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 1.93 by root, Mon Jul 18 01:27:03 2011 UTC

		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
		5	* All rights reserved.
		6	*
		7	* Redistribution and use in source and binary forms, with or without modifica-
		8	* tion, are permitted provided that the following conditions are met:
		9	*
		10	* 1. Redistributions of source code must retain the above copyright notice,
		11	* this list of conditions and the following disclaimer.
		12	*
		13	* 2. Redistributions in binary form must reproduce the above copyright
		14	* notice, this list of conditions and the following disclaimer in the
		15	* documentation and/or other materials provided with the distribution.
		16	*
		17	* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
		18	* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
		19	* CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
		20	* EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
		21	* CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
		22	* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
		23	* OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
		24	* WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
		25	* ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
		26	* OF THE POSSIBILITY OF SUCH DAMAGE.
		27	*
		28	* Alternatively, the contents of this file may be used under the terms of
		29	* the GNU General Public License ("GPL") version 2 or any later version,
		30	* in which case the provisions of the GPL are applicable instead of
		31	* the above. If you wish to allow the use of your version of this file
		32	* only under the terms of the GPL and not to allow others to use your
		33	* version of this file under the BSD license, indicate your decision
		34	* by deleting the provisions above and replace them with the notice
		35	* and other provisions required by the GPL. If you do not delete the
		36	* provisions above, a recipient may use your version of this file under
		37	* either the BSD or the GPL.
		38	*/
		39
		40	#ifndef _WIN32
		41	# include "config.h"
		42	#endif
		43
1	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
		46
		47	#ifdef EIO_STACKSIZE
		48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
		49	#endif
2	#include "xthread.h"	50	#include "xthread.h"
3		51
4	#include <errno.h>	52	#include <errno.h>
5	#include <stddef.h>	53	#include <stddef.h>
6	#include <stdlib.h>	54	#include <stdlib.h>
…		…
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	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	static void eio_destroy (eio_req *req);
		77
15	#ifndef EIO_FINISH	78	#ifndef EIO_FINISH
16	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	79	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17	#endif	80	#endif
18		81
19	#ifndef EIO_DESTROY	82	#ifndef EIO_DESTROY
…		…
22		85
23	#ifndef EIO_FEED	86	#ifndef EIO_FEED
24	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	87	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
25	#endif	88	#endif
26		89
		90	#ifndef EIO_FD_TO_WIN32_HANDLE
		91	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		92	#endif
		93	#ifndef EIO_WIN32_HANDLE_TO_FD
		94	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		95	#endif
		96
		97	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		98
		99	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		100
27	#ifdef _WIN32	101	#ifdef _WIN32
28		102
29	/*doh*/	103	#undef PAGESIZE
		104	#define PAGESIZE 4096 /* GetSystemInfo? */
		105
		106	#ifdef EIO_STRUCT_STATI64
		107	#define stat(path,buf) _stati64 (path,buf)
		108	#define fstat(fd,buf) _fstati64 (path,buf)
		109	#endif
		110	#define lstat(path,buf) stat (path,buf)
		111	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		112	#define mkdir(path,mode) _mkdir (path)
		113	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		114
		115	#define chmod(path,mode) _chmod (path, mode)
		116	#define dup(fd) _dup (fd)
		117	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		118
		119	#define fchmod(fd,mode) EIO_ENOSYS ()
		120	#define chown(path,uid,gid) EIO_ENOSYS ()
		121	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		122	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		123	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		124	#define mknod(path,mode,dev) EIO_ENOSYS ()
		125	#define sync() EIO_ENOSYS ()
		126	#define readlink(path,buf,s) EIO_ENOSYS ()
		127	#define statvfs(path,buf) EIO_ENOSYS ()
		128	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		129
		130	/* we could even stat and see if it exists */
		131	static int
		132	symlink (const char *old, const char *neu)
		133	{
		134	#if WINVER >= 0x0600
		135	if (CreateSymbolicLink (neu, old, 1))
		136	return 0;
		137
		138	if (CreateSymbolicLink (neu, old, 0))
		139	return 0;
		140	#endif
		141
		142	return EIO_ERRNO (ENOENT, -1);
		143	}
		144
		145	/* POSIX API only */
		146	#define CreateHardLink(neu,old,flags) 0
		147	#define CreateSymbolicLink(neu,old,flags) 0
		148
		149	struct statvfs
		150	{
		151	int dummy;
		152	};
		153
		154	#define DT_DIR EIO_DT_DIR
		155	#define DT_REG EIO_DT_REG
		156	#define D_NAME(entp) entp.cFileName
		157	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
30		158
31	#else	159	#else
32		160
33	# include "config.h"
34	# include <sys/time.h>	161	#include <sys/time.h>
35	# include <sys/select.h>	162	#include <sys/select.h>
		163	#include <sys/statvfs.h>
36	# include <unistd.h>	164	#include <unistd.h>
37	# include <utime.h>	165	#include <utime.h>
38	# include <signal.h>	166	#include <signal.h>
39	# include <dirent.h>	167	#include <dirent.h>
40		168
		169	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		170	#include <sys/mman.h>
		171	#endif
		172
		173	#define D_NAME(entp) entp->d_name
		174
		175	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		176	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		177	#define _DIRENT_HAVE_D_TYPE /* sigh */
		178	#define D_INO(de) (de)->d_fileno
		179	#define D_NAMLEN(de) (de)->d_namlen
		180	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		181	#define D_INO(de) (de)->d_ino
		182	#endif
		183
		184	#ifdef _D_EXACT_NAMLEN
		185	#undef D_NAMLEN
		186	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		187	#endif
		188
		189	#ifdef _DIRENT_HAVE_D_TYPE
		190	#define D_TYPE(de) (de)->d_type
		191	#endif
		192
41	# ifndef EIO_STRUCT_DIRENT	193	#ifndef EIO_STRUCT_DIRENT
42	# define EIO_STRUCT_DIRENT struct dirent	194	#define EIO_STRUCT_DIRENT struct dirent
43	# endif	195	#endif
44		196
45	#endif	197	#endif
46		198
47	#if HAVE_SENDFILE	199	#if HAVE_SENDFILE
48	# if __linux	200	# if __linux
49	# include <sys/sendfile.h>	201	# include <sys/sendfile.h>
50	# elif __freebsd	202	# elif __FreeBSD__ || defined __APPLE__
51	# include <sys/socket.h>	203	# include <sys/socket.h>
52	# include <sys/uio.h>	204	# include <sys/uio.h>
53	# elif __hpux	205	# elif __hpux
54	# include <sys/socket.h>	206	# include <sys/socket.h>
55	# elif __solaris /* not yet */	207	# elif __solaris
56	# include <sys/sendfile.h>	208	# include <sys/sendfile.h>
57	# else	209	# else
58	# error sendfile support requested but not available	210	# error sendfile support requested but not available
59	# endif	211	# endif
60	#endif	212	#endif
61		213
62	/* number of seconds after which an idle threads exit */	214	#ifndef D_TYPE
63	#define IDLE_TIMEOUT 10	215	# define D_TYPE(de) 0
		216	#endif
		217	#ifndef D_INO
		218	# define D_INO(de) 0
		219	#endif
		220	#ifndef D_NAMLEN
		221	# define D_NAMLEN(entp) strlen (D_NAME (entp))
		222	#endif
64		223
65	/* used for struct dirent, AIX doesn't provide it */	224	/* used for struct dirent, AIX doesn't provide it */
66	#ifndef NAME_MAX	225	#ifndef NAME_MAX
67	# define NAME_MAX 4096	226	# define NAME_MAX 4096
68	#endif	227	#endif
69		228
		229	/* used for readlink etc. */
		230	#ifndef PATH_MAX
		231	# define PATH_MAX 4096
		232	#endif
		233
70	/* buffer size for various temporary buffers */	234	/* buffer size for various temporary buffers */
71	#define EIO_BUFSIZE 65536	235	#define EIO_BUFSIZE 65536
72		236
73	#define dBUF \	237	#define dBUF \
74	char *eio_buf; \	238	char *eio_buf; \
75	X_LOCK (etplock); \	239	ETP_WORKER_LOCK (self); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	240	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (etplock); \	241	ETP_WORKER_UNLOCK (self); \
78	errno = ENOMEM; \	242	errno = ENOMEM; \
79	if (!eio_buf) \	243	if (!eio_buf) \
80	return -1;	244	return -1;
81		245
82	#define EIO_TICKS ((1000000 + 1023) >> 10)	246	#define EIO_TICKS ((1000000 + 1023) >> 10)
83		247
		248	#define ETP_PRI_MIN EIO_PRI_MIN
		249	#define ETP_PRI_MAX EIO_PRI_MAX
		250
		251	struct etp_worker;
		252
		253	#define ETP_REQ eio_req
		254	#define ETP_DESTROY(req) eio_destroy (req)
		255	static int eio_finish (eio_req *req);
		256	#define ETP_FINISH(req) eio_finish (req)
		257	static void eio_execute (struct etp_worker *self, eio_req *req);
		258	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		259
		260	#define ETP_WORKER_CLEAR(req) \
		261	if (wrk->dbuf) \
		262	{ \
		263	free (wrk->dbuf); \
		264	wrk->dbuf = 0; \
		265	}
		266
		267	#define ETP_WORKER_COMMON \
		268	void *dbuf;
		269
84	/*****************************************************************************/	270	/*****************************************************************************/
85		271
		272	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		273
86	/* calculcate time difference in ~1/EIO_TICKS of a second */	274	/* calculate time difference in ~1/EIO_TICKS of a second */
		275	ecb_inline int
87	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	276	tvdiff (struct timeval *tv1, struct timeval *tv2)
88	{	277	{
89	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	278	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
90	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	279	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
91	}	280	}
92		281
93	static unsigned int started, idle, wanted = 4;	282	static unsigned int started, idle, wanted = 4;
94		283
95	typedef struct etp_pool
96	{
97	void (*want_poll_cb) (void);	284	static void (*want_poll_cb) (void);
98	void (*done_poll_cb) (void);	285	static void (*done_poll_cb) (void);
99		286
100	unsigned int max_poll_time;	287	static unsigned int max_poll_time; /* reslock */
101	unsigned int max_poll_reqs;	288	static unsigned int max_poll_reqs; /* reslock */
102	} etp_pool;
103		289
104	static volatile unsigned int nreqs, nready, npending;	290	static volatile unsigned int nreqs; /* reqlock */
105	static volatile unsigned int max_idle = 4;	291	static volatile unsigned int nready; /* reqlock */
		292	static volatile unsigned int npending; /* reqlock */
		293	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		294	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
106		295
107	static mutex_t etplock = X_MUTEX_INIT;	296	static xmutex_t wrklock;
108	static mutex_t reslock = X_MUTEX_INIT;	297	static xmutex_t reslock;
109	static mutex_t reqlock = X_MUTEX_INIT;	298	static xmutex_t reqlock;
110	static cond_t reqwait = X_COND_INIT;	299	static xcond_t reqwait;
111		300
		301	#if !HAVE_PREADWRITE
		302	/*
		303	* make our pread/pwrite emulation safe against themselves, but not against
		304	* normal read/write by using a mutex. slows down execution a lot,
		305	* but that's your problem, not mine.
