ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.11 by root, Tue May 13 18:50:27 2008 UTC vs.
Revision 1.54 by root, Tue Feb 23 14:32:45 2010 UTC

		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008,2009,2010 Marc Alexander Lehmann <libeio@schmorp.de>
		5	* All rights reserved.
		6	*
		7	* Redistribution and use in source and binary forms, with or without modifica-
		8	* tion, are permitted provided that the following conditions are met:
		9	*
		10	* 1. Redistributions of source code must retain the above copyright notice,
		11	* this list of conditions and the following disclaimer.
		12	*
		13	* 2. Redistributions in binary form must reproduce the above copyright
		14	* notice, this list of conditions and the following disclaimer in the
		15	* documentation and/or other materials provided with the distribution.
		16	*
		17	* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
		18	* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
		19	* CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
		20	* EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
		21	* CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
		22	* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
		23	* OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
		24	* WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
		25	* ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
		26	* OF THE POSSIBILITY OF SUCH DAMAGE.
		27	*
		28	* Alternatively, the contents of this file may be used under the terms of
		29	* the GNU General Public License ("GPL") version 2 or any later version,
		30	* in which case the provisions of the GPL are applicable instead of
		31	* the above. If you wish to allow the use of your version of this file
		32	* only under the terms of the GPL and not to allow others to use your
		33	* version of this file under the BSD license, indicate your decision
		34	* by deleting the provisions above and replace them with the notice
		35	* and other provisions required by the GPL. If you do not delete the
		36	* provisions above, a recipient may use your version of this file under
		37	* either the BSD or the GPL.
		38	*/
		39
1	#include "eio.h"	40	#include "eio.h"
		41
		42	#ifdef EIO_STACKSIZE
		43	# define XTHREAD_STACKSIZE EIO_STACKSIZE
		44	#endif
2	#include "xthread.h"	45	#include "xthread.h"
3		46
4	#include <errno.h>	47	#include <errno.h>
5	#include <stddef.h>	48	#include <stddef.h>
6	#include <stdlib.h>	49	#include <stdlib.h>
7	#include <string.h>	50	#include <string.h>
8	#include <errno.h>	51	#include <errno.h>
9	#include <sys/types.h>	52	#include <sys/types.h>
10	#include <sys/stat.h>	53	#include <sys/stat.h>
		54	#include <sys/statvfs.h>
11	#include <limits.h>	55	#include <limits.h>
12	#include <fcntl.h>	56	#include <fcntl.h>
13	#include <assert.h>	57	#include <assert.h>
14		58
15	#ifndef EIO_FINISH	59	#ifndef EIO_FINISH
…		…
25	#endif	69	#endif
26		70
27	#ifdef _WIN32	71	#ifdef _WIN32
28		72
29	/*doh*/	73	/*doh*/
30
31	#else	74	#else
32		75
33	# include "config.h"	76	# include "config.h"
34	# include <sys/time.h>	77	# include <sys/time.h>
35	# include <sys/select.h>	78	# include <sys/select.h>
		79	# include <sys/mman.h>
36	# include <unistd.h>	80	# include <unistd.h>
37	# include <utime.h>	81	# include <utime.h>
38	# include <signal.h>	82	# include <signal.h>
39	# include <dirent.h>	83	# include <dirent.h>
40		84
		85	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		86	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		87	# define _DIRENT_HAVE_D_TYPE /* sigh */
		88	# define D_INO(de) (de)->d_fileno
		89	# define D_NAMLEN(de) (de)->d_namlen
		90	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		91	# define D_INO(de) (de)->d_ino
		92	# endif
		93
		94	#ifdef _D_EXACT_NAMLEN
		95	# undef D_NAMLEN
		96	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		97	#endif
		98
		99	# ifdef _DIRENT_HAVE_D_TYPE
		100	# define D_TYPE(de) (de)->d_type
		101	# endif
		102
41	# ifndef EIO_STRUCT_DIRENT	103	# ifndef EIO_STRUCT_DIRENT
42	# define EIO_STRUCT_DIRENT struct dirent	104	# define EIO_STRUCT_DIRENT struct dirent
43	# endif	105	# endif
44		106
45	#endif	107	#endif
46		108
47	#if HAVE_SENDFILE	109	#if HAVE_SENDFILE
48	# if __linux	110	# if __linux
49	# include <sys/sendfile.h>	111	# include <sys/sendfile.h>
50	# elif __freebsd	112	# elif __FreeBSD__ || defined __APPLE__
51	# include <sys/socket.h>	113	# include <sys/socket.h>
52	# include <sys/uio.h>	114	# include <sys/uio.h>
53	# elif __hpux	115	# elif __hpux
54	# include <sys/socket.h>	116	# include <sys/socket.h>
55	# elif __solaris /* not yet */	117	# elif __solaris
56	# include <sys/sendfile.h>	118	# include <sys/sendfile.h>
57	# else	119	# else
58	# error sendfile support requested but not available	120	# error sendfile support requested but not available
59	# endif	121	# endif
60	#endif	122	#endif
61		123
		124	#ifndef D_TYPE
		125	# define D_TYPE(de) 0
		126	#endif
		127	#ifndef D_INO
		128	# define D_INO(de) 0
		129	#endif
		130	#ifndef D_NAMLEN
		131	# define D_NAMLEN(de) strlen ((de)->d_name)
		132	#endif
		133
62	/* number of seconds after which an idle threads exit */	134	/* number of seconds after which an idle threads exit */
63	#define IDLE_TIMEOUT 10	135	#define IDLE_TIMEOUT 10
64		136
65	/* used for struct dirent, AIX doesn't provide it */	137	/* used for struct dirent, AIX doesn't provide it */
66	#ifndef NAME_MAX	138	#ifndef NAME_MAX
67	# define NAME_MAX 4096	139	# define NAME_MAX 4096
68	#endif	140	#endif
69		141
		142	/* used for readlink etc. */
		143	#ifndef PATH_MAX
		144	# define PATH_MAX 4096
		145	#endif
		146
70	/* buffer size for various temporary buffers */	147	/* buffer size for various temporary buffers */
71	#define EIO_BUFSIZE 65536	148	#define EIO_BUFSIZE 65536
72		149
73	#define dBUF \	150	#define dBUF \
74	char *eio_buf; \	151	char *eio_buf; \
75	X_LOCK (wrklock); \	152	ETP_WORKER_LOCK (self); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	153	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (wrklock); \	154	ETP_WORKER_UNLOCK (self); \
78	errno = ENOMEM; \	155	errno = ENOMEM; \
79	if (!eio_buf) \	156	if (!eio_buf) \
80	return -1;	157	return -1;
81		158
82	#define EIO_TICKS ((1000000 + 1023) >> 10)	159	#define EIO_TICKS ((1000000 + 1023) >> 10)
83		160
		161	/*****************************************************************************/
		162
		163	#if __GNUC__ >= 3
		164	# define expect(expr,value) __builtin_expect ((expr),(value))
		165	#else
		166	# define expect(expr,value) (expr)
		167	#endif
		168
		169	#define expect_false(expr) expect ((expr) != 0, 0)
		170	#define expect_true(expr) expect ((expr) != 0, 1)
		171
		172	/*****************************************************************************/
		173
		174	#define ETP_PRI_MIN EIO_PRI_MIN
		175	#define ETP_PRI_MAX EIO_PRI_MAX
		176
		177	struct etp_worker;
		178
		179	#define ETP_REQ eio_req
		180	#define ETP_DESTROY(req) eio_destroy (req)
		181	static int eio_finish (eio_req *req);
		182	#define ETP_FINISH(req) eio_finish (req)
		183	static void eio_execute (struct etp_worker *self, eio_req *req);
