ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.12 by root, Tue May 13 18:54:52 2008 UTC vs.
Revision 1.139 by root, Thu Jun 25 18:14:19 2015 UTC

		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008,2009,2010,2011,2012,2013 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 X_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	#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
		77	# define ENOTSOCK WSAENOTSOCK
		78	#endif
		79
		80	static void eio_destroy (eio_req *req);
		81
15	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
16	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	83	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17	#endif	84	#endif
18		85
19	#ifndef EIO_DESTROY	86	#ifndef EIO_DESTROY
…		…
22		89
23	#ifndef EIO_FEED	90	#ifndef EIO_FEED
24	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	91	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
25	#endif	92	#endif
26		93
		94	#ifndef EIO_FD_TO_WIN32_HANDLE
		95	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		96	#endif
		97	#ifndef EIO_WIN32_HANDLE_TO_FD
		98	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		99	#endif
		100
		101	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		102
		103	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		104
27	#ifdef _WIN32	105	#ifdef _WIN32
28		106
29	/*doh*/	107	#undef PAGESIZE
		108	#define PAGESIZE 4096 /* GetSystemInfo? */
		109
		110	/* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
		111
		112	#ifdef EIO_STRUCT_STATI64
		113	/* look at perl's non-sloppy stat */
		114	#define stat(path,buf) _stati64 (path,buf)
		115	#define fstat(fd,buf) _fstati64 (fd,buf)
		116	#endif
		117	#define lstat(path,buf) stat (path,buf)
		118	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		119	#define mkdir(path,mode) _mkdir (path)
		120	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		121
		122	#define chmod(path,mode) _chmod (path, mode)
		123	#define dup(fd) _dup (fd)
		124	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		125	#define pipe(fds) _pipe (fds, 4096, O_BINARY)
		126
		127	#define fchmod(fd,mode) EIO_ENOSYS ()
		128	#define chown(path,uid,gid) EIO_ENOSYS ()
		129	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		130	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		131	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		132	#define mknod(path,mode,dev) EIO_ENOSYS ()
		133	#define sync() EIO_ENOSYS ()
		134	#define readlink(path,buf,s) EIO_ENOSYS ()
		135	#define statvfs(path,buf) EIO_ENOSYS ()
		136	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		137
		138	#define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
		139	#define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
		140
		141	#if __GNUC__
		142	typedef long long eio_off_t; /* signed for compatibility to msvc */
		143	#else
		144	typedef __int64 eio_off_t; /* unsigned not supported by msvc */
		145	#endif
		146
		147	static eio_ssize_t
		148	eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
		149	{
		150	OVERLAPPED o = { 0 };
		151	DWORD got;
		152
		153	o.Offset = offset;
		154	o.OffsetHigh = offset >> 32;
		155
		156	return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
		157	? got : -1;
		158	}
		159
		160	static eio_ssize_t
		161	eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
		162	{
		163	OVERLAPPED o = { 0 };
		164	DWORD got;
		165
		166	o.Offset = offset;
		167	o.OffsetHigh = offset >> 32;
		168
		169	return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
		170	? got : -1;
		171	}
		172
		173	/* rename() uses MoveFile, which fails to overwrite */
		174	#define rename(old,neu) eio__rename (old, neu)
		175
		176	static int
		177	eio__rename (const char *old, const char *neu)
		178	{
		179	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		180	return 0;
		181
		182	/* should steal _dosmaperr */
		183	switch (GetLastError ())
		184	{
		185	case ERROR_FILE_NOT_FOUND:
		186	case ERROR_PATH_NOT_FOUND:
		187	case ERROR_INVALID_DRIVE:
		188	case ERROR_NO_MORE_FILES:
		189	case ERROR_BAD_NETPATH:
		190	case ERROR_BAD_NET_NAME:
		191	case ERROR_BAD_PATHNAME:
		192	case ERROR_FILENAME_EXCED_RANGE:
		193	errno = ENOENT;
		194	break;
		195
		196	default:
		197	errno = EACCES;
		198	break;
		199	}
		200
		201	return -1;
		202	}
		203
		204	/* we could even stat and see if it exists */
		205	static int
		206	symlink (const char *old, const char *neu)
		207	{
		208	#if WINVER >= 0x0600
		209	if (CreateSymbolicLink (neu, old, 1))
		210	return 0;
		211
		212	if (CreateSymbolicLink (neu, old, 0))
		213	return 0;
		214	#endif
		215
		216	return EIO_ERRNO (ENOENT, -1);
		217	}
		218
		219	/* POSIX API only, causing trouble for win32 apps */
		220	#define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
		221	#define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
		222
		223	struct statvfs
		224	{
		225	int dummy;
		226	};
		227
		228	#define DT_DIR EIO_DT_DIR
		229	#define DT_REG EIO_DT_REG
		230	#define D_NAME(entp) entp.cFileName
		231	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
30		232
31	#else	233	#else
32		234
33	# include "config.h"
34	# include <sys/time.h>	235	#include <sys/time.h>
35	# include <sys/select.h>	236	#include <sys/select.h>
36	# include <unistd.h>	237	#include <unistd.h>
		238	#include <signal.h>
		239	#include <dirent.h>
		240
		241	#ifdef ANDROID
		242	#include <sys/vfs.h>
		243	#define statvfs statfs
		244	#define fstatvfs fstatfs
		245	#include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
		246	#else
		247	#include <sys/statvfs.h>
		248	#endif
		249
		250	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		251	#include <sys/mman.h>
		252	#endif
		253
		254	#define D_NAME(entp) entp->d_name
		255
		256	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		257	#if __FreeBSD__ || __NetBSD__ || __OpenBSD__
		258	#define _DIRENT_HAVE_D_TYPE /* sigh */
		259	#define D_INO(de) (de)->d_fileno
		260	#define D_NAMLEN(de) (de)->d_namlen
		261	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		262	#define D_INO(de) (de)->d_ino
		263	#endif
		264
		265	#ifdef _D_EXACT_NAMLEN
		266	#undef D_NAMLEN
		267	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		268	#endif
		269
		270	#ifdef _DIRENT_HAVE_D_TYPE
		271	#define D_TYPE(de) (de)->d_type
		272	#endif
		273
		274	#ifndef EIO_STRUCT_DIRENT
		275	#define EIO_STRUCT_DIRENT struct dirent
		276	#endif
		277
		278	#endif
		279
		280	#if HAVE_UTIMES
37	# include <utime.h>	281	# include <utime.h>
38	# include <signal.h>
39	# include <dirent.h>
40
41	# ifndef EIO_STRUCT_DIRENT
42	# define EIO_STRUCT_DIRENT struct dirent
43	# endif	282	#endif
44		283
		284	#if HAVE_SYS_SYSCALL_H
		285	# include <sys/syscall.h>
		286	#endif
		287
		288	#if HAVE_SYS_PRCTL_H
		289	# include <sys/prctl.h>
45	#endif	290	#endif
46		291
47	#if HAVE_SENDFILE	292	#if HAVE_SENDFILE
48	# if __linux	293	# if __linux
49	# include <sys/sendfile.h>	294	# include <sys/sendfile.h>
50	# elif __freebsd	295	# elif __FreeBSD__ || defined __APPLE__
51	# include <sys/socket.h>	296	# include <sys/socket.h>
52	# include <sys/uio.h>	297	# include <sys/uio.h>
53	# elif __hpux	298	# elif __hpux
54	# include <sys/socket.h>	299	# include <sys/socket.h>
55	# elif __solaris /* not yet */	300	# elif __solaris
56	# include <sys/sendfile.h>	301	# include <sys/sendfile.h>
57	# else	302	# else
58	# error sendfile support requested but not available	303	# error sendfile support requested but not available
59	# endif	304	# endif
60	#endif	305	#endif
61		306
62	/* number of seconds after which an idle threads exit */	307	#ifndef D_TYPE
63	#define IDLE_TIMEOUT 10	308	# define D_TYPE(de) 0
		309	#endif
		310	#ifndef D_INO
		311	# define D_INO(de) 0
		312	#endif
		313	#ifndef D_NAMLEN
		314	# define D_NAMLEN(entp) strlen (D_NAME (entp))
		315	#endif
64		316
65	/* used for struct dirent, AIX doesn't provide it */	317	/* used for struct dirent, AIX doesn't provide it */
66	#ifndef NAME_MAX	318	#ifndef NAME_MAX
67	# define NAME_MAX 4096	319	# define NAME_MAX 4096
68	#endif	320	#endif
69		321
		322	/* used for readlink etc. */
		323	#ifndef PATH_MAX
		324	# define PATH_MAX 4096
		325	#endif
		326
70	/* buffer size for various temporary buffers */	327	/* buffer size for various temporary buffers */
71	#define EIO_BUFSIZE 65536	328	#define EIO_BUFSIZE 65536
72		329
73	#define dBUF \	330	#define dBUF \
74	char *eio_buf; \
75	X_LOCK (wrklock); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	331	char *eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (wrklock); \
78	errno = ENOMEM; \	332	errno = ENOMEM; \
79	if (!eio_buf) \	333	if (!eio_buf) \
80	return -1;	334	return -1
81		335
82	#define EIO_TICKS ((1000000 + 1023) >> 10)	336	#define FUBd \
		337	free (eio_buf)
83		338
84	static void (*want_poll_cb) (void);	339	/*****************************************************************************/
85	static void (*done_poll_cb) (void);
86		340
87	static unsigned int max_poll_time = 0;	341	struct etp_tmpbuf;
88	static unsigned int max_poll_reqs = 0;
89		342
90	/* calculcate time difference in ~1/EIO_TICKS of a second */	343	#if _POSIX_VERSION >= 200809L
91	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	344	#define HAVE_AT 1
92	{	345	#define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
93	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	346	#ifndef O_SEARCH
94	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	347	#define O_SEARCH O_RDONLY
95	}	348	#endif
		349	#else
		350	#define HAVE_AT 0
		351	static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
		352	#endif
96		353
97	static unsigned int started, idle, wanted = 4;	354	struct eio_pwd
98
99	/* worker threads management */
100	static mutex_t wrklock = X_MUTEX_INIT;
101
102	typedef struct worker
103	{	355	{
104	/* locked by wrklock */	356	#if HAVE_AT
105	struct worker *prev, *next;	357	int fd;
106		358	#endif
107	thread_t tid;
108
109	/* locked by reslock, reqlock or wrklock */
110	eio_req *req; /* currently processed request */
111	void *dbuf;
112	DIR *dirp;
113	} worker;
114
115	static worker wrk_first = { &wrk_first, &wrk_first, 0 };
116
117	static void worker_clear (worker *wrk)
118	{
119	if (wrk->dirp)
120	{
121	closedir (wrk->dirp);
122	wrk->dirp = 0;
123	}
124
125	if (wrk->dbuf)
126	{
127	free (wrk->dbuf);
128	wrk->dbuf = 0;
129	}
130	}
131
132	static void worker_free (worker *wrk)
133	{
134	wrk->next->prev = wrk->prev;
135	wrk->prev->next = wrk->next;
136
137	free (wrk);
138	}
139
140	static volatile unsigned int nreqs, nready, npending;
141	static volatile unsigned int max_idle = 4;
142
143	static mutex_t reslock = X_MUTEX_INIT;
144	static mutex_t reqlock = X_MUTEX_INIT;
145	static cond_t reqwait = X_COND_INIT;
146
147	unsigned int eio_nreqs (void)
148	{
149	return nreqs;
150	}
151
152	unsigned int eio_nready (void)
153	{
154	unsigned int retval;
155
156	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157	retval = nready;
158	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159
160	return retval;
161	}
162
163	unsigned int eio_npending (void)
164	{
165	unsigned int retval;
166
167	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168	retval = npending;
169	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170
171	return retval;
172	}
173
174	unsigned int eio_nthreads (void)
175	{
176	unsigned int retval;
177
178	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179	retval = started;
180	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
181
182	return retval;
183	}
184
185	/*
186	* a somewhat faster data structure might be nice, but
187	* with 8 priorities this actually needs <20 insns
188	* per shift, the most expensive operation.
