ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.13 by root, Tue May 13 19:34:11 2008 UTC vs.
Revision 1.89 by root, Thu Jul 14 22:36:17 2011 UTC

		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
		5	* All rights reserved.
		6	*
		7	* Redistribution and use in source and binary forms, with or without modifica-
		8	* tion, are permitted provided that the following conditions are met:
		9	*
		10	* 1. Redistributions of source code must retain the above copyright notice,
		11	* this list of conditions and the following disclaimer.
		12	*
		13	* 2. Redistributions in binary form must reproduce the above copyright
		14	* notice, this list of conditions and the following disclaimer in the
		15	* documentation and/or other materials provided with the distribution.
		16	*
		17	* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
		18	* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
		19	* CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
		20	* EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
		21	* CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
		22	* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
		23	* OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
		24	* WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
		25	* ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
		26	* OF THE POSSIBILITY OF SUCH DAMAGE.
		27	*
		28	* Alternatively, the contents of this file may be used under the terms of
		29	* the GNU General Public License ("GPL") version 2 or any later version,
		30	* in which case the provisions of the GPL are applicable instead of
		31	* the above. If you wish to allow the use of your version of this file
		32	* only under the terms of the GPL and not to allow others to use your
		33	* version of this file under the BSD license, indicate your decision
		34	* by deleting the provisions above and replace them with the notice
		35	* and other provisions required by the GPL. If you do not delete the
		36	* provisions above, a recipient may use your version of this file under
		37	* either the BSD or the GPL.
		38	*/
		39
		40	#ifndef _WIN32
		41	# include "config.h"
		42	#endif
		43
1	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
		46
		47	#ifdef EIO_STACKSIZE
		48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
		49	#endif
2	#include "xthread.h"	50	#include "xthread.h"
3		51
4	#include <errno.h>	52	#include <errno.h>
5	#include <stddef.h>	53	#include <stddef.h>
6	#include <stdlib.h>	54	#include <stdlib.h>
…		…
10	#include <sys/stat.h>	58	#include <sys/stat.h>
11	#include <limits.h>	59	#include <limits.h>
12	#include <fcntl.h>	60	#include <fcntl.h>
13	#include <assert.h>	61	#include <assert.h>
14		62
		63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
		64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
		65	#if HAVE_STDINT_H
		66	# include <stdint.h>
		67	#endif
		68
		69	#ifndef ECANCELED
		70	# define ECANCELED EDOM
		71	#endif
		72
		73	static void eio_destroy (eio_req *req);
		74
15	#ifndef EIO_FINISH	75	#ifndef EIO_FINISH
16	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	76	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17	#endif	77	#endif
18		78
19	#ifndef EIO_DESTROY	79	#ifndef EIO_DESTROY
…		…
22		82
23	#ifndef EIO_FEED	83	#ifndef EIO_FEED
24	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	84	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
25	#endif	85	#endif
26		86
		87	#ifndef EIO_FD_TO_WIN32_HANDLE
		88	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		89	#endif
		90	#ifndef EIO_WIN32_HANDLE_TO_FD
		91	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		92	#endif
		93
		94	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		95
		96	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		97
27	#ifdef _WIN32	98	#ifdef _WIN32
28		99
29	/*doh*/	100	#define PAGESIZE 4096 /* GetSystemInfo? */
		101
		102	#ifdef EIO_STRUCT_STATI64
		103	#define stat(path,buf) _stati64 (path,buf)
		104	#define fstat(fd,buf) _fstati64 (path,buf)
		105	#endif
		106	#define lstat(path,buf) stat (path,buf)
		107	#define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		108	#define mkdir(path,mode) _mkdir (path)
		109	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		110
		111	#define chown(path,uid,gid) EIO_ENOSYS ()
		112	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		113	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		114	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		115	#define mknod(path,mode,dev) EIO_ENOSYS ()
		116	#define sync() EIO_ENOSYS ()
		117
		118	/* we could even stat and see if it exists */
		119	static int
		120	symlink (const char *old, const char *neu)
		121	{
		122	if (CreateSymbolicLink (neu, old, 1))
		123	return 0;
		124
		125	if (CreateSymbolicLink (neu, old, 0))
		126	return 0;
		127
		128	return EIO_ERRNO (ENOENT, -1);
		129	}
30		130
31	#else	131	#else
32		132
33	# include "config.h"
34	# include <sys/time.h>	133	#include <sys/time.h>
35	# include <sys/select.h>	134	#include <sys/select.h>
		135	#include <sys/statvfs.h>
36	# include <unistd.h>	136	#include <unistd.h>
37	# include <utime.h>	137	#include <utime.h>
38	# include <signal.h>	138	#include <signal.h>
39	# include <dirent.h>	139	#include <dirent.h>
40		140
		141	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		142	#include <sys/mman.h>
		143	#endif
		144
		145	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		146	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		147	#define _DIRENT_HAVE_D_TYPE /* sigh */
		148	#define D_INO(de) (de)->d_fileno
		149	#define D_NAMLEN(de) (de)->d_namlen
		150	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		151	#define D_INO(de) (de)->d_ino
		152	#endif
		153
		154	#ifdef _D_EXACT_NAMLEN
		155	#undef D_NAMLEN
		156	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		157	#endif
		158
		159	#ifdef _DIRENT_HAVE_D_TYPE
		160	#define D_TYPE(de) (de)->d_type
		161	#endif
		162
41	# ifndef EIO_STRUCT_DIRENT	163	#ifndef EIO_STRUCT_DIRENT
42	# define EIO_STRUCT_DIRENT struct dirent	164	#define EIO_STRUCT_DIRENT struct dirent
43	# endif	165	#endif
44		166
45	#endif	167	#endif
46		168
47	#if HAVE_SENDFILE	169	#if HAVE_SENDFILE
48	# if __linux	170	# if __linux
49	# include <sys/sendfile.h>	171	# include <sys/sendfile.h>
50	# elif __freebsd	172	# elif __FreeBSD__ || defined __APPLE__
51	# include <sys/socket.h>	173	# include <sys/socket.h>
52	# include <sys/uio.h>	174	# include <sys/uio.h>
53	# elif __hpux	175	# elif __hpux
54	# include <sys/socket.h>	176	# include <sys/socket.h>
55	# elif __solaris /* not yet */	177	# elif __solaris
56	# include <sys/sendfile.h>	178	# include <sys/sendfile.h>
57	# else	179	# else
58	# error sendfile support requested but not available	180	# error sendfile support requested but not available
59	# endif	181	# endif
60	#endif	182	#endif
61		183
62	/* number of seconds after which an idle threads exit */	184	#ifndef D_TYPE
63	#define IDLE_TIMEOUT 10	185	# define D_TYPE(de) 0
		186	#endif
		187	#ifndef D_INO
		188	# define D_INO(de) 0
		189	#endif
		190	#ifndef D_NAMLEN
		191	# define D_NAMLEN(de) strlen ((de)->d_name)
		192	#endif
64		193
65	/* used for struct dirent, AIX doesn't provide it */	194	/* used for struct dirent, AIX doesn't provide it */
66	#ifndef NAME_MAX	195	#ifndef NAME_MAX
67	# define NAME_MAX 4096	196	# define NAME_MAX 4096
68	#endif	197	#endif
69		198
		199	/* used for readlink etc. */
		200	#ifndef PATH_MAX
		201	# define PATH_MAX 4096
		202	#endif
		203
70	/* buffer size for various temporary buffers */	204	/* buffer size for various temporary buffers */
71	#define EIO_BUFSIZE 65536	205	#define EIO_BUFSIZE 65536
72		206
73	#define dBUF \	207	#define dBUF \
74	char *eio_buf; \	208	char *eio_buf; \
75	X_LOCK (etplock); \	209	ETP_WORKER_LOCK (self); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	210	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (etplock); \	211	ETP_WORKER_UNLOCK (self); \
78	errno = ENOMEM; \	212	errno = ENOMEM; \
79	if (!eio_buf) \	213	if (!eio_buf) \
80	return -1;	214	return -1;
81		215
82	#define EIO_TICKS ((1000000 + 1023) >> 10)	216	#define EIO_TICKS ((1000000 + 1023) >> 10)
83		217
		218	#define ETP_PRI_MIN EIO_PRI_MIN
		219	#define ETP_PRI_MAX EIO_PRI_MAX
		220
		221	struct etp_worker;
		222
		223	#define ETP_REQ eio_req
		224	#define ETP_DESTROY(req) eio_destroy (req)
		225	static int eio_finish (eio_req *req);
		226	#define ETP_FINISH(req) eio_finish (req)
		227	static void eio_execute (struct etp_worker *self, eio_req *req);
		228	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		229
		230	#define ETP_WORKER_CLEAR(req) \
		231	if (wrk->dbuf) \
		232	{ \
		233	free (wrk->dbuf); \
		234	wrk->dbuf = 0; \
		235	} \
		236	\
		237	if (wrk->dirp) \
		238	{ \
		239	closedir (wrk->dirp); \
		240	wrk->dirp = 0; \
		241	}
		242
		243	#define ETP_WORKER_COMMON \
		244	void *dbuf; \
		245	DIR *dirp;
		246
84	/*****************************************************************************/	247	/*****************************************************************************/
85		248
		249	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		250
86	/* calculcate time difference in ~1/EIO_TICKS of a second */	251	/* calculate time difference in ~1/EIO_TICKS of a second */
		252	ecb_inline int
87	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	253	tvdiff (struct timeval *tv1, struct timeval *tv2)
88	{	254	{
89	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	255	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
90	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	256	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
91	}	257	}
92		258
93	static unsigned int started, idle, wanted = 4;	259	static unsigned int started, idle, wanted = 4;
94		260
95	typedef struct etp_pool
96	{
97	void (*want_poll_cb) (void);	261	static void (*want_poll_cb) (void);
98	void (*done_poll_cb) (void);	262	static void (*done_poll_cb) (void);
99		263
100	unsigned int max_poll_time;	264	static unsigned int max_poll_time; /* reslock */
101	unsigned int max_poll_reqs;	265	static unsigned int max_poll_reqs; /* reslock */
102	} etp_pool;
103		266
104	static volatile unsigned int nreqs, nready, npending;	267	static volatile unsigned int nreqs; /* reqlock */
105	static volatile unsigned int max_idle = 4;	268	static volatile unsigned int nready; /* reqlock */
		269	static volatile unsigned int npending; /* reqlock */
		270	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		271	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
106		272
		273	static xmutex_t wrklock;
		274	static xmutex_t reslock;
		275	static xmutex_t reqlock;
		276	static xcond_t reqwait;
		277
		278	#if !HAVE_PREADWRITE
		279	/*
		280	* make our pread/pwrite emulation safe against themselves, but not against
		281	* normal read/write by using a mutex. slows down execution a lot,
		282	* but that's your problem, not mine.