		184	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		185
		186	#define ETP_WORKER_CLEAR(req) \
		187	if (wrk->dbuf) \
		188	{ \
		189	free (wrk->dbuf); \
		190	wrk->dbuf = 0; \
		191	} \
		192	\
		193	if (wrk->dirp) \
		194	{ \
		195	closedir (wrk->dirp); \
		196	wrk->dirp = 0; \
		197	}
		198
		199	#define ETP_WORKER_COMMON \
		200	void *dbuf; \
		201	DIR *dirp;
		202
		203	/*****************************************************************************/
		204
		205	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		206
		207	/* calculate time difference in ~1/EIO_TICKS of a second */
		208	static int tvdiff (struct timeval *tv1, struct timeval *tv2)
		209	{
		210	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
		211	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
		212	}
		213
		214	static unsigned int started, idle, wanted = 4;
		215
84	static void (*want_poll_cb) (void);	216	static void (*want_poll_cb) (void);
85	static void (*done_poll_cb) (void);	217	static void (*done_poll_cb) (void);
86		218
87	static unsigned int max_poll_time = 0;	219	static unsigned int max_poll_time; /* reslock */
88	static unsigned int max_poll_reqs = 0;	220	static unsigned int max_poll_reqs; /* reslock */
89		221
90	/* calculcate time difference in ~1/EIO_TICKS of a second */	222	static volatile unsigned int nreqs; /* reqlock */
91	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	223	static volatile unsigned int nready; /* reqlock */
92	{	224	static volatile unsigned int npending; /* reqlock */
93	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	225	static volatile unsigned int max_idle = 4;
94	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
95	}
96		226
97	static unsigned int started, idle, wanted = 4;
98
99	/* worker threads management */
100	static mutex_t wrklock = X_MUTEX_INIT;	227	static mutex_t wrklock = X_MUTEX_INIT;
101
102	typedef struct worker
103	{
104	/* locked by wrklock */
105	struct worker *prev, *next;
106
107	thread_t tid;
108
109	/* locked by reslock, reqlock or wrklock */
110	eio_req *req; /* currently processed request */
111	void *dbuf;
112	DIR *dirp;
113	} worker;
114
115	static worker wrk_first = { &wrk_first, &wrk_first, 0 };
116
117	static void worker_clear (worker *wrk)
118	{
119	if (wrk->dirp)
120	{
121	closedir (wrk->dirp);
122	wrk->dirp = 0;
123	}
124
125	if (wrk->dbuf)
126	{
127	free (wrk->dbuf);
128	wrk->dbuf = 0;
129	}
130	}
131
132	static void worker_free (worker *wrk)
133	{
134	wrk->next->prev = wrk->prev;
135	wrk->prev->next = wrk->next;
136
137	free (wrk);
138	}
139
140	static volatile unsigned int nreqs, nready, npending;
141	static volatile unsigned int max_idle = 4;
142
143	static mutex_t reslock = X_MUTEX_INIT;	228	static mutex_t reslock = X_MUTEX_INIT;
144	static mutex_t reqlock = X_MUTEX_INIT;	229	static mutex_t reqlock = X_MUTEX_INIT;
145	static cond_t reqwait = X_COND_INIT;	230	static cond_t reqwait = X_COND_INIT;
146		231
		232	#if !HAVE_PREADWRITE
		233	/*
		234	* make our pread/pwrite emulation safe against themselves, but not against
		235	* normal read/write by using a mutex. slows down execution a lot,
		236	* but that's your problem, not mine.
		237	*/
		238	static mutex_t preadwritelock = X_MUTEX_INIT;
		239	#endif
		240
		241	typedef struct etp_worker
		242	{
		243	/* locked by wrklock */
		244	struct etp_worker *prev, *next;
		245
		246	thread_t tid;
		247
		248	/* locked by reslock, reqlock or wrklock */
		249	ETP_REQ *req; /* currently processed request */
		250
		251	ETP_WORKER_COMMON
		252	} etp_worker;
		253
		254	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
		255
		256	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		257	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
		258
		259	/* worker threads management */
		260
		261	static void etp_worker_clear (etp_worker *wrk)
		262	{
		263	ETP_WORKER_CLEAR (wrk);
		264	}
		265
		266	static void etp_worker_free (etp_worker *wrk)
		267	{
		268	wrk->next->prev = wrk->prev;
		269	wrk->prev->next = wrk->next;
		270
		271	free (wrk);
		272	}
		273
147	unsigned int eio_nreqs (void)	274	static unsigned int etp_nreqs (void)
148	{	275	{
		276	int retval;
		277	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		278	retval = nreqs;
		279	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
149	return nreqs;	280	return retval;
150	}	281	}
151		282
152	unsigned int eio_nready (void)	283	static unsigned int etp_nready (void)
153	{	284	{
154	unsigned int retval;	285	unsigned int retval;
155		286
156	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	287	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157	retval = nready;	288	retval = nready;
158	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	289	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159		290
160	return retval;	291	return retval;
161	}	292	}
162		293
163	unsigned int eio_npending (void)	294	static unsigned int etp_npending (void)
164	{	295	{
165	unsigned int retval;	296	unsigned int retval;
166		297
167	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	298	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168	retval = npending;	299	retval = npending;
169	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	300	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170		301
171	return retval;	302	return retval;
172	}	303	}
173		304
174	unsigned int eio_nthreads (void)	305	static unsigned int etp_nthreads (void)
175	{	306	{
176	unsigned int retval;	307	unsigned int retval;
177		308
178	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	309	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179	retval = started;	310	retval = started;
…		…
186	* a somewhat faster data structure might be nice, but	317	* a somewhat faster data structure might be nice, but
187	* with 8 priorities this actually needs <20 insns	318	* with 8 priorities this actually needs <20 insns
188	* per shift, the most expensive operation.	319	* per shift, the most expensive operation.
189	*/	320	*/
190	typedef struct {	321	typedef struct {
191	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	322	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
192	int size;	323	int size;
193	} reqq;	324	} etp_reqq;
194		325
195	static reqq req_queue;	326	static etp_reqq req_queue;
196	static reqq res_queue;	327	static etp_reqq res_queue;
197		328
198	static int reqq_push (reqq *q, eio_req *req)	329	static int reqq_push (etp_reqq *q, ETP_REQ *req)
199	{	330	{
200	int pri = req->pri;	331	int pri = req->pri;
201	req->next = 0;	332	req->next = 0;
202		333
203	if (q->qe[pri])	334	if (q->qe[pri])
…		…
209	q->qe[pri] = q->qs[pri] = req;	340	q->qe[pri] = q->qs[pri] = req;
210		341
211	return q->size++;	342	return q->size++;
212	}	343	}
213		344
214	static eio_req *reqq_shift (reqq *q)	345	static ETP_REQ *reqq_shift (etp_reqq *q)
215	{	346	{
216	int pri;	347	int pri;
217		348
218	if (!q->size)	349	if (!q->size)
219	return 0;	350	return 0;
220		351
221	--q->size;	352	--q->size;
222		353
223	for (pri = EIO_NUM_PRI; pri--; )	354	for (pri = ETP_NUM_PRI; pri--; )
224	{	355	{