189	*/
190	typedef struct {
191	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */
192	int size;	359	int len;
193	} reqq;	360	char str[1]; /* actually, a 0-terminated canonical path */
		361	};
194		362
195	static reqq req_queue;	363	/*****************************************************************************/
196	static reqq res_queue;
197		364
198	static int reqq_push (reqq *q, eio_req *req)	365	#define ETP_PRI_MIN EIO_PRI_MIN
199	{	366	#define ETP_PRI_MAX EIO_PRI_MAX
200	int pri = req->pri;
201	req->next = 0;
202		367
203	if (q->qe[pri])	368	#define ETP_TYPE_QUIT -1
204	{	369	#define ETP_TYPE_GROUP EIO_GROUP
205	q->qe[pri]->next = req;
206	q->qe[pri] = req;
207	}
208	else
209	q->qe[pri] = q->qs[pri] = req;
210		370
211	return q->size++;	371	static void eio_nop_callback (void) { }
212	}	372	static void (*eio_want_poll_cb)(void) = eio_nop_callback;
		373	static void (*eio_done_poll_cb)(void) = eio_nop_callback;
213		374
214	static eio_req *reqq_shift (reqq *q)	375	#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
215	{	376	#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
216	int pri;
217		377
218	if (!q->size)	378	struct etp_worker;
219	return 0;	379	#define ETP_REQ eio_req
		380	#define ETP_DESTROY(req) eio_destroy (req)
		381	static int eio_finish (eio_req *req);
		382	#define ETP_FINISH(req) eio_finish (req)
		383	static void eio_execute (struct etp_worker *self, eio_req *req);
		384	#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
220		385
221	--q->size;	386	#include "etp.c"
222		387
223	for (pri = EIO_NUM_PRI; pri--; )	388	static struct etp_pool eio_pool;
224	{	389	#define EIO_POOL (&eio_pool)
225	eio_req *req = q->qs[pri];
226		390
227	if (req)	391	/*****************************************************************************/
228	{
229	if (!(q->qs[pri] = (eio_req *)req->next))
230	q->qe[pri] = 0;
231		392
232	return req;	393	static void
233	}
234	}
235
236	abort ();
237	}
238
239	static void grp_try_feed (eio_req *grp)	394	grp_try_feed (eio_req *grp)
240	{	395	{
241	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	396	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
242	{	397	{
243	int old_len = grp->size;	398	grp->flags &= ~ETP_FLAG_GROUPADD;
244		399
245	EIO_FEED (grp);	400	EIO_FEED (grp);
246		401
247	/* stop if no progress has been made */	402	/* stop if no progress has been made */
248	if (old_len == grp->size)	403	if (!(grp->flags & ETP_FLAG_GROUPADD))
249	{	404	{
250	grp->feed = 0;	405	grp->feed = 0;
251	break;	406	break;
252	}	407	}
253	}	408	}
254	}	409	}
255		410
256	static int eio_finish (eio_req *req);	411	static int
257
258	static int grp_dec (eio_req *grp)	412	grp_dec (eio_req *grp)
259	{	413	{
260	--grp->size;	414	--grp->size;
261		415
262	/* call feeder, if applicable */	416	/* call feeder, if applicable */
263	grp_try_feed (grp);	417	grp_try_feed (grp);
264		418
265	/* finish, if done */	419	/* finish, if done */
266	if (!grp->size && grp->int1)	420	if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
267	return eio_finish (grp);	421	return eio_finish (grp);
268	else	422	else
269	return 0;	423	return 0;
270	}	424	}
271		425
		426	static void
272	void eio_destroy (eio_req *req)	427	eio_destroy (eio_req *req)
273	{	428	{
274	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	429	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
275	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	430	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
276		431
277	EIO_DESTROY (req);	432	EIO_DESTROY (req);
278	}	433	}
279		434
		435	static int
280	static int eio_finish (eio_req *req)	436	eio_finish (eio_req *req)
281	{	437	{
282	int res = EIO_FINISH (req);	438	int res = EIO_FINISH (req);
283		439
284	if (req->grp)	440	if (req->grp)
285	{	441	{
…		…
293	if (grp->grp_first == req)	449	if (grp->grp_first == req)
294	grp->grp_first = req->grp_next;	450	grp->grp_first = req->grp_next;
295		451
296	res2 = grp_dec (grp);	452	res2 = grp_dec (grp);
297		453
298	if (!res && res2)	454	if (!res)
299	res = res2;	455	res = res2;
300	}	456	}
301		457
302	eio_destroy (req);	458	eio_destroy (req);
303		459
304	return res;	460	return res;
305	}	461	}
306		462
		463	void
307	void eio_grp_cancel (eio_req *grp)	464	eio_grp_cancel (eio_req *grp)
308	{	465	{
309	for (grp = grp->grp_first; grp; grp = grp->grp_next)	466	etp_grp_cancel (EIO_POOL, grp);
310	eio_cancel (grp);
311	}	467	}
312		468
		469	void
313	void eio_cancel (eio_req *req)	470	eio_cancel (eio_req *req)
314	{	471	{
315	X_LOCK (wrklock);	472	etp_cancel (EIO_POOL, req);
316	req->flags |= EIO_FLAG_CANCELLED;
317	X_UNLOCK (wrklock);
318
319	eio_grp_cancel (req);
320	}	473	}
321		474
322	X_THREAD_PROC (eio_proc);	475	void
323
324	static void start_thread (void)
325	{
326	worker *wrk = calloc (1, sizeof (worker));
327
328	/*TODO*/
329	assert (("unable to allocate worker thread data", wrk));
330
331	X_LOCK (wrklock);
332
333	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))
334	{
335	wrk->prev = &wrk_first;
336	wrk->next = wrk_first.next;
337	wrk_first.next->prev = wrk;
338	wrk_first.next = wrk;
339	++started;
340	}
341	else
342	free (wrk);
343
344	X_UNLOCK (wrklock);
345	}
346
347	static void maybe_start_thread (void)
348	{
349	if (eio_nthreads () >= wanted)
350	return;
351
352	/* todo: maybe use idle here, but might be less exact */
353	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())
354	return;
355
356	start_thread ();
357	}
358
359	void eio_submit (eio_req *req)	476	eio_submit (eio_req *req)
360	{	477	{
361	req->pri += EIO_PRI_BIAS;	478	etp_submit (EIO_POOL, req);
362
363	if (req->pri < EIO_PRI_MIN + EIO_PRI_BIAS) req->pri = EIO_PRI_MIN + EIO_PRI_BIAS;
364	if (req->pri > EIO_PRI_MAX + EIO_PRI_BIAS) req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
365
366	++nreqs;
367
368	X_LOCK (reqlock);
369	++nready;
370	reqq_push (&req_queue, req);
371	X_COND_SIGNAL (reqwait);
372	X_UNLOCK (reqlock);
373
374	maybe_start_thread ();
375	}	479	}
376		480
377	static void end_thread (void)	481	unsigned int
		482	eio_nreqs (void)
378	{	483	{
379	eio_req *req = calloc (1, sizeof (eio_req));	484	return etp_nreqs (EIO_POOL);
380
381	req->type = EIO_QUIT;
382	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
383
384	X_LOCK (reqlock);
385	reqq_push (&req_queue, req);
386	X_COND_SIGNAL (reqwait);
387	X_UNLOCK (reqlock);
388
389	X_LOCK (wrklock);
390	--started;
391	X_UNLOCK (wrklock);
392	}	485	}
393		486
		487	unsigned int
		488	eio_nready (void)
		489	{
		490	return etp_nready (EIO_POOL);
		491	}
		492
		493	unsigned int
		494	eio_npending (void)
		495	{
		496	return etp_npending (EIO_POOL);
		497	}
		498
		499	unsigned int ecb_cold
		500	eio_nthreads (void)
		501	{
		502	return etp_nthreads (EIO_POOL);
		503	}
		504
		505	void ecb_cold
394	void eio_set_max_poll_time (double nseconds)	506	eio_set_max_poll_time (double nseconds)
395	{	507	{
396	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	508	etp_set_max_poll_time (EIO_POOL, nseconds);
397	max_poll_time = nseconds;
398	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
399	}	509	}
400		510
		511	void ecb_cold
401	void eio_set_max_poll_reqs (unsigned int maxreqs)	512	eio_set_max_poll_reqs (unsigned int maxreqs)
402	{	513	{
403	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	514	etp_set_max_poll_reqs (EIO_POOL, maxreqs);
404	max_poll_reqs = maxreqs;
405	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
406	}	515	}
407		516
		517	void ecb_cold
408	void eio_set_max_idle (unsigned int nthreads)	518	eio_set_max_idle (unsigned int nthreads)
409	{	519	{
410	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	520	etp_set_max_idle (EIO_POOL, nthreads);
411	max_idle = nthreads <= 0 ? 1 : nthreads;
412	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
413	}	521	}
414		522
		523	void ecb_cold
		524	eio_set_idle_timeout (unsigned int seconds)
		525	{
		526	etp_set_idle_timeout (EIO_POOL, seconds);
		527	}
		528
		529	void ecb_cold