		306	*/
		307	static xmutex_t preadwritelock;
		308	#endif
		309
112	typedef struct worker	310	typedef struct etp_worker
113	{	311	{
114	/* locked by etplock */	312	/* locked by wrklock */
115	struct worker *prev, *next;	313	struct etp_worker *prev, *next;
116		314
117	thread_t tid;	315	xthread_t tid;
118		316
119	/* locked by reslock, reqlock or etplock */	317	/* locked by reslock, reqlock or wrklock */
120	eio_req *req; /* currently processed request */	318	ETP_REQ *req; /* currently processed request */
121	void *dbuf;	319
122	DIR *dirp;	320	ETP_WORKER_COMMON
123	} worker;	321	} etp_worker;
124		322
125	static worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	323	static etp_worker wrk_first; /* NOT etp */
		324
		325	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		326	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
126		327
127	/* worker threads management */	328	/* worker threads management */
128		329
		330	static void ecb_cold
129	static void worker_clear (worker *wrk)	331	etp_worker_clear (etp_worker *wrk)
130	{	332	{
131	if (wrk->dirp)	333	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	}	334	}
143		335
144	static void worker_free (worker *wrk)	336	static void ecb_cold
		337	etp_worker_free (etp_worker *wrk)
145	{	338	{
146	wrk->next->prev = wrk->prev;	339	wrk->next->prev = wrk->prev;
147	wrk->prev->next = wrk->next;	340	wrk->prev->next = wrk->next;
148		341
149	free (wrk);	342	free (wrk);
150	}	343	}
151		344
152	unsigned int eio_nreqs (void)	345	static unsigned int
		346	etp_nreqs (void)
153	{	347	{
		348	int retval;
		349	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		350	retval = nreqs;
		351	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154	return nreqs;	352	return retval;
155	}	353	}
156		354
157	unsigned int eio_nready (void)	355	static unsigned int
		356	etp_nready (void)
158	{	357	{
159	unsigned int retval;	358	unsigned int retval;
160		359
161	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	360	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162	retval = nready;	361	retval = nready;
163	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	362	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164		363
165	return retval;	364	return retval;
166	}	365	}
167		366
168	unsigned int eio_npending (void)	367	static unsigned int
		368	etp_npending (void)
169	{	369	{
170	unsigned int retval;	370	unsigned int retval;
171		371
172	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	372	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173	retval = npending;	373	retval = npending;
174	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	374	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175		375
176	return retval;	376	return retval;
177	}	377	}
178		378
179	unsigned int eio_nthreads (void)	379	static unsigned int
		380	etp_nthreads (void)
180	{	381	{
181	unsigned int retval;	382	unsigned int retval;
182		383
183	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	384	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184	retval = started;	385	retval = started;
…		…
191	* a somewhat faster data structure might be nice, but	392	* a somewhat faster data structure might be nice, but
192	* with 8 priorities this actually needs <20 insns	393	* with 8 priorities this actually needs <20 insns
193	* per shift, the most expensive operation.	394	* per shift, the most expensive operation.
194	*/	395	*/
195	typedef struct {	396	typedef struct {
196	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	397	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197	int size;	398	int size;
198	} reqq;	399	} etp_reqq;
199		400
200	static reqq req_queue;	401	static etp_reqq req_queue;
201	static reqq res_queue;	402	static etp_reqq res_queue;
202		403
203	static int reqq_push (reqq *q, eio_req *req)	404	static void ecb_noinline ecb_cold
		405	reqq_init (etp_reqq *q)
		406	{
		407	int pri;
		408
		409	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		410	q->qs[pri] = q->qe[pri] = 0;
		411
		412	q->size = 0;
		413	}
		414
		415	static int ecb_noinline
		416	reqq_push (etp_reqq *q, ETP_REQ *req)
204	{	417	{
205	int pri = req->pri;	418	int pri = req->pri;
206	req->next = 0;	419	req->next = 0;
207		420
208	if (q->qe[pri])	421	if (q->qe[pri])
…		…
214	q->qe[pri] = q->qs[pri] = req;	427	q->qe[pri] = q->qs[pri] = req;
215		428
216	return q->size++;	429	return q->size++;
217	}	430	}
218		431
219	static eio_req *reqq_shift (reqq *q)	432	static ETP_REQ * ecb_noinline
		433	reqq_shift (etp_reqq *q)
220	{	434	{
221	int pri;	435	int pri;
222		436
223	if (!q->size)	437	if (!q->size)
224	return 0;	438	return 0;
225		439
226	--q->size;	440	--q->size;
227		441
228	for (pri = EIO_NUM_PRI; pri--; )	442	for (pri = ETP_NUM_PRI; pri--; )
229	{	443	{
230	eio_req *req = q->qs[pri];	444	eio_req *req = q->qs[pri];
231		445
232	if (req)	446	if (req)
233	{	447	{
…		…
239	}	453	}
240		454
241	abort ();	455	abort ();
242	}	456	}
243		457
244	static void etp_atfork_prepare (void)	458	static int ecb_cold
		459	etp_init (void (*want_poll)(void), void (*done_poll)(void))
245	{	460	{
246	X_LOCK (etplock);	461	X_MUTEX_CREATE (wrklock);
247	X_LOCK (reqlock);	462	X_MUTEX_CREATE (reslock);
248	X_LOCK (reslock);	463	X_MUTEX_CREATE (reqlock);
249	#if !HAVE_PREADWRITE	464	X_COND_CREATE (reqwait);
250	X_LOCK (preadwritelock);
251	#endif
252	#if !HAVE_READDIR_R
253	X_LOCK (readdirlock);
254	#endif
255	}
256		465
257	static void etp_atfork_parent (void)	466	reqq_init (&req_queue);
258	{	467	reqq_init (&res_queue);
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	}
269		468
270	static void etp_atfork_child (void)	469	wrk_first.next =
271	{	470	wrk_first.prev = &wrk_first;
272	eio_req *prv;
273
274	while (prv = reqq_shift (&req_queue))
275	eio_destroy (prv);
276
277	while (prv = reqq_shift (&res_queue))
278	eio_destroy (prv);
279
280	while (wrk_first.next != &wrk_first)
281	{
282	worker *wrk = wrk_first.next;
283
284	if (wrk->req)
285	eio_destroy (wrk->req);
286
287	worker_clear (wrk);
288	worker_free (wrk);
289	}
290		471
291	started = 0;	472	started = 0;
292	idle = 0;	473	idle = 0;
293	nreqs = 0;	474	nreqs = 0;
294	nready = 0;	475	nready = 0;
295	npending = 0;	476	npending = 0;
296		477
297	etp_atfork_parent ();
298	}
299
300	static void
301	etp_once_init (void)
302	{
303	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
304	}
305
306	static int
307	etp_init (etp_pool *etp, void (*want_poll)(void), void (*done_poll)(void))
308	{
309	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
310
311	pthread_once (&doinit, etp_once_init);
312
313	memset (etp, 0, sizeof *etp);
314
315	etp->want_poll_cb = want_poll;	478	want_poll_cb = want_poll;
316	etp->done_poll_cb = done_poll;	479	done_poll_cb = done_poll;
317	}
318		480
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;	481	return 0;
354	}	482	}
355		483
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);	484	X_THREAD_PROC (etp_proc);
407		485
408	static void start_thread (void)	486	static void ecb_cold
		487	etp_start_thread (void)
409	{	488	{
410	worker *wrk = calloc (1, sizeof (worker));	489	etp_worker *wrk = calloc (1, sizeof (etp_worker));
411		490
412	/*TODO*/	491	/*TODO*/
413	assert (("unable to allocate worker thread data", wrk));	492	assert (("unable to allocate worker thread data", wrk));
414		493
415	X_LOCK (etplock);	494	X_LOCK (wrklock);
416		495
417	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	496	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
418	{	497	{
419	wrk->prev = &wrk_first;	498	wrk->prev = &wrk_first;
