others to use your
		33	* version of this file under the BSD license, indicate your decision
		34	* by deleting the provisions above and replace them with the notice
		35	* and other provisions required by the GPL. If you do not delete the
		36	* provisions above, a recipient may use your version of this file under
		37	* either the BSD or the GPL.
		38	*/
		39
1	#include "eio.h"	40	#include "eio.h"
		41
		42	#ifdef EIO_STACKSIZE
		43	# define XTHREAD_STACKSIZE EIO_STACKSIZE
		44	#endif
2	#include "xthread.h"	45	#include "xthread.h"
3		46
4	#include <errno.h>	47	#include <errno.h>
5
6	#include "EXTERN.h"
7	#include "perl.h"
8	#include "XSUB.h"
9
10	#include <stddef.h>	48	#include <stddef.h>
11	#include <stdlib.h>	49	#include <stdlib.h>
		50	#include <string.h>
12	#include <errno.h>	51	#include <errno.h>
13	#include <sys/types.h>	52	#include <sys/types.h>
14	#include <sys/stat.h>	53	#include <sys/stat.h>
15	#include <limits.h>	54	#include <limits.h>
16	#include <fcntl.h>	55	#include <fcntl.h>
17	#include <sched.h>	56	#include <assert.h>
18		57
19	#ifndef EIO_FINISH	58	#ifndef EIO_FINISH
20	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	59	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
21	#endif	60	#endif
22		61
…		…
29	#endif	68	#endif
30		69
31	#ifdef _WIN32	70	#ifdef _WIN32
32		71
33	/*doh*/	72	/*doh*/
34
35	#else	73	#else
36		74
37	# include "config.h"	75	# include "config.h"
38	# include <sys/time.h>	76	# include <sys/time.h>
39	# include <sys/select.h>	77	# include <sys/select.h>
40	# include <unistd.h>	78	# include <unistd.h>
41	# include <utime.h>	79	# include <utime.h>
42	# include <signal.h>	80	# include <signal.h>
		81	# include <dirent.h>
		82
		83	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		84	# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
		85	# define _DIRENT_HAVE_D_TYPE /* sigh */
		86	# define D_INO(de) (de)->d_fileno
		87	# define D_NAMLEN(de) (de)->d_namlen
		88	# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
		89	# define D_INO(de) (de)->d_ino
		90	# endif
		91
		92	#ifdef _D_EXACT_NAMLEN
		93	# undef D_NAMLEN
		94	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		95	#endif
		96
		97	# ifdef _DIRENT_HAVE_D_TYPE
		98	# define D_TYPE(de) (de)->d_type
		99	# endif
43		100
44	# ifndef EIO_STRUCT_DIRENT	101	# ifndef EIO_STRUCT_DIRENT
45	# define EIO_STRUCT_DIRENT struct dirent	102	# define EIO_STRUCT_DIRENT struct dirent
46	# endif	103	# endif
47		104
48	#endif	105	#endif
49
50	# ifndef EIO_STRUCT_STAT
51	# define EIO_STRUCT_STAT struct stat
52	# endif
53		106
54	#if HAVE_SENDFILE	107	#if HAVE_SENDFILE
55	# if __linux	108	# if __linux
56	# include <sys/sendfile.h>	109	# include <sys/sendfile.h>
57	# elif __freebsd	110	# elif __freebsd
…		…
64	# else	117	# else
65	# error sendfile support requested but not available	118	# error sendfile support requested but not available
66	# endif	119	# endif
67	#endif	120	#endif
68		121
		122	#ifndef D_TYPE
		123	# define D_TYPE(de) 0
		124	#endif
		125	#ifndef D_INO
		126	# define D_INO(de) 0
		127	#endif
		128	#ifndef D_NAMLEN
		129	# define D_NAMLEN(de) strlen ((de)->d_name)
		130	#endif
		131
69	/* number of seconds after which an idle threads exit */	132	/* number of seconds after which an idle threads exit */
70	#define IDLE_TIMEOUT 10	133	#define IDLE_TIMEOUT 10
71		134
72	/* used for struct dirent, AIX doesn't provide it */	135	/* used for struct dirent, AIX doesn't provide it */
73	#ifndef NAME_MAX	136	#ifndef NAME_MAX
…		…
77	/* buffer size for various temporary buffers */	140	/* buffer size for various temporary buffers */
78	#define EIO_BUFSIZE 65536	141	#define EIO_BUFSIZE 65536
79		142
80	#define dBUF \	143	#define dBUF \
81	char *eio_buf; \	144	char *eio_buf; \
82	X_LOCK (wrklock); \	145	ETP_WORKER_LOCK (self); \
83	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	146	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
84	X_UNLOCK (wrklock); \	147	ETP_WORKER_UNLOCK (self); \
		148	errno = ENOMEM; \
85	if (!eio_buf) \	149	if (!eio_buf) \
86	return -1;	150	return -1;
87		151
88	#define EIO_TICKS ((1000000 + 1023) >> 10)	152	#define EIO_TICKS ((1000000 + 1023) >> 10)
89		153
		154	/*****************************************************************************/
		155
		156	#if __GNUC__ >= 3
		157	# define expect(expr,value) __builtin_expect ((expr),(value))
		158	#else
		159	# define expect(expr,value) (expr)
		160	#endif
		161
		162	#define expect_false(expr) expect ((expr) != 0, 0)
		163	#define expect_true(expr) expect ((expr) != 0, 1)
		164
		165	/*****************************************************************************/
		166
		167	#define ETP_PRI_MIN EIO_PRI_MIN
		168	#define ETP_PRI_MAX EIO_PRI_MAX
		169
		170	struct etp_worker;
		171
		172	#define ETP_REQ eio_req
		173	#define ETP_DESTROY(req) eio_destroy (req)
		174	static int eio_finish (eio_req *req);
		175	#define ETP_FINISH(req) eio_finish (req)
		176	static void eio_execute (struct etp_worker *self, eio_req *req);
		177	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		178
		179	#define ETP_WORKER_CLEAR(req) \
		180	if (wrk->dbuf) \
		181	{ \
		182	free (wrk->dbuf); \
		183	wrk->dbuf = 0; \
		184	} \
		185	\
		186	if (wrk->dirp) \
		187	{ \
		188	closedir (wrk->dirp); \
		189	wrk->dirp = 0; \
		190	}
		191
		192	#define ETP_WORKER_COMMON \
		193	void *dbuf; \
		194	DIR *dirp;
		195
		196	/*****************************************************************************/
		197
		198	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		199
		200	/* calculcate time difference in ~1/EIO_TICKS of a second */
		201	static int tvdiff (struct timeval *tv1, struct timeval *tv2)
		202	{
		203	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
		204	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
		205	}
		206
		207	static unsigned int started, idle, wanted = 4;
		208
90	static void (*want_poll_cb) (void);	209	static void (*want_poll_cb) (void);
91	static void (*done_poll_cb) (void);	210	static void (*done_poll_cb) (void);
92		211
93	static unsigned int max_poll_time = 0;	212	static unsigned int max_poll_time; /* reslock */
94	static unsigned int max_poll_reqs = 0;	213	static unsigned int max_poll_reqs; /* reslock */
95		214
96	/* calculcate time difference in ~1/EIO_TICKS of a second */	215	static volatile unsigned int nreqs; /* reqlock */
97	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	216	static volatile unsigned int nready; /* reqlock */
98	{	217	static volatile unsigned int npending; /* reqlock */
99	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	218	static volatile unsigned int max_idle = 4;
100	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
101	}
102		219
103	static unsigned int started, idle, wanted;
104
105	/* worker threads management */
106	static mutex_t wrklock = X_MUTEX_INIT;	220	static mutex_t wrklock = X_MUTEX_INIT;
107
108	typedef struct worker {
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)
123	{
124	if (wrk->dirp)
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	}
136
137	static void worker_free (worker *wrk)
138	{
139	wrk->next->prev = wrk->prev;
140	wrk->prev->next = wrk->next;
141
142	free (wrk);
143	}
144
145	static volatile unsigned int nreqs, nready, npending;
146	static volatile unsigned int max_idle = 4;
147
148	static mutex_t reslock = X_MUTEX_INIT;	221	static mutex_t reslock = X_MUTEX_INIT;
149	static mutex_t reqlock = X_MUTEX_INIT;	222	static mutex_t reqlock = X_MUTEX_INIT;
150	static cond_t reqwait = X_COND_INIT;	223	static cond_t reqwait = X_COND_INIT;
151		224
		225	#if !HAVE_PREADWRITE
		226	/*
		227	* make our pread/pwrite emulation safe against themselves, but not against
		228	* normal read/write by using a mutex. slows down execution a lot,
		229	* but that's your problem, not mine.