415	void eio_set_min_parallel (unsigned int nthreads)	530	eio_set_min_parallel (unsigned int nthreads)
416	{	531	{
417	if (wanted < nthreads)	532	etp_set_min_parallel (EIO_POOL, nthreads);
418	wanted = nthreads;
419	}	533	}
420		534
		535	void ecb_cold
421	void eio_set_max_parallel (unsigned int nthreads)	536	eio_set_max_parallel (unsigned int nthreads)
422	{	537	{
423	if (wanted > nthreads)	538	etp_set_max_parallel (EIO_POOL, nthreads);
424	wanted = nthreads;
425
426	while (started > wanted)
427	end_thread ();
428	}	539	}
429		540
430	int eio_poll (void)	541	int eio_poll (void)
431	{	542	{
432	int maxreqs = max_poll_reqs;	543	return etp_poll (EIO_POOL);
433	struct timeval tv_start, tv_now;
434	eio_req *req;
435
436	if (max_poll_time)
437	gettimeofday (&tv_start, 0);
438
439	for (;;)
440	{
441	maybe_start_thread ();
442
443	X_LOCK (reslock);
444	req = reqq_shift (&res_queue);
445
446	if (req)
447	{
448	--npending;
449
450	if (!res_queue.size && done_poll_cb)
451	done_poll_cb ();
452	}
453
454	X_UNLOCK (reslock);
455
456	if (!req)
457	return 0;
458
459	--nreqs;
460
461	if (req->type == EIO_GROUP && req->size)
462	{
463	req->int1 = 1; /* mark request as delayed */
464	continue;
465	}
466	else
467	{
468	int res = eio_finish (req);
469	if (res)
470	return res;
471	}
472
473	if (maxreqs && !--maxreqs)
474	break;
475
476	if (max_poll_time)
477	{
478	gettimeofday (&tv_now, 0);
479
480	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)
481	break;
482	}
483	}
484
485	errno = EAGAIN;
486	return -1;
487	}	544	}
488		545
489	/*****************************************************************************/	546	/*****************************************************************************/
490	/* work around various missing functions */	547	/* work around various missing functions */
491		548
492	#if !HAVE_PREADWRITE
493	# define pread eio__pread
494	# define pwrite eio__pwrite
495
496	/*
497	* make our pread/pwrite safe against themselves, but not against
498	* normal read/write by using a mutex. slows down execution a lot,
499	* but that's your problem, not mine.
500	*/
501	static mutex_t preadwritelock = X_MUTEX_INIT;
502
503	static ssize_t
504	eio__pread (int fd, void *buf, size_t count, off_t offset)
505	{
506	ssize_t res;
507	off_t ooffset;
508
509	X_LOCK (preadwritelock);
510	ooffset = lseek (fd, 0, SEEK_CUR);
511	lseek (fd, offset, SEEK_SET);
512	res = read (fd, buf, count);
513	lseek (fd, ooffset, SEEK_SET);
514	X_UNLOCK (preadwritelock);
515
516	return res;
517	}
518
519	static ssize_t
520	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
521	{
522	ssize_t res;
523	off_t ooffset;
524
525	X_LOCK (preadwritelock);
526	ooffset = lseek (fd, 0, SEEK_CUR);
527	lseek (fd, offset, SEEK_SET);
528	res = write (fd, buf, count);
529	lseek (fd, offset, SEEK_SET);
530	X_UNLOCK (preadwritelock);
531
532	return res;
533	}
534	#endif
535
536	#ifndef HAVE_FUTIMES	549	#ifndef HAVE_UTIMES
537		550
		551	# undef utimes
538	# define utimes(path,times) eio__utimes (path, times)	552	# define utimes(path,times) eio__utimes (path, times)
539	# define futimes(fd,times) eio__futimes (fd, times)
540		553
541	static int	554	static int
542	eio__utimes (const char *filename, const struct timeval times[2])	555	eio__utimes (const char *filename, const struct timeval times[2])
543	{	556	{
544	if (times)	557	if (times)
…		…
552	}	565	}
553	else	566	else
554	return utime (filename, 0);	567	return utime (filename, 0);
555	}	568	}
556		569
		570	#endif
		571
		572	#ifndef HAVE_FUTIMES
		573
		574	# undef futimes
		575	# define futimes(fd,times) eio__futimes (fd, times)
		576
		577	static int
557	static int eio__futimes (int fd, const struct timeval tv[2])	578	eio__futimes (int fd, const struct timeval tv[2])
558	{	579	{
559	errno = ENOSYS;	580	errno = ENOSYS;
560	return -1;	581	return -1;
561	}	582	}
562		583
563	#endif	584	#endif
564		585
565	#if !HAVE_FDATASYNC	586	#if !HAVE_FDATASYNC
		587	# undef fdatasync
566	# define fdatasync fsync	588	# define fdatasync(fd) fsync (fd)
567	#endif	589	#endif
		590
		591	static int
		592	eio__syncfs (int fd)
		593	{
		594	int res;
		595
		596	#if HAVE_SYS_SYNCFS
		597	res = (int)syscall (__NR_syncfs, (int)(fd));
		598	#else
		599	res = EIO_ENOSYS ();
		600	#endif
		601
		602	if (res < 0 && errno == ENOSYS && fd >= 0)
		603	sync ();
		604
		605	return res;
		606	}
		607
		608	/* sync_file_range always needs emulation */
		609	static int
		610	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		611	{
		612	#if HAVE_SYNC_FILE_RANGE
		613	int res;
		614
		615	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		616	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		617	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		618	{
		619	flags = 0
		620	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		621	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		622	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		623	}
		624
		625	res = sync_file_range (fd, offset, nbytes, flags);
		626
		627	if (!res || errno != ENOSYS)
		628	return res;
		629	#endif
		630
		631	/* even though we could play tricks with the flags, it's better to always
		632	* call fdatasync, as that matches the expectation of its users best */
		633	return fdatasync (fd);
		634	}
		635
		636	static int
		637	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		638	{
		639	#if HAVE_LINUX_FALLOCATE
		640	return fallocate (fd, mode, offset, len);
		641	#else
		642	return EIO_ENOSYS ();
		643	#endif
		644	}
568		645
569	#if !HAVE_READAHEAD	646	#if !HAVE_READAHEAD
		647	# undef readahead
570	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	648	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
571		649
572	static ssize_t	650	static eio_ssize_t
573	eio__readahead (int fd, off_t offset, size_t count, worker *self)	651	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
574	{	652	{
575	size_t todo = count;	653	size_t todo = count;
576	dBUF;	654	dBUF;
577		655
578	while (todo > 0)	656	while (todo > 0)
…		…
582	pread (fd, eio_buf, len, offset);	660	pread (fd, eio_buf, len, offset);
583	offset += len;	661	offset += len;
584	todo -= len;	662	todo -= len;
585	}	663	}
586		664
587	errno = 0;	665	FUBd;
		666
		667	/* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
		668	/* but not for e.g. EIO or eof, so we also never fail */
588	return count;	669	return 0;
589	}	670	}
590		671
591	#endif
592
593	#if !HAVE_READDIR_R
594	# define readdir_r eio__readdir_r
595
596	static mutex_t readdirlock = X_MUTEX_INIT;
597
598	static int
599	eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
600	{
601	EIO_STRUCT_DIRENT *e;
602	int errorno;
603
604	X_LOCK (readdirlock);
605
606	e = readdir (dirp);
607	errorno = errno;
608
609	if (e)
610	{
611	*res = ent;
612	strcpy (ent->d_name, e->d_name);
613	}
614	else
615	*res = 0;
616
617	X_UNLOCK (readdirlock);
618
619	errno = errorno;
620	return e ? 0 : -1;
621	}
622	#endif	672	#endif
623		673
624	/* sendfile always needs emulation */	674	/* sendfile always needs emulation */
625	static ssize_t	675	static eio_ssize_t
626	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self)	676	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
627	{	677	{
		678	eio_ssize_t written = 0;
628	ssize_t res;	679	eio_ssize_t res;
629		680
630	if (!count)	681	if (!count)
631	return 0;	682	return 0;
632		683
		684	for (;;)
		685	{
		686	#ifdef __APPLE__
		687	# undef HAVE_SENDFILE /* broken, as everything on os x */
		688	#endif
633	#if HAVE_SENDFILE	689	#if HAVE_SENDFILE
634	# if __linux	690	# if __linux
		691	off_t soffset = offset;
635	res = sendfile (ofd, ifd, &offset, count);	692	res = sendfile (ofd, ifd, &soffset, count);
636		693
637	# elif __freebsd	694	# elif __FreeBSD__
638	/*	695	/*
639	* Of course, the freebsd sendfile is a dire hack with no thoughts	696	* Of course, the freebsd sendfile is a dire hack with no thoughts
640	* wasted on making it similar to other I/O functions.	697	* wasted on making it similar to other I/O functions.