		283	*/
107	static mutex_t etplock = X_MUTEX_INIT;	284	static xmutex_t preadwritelock = X_MUTEX_INIT;
108	static mutex_t reslock = X_MUTEX_INIT;	285	#endif
109	static mutex_t reqlock = X_MUTEX_INIT;
110	static cond_t reqwait = X_COND_INIT;
111		286
112	typedef struct worker	287	typedef struct etp_worker
113	{	288	{
114	/* locked by etplock */	289	/* locked by wrklock */
115	struct worker *prev, *next;	290	struct etp_worker *prev, *next;
116		291
117	thread_t tid;	292	xthread_t tid;
118		293
119	/* locked by reslock, reqlock or etplock */	294	/* locked by reslock, reqlock or wrklock */
120	eio_req *req; /* currently processed request */	295	ETP_REQ *req; /* currently processed request */
121	void *dbuf;	296
122	DIR *dirp;	297	ETP_WORKER_COMMON
123	} worker;	298	} etp_worker;
124		299
125	static worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	300	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
		301
		302	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		303	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
126		304
127	/* worker threads management */	305	/* worker threads management */
128		306
		307	static void ecb_cold
129	static void worker_clear (worker *wrk)	308	etp_worker_clear (etp_worker *wrk)
130	{	309	{
131	if (wrk->dirp)	310	ETP_WORKER_CLEAR (wrk);
132	{
133	closedir (wrk->dirp);
134	wrk->dirp = 0;
135	}
136
137	if (wrk->dbuf)
138	{
139	free (wrk->dbuf);
140	wrk->dbuf = 0;
141	}
142	}	311	}
143		312
144	static void worker_free (worker *wrk)	313	static void ecb_cold
		314	etp_worker_free (etp_worker *wrk)
145	{	315	{
146	wrk->next->prev = wrk->prev;	316	wrk->next->prev = wrk->prev;
147	wrk->prev->next = wrk->next;	317	wrk->prev->next = wrk->next;
148		318
149	free (wrk);	319	free (wrk);
150	}	320	}
151		321
152	unsigned int eio_nreqs (void)	322	static unsigned int
		323	etp_nreqs (void)
153	{	324	{
		325	int retval;
		326	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		327	retval = nreqs;
		328	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154	return nreqs;	329	return retval;
155	}	330	}
156		331
157	unsigned int eio_nready (void)	332	static unsigned int
		333	etp_nready (void)
158	{	334	{
159	unsigned int retval;	335	unsigned int retval;
160		336
161	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	337	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162	retval = nready;	338	retval = nready;
163	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	339	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164		340
165	return retval;	341	return retval;
166	}	342	}
167		343
168	unsigned int eio_npending (void)	344	static unsigned int
		345	etp_npending (void)
169	{	346	{
170	unsigned int retval;	347	unsigned int retval;
171		348
172	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	349	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173	retval = npending;	350	retval = npending;
174	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	351	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175		352
176	return retval;	353	return retval;
177	}	354	}
178		355
179	unsigned int eio_nthreads (void)	356	static unsigned int
		357	etp_nthreads (void)
180	{	358	{
181	unsigned int retval;	359	unsigned int retval;
182		360
183	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	361	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184	retval = started;	362	retval = started;
…		…
191	* a somewhat faster data structure might be nice, but	369	* a somewhat faster data structure might be nice, but
192	* with 8 priorities this actually needs <20 insns	370	* with 8 priorities this actually needs <20 insns
193	* per shift, the most expensive operation.	371	* per shift, the most expensive operation.
194	*/	372	*/
195	typedef struct {	373	typedef struct {
196	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	374	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197	int size;	375	int size;
198	} reqq;	376	} etp_reqq;
199		377
200	static reqq req_queue;	378	static etp_reqq req_queue;
201	static reqq res_queue;	379	static etp_reqq res_queue;
202		380
203	static int reqq_push (reqq *q, eio_req *req)	381	static int ecb_noinline
		382	reqq_push (etp_reqq *q, ETP_REQ *req)
204	{	383	{
205	int pri = req->pri;	384	int pri = req->pri;
206	req->next = 0;	385	req->next = 0;
207		386
208	if (q->qe[pri])	387	if (q->qe[pri])
…		…
214	q->qe[pri] = q->qs[pri] = req;	393	q->qe[pri] = q->qs[pri] = req;
215		394
216	return q->size++;	395	return q->size++;
217	}	396	}
218		397
219	static eio_req *reqq_shift (reqq *q)	398	static ETP_REQ * ecb_noinline
		399	reqq_shift (etp_reqq *q)
220	{	400	{
221	int pri;	401	int pri;
222		402
223	if (!q->size)	403	if (!q->size)
224	return 0;	404	return 0;
225		405
226	--q->size;	406	--q->size;
227		407
228	for (pri = EIO_NUM_PRI; pri--; )	408	for (pri = ETP_NUM_PRI; pri--; )
229	{	409	{
230	eio_req *req = q->qs[pri];	410	eio_req *req = q->qs[pri];
231		411
232	if (req)	412	if (req)
233	{	413	{
…		…
239	}	419	}
240		420
241	abort ();	421	abort ();
242	}	422	}
243		423
244	static void etp_atfork_prepare (void)	424	static void ecb_cold
		425	etp_thread_init (void)
245	{	426	{
246	X_LOCK (etplock);
247	X_LOCK (reqlock);
248	X_LOCK (reslock);
249	#if !HAVE_PREADWRITE	427	#if !HAVE_PREADWRITE
250	X_LOCK (preadwritelock);	428	X_MUTEX_CREATE (preadwritelock);
251	#endif	429	#endif
252	#if !HAVE_READDIR_R	430	X_MUTEX_CREATE (wrklock);
253	X_LOCK (readdirlock);	431	X_MUTEX_CREATE (reslock);
254	#endif	432	X_MUTEX_CREATE (reqlock);
		433	X_COND_CREATE (reqwait);
255	}	434	}
256		435
		436	static void ecb_cold
		437	etp_atfork_prepare (void)
		438	{
		439	}
		440
		441	static void ecb_cold
257	static void etp_atfork_parent (void)	442	etp_atfork_parent (void)
258	{	443	{
259	#if !HAVE_READDIR_R
260	X_UNLOCK (readdirlock);
261	#endif
262	#if !HAVE_PREADWRITE
263	X_UNLOCK (preadwritelock);
264	#endif
265	X_UNLOCK (reslock);
266	X_UNLOCK (reqlock);
267	X_UNLOCK (etplock);
268	}	444	}
269		445
		446	static void ecb_cold
270	static void etp_atfork_child (void)	447	etp_atfork_child (void)
271	{	448	{
272	eio_req *prv;	449	ETP_REQ *prv;
273		450
274	while (prv = reqq_shift (&req_queue))	451	while ((prv = reqq_shift (&req_queue)))
275	eio_destroy (prv);	452	ETP_DESTROY (prv);
276		453
277	while (prv = reqq_shift (&res_queue))	454	while ((prv = reqq_shift (&res_queue)))
278	eio_destroy (prv);	455	ETP_DESTROY (prv);
279		456
280	while (wrk_first.next != &wrk_first)	457	while (wrk_first.next != &wrk_first)
281	{	458	{
282	worker *wrk = wrk_first.next;	459	etp_worker *wrk = wrk_first.next;
283		460
284	if (wrk->req)	461	if (wrk->req)
285	eio_destroy (wrk->req);	462	ETP_DESTROY (wrk->req);
286		463
287	worker_clear (wrk);	464	etp_worker_clear (wrk);
288	worker_free (wrk);	465	etp_worker_free (wrk);
289	}	466	}
290		467
291	started = 0;	468	started = 0;
292	idle = 0;	469	idle = 0;
293	nreqs = 0;	470	nreqs = 0;
294	nready = 0;	471	nready = 0;
295	npending = 0;	472	npending = 0;
296		473
297	etp_atfork_parent ();	474	etp_thread_init ();
298	}	475	}
299		476
300	static void	477	static void ecb_cold
301	etp_once_init (void)	478	etp_once_init (void)
302	{	479	{
		480	etp_thread_init ();
303	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);	481	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
304	}	482	}
305		483
306	static int	484	static int ecb_cold
307	etp_init (etp_pool *etp, void (*want_poll)(void), void (*done_poll)(void))	485	etp_init (void (*want_poll)(void), void (*done_poll)(void))
308	{	486	{
309	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	487	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
310		488
311	pthread_once (&doinit, etp_once_init);	489	pthread_once (&doinit, etp_once_init);
312		490
313	memset (etp, 0, sizeof *etp);
314
315	etp->want_poll_cb = want_poll;	491	want_poll_cb = want_poll;
316	etp->done_poll_cb = done_poll;	492	done_poll_cb = done_poll;
317	}
318		493
319	static etp_pool etp;
320
321	/*****************************************************************************/
322
323	static void grp_try_feed (eio_req *grp)
324	{
325	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
326	{
327	int old_len = grp->size;
328
329	EIO_FEED (grp);
330
331	/* stop if no progress has been made */
332	if (old_len == grp->size)
333	{
334	grp->feed = 0;
335	break;
336	}
337	}
338	}
339
340	static int eio_finish (eio_req *req);
341
342	static int grp_dec (eio_req *grp)
343	{
344	--grp->size;
345
346	/* call feeder, if applicable */
347	grp_try_feed (grp);
348
349	/* finish, if done */
350	if (!grp->size && grp->int1)
351	return eio_finish (grp);
352	else
353	return 0;	494	return 0;
354	}	495	}
355		496
356	void eio_destroy (eio_req *req)
357	{
358	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
359	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
360
361	EIO_DESTROY (req);
362	}
363
364	static int eio_finish (eio_req *req)
365	{
366	int res = EIO_FINISH (req);
367
368	if (req->grp)
369	{
370	int res2;
371	eio_req *grp = req->grp;
372
373	/* unlink request */
374	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
375	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
376
377	if (grp->grp_first == req)