		230	*/
		231	static mutex_t preadwritelock = X_MUTEX_INIT;
		232	#endif
		233
		234	typedef struct etp_worker
		235	{
		236	/* locked by wrklock */
		237	struct etp_worker *prev, *next;
		238
		239	thread_t tid;
		240
		241	/* locked by reslock, reqlock or wrklock */
		242	ETP_REQ *req; /* currently processed request */
		243
		244	ETP_WORKER_COMMON
		245	} etp_worker;
		246
		247	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
		248
		249	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		250	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
		251
		252	/* worker threads management */
		253
		254	static void etp_worker_clear (etp_worker *wrk)
		255	{
		256	ETP_WORKER_CLEAR (wrk);
		257	}
		258
		259	static void etp_worker_free (etp_worker *wrk)
		260	{
		261	wrk->next->prev = wrk->prev;
		262	wrk->prev->next = wrk->next;
		263
		264	free (wrk);
		265	}
		266
152	unsigned int eio_nreqs (void)	267	static unsigned int etp_nreqs (void)
153	{	268	{
		269	int retval;
		270	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		271	retval = nreqs;
		272	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154	return nreqs;	273	return retval;
155	}	274	}
156		275
157	unsigned int eio_nready (void)	276	static unsigned int etp_nready (void)
158	{	277	{
159	unsigned int retval;	278	unsigned int retval;
160		279
161	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	280	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162	retval = nready;	281	retval = nready;
163	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	282	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164		283
165	return retval;	284	return retval;
166	}	285	}
167		286
168	unsigned int eio_npending (void)	287	static unsigned int etp_npending (void)
169	{	288	{
170	unsigned int retval;	289	unsigned int retval;
171		290
172	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	291	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173	retval = npending;	292	retval = npending;
174	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	293	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175		294
176	return retval;	295	return retval;
177	}	296	}
178		297
179	unsigned int eio_nthreads (void)	298	static unsigned int etp_nthreads (void)
180	{	299	{
181	unsigned int retval;	300	unsigned int retval;
182		301
183	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	302	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184	retval = started;	303	retval = started;
…		…
191	* a somewhat faster data structure might be nice, but	310	* a somewhat faster data structure might be nice, but
192	* with 8 priorities this actually needs <20 insns	311	* with 8 priorities this actually needs <20 insns
193	* per shift, the most expensive operation.	312	* per shift, the most expensive operation.
194	*/	313	*/
195	typedef struct {	314	typedef struct {
196	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	315	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197	int size;	316	int size;
198	} reqq;	317	} etp_reqq;
199		318
200	static reqq req_queue;	319	static etp_reqq req_queue;
201	static reqq res_queue;	320	static etp_reqq res_queue;
202		321
203	static int reqq_push (reqq *q, eio_req *req)	322	static int reqq_push (etp_reqq *q, ETP_REQ *req)
204	{	323	{
205	int pri = req->pri;	324	int pri = req->pri;
206	req->next = 0;	325	req->next = 0;
207		326
208	if (q->qe[pri])	327	if (q->qe[pri])
…		…
214	q->qe[pri] = q->qs[pri] = req;	333	q->qe[pri] = q->qs[pri] = req;
215		334
216	return q->size++;	335	return q->size++;
217	}	336	}
218		337
219	static eio_req *reqq_shift (reqq *q)	338	static ETP_REQ *reqq_shift (etp_reqq *q)
220	{	339	{
221	int pri;	340	int pri;
222		341
223	if (!q->size)	342	if (!q->size)
224	return 0;	343	return 0;
225		344
226	--q->size;	345	--q->size;
227		346
228	for (pri = EIO_NUM_PRI; pri--; )	347	for (pri = ETP_NUM_PRI; pri--; )
229	{	348	{
230	eio_req *req = q->qs[pri];	349	eio_req *req = q->qs[pri];
231		350
232	if (req)	351	if (req)
233	{	352	{
…		…
239	}	358	}
240		359
241	abort ();	360	abort ();
242	}	361	}
243		362
244	static void grp_try_feed (eio_req *grp)	363	static void etp_atfork_prepare (void)
245	{	364	{
246	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	365	X_LOCK (wrklock);
		366	X_LOCK (reqlock);
		367	X_LOCK (reslock);
		368	#if !HAVE_PREADWRITE
		369	X_LOCK (preadwritelock);
		370	#endif
		371	}
		372
		373	static void etp_atfork_parent (void)
		374	{
		375	#if !HAVE_PREADWRITE
		376	X_UNLOCK (preadwritelock);
		377	#endif
		378	X_UNLOCK (reslock);
		379	X_UNLOCK (reqlock);
		380	X_UNLOCK (wrklock);
		381	}
		382
		383	static void etp_atfork_child (void)
		384	{
		385	ETP_REQ *prv;
		386
		387	while ((prv = reqq_shift (&req_queue)))
		388	ETP_DESTROY (prv);
		389
		390	while ((prv = reqq_shift (&res_queue)))
		391	ETP_DESTROY (prv);
		392
		393	while (wrk_first.next != &wrk_first)
247	{	394	{
248	int old_len = grp->size;	395	etp_worker *wrk = wrk_first.next;
249		396
250	EIO_FEED (grp);	397	if (wrk->req)
		398	ETP_DESTROY (wrk->req);
251		399
252	/* stop if no progress has been made */	400	etp_worker_clear (wrk);
253	if (old_len == grp->size)	401	etp_worker_free (wrk);
254	{
255	grp->feed = 0;
256	break;
257	}
258	}	402	}
259	}
260		403
261	static int eio_finish (eio_req *req);	404	started = 0;
		405	idle = 0;
		406	nreqs = 0;
		407	nready = 0;
		408	npending = 0;
262		409
263	static int grp_dec (eio_req *grp)	410	etp_atfork_parent ();
264	{	411	}
265	--grp->size;
266		412
267	/* call feeder, if applicable */	413	static void
268	grp_try_feed (grp);	414	etp_once_init (void)
		415	{
		416	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
		417	}
269		418
270	/* finish, if done */	419	static int
271	if (!grp->size && grp->int1)	420	etp_init (void (*want_poll)(void), void (*done_poll)(void))
272	return eio_finish (grp);	421	{
273	else	422	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
		423
		424	pthread_once (&doinit, etp_once_init);
		425
		426	want_poll_cb = want_poll;
		427	done_poll_cb = done_poll;
		428
274	return 0;	429	return 0;
275	}	430	}
276		431
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);	432	X_THREAD_PROC (etp_proc);
326		433
327	static void start_thread (void)	434	static void etp_start_thread (void)
328	{	435	{
329	worker *wrk = calloc (1, sizeof (worker));	436	etp_worker *wrk = calloc (1, sizeof (etp_worker));