420	wrk->next = wrk_first.next;	499	wrk->next = wrk_first.next;
421	wrk_first.next->prev = wrk;	500	wrk_first.next->prev = wrk;
422	wrk_first.next = wrk;	501	wrk_first.next = wrk;
423	++started;	502	++started;
424	}	503	}
425	else	504	else
426	free (wrk);	505	free (wrk);
427		506
428	X_UNLOCK (etplock);	507	X_UNLOCK (wrklock);
429	}	508	}
430		509
		510	static void
431	static void maybe_start_thread (void)	511	etp_maybe_start_thread (void)
432	{	512	{
433	if (eio_nthreads () >= wanted)	513	if (ecb_expect_true (etp_nthreads () >= wanted))
434	return;	514	return;
435		515
436	/* todo: maybe use idle here, but might be less exact */	516	/* todo: maybe use idle here, but might be less exact */
437	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	517	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
438	return;	518	return;
439		519
440	start_thread ();	520	etp_start_thread ();
441	}	521	}
442		522
443	void eio_submit (eio_req *req)	523	static void ecb_cold
444	{	524	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	{	525	{
463	eio_req *req = calloc (1, sizeof (eio_req));	526	eio_req *req = calloc (1, sizeof (eio_req));
464		527
465	req->type = EIO_QUIT;	528	req->type = -1;
466	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	529	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
467		530
468	X_LOCK (reqlock);	531	X_LOCK (reqlock);
469	reqq_push (&req_queue, req);	532	reqq_push (&req_queue, req);
470	X_COND_SIGNAL (reqwait);	533	X_COND_SIGNAL (reqwait);
471	X_UNLOCK (reqlock);	534	X_UNLOCK (reqlock);
472		535
473	X_LOCK (etplock);	536	X_LOCK (wrklock);
474	--started;	537	--started;
475	X_UNLOCK (etplock);	538	X_UNLOCK (wrklock);
476	}	539	}
477		540
478	void eio_set_max_poll_time (double nseconds)	541	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)	542	etp_poll (void)
515	{	543	{
516	int maxreqs = etp.max_poll_reqs;	544	unsigned int maxreqs;
		545	unsigned int maxtime;
517	struct timeval tv_start, tv_now;	546	struct timeval tv_start, tv_now;
518	eio_req *req;
519		547
520	if (etp.max_poll_time)	548	X_LOCK (reslock);
		549	maxreqs = max_poll_reqs;
		550	maxtime = max_poll_time;
		551	X_UNLOCK (reslock);
		552
		553	if (maxtime)
521	gettimeofday (&tv_start, 0);	554	gettimeofday (&tv_start, 0);
522		555
523	for (;;)	556	for (;;)
524	{	557	{
		558	ETP_REQ *req;
		559
525	maybe_start_thread ();	560	etp_maybe_start_thread ();
526		561
527	X_LOCK (reslock);	562	X_LOCK (reslock);
528	req = reqq_shift (&res_queue);	563	req = reqq_shift (&res_queue);
529		564
530	if (req)	565	if (req)
531	{	566	{
532	--npending;	567	--npending;
533		568
534	if (!res_queue.size && etp.done_poll_cb)	569	if (!res_queue.size && done_poll_cb)
535	etp.done_poll_cb ();	570	done_poll_cb ();
536	}	571	}
537		572
538	X_UNLOCK (reslock);	573	X_UNLOCK (reslock);
539		574
540	if (!req)	575	if (!req)
541	return 0;	576	return 0;
542		577
		578	X_LOCK (reqlock);
543	--nreqs;	579	--nreqs;
		580	X_UNLOCK (reqlock);
544		581
545	if (req->type == EIO_GROUP && req->size)	582	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
546	{	583	{
547	req->int1 = 1; /* mark request as delayed */	584	req->int1 = 1; /* mark request as delayed */
548	continue;	585	continue;
549	}	586	}
550	else	587	else
551	{	588	{
552	int res = eio_finish (req);	589	int res = ETP_FINISH (req);
553	if (res)	590	if (ecb_expect_false (res))
554	return res;	591	return res;
555	}	592	}
556		593
557	if (maxreqs && !--maxreqs)	594	if (ecb_expect_false (maxreqs && !--maxreqs))
558	break;	595	break;
559		596
560	if (etp.max_poll_time)	597	if (maxtime)
561	{	598	{
562	gettimeofday (&tv_now, 0);	599	gettimeofday (&tv_now, 0);
563		600
564	if (tvdiff (&tv_start, &tv_now) >= etp.max_poll_time)	601	if (tvdiff (&tv_start, &tv_now) >= maxtime)
565	break;	602	break;
566	}	603	}
567	}	604	}
568		605
569	errno = EAGAIN;	606	errno = EAGAIN;
570	return -1;	607	return -1;
571	}	608	}
572		609
		610	static void
		611	etp_cancel (ETP_REQ *req)
		612	{
		613	req->cancelled = 1;
		614
		615	eio_grp_cancel (req);
		616	}
		617
		618	static void
		619	etp_submit (ETP_REQ *req)
		620	{
		621	req->pri -= ETP_PRI_MIN;
		622
		623	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		624	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		625
		626	if (ecb_expect_false (req->type == EIO_GROUP))
		627	{
		628	/* I hope this is worth it :/ */
		629	X_LOCK (reqlock);
		630	++nreqs;
		631	X_UNLOCK (reqlock);
		632
		633	X_LOCK (reslock);
		634
		635	++npending;
		636
		637	if (!reqq_push (&res_queue, req) && want_poll_cb)
		638	want_poll_cb ();
		639
		640	X_UNLOCK (reslock);
		641	}
		642	else
		643	{
		644	X_LOCK (reqlock);
		645	++nreqs;
		646	++nready;
		647	reqq_push (&req_queue, req);
		648	X_COND_SIGNAL (reqwait);
		649	X_UNLOCK (reqlock);
		650
		651	etp_maybe_start_thread ();
		652	}
		653	}
		654
		655	static void ecb_cold
		656	etp_set_max_poll_time (double nseconds)
		657	{
		658	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		659	max_poll_time = nseconds * EIO_TICKS;
		660	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		661	}
		662
		663	static void ecb_cold
		664	etp_set_max_poll_reqs (unsigned int maxreqs)
		665	{
		666	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		667	max_poll_reqs = maxreqs;
		668	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		669	}
		670
		671	static void ecb_cold
		672	etp_set_max_idle (unsigned int nthreads)
		673	{
		674	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		675	max_idle = nthreads;
		676	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		677	}
		678
		679	static void ecb_cold
		680	etp_set_idle_timeout (unsigned int seconds)
		681	{
		682	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		683	idle_timeout = seconds;
		684	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		685	}
		686
		687	static void ecb_cold
		688	etp_set_min_parallel (unsigned int nthreads)
		689	{
		690	if (wanted < nthreads)
		691	wanted = nthreads;
		692	}
		693
		694	static void ecb_cold
		695	etp_set_max_parallel (unsigned int nthreads)
		696	{
		697	if (wanted > nthreads)
		698	wanted = nthreads;
		699
		700	while (started > wanted)
		701	etp_end_thread ();
		702	}
		703
		704	/*****************************************************************************/
		705
		706	static void
		707	grp_try_feed (eio_req *grp)
		708	{
		709	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		710	{
		711	grp->flags &= ~EIO_FLAG_GROUPADD;
		712
		713	EIO_FEED (grp);
		714
		715	/* stop if no progress has been made */
		716	if (!(grp->flags & EIO_FLAG_GROUPADD))
		717	{
		718	grp->feed = 0;
		719	break;
		720	}
		721	}
		722	}
		723
		724	static int
		725	grp_dec (eio_req *grp)
		726	{
		727	--grp->size;
		728
		729	/* call feeder, if applicable */
		730	grp_try_feed (grp);
		731
		732	/* finish, if done */
		733	if (!grp->size && grp->int1)
		734	return eio_finish (grp);
		735	else
		736	return 0;
		737	}
		738
		739	static void
		740	eio_destroy (eio_req *req)
		741	{
		742	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		743	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		744
		745	EIO_DESTROY (req);
		746	}
		747
		748	static int
		749	eio_finish (eio_req *req)
		750	{
		751	int res = EIO_FINISH (req);
		752
		753	if (req->grp)
		754	{
		755	int res2;
		756	eio_req *grp = req->grp;
		757
		758	/* unlink request */
		759	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		760	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		761
		762	if (grp->grp_first == req)
		763	grp->grp_first = req->grp_next;
		764
		765	res2 = grp_dec (grp);