378	grp->grp_first = req->grp_next;
379
380	res2 = grp_dec (grp);
381
382	if (!res && res2)
383	res = res2;
384	}
385
386	eio_destroy (req);
387
388	return res;
389	}
390
391	void eio_grp_cancel (eio_req *grp)
392	{
393	for (grp = grp->grp_first; grp; grp = grp->grp_next)
394	eio_cancel (grp);
395	}
396
397	void eio_cancel (eio_req *req)
398	{
399	X_LOCK (etplock);
400	req->flags |= EIO_FLAG_CANCELLED;
401	X_UNLOCK (etplock);
402
403	eio_grp_cancel (req);
404	}
405
406	X_THREAD_PROC (eio_proc);	497	X_THREAD_PROC (etp_proc);
407		498
408	static void start_thread (void)	499	static void ecb_cold
		500	etp_start_thread (void)
409	{	501	{
410	worker *wrk = calloc (1, sizeof (worker));	502	etp_worker *wrk = calloc (1, sizeof (etp_worker));
411		503
412	/*TODO*/	504	/*TODO*/
413	assert (("unable to allocate worker thread data", wrk));	505	assert (("unable to allocate worker thread data", wrk));
414		506
415	X_LOCK (etplock);	507	X_LOCK (wrklock);
416		508
417	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	509	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
418	{	510	{
419	wrk->prev = &wrk_first;	511	wrk->prev = &wrk_first;
420	wrk->next = wrk_first.next;	512	wrk->next = wrk_first.next;
421	wrk_first.next->prev = wrk;	513	wrk_first.next->prev = wrk;
422	wrk_first.next = wrk;	514	wrk_first.next = wrk;
423	++started;	515	++started;
424	}	516	}
425	else	517	else
426	free (wrk);	518	free (wrk);
427		519
428	X_UNLOCK (etplock);	520	X_UNLOCK (wrklock);
429	}	521	}
430		522
		523	static void
431	static void maybe_start_thread (void)	524	etp_maybe_start_thread (void)
432	{	525	{
433	if (eio_nthreads () >= wanted)	526	if (ecb_expect_true (etp_nthreads () >= wanted))
434	return;	527	return;
435		528
436	/* todo: maybe use idle here, but might be less exact */	529	/* todo: maybe use idle here, but might be less exact */
437	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	530	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
438	return;	531	return;
439		532
440	start_thread ();	533	etp_start_thread ();
441	}	534	}
442		535
443	void eio_submit (eio_req *req)	536	static void ecb_cold
444	{	537	etp_end_thread (void)
445	req->pri += EIO_PRI_BIAS;
446
447	if (req->pri < EIO_PRI_MIN + EIO_PRI_BIAS) req->pri = EIO_PRI_MIN + EIO_PRI_BIAS;
448	if (req->pri > EIO_PRI_MAX + EIO_PRI_BIAS) req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
449
450	++nreqs;
451
452	X_LOCK (reqlock);
453	++nready;
454	reqq_push (&req_queue, req);
455	X_COND_SIGNAL (reqwait);
456	X_UNLOCK (reqlock);
457
458	maybe_start_thread ();
459	}
460
461	static void end_thread (void)
462	{	538	{
463	eio_req *req = calloc (1, sizeof (eio_req));	539	eio_req *req = calloc (1, sizeof (eio_req));
464		540
465	req->type = EIO_QUIT;	541	req->type = -1;
466	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	542	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
467		543
468	X_LOCK (reqlock);	544	X_LOCK (reqlock);
469	reqq_push (&req_queue, req);	545	reqq_push (&req_queue, req);
470	X_COND_SIGNAL (reqwait);	546	X_COND_SIGNAL (reqwait);
471	X_UNLOCK (reqlock);	547	X_UNLOCK (reqlock);
472		548
473	X_LOCK (etplock);	549	X_LOCK (wrklock);
474	--started;	550	--started;
475	X_UNLOCK (etplock);	551	X_UNLOCK (wrklock);
476	}	552	}
477		553
478	void eio_set_max_poll_time (double nseconds)	554	static int
479	{
480	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
481	etp.max_poll_time = nseconds;
482	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
483	}
484
485	void eio_set_max_poll_reqs (unsigned int maxreqs)
486	{
487	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
488	etp.max_poll_reqs = maxreqs;
489	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
490	}
491
492	void eio_set_max_idle (unsigned int nthreads)
493	{
494	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
495	max_idle = nthreads <= 0 ? 1 : nthreads;
496	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
497	}
498
499	void eio_set_min_parallel (unsigned int nthreads)
500	{
501	if (wanted < nthreads)
502	wanted = nthreads;
503	}
504
505	void eio_set_max_parallel (unsigned int nthreads)
506	{
507	if (wanted > nthreads)
508	wanted = nthreads;
509
510	while (started > wanted)
511	end_thread ();
512	}
513
514	int eio_poll (void)	555	etp_poll (void)
515	{	556	{
516	int maxreqs = etp.max_poll_reqs;	557	unsigned int maxreqs;
		558	unsigned int maxtime;
517	struct timeval tv_start, tv_now;	559	struct timeval tv_start, tv_now;
518	eio_req *req;
519		560
520	if (etp.max_poll_time)	561	X_LOCK (reslock);
		562	maxreqs = max_poll_reqs;
		563	maxtime = max_poll_time;
		564	X_UNLOCK (reslock);
		565
		566	if (maxtime)
521	gettimeofday (&tv_start, 0);	567	gettimeofday (&tv_start, 0);
522		568
523	for (;;)	569	for (;;)
524	{	570	{
		571	ETP_REQ *req;
		572
525	maybe_start_thread ();	573	etp_maybe_start_thread ();
526		574
527	X_LOCK (reslock);	575	X_LOCK (reslock);
528	req = reqq_shift (&res_queue);	576	req = reqq_shift (&res_queue);
529		577
530	if (req)	578	if (req)
531	{	579	{
532	--npending;	580	--npending;
533		581
534	if (!res_queue.size && etp.done_poll_cb)	582	if (!res_queue.size && done_poll_cb)
535	etp.done_poll_cb ();	583	done_poll_cb ();
536	}	584	}
537		585
538	X_UNLOCK (reslock);	586	X_UNLOCK (reslock);
539		587
540	if (!req)	588	if (!req)
541	return 0;	589	return 0;
542		590
		591	X_LOCK (reqlock);
543	--nreqs;	592	--nreqs;
		593	X_UNLOCK (reqlock);
544		594
545	if (req->type == EIO_GROUP && req->size)	595	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
546	{	596	{
547	req->int1 = 1; /* mark request as delayed */	597	req->int1 = 1; /* mark request as delayed */
548	continue;	598	continue;
549	}	599	}
550	else	600	else
551	{	601	{
552	int res = eio_finish (req);	602	int res = ETP_FINISH (req);
553	if (res)	603	if (ecb_expect_false (res))
554	return res;	604	return res;
555	}	605	}
556		606
557	if (maxreqs && !--maxreqs)	607	if (ecb_expect_false (maxreqs && !--maxreqs))
558	break;	608	break;
559		609
560	if (etp.max_poll_time)	610	if (maxtime)
561	{	611	{
562	gettimeofday (&tv_now, 0);	612	gettimeofday (&tv_now, 0);
563		613
564	if (tvdiff (&tv_start, &tv_now) >= etp.max_poll_time)	614	if (tvdiff (&tv_start, &tv_now) >= maxtime)
565	break;	615	break;
566	}	616	}
567	}	617	}
568		618
569	errno = EAGAIN;	619	errno = EAGAIN;
570	return -1;	620	return -1;
571	}	621	}
572		622
		623	static void
		624	etp_cancel (ETP_REQ *req)
		625	{
		626	req->cancelled = 1;
		627
		628	eio_grp_cancel (req);
		629	}
		630
		631	static void
		632	etp_submit (ETP_REQ *req)
		633	{
		634	req->pri -= ETP_PRI_MIN;
		635
		636	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		637	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		638
		639	if (ecb_expect_false (req->type == EIO_GROUP))
		640	{
		641	/* I hope this is worth it :/ */
		642	X_LOCK (reqlock);
		643	++nreqs;
		644	X_UNLOCK (reqlock);
		645
		646	X_LOCK (reslock);
		647
		648	++npending;
		649
		650	if (!reqq_push (&res_queue, req) && want_poll_cb)
		651	want_poll_cb ();
		652
		653	X_UNLOCK (reslock);
		654	}
		655	else
		656	{
		657	X_LOCK (reqlock);
		658	++nreqs;
		659	++nready;
		660	reqq_push (&req_queue, req);
		661	X_COND_SIGNAL (reqwait);
		662	X_UNLOCK (reqlock);
		663
		664	etp_maybe_start_thread ();
		665	}
		666	}
		667
		668	static void ecb_cold
		669	etp_set_max_poll_time (double nseconds)
		670	{
		671	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		672	max_poll_time = nseconds * EIO_TICKS;
		673	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		674	}
		675
		676	static void ecb_cold
		677	etp_set_max_poll_reqs (unsigned int maxreqs)
		678	{
		679	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		680	max_poll_reqs = maxreqs;
		681	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		682	}
		683
		684	static void ecb_cold
		685	etp_set_max_idle (unsigned int nthreads)
		686	{
		687	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		688	max_idle = nthreads;
		689	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		690	}
		691
		692	static void ecb_cold
		693	etp_set_idle_timeout (unsigned int seconds)
		694	{
		695	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		696	idle_timeout = seconds;
		697	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		698	}
		699
		700	static void ecb_cold
		701	etp_set_min_parallel (unsigned int nthreads)
		702	{
		703	if (wanted < nthreads)
		704	wanted = nthreads;
		705	}
		706
		707	static void ecb_cold
		708	etp_set_max_parallel (unsigned int nthreads)
		709	{
		710	if (wanted > nthreads)
		711	wanted = nthreads;
		712
		713	while (started > wanted)
		714	etp_end_thread ();
		715	}
		716
		717	/*****************************************************************************/
		718
		719	static void
		720	grp_try_feed (eio_req *grp)
		721	{
		722	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		723	{
		724	grp->flags &= ~EIO_FLAG_GROUPADD;
		725
		726	EIO_FEED (grp);
		727
		728	/* stop if no progress has been made */
		729	if (!(grp->flags & EIO_FLAG_GROUPADD))
		730	{
		731	grp->feed = 0;
		732	break;
		733	}
		734	}
		735	}
		736
		737	static int
		738	grp_dec (eio_req *grp)
		739	{
		740	--grp->size;