330		437
331	if (!wrk)	438	/*TODO*/
332	croak ("unable to allocate worker thread data");	439	assert (("unable to allocate worker thread data", wrk));
333		440
334	X_LOCK (wrklock);	441	X_LOCK (wrklock);
335		442
336	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	443	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
337	{	444	{
338	wrk->prev = &wrk_first;	445	wrk->prev = &wrk_first;
339	wrk->next = wrk_first.next;	446	wrk->next = wrk_first.next;
340	wrk_first.next->prev = wrk;	447	wrk_first.next->prev = wrk;
341	wrk_first.next = wrk;	448	wrk_first.next = wrk;
…		…
345	free (wrk);	452	free (wrk);
346		453
347	X_UNLOCK (wrklock);	454	X_UNLOCK (wrklock);
348	}	455	}
349		456
350	static void maybe_start_thread (void)	457	static void etp_maybe_start_thread (void)
351	{	458	{
352	if (eio_nthreads () >= wanted)	459	if (expect_true (etp_nthreads () >= wanted))
353	return;	460	return;
354		461
355	/* todo: maybe use idle here, but might be less exact */	462	/* todo: maybe use idle here, but might be less exact */
356	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	463	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
357	return;	464	return;
358		465
359	start_thread ();	466	etp_start_thread ();
360	}	467	}
361		468
362	void eio_submit (eio_req *req)
363	{
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)	469	static void etp_end_thread (void)
376	{	470	{
377	eio_req *req = calloc (1, sizeof (eio_req));	471	eio_req *req = calloc (1, sizeof (eio_req));
378		472
379	req->type = EIO_QUIT;	473	req->type = -1;
380	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	474	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381		475
382	X_LOCK (reqlock);	476	X_LOCK (reqlock);
383	reqq_push (&req_queue, req);	477	reqq_push (&req_queue, req);
384	X_COND_SIGNAL (reqwait);	478	X_COND_SIGNAL (reqwait);
385	X_UNLOCK (reqlock);	479	X_UNLOCK (reqlock);
…		…
387	X_LOCK (wrklock);	481	X_LOCK (wrklock);
388	--started;	482	--started;
389	X_UNLOCK (wrklock);	483	X_UNLOCK (wrklock);
390	}	484	}
391		485
392	void eio_set_max_poll_time (double nseconds)
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)	486	static int etp_poll (void)
429	{	487	{
430	int maxreqs = max_poll_reqs;	488	unsigned int maxreqs;
		489	unsigned int maxtime;
431	struct timeval tv_start, tv_now;	490	struct timeval tv_start, tv_now;
432	eio_req *req;
433		491
		492	X_LOCK (reslock);
		493	maxreqs = max_poll_reqs;
		494	maxtime = max_poll_time;
		495	X_UNLOCK (reslock);
		496
434	if (max_poll_time)	497	if (maxtime)
435	gettimeofday (&tv_start, 0);	498	gettimeofday (&tv_start, 0);
436		499
437	for (;;)	500	for (;;)
438	{	501	{
		502	ETP_REQ *req;
		503
439	maybe_start_thread ();	504	etp_maybe_start_thread ();
440		505
441	X_LOCK (reslock);	506	X_LOCK (reslock);
442	req = reqq_shift (&res_queue);	507	req = reqq_shift (&res_queue);
443		508
444	if (req)	509	if (req)
…		…
452	X_UNLOCK (reslock);	517	X_UNLOCK (reslock);
453		518
454	if (!req)	519	if (!req)
455	return 0;	520	return 0;
456		521
		522	X_LOCK (reqlock);
457	--nreqs;	523	--nreqs;
		524	X_UNLOCK (reqlock);
458		525
459	if (req->type == EIO_GROUP && req->size)	526	if (expect_false (req->type == EIO_GROUP && req->size))
460	{	527	{
461	req->int1 = 1; /* mark request as delayed */	528	req->int1 = 1; /* mark request as delayed */
462	continue;	529	continue;
463	}	530	}
464	else	531	else
465	{	532	{
466	int res = eio_finish (req);	533	int res = ETP_FINISH (req);
467	if (res)	534	if (expect_false (res))
468	return res;	535	return res;
469	}	536	}
470		537
471	if (maxreqs && !--maxreqs)	538	if (expect_false (maxreqs && !--maxreqs))
472	break;	539	break;
473		540
474	if (max_poll_time)	541	if (maxtime)
475	{	542	{
476	gettimeofday (&tv_now, 0);	543	gettimeofday (&tv_now, 0);
477		544
478	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	545	if (tvdiff (&tv_start, &tv_now) >= maxtime)
479	break;	546	break;
480	}	547	}
481	}	548	}
482		549
483	errno = EAGAIN;	550	errno = EAGAIN;
484	return -1;	551	return -1;
485	}	552	}
486		553
		554	static void etp_cancel (ETP_REQ *req)
		555	{
		556	X_LOCK (wrklock);
		557	req->flags |= EIO_FLAG_CANCELLED;
		558	X_UNLOCK (wrklock);
		559
		560	eio_grp_cancel (req);
		561	}
		562
		563	static void etp_submit (ETP_REQ *req)
		564	{
		565	req->pri -= ETP_PRI_MIN;
		566
		567	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		568	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		569
		570	if (expect_false (req->type == EIO_GROUP))
		571	{
		572	/* I hope this is worth it :/ */
		573	X_LOCK (reqlock);
		574	++nreqs;
		575	X_UNLOCK (reqlock);
		576
		577	X_LOCK (reslock);
		578
		579	++npending;
		580
		581	if (!reqq_push (&res_queue, req) && want_poll_cb)
		582	want_poll_cb ();
		583
		584	X_UNLOCK (reslock);
		585	}
		586	else
		587	{
		588	X_LOCK (reqlock);
		589	++nreqs;
		590	++nready;
		591	reqq_push (&req_queue, req);
		592	X_COND_SIGNAL (reqwait);
		593	X_UNLOCK (reqlock);
		594
		595	etp_maybe_start_thread ();
		596	}
		597	}
		598
		599	static void etp_set_max_poll_time (double nseconds)
		600	{
		601	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		602	max_poll_time = nseconds;
		603	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		604	}
		605
		606	static void etp_set_max_poll_reqs (unsigned int maxreqs)
		607	{
		608	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		609	max_poll_reqs = maxreqs;
		610	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		611	}
		612
		613	static void etp_set_max_idle (unsigned int nthreads)
		614	{
		615	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		616	max_idle = nthreads <= 0 ? 1 : nthreads;
		617	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		618	}
		619
		620	static void etp_set_min_parallel (unsigned int nthreads)
		621	{
		622	if (wanted < nthreads)
		623	wanted = nthreads;
		624	}
		625
		626	static void etp_set_max_parallel (unsigned int nthreads)
		627	{
		628	if (wanted > nthreads)
		629	wanted = nthreads;
		630
		631	while (started > wanted)
		632	etp_end_thread ();
		633	}
		634
		635	/*****************************************************************************/
		636
		637	static void grp_try_feed (eio_req *grp)
		638	{
		639	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		640	{
		641	grp->flags &= ~EIO_FLAG_GROUPADD;
		642
		643	EIO_FEED (grp);
		644
		645	/* stop if no progress has been made */
