ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.10 by root, Tue May 13 17:08:15 2008 UTC vs.
Revision 1.126 by root, Fri Dec 28 07:33:41 2012 UTC

		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008,2009,2010,2011,2012 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
		126	#define fchmod(fd,mode) EIO_ENOSYS ()
		127	#define chown(path,uid,gid) EIO_ENOSYS ()
		128	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		129	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		130	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		131	#define mknod(path,mode,dev) EIO_ENOSYS ()
		132	#define sync() EIO_ENOSYS ()
		133	#define readlink(path,buf,s) EIO_ENOSYS ()
		134	#define statvfs(path,buf) EIO_ENOSYS ()
		135	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		136
		137	#define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
		138	#define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
		139
		140	#if __GNUC__
		141	typedef long long eio_off_t; /* signed for compatibility to msvc */
		142	#else
		143	typedef __int64 eio_off_t; /* unsigned not supported by msvc */
		144	#endif
		145
		146	static eio_ssize_t
		147	eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
		148	{
		149	OVERLAPPED o = { 0 };
		150	DWORD got;
		151
		152	o.Offset = offset;
		153	o.OffsetHigh = offset >> 32;
		154
		155	return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
		156	? got : -1;
		157	}
		158
		159	static eio_ssize_t
		160	eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
		161	{
		162	OVERLAPPED o = { 0 };
		163	DWORD got;
		164
		165	o.Offset = offset;
		166	o.OffsetHigh = offset >> 32;
		167
		168	return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
		169	? got : -1;
		170	}
		171
		172	/* rename() uses MoveFile, which fails to overwrite */
		173	#define rename(old,neu) eio__rename (old, neu)
		174
		175	static int
		176	eio__rename (const char *old, const char *neu)
		177	{
		178	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		179	return 0;
		180
		181	/* should steal _dosmaperr */
		182	switch (GetLastError ())
		183	{
		184	case ERROR_FILE_NOT_FOUND:
		185	case ERROR_PATH_NOT_FOUND:
		186	case ERROR_INVALID_DRIVE:
		187	case ERROR_NO_MORE_FILES:
		188	case ERROR_BAD_NETPATH:
		189	case ERROR_BAD_NET_NAME:
		190	case ERROR_BAD_PATHNAME:
		191	case ERROR_FILENAME_EXCED_RANGE:
		192	errno = ENOENT;
		193	break;
		194
		195	default:
		196	errno = EACCES;
		197	break;
		198	}
		199
		200	return -1;
		201	}
		202
		203	/* we could even stat and see if it exists */
		204	static int
		205	symlink (const char *old, const char *neu)
		206	{
		207	#if WINVER >= 0x0600
		208	if (CreateSymbolicLink (neu, old, 1))
		209	return 0;
		210
		211	if (CreateSymbolicLink (neu, old, 0))
		212	return 0;
		213	#endif
		214
		215	return EIO_ERRNO (ENOENT, -1);
		216	}
		217
		218	/* POSIX API only */
		219	#define CreateHardLink(neu,old,flags) 0
		220	#define CreateSymbolicLink(neu,old,flags) 0
		221
		222	struct statvfs
		223	{
		224	int dummy;
		225	};
		226
		227	#define DT_DIR EIO_DT_DIR
		228	#define DT_REG EIO_DT_REG
		229	#define D_NAME(entp) entp.cFileName
		230	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
30		231
31	#else	232	#else
32		233
33	# include "config.h"
34	# include <sys/time.h>	234	#include <sys/time.h>
35	# include <sys/select.h>	235	#include <sys/select.h>
		236	#include <sys/statvfs.h>
36	# include <unistd.h>	237	#include <unistd.h>
		238	#include <signal.h>
		239	#include <dirent.h>
		240
		241	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		242	#include <sys/mman.h>
		243	#endif
		244
		245	#define D_NAME(entp) entp->d_name
		246
		247	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		248	#if __FreeBSD__ || __NetBSD__ || __OpenBSD__
		249	#define _DIRENT_HAVE_D_TYPE /* sigh */
		250	#define D_INO(de) (de)->d_fileno
		251	#define D_NAMLEN(de) (de)->d_namlen
		252	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		253	#define D_INO(de) (de)->d_ino
		254	#endif
		255
		256	#ifdef _D_EXACT_NAMLEN
		257	#undef D_NAMLEN
		258	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		259	#endif
		260
		261	#ifdef _DIRENT_HAVE_D_TYPE
		262	#define D_TYPE(de) (de)->d_type
		263	#endif
		264
		265	#ifndef EIO_STRUCT_DIRENT
		266	#define EIO_STRUCT_DIRENT struct dirent
		267	#endif
		268
		269	#endif
		270
		271	#if HAVE_UTIMES
37	# include <utime.h>	272	# 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	273	#endif
44		274
		275	#if HAVE_SYS_SYSCALL_H
		276	# include <sys/syscall.h>
		277	#endif
		278
		279	#if HAVE_SYS_PRCTL_H
		280	# include <sys/prctl.h>
45	#endif	281	#endif
46		282
47	#if HAVE_SENDFILE	283	#if HAVE_SENDFILE
48	# if __linux	284	# if __linux
49	# include <sys/sendfile.h>	285	# include <sys/sendfile.h>
50	# elif __freebsd	286	# elif __FreeBSD__ || defined __APPLE__
51	# include <sys/socket.h>	287	# include <sys/socket.h>
52	# include <sys/uio.h>	288	# include <sys/uio.h>
53	# elif __hpux	289	# elif __hpux
54	# include <sys/socket.h>	290	# include <sys/socket.h>
55	# elif __solaris /* not yet */	291	# elif __solaris
56	# include <sys/sendfile.h>	292	# include <sys/sendfile.h>
57	# else	293	# else
58	# error sendfile support requested but not available	294	# error sendfile support requested but not available
59	# endif	295	# endif
60	#endif	296	#endif
61		297
62	/* number of seconds after which an idle threads exit */	298	#ifndef D_TYPE
63	#define IDLE_TIMEOUT 10	299	# define D_TYPE(de) 0
		300	#endif
		301	#ifndef D_INO
		302	# define D_INO(de) 0
		303	#endif
		304	#ifndef D_NAMLEN
		305	# define D_NAMLEN(entp) strlen (D_NAME (entp))
		306	#endif
64		307
65	/* used for struct dirent, AIX doesn't provide it */	308	/* used for struct dirent, AIX doesn't provide it */
66	#ifndef NAME_MAX	309	#ifndef NAME_MAX
67	# define NAME_MAX 4096	310	# define NAME_MAX 4096
68	#endif	311	#endif
69		312
		313	/* used for readlink etc. */
		314	#ifndef PATH_MAX
		315	# define PATH_MAX 4096
		316	#endif
		317
70	/* buffer size for various temporary buffers */	318	/* buffer size for various temporary buffers */
71	#define EIO_BUFSIZE 65536	319	#define EIO_BUFSIZE 65536
72		320
73	#define dBUF \	321	#define dBUF \
74	char *eio_buf; \
75	X_LOCK (wrklock); \
76	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	322	char *eio_buf = malloc (EIO_BUFSIZE); \
77	X_UNLOCK (wrklock); \
78	errno = ENOMEM; \	323	errno = ENOMEM; \
79	if (!eio_buf) \	324	if (!eio_buf) \
80	return -1;	325	return -1
		326
		327	#define FUBd \
		328	free (eio_buf)
81		329
82	#define EIO_TICKS ((1000000 + 1023) >> 10)	330	#define EIO_TICKS ((1000000 + 1023) >> 10)