		766
		767	if (!res)
		768	res = res2;
		769	}
		770
		771	eio_destroy (req);
		772
		773	return res;
		774	}
		775
		776	void
		777	eio_grp_cancel (eio_req *grp)
		778	{
		779	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		780	eio_cancel (grp);
		781	}
		782
		783	void
		784	eio_cancel (eio_req *req)
		785	{
		786	etp_cancel (req);
		787	}
		788
		789	void
		790	eio_submit (eio_req *req)
		791	{
		792	etp_submit (req);
		793	}
		794
		795	unsigned int
		796	eio_nreqs (void)
		797	{
		798	return etp_nreqs ();
		799	}
		800
		801	unsigned int
		802	eio_nready (void)
		803	{
		804	return etp_nready ();
		805	}
		806
		807	unsigned int
		808	eio_npending (void)
		809	{
		810	return etp_npending ();
		811	}
		812
		813	unsigned int ecb_cold
		814	eio_nthreads (void)
		815	{
		816	return etp_nthreads ();
		817	}
		818
		819	void ecb_cold
		820	eio_set_max_poll_time (double nseconds)
		821	{
		822	etp_set_max_poll_time (nseconds);
		823	}
		824
		825	void ecb_cold
		826	eio_set_max_poll_reqs (unsigned int maxreqs)
		827	{
		828	etp_set_max_poll_reqs (maxreqs);
		829	}
		830
		831	void ecb_cold
		832	eio_set_max_idle (unsigned int nthreads)
		833	{
		834	etp_set_max_idle (nthreads);
		835	}
		836
		837	void ecb_cold
		838	eio_set_idle_timeout (unsigned int seconds)
		839	{
		840	etp_set_idle_timeout (seconds);
		841	}
		842
		843	void ecb_cold
		844	eio_set_min_parallel (unsigned int nthreads)
		845	{
		846	etp_set_min_parallel (nthreads);
		847	}
		848
		849	void ecb_cold
		850	eio_set_max_parallel (unsigned int nthreads)
		851	{
		852	etp_set_max_parallel (nthreads);
		853	}
		854
		855	int eio_poll (void)
		856	{
		857	return etp_poll ();
		858	}
		859
573	/*****************************************************************************/	860	/*****************************************************************************/
574	/* work around various missing functions */	861	/* work around various missing functions */
575		862
576	#if !HAVE_PREADWRITE	863	#if !HAVE_PREADWRITE
		864	# undef pread
		865	# undef pwrite
577	# define pread eio__pread	866	# define pread eio__pread
578	# define pwrite eio__pwrite	867	# define pwrite eio__pwrite
579		868
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	869	static eio_ssize_t
588	eio__pread (int fd, void *buf, size_t count, off_t offset)	870	eio__pread (int fd, void *buf, size_t count, off_t offset)
589	{	871	{
590	ssize_t res;	872	eio_ssize_t res;
591	off_t ooffset;	873	off_t ooffset;
592		874
593	X_LOCK (preadwritelock);	875	X_LOCK (preadwritelock);
594	ooffset = lseek (fd, 0, SEEK_CUR);	876	ooffset = lseek (fd, 0, SEEK_CUR);
595	lseek (fd, offset, SEEK_SET);	877	lseek (fd, offset, SEEK_SET);
…		…
598	X_UNLOCK (preadwritelock);	880	X_UNLOCK (preadwritelock);
599		881
600	return res;	882	return res;
601	}	883	}
602		884
603	static ssize_t	885	static eio_ssize_t
604	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	886	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
605	{	887	{
606	ssize_t res;	888	eio_ssize_t res;
607	off_t ooffset;	889	off_t ooffset;
608		890
609	X_LOCK (preadwritelock);	891	X_LOCK (preadwritelock);
610	ooffset = lseek (fd, 0, SEEK_CUR);	892	ooffset = lseek (fd, 0, SEEK_CUR);
611	lseek (fd, offset, SEEK_SET);	893	lseek (fd, offset, SEEK_SET);
612	res = write (fd, buf, count);	894	res = write (fd, buf, count);
613	lseek (fd, offset, SEEK_SET);	895	lseek (fd, ooffset, SEEK_SET);
614	X_UNLOCK (preadwritelock);	896	X_UNLOCK (preadwritelock);
615		897
616	return res;	898	return res;
617	}	899	}
618	#endif	900	#endif
619		901
620	#ifndef HAVE_FUTIMES	902	#ifndef HAVE_UTIMES
621		903
		904	# undef utimes
622	# define utimes(path,times) eio__utimes (path, times)	905	# define utimes(path,times) eio__utimes (path, times)
623	# define futimes(fd,times) eio__futimes (fd, times)
624		906
625	static int	907	static int
626	eio__utimes (const char *filename, const struct timeval times[2])	908	eio__utimes (const char *filename, const struct timeval times[2])
627	{	909	{
628	if (times)	910	if (times)
…		…
636	}	918	}
637	else	919	else
638	return utime (filename, 0);	920	return utime (filename, 0);
639	}	921	}
640		922
		923	#endif
		924
		925	#ifndef HAVE_FUTIMES
		926
		927	# undef futimes
		928	# define futimes(fd,times) eio__futimes (fd, times)
		929
		930	static int
641	static int eio__futimes (int fd, const struct timeval tv[2])	931	eio__futimes (int fd, const struct timeval tv[2])
642	{	932	{
643	errno = ENOSYS;	933	errno = ENOSYS;
644	return -1;	934	return -1;
645	}	935	}
646		936
647	#endif	937	#endif
648		938
649	#if !HAVE_FDATASYNC	939	#if !HAVE_FDATASYNC
		940	# undef fdatasync
650	# define fdatasync fsync	941	# define fdatasync(fd) fsync (fd)
651	#endif	942	#endif
		943
		944	/* sync_file_range always needs emulation */
		945	static int
		946	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		947	{
		948	#if HAVE_SYNC_FILE_RANGE
		949	int res;
		950
		951	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		952	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		953	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		954	{
		955	flags = 0
		956	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		957	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		958	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		959	}
		960
		961	res = sync_file_range (fd, offset, nbytes, flags);
		962
		963	if (!res || errno != ENOSYS)
		964	return res;
		965	#endif
		966
		967	/* even though we could play tricks with the flags, it's better to always
		968	* call fdatasync, as that matches the expectation of its users best */
		969	return fdatasync (fd);
		970	}
		971
		972	static int
		973	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		974	{
		975	#if HAVE_FALLOCATE
		976	return fallocate (fd, mode, offset, len);
		977	#else
		978	errno = ENOSYS;
		979	return -1;
		980	#endif
		981	}
652		982
653	#if !HAVE_READAHEAD	983	#if !HAVE_READAHEAD
		984	# undef readahead
654	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	985	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
655		986
656	static ssize_t	987	static eio_ssize_t
657	eio__readahead (int fd, off_t offset, size_t count, worker *self)	988	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
658	{	989	{
659	size_t todo = count;	990	size_t todo = count;
660	dBUF;	991	dBUF;
661		992
662	while (todo > 0)	993	while (todo > 0)
…		…
672	return count;	1003	return count;
673	}	1004	}
674		1005
675	#endif	1006	#endif
676		1007
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 */	1008	/* sendfile always needs emulation */
709	static ssize_t	1009	static eio_ssize_t
710	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self)	1010	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
711	{	1011	{
		1012	eio_ssize_t written = 0;
712	ssize_t res;	1013	eio_ssize_t res;
713		1014
714	if (!count)	1015	if (!count)
715	return 0;	1016	return 0;
716		1017
		1018	for (;;)
		1019	{
		1020	#ifdef __APPLE__
		1021	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1022	#endif
717	#if HAVE_SENDFILE	1023	#if HAVE_SENDFILE
718	# if __linux	1024	# if __linux
		1025	off_t soffset = offset;
719	res = sendfile (ofd, ifd, &offset, count);	1026	res = sendfile (ofd, ifd, &soffset, count);
720		1027
721	# elif __freebsd	1028	# elif __FreeBSD__
722	/*	1029	/*
723	* Of course, the freebsd sendfile is a dire hack with no thoughts	1030	* Of course, the freebsd sendfile is a dire hack with no thoughts
724	* wasted on making it similar to other I/O functions.	1031	* wasted on making it similar to other I/O functions.