		646	if (!(grp->flags & EIO_FLAG_GROUPADD))
		647	{
		648	grp->feed = 0;
		649	break;
		650	}
		651	}
		652	}
		653
		654	static int grp_dec (eio_req *grp)
		655	{
		656	--grp->size;
		657
		658	/* call feeder, if applicable */
		659	grp_try_feed (grp);
		660
		661	/* finish, if done */
		662	if (!grp->size && grp->int1)
		663	return eio_finish (grp);
		664	else
		665	return 0;
		666	}
		667
		668	void eio_destroy (eio_req *req)
		669	{
		670	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		671	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		672
		673	EIO_DESTROY (req);
		674	}
		675
		676	static int eio_finish (eio_req *req)
		677	{
		678	int res = EIO_FINISH (req);
		679
		680	if (req->grp)
		681	{
		682	int res2;
		683	eio_req *grp = req->grp;
		684
		685	/* unlink request */
		686	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		687	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		688
		689	if (grp->grp_first == req)
		690	grp->grp_first = req->grp_next;
		691
		692	res2 = grp_dec (grp);
		693
		694	if (!res && res2)
		695	res = res2;
		696	}
		697
		698	eio_destroy (req);
		699
		700	return res;
		701	}
		702
		703	void eio_grp_cancel (eio_req *grp)
		704	{
		705	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		706	eio_cancel (grp);
		707	}
		708
		709	void eio_cancel (eio_req *req)
		710	{
		711	etp_cancel (req);
		712	}
		713
		714	void eio_submit (eio_req *req)
		715	{
		716	etp_submit (req);
		717	}
		718
		719	unsigned int eio_nreqs (void)
		720	{
		721	return etp_nreqs ();
		722	}
		723
		724	unsigned int eio_nready (void)
		725	{
		726	return etp_nready ();
		727	}
		728
		729	unsigned int eio_npending (void)
		730	{
		731	return etp_npending ();
		732	}
		733
		734	unsigned int eio_nthreads (void)
		735	{
		736	return etp_nthreads ();
		737	}
		738
		739	void eio_set_max_poll_time (double nseconds)
		740	{
		741	etp_set_max_poll_time (nseconds);
		742	}
		743
		744	void eio_set_max_poll_reqs (unsigned int maxreqs)
		745	{
		746	etp_set_max_poll_reqs (maxreqs);
		747	}
		748
		749	void eio_set_max_idle (unsigned int nthreads)
		750	{
		751	etp_set_max_idle (nthreads);
		752	}
		753
		754	void eio_set_min_parallel (unsigned int nthreads)
		755	{
		756	etp_set_min_parallel (nthreads);
		757	}
		758
		759	void eio_set_max_parallel (unsigned int nthreads)
		760	{
		761	etp_set_max_parallel (nthreads);
		762	}
		763
		764	int eio_poll (void)
		765	{
		766	return etp_poll ();
		767	}
		768
487	/*****************************************************************************/	769	/*****************************************************************************/
488	/* work around various missing functions */	770	/* work around various missing functions */
489		771
490	#if !HAVE_PREADWRITE	772	#if !HAVE_PREADWRITE
		773	# undef pread
		774	# undef pwrite
491	# define pread aio_pread	775	# define pread eio__pread
492	# define pwrite aio_pwrite	776	# define pwrite eio__pwrite
493		777
494	/*	778	static ssize_t
495	* make our pread/pwrite safe against themselves, but not against	779	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	{	780	{
503	ssize_t res;	781	ssize_t res;
504	off_t ooffset;	782	off_t ooffset;
505		783
506	X_LOCK (preadwritelock);	784	X_LOCK (preadwritelock);
…		…
511	X_UNLOCK (preadwritelock);	789	X_UNLOCK (preadwritelock);
512		790
513	return res;	791	return res;
514	}	792	}
515		793
516	static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset)	794	static ssize_t
		795	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
517	{	796	{
518	ssize_t res;	797	ssize_t res;
519	off_t ooffset;	798	off_t ooffset;
520		799
521	X_LOCK (preadwritelock);	800	X_LOCK (preadwritelock);
522	ooffset = lseek (fd, 0, SEEK_CUR);	801	ooffset = lseek (fd, 0, SEEK_CUR);
523	lseek (fd, offset, SEEK_SET);	802	lseek (fd, offset, SEEK_SET);
524	res = write (fd, buf, count);	803	res = write (fd, buf, count);
525	lseek (fd, offset, SEEK_SET);	804	lseek (fd, ooffset, SEEK_SET);
526	X_UNLOCK (preadwritelock);	805	X_UNLOCK (preadwritelock);
527		806
528	return res;	807	return res;
529	}	808	}
530	#endif	809	#endif
531		810
532	#ifndef HAVE_FUTIMES	811	#ifndef HAVE_FUTIMES
533		812
		813	# undef utimes
		814	# undef futimes
534	# define utimes(path,times) aio_utimes (path, times)	815	# define utimes(path,times) eio__utimes (path, times)
535	# define futimes(fd,times) aio_futimes (fd, times)	816	# define futimes(fd,times) eio__futimes (fd, times)
536		817
		818	static int
537	static int aio_utimes (const char *filename, const struct timeval times[2])	819	eio__utimes (const char *filename, const struct timeval times[2])
538	{	820	{
539	if (times)	821	if (times)
540	{	822	{
541	struct utimbuf buf;	823	struct utimbuf buf;
542		824
…		…
547	}	829	}
548	else	830	else
549	return utime (filename, 0);	831	return utime (filename, 0);
550	}	832	}
551		833
552	static int aio_futimes (int fd, const struct timeval tv[2])	834	static int eio__futimes (int fd, const struct timeval tv[2])
553	{	835	{
554	errno = ENOSYS;	836	errno = ENOSYS;
555	return -1;	837	return -1;
556	}	838	}
557		839
558	#endif	840	#endif
559		841
560	#if !HAVE_FDATASYNC	842	#if !HAVE_FDATASYNC
		843	# undef fdatasync
561	# define fdatasync fsync	844	# define fdatasync(fd) fsync (fd)
562	#endif	845	#endif
		846
		847	/* sync_file_range always needs emulation */
		848	int
		849	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		850	{
		851	#if HAVE_SYNC_FILE_RANGE
		852	int res;
		853
		854	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		855	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		856	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		857	{
		858	flags = 0
		859	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		860	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		861	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		862	}
		863
		864	res = sync_file_range (fd, offset, nbytes, flags);
		865
		866	if (!res || errno != ENOSYS)
		867	return res;
		868	#endif
		869
		870	/* even though we could play tricks with the flags, it's better to always
		871	* call fdatasync, as thta matches the expectation of it's users best */
		872	return fdatasync (fd);
		873	}
563		874
564	#if !HAVE_READAHEAD	875	#if !HAVE_READAHEAD
		876	# undef readahead
565	# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self)	877	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566		878