225	eio_req *req = q->qs[pri];	356	eio_req *req = q->qs[pri];
226		357
227	if (req)	358	if (req)
228	{	359	{
…		…
234	}	365	}
235		366
236	abort ();	367	abort ();
237	}	368	}
238		369
239	static void grp_try_feed (eio_req *grp)	370	static void etp_atfork_prepare (void)
240	{	371	{
241	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	372	X_LOCK (wrklock);
		373	X_LOCK (reqlock);
		374	X_LOCK (reslock);
		375	#if !HAVE_PREADWRITE
		376	X_LOCK (preadwritelock);
		377	#endif
		378	}
		379
		380	static void etp_atfork_parent (void)
		381	{
		382	#if !HAVE_PREADWRITE
		383	X_UNLOCK (preadwritelock);
		384	#endif
		385	X_UNLOCK (reslock);
		386	X_UNLOCK (reqlock);
		387	X_UNLOCK (wrklock);
		388	}
		389
		390	static void etp_atfork_child (void)
		391	{
		392	ETP_REQ *prv;
		393
		394	while ((prv = reqq_shift (&req_queue)))
		395	ETP_DESTROY (prv);
		396
		397	while ((prv = reqq_shift (&res_queue)))
		398	ETP_DESTROY (prv);
		399
		400	while (wrk_first.next != &wrk_first)
242	{	401	{
243	int old_len = grp->size;	402	etp_worker *wrk = wrk_first.next;
244		403
245	EIO_FEED (grp);	404	if (wrk->req)
		405	ETP_DESTROY (wrk->req);
246		406
247	/* stop if no progress has been made */	407	etp_worker_clear (wrk);
248	if (old_len == grp->size)	408	etp_worker_free (wrk);
249	{
250	grp->feed = 0;
251	break;
252	}
253	}	409	}
254	}
255		410
256	static int eio_finish (eio_req *req);	411	started = 0;
		412	idle = 0;
		413	nreqs = 0;
		414	nready = 0;
		415	npending = 0;
257		416
258	static int grp_dec (eio_req *grp)	417	etp_atfork_parent ();
259	{	418	}
260	--grp->size;
261		419
262	/* call feeder, if applicable */	420	static void
263	grp_try_feed (grp);	421	etp_once_init (void)
		422	{
		423	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
		424	}
264		425
265	/* finish, if done */	426	static int
266	if (!grp->size && grp->int1)	427	etp_init (void (*want_poll)(void), void (*done_poll)(void))
267	return eio_finish (grp);	428	{
268	else	429	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
		430
		431	pthread_once (&doinit, etp_once_init);
		432
		433	want_poll_cb = want_poll;
		434	done_poll_cb = done_poll;
		435
269	return 0;	436	return 0;
270	}	437	}
271		438
272	void eio_destroy (eio_req *req)
273	{
274	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
275	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
276
277	EIO_DESTROY (req);
278	}
279
280	static int eio_finish (eio_req *req)
281	{
282	int res = EIO_FINISH (req);
283
284	if (req->grp)
285	{
286	int res2;
287	eio_req *grp = req->grp;
288
289	/* unlink request */
290	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
291	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
292
293	if (grp->grp_first == req)
294	grp->grp_first = req->grp_next;
295
296	res2 = grp_dec (grp);
297
298	if (!res && res2)
299	res = res2;
300	}
301
302	eio_destroy (req);
303
304	return res;
305	}
306
307	void eio_grp_cancel (eio_req *grp)
308	{
309	for (grp = grp->grp_first; grp; grp = grp->grp_next)
310	eio_cancel (grp);
311	}
312
313	void eio_cancel (eio_req *req)
314	{
315	X_LOCK (wrklock);
316	req->flags |= EIO_FLAG_CANCELLED;
317	X_UNLOCK (wrklock);
318
319	eio_grp_cancel (req);
320	}
321
322	X_THREAD_PROC (eio_proc);	439	X_THREAD_PROC (etp_proc);
323		440
324	static void start_thread (void)	441	static void etp_start_thread (void)
325	{	442	{
326	worker *wrk = calloc (1, sizeof (worker));	443	etp_worker *wrk = calloc (1, sizeof (etp_worker));
327		444
328	/*TODO*/	445	/*TODO*/
329	assert (("unable to allocate worker thread data", wrk));	446	assert (("unable to allocate worker thread data", wrk));
330		447
331	X_LOCK (wrklock);	448	X_LOCK (wrklock);
332		449
333	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	450	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334	{	451	{
335	wrk->prev = &wrk_first;	452	wrk->prev = &wrk_first;
336	wrk->next = wrk_first.next;	453	wrk->next = wrk_first.next;
337	wrk_first.next->prev = wrk;	454	wrk_first.next->prev = wrk;
338	wrk_first.next = wrk;	455	wrk_first.next = wrk;
…		…
342	free (wrk);	459	free (wrk);
343		460
344	X_UNLOCK (wrklock);	461	X_UNLOCK (wrklock);
345	}	462	}
346		463
347	static void maybe_start_thread (void)	464	static void etp_maybe_start_thread (void)
348	{	465	{
349	if (eio_nthreads () >= wanted)	466	if (expect_true (etp_nthreads () >= wanted))
350	return;	467	return;
351		468
352	/* todo: maybe use idle here, but might be less exact */	469	/* todo: maybe use idle here, but might be less exact */
353	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	470	if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354	return;	471	return;
355		472
356	start_thread ();	473	etp_start_thread ();
357	}	474	}
358		475
359	void eio_submit (eio_req *req)
360	{
361	req->pri += EIO_PRI_BIAS;
362
363	if (req->pri < EIO_PRI_MIN + EIO_PRI_BIAS) req->pri = EIO_PRI_MIN + EIO_PRI_BIAS;
364	if (req->pri > EIO_PRI_MAX + EIO_PRI_BIAS) req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
365
366	++nreqs;
367
368	X_LOCK (reqlock);
369	++nready;
370	reqq_push (&req_queue, req);
371	X_COND_SIGNAL (reqwait);
372	X_UNLOCK (reqlock);
373
374	maybe_start_thread ();
375	}
376
377	static void end_thread (void)	476	static void etp_end_thread (void)
378	{	477	{
379	eio_req *req = calloc (1, sizeof (eio_req));	478	eio_req *req = calloc (1, sizeof (eio_req));
380		479
381	req->type = EIO_QUIT;	480	req->type = -1;
382	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	481	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
383		482
384	X_LOCK (reqlock);	483	X_LOCK (reqlock);
385	reqq_push (&req_queue, req);	484	reqq_push (&req_queue, req);
386	X_COND_SIGNAL (reqwait);	485	X_COND_SIGNAL (reqwait);
387	X_UNLOCK (reqlock);	486	X_UNLOCK (reqlock);
…		…
389	X_LOCK (wrklock);	488	X_LOCK (wrklock);
390	--started;	489	--started;
391	X_UNLOCK (wrklock);	490	X_UNLOCK (wrklock);
392	}	491	}
393		492
394	void eio_set_max_poll_time (double nseconds)
395	{
396	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
397	max_poll_time = nseconds;
398	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
399	}
400
401	void eio_set_max_poll_reqs (unsigned int maxreqs)
402	{
403	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
404	max_poll_reqs = maxreqs;
405	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
406	}
407
408	void eio_set_max_idle (unsigned int nthreads)
409	{
410	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
411	max_idle = nthreads <= 0 ? 1 : nthreads;
412	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
413	}
414
415	void eio_set_min_parallel (unsigned int nthreads)
416	{
417	if (wanted < nthreads)
418	wanted = nthreads;
419	}
420
421	void eio_set_max_parallel (unsigned int nthreads)
422	{
423	if (wanted > nthreads)
424	wanted = nthreads;
425
426	while (started > wanted)
427	end_thread ();
428	}
429
430	int eio_poll (void)	493	static int etp_poll (void)
431	{	494	{
432	int maxreqs = max_poll_reqs;	495	unsigned int maxreqs;
		496	unsigned int maxtime;
433	struct timeval tv_start, tv_now;	497	struct timeval tv_start, tv_now;
434	eio_req *req;
435		498
		499	X_LOCK (reslock);
		500	maxreqs = max_poll_reqs;