725	*/	1032	*/
726	{
727	off_t sbytes;	1033	off_t sbytes;
728	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1034	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
729		1035
730	if (res < 0 && sbytes)	1036	#if 0 /* according to the manpage, this is correct, but broken behaviour */
731	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1037	/* freebsd' sendfile will return 0 on success */
		1038	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1039	/* not on e.g. EIO or EPIPE - sounds broken */
		1040	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
732	res = sbytes;	1041	res = sbytes;
733	}	1042	#endif
		1043
		1044	/* according to source inspection, this is correct, and useful behaviour */
		1045	if (sbytes)
		1046	res = sbytes;
		1047
		1048	# elif defined (__APPLE__)
		1049	off_t sbytes = count;
		1050	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1051
		1052	/* according to the manpage, sbytes is always valid */
		1053	if (sbytes)
		1054	res = sbytes;
734		1055
735	# elif __hpux	1056	# elif __hpux
736	res = sendfile (ofd, ifd, offset, count, 0, 0);	1057	res = sendfile (ofd, ifd, offset, count, 0, 0);
737		1058
738	# elif __solaris	1059	# elif __solaris
739	{
740	struct sendfilevec vec;	1060	struct sendfilevec vec;
741	size_t sbytes;	1061	size_t sbytes;
742		1062
743	vec.sfv_fd = ifd;	1063	vec.sfv_fd = ifd;
744	vec.sfv_flag = 0;	1064	vec.sfv_flag = 0;
745	vec.sfv_off = offset;	1065	vec.sfv_off = offset;
746	vec.sfv_len = count;	1066	vec.sfv_len = count;
747		1067
748	res = sendfilev (ofd, &vec, 1, &sbytes);	1068	res = sendfilev (ofd, &vec, 1, &sbytes);
749		1069
750	if (res < 0 && sbytes)	1070	if (res < 0 && sbytes)
751	res = sbytes;	1071	res = sbytes;
752	}
753		1072
754	# endif	1073	# endif
		1074
		1075	#elif defined (_WIN32) && 0
		1076	/* does not work, just for documentation of what would need to be done */
		1077	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1078	/* libeio guarantees that the file offset does not change, and windows */
		1079	/* has no way to get an independent handle to the same file description */
		1080	HANDLE h = TO_SOCKET (ifd);
		1081	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1082	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1083
755	#else	1084	#else
756	res = -1;	1085	res = -1;
757	errno = ENOSYS;	1086	errno = ENOSYS;
758	#endif	1087	#endif
759		1088
		1089	/* we assume sendfile can copy at least 128mb in one go */
		1090	if (res <= 128 * 1024 * 1024)
		1091	{
		1092	if (res > 0)
		1093	written += res;
		1094
		1095	if (written)
		1096	return written;
		1097
		1098	break;
		1099	}
		1100	else
		1101	{
		1102	/* if we requested more, then probably the kernel was lazy */
		1103	written += res;
		1104	offset += res;
		1105	count -= res;
		1106
		1107	if (!count)
		1108	return written;
		1109	}
		1110	}
		1111
760	if (res < 0	1112	if (res < 0
761	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1113	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1114	/* BSDs */
		1115	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1116	|| errno == ENOTSUP
		1117	#endif
		1118	#ifdef EOPNOTSUPP /* windows */
		1119	|| errno == EOPNOTSUPP /* BSDs */
		1120	#endif
762	#if __solaris	1121	#if __solaris
763	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1122	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
764	#endif	1123	#endif
765	)	1124	)
766	)	1125	)
…		…
770		1129
771	res = 0;	1130	res = 0;
772		1131
773	while (count)	1132	while (count)
774	{	1133	{
775	ssize_t cnt;	1134	eio_ssize_t cnt;
776		1135
777	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1136	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
778		1137
779	if (cnt <= 0)	1138	if (cnt <= 0)
780	{	1139	{
…		…
797	}	1156	}
798		1157
799	return res;	1158	return res;
800	}	1159	}
801		1160
		1161	#ifdef PAGESIZE
		1162	# define eio_pagesize() PAGESIZE
		1163	#else
		1164	static intptr_t
		1165	eio_pagesize (void)
		1166	{
		1167	static intptr_t page;
		1168
		1169	if (!page)
		1170	page = sysconf (_SC_PAGESIZE);
		1171
		1172	return page;
		1173	}
		1174	#endif
		1175
		1176	static void
		1177	eio_page_align (void **addr, size_t *length)
		1178	{
		1179	intptr_t mask = eio_pagesize () - 1;
		1180
		1181	/* round down addr */
		1182	intptr_t adj = mask & (intptr_t)*addr;
		1183
		1184	*addr = (void *)((intptr_t)*addr - adj);
		1185	*length += adj;
		1186
		1187	/* round up length */
		1188	*length = (*length + mask) & ~mask;
		1189	}
		1190
		1191	#if !_POSIX_MEMLOCK
		1192	# define eio__mlockall(a) EIO_ENOSYS ()
		1193	#else
		1194
		1195	static int
		1196	eio__mlockall (int flags)
		1197	{
		1198	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1199	extern int mallopt (int, int);
		1200	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1201	#endif
		1202
		1203	if (EIO_MCL_CURRENT != MCL_CURRENT
		1204	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1205	{
		1206	flags = 0
		1207	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1208	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1209	}
		1210
		1211	return mlockall (flags);
		1212	}
		1213	#endif
		1214
		1215	#if !_POSIX_MEMLOCK_RANGE
		1216	# define eio__mlock(a,b) EIO_ENOSYS ()
		1217	#else
		1218
		1219	static int
		1220	eio__mlock (void *addr, size_t length)
		1221	{
		1222	eio_page_align (&addr, &length);
		1223
		1224	return mlock (addr, length);
		1225	}
		1226
		1227	#endif
		1228
		1229	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1230	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1231	#else
		1232
		1233	static int
		1234	eio__msync (void *mem, size_t len, int flags)
		1235	{
		1236	eio_page_align (&mem, &len);
		1237
		1238	if (EIO_MS_ASYNC != MS_SYNC
		1239	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1240	|| EIO_MS_SYNC != MS_SYNC)
		1241	{
		1242	flags = 0
		1243	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1244	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1245	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1246	}
		1247
		1248	return msync (mem, len, flags);
		1249	}
		1250
		1251	#endif
		1252
		1253	static int
		1254	eio__mtouch (eio_req *req)
		1255	{
		1256	void *mem = req->ptr2;
		1257	size_t len = req->size;
		1258	int flags = req->int1;
		1259
		1260	eio_page_align (&mem, &len);
		1261
		1262	{
		1263	intptr_t addr = (intptr_t)mem;
		1264	intptr_t end = addr + len;
		1265	intptr_t page = eio_pagesize ();
		1266
		1267	if (addr < end)
		1268	if (flags & EIO_MT_MODIFY) /* modify */
		1269	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1270	else
		1271	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1272	}
		1273
		1274	return 0;
		1275	}
		1276
		1277	/*****************************************************************************/
		1278	/* requests implemented outside eio_execute, because they are so large */
		1279
		1280	static void
		1281	eio__realpath (eio_req *req, etp_worker *self)
		1282	{
		1283	char *rel = req->ptr1;
		1284	char *res;
		1285	char *tmp1, *tmp2;
		1286	#if SYMLOOP_MAX > 32
		1287	int symlinks = SYMLOOP_MAX;
		1288	#else
		1289	int symlinks = 32;
		1290	#endif
		1291
		1292	req->result = -1;
		1293
		1294	errno = EINVAL;
		1295	if (!rel)
		1296	return;
		1297
		1298	errno = ENOENT;
		1299	if (!*rel)
		1300	return;
		1301
		1302	if (!req->ptr2)
		1303	{
		1304	X_LOCK (wrklock);
		1305	req->flags |= EIO_FLAG_PTR2_FREE;
		1306	X_UNLOCK (wrklock);
		1307	req->ptr2 = malloc (PATH_MAX * 3);
		1308
		1309	errno = ENOMEM;
		1310	if (!req->ptr2)
		1311	return;
		1312	}
		1313
		1314	res = req->ptr2;
		1315	tmp1 = res + PATH_MAX;
		1316	tmp2 = tmp1 + PATH_MAX;
		1317
		1318	#if 0 /* disabled, the musl way to do things is just too racy */
		1319	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1320	/* on linux we may be able to ask the kernel */
		1321	{
		1322	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1323
		1324	if (fd >= 0)
		1325	{
		1326	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1327	req->result = readlink (tmp1, res, PATH_MAX);
		1328	close (fd);
		1329
		1330	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1331
		1332	if (req->result > 0)
		1333	goto done;
		1334	}
		1335	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1336	return;
		1337	}
		1338	#endif
		1339	#endif
		1340
		1341	if (*rel != '/')
		1342	{
		1343	if (!getcwd (res, PATH_MAX))
		1344	return;
		1345
		1346	if (res [1]) /* only use if not / */
		1347	res += strlen (res);
		1348	}
		1349
		1350	while (*rel)
		1351	{
		1352	eio_ssize_t len, linklen;
		1353	char *beg = rel;
		1354
		1355	while (*rel && *rel != '/')
		1356	++rel;
		1357
		1358	len = rel - beg;
		1359
		1360	if (!len) /* skip slashes */
		1361	{
		1362	++rel;
		1363	continue;
		1364	}
		1365
		1366	if (beg [0] == '.')