		331
		332	/*****************************************************************************/
		333
		334	struct tmpbuf
		335	{
		336	void *ptr;
		337	int len;
		338	};
		339
		340	static void *
		341	tmpbuf_get (struct tmpbuf *buf, int len)
		342	{
		343	if (buf->len < len)
		344	{
		345	free (buf->ptr);
		346	buf->ptr = malloc (buf->len = len);
		347	}
		348
		349	return buf->ptr;
		350	}
		351
		352	struct tmpbuf;
		353
		354	#if _POSIX_VERSION >= 200809L
		355	#define HAVE_AT 1
		356	#define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
		357	#ifndef O_SEARCH
		358	#define O_SEARCH O_RDONLY
		359	#endif
		360	#else
		361	#define HAVE_AT 0
		362	static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
		363	#endif
		364
		365	struct eio_pwd
		366	{
		367	#if HAVE_AT
		368	int fd;
		369	#endif
		370	int len;
		371	char str[1]; /* actually, a 0-terminated canonical path */
		372	};
		373
		374	/*****************************************************************************/
		375
		376	#define ETP_PRI_MIN EIO_PRI_MIN
		377	#define ETP_PRI_MAX EIO_PRI_MAX
		378
		379	struct etp_worker;
		380
		381	#define ETP_REQ eio_req
		382	#define ETP_DESTROY(req) eio_destroy (req)
		383	static int eio_finish (eio_req *req);
		384	#define ETP_FINISH(req) eio_finish (req)
		385	static void eio_execute (struct etp_worker *self, eio_req *req);
		386	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		387
		388	/*****************************************************************************/
		389
		390	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		391
		392	/* calculate time difference in ~1/EIO_TICKS of a second */
		393	ecb_inline int
		394	tvdiff (struct timeval *tv1, struct timeval *tv2)
		395	{
		396	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
		397	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
		398	}
		399
		400	static unsigned int started, idle, wanted = 4;
83		401
84	static void (*want_poll_cb) (void);	402	static void (*want_poll_cb) (void);
85	static void (*done_poll_cb) (void);	403	static void (*done_poll_cb) (void);
86		404
87	static unsigned int max_poll_time = 0;	405	static unsigned int max_poll_time; /* reslock */
88	static unsigned int max_poll_reqs = 0;	406	static unsigned int max_poll_reqs; /* reslock */
89		407
90	/* calculcate time difference in ~1/EIO_TICKS of a second */	408	static unsigned int nreqs; /* reqlock */
91	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	409	static unsigned int nready; /* reqlock */
92	{	410	static unsigned int npending; /* reqlock */
93	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	411	static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
94	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	412	static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
95	}
96		413
97	static unsigned int started, idle, wanted = 4;	414	static xmutex_t wrklock;
		415	static xmutex_t reslock;
		416	static xmutex_t reqlock;
		417	static xcond_t reqwait;
		418
		419	typedef struct etp_worker
		420	{
		421	struct tmpbuf tmpbuf;
		422
		423	/* locked by wrklock */
		424	struct etp_worker *prev, *next;
		425
		426	xthread_t tid;
		427
		428	#ifdef ETP_WORKER_COMMON
		429	ETP_WORKER_COMMON
		430	#endif
		431	} etp_worker;
		432
		433	static etp_worker wrk_first; /* NOT etp */
		434
		435	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		436	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
98		437
99	/* worker threads management */	438	/* worker threads management */
100	static mutex_t wrklock = X_MUTEX_INIT;
101		439
102	typedef struct worker	440	static void
103	{
104	/* locked by wrklock */
105	struct worker *prev, *next;
106
107	thread_t tid;
108
109	/* locked by reslock, reqlock or wrklock */
110	eio_req *req; /* currently processed request */
111	void *dbuf;
112	DIR *dirp;
113	} worker;
114
115	static worker wrk_first = { &wrk_first, &wrk_first, 0 };
116
117	static void worker_clear (worker *wrk)	441	etp_worker_clear (etp_worker *wrk)
118	{	442	{
119	if (wrk->dirp)	443	}
120	{
121	closedir (wrk->dirp);
122	wrk->dirp = 0;
123	}
124		444
125	if (wrk->dbuf)	445	static void ecb_cold
126	{	446	etp_worker_free (etp_worker *wrk)
		447	{
127	free (wrk->dbuf);	448	free (wrk->tmpbuf.ptr);
128	wrk->dbuf = 0;
129	}
130	}
131		449
132	static void worker_free (worker *wrk)
133	{
134	wrk->next->prev = wrk->prev;	450	wrk->next->prev = wrk->prev;
135	wrk->prev->next = wrk->next;	451	wrk->prev->next = wrk->next;
136		452
137	free (wrk);	453	free (wrk);
138	}	454	}
139		455
140	static volatile unsigned int nreqs, nready, npending;	456	static unsigned int
141	static volatile unsigned int max_idle = 4;	457	etp_nreqs (void)
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	{	458	{
		459	int retval;
		460	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		461	retval = nreqs;
		462	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
149	return nreqs;	463	return retval;
150	}	464	}
151		465
152	unsigned int eio_nready (void)	466	static unsigned int
		467	etp_nready (void)
153	{	468	{
154	unsigned int retval;	469	unsigned int retval;
155		470
156	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	471	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157	retval = nready;	472	retval = nready;
158	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	473	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159		474
160	return retval;	475	return retval;
161	}	476	}
162		477
163	unsigned int eio_npending (void)	478	static unsigned int
		479	etp_npending (void)
164	{	480	{
165	unsigned int retval;	481	unsigned int retval;
166		482
167	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	483	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168	retval = npending;	484	retval = npending;
169	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	485	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170		486
171	return retval;	487	return retval;
172	}	488	}
173		489
174	unsigned int eio_nthreads (void)	490	static unsigned int
		491	etp_nthreads (void)
175	{	492	{
176	unsigned int retval;	493	unsigned int retval;
177		494
178	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	495	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179	retval = started;	496	retval = started;
…		…
186	* a somewhat faster data structure might be nice, but	503	* a somewhat faster data structure might be nice, but
187	* with 8 priorities this actually needs <20 insns	504	* with 8 priorities this actually needs <20 insns
188	* per shift, the most expensive operation.	505	* per shift, the most expensive operation.