		741
		742	/* call feeder, if applicable */
		743	grp_try_feed (grp);
		744
		745	/* finish, if done */
		746	if (!grp->size && grp->int1)
		747	return eio_finish (grp);
		748	else
		749	return 0;
		750	}
		751
		752	static void
		753	eio_destroy (eio_req *req)
		754	{
		755	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		756	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		757
		758	EIO_DESTROY (req);
		759	}
		760
		761	static int
		762	eio_finish (eio_req *req)
		763	{
		764	int res = EIO_FINISH (req);
		765
		766	if (req->grp)
		767	{
		768	int res2;
		769	eio_req *grp = req->grp;
		770
		771	/* unlink request */
		772	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		773	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		774
		775	if (grp->grp_first == req)
		776	grp->grp_first = req->grp_next;
		777
		778	res2 = grp_dec (grp);
		779
		780	if (!res)
		781	res = res2;
		782	}
		783
		784	eio_destroy (req);
		785
		786	return res;
		787	}
		788
		789	void
		790	eio_grp_cancel (eio_req *grp)
		791	{
		792	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		793	eio_cancel (grp);
		794	}
		795
		796	void
		797	eio_cancel (eio_req *req)
		798	{
		799	etp_cancel (req);
		800	}
		801
		802	void
		803	eio_submit (eio_req *req)
		804	{
		805	etp_submit (req);
		806	}
		807
		808	unsigned int
		809	eio_nreqs (void)
		810	{
		811	return etp_nreqs ();
		812	}
		813
		814	unsigned int
		815	eio_nready (void)
		816	{
		817	return etp_nready ();
		818	}
		819
		820	unsigned int
		821	eio_npending (void)
		822	{
		823	return etp_npending ();
		824	}
		825
		826	unsigned int ecb_cold
		827	eio_nthreads (void)
		828	{
		829	return etp_nthreads ();
		830	}
		831
		832	void ecb_cold
		833	eio_set_max_poll_time (double nseconds)
		834	{
		835	etp_set_max_poll_time (nseconds);
		836	}
		837
		838	void ecb_cold
		839	eio_set_max_poll_reqs (unsigned int maxreqs)
		840	{
		841	etp_set_max_poll_reqs (maxreqs);
		842	}
		843
		844	void ecb_cold
		845	eio_set_max_idle (unsigned int nthreads)
		846	{
		847	etp_set_max_idle (nthreads);
		848	}
		849
		850	void ecb_cold
		851	eio_set_idle_timeout (unsigned int seconds)
		852	{
		853	etp_set_idle_timeout (seconds);
		854	}
		855
		856	void ecb_cold
		857	eio_set_min_parallel (unsigned int nthreads)
		858	{
		859	etp_set_min_parallel (nthreads);
		860	}
		861
		862	void ecb_cold
		863	eio_set_max_parallel (unsigned int nthreads)
		864	{
		865	etp_set_max_parallel (nthreads);
		866	}
		867
		868	int eio_poll (void)
		869	{
		870	return etp_poll ();
		871	}
		872
573	/*****************************************************************************/	873	/*****************************************************************************/
574	/* work around various missing functions */	874	/* work around various missing functions */
575		875
576	#if !HAVE_PREADWRITE	876	#if !HAVE_PREADWRITE
		877	# undef pread
		878	# undef pwrite
577	# define pread eio__pread	879	# define pread eio__pread
578	# define pwrite eio__pwrite	880	# define pwrite eio__pwrite
579		881
580	/*
581	* make our pread/pwrite safe against themselves, but not against
582	* normal read/write by using a mutex. slows down execution a lot,
583	* but that's your problem, not mine.
584	*/
585	static mutex_t preadwritelock = X_MUTEX_INIT;
586
587	static ssize_t	882	static eio_ssize_t
588	eio__pread (int fd, void *buf, size_t count, off_t offset)	883	eio__pread (int fd, void *buf, size_t count, off_t offset)
589	{	884	{
590	ssize_t res;	885	eio_ssize_t res;
591	off_t ooffset;	886	off_t ooffset;
592		887
593	X_LOCK (preadwritelock);	888	X_LOCK (preadwritelock);
594	ooffset = lseek (fd, 0, SEEK_CUR);	889	ooffset = lseek (fd, 0, SEEK_CUR);
595	lseek (fd, offset, SEEK_SET);	890	lseek (fd, offset, SEEK_SET);
…		…
598	X_UNLOCK (preadwritelock);	893	X_UNLOCK (preadwritelock);
599		894
600	return res;	895	return res;
601	}	896	}
602		897
603	static ssize_t	898	static eio_ssize_t
604	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	899	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
605	{	900	{
606	ssize_t res;	901	eio_ssize_t res;
607	off_t ooffset;	902	off_t ooffset;
608		903
609	X_LOCK (preadwritelock);	904	X_LOCK (preadwritelock);
610	ooffset = lseek (fd, 0, SEEK_CUR);	905	ooffset = lseek (fd, 0, SEEK_CUR);
611	lseek (fd, offset, SEEK_SET);	906	lseek (fd, offset, SEEK_SET);
612	res = write (fd, buf, count);	907	res = write (fd, buf, count);
613	lseek (fd, offset, SEEK_SET);	908	lseek (fd, ooffset, SEEK_SET);
614	X_UNLOCK (preadwritelock);	909	X_UNLOCK (preadwritelock);
615		910
616	return res;	911	return res;
617	}	912	}
618	#endif	913	#endif
619		914
620	#ifndef HAVE_FUTIMES	915	#ifndef HAVE_UTIMES
621		916
		917	# undef utimes
622	# define utimes(path,times) eio__utimes (path, times)	918	# define utimes(path,times) eio__utimes (path, times)
623	# define futimes(fd,times) eio__futimes (fd, times)
624		919
625	static int	920	static int
626	eio__utimes (const char *filename, const struct timeval times[2])	921	eio__utimes (const char *filename, const struct timeval times[2])
627	{	922	{
628	if (times)	923	if (times)
…		…
636	}	931	}
637	else	932	else
638	return utime (filename, 0);	933	return utime (filename, 0);
639	}	934	}
640		935
		936	#endif
		937
		938	#ifndef HAVE_FUTIMES
		939
		940	# undef futimes
		941	# define futimes(fd,times) eio__futimes (fd, times)
		942
		943	static int
641	static int eio__futimes (int fd, const struct timeval tv[2])	944	eio__futimes (int fd, const struct timeval tv[2])
642	{	945	{
643	errno = ENOSYS;	946	errno = ENOSYS;
644	return -1;	947	return -1;
645	}	948	}
646		949
647	#endif	950	#endif
648		951
649	#if !HAVE_FDATASYNC	952	#if !HAVE_FDATASYNC
		953	# undef fdatasync
650	# define fdatasync fsync	954	# define fdatasync(fd) fsync (fd)
651	#endif	955	#endif
		956
		957	/* sync_file_range always needs emulation */
		958	static int
		959	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		960	{
		961	#if HAVE_SYNC_FILE_RANGE
		962	int res;
		963
		964	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		965	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		966	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		967	{
		968	flags = 0
		969	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		970	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		971	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		972	}
		973
		974	res = sync_file_range (fd, offset, nbytes, flags);
		975
		976	if (!res || errno != ENOSYS)
		977	return res;
		978	#endif
		979
		980	/* even though we could play tricks with the flags, it's better to always
		981	* call fdatasync, as that matches the expectation of its users best */
		982	return fdatasync (fd);
		983	}
		984
		985	static int
		986	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		987	{
		988	#if HAVE_FALLOCATE
		989	return fallocate (fd, mode, offset, len);
		990	#else
		991	errno = ENOSYS;
		992	return -1;
		993	#endif
		994	}
652		995
653	#if !HAVE_READAHEAD	996	#if !HAVE_READAHEAD
		997	# undef readahead
654	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	998	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
655		999
656	static ssize_t	1000	static eio_ssize_t
657	eio__readahead (int fd, off_t offset, size_t count, worker *self)	1001	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
658	{	1002	{
659	size_t todo = count;	1003	size_t todo = count;
660	dBUF;	1004	dBUF;
661		1005
662	while (todo > 0)	1006	while (todo > 0)
…		…
672	return count;	1016	return count;
673	}	1017	}
674		1018
675	#endif	1019	#endif
676		1020
677	#if !HAVE_READDIR_R
678	# define readdir_r eio__readdir_r
679
680	static mutex_t readdirlock = X_MUTEX_INIT;
681
682	static int
683	eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
684	{
685	EIO_STRUCT_DIRENT *e;
686	int errorno;
687
688	X_LOCK (readdirlock);
689
690	e = readdir (dirp);
691	errorno = errno;
692
693	if (e)
694	{
695	*res = ent;
696	strcpy (ent->d_name, e->d_name);
697	}
698	else
699	*res = 0;
700
701	X_UNLOCK (readdirlock);
702
703	errno = errorno;
704	return e ? 0 : -1;
705	}
706	#endif
707
708	/* sendfile always needs emulation */	1021	/* sendfile always needs emulation */
709	static ssize_t	1022	static eio_ssize_t
710	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self)	1023	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
711	{	1024	{
		1025	eio_ssize_t written = 0;
712	ssize_t res;	1026	eio_ssize_t res;
713		1027
714	if (!count)	1028	if (!count)
715	return 0;	1029	return 0;
716		1030
		1031	for (;;)
		1032	{
		1033	#ifdef __APPLE__
		1034	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1035	#endif
717	#if HAVE_SENDFILE	1036	#if HAVE_SENDFILE
718	# if __linux	1037	# if __linux
		1038	off_t soffset = offset;
719	res = sendfile (ofd, ifd, &offset, count);	1039	res = sendfile (ofd, ifd, &soffset, count);
720		1040
721	# elif __freebsd	1041	# elif __FreeBSD__
722	/*	1042	/*
723	* Of course, the freebsd sendfile is a dire hack with no thoughts	1043	* Of course, the freebsd sendfile is a dire hack with no thoughts
724	* wasted on making it similar to other I/O functions.	1044	* wasted on making it similar to other I/O functions.