		1367	{
		1368	if (len == 1)
		1369	continue; /* . - nop */
		1370
		1371	if (beg [1] == '.' && len == 2)
		1372	{
		1373	/* .. - back up one component, if possible */
		1374
		1375	while (res != req->ptr2)
		1376	if (*--res == '/')
		1377	break;
		1378
		1379	continue;
		1380	}
		1381	}
		1382
		1383	errno = ENAMETOOLONG;
		1384	if (res + 1 + len + 1 >= tmp1)
		1385	return;
		1386
		1387	/* copy one component */
		1388	*res = '/';
		1389	memcpy (res + 1, beg, len);
		1390
		1391	/* zero-terminate, for readlink */
		1392	res [len + 1] = 0;
		1393
		1394	/* now check if it's a symlink */
		1395	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1396
		1397	if (linklen < 0)
		1398	{
		1399	if (errno != EINVAL)
		1400	return;
		1401
		1402	/* it's a normal directory. hopefully */
		1403	res += len + 1;
		1404	}
		1405	else
		1406	{
		1407	/* yay, it was a symlink - build new path in tmp2 */
		1408	int rellen = strlen (rel);
		1409
		1410	errno = ENAMETOOLONG;
		1411	if (linklen + 1 + rellen >= PATH_MAX)
		1412	return;
		1413
		1414	errno = ELOOP;
		1415	if (!--symlinks)
		1416	return;
		1417
		1418	if (*tmp1 == '/')
		1419	res = req->ptr2; /* symlink resolves to an absolute path */
		1420
		1421	/* we need to be careful, as rel might point into tmp2 already */
		1422	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1423	tmp2 [linklen] = '/';
		1424	memcpy (tmp2, tmp1, linklen);
		1425
		1426	rel = tmp2;
		1427	}
		1428	}
		1429
		1430	/* special case for the lone root path */
		1431	if (res == req->ptr2)
		1432	*res++ = '/';
		1433
		1434	req->result = res - (char *)req->ptr2;
		1435
		1436	done:
		1437	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1438	}
		1439
		1440	static signed char
		1441	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1442	{
		1443	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1444	: a->inode < b->inode ? -1
		1445	: a->inode > b->inode ? 1
		1446	: 0;
		1447	}
		1448
		1449	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1450
		1451	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1452	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1453
		1454	static void
		1455	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1456	{
		1457	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
		1458	unsigned char *bit = bits;
		1459
		1460	assert (CHAR_BIT == 8);
		1461	assert (sizeof (eio_dirent) * 8 < 256);
		1462	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1463	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1464
		1465	if (size <= EIO_SORT_FAST)
		1466	return;
		1467
		1468	/* first prepare an array of bits to test in our radix sort */
		1469	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
		1470	/* inode_bits must contain all inodes ORed together */
		1471	/* which is used to skip bits that are 0 everywhere, which is very common */
		1472	{
		1473	eio_ino_t endianness;
		1474	int i, j;
		1475
		1476	/* we store the byte offset of byte n into byte n of "endianness" */
		1477	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1478	((unsigned char *)&endianness)[i] = i;
		1479
		1480	*bit++ = 0;
		1481
		1482	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1483	{
		1484	/* shifting off the byte offsets out of "endianness" */
		1485	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1486	endianness >>= 8;
		1487
		1488	for (j = 0; j < 8; ++j)
		1489	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
		1490	*bit++ = offs + j;
		1491	}
		1492
		1493	for (j = 0; j < 8; ++j)
		1494	if (score_bits & (1 << j))
		1495	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1496	}
		1497
		1498	/* now actually do the sorting (a variant of MSD radix sort) */
		1499	{
		1500	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
		1501	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
		1502	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
		1503	int stk_idx = 0;
		1504
		1505	base_stk [stk_idx] = dents;
		1506	end_stk [stk_idx] = dents + size;
		1507	bit_stk [stk_idx] = bit - 1;
		1508
		1509	do
		1510	{
		1511	base = base_stk [stk_idx];
		1512	end = end_stk [stk_idx];
		1513	bit = bit_stk [stk_idx];
		1514
		1515	for (;;)
		1516	{
		1517	unsigned char O = *bit >> 3;
		1518	unsigned char M = 1 << (*bit & 7);
		1519
		1520	eio_dirent *a = base;
		1521	eio_dirent *b = end;
		1522
		1523	if (b - a < EIO_SORT_CUTOFF)
		1524	break;
		1525
		1526	/* now bit-partition the array on the bit */
		1527	/* this ugly asymmetric loop seems to perform much better than typical */
		1528	/* partition algos found in the literature */
		1529	do
		1530	if (!(((unsigned char *)a)[O] & M))
		1531	++a;
		1532	else if (!(((unsigned char *)--b)[O] & M))
		1533	{
		1534	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1535	++a;
		1536	}
		1537	while (b > a);
		1538
		1539	/* next bit, or stop, if no bits left in this path */
		1540	if (!*--bit)
		1541	break;
		1542
		1543	base_stk [stk_idx] = a;
		1544	end_stk [stk_idx] = end;
		1545	bit_stk [stk_idx] = bit;
		1546	++stk_idx;
		1547
		1548	end = a;
		1549	}
		1550	}
		1551	while (stk_idx--);
		1552	}
		1553	}
		1554
		1555	static void
		1556	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1557	{
		1558	/* first move the smallest element to the front, to act as a sentinel */
		1559	{
		1560	int i;
		1561	eio_dirent *min = dents;
		1562
		1563	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1564	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1565	if (EIO_DENT_CMP (dents [i], <, *min))
		1566	min = &dents [i];
		1567
		1568	/* swap elements 0 and j (minimum) */
		1569	{
		1570	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1571	}
		1572	}
		1573
		1574	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1575	{
		1576	eio_dirent *i, *j;
		1577
		1578	for (i = dents + 1; i < dents + size; ++i)
		1579	{
		1580	eio_dirent value = *i;
		1581
		1582	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1583	j [1] = j [0];
		1584
		1585	j [1] = value;
		1586	}
		1587	}
		1588	}
		1589
		1590	static void
		1591	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1592	{
		1593	if (size <= 1)
		1594	return; /* our insertion sort relies on size > 0 */
		1595
		1596	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1597	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1598	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1599
		1600	/* use an insertion sort at the end, or for small arrays, */
		1601	/* as insertion sort is more efficient for small partitions */
		1602	eio_dent_insertion_sort (dents, size);
		1603	}
		1604
802	/* read a full directory */	1605	/* read a full directory */
803	static void	1606	static void
804	eio__scandir (eio_req *req, worker *self)	1607	eio__scandir (eio_req *req, etp_worker *self)
805	{	1608	{
		1609	char *name, *names;
		1610	int namesalloc = 4096 - sizeof (void *) * 4;
		1611	int namesoffs = 0;
		1612	int flags = req->int1;
		1613	eio_dirent *dents = 0;
		1614	int dentalloc = 128;
		1615	int dentoffs = 0;
		1616	eio_ino_t inode_bits = 0;
		1617	#ifdef _WIN32
		1618	HANDLE dirp;
		1619	WIN32_FIND_DATA entp;
		1620	#else
806	DIR *dirp;	1621	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;	1622	EIO_STRUCT_DIRENT *entp;
813	char *name, *names;	1623	#endif
814	int memlen = 4096;
815	int memofs = 0;
816	int res = 0;
817		1624
818	X_LOCK (etplock);	1625	req->result = -1;
		1626
		1627	if (!(flags & EIO_READDIR_DENTS))
		1628	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
		1629
		1630	#ifdef _WIN32
		1631	{
		1632	int len = strlen ((const char *)req->ptr1);
		1633	char *path = malloc (MAX_PATH);
		1634	const char *fmt;
		1635
		1636	if (!len)
		1637	fmt = "./*";
		1638	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
		1639	fmt = "%s*";
		1640	else
		1641	fmt = "%s/*";
		1642
		1643	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
		1644	dirp = FindFirstFile (path, &entp);
		1645	free (path);
		1646
		1647	if (dirp == INVALID_HANDLE_VALUE)
		1648	{
		1649	dirp = 0;
		1650
		1651	switch (GetLastError ())
		1652	{
		1653	case ERROR_FILE_NOT_FOUND:
		1654	req->result = 0;
		1655	break;
		1656
		1657	case ERROR_INVALID_NAME:
		1658	case ERROR_PATH_NOT_FOUND:
		1659	case ERROR_NO_MORE_FILES:
		1660	errno = ENOENT;
		1661	break;
		1662
		1663	case ERROR_NOT_ENOUGH_MEMORY:
		1664	errno = ENOMEM;
		1665	break;
		1666
		1667	default:
		1668	errno = EINVAL;
		1669	break;
		1670	}
		1671	}
		1672	}
		1673	#else
819	self->dirp = dirp = opendir (req->ptr1);	1674	dirp = opendir (req->ptr1);
820	self->dbuf = u = malloc (sizeof (*u));	1675	#endif
		1676
821	req->flags |= EIO_FLAG_PTR2_FREE;	1677	if (req->flags & EIO_FLAG_PTR1_FREE)
		1678	free (req->ptr1);
		1679
		1680	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1681	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
822	req->ptr2 = names = malloc (memlen);	1682	req->ptr2 = names = malloc (namesalloc);
823	X_UNLOCK (etplock);
824		1683
825	if (dirp && u && names)	1684	if (dirp && names && (!flags || dents))
826	for (;;)	1685	for (;;)
827	{	1686	{
		1687	int done;
		1688
		1689	#ifdef _WIN32
		1690	done = !dirp;
		1691	#else
828	errno = 0;	1692	errno = 0;
829	readdir_r (dirp, &u->d, &entp);	1693	entp = readdir (dirp);
		1694	done = !entp;
		1695	#endif
830		1696
831	if (!entp)	1697	if (done)
		1698	{
		1699	#ifndef _WIN32
		1700	int old_errno = errno;
		1701	closedir (dirp);
		1702	errno = old_errno;
		1703
		1704	if (errno)
		1705	break;
		1706	#endif
		1707
		1708	/* sort etc. */
		1709	req->int1 = flags;
		1710	req->result = dentoffs;
		1711
		1712	if (flags & EIO_READDIR_STAT_ORDER)
		1713	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1714	else if (flags & EIO_READDIR_DIRS_FIRST)
		1715	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1716	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1717	else
		1718	{
		1719	/* in this case, all is known, and we just put dirs first and sort them */
		1720	eio_dirent *oth = dents + dentoffs;
		1721	eio_dirent *dir = dents;
		1722
		1723	/* now partition dirs to the front, and non-dirs to the back */
		1724	/* by walking from both sides and swapping if necessary */
		1725	while (oth > dir)
		1726	{
		1727	if (dir->type == EIO_DT_DIR)
		1728	++dir;
		1729	else if ((--oth)->type == EIO_DT_DIR)
		1730	{
		1731	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1732
		1733	++dir;
		1734	}
		1735	}
		1736
		1737	/* now sort the dirs only (dirs all have the same score) */
		1738	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1739	}
		1740
832	break;	1741	break;
		1742	}
833		1743
834	name = entp->d_name;	1744	/* now add the entry to our list(s) */
		1745	name = D_NAME (entp);
835		1746
		1747	/* skip . and .. entries */
836	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1748	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
837	{	1749	{
838	int len = strlen (name) + 1;	1750	int len = D_NAMLEN (entp) + 1;
839		1751
840	res++;	1752	while (ecb_expect_false (namesoffs + len > namesalloc))
841
842	while (memofs + len > memlen)
843	{	1753	{
844	memlen *= 2;	1754	namesalloc *= 2;
845	X_LOCK (etplock);
846	req->ptr2 = names = realloc (names, memlen);	1755	req->ptr2 = names = realloc (names, namesalloc);