641	*/	698	*/
642	{
643	off_t sbytes;	699	off_t sbytes;
644	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	700	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
645		701
646	if (res < 0 && sbytes)	702	#if 0 /* according to the manpage, this is correct, but broken behaviour */
647	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	703	/* freebsd' sendfile will return 0 on success */
		704	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		705	/* not on e.g. EIO or EPIPE - sounds broken */
		706	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
648	res = sbytes;	707	res = sbytes;
649	}	708	#endif
		709
		710	/* according to source inspection, this is correct, and useful behaviour */
		711	if (sbytes)
		712	res = sbytes;
		713
		714	# elif defined __APPLE__
		715	off_t sbytes = count;
		716	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		717
		718	/* according to the manpage, sbytes is always valid */
		719	if (sbytes)
		720	res = sbytes;
650		721
651	# elif __hpux	722	# elif __hpux
652	res = sendfile (ofd, ifd, offset, count, 0, 0);	723	res = sendfile (ofd, ifd, offset, count, 0, 0);
653		724
654	# elif __solaris	725	# elif __solaris
655	{
656	struct sendfilevec vec;	726	struct sendfilevec vec;
657	size_t sbytes;	727	size_t sbytes;
658		728
659	vec.sfv_fd = ifd;	729	vec.sfv_fd = ifd;
660	vec.sfv_flag = 0;	730	vec.sfv_flag = 0;
661	vec.sfv_off = offset;	731	vec.sfv_off = offset;
662	vec.sfv_len = count;	732	vec.sfv_len = count;
663		733
664	res = sendfilev (ofd, &vec, 1, &sbytes);	734	res = sendfilev (ofd, &vec, 1, &sbytes);
665		735
666	if (res < 0 && sbytes)	736	if (res < 0 && sbytes)
667	res = sbytes;	737	res = sbytes;
668	}
669		738
670	# endif	739	# endif
		740
		741	#elif defined (_WIN32) && 0
		742	/* does not work, just for documentation of what would need to be done */
		743	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		744	/* libeio guarantees that the file offset does not change, and windows */
		745	/* has no way to get an independent handle to the same file description */
		746	HANDLE h = TO_SOCKET (ifd);
		747	SetFilePointer (h, offset, 0, FILE_BEGIN);
		748	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		749
671	#else	750	#else
672	res = -1;	751	res = EIO_ENOSYS ();
673	errno = ENOSYS;
674	#endif	752	#endif
675		753
		754	/* we assume sendfile can copy at least 128mb in one go */
		755	if (res <= 128 * 1024 * 1024)
		756	{
		757	if (res > 0)
		758	written += res;
		759
		760	if (written)
		761	return written;
		762
		763	break;
		764	}
		765	else
		766	{
		767	/* if we requested more, then probably the kernel was lazy */
		768	written += res;
		769	offset += res;
		770	count -= res;
		771
		772	if (!count)
		773	return written;
		774	}
		775	}
		776
676	if (res < 0	777	if (res < 0
677	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	778	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		779	/* BSDs */
		780	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		781	|| errno == ENOTSUP
		782	#endif
		783	#ifdef EOPNOTSUPP /* windows */
		784	|| errno == EOPNOTSUPP /* BSDs */
		785	#endif
678	#if __solaris	786	#if __solaris
679	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	787	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
680	#endif	788	#endif
681	)	789	)
682	)	790	)
…		…
686		794
687	res = 0;	795	res = 0;
688		796
689	while (count)	797	while (count)
690	{	798	{
691	ssize_t cnt;	799	eio_ssize_t cnt;
692		800
693	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	801	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
694		802
695	if (cnt <= 0)	803	if (cnt <= 0)
696	{	804	{
…		…
708		816
709	offset += cnt;	817	offset += cnt;
710	res += cnt;	818	res += cnt;
711	count -= cnt;	819	count -= cnt;
712	}	820	}
		821
		822	FUBd;
713	}	823	}
714		824
715	return res;	825	return res;
		826	}
		827
		828	#ifdef PAGESIZE
		829	# define eio_pagesize() PAGESIZE
		830	#else
		831	static intptr_t
		832	eio_pagesize (void)
		833	{
		834	static intptr_t page;
		835
		836	if (!page)
		837	page = sysconf (_SC_PAGESIZE);
		838
		839	return page;
		840	}
		841	#endif
		842
		843	static void
		844	eio_page_align (void **addr, size_t *length)
		845	{
		846	intptr_t mask = eio_pagesize () - 1;
		847
		848	/* round down addr */
		849	intptr_t adj = mask & (intptr_t)*addr;
		850
		851	*addr = (void *)((intptr_t)*addr - adj);
		852	*length += adj;
		853
		854	/* round up length */
		855	*length = (*length + mask) & ~mask;
		856	}
		857
		858	#if !_POSIX_MEMLOCK
		859	# define eio__mlockall(a) EIO_ENOSYS ()
		860	#else
		861
		862	static int
		863	eio__mlockall (int flags)
		864	{
		865	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		866	extern int mallopt (int, int);
		867	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		868	#endif
		869
		870	if (EIO_MCL_CURRENT != MCL_CURRENT
		871	|| EIO_MCL_FUTURE != MCL_FUTURE)
		872	{
		873	flags = 0
		874	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		875	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		876	}
		877
		878	return mlockall (flags);
		879	}
		880	#endif
		881
		882	#if !_POSIX_MEMLOCK_RANGE
		883	# define eio__mlock(a,b) EIO_ENOSYS ()
		884	#else
		885
		886	static int
		887	eio__mlock (void *addr, size_t length)
		888	{
		889	eio_page_align (&addr, &length);
		890
		891	return mlock (addr, length);
		892	}
		893
		894	#endif
		895
		896	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		897	# define eio__msync(a,b,c) EIO_ENOSYS ()
		898	#else
		899
		900	static int
		901	eio__msync (void *mem, size_t len, int flags)
		902	{
		903	eio_page_align (&mem, &len);
		904
		905	if (EIO_MS_ASYNC != MS_SYNC
		906	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		907	|| EIO_MS_SYNC != MS_SYNC)
		908	{
		909	flags = 0
		910	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		911	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		912	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		913	}
		914
		915	return msync (mem, len, flags);
		916	}
		917
		918	#endif
		919
		920	static int
		921	eio__mtouch (eio_req *req)
		922	{
		923	void *mem = req->ptr2;
		924	size_t len = req->size;
		925	int flags = req->int1;
		926
		927	eio_page_align (&mem, &len);
		928
		929	{
		930	intptr_t addr = (intptr_t)mem;
		931	intptr_t end = addr + len;
		932	intptr_t page = eio_pagesize ();
		933
		934	if (addr < end)
		935	if (flags & EIO_MT_MODIFY) /* modify */
		936	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		937	else
		938	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		939	}
		940
		941	return 0;
		942	}
		943
		944	/*****************************************************************************/
		945	/* requests implemented outside eio_execute, because they are so large */
		946
		947	static void
		948	eio__lseek (eio_req *req)
		949	{
		950	/* this usually gets optimised away completely, or your compiler sucks, */
		951	/* or the whence constants really are not 0, 1, 2 */
		952	int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
		953	: req->int2 == EIO_SEEK_CUR ? SEEK_CUR
		954	: req->int2 == EIO_SEEK_END ? SEEK_END
		955	: req->int2;
		956
		957	req->offs = lseek (req->int1, req->offs, whence);
		958	req->result = req->offs == (off_t)-1 ? -1 : 0;
		959	}
		960
		961	/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
		962	static int
		963	eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
		964	{
		965	char *res;
		966	const char *rel = path;
		967	char *tmp1, *tmp2;
		968	#if SYMLOOP_MAX > 32
		969	int symlinks = SYMLOOP_MAX;
		970	#else
		971	int symlinks = 32;
		972	#endif
		973
		974	errno = EINVAL;
		975	if (!rel)
		976	return -1;
		977
		978	errno = ENOENT;
		979	if (!*rel)
		980	return -1;
		981
		982	res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
		983	#ifdef _WIN32
		984	if (_access (rel, 4) != 0)
		985	return -1;
		986
		987	symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
		988
		989	errno = ENAMETOOLONG;
		990	if (symlinks >= PATH_MAX * 3)
		991	return -1;
		992
		993	errno = EIO;
		994	if (symlinks <= 0)
		995	return -1;
		996
		997	return symlinks;
		998
		999	#else
		1000	tmp1 = res + PATH_MAX;
		1001	tmp2 = tmp1 + PATH_MAX;
		1002
		1003	#if 0 /* disabled, the musl way to do things is just too racy */
		1004	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1005	/* on linux we may be able to ask the kernel */
		1006	{
		1007	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1008
		1009	if (fd >= 0)
		1010	{
		1011	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1012	req->result = readlink (tmp1, res, PATH_MAX);
		1013	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1014	close (fd);
		1015
		1016	if (req->result > 0)
		1017	goto done;
		1018	}
		1019	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1020	return -1;
		1021	}
		1022	#endif
		1023	#endif
		1024
		1025	if (*rel != '/')
		1026	{
		1027	int len;
		1028
		1029	errno = ENOENT;
		1030	if (wd == EIO_INVALID_WD)
		1031	return -1;
		1032
		1033	if (wd == EIO_CWD)
		1034	{
		1035	if (!getcwd (res, PATH_MAX))
		1036	return -1;
		1037
		1038	len = strlen (res);
		1039	}
		1040	else
		1041	memcpy (res, wd->str, len = wd->len);
		1042
		1043	if (res [1]) /* only use if not / */
		1044	res += len;
		1045	}
		1046
		1047	while (*rel)
		1048	{
		1049	eio_ssize_t len, linklen;
		1050	const char *beg = rel;
		1051
		1052	while (*rel && *rel != '/')
		1053	++rel;
		1054
		1055	len = rel - beg;
		1056
		1057	if (!len) /* skip slashes */
		1058	{
		1059	++rel;
		1060	continue;
		1061	}
		1062
		1063	if (beg [0] == '.')