189	*/	506	*/
190	typedef struct {	507	typedef struct {
191	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	508	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
192	int size;	509	int size;
193	} reqq;	510	} etp_reqq;
194		511
195	static reqq req_queue;	512	static etp_reqq req_queue;
196	static reqq res_queue;	513	static etp_reqq res_queue;
197		514
198	static int reqq_push (reqq *q, eio_req *req)	515	static void ecb_noinline ecb_cold
		516	reqq_init (etp_reqq *q)
		517	{
		518	int pri;
		519
		520	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		521	q->qs[pri] = q->qe[pri] = 0;
		522
		523	q->size = 0;
		524	}
		525
		526	static int ecb_noinline
		527	reqq_push (etp_reqq *q, ETP_REQ *req)
199	{	528	{
200	int pri = req->pri;	529	int pri = req->pri;
201	req->next = 0;	530	req->next = 0;
202		531
203	if (q->qe[pri])	532	if (q->qe[pri])
…		…
209	q->qe[pri] = q->qs[pri] = req;	538	q->qe[pri] = q->qs[pri] = req;
210		539
211	return q->size++;	540	return q->size++;
212	}	541	}
213		542
214	static eio_req *reqq_shift (reqq *q)	543	static ETP_REQ * ecb_noinline
		544	reqq_shift (etp_reqq *q)
215	{	545	{
216	int pri;	546	int pri;
217		547
218	if (!q->size)	548	if (!q->size)
219	return 0;	549	return 0;
220		550
221	--q->size;	551	--q->size;
222		552
223	for (pri = EIO_NUM_PRI; pri--; )	553	for (pri = ETP_NUM_PRI; pri--; )
224	{	554	{
225	eio_req *req = q->qs[pri];	555	eio_req *req = q->qs[pri];
226		556
227	if (req)	557	if (req)
228	{	558	{
…		…
234	}	564	}
235		565
236	abort ();	566	abort ();
237	}	567	}
238		568
239	static void grp_try_feed (eio_req *grp)	569	static int ecb_cold
		570	etp_init (void (*want_poll)(void), void (*done_poll)(void))
240	{	571	{
241	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	572	X_MUTEX_CREATE (wrklock);
242	{	573	X_MUTEX_CREATE (reslock);
243	int old_len = grp->size;	574	X_MUTEX_CREATE (reqlock);
		575	X_COND_CREATE (reqwait);
244		576
245	EIO_FEED (grp);	577	reqq_init (&req_queue);
		578	reqq_init (&res_queue);
246		579
247	/* stop if no progress has been made */	580	wrk_first.next =
248	if (old_len == grp->size)	581	wrk_first.prev = &wrk_first;
249	{
250	grp->feed = 0;
251	break;
252	}
253	}
254	}
255		582
256	static int eio_finish (eio_req *req);	583	started = 0;
		584	idle = 0;
		585	nreqs = 0;
		586	nready = 0;
		587	npending = 0;
257		588
258	static int grp_dec (eio_req *grp)	589	want_poll_cb = want_poll;
259	{	590	done_poll_cb = done_poll;
260	--grp->size;
261		591
262	/* call feeder, if applicable */
263	grp_try_feed (grp);
264
265	/* finish, if done */
266	if (!grp->size && grp->int1)
267	return eio_finish (grp);
268	else
269	return 0;	592	return 0;
270	}	593	}
271		594
272	void eio_destroy (eio_req *req)
273	{
274	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
275	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
276
277	EIO_DESTROY (req);
278	}
279
280	static int eio_finish (eio_req *req)
281	{
282	int res = EIO_FINISH (req);
283
284	if (req->grp)
285	{
286	int res2;
287	eio_req *grp = req->grp;
288
289	/* unlink request */
290	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
291	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
292
293	if (grp->grp_first == req)
294	grp->grp_first = req->grp_next;
295
296	res2 = grp_dec (grp);
297
298	if (!res && res2)
299	res = res2;
300	}
301
302	eio_destroy (req);
303
304	return res;
305	}
306
307	void eio_grp_cancel (eio_req *grp)
308	{
309	for (grp = grp->grp_first; grp; grp = grp->grp_next)
310	eio_cancel (grp);
311	}
312
313	void eio_cancel (eio_req *req)
314	{
315	X_LOCK (wrklock);
316	req->flags |= EIO_FLAG_CANCELLED;
317	X_UNLOCK (wrklock);
318
319	eio_grp_cancel (req);
320	}
321
322	X_THREAD_PROC (eio_proc);	595	X_THREAD_PROC (etp_proc);
323		596
324	static void start_thread (void)	597	static void ecb_cold
		598	etp_start_thread (void)
325	{	599	{
326	worker *wrk = calloc (1, sizeof (worker));	600	etp_worker *wrk = calloc (1, sizeof (etp_worker));
327		601
328	/*TODO*/	602	/*TODO*/
329	assert (("unable to allocate worker thread data", wrk));	603	assert (("unable to allocate worker thread data", wrk));
330		604
331	X_LOCK (wrklock);	605	X_LOCK (wrklock);
332		606
333	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	607	if (xthread_create (&wrk->tid, etp_proc, (void *)wrk))
334	{	608	{
335	wrk->prev = &wrk_first;	609	wrk->prev = &wrk_first;
336	wrk->next = wrk_first.next;	610	wrk->next = wrk_first.next;
337	wrk_first.next->prev = wrk;	611	wrk_first.next->prev = wrk;
338	wrk_first.next = wrk;	612	wrk_first.next = wrk;
…		…
342	free (wrk);	616	free (wrk);
343		617
344	X_UNLOCK (wrklock);	618	X_UNLOCK (wrklock);
345	}	619	}
346		620
		621	static void
347	static void maybe_start_thread (void)	622	etp_maybe_start_thread (void)
348	{	623	{
349	if (eio_nthreads () >= wanted)	624	if (ecb_expect_true (etp_nthreads () >= wanted))
350	return;	625	return;
351		626
352	/* todo: maybe use idle here, but might be less exact */	627	/* todo: maybe use idle here, but might be less exact */
353	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	628	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354	return;	629	return;
355		630
356	start_thread ();	631	etp_start_thread ();
357	}	632	}
358		633
359	void eio_submit (eio_req *req)	634	static void ecb_cold
		635	etp_end_thread (void)
360	{	636	{
361	++nreqs;
362
363	X_LOCK (reqlock);
364	++nready;
365	reqq_push (&req_queue, req);
366	X_COND_SIGNAL (reqwait);
367	X_UNLOCK (reqlock);
368
369	maybe_start_thread ();
370	}
371
372	static void end_thread (void)
373	{
374	eio_req *req = calloc (1, sizeof (eio_req));	637	eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
375		638
376	req->type = EIO_QUIT;	639	req->type = -1;
377	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	640	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
378		641
379	X_LOCK (reqlock);	642	X_LOCK (reqlock);
380	reqq_push (&req_queue, req);	643	reqq_push (&req_queue, req);
381	X_COND_SIGNAL (reqwait);	644	X_COND_SIGNAL (reqwait);
382	X_UNLOCK (reqlock);	645	X_UNLOCK (reqlock);
…		…
384	X_LOCK (wrklock);	647	X_LOCK (wrklock);