725	*/	1045	*/
726	{
727	off_t sbytes;	1046	off_t sbytes;
728	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1047	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
729		1048
730	if (res < 0 && sbytes)	1049	#if 0 /* according to the manpage, this is correct, but broken behaviour */
731	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1050	/* freebsd' sendfile will return 0 on success */
		1051	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1052	/* not on e.g. EIO or EPIPE - sounds broken */
		1053	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
732	res = sbytes;	1054	res = sbytes;
733	}	1055	#endif
		1056
		1057	/* according to source inspection, this is correct, and useful behaviour */
		1058	if (sbytes)
		1059	res = sbytes;
		1060
		1061	# elif defined (__APPLE__)
		1062	off_t sbytes = count;
		1063	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1064
		1065	/* according to the manpage, sbytes is always valid */
		1066	if (sbytes)
		1067	res = sbytes;
734		1068
735	# elif __hpux	1069	# elif __hpux
736	res = sendfile (ofd, ifd, offset, count, 0, 0);	1070	res = sendfile (ofd, ifd, offset, count, 0, 0);
737		1071
738	# elif __solaris	1072	# elif __solaris
739	{
740	struct sendfilevec vec;	1073	struct sendfilevec vec;
741	size_t sbytes;	1074	size_t sbytes;
742		1075
743	vec.sfv_fd = ifd;	1076	vec.sfv_fd = ifd;
744	vec.sfv_flag = 0;	1077	vec.sfv_flag = 0;
745	vec.sfv_off = offset;	1078	vec.sfv_off = offset;
746	vec.sfv_len = count;	1079	vec.sfv_len = count;
747		1080
748	res = sendfilev (ofd, &vec, 1, &sbytes);	1081	res = sendfilev (ofd, &vec, 1, &sbytes);
749		1082
750	if (res < 0 && sbytes)	1083	if (res < 0 && sbytes)
751	res = sbytes;	1084	res = sbytes;
752	}
753		1085
754	# endif	1086	# endif
		1087
		1088	#elif defined (_WIN32) && 0
		1089	/* does not work, just for documentation of what would need to be done */
		1090	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1091	/* libeio guarantees that the file offset does not change, and windows */
		1092	/* has no way to get an independent handle to the same file description */
		1093	HANDLE h = TO_SOCKET (ifd);
		1094	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1095	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1096
755	#else	1097	#else
756	res = -1;	1098	res = -1;
757	errno = ENOSYS;	1099	errno = ENOSYS;
758	#endif	1100	#endif
759		1101
		1102	/* we assume sendfile can copy at least 128mb in one go */
		1103	if (res <= 128 * 1024 * 1024)
		1104	{
		1105	if (res > 0)
		1106	written += res;
		1107
		1108	if (written)
		1109	return written;
		1110
		1111	break;
		1112	}
		1113	else
		1114	{
		1115	/* if we requested more, then probably the kernel was lazy */
		1116	written += res;
		1117	offset += res;
		1118	count -= res;
		1119
		1120	if (!count)
		1121	return written;
		1122	}
		1123	}
		1124
760	if (res < 0	1125	if (res < 0
761	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1126	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1127	/* BSDs */
		1128	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1129	|| errno == ENOTSUP
		1130	#endif
		1131	|| errno == EOPNOTSUPP /* BSDs */
762	#if __solaris	1132	#if __solaris
763	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1133	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
764	#endif	1134	#endif
765	)	1135	)
766	)	1136	)
…		…
770		1140
771	res = 0;	1141	res = 0;
772		1142
773	while (count)	1143	while (count)
774	{	1144	{
775	ssize_t cnt;	1145	eio_ssize_t cnt;
776		1146
777	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1147	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
778		1148
779	if (cnt <= 0)	1149	if (cnt <= 0)
780	{	1150	{
…		…
797	}	1167	}
798		1168
799	return res;	1169	return res;
800	}	1170	}
801		1171
		1172	#ifdef PAGESIZE
		1173	# define eio_pagesize() PAGESIZE
		1174	#else
		1175	static intptr_t
		1176	eio_pagesize (void)
		1177	{
		1178	static intptr_t page;
		1179
		1180	if (!page)
		1181	page = sysconf (_SC_PAGESIZE);
		1182
		1183	return page;
		1184	}
		1185	#endif
		1186
		1187	static void
		1188	eio_page_align (void **addr, size_t *length)
		1189	{
		1190	intptr_t mask = eio_pagesize () - 1;
		1191
		1192	/* round down addr */
		1193	intptr_t adj = mask & (intptr_t)*addr;
		1194
		1195	*addr = (void *)((intptr_t)*addr - adj);
		1196	*length += adj;
		1197
		1198	/* round up length */
		1199	*length = (*length + mask) & ~mask;
		1200	}
		1201
		1202	#if !_POSIX_MEMLOCK
		1203	# define eio__mlockall(a) EIO_ENOSYS ()
		1204	#else
		1205
		1206	static int
		1207	eio__mlockall (int flags)
		1208	{
		1209	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1210	extern int mallopt (int, int);
		1211	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1212	#endif
		1213
		1214	if (EIO_MCL_CURRENT != MCL_CURRENT
		1215	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1216	{
		1217	flags = 0
		1218	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1219	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1220	}
		1221
		1222	return mlockall (flags);
		1223	}
		1224	#endif
		1225
		1226	#if !_POSIX_MEMLOCK_RANGE
		1227	# define eio__mlock(a,b) EIO_ENOSYS ()
		1228	#else
		1229
		1230	static int
		1231	eio__mlock (void *addr, size_t length)
		1232	{
		1233	eio_page_align (&addr, &length);
		1234
		1235	return mlock (addr, length);
		1236	}
		1237
		1238	#endif
		1239
		1240	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1241	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1242	#else
		1243
		1244	static int
		1245	eio__msync (void *mem, size_t len, int flags)
		1246	{
		1247	eio_page_align (&mem, &len);
		1248
		1249	if (EIO_MS_ASYNC != MS_SYNC
		1250	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1251	|| EIO_MS_SYNC != MS_SYNC)
		1252	{
		1253	flags = 0
		1254	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1255	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1256	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1257	}
		1258
		1259	return msync (mem, len, flags);
		1260	}
		1261
		1262	#endif
		1263
		1264	static int
		1265	eio__mtouch (eio_req *req)
		1266	{
		1267	void *mem = req->ptr2;
		1268	size_t len = req->size;
		1269	int flags = req->int1;
		1270
		1271	eio_page_align (&mem, &len);
		1272
		1273	{
		1274	intptr_t addr = (intptr_t)mem;
		1275	intptr_t end = addr + len;
		1276	intptr_t page = eio_pagesize ();
		1277
		1278	if (addr < end)
		1279	if (flags & EIO_MT_MODIFY) /* modify */
		1280	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1281	else
		1282	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1283	}
		1284
		1285	return 0;
		1286	}
		1287
		1288	/*****************************************************************************/
		1289	/* requests implemented outside eio_execute, because they are so large */
		1290
		1291	static void
		1292	eio__realpath (eio_req *req, etp_worker *self)
		1293	{
		1294	char *rel = req->ptr1;
		1295	char *res;
		1296	char *tmp1, *tmp2;
		1297	#if SYMLOOP_MAX > 32
		1298	int symlinks = SYMLOOP_MAX;
		1299	#else
		1300	int symlinks = 32;
		1301	#endif
		1302
		1303	req->result = -1;
		1304
		1305	errno = EINVAL;
		1306	if (!rel)
		1307	return;
		1308
		1309	errno = ENOENT;
		1310	if (!*rel)
		1311	return;
		1312
		1313	if (!req->ptr2)
		1314	{
		1315	X_LOCK (wrklock);
		1316	req->flags |= EIO_FLAG_PTR2_FREE;
		1317	X_UNLOCK (wrklock);
		1318	req->ptr2 = malloc (PATH_MAX * 3);
		1319
		1320	errno = ENOMEM;
		1321	if (!req->ptr2)
		1322	return;
		1323	}
		1324
		1325	res = req->ptr2;
		1326	tmp1 = res + PATH_MAX;
		1327	tmp2 = tmp1 + PATH_MAX;
		1328
		1329	#if 0 /* disabled, the musl way to do things is just too racy */
		1330	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1331	/* on linux we may be able to ask the kernel */
		1332	{
		1333	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1334
		1335	if (fd >= 0)
		1336	{
		1337	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1338	req->result = readlink (tmp1, res, PATH_MAX);
		1339	close (fd);
		1340
		1341	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1342
		1343	if (req->result > 0)
		1344	goto done;
		1345	}
		1346	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1347	return;
		1348	}
		1349	#endif
		1350	#endif
		1351
		1352	if (*rel != '/')
		1353	{
		1354	if (!getcwd (res, PATH_MAX))
		1355	return;
		1356
		1357	if (res [1]) /* only use if not / */
		1358	res += strlen (res);
		1359	}
		1360
		1361	while (*rel)
		1362	{
		1363	eio_ssize_t len, linklen;
		1364	char *beg = rel;
		1365
		1366	while (*rel && *rel != '/')
		1367	++rel;
		1368
		1369	len = rel - beg;
		1370
		1371	if (!len) /* skip slashes */
		1372	{
		1373	++rel;
		1374	continue;
		1375	}
		1376
		1377	if (beg [0] == '.')