		1064	{
		1065	if (len == 1)
		1066	continue; /* . - nop */
		1067
		1068	if (beg [1] == '.' && len == 2)
		1069	{
		1070	/* .. - back up one component, if possible */
		1071
		1072	while (res != tmpbuf->ptr)
		1073	if (*--res == '/')
		1074	break;
		1075
		1076	continue;
		1077	}
		1078	}
		1079
		1080	errno = ENAMETOOLONG;
		1081	if (res + 1 + len + 1 >= tmp1)
		1082	return -1;
		1083
		1084	/* copy one component */
		1085	*res = '/';
		1086	memcpy (res + 1, beg, len);
		1087
		1088	/* zero-terminate, for readlink */
		1089	res [len + 1] = 0;
		1090
		1091	/* now check if it's a symlink */
		1092	linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
		1093
		1094	if (linklen < 0)
		1095	{
		1096	if (errno != EINVAL)
		1097	return -1;
		1098
		1099	/* it's a normal directory. hopefully */
		1100	res += len + 1;
		1101	}
		1102	else
		1103	{
		1104	/* yay, it was a symlink - build new path in tmp2 */
		1105	int rellen = strlen (rel);
		1106
		1107	errno = ENAMETOOLONG;
		1108	if (linklen + 1 + rellen >= PATH_MAX)
		1109	return -1;
		1110
		1111	errno = ELOOP;
		1112	if (!--symlinks)
		1113	return -1;
		1114
		1115	if (*tmp1 == '/')
		1116	res = tmpbuf->ptr; /* symlink resolves to an absolute path */
		1117
		1118	/* we need to be careful, as rel might point into tmp2 already */
		1119	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1120	tmp2 [linklen] = '/';
		1121	memcpy (tmp2, tmp1, linklen);
		1122
		1123	rel = tmp2;
		1124	}
		1125	}
		1126
		1127	/* special case for the lone root path */
		1128	if (res == tmpbuf->ptr)
		1129	*res++ = '/';
		1130
		1131	return res - (char *)tmpbuf->ptr;
		1132	#endif
		1133	}
		1134
		1135	static signed char
		1136	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1137	{
		1138	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1139	: a->inode < b->inode ? -1
		1140	: a->inode > b->inode ? 1
		1141	: 0;
		1142	}
		1143
		1144	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1145
		1146	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1147	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1148
		1149	static void
		1150	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1151	{
		1152	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
		1153	unsigned char *bit = bits;
		1154
		1155	assert (CHAR_BIT == 8);
		1156	assert (sizeof (eio_dirent) * 8 < 256);
		1157	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1158	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1159
		1160	if (size <= EIO_SORT_FAST)
		1161	return;
		1162
		1163	/* first prepare an array of bits to test in our radix sort */
		1164	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
		1165	/* inode_bits must contain all inodes ORed together */
		1166	/* which is used to skip bits that are 0 everywhere, which is very common */
		1167	{
		1168	eio_ino_t endianness;
		1169	int i, j;
		1170
		1171	/* we store the byte offset of byte n into byte n of "endianness" */
		1172	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1173	((unsigned char *)&endianness)[i] = i;
		1174
		1175	*bit++ = 0;
		1176
		1177	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1178	{
		1179	/* shifting off the byte offsets out of "endianness" */
		1180	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1181	endianness >>= 8;
		1182
		1183	for (j = 0; j < 8; ++j)
		1184	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
		1185	*bit++ = offs + j;
		1186	}
		1187
		1188	for (j = 0; j < 8; ++j)
		1189	if (score_bits & (1 << j))
		1190	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1191	}
		1192
		1193	/* now actually do the sorting (a variant of MSD radix sort) */
		1194	{
		1195	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
		1196	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
		1197	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
		1198	int stk_idx = 0;
		1199
		1200	base_stk [stk_idx] = dents;
		1201	end_stk [stk_idx] = dents + size;
		1202	bit_stk [stk_idx] = bit - 1;
		1203
		1204	do
		1205	{
		1206	base = base_stk [stk_idx];
		1207	end = end_stk [stk_idx];
		1208	bit = bit_stk [stk_idx];
		1209
		1210	for (;;)
		1211	{
		1212	unsigned char O = *bit >> 3;
		1213	unsigned char M = 1 << (*bit & 7);
		1214
		1215	eio_dirent *a = base;
		1216	eio_dirent *b = end;
		1217
		1218	if (b - a < EIO_SORT_CUTOFF)
		1219	break;
		1220
		1221	/* now bit-partition the array on the bit */
		1222	/* this ugly asymmetric loop seems to perform much better than typical */
		1223	/* partition algos found in the literature */
		1224	do
		1225	if (!(((unsigned char *)a)[O] & M))
		1226	++a;
		1227	else if (!(((unsigned char *)--b)[O] & M))
		1228	{
		1229	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1230	++a;
		1231	}
		1232	while (b > a);
		1233
		1234	/* next bit, or stop, if no bits left in this path */
		1235	if (!*--bit)
		1236	break;
		1237
		1238	base_stk [stk_idx] = a;
		1239	end_stk [stk_idx] = end;
		1240	bit_stk [stk_idx] = bit;
		1241	++stk_idx;
		1242
		1243	end = a;
		1244	}
		1245	}
		1246	while (stk_idx--);
		1247	}
		1248	}
		1249
		1250	static void
		1251	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1252	{
		1253	/* first move the smallest element to the front, to act as a sentinel */
		1254	{
		1255	int i;
		1256	eio_dirent *min = dents;
		1257
		1258	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1259	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1260	if (EIO_DENT_CMP (dents [i], <, *min))
		1261	min = &dents [i];
		1262
		1263	/* swap elements 0 and j (minimum) */
		1264	{
		1265	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1266	}
		1267	}
		1268
		1269	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1270	{
		1271	eio_dirent *i, *j;
		1272
		1273	for (i = dents + 1; i < dents + size; ++i)
		1274	{
		1275	eio_dirent value = *i;
		1276
		1277	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1278	j [1] = j [0];
		1279
		1280	j [1] = value;
		1281	}
		1282	}
		1283	}
		1284
		1285	static void
		1286	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1287	{
		1288	if (size <= 1)
		1289	return; /* our insertion sort relies on size > 0 */
		1290
		1291	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1292	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1293	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1294
		1295	/* use an insertion sort at the end, or for small arrays, */
		1296	/* as insertion sort is more efficient for small partitions */
		1297	eio_dent_insertion_sort (dents, size);
716	}	1298	}
717		1299
718	/* read a full directory */	1300	/* read a full directory */
719	static void	1301	static void
720	eio__scandir (eio_req *req, worker *self)	1302	eio__scandir (eio_req *req, etp_worker *self)
721	{	1303	{
		1304	char *name, *names;
		1305	int namesalloc = 4096 - sizeof (void *) * 4;
		1306	int namesoffs = 0;
		1307	int flags = req->int1;
		1308	eio_dirent *dents = 0;
		1309	int dentalloc = 128;
		1310	int dentoffs = 0;
		1311	eio_ino_t inode_bits = 0;
		1312	#ifdef _WIN32
		1313	HANDLE dirp;
		1314	WIN32_FIND_DATA entp;
		1315	#else
722	DIR *dirp;	1316	DIR *dirp;
723	union
724	{
725	EIO_STRUCT_DIRENT d;
726	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
727	} *u;
728	EIO_STRUCT_DIRENT *entp;	1317	EIO_STRUCT_DIRENT *entp;
729	char *name, *names;	1318	#endif
730	int memlen = 4096;
731	int memofs = 0;
732	int res = 0;
733		1319
734	X_LOCK (wrklock);	1320	req->result = -1;
735	self->dirp = dirp = opendir (req->ptr1);
736	self->dbuf = u = malloc (sizeof (*u));
737	req->flags |= EIO_FLAG_PTR2_FREE;
738	req->ptr2 = names = malloc (memlen);
739	X_UNLOCK (wrklock);
740		1321
741	if (dirp && u && names)	1322	if (!(flags & EIO_READDIR_DENTS))
742	for (;;)	1323	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
		1324
		1325	#ifdef _WIN32
		1326	{
		1327	int len = strlen ((const char *)req->ptr1);
		1328	char *path = malloc (MAX_PATH);
		1329	const char *fmt;
		1330	const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		1331
		1332	if (!len)
		1333	fmt = "./*";
		1334	else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
		1335	fmt = "%s*";
		1336	else
		1337	fmt = "%s/*";
		1338
		1339	_snprintf (path, MAX_PATH, fmt, reqpath);
		1340	dirp = FindFirstFile (path, &entp);
		1341	free (path);
		1342
		1343	if (dirp == INVALID_HANDLE_VALUE)
		1344	{
		1345	/* should steal _dosmaperr */
		1346	switch (GetLastError ())
		1347	{
		1348	case ERROR_FILE_NOT_FOUND:
		1349	req->result = 0;
		1350	break;
		1351
		1352	case ERROR_INVALID_NAME:
		1353	case ERROR_PATH_NOT_FOUND:
		1354	case ERROR_NO_MORE_FILES:
		1355	errno = ENOENT;
		1356	break;
		1357
		1358	case ERROR_NOT_ENOUGH_MEMORY:
		1359	errno = ENOMEM;
		1360	break;
		1361
		1362	default:
		1363	errno = EINVAL;
		1364	break;
		1365	}
		1366
		1367	return;
		1368	}
		1369	}
		1370	#else
		1371	#if HAVE_AT
		1372	if (req->wd)
743	{	1373	{
		1374	int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1375
		1376	if (fd < 0)
		1377	return;
		1378
		1379	dirp = fdopendir (fd);
		1380
		1381	if (!dirp)
		1382	close (fd);
		1383	}
		1384	else
		1385	dirp = opendir (req->ptr1);
		1386	#else
		1387	dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
		1388	#endif
		1389
		1390	if (!dirp)
		1391	return;
		1392	#endif
		1393
		1394	if (req->flags & EIO_FLAG_PTR1_FREE)
		1395	free (req->ptr1);
		1396
		1397	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1398	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
		1399	req->ptr2 = names = malloc (namesalloc);