385	--started;	648	--started;
386	X_UNLOCK (wrklock);	649	X_UNLOCK (wrklock);
387	}	650	}
388		651
389	void eio_set_max_poll_time (double nseconds)	652	static int
390	{
391	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
392	max_poll_time = nseconds;
393	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
394	}
395
396	void eio_set_max_poll_reqs (unsigned int maxreqs)
397	{
398	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
399	max_poll_reqs = maxreqs;
400	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
401	}
402
403	void eio_set_max_idle (unsigned int nthreads)
404	{
405	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
406	max_idle = nthreads <= 0 ? 1 : nthreads;
407	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
408	}
409
410	void eio_set_min_parallel (unsigned int nthreads)
411	{
412	if (wanted < nthreads)
413	wanted = nthreads;
414	}
415
416	void eio_set_max_parallel (unsigned int nthreads)
417	{
418	if (wanted > nthreads)
419	wanted = nthreads;
420
421	while (started > wanted)
422	end_thread ();
423	}
424
425	int eio_poll (void)	653	etp_poll (void)
426	{	654	{
427	int maxreqs = max_poll_reqs;	655	unsigned int maxreqs;
		656	unsigned int maxtime;
428	struct timeval tv_start, tv_now;	657	struct timeval tv_start, tv_now;
429	eio_req *req;
430		658
		659	X_LOCK (reslock);
		660	maxreqs = max_poll_reqs;
		661	maxtime = max_poll_time;
		662	X_UNLOCK (reslock);
		663
431	if (max_poll_time)	664	if (maxtime)
432	gettimeofday (&tv_start, 0);	665	gettimeofday (&tv_start, 0);
433		666
434	for (;;)	667	for (;;)
435	{	668	{
		669	ETP_REQ *req;
		670
436	maybe_start_thread ();	671	etp_maybe_start_thread ();
437		672
438	X_LOCK (reslock);	673	X_LOCK (reslock);
439	req = reqq_shift (&res_queue);	674	req = reqq_shift (&res_queue);
440		675
441	if (req)	676	if (req)
…		…
449	X_UNLOCK (reslock);	684	X_UNLOCK (reslock);
450		685
451	if (!req)	686	if (!req)
452	return 0;	687	return 0;
453		688
		689	X_LOCK (reqlock);
454	--nreqs;	690	--nreqs;
		691	X_UNLOCK (reqlock);
455		692
456	if (req->type == EIO_GROUP && req->size)	693	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
457	{	694	{
458	req->int1 = 1; /* mark request as delayed */	695	req->int1 = 1; /* mark request as delayed */
459	continue;	696	continue;
460	}	697	}
461	else	698	else
462	{	699	{
463	int res = eio_finish (req);	700	int res = ETP_FINISH (req);
464	if (res)	701	if (ecb_expect_false (res))
465	return res;	702	return res;
466	}	703	}
467		704
468	if (maxreqs && !--maxreqs)	705	if (ecb_expect_false (maxreqs && !--maxreqs))
469	break;	706	break;
470		707
471	if (max_poll_time)	708	if (maxtime)
472	{	709	{
473	gettimeofday (&tv_now, 0);	710	gettimeofday (&tv_now, 0);
474		711
475	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	712	if (tvdiff (&tv_start, &tv_now) >= maxtime)
476	break;	713	break;
477	}	714	}
478	}	715	}
479		716
480	errno = EAGAIN;	717	errno = EAGAIN;
481	return -1;	718	return -1;
482	}	719	}
483		720
		721	static void
		722	etp_cancel (ETP_REQ *req)
		723	{
		724	req->cancelled = 1;
		725
		726	eio_grp_cancel (req);
		727	}
		728
		729	static void
		730	etp_submit (ETP_REQ *req)
		731	{
		732	req->pri -= ETP_PRI_MIN;
		733
		734	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		735	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		736
		737	if (ecb_expect_false (req->type == EIO_GROUP))
		738	{
		739	/* I hope this is worth it :/ */
		740	X_LOCK (reqlock);
		741	++nreqs;
		742	X_UNLOCK (reqlock);
		743
		744	X_LOCK (reslock);
		745
		746	++npending;
		747
		748	if (!reqq_push (&res_queue, req) && want_poll_cb)
		749	want_poll_cb ();
		750
		751	X_UNLOCK (reslock);
		752	}
		753	else
		754	{
		755	X_LOCK (reqlock);
		756	++nreqs;
		757	++nready;
		758	reqq_push (&req_queue, req);
		759	X_COND_SIGNAL (reqwait);
		760	X_UNLOCK (reqlock);
		761
		762	etp_maybe_start_thread ();
		763	}
		764	}
		765
		766	static void ecb_cold
		767	etp_set_max_poll_time (double nseconds)
		768	{
		769	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		770	max_poll_time = nseconds * EIO_TICKS;
		771	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		772	}
		773
		774	static void ecb_cold
		775	etp_set_max_poll_reqs (unsigned int maxreqs)
		776	{
		777	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		778	max_poll_reqs = maxreqs;
		779	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		780	}
		781
		782	static void ecb_cold
		783	etp_set_max_idle (unsigned int nthreads)
		784	{
		785	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		786	max_idle = nthreads;
		787	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		788	}
		789
		790	static void ecb_cold
		791	etp_set_idle_timeout (unsigned int seconds)
		792	{
		793	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		794	idle_timeout = seconds;
		795	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		796	}
		797
		798	static void ecb_cold
		799	etp_set_min_parallel (unsigned int nthreads)
		800	{
		801	if (wanted < nthreads)
		802	wanted = nthreads;
		803	}
		804
		805	static void ecb_cold
		806	etp_set_max_parallel (unsigned int nthreads)
		807	{
		808	if (wanted > nthreads)
		809	wanted = nthreads;
		810
		811	while (started > wanted)
		812	etp_end_thread ();
		813	}
		814
		815	/*****************************************************************************/
		816
		817	static void
		818	grp_try_feed (eio_req *grp)
		819	{
		820	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		821	{
		822	grp->flags &= ~EIO_FLAG_GROUPADD;
		823
		824	EIO_FEED (grp);
		825
		826	/* stop if no progress has been made */
		827	if (!(grp->flags & EIO_FLAG_GROUPADD))
		828	{
		829	grp->feed = 0;
		830	break;
		831	}
		832	}
		833	}
		834
		835	static int
		836	grp_dec (eio_req *grp)
		837	{
		838	--grp->size;
		839
		840	/* call feeder, if applicable */
		841	grp_try_feed (grp);
		842
		843	/* finish, if done */
		844	if (!grp->size && grp->int1)
		845	return eio_finish (grp);
		846	else
		847	return 0;
		848	}
		849
		850	static void
		851	eio_destroy (eio_req *req)
		852	{