		1378	{
		1379	if (len == 1)
		1380	continue; /* . - nop */
		1381
		1382	if (beg [1] == '.' && len == 2)
		1383	{
		1384	/* .. - back up one component, if possible */
		1385
		1386	while (res != req->ptr2)
		1387	if (*--res == '/')
		1388	break;
		1389
		1390	continue;
		1391	}
		1392	}
		1393
		1394	errno = ENAMETOOLONG;
		1395	if (res + 1 + len + 1 >= tmp1)
		1396	return;
		1397
		1398	/* copy one component */
		1399	*res = '/';
		1400	memcpy (res + 1, beg, len);
		1401
		1402	/* zero-terminate, for readlink */
		1403	res [len + 1] = 0;
		1404
		1405	/* now check if it's a symlink */
		1406	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1407
		1408	if (linklen < 0)
		1409	{
		1410	if (errno != EINVAL)
		1411	return;
		1412
		1413	/* it's a normal directory. hopefully */
		1414	res += len + 1;
		1415	}
		1416	else
		1417	{
		1418	/* yay, it was a symlink - build new path in tmp2 */
		1419	int rellen = strlen (rel);
		1420
		1421	errno = ENAMETOOLONG;
		1422	if (linklen + 1 + rellen >= PATH_MAX)
		1423	return;
		1424
		1425	errno = ELOOP;
		1426	if (!--symlinks)
		1427	return;
		1428
		1429	if (*tmp1 == '/')
		1430	res = req->ptr2; /* symlink resolves to an absolute path */
		1431
		1432	/* we need to be careful, as rel might point into tmp2 already */
		1433	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1434	tmp2 [linklen] = '/';
		1435	memcpy (tmp2, tmp1, linklen);
		1436
		1437	rel = tmp2;
		1438	}
		1439	}
		1440
		1441	/* special case for the lone root path */
		1442	if (res == req->ptr2)
		1443	*res++ = '/';
		1444
		1445	req->result = res - (char *)req->ptr2;
		1446
		1447	done:
		1448	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1449	}
		1450
		1451	static signed char
		1452	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1453	{
		1454	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1455	: a->inode < b->inode ? -1
		1456	: a->inode > b->inode ? 1
		1457	: 0;
		1458	}
		1459
		1460	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1461
		1462	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1463	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1464
		1465	static void
		1466	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1467	{
		1468	unsigned char bits [9 + sizeof (ino_t) * 8];
		1469	unsigned char *bit = bits;
		1470
		1471	assert (CHAR_BIT == 8);
		1472	assert (sizeof (eio_dirent) * 8 < 256);
		1473	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1474	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1475
		1476	if (size <= EIO_SORT_FAST)
		1477	return;
		1478
		1479	/* first prepare an array of bits to test in our radix sort */
		1480	/* try to take endianness into account, as well as differences in ino_t sizes */
		1481	/* inode_bits must contain all inodes ORed together */
		1482	/* which is used to skip bits that are 0 everywhere, which is very common */
		1483	{
		1484	ino_t endianness;
		1485	int i, j;
		1486
		1487	/* we store the byte offset of byte n into byte n of "endianness" */
		1488	for (i = 0; i < sizeof (ino_t); ++i)
		1489	((unsigned char *)&endianness)[i] = i;
		1490
		1491	*bit++ = 0;
		1492
		1493	for (i = 0; i < sizeof (ino_t); ++i)
		1494	{
		1495	/* shifting off the byte offsets out of "endianness" */
		1496	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1497	endianness >>= 8;
		1498
		1499	for (j = 0; j < 8; ++j)
		1500	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1501	*bit++ = offs + j;
		1502	}
		1503
		1504	for (j = 0; j < 8; ++j)
		1505	if (score_bits & (1 << j))
		1506	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1507	}
		1508
		1509	/* now actually do the sorting (a variant of MSD radix sort) */
		1510	{
		1511	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1512	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1513	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1514	int stk_idx = 0;
		1515
		1516	base_stk [stk_idx] = dents;
		1517	end_stk [stk_idx] = dents + size;
		1518	bit_stk [stk_idx] = bit - 1;
		1519
		1520	do
		1521	{
		1522	base = base_stk [stk_idx];
		1523	end = end_stk [stk_idx];
		1524	bit = bit_stk [stk_idx];
		1525
		1526	for (;;)
		1527	{
		1528	unsigned char O = *bit >> 3;
		1529	unsigned char M = 1 << (*bit & 7);
		1530
		1531	eio_dirent *a = base;
		1532	eio_dirent *b = end;
		1533
		1534	if (b - a < EIO_SORT_CUTOFF)
		1535	break;
		1536
		1537	/* now bit-partition the array on the bit */
		1538	/* this ugly asymmetric loop seems to perform much better than typical */
		1539	/* partition algos found in the literature */
		1540	do
		1541	if (!(((unsigned char *)a)[O] & M))
		1542	++a;
		1543	else if (!(((unsigned char *)--b)[O] & M))
		1544	{
		1545	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1546	++a;
		1547	}
		1548	while (b > a);
		1549
		1550	/* next bit, or stop, if no bits left in this path */
		1551	if (!*--bit)
		1552	break;
		1553
		1554	base_stk [stk_idx] = a;
		1555	end_stk [stk_idx] = end;
		1556	bit_stk [stk_idx] = bit;
		1557	++stk_idx;
		1558
		1559	end = a;
		1560	}
		1561	}
		1562	while (stk_idx--);
		1563	}
		1564	}
		1565
		1566	static void
		1567	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1568	{
		1569	/* first move the smallest element to the front, to act as a sentinel */
		1570	{
		1571	int i;
		1572	eio_dirent *min = dents;
		1573
		1574	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1575	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1576	if (EIO_DENT_CMP (dents [i], <, *min))
		1577	min = &dents [i];
		1578
		1579	/* swap elements 0 and j (minimum) */
		1580	{
		1581	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1582	}
		1583	}
		1584
		1585	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1586	{
		1587	eio_dirent *i, *j;
		1588
		1589	for (i = dents + 1; i < dents + size; ++i)
		1590	{
		1591	eio_dirent value = *i;
		1592
		1593	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1594	j [1] = j [0];
		1595
		1596	j [1] = value;
		1597	}
		1598	}
		1599	}
		1600
		1601	static void
		1602	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1603	{
		1604	if (size <= 1)
		1605	return; /* our insertion sort relies on size > 0 */
		1606
		1607	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1608	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1609	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1610
		1611	/* use an insertion sort at the end, or for small arrays, */
		1612	/* as insertion sort is more efficient for small partitions */
		1613	eio_dent_insertion_sort (dents, size);
		1614	}
		1615
802	/* read a full directory */	1616	/* read a full directory */
803	static void	1617	static void
804	eio__scandir (eio_req *req, worker *self)	1618	eio__scandir (eio_req *req, etp_worker *self)
805	{	1619	{
806	DIR *dirp;	1620	DIR *dirp;
807	union
808	{
809	EIO_STRUCT_DIRENT d;
810	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
811	} *u;
812	EIO_STRUCT_DIRENT *entp;	1621	EIO_STRUCT_DIRENT *entp;
813	char *name, *names;	1622	char *name, *names;
814	int memlen = 4096;	1623	int namesalloc = 4096;
815	int memofs = 0;	1624	int namesoffs = 0;
		1625	int flags = req->int1;
		1626	eio_dirent *dents = 0;
		1627	int dentalloc = 128;
816	int res = 0;	1628	int dentoffs = 0;
		1629	ino_t inode_bits = 0;
817		1630
		1631	req->result = -1;
		1632
		1633	if (!(flags & EIO_READDIR_DENTS))
		1634	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
		1635
818	X_LOCK (etplock);	1636	X_LOCK (wrklock);
		1637	/* the corresponding closedir is in ETP_WORKER_CLEAR */
819	self->dirp = dirp = opendir (req->ptr1);	1638	self->dirp = dirp = opendir (req->ptr1);
820	self->dbuf = u = malloc (sizeof (*u));	1639
821	req->flags |= EIO_FLAG_PTR2_FREE;	1640	if (req->flags & EIO_FLAG_PTR1_FREE)
		1641	free (req->ptr1);
		1642
		1643	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1644	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
822	req->ptr2 = names = malloc (memlen);	1645	req->ptr2 = names = malloc (namesalloc);
823	X_UNLOCK (etplock);	1646	X_UNLOCK (wrklock);
824		1647
825	if (dirp && u && names)	1648	if (dirp && names && (!flags || dents))
826	for (;;)	1649	for (;;)
827	{	1650	{
828	errno = 0;	1651	errno = 0;
829	readdir_r (dirp, &u->d, &entp);	1652	entp = readdir (dirp);
830		1653
831	if (!entp)	1654	if (!entp)
		1655	{
		1656	if (errno)
		1657	break;
		1658
		1659	/* sort etc. */
		1660	req->int1 = flags;
		1661	req->result = dentoffs;
		1662
		1663	if (flags & EIO_READDIR_STAT_ORDER)
		1664	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1665	else if (flags & EIO_READDIR_DIRS_FIRST)
		1666	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1667	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1668	else
		1669	{
		1670	/* in this case, all is known, and we just put dirs first and sort them */
		1671	eio_dirent *oth = dents + dentoffs;
		1672	eio_dirent *dir = dents;
		1673
		1674	/* now partition dirs to the front, and non-dirs to the back */
		1675	/* by walking from both sides and swapping if necessary */
		1676	while (oth > dir)
		1677	{
		1678	if (dir->type == EIO_DT_DIR)
		1679	++dir;
		1680	else if ((--oth)->type == EIO_DT_DIR)
		1681	{
		1682	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1683
		1684	++dir;
		1685	}
		1686	}