		1400
		1401	if (!names || (flags && !dents))
		1402	return;
		1403
		1404	for (;;)
		1405	{
		1406	int done;
		1407
		1408	#ifdef _WIN32
		1409	done = !dirp;
		1410	#else
744	errno = 0;	1411	errno = 0;
745	readdir_r (dirp, &u->d, &entp);	1412	entp = readdir (dirp);
		1413	done = !entp;
		1414	#endif
746		1415
		1416	if (done)
		1417	{
		1418	#ifndef _WIN32
		1419	int old_errno = errno;
		1420	closedir (dirp);
		1421	errno = old_errno;
		1422
747	if (!entp)	1423	if (errno)
		1424	break;
		1425	#endif
		1426
		1427	/* sort etc. */
		1428	req->int1 = flags;
		1429	req->result = dentoffs;
		1430
		1431	if (flags & EIO_READDIR_STAT_ORDER)
		1432	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1433	else if (flags & EIO_READDIR_DIRS_FIRST)
		1434	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1435	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1436	else
		1437	{
		1438	/* in this case, all is known, and we just put dirs first and sort them */
		1439	eio_dirent *oth = dents + dentoffs;
		1440	eio_dirent *dir = dents;
		1441
		1442	/* now partition dirs to the front, and non-dirs to the back */
		1443	/* by walking from both sides and swapping if necessary */
		1444	while (oth > dir)
		1445	{
		1446	if (dir->type == EIO_DT_DIR)
		1447	++dir;
		1448	else if ((--oth)->type == EIO_DT_DIR)
		1449	{
		1450	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1451
		1452	++dir;
		1453	}
		1454	}
		1455
		1456	/* now sort the dirs only (dirs all have the same score) */
		1457	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1458	}
		1459
748	break;	1460	break;
		1461	}
749		1462
750	name = entp->d_name;	1463	/* now add the entry to our list(s) */
		1464	name = D_NAME (entp);
751		1465
		1466	/* skip . and .. entries */
752	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1467	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
753	{	1468	{
754	int len = strlen (name) + 1;	1469	int len = D_NAMLEN (entp) + 1;
755		1470
756	res++;	1471	while (ecb_expect_false (namesoffs + len > namesalloc))
757
758	while (memofs + len > memlen)
759	{	1472	{
760	memlen *= 2;	1473	namesalloc *= 2;
761	X_LOCK (wrklock);
762	req->ptr2 = names = realloc (names, memlen);	1474	req->ptr2 = names = realloc (names, namesalloc);
763	X_UNLOCK (wrklock);
764		1475
765	if (!names)	1476	if (!names)
		1477	break;
		1478	}
		1479
		1480	memcpy (names + namesoffs, name, len);
		1481
		1482	if (dents)
		1483	{
		1484	struct eio_dirent *ent;
		1485
		1486	if (ecb_expect_false (dentoffs == dentalloc))
		1487	{
		1488	dentalloc *= 2;
		1489	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1490
		1491	if (!dents)
766	break;	1492	break;
767	}	1493	}
768		1494
769	memcpy (names + memofs, name, len);	1495	ent = dents + dentoffs;
770	memofs += len;	1496
		1497	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1498	ent->namelen = len - 1;
		1499	ent->inode = D_INO (entp);
		1500
		1501	inode_bits |= ent->inode;
		1502
		1503	switch (D_TYPE (entp))
		1504	{
		1505	default:
		1506	ent->type = EIO_DT_UNKNOWN;
		1507	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1508	break;
		1509
		1510	#ifdef DT_FIFO
		1511	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1512	#endif
		1513	#ifdef DT_CHR
		1514	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1515	#endif
		1516	#ifdef DT_MPC
		1517	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1518	#endif
		1519	#ifdef DT_DIR
		1520	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1521	#endif
		1522	#ifdef DT_NAM
		1523	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1524	#endif
		1525	#ifdef DT_BLK
		1526	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1527	#endif
		1528	#ifdef DT_MPB
		1529	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1530	#endif
		1531	#ifdef DT_REG
		1532	case DT_REG: ent->type = EIO_DT_REG; break;
		1533	#endif
		1534	#ifdef DT_NWK
		1535	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1536	#endif
		1537	#ifdef DT_CMP
		1538	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1539	#endif
		1540	#ifdef DT_LNK
		1541	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1542	#endif
		1543	#ifdef DT_SOCK
		1544	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1545	#endif
		1546	#ifdef DT_DOOR
		1547	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1548	#endif
		1549	#ifdef DT_WHT
		1550	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1551	#endif
		1552	}
		1553
		1554	ent->score = 7;
		1555
		1556	if (flags & EIO_READDIR_DIRS_FIRST)
		1557	{
		1558	if (ent->type == EIO_DT_UNKNOWN)
		1559	{
		1560	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1561	ent->score = 1;
		1562	else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
		1563	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1564	}
		1565	else if (ent->type == EIO_DT_DIR)
		1566	ent->score = 0;
		1567	}
771	}	1568	}
		1569
		1570	namesoffs += len;
		1571	++dentoffs;
772	}	1572	}
773		1573
774	if (errno)	1574	if (EIO_CANCELLED (req))
775	res = -1;	1575	{
		1576	errno = ECANCELED;
		1577	break;
		1578	}
		1579
		1580	#ifdef _WIN32
		1581	if (!FindNextFile (dirp, &entp))
		1582	{
		1583	FindClose (dirp);
		1584	dirp = 0;
		1585	}
		1586	#endif
		1587	}
		1588	}
		1589
		1590	/*****************************************************************************/
		1591	/* working directory stuff */
		1592	/* various deficiencies in the posix 2008 api force us to */
		1593	/* keep the absolute path in string form at all times */
		1594	/* fuck yeah. */
		1595
		1596	#if !HAVE_AT
		1597
		1598	/* a bit like realpath, but usually faster because it doesn'T have to return */
		1599	/* an absolute or canonical path */
		1600	static const char *
		1601	wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1602	{
		1603	if (!wd || *path == '/')
		1604	return path;
		1605
		1606	if (path [0] == '.' && !path [1])
		1607	return wd->str;
		1608
776		1609	{
777	req->result = res;	1610	int l1 = wd->len;
		1611	int l2 = strlen (path);
		1612
		1613	char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
		1614
		1615	memcpy (res, wd->str, l1);
		1616	res [l1] = '/';
		1617	memcpy (res + l1 + 1, path, l2 + 1);
		1618
		1619	return res;
		1620	}
778	}	1621	}
		1622
		1623	#endif
		1624
		1625	static eio_wd
		1626	eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1627	{
		1628	int fd;
		1629	eio_wd res;
		1630	int len = eio__realpath (tmpbuf, wd, path);
		1631
		1632	if (len < 0)
		1633	return EIO_INVALID_WD;
		1634
		1635	#if HAVE_AT
		1636	fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1637
		1638	if (fd < 0)
		1639	return EIO_INVALID_WD;
		1640	#endif
		1641
		1642	res = malloc (sizeof (*res) + len); /* one extra 0-byte */
		1643
		1644	#if HAVE_AT
		1645	res->fd = fd;
		1646	#endif
		1647
		1648	res->len = len;
		1649	memcpy (res->str, tmpbuf->ptr, len);
		1650	res->str [len] = 0;
		1651
		1652	return res;
		1653	}
		1654
		1655	eio_wd
		1656	eio_wd_open_sync (eio_wd wd, const char *path)
		1657	{
		1658	struct etp_tmpbuf tmpbuf = { };
		1659	wd = eio__wd_open_sync (&tmpbuf, wd, path);
		1660	free (tmpbuf.ptr);
		1661
		1662	return wd;
		1663	}
		1664
		1665	void
		1666	eio_wd_close_sync (eio_wd wd)
		1667	{
		1668	if (wd != EIO_INVALID_WD && wd != EIO_CWD)
		1669	{
		1670	#if HAVE_AT
		1671	close (wd->fd);
		1672	#endif
		1673	free (wd);
		1674	}
		1675	}
		1676
		1677	#if HAVE_AT
		1678
		1679	/* they forgot these */
		1680
		1681	static int
		1682	eio__truncateat (int dirfd, const char *path, off_t length)
		1683	{
		1684	int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
		1685	int res;
		1686
		1687	if (fd < 0)
		1688	return fd;
		1689
		1690	res = ftruncate (fd, length);
		1691	close (fd);
		1692	return res;
		1693	}
		1694
		1695	static int
		1696	eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
		1697	{
		1698	int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
		1699	int res;
		1700
		1701	if (fd < 0)
		1702	return fd;
		1703
		1704	res = fstatvfs (fd, buf);
		1705	close (fd);
		1706	return res;
		1707
		1708	}
		1709
		1710	#endif
779		1711
780	/*****************************************************************************/	1712	/*****************************************************************************/
781		1713
782	#define ALLOC(len) \	1714	#define ALLOC(len) \
783	if (!req->ptr2) \	1715	if (!req->ptr2) \
784	{ \	1716	{ \
785	X_LOCK (wrklock); \	1717	X_LOCK (EIO_POOL->wrklock); \
786	req->flags |= EIO_FLAG_PTR2_FREE; \	1718	req->flags |= EIO_FLAG_PTR2_FREE; \
787	X_UNLOCK (wrklock); \	1719	X_UNLOCK (EIO_POOL->wrklock); \
788	req->ptr2 = malloc (len); \	1720	req->ptr2 = malloc (len); \
789	if (!req->ptr2) \	1721	if (!req->ptr2) \
790	{ \	1722	{ \
791	errno = ENOMEM; \	1723	errno = ENOMEM; \
792	req->result = -1; \	1724	req->result = -1; \
793	break; \	1725	break; \
794	} \	1726	} \
795	}	1727	}
796		1728
797	X_THREAD_PROC (eio_proc)
798	{
799	eio_req *req;
800	struct timespec ts;
801	worker *self = (worker *)thr_arg;
802
803	/* try to distribute timeouts somewhat randomly */
804	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
805
806	for (;;)
807	{
808	ts.tv_sec = time (0) + IDLE_TIMEOUT;
809
810	X_LOCK (reqlock);
811
812	for (;;)
813	{
814	self->req = req = reqq_shift (&req_queue);
815
816	if (req)
817	break;
818
819	++idle;
820
821	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
822	{
823	if (idle > max_idle)
824	{
825	--idle;
826	X_UNLOCK (reqlock);
827	X_LOCK (wrklock);
828	--started;
829	X_UNLOCK (wrklock);