		501	maxtime = max_poll_time;
		502	X_UNLOCK (reslock);
		503
436	if (max_poll_time)	504	if (maxtime)
437	gettimeofday (&tv_start, 0);	505	gettimeofday (&tv_start, 0);
438		506
439	for (;;)	507	for (;;)
440	{	508	{
		509	ETP_REQ *req;
		510
441	maybe_start_thread ();	511	etp_maybe_start_thread ();
442		512
443	X_LOCK (reslock);	513	X_LOCK (reslock);
444	req = reqq_shift (&res_queue);	514	req = reqq_shift (&res_queue);
445		515
446	if (req)	516	if (req)
…		…
454	X_UNLOCK (reslock);	524	X_UNLOCK (reslock);
455		525
456	if (!req)	526	if (!req)
457	return 0;	527	return 0;
458		528
		529	X_LOCK (reqlock);
459	--nreqs;	530	--nreqs;
		531	X_UNLOCK (reqlock);
460		532
461	if (req->type == EIO_GROUP && req->size)	533	if (expect_false (req->type == EIO_GROUP && req->size))
462	{	534	{
463	req->int1 = 1; /* mark request as delayed */	535	req->int1 = 1; /* mark request as delayed */
464	continue;	536	continue;
465	}	537	}
466	else	538	else
467	{	539	{
468	int res = eio_finish (req);	540	int res = ETP_FINISH (req);
469	if (res)	541	if (expect_false (res))
470	return res;	542	return res;
471	}	543	}
472		544
473	if (maxreqs && !--maxreqs)	545	if (expect_false (maxreqs && !--maxreqs))
474	break;	546	break;
475		547
476	if (max_poll_time)	548	if (maxtime)
477	{	549	{
478	gettimeofday (&tv_now, 0);	550	gettimeofday (&tv_now, 0);
479		551
480	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	552	if (tvdiff (&tv_start, &tv_now) >= maxtime)
481	break;	553	break;
482	}	554	}
483	}	555	}
484		556
485	errno = EAGAIN;	557	errno = EAGAIN;
486	return -1;	558	return -1;
487	}	559	}
488		560
		561	static void etp_cancel (ETP_REQ *req)
		562	{
		563	X_LOCK (wrklock);
		564	req->flags |= EIO_FLAG_CANCELLED;
		565	X_UNLOCK (wrklock);
		566
		567	eio_grp_cancel (req);
		568	}
		569
		570	static void etp_submit (ETP_REQ *req)
		571	{
		572	req->pri -= ETP_PRI_MIN;
		573
		574	if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		575	if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		576
		577	if (expect_false (req->type == EIO_GROUP))
		578	{
		579	/* I hope this is worth it :/ */
		580	X_LOCK (reqlock);
		581	++nreqs;
		582	X_UNLOCK (reqlock);
		583
		584	X_LOCK (reslock);
		585
		586	++npending;
		587
		588	if (!reqq_push (&res_queue, req) && want_poll_cb)
		589	want_poll_cb ();
		590
		591	X_UNLOCK (reslock);
		592	}
		593	else
		594	{
		595	X_LOCK (reqlock);
		596	++nreqs;
		597	++nready;
		598	reqq_push (&req_queue, req);
		599	X_COND_SIGNAL (reqwait);
		600	X_UNLOCK (reqlock);
		601
		602	etp_maybe_start_thread ();
		603	}
		604	}
		605
		606	static void etp_set_max_poll_time (double nseconds)
		607	{
		608	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		609	max_poll_time = nseconds * EIO_TICKS;
		610	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		611	}
		612
		613	static void etp_set_max_poll_reqs (unsigned int maxreqs)
		614	{
		615	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		616	max_poll_reqs = maxreqs;
		617	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		618	}
		619
		620	static void etp_set_max_idle (unsigned int nthreads)
		621	{
		622	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		623	max_idle = nthreads <= 0 ? 1 : nthreads;
		624	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		625	}
		626
		627	static void etp_set_min_parallel (unsigned int nthreads)
		628	{
		629	if (wanted < nthreads)
		630	wanted = nthreads;
		631	}
		632
		633	static void etp_set_max_parallel (unsigned int nthreads)
		634	{
		635	if (wanted > nthreads)
		636	wanted = nthreads;
		637
		638	while (started > wanted)
		639	etp_end_thread ();
		640	}
		641
		642	/*****************************************************************************/
		643
		644	static void grp_try_feed (eio_req *grp)
		645	{
		646	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		647	{
		648	grp->flags &= ~EIO_FLAG_GROUPADD;
		649
		650	EIO_FEED (grp);
		651
		652	/* stop if no progress has been made */
		653	if (!(grp->flags & EIO_FLAG_GROUPADD))
		654	{
		655	grp->feed = 0;
		656	break;
		657	}
		658	}
		659	}
		660
		661	static int grp_dec (eio_req *grp)
		662	{
		663	--grp->size;
		664
		665	/* call feeder, if applicable */
		666	grp_try_feed (grp);
		667
		668	/* finish, if done */
		669	if (!grp->size && grp->int1)
		670	return eio_finish (grp);
		671	else
		672	return 0;
		673	}
		674
		675	void eio_destroy (eio_req *req)
		676	{
		677	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		678	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		679
		680	EIO_DESTROY (req);
		681	}
		682
		683	static int eio_finish (eio_req *req)
		684	{
		685	int res = EIO_FINISH (req);
		686
		687	if (req->grp)
		688	{
		689	int res2;
		690	eio_req *grp = req->grp;
		691
		692	/* unlink request */
		693	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		694	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		695
		696	if (grp->grp_first == req)
		697	grp->grp_first = req->grp_next;
		698
		699	res2 = grp_dec (grp);
		700
		701	if (!res && res2)
		702	res = res2;
		703	}
		704
		705	eio_destroy (req);
		706
		707	return res;
		708	}
		709
		710	void eio_grp_cancel (eio_req *grp)
		711	{
		712	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		713	eio_cancel (grp);
		714	}
		715
		716	void eio_cancel (eio_req *req)
		717	{
		718	etp_cancel (req);
		719	}
		720
		721	void eio_submit (eio_req *req)
		722	{
		723	etp_submit (req);
		724	}
		725
		726	unsigned int eio_nreqs (void)
		727	{
		728	return etp_nreqs ();
		729	}
		730
		731	unsigned int eio_nready (void)
		732	{
		733	return etp_nready ();
		734	}
		735
		736	unsigned int eio_npending (void)
		737	{
		738	return etp_npending ();
		739	}
		740
		741	unsigned int eio_nthreads (void)
		742	{
		743	return etp_nthreads ();
		744	}
		745
		746	void eio_set_max_poll_time (double nseconds)
		747	{
		748	etp_set_max_poll_time (nseconds);
		749	}
		750
		751	void eio_set_max_poll_reqs (unsigned int maxreqs)
		752	{
		753	etp_set_max_poll_reqs (maxreqs);
		754	}
		755
		756	void eio_set_max_idle (unsigned int nthreads)
		757	{
		758	etp_set_max_idle (nthreads);
		759	}
		760
		761	void eio_set_min_parallel (unsigned int nthreads)
		762	{
		763	etp_set_min_parallel (nthreads);
		764	}
		765
		766	void eio_set_max_parallel (unsigned int nthreads)
		767	{
		768	etp_set_max_parallel (nthreads);
		769	}
		770
		771	int eio_poll (void)
		772	{
		773	return etp_poll ();
		774	}
		775
489	/*****************************************************************************/	776	/*****************************************************************************/
490	/* work around various missing functions */	777	/* work around various missing functions */
491		778
492	#if !HAVE_PREADWRITE	779	#if !HAVE_PREADWRITE
		780	# undef pread
		781	# undef pwrite
493	# define pread eio__pread	782	# define pread eio__pread
494	# define pwrite eio__pwrite	783	# define pwrite eio__pwrite
495
496	/*
497	* make our pread/pwrite safe against themselves, but not against
498	* normal read/write by using a mutex. slows down execution a lot,
499	* but that's your problem, not mine.