		1687
		1688	/* now sort the dirs only (dirs all have the same score) */
		1689	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1690	}
		1691
832	break;	1692	break;
		1693	}
833		1694
		1695	/* now add the entry to our list(s) */
834	name = entp->d_name;	1696	name = entp->d_name;
835		1697
		1698	/* skip . and .. entries */
836	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1699	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
837	{	1700	{
838	int len = strlen (name) + 1;	1701	int len = D_NAMLEN (entp) + 1;
839		1702
840	res++;	1703	while (ecb_expect_false (namesoffs + len > namesalloc))
841
842	while (memofs + len > memlen)
843	{	1704	{
844	memlen *= 2;	1705	namesalloc *= 2;
845	X_LOCK (etplock);	1706	X_LOCK (wrklock);
846	req->ptr2 = names = realloc (names, memlen);	1707	req->ptr2 = names = realloc (names, namesalloc);
847	X_UNLOCK (etplock);	1708	X_UNLOCK (wrklock);
848		1709
849	if (!names)	1710	if (!names)
850	break;	1711	break;
851	}	1712	}
852		1713
853	memcpy (names + memofs, name, len);	1714	memcpy (names + namesoffs, name, len);
		1715
		1716	if (dents)
		1717	{
		1718	struct eio_dirent *ent;
		1719
		1720	if (ecb_expect_false (dentoffs == dentalloc))
		1721	{
		1722	dentalloc *= 2;
		1723	X_LOCK (wrklock);
		1724	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1725	X_UNLOCK (wrklock);
		1726
		1727	if (!dents)
		1728	break;
		1729	}
		1730
		1731	ent = dents + dentoffs;
		1732
		1733	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1734	ent->namelen = len - 1;
		1735	ent->inode = D_INO (entp);
		1736
		1737	inode_bits |= ent->inode;
		1738
		1739	switch (D_TYPE (entp))
		1740	{
		1741	default:
		1742	ent->type = EIO_DT_UNKNOWN;
		1743	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1744	break;
		1745
		1746	#ifdef DT_FIFO
		1747	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1748	#endif
		1749	#ifdef DT_CHR
		1750	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1751	#endif
		1752	#ifdef DT_MPC
		1753	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1754	#endif
		1755	#ifdef DT_DIR
		1756	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1757	#endif
		1758	#ifdef DT_NAM
		1759	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1760	#endif
		1761	#ifdef DT_BLK
		1762	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1763	#endif
		1764	#ifdef DT_MPB
		1765	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1766	#endif
		1767	#ifdef DT_REG
		1768	case DT_REG: ent->type = EIO_DT_REG; break;
		1769	#endif
		1770	#ifdef DT_NWK
		1771	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1772	#endif
		1773	#ifdef DT_CMP
		1774	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1775	#endif
		1776	#ifdef DT_LNK
		1777	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1778	#endif
		1779	#ifdef DT_SOCK
		1780	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1781	#endif
		1782	#ifdef DT_DOOR
		1783	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1784	#endif
		1785	#ifdef DT_WHT
		1786	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1787	#endif
		1788	}
		1789
		1790	ent->score = 7;
		1791
		1792	if (flags & EIO_READDIR_DIRS_FIRST)
		1793	{
		1794	if (ent->type == EIO_DT_UNKNOWN)
		1795	{
		1796	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1797	ent->score = 1;
		1798	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1799	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1800	}
		1801	else if (ent->type == EIO_DT_DIR)
		1802	ent->score = 0;
		1803	}
		1804	}
		1805
854	memofs += len;	1806	namesoffs += len;
		1807	++dentoffs;
		1808	}
		1809
		1810	if (EIO_CANCELLED (req))
		1811	{
		1812	errno = ECANCELED;
		1813	break;
855	}	1814	}
856	}	1815	}
857
858	if (errno)
859	res = -1;
860
861	req->result = res;
862	}	1816	}
863		1817
864	/*****************************************************************************/	1818	/*****************************************************************************/
865		1819
866	#define ALLOC(len) \	1820	#define ALLOC(len) \
867	if (!req->ptr2) \	1821	if (!req->ptr2) \
868	{ \	1822	{ \
869	X_LOCK (etplock); \	1823	X_LOCK (wrklock); \
870	req->flags |= EIO_FLAG_PTR2_FREE; \	1824	req->flags |= EIO_FLAG_PTR2_FREE; \
871	X_UNLOCK (etplock); \	1825	X_UNLOCK (wrklock); \
872	req->ptr2 = malloc (len); \	1826	req->ptr2 = malloc (len); \
873	if (!req->ptr2) \	1827	if (!req->ptr2) \
874	{ \	1828	{ \
875	errno = ENOMEM; \	1829	errno = ENOMEM; \
876	req->result = -1; \	1830	req->result = -1; \
877	break; \	1831	break; \
878	} \	1832	} \
879	}	1833	}
880		1834
881	X_THREAD_PROC (eio_proc)	1835	X_THREAD_PROC (etp_proc)
882	{	1836	{
883	eio_req *req;	1837	ETP_REQ *req;
884	struct timespec ts;	1838	struct timespec ts;
885	worker *self = (worker *)thr_arg;	1839	etp_worker *self = (etp_worker *)thr_arg;
		1840	int timeout;
886		1841
887	/* try to distribute timeouts somewhat randomly */	1842	/* try to distribute timeouts somewhat evenly */
888	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1843	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
889		1844
890	for (;;)	1845	for (;;)
891	{	1846	{
892	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1847	ts.tv_sec = 0;
893		1848
894	X_LOCK (reqlock);	1849	X_LOCK (reqlock);
895		1850
896	for (;;)	1851	for (;;)
897	{	1852	{
898	self->req = req = reqq_shift (&req_queue);	1853	self->req = req = reqq_shift (&req_queue);
899		1854
900	if (req)	1855	if (req)
901	break;	1856	break;
902		1857
		1858	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1859	{
		1860	X_UNLOCK (reqlock);
		1861	X_LOCK (wrklock);
		1862	--started;
		1863	X_UNLOCK (wrklock);
		1864	goto quit;
		1865	}
		1866
903	++idle;	1867	++idle;
904		1868
905	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1869	if (idle <= max_idle)
		1870	/* we are allowed to idle, so do so without any timeout */
		1871	X_COND_WAIT (reqwait, reqlock);
		1872	else
906	{	1873	{
907	if (idle > max_idle)	1874	/* initialise timeout once */
908	{	1875	if (!ts.tv_sec)
909	--idle;	1876	ts.tv_sec = time (0) + idle_timeout;
910	X_UNLOCK (reqlock);
911	X_LOCK (etplock);
912	--started;
913	X_UNLOCK (etplock);
914	goto quit;
915	}
916		1877
917	/* we are allowed to idle, so do so without any timeout */
918	X_COND_WAIT (reqwait, reqlock);	1878	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
919	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1879	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
920	}	1880	}
921		1881
922	--idle;	1882	--idle;
923	}	1883	}
924		1884
925	--nready;	1885	--nready;
926		1886
927	X_UNLOCK (reqlock);	1887	X_UNLOCK (reqlock);
928		1888
929	errno = 0; /* strictly unnecessary */
930
931	if (!EIO_CANCELLED (req))
932	switch (req->type)	1889	if (req->type < 0)
933	{
934	case EIO_READ: ALLOC (req->size);
935	req->result = req->offs >= 0
936	? pread (req->int1, req->ptr2, req->size, req->offs)
937	: read (req->int1, req->ptr2, req->size); break;
938	case EIO_WRITE: req->result = req->offs >= 0
939	? pwrite (req->int1, req->ptr2, req->size, req->offs)
940	: write (req->int1, req->ptr2, req->size); break;
941
942	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
943	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
944
945	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
946	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
947	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
948	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
949	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
950	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
951
952	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
953	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
954	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
955	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
956	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
957	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
958
959	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
960	case EIO_CLOSE: req->result = close (req->int1); break;
961	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
962	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
963	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
964	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
965	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
966	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
967	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
968	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
969
970	case EIO_READLINK: ALLOC (NAME_MAX);
971	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
972
973	case EIO_SYNC: req->result = 0; sync (); break;
974	case EIO_FSYNC: req->result = fsync (req->int1); break;
975	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
976
977	case EIO_READDIR: eio__scandir (req, self); break;
978
979	case EIO_BUSY:
980	#ifdef _WIN32
981	Sleep (req->nv1 * 1000.);
982	#else
983	{
984	struct timeval tv;
985
986	tv.tv_sec = req->nv1;
987	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
988
989	req->result = select (0, 0, 0, 0, &tv);
990	}
991	#endif
992	break;
993
994	case EIO_UTIME:
995	case EIO_FUTIME:
996	{
997	struct timeval tv[2];
998	struct timeval *times;
999
1000	if (req->nv1 != -1. || req->nv2 != -1.)