830	goto quit;
831	}
832
833	/* we are allowed to idle, so do so without any timeout */
834	X_COND_WAIT (reqwait, reqlock);
835	ts.tv_sec = time (0) + IDLE_TIMEOUT;
836	}
837
838	--idle;
839	}
840
841	--nready;
842
843	X_UNLOCK (reqlock);
844
845	errno = 0; /* strictly unnecessary */
846
847	if (!EIO_CANCELLED (req))
848	switch (req->type)
849	{
850	case EIO_READ: ALLOC (req->size);
851	req->result = req->offs >= 0
852	? pread (req->int1, req->ptr2, req->size, req->offs)
853	: read (req->int1, req->ptr2, req->size); break;
854	case EIO_WRITE: req->result = req->offs >= 0
855	? pwrite (req->int1, req->ptr2, req->size, req->offs)
856	: write (req->int1, req->ptr2, req->size); break;
857
858	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
859	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
860
861	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
862	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
863	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
864	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
865	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
866	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
867
868	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
869	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
870	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
871	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
872	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
873	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
874
875	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
876	case EIO_CLOSE: req->result = close (req->int1); break;
877	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
878	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
879	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
880	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
881	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
882	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
883	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
884	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
885
886	case EIO_READLINK: ALLOC (NAME_MAX);
887	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
888
889	case EIO_SYNC: req->result = 0; sync (); break;
890	case EIO_FSYNC: req->result = fsync (req->int1); break;
891	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
892
893	case EIO_READDIR: eio__scandir (req, self); break;
894
895	case EIO_BUSY:
896	#ifdef _WIN32
897	Sleep (req->nv1 * 1000.);
898	#else
899	{
900	struct timeval tv;
901
902	tv.tv_sec = req->nv1;
903	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
904
905	req->result = select (0, 0, 0, 0, &tv);
906	}
907	#endif
908	break;
909
910	case EIO_UTIME:
911	case EIO_FUTIME:
912	{
913	struct timeval tv[2];
914	struct timeval *times;
915
916	if (req->nv1 != -1. || req->nv2 != -1.)
917	{
918	tv[0].tv_sec = req->nv1;
919	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
920	tv[1].tv_sec = req->nv2;
921	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
922
923	times = tv;
924	}
925	else
926	times = 0;
927
928
929	req->result = req->type == EIO_FUTIME
930	? futimes (req->int1, times)
931	: utimes (req->ptr1, times);
932	}
933
934	case EIO_GROUP:
935	case EIO_NOP:
936	req->result = 0;
937	break;
938
939	case EIO_QUIT:
940	goto quit;
941
942	default:
943	req->result = -1;
944	break;
945	}
946
947	req->errorno = errno;
948
949	X_LOCK (reslock);
950
951	++npending;
952
953	if (!reqq_push (&res_queue, req) && want_poll_cb)
954	want_poll_cb ();
955
956	self->req = 0;
957	worker_clear (self);
958
959	X_UNLOCK (reslock);
960	}
961
962	quit:
963	X_LOCK (wrklock);
964	worker_free (self);
965	X_UNLOCK (wrklock);
966
967	return 0;
968	}
969
970	/*****************************************************************************/	1729	/*****************************************************************************/
971		1730
972	static void eio_atfork_prepare (void)	1731	int ecb_cold
973	{
974	X_LOCK (wrklock);
975	X_LOCK (reqlock);
976	X_LOCK (reslock);
977	#if !HAVE_PREADWRITE
978	X_LOCK (preadwritelock);
979	#endif
980	#if !HAVE_READDIR_R
981	X_LOCK (readdirlock);
982	#endif
983	}
984
985	static void eio_atfork_parent (void)
986	{
987	#if !HAVE_READDIR_R
988	X_UNLOCK (readdirlock);
989	#endif
990	#if !HAVE_PREADWRITE
991	X_UNLOCK (preadwritelock);
992	#endif
993	X_UNLOCK (reslock);
994	X_UNLOCK (reqlock);
995	X_UNLOCK (wrklock);
996	}
997
998	static void eio_atfork_child (void)
999	{
1000	eio_req *prv;
1001
1002	while (prv = reqq_shift (&req_queue))
1003	eio_destroy (prv);
1004
1005	while (prv = reqq_shift (&res_queue))
1006	eio_destroy (prv);
1007
1008	while (wrk_first.next != &wrk_first)
1009	{
1010	worker *wrk = wrk_first.next;
1011
1012	if (wrk->req)
1013	eio_destroy (wrk->req);
1014
1015	worker_clear (wrk);
1016	worker_free (wrk);
1017	}
1018
1019	started = 0;
1020	idle = 0;
1021	nreqs = 0;
1022	nready = 0;
1023	npending = 0;
1024
1025	eio_atfork_parent ();
1026	}
1027
1028	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1732	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1029	{	1733	{
1030	want_poll_cb = want_poll;	1734	eio_want_poll_cb = want_poll;
1031	done_poll_cb = done_poll;	1735	eio_done_poll_cb = done_poll;
1032		1736
1033	#ifdef _WIN32	1737	return etp_init (EIO_POOL, 0, 0, 0);
1034	X_MUTEX_CHECK (wrklock);
1035	X_MUTEX_CHECK (reslock);
1036	X_MUTEX_CHECK (reqlock);
1037	X_MUTEX_CHECK (reqwait);
1038	X_MUTEX_CHECK (preadwritelock);
1039	X_MUTEX_CHECK (readdirlock);
1040
1041	X_COND_CHECK (reqwait);
1042	#endif
1043
1044	X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1045	}	1738	}
1046		1739
		1740	ecb_inline void
1047	static void eio_api_destroy (eio_req *req)	1741	eio_api_destroy (eio_req *req)
1048	{	1742	{
1049	free (req);	1743	free (req);
1050	}	1744	}
1051		1745
1052	#define REQ(rtype) \	1746	#define REQ(rtype) \
1053	eio_req *req; \	1747	eio_req *req; \
1054	\	1748	\
1055	req = (eio_req *)calloc (1, sizeof *req); \	1749	req = (eio_req *)calloc (1, sizeof *req); \
1056	if (!req) \	1750	if (!req) \
1057	return 0; \	1751	return 0; \
…		…
1071	{ \	1765	{ \
1072	eio_api_destroy (req); \	1766	eio_api_destroy (req); \
1073	return 0; \	1767	return 0; \
1074	}	1768	}
1075		1769
		1770	#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
		1771
		1772	static void
		1773	eio_execute (etp_worker *self, eio_req *req)
		1774	{
		1775	#if HAVE_AT
		1776	int dirfd;
		1777	#else
		1778	const char *path;
		1779	#endif
		1780
		1781	if (ecb_expect_false (EIO_CANCELLED (req)))
		1782	{
		1783	req->result = -1;
		1784	req->errorno = ECANCELED;
		1785	return;
		1786	}
		1787
		1788	if (ecb_expect_false (req->wd == EIO_INVALID_WD))
		1789	{
		1790	req->result = -1;
		1791	req->errorno = ENOENT;
		1792	return;
		1793	}
		1794
		1795	if (req->type >= EIO_OPEN)
		1796	{
		1797	#if HAVE_AT
		1798	dirfd = WD2FD (req->wd);
		1799	#else
		1800	path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		1801	#endif
		1802	}
		1803
		1804	switch (req->type)
		1805	{
		1806	case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
		1807	req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
		1808	break;
		1809	case EIO_WD_CLOSE: req->result = 0;
		1810	eio_wd_close_sync (req->wd); break;
		1811
		1812	case EIO_SEEK: eio__lseek (req); break;
		1813	case EIO_READ: ALLOC (req->size);
		1814	req->result = req->offs >= 0
		1815	? pread (req->int1, req->ptr2, req->size, req->offs)
		1816	: read (req->int1, req->ptr2, req->size); break;
		1817	case EIO_WRITE: req->result = req->offs >= 0
		1818	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		1819	: write (req->int1, req->ptr2, req->size); break;
		1820
		1821	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		1822	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
		1823
		1824	#if HAVE_AT
		1825
		1826	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1827	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
		1828	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1829	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
		1830	case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
		1831	case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
		1832	case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
		1833	case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
		1834
		1835	case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
		1836	case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
		1837	req->result = req->wd && SINGLEDOT (req->ptr1)
		1838	? rmdir (req->wd->str)
		1839	: unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
		1840	case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
		1841	case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
		1842	req->result = req->wd && SINGLEDOT (req->ptr1)
		1843	? rename (req->wd->str, req->ptr2)
		1844	: renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
		1845	case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
		1846	case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
		1847	case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		1848	case EIO_READLINK: ALLOC (PATH_MAX);
		1849	req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
		1850	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1851	req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1852	case EIO_UTIME:
		1853	case EIO_FUTIME:
		1854	{
		1855	struct timespec ts[2];
		1856	struct timespec *times;
		1857
		1858	if (req->nv1 != -1. || req->nv2 != -1.)