		853	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		854	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		855
		856	EIO_DESTROY (req);
		857	}
		858
		859	static int
		860	eio_finish (eio_req *req)
		861	{
		862	int res = EIO_FINISH (req);
		863
		864	if (req->grp)
		865	{
		866	int res2;
		867	eio_req *grp = req->grp;
		868
		869	/* unlink request */
		870	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		871	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		872
		873	if (grp->grp_first == req)
		874	grp->grp_first = req->grp_next;
		875
		876	res2 = grp_dec (grp);
		877
		878	if (!res)
		879	res = res2;
		880	}
		881
		882	eio_destroy (req);
		883
		884	return res;
		885	}
		886
		887	void
		888	eio_grp_cancel (eio_req *grp)
		889	{
		890	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		891	eio_cancel (grp);
		892	}
		893
		894	void
		895	eio_cancel (eio_req *req)
		896	{
		897	etp_cancel (req);
		898	}
		899
		900	void
		901	eio_submit (eio_req *req)
		902	{
		903	etp_submit (req);
		904	}
		905
		906	unsigned int
		907	eio_nreqs (void)
		908	{
		909	return etp_nreqs ();
		910	}
		911
		912	unsigned int
		913	eio_nready (void)
		914	{
		915	return etp_nready ();
		916	}
		917
		918	unsigned int
		919	eio_npending (void)
		920	{
		921	return etp_npending ();
		922	}
		923
		924	unsigned int ecb_cold
		925	eio_nthreads (void)
		926	{
		927	return etp_nthreads ();
		928	}
		929
		930	void ecb_cold
		931	eio_set_max_poll_time (double nseconds)
		932	{
		933	etp_set_max_poll_time (nseconds);
		934	}
		935
		936	void ecb_cold
		937	eio_set_max_poll_reqs (unsigned int maxreqs)
		938	{
		939	etp_set_max_poll_reqs (maxreqs);
		940	}
		941
		942	void ecb_cold
		943	eio_set_max_idle (unsigned int nthreads)
		944	{
		945	etp_set_max_idle (nthreads);
		946	}
		947
		948	void ecb_cold
		949	eio_set_idle_timeout (unsigned int seconds)
		950	{
		951	etp_set_idle_timeout (seconds);
		952	}
		953
		954	void ecb_cold
		955	eio_set_min_parallel (unsigned int nthreads)
		956	{
		957	etp_set_min_parallel (nthreads);
		958	}
		959
		960	void ecb_cold
		961	eio_set_max_parallel (unsigned int nthreads)
		962	{
		963	etp_set_max_parallel (nthreads);
		964	}
		965
		966	int eio_poll (void)
		967	{
		968	return etp_poll ();
		969	}
		970
484	/*****************************************************************************/	971	/*****************************************************************************/
485	/* work around various missing functions */	972	/* work around various missing functions */
486		973
487	#if !HAVE_PREADWRITE
488	# define pread eio__pread
489	# define pwrite eio__pwrite
490
491	/*
492	* make our pread/pwrite safe against themselves, but not against
493	* normal read/write by using a mutex. slows down execution a lot,
494	* but that's your problem, not mine.
495	*/
496	static mutex_t preadwritelock = X_MUTEX_INIT;
497
498	static ssize_t
499	eio__pread (int fd, void *buf, size_t count, off_t offset)
500	{
501	ssize_t res;
502	off_t ooffset;
503
504	X_LOCK (preadwritelock);
505	ooffset = lseek (fd, 0, SEEK_CUR);
506	lseek (fd, offset, SEEK_SET);
507	res = read (fd, buf, count);
508	lseek (fd, ooffset, SEEK_SET);
509	X_UNLOCK (preadwritelock);
510
511	return res;
512	}
513
514	static ssize_t
515	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
516	{
517	ssize_t res;
518	off_t ooffset;
519
520	X_LOCK (preadwritelock);
521	ooffset = lseek (fd, 0, SEEK_CUR);
522	lseek (fd, offset, SEEK_SET);
523	res = write (fd, buf, count);
524	lseek (fd, offset, SEEK_SET);
525	X_UNLOCK (preadwritelock);
526
527	return res;
528	}
529	#endif
530
531	#ifndef HAVE_FUTIMES	974	#ifndef HAVE_UTIMES
532		975
		976	# undef utimes
533	# define utimes(path,times) eio__utimes (path, times)	977	# define utimes(path,times) eio__utimes (path, times)
534	# define futimes(fd,times) eio__futimes (fd, times)
535		978
536	static int	979	static int
537	eio__utimes (const char *filename, const struct timeval times[2])	980	eio__utimes (const char *filename, const struct timeval times[2])
538	{	981	{
539	if (times)	982	if (times)
…		…
547	}	990	}
548	else	991	else
549	return utime (filename, 0);	992	return utime (filename, 0);
550	}	993	}
551		994
		995	#endif
		996
		997	#ifndef HAVE_FUTIMES
		998
		999	# undef futimes
		1000	# define futimes(fd,times) eio__futimes (fd, times)
		1001
		1002	static int
552	static int eio__futimes (int fd, const struct timeval tv[2])	1003	eio__futimes (int fd, const struct timeval tv[2])
553	{	1004	{
554	errno = ENOSYS;	1005	errno = ENOSYS;
555	return -1;	1006	return -1;
556	}	1007	}
557		1008
558	#endif	1009	#endif
559		1010
560	#if !HAVE_FDATASYNC	1011	#if !HAVE_FDATASYNC
		1012	# undef fdatasync
561	# define fdatasync fsync	1013	# define fdatasync(fd) fsync (fd)
562	#endif	1014	#endif
		1015
		1016	static int
		1017	eio__syncfs (int fd)
		1018	{
		1019	int res;
		1020
		1021	#if HAVE_SYS_SYNCFS
		1022	res = (int)syscall (__NR_syncfs, (int)(fd));
		1023	#else
		1024	res = EIO_ENOSYS ();
		1025	#endif
		1026
		1027	if (res < 0 && errno == ENOSYS && fd >= 0)
		1028	sync ();
		1029
		1030	return res;
		1031	}
		1032
		1033	/* sync_file_range always needs emulation */
		1034	static int
		1035	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		1036	{
		1037	#if HAVE_SYNC_FILE_RANGE
		1038	int res;
		1039
		1040	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		1041	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		1042	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		1043	{
		1044	flags = 0
		1045	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		1046	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		1047	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		1048	}
		1049
		1050	res = sync_file_range (fd, offset, nbytes, flags);
		1051
		1052	if (!res || errno != ENOSYS)
		1053	return res;
		1054	#endif
		1055
		1056	/* even though we could play tricks with the flags, it's better to always
		1057	* call fdatasync, as that matches the expectation of its users best */
		1058	return fdatasync (fd);
		1059	}
		1060
		1061	static int
		1062	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		1063	{