1001	{
1002	tv[0].tv_sec = req->nv1;
1003	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1004	tv[1].tv_sec = req->nv2;
1005	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1006
1007	times = tv;
1008	}
1009	else
1010	times = 0;
1011
1012
1013	req->result = req->type == EIO_FUTIME
1014	? futimes (req->int1, times)
1015	: utimes (req->ptr1, times);
1016	}
1017
1018	case EIO_GROUP:
1019	case EIO_NOP:
1020	req->result = 0;
1021	break;
1022
1023	case EIO_QUIT:
1024	goto quit;	1890	goto quit;
1025		1891
1026	default:	1892	ETP_EXECUTE (self, req);
1027	req->result = -1;
1028	break;
1029	}
1030
1031	req->errorno = errno;
1032		1893
1033	X_LOCK (reslock);	1894	X_LOCK (reslock);
1034		1895
1035	++npending;	1896	++npending;
1036		1897
1037	if (!reqq_push (&res_queue, req) && etp.want_poll_cb)	1898	if (!reqq_push (&res_queue, req) && want_poll_cb)
1038	etp.want_poll_cb ();	1899	want_poll_cb ();
1039		1900
1040	self->req = 0;	1901	self->req = 0;
1041	worker_clear (self);	1902	etp_worker_clear (self);
1042		1903
1043	X_UNLOCK (reslock);	1904	X_UNLOCK (reslock);
1044	}	1905	}
1045		1906
1046	quit:	1907	quit:
1047	X_LOCK (etplock);	1908	X_LOCK (wrklock);
1048	worker_free (self);	1909	etp_worker_free (self);
1049	X_UNLOCK (etplock);	1910	X_UNLOCK (wrklock);
1050		1911
1051	return 0;	1912	return 0;
1052	}	1913	}
1053		1914
1054	/*****************************************************************************/	1915	/*****************************************************************************/
1055		1916
		1917	int ecb_cold
1056	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1918	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1057	{	1919	{
1058	etp_init (&etp, want_poll, done_poll);	1920	return etp_init (want_poll, done_poll);
1059	}	1921	}
1060		1922
		1923	ecb_inline void
1061	static void eio_api_destroy (eio_req *req)	1924	eio_api_destroy (eio_req *req)
1062	{	1925	{
1063	free (req);	1926	free (req);
1064	}	1927	}
1065		1928
1066	#define REQ(rtype) \	1929	#define REQ(rtype) \
…		…
1085	{ \	1948	{ \
1086	eio_api_destroy (req); \	1949	eio_api_destroy (req); \
1087	return 0; \	1950	return 0; \
1088	}	1951	}
1089		1952
		1953	static void
		1954	eio_execute (etp_worker *self, eio_req *req)
		1955	{
		1956	if (ecb_expect_false (EIO_CANCELLED (req)))
		1957	{
		1958	req->result = -1;
		1959	req->errorno = ECANCELED;
		1960	return;
		1961	}
		1962
		1963	switch (req->type)
		1964	{
		1965	case EIO_READ: ALLOC (req->size);
		1966	req->result = req->offs >= 0
		1967	? pread (req->int1, req->ptr2, req->size, req->offs)
		1968	: read (req->int1, req->ptr2, req->size); break;
		1969	case EIO_WRITE: req->result = req->offs >= 0
		1970	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1971	: write (req->int1, req->ptr2, req->size); break;
		1972
		1973	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		1974	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
		1975
		1976	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1977	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1978	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1979	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1980	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1981	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1982
		1983	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1984	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1985	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1986	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1987
		1988	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		1989	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1990	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		1991	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1992	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		1993	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1994
		1995	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		1996	case EIO_CLOSE: req->result = close (req->int1); break;
		1997	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1998	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		1999	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		2000	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		2001	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		2002	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		2003	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		2004	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2005
		2006	case EIO_REALPATH: eio__realpath (req, self); break;
		2007
		2008	case EIO_READLINK: ALLOC (PATH_MAX);
		2009	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
		2010
		2011	case EIO_SYNC: req->result = 0; sync (); break;
		2012	case EIO_FSYNC: req->result = fsync (req->int1); break;
		2013	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2014	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2015	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		2016	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2017	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		2018	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2019	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
		2020
		2021	case EIO_READDIR: eio__scandir (req, self); break;
		2022
		2023	case EIO_BUSY:
		2024	#ifdef _WIN32
		2025	Sleep (req->nv1 * 1e3);
		2026	#else
		2027	{
		2028	struct timeval tv;
		2029
		2030	tv.tv_sec = req->nv1;
		2031	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
		2032
		2033	req->result = select (0, 0, 0, 0, &tv);
		2034	}
		2035	#endif
		2036	break;
		2037
		2038	case EIO_UTIME:
		2039	case EIO_FUTIME:
		2040	{
		2041	struct timeval tv[2];
		2042	struct timeval *times;
		2043
		2044	if (req->nv1 != -1. || req->nv2 != -1.)
		2045	{
		2046	tv[0].tv_sec = req->nv1;
		2047	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		2048	tv[1].tv_sec = req->nv2;
		2049	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		2050
		2051	times = tv;
		2052	}
		2053	else
		2054	times = 0;
		2055
		2056	req->result = req->type == EIO_FUTIME
		2057	? futimes (req->int1, times)
		2058	: utimes (req->ptr1, times);
		2059	}
		2060	break;
		2061
		2062	case EIO_GROUP:
		2063	abort (); /* handled in eio_request */
		2064
		2065	case EIO_NOP:
		2066	req->result = 0;
		2067	break;
		2068
		2069	case EIO_CUSTOM:
		2070	req->feed (req);
		2071	break;
		2072
		2073	default:
		2074	errno = ENOSYS;
		2075	req->result = -1;
		2076	break;
		2077	}
		2078
		2079	req->errorno = errno;
		2080	}
		2081
1090	#ifndef EIO_NO_WRAPPERS	2082	#ifndef EIO_NO_WRAPPERS
1091		2083
1092	eio_req *eio_nop (int pri, eio_cb cb, void *data)	2084	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1093	{	2085	{
1094	REQ (EIO_NOP); SEND;	2086	REQ (EIO_NOP); SEND;
…		…
1107	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	2099	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1108	{	2100	{
1109	REQ (EIO_FSYNC); req->int1 = fd; SEND;	2101	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1110	}	2102	}
1111		2103
		2104	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2105	{
		2106	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2107	}
		2108
		2109	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2110	{
		2111	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2112	}
		2113
		2114	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2115	{
		2116	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2117	}
		2118
		2119	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2120	{
		2121	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2122	}
		2123
		2124	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2125	{
		2126	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2127	}
		2128
		2129	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2130	{
		2131	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2132	}
		2133
1112	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2134	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1113	{	2135	{
1114	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2136	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1115	}	2137	}
1116		2138
…		…
1137	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2159	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1138	{	2160	{
1139	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2161	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1140	}	2162	}
1141		2163
		2164	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2165	{
		2166	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2167	}
		2168
1142	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2169	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1143	{	2170	{
1144	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2171	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1145	}	2172	}
1146		2173
…		…
1152	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2179	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1153	{	2180	{
1154	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2181	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1155	}	2182	}
1156		2183
1157	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2184	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1158	{	2185	{
1159	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2186	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1160	}	2187	}
1161		2188
1162	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2189	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1182	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2209	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1183	{	2210	{
1184	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2211	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1185	}	2212	}
1186		2213
1187	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2214	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1188	{	2215	{
1189	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2216	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1190	}	2217	}
1191		2218
1192	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2219	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1208	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2235	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1209	{	2236	{
1210	return eio__1path (EIO_READLINK, path, pri, cb, data);	2237	return eio__1path (EIO_READLINK, path, pri, cb, data);
1211	}	2238	}
1212		2239
		2240	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2241	{
		2242	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2243	}
		2244
1213	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2245	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1214	{	2246	{
1215	return eio__1path (EIO_STAT, path, pri, cb, data);	2247	return eio__1path (EIO_STAT, path, pri, cb, data);
1216	}	2248	}
1217		2249
1218	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2250	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1219	{	2251	{
1220	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2252	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1221	}	2253	}
1222		2254
		2255	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2256	{
		2257	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2258	}
		2259
1223	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2260	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1224	{	2261	{
1225	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2262	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1226	}	2263	}
1227		2264
1228	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	2265	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1229	{	2266	{
1230	return eio__1path (EIO_RMDIR, path, pri, cb, data);	2267	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1231	}	2268	}
1232		2269
1233	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	2270	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1234	{	2271	{
1235	return eio__1path (EIO_READDIR, path, pri, cb, data);	2272	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1236	}	2273	}
1237		2274
1238	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2275	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1239	{	2276	{
1240	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	2277	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1241	}	2278	}
1242		2279
1243	static eio_req *	2280	static eio_req *
1244	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2281	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1245	{	2282	{
…		…
1269	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2306	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1270	{	2307	{
1271	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2308	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1272	}	2309	}
1273		2310
		2311	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
		2312	{
		2313	REQ (EIO_CUSTOM); req->feed = execute; SEND;
		2314	}
		2315
1274	#endif	2316	#endif
1275		2317
1276	eio_req *eio_grp (eio_cb cb, void *data)	2318	eio_req *eio_grp (eio_cb cb, void *data)
1277	{	2319	{
1278	const int pri = EIO_PRI_MAX;	2320	const int pri = EIO_PRI_MAX;
…		…
1285	#undef SEND	2327	#undef SEND
1286		2328
1287	/*****************************************************************************/	2329	/*****************************************************************************/
1288	/* grp functions */	2330	/* grp functions */
1289		2331
		2332	void
1290	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2333	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1291	{	2334	{
1292	grp->int2 = limit;	2335	grp->int2 = limit;
1293	grp->feed = feed;	2336	grp->feed = feed;
1294		2337
1295	grp_try_feed (grp);	2338	grp_try_feed (grp);
1296	}	2339	}
1297		2340
		2341	void
1298	void eio_grp_limit (eio_req *grp, int limit)	2342	eio_grp_limit (eio_req *grp, int limit)
1299	{	2343	{
1300	grp->int2 = limit;	2344	grp->int2 = limit;
1301		2345
1302	grp_try_feed (grp);	2346	grp_try_feed (grp);
1303	}	2347	}
1304		2348
		2349	void
1305	void eio_grp_add (eio_req *grp, eio_req *req)	2350	eio_grp_add (eio_req *grp, eio_req *req)
1306	{	2351	{
1307	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2352	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
		2353
		2354	grp->flags |= EIO_FLAG_GROUPADD;
1308		2355
1309	++grp->size;	2356	++grp->size;
1310	req->grp = grp;	2357	req->grp = grp;
1311		2358
1312	req->grp_prev = 0;	2359	req->grp_prev = 0;
…		…
1319	}	2366	}
1320		2367
1321	/*****************************************************************************/	2368	/*****************************************************************************/
1322	/* misc garbage */	2369	/* misc garbage */
1323		2370
		2371	eio_ssize_t
1324	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2372	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1325	{	2373	{
1326	worker wrk;	2374	etp_worker wrk;
		2375	eio_ssize_t ret;
1327		2376
1328	wrk.dbuf = 0;	2377	wrk.dbuf = 0;
1329		2378
1330	eio__sendfile (ofd, ifd, offset, count, &wrk);	2379	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1331		2380
1332	if (wrk.dbuf)	2381	if (wrk.dbuf)
1333	free (wrk.dbuf);	2382	free (wrk.dbuf);
1334	}
1335		2383
		2384	return ret;
		2385	}
		2386

Diff Legend

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