847	X_UNLOCK (etplock);
848		1756
849	if (!names)	1757	if (!names)
850	break;	1758	break;
851	}	1759	}
852		1760
853	memcpy (names + memofs, name, len);	1761	memcpy (names + namesoffs, name, len);
		1762
		1763	if (dents)
		1764	{
		1765	struct eio_dirent *ent;
		1766
		1767	if (ecb_expect_false (dentoffs == dentalloc))
		1768	{
		1769	dentalloc *= 2;
		1770	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1771
		1772	if (!dents)
		1773	break;
		1774	}
		1775
		1776	ent = dents + dentoffs;
		1777
		1778	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1779	ent->namelen = len - 1;
		1780	ent->inode = D_INO (entp);
		1781
		1782	inode_bits |= ent->inode;
		1783
		1784	switch (D_TYPE (entp))
		1785	{
		1786	default:
		1787	ent->type = EIO_DT_UNKNOWN;
		1788	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1789	break;
		1790
		1791	#ifdef DT_FIFO
		1792	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1793	#endif
		1794	#ifdef DT_CHR
		1795	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1796	#endif
		1797	#ifdef DT_MPC
		1798	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1799	#endif
		1800	#ifdef DT_DIR
		1801	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1802	#endif
		1803	#ifdef DT_NAM
		1804	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1805	#endif
		1806	#ifdef DT_BLK
		1807	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1808	#endif
		1809	#ifdef DT_MPB
		1810	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1811	#endif
		1812	#ifdef DT_REG
		1813	case DT_REG: ent->type = EIO_DT_REG; break;
		1814	#endif
		1815	#ifdef DT_NWK
		1816	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1817	#endif
		1818	#ifdef DT_CMP
		1819	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1820	#endif
		1821	#ifdef DT_LNK
		1822	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1823	#endif
		1824	#ifdef DT_SOCK
		1825	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1826	#endif
		1827	#ifdef DT_DOOR
		1828	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1829	#endif
		1830	#ifdef DT_WHT
		1831	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1832	#endif
		1833	}
		1834
		1835	ent->score = 7;
		1836
		1837	if (flags & EIO_READDIR_DIRS_FIRST)
		1838	{
		1839	if (ent->type == EIO_DT_UNKNOWN)
		1840	{
		1841	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1842	ent->score = 1;
		1843	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1844	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1845	}
		1846	else if (ent->type == EIO_DT_DIR)
		1847	ent->score = 0;
		1848	}
		1849	}
		1850
854	memofs += len;	1851	namesoffs += len;
		1852	++dentoffs;
855	}	1853	}
		1854
		1855	if (EIO_CANCELLED (req))
		1856	{
		1857	errno = ECANCELED;
		1858	break;
		1859	}
		1860
		1861	#ifdef _WIN32
		1862	if (!FindNextFile (dirp, &entp))
		1863	{
		1864	FindClose (dirp);
		1865	dirp = 0;
		1866	}
		1867	#endif
856	}	1868	}
857
858	if (errno)
859	res = -1;
860
861	req->result = res;
862	}	1869	}
863		1870
864	/*****************************************************************************/	1871	/*****************************************************************************/
865		1872
866	#define ALLOC(len) \	1873	#define ALLOC(len) \
867	if (!req->ptr2) \	1874	if (!req->ptr2) \
868	{ \	1875	{ \
869	X_LOCK (etplock); \	1876	X_LOCK (wrklock); \
870	req->flags |= EIO_FLAG_PTR2_FREE; \	1877	req->flags |= EIO_FLAG_PTR2_FREE; \
871	X_UNLOCK (etplock); \	1878	X_UNLOCK (wrklock); \
872	req->ptr2 = malloc (len); \	1879	req->ptr2 = malloc (len); \
873	if (!req->ptr2) \	1880	if (!req->ptr2) \
874	{ \	1881	{ \
875	errno = ENOMEM; \	1882	errno = ENOMEM; \
876	req->result = -1; \	1883	req->result = -1; \
877	break; \	1884	break; \
878	} \	1885	} \
879	}	1886	}
880		1887
881	X_THREAD_PROC (eio_proc)	1888	X_THREAD_PROC (etp_proc)
882	{	1889	{
883	eio_req *req;	1890	ETP_REQ *req;
884	struct timespec ts;	1891	struct timespec ts;
885	worker *self = (worker *)thr_arg;	1892	etp_worker *self = (etp_worker *)thr_arg;
886		1893
887	/* try to distribute timeouts somewhat randomly */	1894	/* try to distribute timeouts somewhat evenly */
888	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1895	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
889		1896
890	for (;;)	1897	for (;;)
891	{	1898	{
892	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1899	ts.tv_sec = 0;
893		1900
894	X_LOCK (reqlock);	1901	X_LOCK (reqlock);
895		1902
896	for (;;)	1903	for (;;)
897	{	1904	{
898	self->req = req = reqq_shift (&req_queue);	1905	self->req = req = reqq_shift (&req_queue);
899		1906
900	if (req)	1907	if (req)
901	break;	1908	break;
902		1909
		1910	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1911	{
		1912	X_UNLOCK (reqlock);
		1913	X_LOCK (wrklock);
		1914	--started;
		1915	X_UNLOCK (wrklock);
		1916	goto quit;
		1917	}
		1918
903	++idle;	1919	++idle;
904		1920
905	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1921	if (idle <= max_idle)
		1922	/* we are allowed to idle, so do so without any timeout */
		1923	X_COND_WAIT (reqwait, reqlock);
		1924	else
906	{	1925	{
907	if (idle > max_idle)	1926	/* initialise timeout once */
908	{	1927	if (!ts.tv_sec)
909	--idle;	1928	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		1929
917	/* we are allowed to idle, so do so without any timeout */
918	X_COND_WAIT (reqwait, reqlock);	1930	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
919	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1931	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
920	}	1932	}
921		1933
922	--idle;	1934	--idle;
923	}	1935	}
924		1936
925	--nready;	1937	--nready;
926		1938
927	X_UNLOCK (reqlock);	1939	X_UNLOCK (reqlock);
928		1940
929	errno = 0; /* strictly unnecessary */
930
931	if (!EIO_CANCELLED (req))
932	switch (req->type)	1941	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;	1942	goto quit;
1025		1943
1026	default:	1944	ETP_EXECUTE (self, req);
1027	req->result = -1;
1028	break;
1029	}
1030
1031	req->errorno = errno;
1032		1945
1033	X_LOCK (reslock);	1946	X_LOCK (reslock);
1034		1947
1035	++npending;	1948	++npending;
1036		1949
1037	if (!reqq_push (&res_queue, req) && etp.want_poll_cb)	1950	if (!reqq_push (&res_queue, req) && want_poll_cb)
1038	etp.want_poll_cb ();	1951	want_poll_cb ();
1039		1952
1040	self->req = 0;	1953	self->req = 0;
1041	worker_clear (self);	1954	etp_worker_clear (self);
1042		1955
1043	X_UNLOCK (reslock);	1956	X_UNLOCK (reslock);
1044	}	1957	}
1045		1958
1046	quit:	1959	quit:
1047	X_LOCK (etplock);	1960	X_LOCK (wrklock);
1048	worker_free (self);	1961	etp_worker_free (self);
1049	X_UNLOCK (etplock);	1962	X_UNLOCK (wrklock);
1050		1963
1051	return 0;	1964	return 0;
1052	}	1965	}
1053		1966
1054	/*****************************************************************************/	1967	/*****************************************************************************/
1055		1968
		1969	int ecb_cold
1056	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1970	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1057	{	1971	{
		1972	#if !HAVE_PREADWRITE
		1973	X_MUTEX_CREATE (preadwritelock);
		1974	#endif
		1975
1058	etp_init (&etp, want_poll, done_poll);	1976	return etp_init (want_poll, done_poll);
1059	}	1977	}
1060		1978
		1979	ecb_inline void
1061	static void eio_api_destroy (eio_req *req)	1980	eio_api_destroy (eio_req *req)
1062	{	1981	{
1063	free (req);	1982	free (req);
1064	}	1983	}
1065		1984
1066	#define REQ(rtype) \	1985	#define REQ(rtype) \
…		…
1085	{ \	2004	{ \
1086	eio_api_destroy (req); \	2005	eio_api_destroy (req); \
1087	return 0; \	2006	return 0; \
1088	}	2007	}
1089		2008
		2009	static void
		2010	eio_execute (etp_worker *self, eio_req *req)
		2011	{
		2012	if (ecb_expect_false (EIO_CANCELLED (req)))
		2013	{
		2014	req->result = -1;
		2015	req->errorno = ECANCELED;
		2016	return;
		2017	}
		2018
		2019	switch (req->type)
		2020	{
		2021	case EIO_READ: ALLOC (req->size);
		2022	req->result = req->offs >= 0
		2023	? pread (req->int1, req->ptr2, req->size, req->offs)
		2024	: read (req->int1, req->ptr2, req->size); break;
		2025	case EIO_WRITE: req->result = req->offs >= 0
		2026	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		2027	: write (req->int1, req->ptr2, req->size); break;
		2028
		2029	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		2030	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
		2031
		2032	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2033	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2034	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2035	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2036	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2037	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2038
		2039	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2040	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2041	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2042	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2043
		2044	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
		2045	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		2046	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
		2047	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		2048	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
		2049	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		2050
		2051	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
		2052	case EIO_CLOSE: req->result = close (req->int1); break;
		2053	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		2054	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
		2055	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
		2056	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
		2057	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
		2058	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
		2059	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
		2060	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2061
		2062	case EIO_REALPATH: eio__realpath (req, self); break;
		2063
		2064	case EIO_READLINK: ALLOC (PATH_MAX);
		2065	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
		2066
		2067	case EIO_SYNC: req->result = 0; sync (); break;
		2068	case EIO_FSYNC: req->result = fsync (req->int1); break;
		2069	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2070	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2071	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		2072	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2073	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		2074	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2075	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
		2076
		2077	case EIO_READDIR: eio__scandir (req, self); break;
		2078
		2079	case EIO_BUSY:
		2080	#ifdef _WIN32
		2081	Sleep (req->nv1 * 1e3);
		2082	#else
		2083	{
		2084	struct timeval tv;
		2085
		2086	tv.tv_sec = req->nv1;
		2087	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
		2088
		2089	req->result = select (0, 0, 0, 0, &tv);
		2090	}
		2091	#endif
		2092	break;
		2093
		2094	case EIO_UTIME:
		2095	case EIO_FUTIME:
		2096	{
		2097	struct timeval tv[2];
		2098	struct timeval *times;
		2099
		2100	if (req->nv1 != -1. || req->nv2 != -1.)