		1064	#if HAVE_LINUX_FALLOCATE
		1065	return fallocate (fd, mode, offset, len);
		1066	#else
		1067	return EIO_ENOSYS ();
		1068	#endif
		1069	}
563		1070
564	#if !HAVE_READAHEAD	1071	#if !HAVE_READAHEAD
		1072	# undef readahead
565	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1073	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566		1074
567	static ssize_t	1075	static eio_ssize_t
568	eio__readahead (int fd, off_t offset, size_t count, worker *self)	1076	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
569	{	1077	{
570	size_t todo = count;	1078	size_t todo = count;
571	dBUF;	1079	dBUF;
572		1080
573	while (todo > 0)	1081	while (todo > 0)
…		…
577	pread (fd, eio_buf, len, offset);	1085	pread (fd, eio_buf, len, offset);
578	offset += len;	1086	offset += len;
579	todo -= len;	1087	todo -= len;
580	}	1088	}
581		1089
582	errno = 0;	1090	FUBd;
		1091
		1092	/* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
		1093	/* but not for e.g. EIO or eof, so we also never fail */
583	return count;	1094	return 0;
584	}	1095	}
585		1096
586	#endif
587
588	#if !HAVE_READDIR_R
589	# define readdir_r eio__readdir_r
590
591	static mutex_t readdirlock = X_MUTEX_INIT;
592
593	static int
594	eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
595	{
596	EIO_STRUCT_DIRENT *e;
597	int errorno;
598
599	X_LOCK (readdirlock);
600
601	e = readdir (dirp);
602	errorno = errno;
603
604	if (e)
605	{
606	*res = ent;
607	strcpy (ent->d_name, e->d_name);
608	}
609	else
610	*res = 0;
611
612	X_UNLOCK (readdirlock);
613
614	errno = errorno;
615	return e ? 0 : -1;
616	}
617	#endif	1097	#endif
618		1098
619	/* sendfile always needs emulation */	1099	/* sendfile always needs emulation */
620	static ssize_t	1100	static eio_ssize_t
621	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self)	1101	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
622	{	1102	{
		1103	eio_ssize_t written = 0;
623	ssize_t res;	1104	eio_ssize_t res;
624		1105
625	if (!count)	1106	if (!count)
626	return 0;	1107	return 0;
627		1108
		1109	for (;;)
		1110	{
		1111	#ifdef __APPLE__
		1112	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1113	#endif
628	#if HAVE_SENDFILE	1114	#if HAVE_SENDFILE
629	# if __linux	1115	# if __linux
		1116	off_t soffset = offset;
630	res = sendfile (ofd, ifd, &offset, count);	1117	res = sendfile (ofd, ifd, &soffset, count);
631		1118
632	# elif __freebsd	1119	# elif __FreeBSD__
633	/*	1120	/*
634	* Of course, the freebsd sendfile is a dire hack with no thoughts	1121	* Of course, the freebsd sendfile is a dire hack with no thoughts
635	* wasted on making it similar to other I/O functions.	1122	* wasted on making it similar to other I/O functions.
636	*/	1123	*/
637	{
638	off_t sbytes;	1124	off_t sbytes;
639	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1125	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
640		1126
641	if (res < 0 && sbytes)	1127	#if 0 /* according to the manpage, this is correct, but broken behaviour */
642	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1128	/* freebsd' sendfile will return 0 on success */
		1129	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1130	/* not on e.g. EIO or EPIPE - sounds broken */
		1131	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
643	res = sbytes;	1132	res = sbytes;
644	}	1133	#endif
		1134
		1135	/* according to source inspection, this is correct, and useful behaviour */
		1136	if (sbytes)
		1137	res = sbytes;
		1138
		1139	# elif defined __APPLE__
		1140	off_t sbytes = count;
		1141	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1142
		1143	/* according to the manpage, sbytes is always valid */
		1144	if (sbytes)
		1145	res = sbytes;
645		1146
646	# elif __hpux	1147	# elif __hpux
647	res = sendfile (ofd, ifd, offset, count, 0, 0);	1148	res = sendfile (ofd, ifd, offset, count, 0, 0);
648		1149
649	# elif __solaris	1150	# elif __solaris
650	{
651	struct sendfilevec vec;	1151	struct sendfilevec vec;
652	size_t sbytes;	1152	size_t sbytes;
653		1153
654	vec.sfv_fd = ifd;	1154	vec.sfv_fd = ifd;
655	vec.sfv_flag = 0;	1155	vec.sfv_flag = 0;
656	vec.sfv_off = offset;	1156	vec.sfv_off = offset;
657	vec.sfv_len = count;	1157	vec.sfv_len = count;
658		1158
659	res = sendfilev (ofd, &vec, 1, &sbytes);	1159	res = sendfilev (ofd, &vec, 1, &sbytes);
660		1160
661	if (res < 0 && sbytes)	1161	if (res < 0 && sbytes)
662	res = sbytes;	1162	res = sbytes;
663	}
664		1163
665	# endif	1164	# endif
		1165
		1166	#elif defined (_WIN32) && 0
		1167	/* does not work, just for documentation of what would need to be done */
		1168	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1169	/* libeio guarantees that the file offset does not change, and windows */
		1170	/* has no way to get an independent handle to the same file description */
		1171	HANDLE h = TO_SOCKET (ifd);
		1172	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1173	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1174
666	#else	1175	#else
667	res = -1;	1176	res = EIO_ENOSYS ();
668	errno = ENOSYS;
669	#endif	1177	#endif
670		1178
		1179	/* we assume sendfile can copy at least 128mb in one go */
		1180	if (res <= 128 * 1024 * 1024)
		1181	{
		1182	if (res > 0)
		1183	written += res;
		1184
		1185	if (written)
		1186	return written;
		1187
		1188	break;
		1189	}
		1190	else
		1191	{
		1192	/* if we requested more, then probably the kernel was lazy */
		1193	written += res;
		1194	offset += res;
		1195	count -= res;
		1196
		1197	if (!count)
		1198	return written;
		1199	}
		1200	}
		1201
671	if (res < 0	1202	if (res < 0
672	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1203	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1204	/* BSDs */
		1205	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1206	|| errno == ENOTSUP
		1207	#endif
		1208	#ifdef EOPNOTSUPP /* windows */
		1209	|| errno == EOPNOTSUPP /* BSDs */
		1210	#endif
673	#if __solaris	1211	#if __solaris
674	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1212	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
675	#endif	1213	#endif
676	)	1214	)
677	)	1215	)