		879	static ssize_t
567	static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self)	880	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568	{	881	{
569	size_t todo = count;	882	size_t todo = count;
570	dBUF;	883	dBUF;
571		884
572	while (todo > 0)	885	while (todo > 0)
573	{	886	{
574	size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;	887	size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575		888
576	pread (fd, aio_buf, len, offset);	889	pread (fd, eio_buf, len, offset);
577	offset += len;	890	offset += len;
578	todo -= len;	891	todo -= len;
579	}	892	}
580		893
581	errno = 0;	894	errno = 0;
582	return count;	895	return count;
583	}	896	}
584		897
585	#endif	898	#endif
586		899
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 */	900	/* sendfile always needs emulation */
		901	static ssize_t
618	static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self)	902	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
619	{	903	{
620	ssize_t res;	904	ssize_t res;
621		905
622	if (!count)	906	if (!count)
623	return 0;	907	return 0;
…		…
705	}	989	}
706		990
707	return res;	991	return res;
708	}	992	}
709		993
		994	static int
		995	eio_dent_cmp (const void *a_, const void *b_)
		996	{
		997	const eio_dirent *a = (const eio_dirent *)a_;
		998	const eio_dirent *b = (const eio_dirent *)b_;
		999
		1000	return (int)b->score - (int)a->score ? (int)b->score - (int)a->score
		1001	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */
		1002	}
		1003
710	/* read a full directory */	1004	/* read a full directory */
		1005	static void
711	static void scandir_ (eio_req *req, worker *self)	1006	eio__scandir (eio_req *req, etp_worker *self)
712	{	1007	{
713	DIR *dirp;	1008	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;	1009	EIO_STRUCT_DIRENT *entp;
720	char *name, *names;	1010	unsigned char *name, *names;
721	int memlen = 4096;	1011	int namesalloc = 4096;
722	int memofs = 0;	1012	int namesoffs = 0;
		1013	int flags = req->int1;
		1014	eio_dirent *dents = 0;
		1015	int dentalloc = 128;
723	int res = 0;	1016	int dentoffs = 0;
		1017
		1018	req->result = -1;
		1019
		1020	if (!(flags & EIO_READDIR_DENTS))
		1021	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
724		1022
725	X_LOCK (wrklock);	1023	X_LOCK (wrklock);
		1024	/* the corresponding closedir is in ETP_WORKER_CLEAR */
726	self->dirp = dirp = opendir (req->ptr1);	1025	self->dirp = dirp = opendir (req->ptr1);
727	self->dbuf = u = malloc (sizeof (*u));
728	req->flags |= EIO_FLAG_PTR2_FREE;	1026	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
729	req->ptr2 = names = malloc (memlen);	1027	req->ptr1 = names = malloc (namesalloc);
		1028	req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
730	X_UNLOCK (wrklock);	1029	X_UNLOCK (wrklock);
731		1030
732	if (dirp && u && names)	1031	if (dirp && names && (!flags || dents))
733	for (;;)	1032	for (;;)
734	{	1033	{
735	errno = 0;	1034	errno = 0;
736	readdir_r (dirp, &u->d, &entp);	1035	entp = readdir (dirp);
737		1036
738	if (!entp)	1037	if (!entp)
		1038	{
		1039	if (errno)
		1040	break;
		1041
		1042	/* sort etc. */
		1043	req->int1 = flags;
		1044	req->result = dentoffs;
		1045
		1046	if (dents)
		1047	{
		1048	eio_dirent *ent = dents + dentoffs;
		1049
		1050	while (ent > dents)
		1051	(--ent)->name = names + (size_t)ent->name;
		1052	}
		1053
		1054	if (flags & EIO_READDIR_STAT_ORDER
		1055	|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))
		1056	{
		1057	/* pray your qsort doesn't use quicksort */
		1058	qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
		1059	}
		1060	else if (flags & EIO_READDIR_DIRS_FIRST)
		1061	{
		1062	/* in this case, all is known, and we just put dirs first and sort them */
		1063	eio_dirent *ent = dents + dentoffs;
		1064	eio_dirent *dir = dents;
		1065
		1066	/* now move dirs to the front, and non-dirs to the back */
		1067	/* by walkign from both sides and swapping if necessary */
		1068	while (ent > dir)
		1069	{
		1070	if (dir->type == DT_DIR)
		1071	++dir;
		1072	else
		1073	{
		1074	--ent;
		1075
		1076	if (ent->type == DT_DIR)
		1077	{
		1078	eio_dirent tmp = *dir;
		1079	*dir = *ent;
		1080	*ent = tmp;
		1081
		1082	++dir;
		1083	}
		1084	}
		1085	}
		1086
		1087	/* now sort the dirs only */
		1088	qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
		1089	}
		1090
		1091	/* only provide the names array unless DENTS is specified */
		1092	if (!(flags & EIO_READDIR_DENTS))
		1093	{
		1094	X_LOCK (wrklock);
		1095	assert (!dents);
		1096	req->ptr1 = 0;
		1097	req->ptr2 = names;
		1098	X_UNLOCK (wrklock);
		1099	}
		1100
739	break;	1101	break;
		1102	}
740		1103
		1104	/* now add the entry to our list(s) */
741	name = entp->d_name;	1105	name = entp->d_name;
742		1106
		1107	/* skip . and .. entries */
743	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1108	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
744	{	1109	{
745	int len = strlen (name) + 1;	1110	int len = D_NAMLEN (entp) + 1;
746		1111
747	res++;	1112	while (expect_false (namesoffs + len > namesalloc))
748
749	while (memofs + len > memlen)
750	{	1113	{
751	memlen *= 2;	1114	namesalloc *= 2;
752	X_LOCK (wrklock);	1115	X_LOCK (wrklock);
753	req->ptr2 = names = realloc (names, memlen);	1116	req->ptr1 = names = realloc (names, namesalloc);
754	X_UNLOCK (wrklock);	1117	X_UNLOCK (wrklock);
755		1118
756	if (!names)	1119	if (!names)
757	break;	1120	break;
758	}	1121	}
759		1122
760	memcpy (names + memofs, name, len);	1123	memcpy (names + namesoffs, name, len);
		1124
		1125	if (dents)
		1126	{
		1127	struct eio_dirent *ent;
		1128
		1129	if (expect_false (dentoffs == dentalloc))
		1130	{
		1131	dentalloc *= 2;
		1132	X_LOCK (wrklock);
		1133	req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1134	X_UNLOCK (wrklock);
		1135
		1136	if (!dents)
		1137	break;
		1138	}
		1139
		1140	ent = dents + dentoffs;
		1141
		1142	ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */
		1143	ent->namelen = len - 1;
		1144	ent->inode = D_INO (entp);
		1145
		1146	switch (D_TYPE (entp))
		1147	{
		1148	default:
		1149	ent->type = EIO_DT_UNKNOWN;
		1150	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1151	break;
		1152
		1153	#ifdef DT_FIFO
		1154	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1155	#endif
		1156	#ifdef DT_CHR