500	*/
501	static mutex_t preadwritelock = X_MUTEX_INIT;
502		784
503	static ssize_t	785	static ssize_t
504	eio__pread (int fd, void *buf, size_t count, off_t offset)	786	eio__pread (int fd, void *buf, size_t count, off_t offset)
505	{	787	{
506	ssize_t res;	788	ssize_t res;
…		…
524		806
525	X_LOCK (preadwritelock);	807	X_LOCK (preadwritelock);
526	ooffset = lseek (fd, 0, SEEK_CUR);	808	ooffset = lseek (fd, 0, SEEK_CUR);
527	lseek (fd, offset, SEEK_SET);	809	lseek (fd, offset, SEEK_SET);
528	res = write (fd, buf, count);	810	res = write (fd, buf, count);
529	lseek (fd, offset, SEEK_SET);	811	lseek (fd, ooffset, SEEK_SET);
530	X_UNLOCK (preadwritelock);	812	X_UNLOCK (preadwritelock);
531		813
532	return res;	814	return res;
533	}	815	}
534	#endif	816	#endif
535		817
536	#ifndef HAVE_FUTIMES	818	#ifndef HAVE_FUTIMES
537		819
		820	# undef utimes
		821	# undef futimes
538	# define utimes(path,times) eio__utimes (path, times)	822	# define utimes(path,times) eio__utimes (path, times)
539	# define futimes(fd,times) eio__futimes (fd, times)	823	# define futimes(fd,times) eio__futimes (fd, times)
540		824
541	static int	825	static int
542	eio__utimes (const char *filename, const struct timeval times[2])	826	eio__utimes (const char *filename, const struct timeval times[2])
…		…
561	}	845	}
562		846
563	#endif	847	#endif
564		848
565	#if !HAVE_FDATASYNC	849	#if !HAVE_FDATASYNC
		850	# undef fdatasync
566	# define fdatasync fsync	851	# define fdatasync(fd) fsync (fd)
567	#endif	852	#endif
		853
		854	/* sync_file_range always needs emulation */
		855	int
		856	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		857	{
		858	#if HAVE_SYNC_FILE_RANGE
		859	int res;
		860
		861	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		862	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		863	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		864	{
		865	flags = 0
		866	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		867	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		868	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		869	}
		870
		871	res = sync_file_range (fd, offset, nbytes, flags);
		872
		873	if (!res || errno != ENOSYS)
		874	return res;
		875	#endif
		876
		877	/* even though we could play tricks with the flags, it's better to always
		878	* call fdatasync, as that matches the expectation of its users best */
		879	return fdatasync (fd);
		880	}
568		881
569	#if !HAVE_READAHEAD	882	#if !HAVE_READAHEAD
		883	# undef readahead
570	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	884	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
571		885
572	static ssize_t	886	static ssize_t
573	eio__readahead (int fd, off_t offset, size_t count, worker *self)	887	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
574	{	888	{
575	size_t todo = count;	889	size_t todo = count;
576	dBUF;	890	dBUF;
577		891
578	while (todo > 0)	892	while (todo > 0)
…		…
588	return count;	902	return count;
589	}	903	}
590		904
591	#endif	905	#endif
592		906
593	#if !HAVE_READDIR_R
594	# define readdir_r eio__readdir_r
595
596	static mutex_t readdirlock = X_MUTEX_INIT;
597
598	static int
599	eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
600	{
601	EIO_STRUCT_DIRENT *e;
602	int errorno;
603
604	X_LOCK (readdirlock);
605
606	e = readdir (dirp);
607	errorno = errno;
608
609	if (e)
610	{
611	*res = ent;
612	strcpy (ent->d_name, e->d_name);
613	}
614	else
615	*res = 0;
616
617	X_UNLOCK (readdirlock);
618
619	errno = errorno;
620	return e ? 0 : -1;
621	}
622	#endif
623
624	/* sendfile always needs emulation */	907	/* sendfile always needs emulation */
625	static ssize_t	908	static ssize_t
626	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self)	909	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
627	{	910	{
628	ssize_t res;	911	ssize_t res;
629		912
630	if (!count)	913	if (!count)
631	return 0;	914	return 0;
632		915
633	#if HAVE_SENDFILE	916	#if HAVE_SENDFILE
634	# if __linux	917	# if __linux
635	res = sendfile (ofd, ifd, &offset, count);	918	res = sendfile (ofd, ifd, &offset, count);
636		919
637	# elif __freebsd	920	# elif __FreeBSD__
638	/*	921	/*
639	* Of course, the freebsd sendfile is a dire hack with no thoughts	922	* Of course, the freebsd sendfile is a dire hack with no thoughts
640	* wasted on making it similar to other I/O functions.	923	* wasted on making it similar to other I/O functions.
641	*/	924	*/
642	{	925	{
643	off_t sbytes;	926	off_t sbytes;
644	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	927	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
645		928
646	if (res < 0 && sbytes)	929	#if 0 /* according to the manpage, this is correct, but broken behaviour */
647	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	930	/* freebsd' sendfile will return 0 on success */
		931	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		932	/* not on e.g. EIO or EPIPE - sounds broken */
		933	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
		934	res = sbytes;
		935	#endif
		936
		937	/* according to source inspection, this is correct, and useful behaviour */
		938	if (sbytes)
		939	res = sbytes;
		940	}
		941
		942	# elif defined (__APPLE__)
		943
		944	{
		945	off_t sbytes = count;
		946	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		947
		948	/* according to the manpage, sbytes is always valid */
		949	if (sbytes)
648	res = sbytes;	950	res = sbytes;
649	}	951	}
650		952
651	# elif __hpux	953	# elif __hpux
652	res = sendfile (ofd, ifd, offset, count, 0, 0);	954	res = sendfile (ofd, ifd, offset, count, 0, 0);
…		…
666	if (res < 0 && sbytes)	968	if (res < 0 && sbytes)
667	res = sbytes;	969	res = sbytes;
668	}	970	}
669		971
670	# endif	972	# endif
		973
		974	#elif defined (_WIN32)
		975
		976	/* does not work, just for documentation of what would need to be done */
		977	{
		978	HANDLE h = TO_SOCKET (ifd);
		979	SetFilePointer (h, offset, 0, FILE_BEGIN);
		980	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		981	}
		982
671	#else	983	#else
672	res = -1;	984	res = -1;
673	errno = ENOSYS;	985	errno = ENOSYS;
674	#endif	986	#endif
675		987
676	if (res < 0	988	if (res < 0
677	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	989	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		990	/* BSDs */
		991	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		992	|| errno == ENOTSUP
		993	#endif
		994	|| errno == EOPNOTSUPP /* BSDs */
678	#if __solaris	995	#if __solaris
679	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	996	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
680	#endif	997	#endif
681	)	998	)
682	)	999	)
…		…
713	}	1030	}
714		1031
715	return res;	1032	return res;
716	}	1033	}
717		1034
		1035	static signed char
		1036	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1037	{
		1038	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1039	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
		1040	}
		1041
		1042	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1043
		1044	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1045	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1046
		1047	static void
		1048	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1049	{
		1050	unsigned char bits [9 + sizeof (ino_t) * 8];
		1051	unsigned char *bit = bits;
		1052
		1053	assert (CHAR_BIT == 8);
		1054	assert (sizeof (eio_dirent) * 8 < 256);
		1055	assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
		1056	assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
		1057
		1058	if (size <= EIO_SORT_FAST)
		1059	return;
		1060
		1061	/* first prepare an array of bits to test in our radix sort */
		1062	/* try to take endianness into account, as well as differences in ino_t sizes */
		1063	/* inode_bits must contain all inodes ORed together */
		1064	/* which is used to skip bits that are 0 everywhere, which is very common */
		1065	{
		1066	ino_t endianness;
		1067	int i, j;
		1068
		1069	/* we store the byte offset of byte n into byte n of "endianness" */
		1070	for (i = 0; i < sizeof (ino_t); ++i)
		1071	((unsigned char *)&endianness)[i] = i;
		1072
		1073	*bit++ = 0;
		1074
		1075	for (i = 0; i < sizeof (ino_t); ++i)
		1076	{
		1077	/* shifting off the byte offsets out of "endianness" */
		1078	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1079	endianness >>= 8;
		1080
		1081	for (j = 0; j < 8; ++j)
		1082	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1083	*bit++ = offs + j;
		1084	}
		1085
		1086	for (j = 0; j < 8; ++j)
		1087	if (score_bits & (1 << j))
		1088	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1089	}
		1090
		1091	/* now actually do the sorting (a variant of MSD radix sort) */
		1092	{
		1093	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1094	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1095	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1096	int stk_idx = 0;
		1097
		1098	base_stk [stk_idx] = dents;
		1099	end_stk [stk_idx] = dents + size;
		1100	bit_stk [stk_idx] = bit - 1;
		1101
		1102	do
		1103	{
		1104	base = base_stk [stk_idx];
		1105	end = end_stk [stk_idx];
		1106	bit = bit_stk [stk_idx];
		1107
		1108	for (;;)
		1109	{
		1110	unsigned char O = *bit >> 3;
		1111	unsigned char M = 1 << (*bit & 7);