		2101	{
		2102	tv[0].tv_sec = req->nv1;
		2103	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
		2104	tv[1].tv_sec = req->nv2;
		2105	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
		2106
		2107	times = tv;
		2108	}
		2109	else
		2110	times = 0;
		2111
		2112	req->result = req->type == EIO_FUTIME
		2113	? futimes (req->int1, times)
		2114	: utimes (req->ptr1, times);
		2115	}
		2116	break;
		2117
		2118	case EIO_GROUP:
		2119	abort (); /* handled in eio_request */
		2120
		2121	case EIO_NOP:
		2122	req->result = 0;
		2123	break;
		2124
		2125	case EIO_CUSTOM:
		2126	req->feed (req);
		2127	break;
		2128
		2129	default:
		2130	errno = ENOSYS;
		2131	req->result = -1;
		2132	break;
		2133	}
		2134
		2135	req->errorno = errno;
		2136	}
		2137
1090	#ifndef EIO_NO_WRAPPERS	2138	#ifndef EIO_NO_WRAPPERS
1091		2139
1092	eio_req *eio_nop (int pri, eio_cb cb, void *data)	2140	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1093	{	2141	{
1094	REQ (EIO_NOP); SEND;	2142	REQ (EIO_NOP); SEND;
…		…
1107	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	2155	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1108	{	2156	{
1109	REQ (EIO_FSYNC); req->int1 = fd; SEND;	2157	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1110	}	2158	}
1111		2159
		2160	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2161	{
		2162	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2163	}
		2164
		2165	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2166	{
		2167	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2168	}
		2169
		2170	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2171	{
		2172	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2173	}
		2174
		2175	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2176	{
		2177	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2178	}
		2179
		2180	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2181	{
		2182	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2183	}
		2184
		2185	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2186	{
		2187	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2188	}
		2189
1112	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2190	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1113	{	2191	{
1114	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2192	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1115	}	2193	}
1116		2194
…		…
1137	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2215	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1138	{	2216	{
1139	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2217	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1140	}	2218	}
1141		2219
		2220	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2221	{
		2222	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2223	}
		2224
1142	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2225	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1143	{	2226	{
1144	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2227	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1145	}	2228	}
1146		2229
…		…
1152	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2235	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1153	{	2236	{
1154	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2237	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1155	}	2238	}
1156		2239
1157	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2240	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1158	{	2241	{
1159	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2242	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1160	}	2243	}
1161		2244
1162	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2245	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)	2265	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1183	{	2266	{
1184	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2267	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1185	}	2268	}
1186		2269
1187	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2270	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1188	{	2271	{
1189	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2272	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1190	}	2273	}
1191		2274
1192	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2275	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)	2291	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1209	{	2292	{
1210	return eio__1path (EIO_READLINK, path, pri, cb, data);	2293	return eio__1path (EIO_READLINK, path, pri, cb, data);
1211	}	2294	}
1212		2295
		2296	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2297	{
		2298	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2299	}
		2300
1213	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2301	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1214	{	2302	{
1215	return eio__1path (EIO_STAT, path, pri, cb, data);	2303	return eio__1path (EIO_STAT, path, pri, cb, data);
1216	}	2304	}
1217		2305
1218	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2306	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1219	{	2307	{
1220	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2308	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1221	}	2309	}
1222		2310
		2311	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2312	{
		2313	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2314	}
		2315
1223	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2316	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1224	{	2317	{
1225	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2318	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1226	}	2319	}
1227		2320
1228	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	2321	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1229	{	2322	{
1230	return eio__1path (EIO_RMDIR, path, pri, cb, data);	2323	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1231	}	2324	}
1232		2325
1233	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	2326	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1234	{	2327	{
1235	return eio__1path (EIO_READDIR, path, pri, cb, data);	2328	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1236	}	2329	}
1237		2330
1238	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2331	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1239	{	2332	{
1240	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	2333	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1241	}	2334	}
1242		2335
1243	static eio_req *	2336	static eio_req *
1244	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2337	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1245	{	2338	{
…		…
1269	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2362	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1270	{	2363	{
1271	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2364	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1272	}	2365	}
1273		2366
		2367	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
		2368	{
		2369	REQ (EIO_CUSTOM); req->feed = execute; SEND;
		2370	}
		2371
1274	#endif	2372	#endif
1275		2373
1276	eio_req *eio_grp (eio_cb cb, void *data)	2374	eio_req *eio_grp (eio_cb cb, void *data)
1277	{	2375	{
1278	const int pri = EIO_PRI_MAX;	2376	const int pri = EIO_PRI_MAX;
…		…
1285	#undef SEND	2383	#undef SEND
1286		2384
1287	/*****************************************************************************/	2385	/*****************************************************************************/
1288	/* grp functions */	2386	/* grp functions */
1289		2387
		2388	void
1290	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2389	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1291	{	2390	{
1292	grp->int2 = limit;	2391	grp->int2 = limit;
1293	grp->feed = feed;	2392	grp->feed = feed;
1294		2393
1295	grp_try_feed (grp);	2394	grp_try_feed (grp);
1296	}	2395	}
1297		2396
		2397	void
1298	void eio_grp_limit (eio_req *grp, int limit)	2398	eio_grp_limit (eio_req *grp, int limit)
1299	{	2399	{
1300	grp->int2 = limit;	2400	grp->int2 = limit;
1301		2401
1302	grp_try_feed (grp);	2402	grp_try_feed (grp);
1303	}	2403	}
1304		2404
		2405	void
1305	void eio_grp_add (eio_req *grp, eio_req *req)	2406	eio_grp_add (eio_req *grp, eio_req *req)
1306	{	2407	{
1307	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2408	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
		2409
		2410	grp->flags |= EIO_FLAG_GROUPADD;
1308		2411
1309	++grp->size;	2412	++grp->size;
1310	req->grp = grp;	2413	req->grp = grp;
1311		2414
1312	req->grp_prev = 0;	2415	req->grp_prev = 0;
…		…
1319	}	2422	}
1320		2423
1321	/*****************************************************************************/	2424	/*****************************************************************************/
1322	/* misc garbage */	2425	/* misc garbage */
1323		2426
		2427	eio_ssize_t
1324	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2428	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1325	{	2429	{
1326	worker wrk;	2430	etp_worker wrk;
		2431	eio_ssize_t ret;
1327		2432
1328	wrk.dbuf = 0;	2433	wrk.dbuf = 0;
1329		2434
1330	eio__sendfile (ofd, ifd, offset, count, &wrk);	2435	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1331		2436
1332	if (wrk.dbuf)	2437	if (wrk.dbuf)
1333	free (wrk.dbuf);	2438	free (wrk.dbuf);
1334	}
1335		2439
		2440	return ret;
		2441	}
		2442

Diff Legend

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