		1157	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1158	#endif
		1159	#ifdef DT_MPC
		1160	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1161	#endif
		1162	#ifdef DT_DIR
		1163	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1164	#endif
		1165	#ifdef DT_NAM
		1166	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1167	#endif
		1168	#ifdef DT_BLK
		1169	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1170	#endif
		1171	#ifdef DT_MPB
		1172	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1173	#endif
		1174	#ifdef DT_REG
		1175	case DT_REG: ent->type = EIO_DT_REG; break;
		1176	#endif
		1177	#ifdef DT_NWK
		1178	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1179	#endif
		1180	#ifdef DT_CMP
		1181	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1182	#endif
		1183	#ifdef DT_LNK
		1184	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1185	#endif
		1186	#ifdef DT_SOCK
		1187	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1188	#endif
		1189	#ifdef DT_DOOR
		1190	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1191	#endif
		1192	#ifdef DT_WHT
		1193	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1194	#endif
		1195	}
		1196
		1197	ent->score = 0;
		1198
		1199	if (flags & EIO_READDIR_DIRS_FIRST)
		1200	{
		1201	if (ent->type == EIO_DT_UNKNOWN)
		1202	{
		1203	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1204	ent->score = 98;
		1205	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1206	ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */
		1207	}
		1208	else if (ent->type == EIO_DT_DIR)
		1209	ent->score = 100;
		1210	}
		1211	}
		1212
761	memofs += len;	1213	namesoffs += len;
		1214	++dentoffs;
		1215	}
		1216
		1217	if (EIO_CANCELLED (req))
		1218	{
		1219	errno = ECANCELED;
		1220	break;
762	}	1221	}
763	}	1222	}
		1223	}
764		1224
765	if (errno)	1225	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
766	res = -1;	1226	# undef msync
767		1227	# define msync(a,b,c) ((errno = ENOSYS), -1)
768	req->result = res;	1228	#endif
		1229
		1230	int
		1231	eio__mtouch (void *mem, size_t len, int flags)
		1232	{
		1233	intptr_t addr = (intptr_t)mem;
		1234	intptr_t end = addr + len;
		1235	#ifdef PAGESIZE
		1236	const intptr_t page = PAGESIZE;
		1237	#else
		1238	static intptr_t page;
		1239
		1240	if (!page)
		1241	page = sysconf (_SC_PAGESIZE);
		1242	#endif
		1243
		1244	addr &= ~(page - 1); /* assume page size is always a power of two */
		1245
		1246	if (addr < end)
		1247	if (flags) /* modify */
		1248	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
		1249	else
		1250	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
		1251
		1252	return 0;
769	}	1253	}
770		1254
771	/*****************************************************************************/	1255	/*****************************************************************************/
772		1256
		1257	#define ALLOC(len) \
		1258	if (!req->ptr2) \
		1259	{ \
		1260	X_LOCK (wrklock); \
		1261	req->flags |= EIO_FLAG_PTR2_FREE; \
		1262	X_UNLOCK (wrklock); \
		1263	req->ptr2 = malloc (len); \
		1264	if (!req->ptr2) \
		1265	{ \
		1266	errno = ENOMEM; \
		1267	req->result = -1; \
		1268	break; \
		1269	} \
		1270	}
		1271
773	X_THREAD_PROC (eio_proc)	1272	X_THREAD_PROC (etp_proc)
774	{	1273	{
775	eio_req *req;	1274	ETP_REQ *req;
776	struct timespec ts;	1275	struct timespec ts;
777	worker *self = (worker *)thr_arg;	1276	etp_worker *self = (etp_worker *)thr_arg;
778		1277
779	/* try to distribute timeouts somewhat randomly */	1278	/* try to distribute timeouts somewhat randomly */
780	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1279	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781		1280
782	for (;;)	1281	for (;;)
783	{	1282	{
784	ts.tv_sec = time (0) + IDLE_TIMEOUT;
785
786	X_LOCK (reqlock);	1283	X_LOCK (reqlock);
787		1284
788	for (;;)	1285	for (;;)
789	{	1286	{
790	self->req = req = reqq_shift (&req_queue);	1287	self->req = req = reqq_shift (&req_queue);
…		…
792	if (req)	1289	if (req)
793	break;	1290	break;
794		1291
795	++idle;	1292	++idle;
796		1293
		1294	ts.tv_sec = time (0) + IDLE_TIMEOUT;
797	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1295	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
798	{	1296	{
799	if (idle > max_idle)	1297	if (idle > max_idle)
800	{	1298	{
801	--idle;	1299	--idle;
…		…
806	goto quit;	1304	goto quit;
807	}	1305	}
808		1306
809	/* we are allowed to idle, so do so without any timeout */	1307	/* we are allowed to idle, so do so without any timeout */
810	X_COND_WAIT (reqwait, reqlock);	1308	X_COND_WAIT (reqwait, reqlock);
811	ts.tv_sec = time (0) + IDLE_TIMEOUT;
812	}	1309	}
813		1310
814	--idle;	1311	--idle;
815	}	1312	}
816		1313
817	--nready;	1314	--nready;
818		1315
819	X_UNLOCK (reqlock);	1316	X_UNLOCK (reqlock);
820		1317
821	errno = 0; /* strictly unnecessary */	1318	if (req->type < 0)
		1319	goto quit;
822		1320
823	if (!EIO_CANCELLED (req))	1321	if (!EIO_CANCELLED (req))
824	switch (req->type)	1322	ETP_EXECUTE (self, req);
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;
910
911	default:
912	req->result = -1;
913	break;
914	}
915
916	req->errorno = errno;
917		1323
918	X_LOCK (reslock);	1324	X_LOCK (reslock);
919		1325
920	++npending;	1326	++npending;
921		1327
922	if (!reqq_push (&res_queue, req) && want_poll_cb)	1328	if (!reqq_push (&res_queue, req) && want_poll_cb)
923	want_poll_cb ();	1329	want_poll_cb ();
924		1330
925	self->req = 0;	1331	self->req = 0;
926	worker_clear (self);	1332	etp_worker_clear (self);
927		1333
928	X_UNLOCK (reslock);	1334	X_UNLOCK (reslock);
929	}	1335	}
930		1336
931	quit:	1337	quit:
932	X_LOCK (wrklock);	1338	X_LOCK (wrklock);
933	worker_free (self);	1339	etp_worker_free (self);
934	X_UNLOCK (wrklock);	1340	X_UNLOCK (wrklock);
935		1341
936	return 0;	1342	return 0;
937	}	1343	}
938		1344
939	/*****************************************************************************/	1345	/*****************************************************************************/
940		1346
941	static void eio_atfork_prepare (void)	1347	int eio_init (void (*want_poll)(void), void (*done_poll)(void))
942	{	1348	{
943	X_LOCK (wrklock);	1349	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	}	1350	}
953		1351
954	static void eio_atfork_parent (void)	1352	static void eio_api_destroy (eio_req *req)
955	{	1353	{
956	#if !HAVE_READDIR_R	1354	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	}	1355	}