		1112
		1113	eio_dirent *a = base;
		1114	eio_dirent *b = end;
		1115
		1116	if (b - a < EIO_SORT_CUTOFF)
		1117	break;
		1118
		1119	/* now bit-partition the array on the bit */
		1120	/* this ugly asymmetric loop seems to perform much better than typical */
		1121	/* partition algos found in the literature */
		1122	do
		1123	if (!(((unsigned char *)a)[O] & M))
		1124	++a;
		1125	else if (!(((unsigned char *)--b)[O] & M))
		1126	{
		1127	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1128	++a;
		1129	}
		1130	while (b > a);
		1131
		1132	/* next bit, or stop, if no bits left in this path */
		1133	if (!*--bit)
		1134	break;
		1135
		1136	base_stk [stk_idx] = a;
		1137	end_stk [stk_idx] = end;
		1138	bit_stk [stk_idx] = bit;
		1139	++stk_idx;
		1140
		1141	end = a;
		1142	}
		1143	}
		1144	while (stk_idx--);
		1145	}
		1146	}
		1147
		1148	static void
		1149	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1150	{
		1151	/* first move the smallest element to the front, to act as a sentinel */
		1152	{
		1153	int i;
		1154	eio_dirent *min = dents;
		1155
		1156	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1157	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1158	if (EIO_DENT_CMP (dents [i], <, *min))
		1159	min = &dents [i];
		1160
		1161	/* swap elements 0 and j (minimum) */
		1162	{
		1163	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1164	}
		1165	}
		1166
		1167	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1168	{
		1169	eio_dirent *i, *j;
		1170
		1171	for (i = dents + 1; i < dents + size; ++i)
		1172	{
		1173	eio_dirent value = *i;
		1174
		1175	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1176	j [1] = j [0];
		1177
		1178	j [1] = value;
		1179	}
		1180	}
		1181	}
		1182
		1183	static void
		1184	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1185	{
		1186	if (size <= 1)
		1187	return; /* our insertion sort relies on size > 0 */
		1188
		1189	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1190	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1191	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1192
		1193	/* use an insertion sort at the end, or for small arrays, */
		1194	/* as insertion sort is more efficient for small partitions */
		1195	eio_dent_insertion_sort (dents, size);
		1196	}
		1197
718	/* read a full directory */	1198	/* read a full directory */
719	static void	1199	static void
720	eio__scandir (eio_req *req, worker *self)	1200	eio__scandir (eio_req *req, etp_worker *self)
721	{	1201	{
722	DIR *dirp;	1202	DIR *dirp;
723	union
724	{
725	EIO_STRUCT_DIRENT d;
726	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
727	} *u;
728	EIO_STRUCT_DIRENT *entp;	1203	EIO_STRUCT_DIRENT *entp;
729	char *name, *names;	1204	char *name, *names;
730	int memlen = 4096;	1205	int namesalloc = 4096;
731	int memofs = 0;	1206	int namesoffs = 0;
		1207	int flags = req->int1;
		1208	eio_dirent *dents = 0;
		1209	int dentalloc = 128;
732	int res = 0;	1210	int dentoffs = 0;
		1211	ino_t inode_bits = 0;
		1212
		1213	req->result = -1;
		1214
		1215	if (!(flags & EIO_READDIR_DENTS))
		1216	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
733		1217
734	X_LOCK (wrklock);	1218	X_LOCK (wrklock);
		1219	/* the corresponding closedir is in ETP_WORKER_CLEAR */
735	self->dirp = dirp = opendir (req->ptr1);	1220	self->dirp = dirp = opendir (req->ptr1);
736	self->dbuf = u = malloc (sizeof (*u));
737	req->flags |= EIO_FLAG_PTR2_FREE;	1221	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1222	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
738	req->ptr2 = names = malloc (memlen);	1223	req->ptr2 = names = malloc (namesalloc);
739	X_UNLOCK (wrklock);	1224	X_UNLOCK (wrklock);
740		1225
741	if (dirp && u && names)	1226	if (dirp && names && (!flags || dents))
742	for (;;)	1227	for (;;)
743	{	1228	{
744	errno = 0;	1229	errno = 0;
745	readdir_r (dirp, &u->d, &entp);	1230	entp = readdir (dirp);
746		1231
747	if (!entp)	1232	if (!entp)
		1233	{
		1234	if (errno)
		1235	break;
		1236
		1237	/* sort etc. */
		1238	req->int1 = flags;
		1239	req->result = dentoffs;
		1240
		1241	if (flags & EIO_READDIR_STAT_ORDER)
		1242	eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
		1243	else if (flags & EIO_READDIR_DIRS_FIRST)
		1244	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1245	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1246	else
		1247	{
		1248	/* in this case, all is known, and we just put dirs first and sort them */
		1249	eio_dirent *oth = dents + dentoffs;
		1250	eio_dirent *dir = dents;
		1251
		1252	/* now partition dirs to the front, and non-dirs to the back */
		1253	/* by walking from both sides and swapping if necessary */
		1254	/* also clear score, so it doesn't influence sorting */
		1255	while (oth > dir)
		1256	{
		1257	if (dir->type == EIO_DT_DIR)
		1258	++dir;
		1259	else if ((--oth)->type == EIO_DT_DIR)
		1260	{
		1261	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1262
		1263	++dir;
		1264	}
		1265	}
		1266
		1267	/* now sort the dirs only */
		1268	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1269	}
		1270
748	break;	1271	break;
		1272	}
749		1273
		1274	/* now add the entry to our list(s) */
750	name = entp->d_name;	1275	name = entp->d_name;
751		1276
		1277	/* skip . and .. entries */
752	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1278	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
753	{	1279	{
754	int len = strlen (name) + 1;	1280	int len = D_NAMLEN (entp) + 1;
755		1281
756	res++;	1282	while (expect_false (namesoffs + len > namesalloc))
757
758	while (memofs + len > memlen)
759	{	1283	{
760	memlen *= 2;	1284	namesalloc *= 2;
761	X_LOCK (wrklock);	1285	X_LOCK (wrklock);
762	req->ptr2 = names = realloc (names, memlen);	1286	req->ptr2 = names = realloc (names, namesalloc);
763	X_UNLOCK (wrklock);	1287	X_UNLOCK (wrklock);
764		1288
765	if (!names)	1289	if (!names)
766	break;	1290	break;
767	}	1291	}
768		1292
769	memcpy (names + memofs, name, len);	1293	memcpy (names + namesoffs, name, len);
		1294
		1295	if (dents)
		1296	{
		1297	struct eio_dirent *ent;
		1298
		1299	if (expect_false (dentoffs == dentalloc))
		1300	{
		1301	dentalloc *= 2;
		1302	X_LOCK (wrklock);
		1303	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1304	X_UNLOCK (wrklock);
		1305
		1306	if (!dents)
		1307	break;
		1308	}
		1309
		1310	ent = dents + dentoffs;
		1311
		1312	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1313	ent->namelen = len - 1;
		1314	ent->inode = D_INO (entp);
		1315
		1316	inode_bits |= ent->inode;
		1317
		1318	switch (D_TYPE (entp))
		1319	{
		1320	default:
		1321	ent->type = EIO_DT_UNKNOWN;
		1322	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1323	break;
		1324
		1325	#ifdef DT_FIFO
		1326	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1327	#endif
		1328	#ifdef DT_CHR
		1329	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1330	#endif
		1331	#ifdef DT_MPC
		1332	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1333	#endif
		1334	#ifdef DT_DIR
		1335	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1336	#endif
		1337	#ifdef DT_NAM
		1338	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1339	#endif
		1340	#ifdef DT_BLK
		1341	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1342	#endif
		1343	#ifdef DT_MPB
		1344	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1345	#endif
		1346	#ifdef DT_REG
		1347	case DT_REG: ent->type = EIO_DT_REG; break;
		1348	#endif
		1349	#ifdef DT_NWK
		1350	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1351	#endif
		1352	#ifdef DT_CMP
		1353	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1354	#endif
		1355	#ifdef DT_LNK
		1356	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1357	#endif
		1358	#ifdef DT_SOCK
		1359	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1360	#endif
		1361	#ifdef DT_DOOR
		1362	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1363	#endif
		1364	#ifdef DT_WHT
		1365	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1366	#endif
		1367	}
		1368
		1369	ent->score = 7;
		1370
		1371	if (flags & EIO_READDIR_DIRS_FIRST)
		1372	{
		1373	if (ent->type == EIO_DT_UNKNOWN)
		1374	{
		1375	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1376	ent->score = 1;
		1377	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1378	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1379	}
		1380	else if (ent->type == EIO_DT_DIR)
		1381	ent->score = 0;
		1382	}
		1383	}
		1384
770	memofs += len;	1385	namesoffs += len;
		1386	++dentoffs;
		1387	}
		1388
		1389	if (EIO_CANCELLED (req))
		1390	{
		1391	errno = ECANCELED;
		1392	break;
771	}	1393	}
772	}	1394	}
		1395	}
773		1396
774	if (errno)	1397	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
775	res = -1;	1398	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
776		1399	#else
777	req->result = res;	1400
		1401	int
		1402	eio__msync (void *mem, size_t len, int flags)
		1403	{
		1404	if (EIO_MS_ASYNC != MS_SYNC
		1405	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1406	|| EIO_MS_SYNC != MS_SYNC)
		1407	{
		1408	flags = 0
		1409	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1410	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1411	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1412	}