…		…
681		1219
682	res = 0;	1220	res = 0;
683		1221
684	while (count)	1222	while (count)
685	{	1223	{
686	ssize_t cnt;	1224	eio_ssize_t cnt;
687		1225
688	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1226	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
689		1227
690	if (cnt <= 0)	1228	if (cnt <= 0)
691	{	1229	{
…		…
703		1241
704	offset += cnt;	1242	offset += cnt;
705	res += cnt;	1243	res += cnt;
706	count -= cnt;	1244	count -= cnt;
707	}	1245	}
		1246
		1247	FUBd;
708	}	1248	}
709		1249
710	return res;	1250	return res;
		1251	}
		1252
		1253	#ifdef PAGESIZE
		1254	# define eio_pagesize() PAGESIZE
		1255	#else
		1256	static intptr_t
		1257	eio_pagesize (void)
		1258	{
		1259	static intptr_t page;
		1260
		1261	if (!page)
		1262	page = sysconf (_SC_PAGESIZE);
		1263
		1264	return page;
		1265	}
		1266	#endif
		1267
		1268	static void
		1269	eio_page_align (void **addr, size_t *length)
		1270	{
		1271	intptr_t mask = eio_pagesize () - 1;
		1272
		1273	/* round down addr */
		1274	intptr_t adj = mask & (intptr_t)*addr;
		1275
		1276	*addr = (void *)((intptr_t)*addr - adj);
		1277	*length += adj;
		1278
		1279	/* round up length */
		1280	*length = (*length + mask) & ~mask;
		1281	}
		1282
		1283	#if !_POSIX_MEMLOCK
		1284	# define eio__mlockall(a) EIO_ENOSYS ()
		1285	#else
		1286
		1287	static int
		1288	eio__mlockall (int flags)
		1289	{
		1290	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1291	extern int mallopt (int, int);
		1292	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1293	#endif
		1294
		1295	if (EIO_MCL_CURRENT != MCL_CURRENT
		1296	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1297	{
		1298	flags = 0
		1299	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1300	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1301	}
		1302
		1303	return mlockall (flags);
		1304	}
		1305	#endif
		1306
		1307	#if !_POSIX_MEMLOCK_RANGE
		1308	# define eio__mlock(a,b) EIO_ENOSYS ()
		1309	#else
		1310
		1311	static int
		1312	eio__mlock (void *addr, size_t length)
		1313	{
		1314	eio_page_align (&addr, &length);
		1315
		1316	return mlock (addr, length);
		1317	}
		1318
		1319	#endif
		1320
		1321	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1322	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1323	#else
		1324
		1325	static int
		1326	eio__msync (void *mem, size_t len, int flags)
		1327	{
		1328	eio_page_align (&mem, &len);
		1329
		1330	if (EIO_MS_ASYNC != MS_SYNC
		1331	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1332	|| EIO_MS_SYNC != MS_SYNC)
		1333	{
		1334	flags = 0
		1335	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1336	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1337	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1338	}
		1339
		1340	return msync (mem, len, flags);
		1341	}
		1342
		1343	#endif
		1344
		1345	static int
		1346	eio__mtouch (eio_req *req)
		1347	{
		1348	void *mem = req->ptr2;
		1349	size_t len = req->size;
		1350	int flags = req->int1;
		1351
		1352	eio_page_align (&mem, &len);
		1353
		1354	{
		1355	intptr_t addr = (intptr_t)mem;
		1356	intptr_t end = addr + len;
		1357	intptr_t page = eio_pagesize ();
		1358
		1359	if (addr < end)
		1360	if (flags & EIO_MT_MODIFY) /* modify */
		1361	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1362	else
		1363	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1364	}
		1365
		1366	return 0;
		1367	}
		1368
		1369	/*****************************************************************************/
		1370	/* requests implemented outside eio_execute, because they are so large */
		1371
		1372	static void
		1373	eio__lseek (eio_req *req)
		1374	{
		1375	/* this usually gets optimised away completely, or your compiler sucks, */
		1376	/* or the whence constants really are not 0, 1, 2 */
		1377	int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
		1378	: req->int2 == EIO_SEEK_CUR ? SEEK_CUR
		1379	: req->int2 == EIO_SEEK_END ? SEEK_END
		1380	: req->int2;
		1381
		1382	req->offs = lseek (req->int1, req->offs, whence);
		1383	req->result = req->offs == (off_t)-1 ? -1 : 0;
		1384	}
		1385
		1386	/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
		1387	static int
		1388	eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1389	{
		1390	const char *rel = path;
		1391	char *res;
		1392	char *tmp1, *tmp2;
		1393	#if SYMLOOP_MAX > 32
		1394	int symlinks = SYMLOOP_MAX;
		1395	#else
		1396	int symlinks = 32;
		1397	#endif
		1398
		1399	errno = EINVAL;
		1400	if (!rel)
		1401	return -1;
		1402
		1403	errno = ENOENT;
		1404	if (!*rel)
		1405	return -1;
		1406
		1407	res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
		1408	tmp1 = res + PATH_MAX;
		1409	tmp2 = tmp1 + PATH_MAX;
		1410
		1411	#if 0 /* disabled, the musl way to do things is just too racy */
		1412	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1413	/* on linux we may be able to ask the kernel */
		1414	{
		1415	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1416
		1417	if (fd >= 0)
		1418	{
		1419	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1420	req->result = readlink (tmp1, res, PATH_MAX);
		1421	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1422	close (fd);
		1423
		1424	if (req->result > 0)
		1425	goto done;
		1426	}
		1427	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1428	return -1;
		1429	}
		1430	#endif
		1431	#endif
		1432
		1433	if (*rel != '/')
		1434	{
		1435	int len;
		1436
		1437	errno = ENOENT;
		1438	if (wd == EIO_INVALID_WD)
		1439	return -1;
		1440
		1441	if (wd == EIO_CWD)
		1442	{
		1443	if (!getcwd (res, PATH_MAX))
		1444	return -1;
		1445
		1446	len = strlen (res);
		1447	}
		1448	else
		1449	memcpy (res, wd->str, len = wd->len);
		1450
		1451	if (res [1]) /* only use if not / */
		1452	res += len;
		1453	}
		1454
		1455	while (*rel)
		1456	{
		1457	eio_ssize_t len, linklen;
		1458	const char *beg = rel;
		1459
		1460	while (*rel && *rel != '/')
		1461	++rel;
		1462
		1463	len = rel - beg;
		1464
		1465	if (!len) /* skip slashes */
		1466	{
		1467	++rel;
		1468	continue;
		1469	}
		1470
		1471	if (beg [0] == '.')