966		1356
967	static void eio_atfork_child (void)	1357	#define REQ(rtype) \
968	{	1358	eio_req *req; \
969	eio_req *prv;	1359	\
		1360	req = (eio_req *)calloc (1, sizeof *req); \
		1361	if (!req) \
		1362	return 0; \
		1363	\
		1364	req->type = rtype; \
		1365	req->pri = pri; \
		1366	req->finish = cb; \
		1367	req->data = data; \
		1368	req->destroy = eio_api_destroy;
970		1369
971	while (prv = reqq_shift (&req_queue))	1370	#define SEND eio_submit (req); return req
972	eio_destroy (prv);
973		1371
974	while (prv = reqq_shift (&res_queue))	1372	#define PATH \
975	eio_destroy (prv);	1373	req->flags |= EIO_FLAG_PTR1_FREE; \
		1374	req->ptr1 = strdup (path); \
		1375	if (!req->ptr1) \
		1376	{ \
		1377	eio_api_destroy (req); \
		1378	return 0; \
		1379	}
976		1380
977	while (wrk_first.next != &wrk_first)	1381	static void eio_execute (etp_worker *self, eio_req *req)
		1382	{
		1383	errno = 0;
		1384
		1385	switch (req->type)
978	{	1386	{
979	worker *wrk = wrk_first.next;	1387	case EIO_READ: ALLOC (req->size);
		1388	req->result = req->offs >= 0
		1389	? pread (req->int1, req->ptr2, req->size, req->offs)
		1390	: read (req->int1, req->ptr2, req->size); break;
		1391	case EIO_WRITE: req->result = req->offs >= 0
		1392	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1393	: write (req->int1, req->ptr2, req->size); break;
980		1394
981	if (wrk->req)	1395	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
982	eio_destroy (wrk->req);	1396	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
983		1397
984	worker_clear (wrk);	1398	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
985	worker_free (wrk);	1399	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
986	}	1400	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1401	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1402	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1403	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
987		1404
988	started = 0;	1405	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
989	idle = 0;	1406	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
990	nreqs = 0;	1407	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
991	nready = 0;	1408	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
992	npending = 0;	1409	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1410	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
993		1411
994	eio_atfork_parent ();	1412	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
995	}	1413	case EIO_CLOSE: req->result = close (req->int1); break;
		1414	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1415	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1416	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1417	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1418	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1419	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1420	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1421	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
996		1422
997	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1423	case EIO_READLINK: ALLOC (NAME_MAX);
998	{	1424	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
999	want_poll_cb = want_poll;
1000	done_poll_cb = done_poll;
1001		1425
		1426	case EIO_SYNC: req->result = 0; sync (); break;
		1427	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1428	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1429	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
		1430	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1431	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1432
		1433	case EIO_READDIR: eio__scandir (req, self); break;
		1434
		1435	case EIO_BUSY:
1002	#ifdef _WIN32	1436	#ifdef _WIN32
1003	X_MUTEX_CHECK (wrklock);	1437	Sleep (req->nv1 * 1000.);
1004	X_MUTEX_CHECK (reslock);	1438	#else
1005	X_MUTEX_CHECK (reqlock);
1006	X_MUTEX_CHECK (reqwait);
1007	X_MUTEX_CHECK (preadwritelock);
1008	X_MUTEX_CHECK (readdirlock);
1009
1010	X_COND_CHECK (reqwait);
1011	#endif
1012
1013	X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1014	}
1015
1016	#if 0
1017
1018	eio_req *eio_fsync (int fd, eio_cb cb);
1019	eio_req *eio_fdatasync (int fd, eio_cb cb);
1020	eio_req *eio_dupclose (int fd, eio_cb cb);
1021	eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb);
1022	eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1023	eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1024	eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1025	eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb);
1026	eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb);
1027	eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb);
1028	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb);
1029	eio_req *eio_dup2 (int fd, int fd2, eio_cb cb);
1030	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb);
1031	eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb);
1032	eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1033	eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1034	eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1035	eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb);
1036	eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb);
1037	eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb);
1038	eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb);
1039	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb);
1040	eio_req *eio_unlink (const char *path, eio_cb cb);
1041	eio_req *eio_rmdir (const char *path, eio_cb cb);
1042	eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1043	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb);
1044	eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */
1045	eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */
1046	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	{	1439	{
1130	dREQ;	1440	struct timeval tv;
1131		1441
1132	req->type = ix;	1442	tv.tv_sec = req->nv1;
1133	req->sv1 = newSVsv (fh);	1443	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
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		1444
1142	if (!SvREADONLY (data))	1445	req->result = select (0, 0, 0, 0, &tv);
		1446	}
		1447	#endif
		1448	break;
		1449
		1450	case EIO_UTIME:
		1451	case EIO_FUTIME:
		1452	{
		1453	struct timeval tv[2];
		1454	struct timeval *times;
		1455
		1456	if (req->nv1 != -1. || req->nv2 != -1.)