		1413
		1414	return msync (mem, len, flags);
		1415	}
		1416
		1417	#endif
		1418
		1419	int
		1420	eio__mtouch (void *mem, size_t len, int flags)
		1421	{
		1422	intptr_t addr = (intptr_t)mem;
		1423	intptr_t end = addr + len;
		1424	#ifdef PAGESIZE
		1425	const intptr_t page = PAGESIZE;
		1426	#else
		1427	static intptr_t page;
		1428
		1429	if (!page)
		1430	page = sysconf (_SC_PAGESIZE);
		1431	#endif
		1432
		1433	/* round down to start of page, although this is probably useless */
		1434	addr &= ~(page - 1); /* assume page size is always a power of two */
		1435
		1436	if (addr < end)
		1437	if (flags & EIO_MT_MODIFY) /* modify */
		1438	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
		1439	else
		1440	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
		1441
		1442	return 0;
778	}	1443	}
779		1444
780	/*****************************************************************************/	1445	/*****************************************************************************/
781		1446
782	#define ALLOC(len) \	1447	#define ALLOC(len) \
…		…
792	req->result = -1; \	1457	req->result = -1; \
793	break; \	1458	break; \
794	} \	1459	} \
795	}	1460	}
796		1461
797	X_THREAD_PROC (eio_proc)	1462	X_THREAD_PROC (etp_proc)
798	{	1463	{
799	eio_req *req;	1464	ETP_REQ *req;
800	struct timespec ts;	1465	struct timespec ts;
801	worker *self = (worker *)thr_arg;	1466	etp_worker *self = (etp_worker *)thr_arg;
802		1467
803	/* try to distribute timeouts somewhat randomly */	1468	/* try to distribute timeouts somewhat randomly */
804	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1469	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
805		1470
806	for (;;)	1471	for (;;)
807	{	1472	{
808	ts.tv_sec = time (0) + IDLE_TIMEOUT;
809
810	X_LOCK (reqlock);	1473	X_LOCK (reqlock);
811		1474
812	for (;;)	1475	for (;;)
813	{	1476	{
814	self->req = req = reqq_shift (&req_queue);	1477	self->req = req = reqq_shift (&req_queue);
…		…
816	if (req)	1479	if (req)
817	break;	1480	break;
818		1481
819	++idle;	1482	++idle;
820		1483
		1484	ts.tv_sec = time (0) + IDLE_TIMEOUT;
821	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1485	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
822	{	1486	{
823	if (idle > max_idle)	1487	if (idle > max_idle)
824	{	1488	{
825	--idle;	1489	--idle;
…		…
830	goto quit;	1494	goto quit;
831	}	1495	}
832		1496
833	/* we are allowed to idle, so do so without any timeout */	1497	/* we are allowed to idle, so do so without any timeout */
834	X_COND_WAIT (reqwait, reqlock);	1498	X_COND_WAIT (reqwait, reqlock);
835	ts.tv_sec = time (0) + IDLE_TIMEOUT;
836	}	1499	}
837		1500
838	--idle;	1501	--idle;
839	}	1502	}
840		1503
841	--nready;	1504	--nready;
842		1505
843	X_UNLOCK (reqlock);	1506	X_UNLOCK (reqlock);
844		1507
845	errno = 0; /* strictly unnecessary */	1508	if (req->type < 0)
		1509	goto quit;
846		1510
847	if (!EIO_CANCELLED (req))	1511	if (!EIO_CANCELLED (req))
848	switch (req->type)	1512	ETP_EXECUTE (self, req);
849	{
850	case EIO_READ: ALLOC (req->size);
851	req->result = req->offs >= 0
852	? pread (req->int1, req->ptr2, req->size, req->offs)
853	: read (req->int1, req->ptr2, req->size); break;
854	case EIO_WRITE: req->result = req->offs >= 0
855	? pwrite (req->int1, req->ptr2, req->size, req->offs)
856	: write (req->int1, req->ptr2, req->size); break;
857
858	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
859	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
860
861	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
862	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
863	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
864	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
865	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
866	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
867
868	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
869	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
870	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
871	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
872	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
873	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
874
875	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
876	case EIO_CLOSE: req->result = close (req->int1); break;
877	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
878	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
879	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
880	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
881	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
882	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
883	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
884	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
885
886	case EIO_READLINK: ALLOC (NAME_MAX);
887	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
888
889	case EIO_SYNC: req->result = 0; sync (); break;
890	case EIO_FSYNC: req->result = fsync (req->int1); break;
891	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
892
893	case EIO_READDIR: eio__scandir (req, self); break;
894
895	case EIO_BUSY:
896	#ifdef _WIN32
897	Sleep (req->nv1 * 1000.);
898	#else
899	{
900	struct timeval tv;
901
902	tv.tv_sec = req->nv1;
903	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
904
905	req->result = select (0, 0, 0, 0, &tv);
906	}
907	#endif
908	break;
909
910	case EIO_UTIME:
911	case EIO_FUTIME:
912	{
913	struct timeval tv[2];
914	struct timeval *times;
915
916	if (req->nv1 != -1. || req->nv2 != -1.)
917	{
918	tv[0].tv_sec = req->nv1;
919	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
920	tv[1].tv_sec = req->nv2;
921	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
922
923	times = tv;
924	}
925	else
926	times = 0;
927
928
929	req->result = req->type == EIO_FUTIME
930	? futimes (req->int1, times)
931	: utimes (req->ptr1, times);
932	}
933
934	case EIO_GROUP:
935	case EIO_NOP:
936	req->result = 0;
937	break;
938
939	case EIO_QUIT:
940	goto quit;
941
942	default:
943	req->result = -1;
944	break;
945	}
946
947	req->errorno = errno;
948		1513
949	X_LOCK (reslock);	1514	X_LOCK (reslock);
950		1515
951	++npending;	1516	++npending;
952		1517
953	if (!reqq_push (&res_queue, req) && want_poll_cb)	1518	if (!reqq_push (&res_queue, req) && want_poll_cb)
954	want_poll_cb ();	1519	want_poll_cb ();
955		1520
956	self->req = 0;	1521	self->req = 0;
957	worker_clear (self);	1522	etp_worker_clear (self);
958		1523
959	X_UNLOCK (reslock);	1524	X_UNLOCK (reslock);
960	}	1525	}
961		1526
962	quit:	1527	quit:
963	X_LOCK (wrklock);	1528	X_LOCK (wrklock);
964	worker_free (self);	1529	etp_worker_free (self);
965	X_UNLOCK (wrklock);	1530	X_UNLOCK (wrklock);
966		1531
967	return 0;	1532	return 0;
968	}	1533	}
969		1534
970	/*****************************************************************************/	1535	/*****************************************************************************/
971		1536
972	static void eio_atfork_prepare (void)
973	{
974	X_LOCK (wrklock);
975	X_LOCK (reqlock);
976	X_LOCK (reslock);
977	#if !HAVE_PREADWRITE
978	X_LOCK (preadwritelock);
979	#endif
980	#if !HAVE_READDIR_R
981	X_LOCK (readdirlock);
982	#endif
983	}
984
985	static void eio_atfork_parent (void)
986	{
987	#if !HAVE_READDIR_R
988	X_UNLOCK (readdirlock);
989	#endif
990	#if !HAVE_PREADWRITE
991	X_UNLOCK (preadwritelock);
992	#endif
993	X_UNLOCK (reslock);
994	X_UNLOCK (reqlock);
995	X_UNLOCK (wrklock);
996	}
997
998	static void eio_atfork_child (void)
999	{
1000	eio_req *prv;
1001
1002	while (prv = reqq_shift (&req_queue))
1003	eio_destroy (prv);
1004
1005	while (prv = reqq_shift (&res_queue))
1006	eio_destroy (prv);
1007
1008	while (wrk_first.next != &wrk_first)
1009	{
1010	worker *wrk = wrk_first.next;
1011
1012	if (wrk->req)
1013	eio_destroy (wrk->req);
1014
1015	worker_clear (wrk);
1016	worker_free (wrk);
1017	}
1018
1019	started = 0;
1020	idle = 0;
1021	nreqs = 0;
1022	nready = 0;
1023	npending = 0;
1024
1025	eio_atfork_parent ();
1026	}
1027
1028	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1537	int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1029	{	1538	{
1030	want_poll_cb = want_poll;	1539	return etp_init (want_poll, done_poll);
1031	done_poll_cb = done_poll;
1032
1033	#ifdef _WIN32
1034	X_MUTEX_CHECK (wrklock);
1035	X_MUTEX_CHECK (reslock);
1036	X_MUTEX_CHECK (reqlock);
1037	X_MUTEX_CHECK (reqwait);
1038	X_MUTEX_CHECK (preadwritelock);
1039	X_MUTEX_CHECK (readdirlock);
1040
1041	X_COND_CHECK (reqwait);
1042	#endif
1043
1044	X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1045	}	1540	}
1046		1541
1047	static void eio_api_destroy (eio_req *req)	1542	static void eio_api_destroy (eio_req *req)
1048	{	1543	{
1049	free (req);	1544	free (req);
…		…
1071	{ \	1566	{ \
1072	eio_api_destroy (req); \	1567	eio_api_destroy (req); \
1073	return 0; \	1568	return 0; \
1074	}	1569	}
1075		1570
		1571	static void eio_execute (etp_worker *self, eio_req *req)
		1572	{
		1573	errno = 0;
		1574
		1575	switch (req->type)
		1576	{
		1577	case EIO_READ: ALLOC (req->size);
		1578	req->result = req->offs >= 0
		1579	? pread (req->int1, req->ptr2, req->size, req->offs)
		1580	: read (req->int1, req->ptr2, req->size); break;