		1859	{
		1860	ts[0].tv_sec = req->nv1;
		1861	ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
		1862	ts[1].tv_sec = req->nv2;
		1863	ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
		1864
		1865	times = ts;
		1866	}
		1867	else
		1868	times = 0;
		1869
		1870	req->result = req->type == EIO_FUTIME
		1871	? futimens (req->int1, times)
		1872	: utimensat (dirfd, req->ptr1, times, 0);
		1873	}
		1874	break;
		1875
		1876	#else
		1877
		1878	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1879	req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		1880	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1881	req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		1882	case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
		1883	case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
		1884	case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
		1885	case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
		1886
		1887	case EIO_UNLINK: req->result = unlink (path ); break;
		1888	case EIO_RMDIR: req->result = rmdir (path ); break;
		1889	case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
		1890	case EIO_RENAME: req->result = rename (path , req->ptr2); break;
		1891	case EIO_LINK: req->result = link (path , req->ptr2); break;
		1892	case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
		1893	case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
		1894	case EIO_READLINK: ALLOC (PATH_MAX);
		1895	req->result = readlink (path, req->ptr2, PATH_MAX); break;
		1896	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1897	req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1898
		1899	case EIO_UTIME:
		1900	case EIO_FUTIME:
		1901	{
		1902	struct timeval tv[2];
		1903	struct timeval *times;
		1904
		1905	if (req->nv1 != -1. || req->nv2 != -1.)
		1906	{
		1907	tv[0].tv_sec = req->nv1;
		1908	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
		1909	tv[1].tv_sec = req->nv2;
		1910	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
		1911
		1912	times = tv;
		1913	}
		1914	else
		1915	times = 0;
		1916
		1917	req->result = req->type == EIO_FUTIME
		1918	? futimes (req->int1, times)
		1919	: utimes (req->ptr1, times);
		1920	}
		1921	break;
		1922
		1923	#endif
		1924
		1925	case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
		1926	{
		1927	ALLOC (req->result);
		1928	memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
		1929	}
		1930	break;
		1931
		1932	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1933	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		1934
		1935	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1936	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1937
		1938	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		1939	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		1940	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		1941
		1942	case EIO_CLOSE: req->result = close (req->int1); break;
		1943	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		1944	case EIO_SYNC: req->result = 0; sync (); break;
		1945	case EIO_FSYNC: req->result = fsync (req->int1); break;
		1946	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1947	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		1948	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		1949	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		1950	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		1951	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		1952	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		1953	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
		1954
		1955	case EIO_READDIR: eio__scandir (req, self); break;
		1956
		1957	case EIO_BUSY:
		1958	#ifdef _WIN32
		1959	Sleep (req->nv1 * 1e3);
		1960	#else
		1961	{
		1962	struct timeval tv;
		1963
		1964	tv.tv_sec = req->nv1;
		1965	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
		1966
		1967	req->result = select (0, 0, 0, 0, &tv);
		1968	}
		1969	#endif
		1970	break;
		1971
		1972	#if 0
		1973	case EIO_GROUP:
		1974	abort (); /* handled in eio_request */
		1975	#endif
		1976
		1977	case EIO_NOP:
		1978	req->result = 0;
		1979	break;
		1980
		1981	case EIO_CUSTOM:
		1982	req->feed (req);
		1983	break;
		1984
		1985	default:
		1986	req->result = EIO_ENOSYS ();
		1987	break;
		1988	}
		1989
		1990	req->errorno = errno;
		1991	}
		1992
1076	#ifndef EIO_NO_WRAPPERS	1993	#ifndef EIO_NO_WRAPPERS
1077		1994
		1995	eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
		1996	{
		1997	REQ (EIO_WD_OPEN); PATH; SEND;
		1998	}
		1999
		2000	eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
		2001	{
		2002	REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
		2003	}
		2004
1078	eio_req *eio_nop (int pri, eio_cb cb, void *data)	2005	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1079	{	2006	{
1080	REQ (EIO_NOP); SEND;	2007	REQ (EIO_NOP); SEND;
1081	}	2008	}
1082		2009
…		…
1093	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	2020	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1094	{	2021	{
1095	REQ (EIO_FSYNC); req->int1 = fd; SEND;	2022	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1096	}	2023	}
1097		2024
		2025	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2026	{
		2027	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2028	}
		2029
1098	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2030	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1099	{	2031	{
1100	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2032	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1101	}	2033	}
1102		2034
		2035	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2036	{
		2037	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2038	}
		2039
		2040	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2041	{
		2042	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2043	}
		2044
		2045	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2046	{
		2047	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2048	}
		2049
		2050	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2051	{
		2052	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2053	}
		2054
		2055	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2056	{
		2057	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2058	}
		2059
		2060	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2061	{
		2062	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2063	}
		2064
1103	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2065	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1104	{	2066	{
1105	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2067	REQ (EIO_CLOSE); req->int1 = fd; SEND;
1106	}	2068	}
1107		2069
1108	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)	2070	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1109	{	2071	{
1110	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;	2072	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1111	}	2073	}
1112		2074
		2075	eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
		2076	{
		2077	REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
		2078	}
		2079
1113	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)	2080	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1114	{	2081	{
1115	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;	2082	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1116	}	2083	}
1117		2084
…		…
1123	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2090	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1124	{	2091	{
1125	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2092	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1126	}	2093	}
1127		2094
		2095	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2096	{
		2097	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2098	}
		2099
1128	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2100	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1129	{	2101	{
1130	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2102	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1131	}	2103	}
1132		2104
…		…
1138	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2110	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1139	{	2111	{
1140	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2112	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1141	}	2113	}
1142		2114
1143	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2115	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1144	{	2116	{
1145	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2117	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1146	}	2118	}
1147		2119
1148	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2120	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1168	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2140	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1169	{	2141	{
1170	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2142	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1171	}	2143	}
1172		2144
1173	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2145	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1174	{	2146	{
1175	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2147	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1176	}	2148	}
1177		2149
1178	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2150	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1194	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2166	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1195	{	2167	{
1196	return eio__1path (EIO_READLINK, path, pri, cb, data);	2168	return eio__1path (EIO_READLINK, path, pri, cb, data);
1197	}	2169	}
1198		2170
		2171	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2172	{
		2173	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2174	}
		2175
1199	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2176	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1200	{	2177	{
1201	return eio__1path (EIO_STAT, path, pri, cb, data);	2178	return eio__1path (EIO_STAT, path, pri, cb, data);
1202	}	2179	}
1203		2180
1204	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2181	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1205	{	2182	{
1206	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2183	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1207	}	2184	}
1208		2185
		2186	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2187	{
		2188	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2189	}
		2190
1209	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2191	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1210	{	2192	{
1211	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2193	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1212	}	2194	}
1213		2195
1214	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	2196	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1215	{	2197	{
1216	return eio__1path (EIO_RMDIR, path, pri, cb, data);	2198	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1217	}	2199	}
1218		2200
1219	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	2201	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1220	{	2202	{
1221	return eio__1path (EIO_READDIR, path, pri, cb, data);	2203	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1222	}	2204	}
1223		2205
1224	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2206	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1225	{	2207	{
1226	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	2208	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1227	}	2209	}
1228		2210
1229	static eio_req *	2211	static eio_req *
1230	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2212	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1231	{	2213	{
…		…
1255	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2237	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1256	{	2238	{
1257	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2239	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1258	}	2240	}
1259		2241
		2242	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
		2243	{
		2244	REQ (EIO_CUSTOM); req->feed = execute; SEND;
		2245	}
		2246
1260	#endif	2247	#endif
1261		2248
1262	eio_req *eio_grp (eio_cb cb, void *data)	2249	eio_req *eio_grp (eio_cb cb, void *data)
1263	{	2250	{
1264	const int pri = EIO_PRI_MAX;	2251	const int pri = EIO_PRI_MAX;
…		…
1271	#undef SEND	2258	#undef SEND
1272		2259
1273	/*****************************************************************************/	2260	/*****************************************************************************/
1274	/* grp functions */	2261	/* grp functions */
1275		2262
		2263	void
1276	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2264	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1277	{	2265	{
1278	grp->int2 = limit;	2266	grp->int2 = limit;
1279	grp->feed = feed;	2267	grp->feed = feed;
1280		2268
1281	grp_try_feed (grp);	2269	grp_try_feed (grp);
1282	}	2270	}
1283		2271
		2272	void
1284	void eio_grp_limit (eio_req *grp, int limit)	2273	eio_grp_limit (eio_req *grp, int limit)
1285	{	2274	{
1286	grp->int2 = limit;	2275	grp->int2 = limit;
1287		2276
1288	grp_try_feed (grp);	2277	grp_try_feed (grp);
1289	}	2278	}
1290		2279
		2280	void
1291	void eio_grp_add (eio_req *grp, eio_req *req)	2281	eio_grp_add (eio_req *grp, eio_req *req)
1292	{	2282	{
1293	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2283	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
		2284
		2285	grp->flags |= ETP_FLAG_GROUPADD;
1294		2286
1295	++grp->size;	2287	++grp->size;
1296	req->grp = grp;	2288	req->grp = grp;
1297		2289
1298	req->grp_prev = 0;	2290	req->grp_prev = 0;
…		…
1305	}	2297	}
1306		2298
1307	/*****************************************************************************/	2299	/*****************************************************************************/
1308	/* misc garbage */	2300	/* misc garbage */
1309		2301
		2302	eio_ssize_t
1310	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2303	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1311	{	2304	{
1312	worker wrk;
1313
1314	wrk.dbuf = 0;
1315
1316	eio__sendfile (ofd, ifd, offset, count, &wrk);	2305	return eio__sendfile (ofd, ifd, offset, count);
1317
1318	if (wrk.dbuf)
1319	free (wrk.dbuf);
1320	}	2306	}
1321		2307

Diff Legend

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