1143	{	1457	{
1144	SvREADONLY_on (data);	1458	tv[0].tv_sec = req->nv1;
1145	req->flags |= FLAG_SV2_RO_OFF;	1459	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		1460	tv[1].tv_sec = req->nv2;
		1461	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		1462
		1463	times = tv;
1146	}	1464	}
		1465	else
		1466	times = 0;
1147		1467
1148	EIO_SEND;	1468
		1469	req->result = req->type == EIO_FUTIME
		1470	? futimes (req->int1, times)
		1471	: utimes (req->ptr1, times);
1149	}	1472	}
1150	}	1473	break;
1151		1474
1152	void	1475	case EIO_GROUP:
1153	aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)	1476	abort (); /* handled in eio_request */
1154	PROTOTYPE: $$;$
1155	PPCODE:
1156	{
1157	SV *data;
1158	dREQ;
1159		1477
1160	data = newSV (NAME_MAX);	1478	case EIO_NOP:
1161	SvPOK_on (data);	1479	req->result = 0;
		1480	break;
1162		1481
1163	req->type = EIO_READLINK;	1482	case EIO_CUSTOM:
1164	req->sv1 = newSVsv (path);	1483	((void (*)(eio_req *))req->feed) (req);
1165	req->ptr1 = SvPVbyte_nolen (req->sv1);	1484	break;
1166	req->sv2 = data;
1167	req->ptr2 = SvPVbyte_nolen (data);
1168		1485
1169	EIO_SEND;	1486	default:
1170	}	1487	req->result = -1;
		1488	break;
		1489	}
1171		1490
1172	void	1491	req->errorno = errno;
1173	aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174	PROTOTYPE: $$$$;$
1175	PPCODE:
1176	{
1177	dREQ;
1178
1179	req->type = EIO_SENDFILE;
1180	req->sv1 = newSVsv (out_fh);
1181	req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182	req->sv2 = newSVsv (in_fh);
1183	req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184	req->offs = SvVAL64 (in_offset);
1185	req->size = length;
1186
1187	EIO_SEND;
1188	}	1492	}
1189		1493
1190	void	1494	#ifndef EIO_NO_WRAPPERS
1191	aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192	PROTOTYPE: $$$;$
1193	PPCODE:
1194	{
1195	dREQ;
1196		1495
1197	req->type = EIO_READAHEAD;	1496	eio_req *eio_nop (int pri, eio_cb cb, void *data)
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	{	1497	{
1213	dREQ;	1498	REQ (EIO_NOP); SEND;
		1499	}
1214		1500
1215	req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));	1501	eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1216	if (!req->ptr2)	1502	{
1217	{	1503	REQ (EIO_BUSY); req->nv1 = delay; SEND;
1218	req_destroy (req);	1504	}
1219	croak ("out of memory during aio_stat statdata allocation");
1220	}
1221		1505
		1506	eio_req *eio_sync (int pri, eio_cb cb, void *data)
		1507	{
		1508	REQ (EIO_SYNC); SEND;
		1509	}
		1510
		1511	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
		1512	{
		1513	REQ (EIO_FSYNC); req->int1 = fd; SEND;
		1514	}
		1515
		1516	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1517	{
		1518	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1519	}
		1520
		1521	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1522	{
		1523	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1524	}
		1525
		1526	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		1527	{
		1528	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		1529	}
		1530
		1531	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		1532	{
		1533	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		1534	}
		1535
		1536	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
		1537	{
		1538	REQ (EIO_CLOSE); req->int1 = fd; SEND;
		1539	}
		1540
		1541	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
		1542	{
		1543	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
		1544	}
		1545
		1546	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
		1547	{
		1548	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
		1549	}
		1550
		1551	eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
		1552	{
		1553	REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
		1554	}
		1555
		1556	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
		1557	{
		1558	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1559	}
		1560
		1561	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
		1562	{
		1563	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
		1564	}
		1565
		1566	eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
		1567	{
		1568	REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
		1569	}
		1570
		1571	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
		1572	{
		1573	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
		1574	}
		1575
		1576	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
		1577	{
		1578	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
		1579	}
		1580
		1581	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
		1582	{
		1583	REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
		1584	}
		1585
		1586	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
		1587	{
		1588	REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
		1589	}
		1590
		1591	eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
		1592	{
		1593	REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
		1594	}
		1595
		1596	eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
		1597	{
		1598	REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
		1599	}
		1600
		1601	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
		1602	{
		1603	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
		1604	}
		1605
		1606	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
		1607	{
		1608	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
		1609	}
		1610
		1611	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
		1612	{
		1613	REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
		1614	}
		1615
		1616	eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
		1617	{
		1618	REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
		1619	}
		1620
		1621	static eio_req *
		1622	eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
		1623	{
		1624	REQ (type); PATH; SEND;
		1625	}
		1626
		1627	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
		1628	{
		1629	return eio__1path (EIO_READLINK, path, pri, cb, data);
		1630	}
		1631
		1632	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
		1633	{
		1634	return eio__1path (EIO_STAT, path, pri, cb, data);
		1635	}
		1636
		1637	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
		1638	{
		1639	return eio__1path (EIO_LSTAT, path, pri, cb, data);
		1640	}
		1641
		1642	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
		1643	{
		1644	return eio__1path (EIO_UNLINK, path, pri, cb, data);
		1645	}
		1646
		1647	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
		1648	{
		1649	return eio__1path (EIO_RMDIR, path, pri, cb, data);
		1650	}
		1651
		1652	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
		1653	{
		1654	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
		1655	}
		1656
		1657	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
		1658	{
		1659	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
		1660	}
		1661
		1662	static eio_req *
		1663	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		1664	{
		1665	REQ (type); PATH;
		1666
1222	req->flags |= FLAG_PTR2_FREE;	1667	req->flags |= EIO_FLAG_PTR2_FREE;
1223	req->sv1 = newSVsv (fh_or_path);	1668	req->ptr2 = strdup (new_path);
		1669	if (!req->ptr2)
		1670	{
		1671	eio_api_destroy (req);
		1672	return 0;
		1673	}
1224		1674
1225	if (SvPOK (fh_or_path))	1675	SEND;
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	}	1676	}
1238		1677
1239	void	1678	eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1240	aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1241	PPCODE:
1242	{	1679	{
1243	dREQ;	1680	return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
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	}	1681	}
1262		1682
1263	void	1683	eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1264	aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1265	PPCODE:
1266	{	1684	{
1267	dREQ;	1685	return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
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	}	1686	}
1285		1687
1286	void	1688	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
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	{	1689	{
1294	dREQ;	1690	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
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	}	1691	}
1307		1692
1308	void	1693	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1309	aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1310	PPCODE:
1311	{	1694	{
1312	dREQ;	1695	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
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	}	1696	}
1331		1697
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	1698	#endif
		1699
		1700	eio_req *eio_grp (eio_cb cb, void *data)
		1701	{
		1702	const int pri = EIO_PRI_MAX;
		1703
		1704	REQ (EIO_GROUP); SEND;
		1705	}
		1706
		1707	#undef REQ
		1708	#undef PATH
		1709	#undef SEND
		1710
		1711	/*****************************************************************************/
		1712	/* grp functions */
1418		1713
1419	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	1714	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420	{	1715	{
1421	grp->int2 = limit;	1716	grp->int2 = limit;
1422	grp->feed = feed;	1717	grp->feed = feed;
…		…
1433		1728
1434	void eio_grp_add (eio_req *grp, eio_req *req)	1729	void eio_grp_add (eio_req *grp, eio_req *req)
1435	{	1730	{
1436	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	1731	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1437		1732
		1733	grp->flags |= EIO_FLAG_GROUPADD;
		1734
1438	++grp->size;	1735	++grp->size;
1439	req->grp = grp;	1736	req->grp = grp;
1440		1737
1441	req->grp_prev = 0;	1738	req->grp_prev = 0;
1442	req->grp_next = grp->grp_first;	1739	req->grp_next = grp->grp_first;
…		…
1445	grp->grp_first->grp_prev = req;	1742	grp->grp_first->grp_prev = req;
1446		1743
1447	grp->grp_first = req;	1744	grp->grp_first = req;
1448	}	1745	}
1449		1746
		1747	/*****************************************************************************/
		1748	/* misc garbage */
1450		1749
		1750	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
		1751	{
		1752	etp_worker wrk;
		1753
		1754	wrk.dbuf = 0;
		1755
		1756	eio__sendfile (ofd, ifd, offset, count, &wrk);
		1757
		1758	if (wrk.dbuf)
		1759	free (wrk.dbuf);
		1760	}
		1761

Diff Legend

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