		1581	case EIO_WRITE: req->result = req->offs >= 0
		1582	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1583	: write (req->int1, req->ptr2, req->size); break;
		1584
		1585	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		1586	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
		1587
		1588	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1589	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1590	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1591	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1592	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1593	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1594
		1595	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1596	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1597	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1598	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1599
		1600	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1601	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1602	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1603	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1604	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1605	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1606
		1607	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1608	case EIO_CLOSE: req->result = close (req->int1); break;
		1609	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1610	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1611	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		1612	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		1613	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		1614	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		1615	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		1616	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
		1617
		1618	case EIO_READLINK: ALLOC (PATH_MAX);
		1619	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
		1620
		1621	case EIO_SYNC: req->result = 0; sync (); break;
		1622	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1623	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1624	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1625	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1626	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1627
		1628	case EIO_READDIR: eio__scandir (req, self); break;
		1629
		1630	case EIO_BUSY:
		1631	#ifdef _WIN32
		1632	Sleep (req->nv1 * 1000.);
		1633	#else
		1634	{
		1635	struct timeval tv;
		1636
		1637	tv.tv_sec = req->nv1;
		1638	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
		1639
		1640	req->result = select (0, 0, 0, 0, &tv);
		1641	}
		1642	#endif
		1643	break;
		1644
		1645	case EIO_UTIME:
		1646	case EIO_FUTIME:
		1647	{
		1648	struct timeval tv[2];
		1649	struct timeval *times;
		1650
		1651	if (req->nv1 != -1. || req->nv2 != -1.)
		1652	{
		1653	tv[0].tv_sec = req->nv1;
		1654	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		1655	tv[1].tv_sec = req->nv2;
		1656	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		1657
		1658	times = tv;
		1659	}
		1660	else
		1661	times = 0;
		1662
		1663
		1664	req->result = req->type == EIO_FUTIME
		1665	? futimes (req->int1, times)
		1666	: utimes (req->ptr1, times);
		1667	}
		1668	break;
		1669
		1670	case EIO_GROUP:
		1671	abort (); /* handled in eio_request */
		1672
		1673	case EIO_NOP:
		1674	req->result = 0;
		1675	break;
		1676
		1677	case EIO_CUSTOM:
		1678	((void (*)(eio_req *))req->feed) (req);
		1679	break;
		1680
		1681	default:
		1682	req->result = -1;
		1683	break;
		1684	}
		1685
		1686	req->errorno = errno;
		1687	}
		1688
		1689	#ifndef EIO_NO_WRAPPERS
		1690
1076	eio_req *eio_nop (int pri, eio_cb cb, void *data)	1691	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1077	{	1692	{
1078	REQ (EIO_NOP); SEND;	1693	REQ (EIO_NOP); SEND;
1079	}	1694	}
1080		1695
…		…
1091	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	1706	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1092	{	1707	{
1093	REQ (EIO_FSYNC); req->int1 = fd; SEND;	1708	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1094	}	1709	}
1095		1710
		1711	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1712	{
		1713	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1714	}
		1715
		1716	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		1717	{
		1718	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		1719	}
		1720
		1721	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		1722	{
		1723	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		1724	}
		1725
1096	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	1726	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1097	{	1727	{
1098	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1728	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1099	}	1729	}
1100		1730
…		…
1119	}	1749	}
1120		1750
1121	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	1751	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1122	{	1752	{
1123	REQ (EIO_FSTAT); req->int1 = fd; SEND;	1753	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1754	}
		1755
		1756	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		1757	{
		1758	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1124	}	1759	}
1125		1760
1126	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	1761	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1127	{	1762	{
1128	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	1763	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
…		…
1202	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	1837	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1203	{	1838	{
1204	return eio__1path (EIO_LSTAT, path, pri, cb, data);	1839	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1205	}	1840	}
1206		1841
		1842	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		1843	{
		1844	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		1845	}
		1846
1207	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	1847	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1208	{	1848	{
1209	return eio__1path (EIO_UNLINK, path, pri, cb, data);	1849	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1210	}	1850	}
1211		1851
1212	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	1852	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1213	{	1853	{
1214	return eio__1path (EIO_RMDIR, path, pri, cb, data);	1854	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1215	}	1855	}
1216		1856
1217	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	1857	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1218	{	1858	{
1219	return eio__1path (EIO_READDIR, path, pri, cb, data);	1859	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1220	}	1860	}
1221		1861
1222	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	1862	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1223	{	1863	{
1224	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	1864	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1225	}	1865	}
1226		1866
1227	static eio_req *	1867	static eio_req *
1228	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1868	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1229	{	1869	{
…		…
1253	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	1893	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1254	{	1894	{
1255	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	1895	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1256	}	1896	}
1257		1897
		1898	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
		1899	{
		1900	REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
		1901	}
		1902
		1903	#endif
		1904
1258	eio_req *eio_grp (eio_cb cb, void *data)	1905	eio_req *eio_grp (eio_cb cb, void *data)
1259	{	1906	{
1260	const int pri = EIO_PRI_MAX;	1907	const int pri = EIO_PRI_MAX;
1261		1908
1262	REQ (EIO_GROUP); SEND;	1909	REQ (EIO_GROUP); SEND;
…		…
1286		1933
1287	void eio_grp_add (eio_req *grp, eio_req *req)	1934	void eio_grp_add (eio_req *grp, eio_req *req)
1288	{	1935	{
1289	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	1936	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1290		1937
		1938	grp->flags |= EIO_FLAG_GROUPADD;
		1939
1291	++grp->size;	1940	++grp->size;
1292	req->grp = grp;	1941	req->grp = grp;
1293		1942
1294	req->grp_prev = 0;	1943	req->grp_prev = 0;
1295	req->grp_next = grp->grp_first;	1944	req->grp_next = grp->grp_first;
…		…
1303	/*****************************************************************************/	1952	/*****************************************************************************/
1304	/* misc garbage */	1953	/* misc garbage */
1305		1954
1306	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	1955	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1307	{	1956	{
1308	worker wrk;	1957	etp_worker wrk;
		1958	ssize_t ret;
1309		1959
1310	wrk.dbuf = 0;	1960	wrk.dbuf = 0;
1311		1961
1312	eio__sendfile (ofd, ifd, offset, count, &wrk);	1962	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1313		1963
1314	if (wrk.dbuf)	1964	if (wrk.dbuf)
1315	free (wrk.dbuf);	1965	free (wrk.dbuf);
1316	}
1317		1966
		1967	return ret;
		1968	}
		1969

Diff Legend

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