		1472	{
		1473	if (len == 1)
		1474	continue; /* . - nop */
		1475
		1476	if (beg [1] == '.' && len == 2)
		1477	{
		1478	/* .. - back up one component, if possible */
		1479
		1480	while (res != tmpbuf->ptr)
		1481	if (*--res == '/')
		1482	break;
		1483
		1484	continue;
		1485	}
		1486	}
		1487
		1488	errno = ENAMETOOLONG;
		1489	if (res + 1 + len + 1 >= tmp1)
		1490	return -1;
		1491
		1492	/* copy one component */
		1493	*res = '/';
		1494	memcpy (res + 1, beg, len);
		1495
		1496	/* zero-terminate, for readlink */
		1497	res [len + 1] = 0;
		1498
		1499	/* now check if it's a symlink */
		1500	linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
		1501
		1502	if (linklen < 0)
		1503	{
		1504	if (errno != EINVAL)
		1505	return -1;
		1506
		1507	/* it's a normal directory. hopefully */
		1508	res += len + 1;
		1509	}
		1510	else
		1511	{
		1512	/* yay, it was a symlink - build new path in tmp2 */
		1513	int rellen = strlen (rel);
		1514
		1515	errno = ENAMETOOLONG;
		1516	if (linklen + 1 + rellen >= PATH_MAX)
		1517	return -1;
		1518
		1519	errno = ELOOP;
		1520	if (!--symlinks)
		1521	return -1;
		1522
		1523	if (*tmp1 == '/')
		1524	res = tmpbuf->ptr; /* symlink resolves to an absolute path */
		1525
		1526	/* we need to be careful, as rel might point into tmp2 already */
		1527	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1528	tmp2 [linklen] = '/';
		1529	memcpy (tmp2, tmp1, linklen);
		1530
		1531	rel = tmp2;
		1532	}
		1533	}
		1534
		1535	/* special case for the lone root path */
		1536	if (res == tmpbuf->ptr)
		1537	*res++ = '/';
		1538
		1539	return res - (char *)tmpbuf->ptr;
		1540	}
		1541
		1542	static signed char
		1543	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1544	{
		1545	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1546	: a->inode < b->inode ? -1
		1547	: a->inode > b->inode ? 1
		1548	: 0;
		1549	}
		1550
		1551	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1552
		1553	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1554	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1555
		1556	static void
		1557	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1558	{
		1559	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
		1560	unsigned char *bit = bits;
		1561
		1562	assert (CHAR_BIT == 8);
		1563	assert (sizeof (eio_dirent) * 8 < 256);
		1564	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1565	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1566
		1567	if (size <= EIO_SORT_FAST)
		1568	return;
		1569
		1570	/* first prepare an array of bits to test in our radix sort */
		1571	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
		1572	/* inode_bits must contain all inodes ORed together */
		1573	/* which is used to skip bits that are 0 everywhere, which is very common */
		1574	{
		1575	eio_ino_t endianness;
		1576	int i, j;
		1577
		1578	/* we store the byte offset of byte n into byte n of "endianness" */
		1579	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1580	((unsigned char *)&endianness)[i] = i;
		1581
		1582	*bit++ = 0;
		1583
		1584	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1585	{
		1586	/* shifting off the byte offsets out of "endianness" */
		1587	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1588	endianness >>= 8;
		1589
		1590	for (j = 0; j < 8; ++j)
		1591	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
		1592	*bit++ = offs + j;
		1593	}
		1594
		1595	for (j = 0; j < 8; ++j)
		1596	if (score_bits & (1 << j))
		1597	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1598	}
		1599
		1600	/* now actually do the sorting (a variant of MSD radix sort) */
		1601	{
		1602	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
		1603	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
		1604	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
		1605	int stk_idx = 0;
		1606
		1607	base_stk [stk_idx] = dents;
		1608	end_stk [stk_idx] = dents + size;
		1609	bit_stk [stk_idx] = bit - 1;
		1610
		1611	do
		1612	{
		1613	base = base_stk [stk_idx];
		1614	end = end_stk [stk_idx];
		1615	bit = bit_stk [stk_idx];
		1616
		1617	for (;;)
		1618	{
		1619	unsigned char O = *bit >> 3;
		1620	unsigned char M = 1 << (*bit & 7);
		1621
		1622	eio_dirent *a = base;
		1623	eio_dirent *b = end;
		1624
		1625	if (b - a < EIO_SORT_CUTOFF)
		1626	break;
		1627
		1628	/* now bit-partition the array on the bit */
		1629	/* this ugly asymmetric loop seems to perform much better than typical */
		1630	/* partition algos found in the literature */
		1631	do
		1632	if (!(((unsigned char *)a)[O] & M))
		1633	++a;
		1634	else if (!(((unsigned char *)--b)[O] & M))
		1635	{
		1636	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1637	++a;
		1638	}
		1639	while (b > a);
		1640
		1641	/* next bit, or stop, if no bits left in this path */
		1642	if (!*--bit)
		1643	break;
		1644
		1645	base_stk [stk_idx] = a;
		1646	end_stk [stk_idx] = end;
		1647	bit_stk [stk_idx] = bit;
		1648	++stk_idx;
		1649
		1650	end = a;
		1651	}
		1652	}
		1653	while (stk_idx--);
		1654	}
		1655	}
		1656
		1657	static void
		1658	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1659	{
		1660	/* first move the smallest element to the front, to act as a sentinel */
		1661	{
		1662	int i;
		1663	eio_dirent *min = dents;
		1664
		1665	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1666	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1667	if (EIO_DENT_CMP (dents [i], <, *min))
		1668	min = &dents [i];
		1669
		1670	/* swap elements 0 and j (minimum) */
		1671	{
		1672	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1673	}
		1674	}
		1675
		1676	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1677	{
		1678	eio_dirent *i, *j;
		1679
		1680	for (i = dents + 1; i < dents + size; ++i)
		1681	{
		1682	eio_dirent value = *i;
		1683
		1684	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1685	j [1] = j [0];
		1686
		1687	j [1] = value;
		1688	}
		1689	}
		1690	}
		1691
		1692	static void
		1693	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1694	{
		1695	if (size <= 1)
		1696	return; /* our insertion sort relies on size > 0 */
		1697
		1698	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1699	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1700	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1701
		1702	/* use an insertion sort at the end, or for small arrays, */
		1703	/* as insertion sort is more efficient for small partitions */
		1704	eio_dent_insertion_sort (dents, size);
711	}	1705	}
712		1706
713	/* read a full directory */	1707	/* read a full directory */
714	static void	1708	static void
715	eio__scandir (eio_req *req, worker *self)	1709	eio__scandir (eio_req *req, etp_worker *self)
716	{	1710	{
		1711	char *name, *names;
		1712	int namesalloc = 4096 - sizeof (void *) * 4;
		1713	int namesoffs = 0;
		1714	int flags = req->int1;
		1715	eio_dirent *dents = 0;
		1716	int dentalloc = 128;
		1717	int dentoffs = 0;
		1718	eio_ino_t inode_bits = 0;
		1719	#ifdef _WIN32
		1720	HANDLE dirp;
		1721	WIN32_FIND_DATA entp;
		1722	#else
717	DIR *dirp;	1723	DIR *dirp;
718	union
719	{
720	EIO_STRUCT_DIRENT d;
721	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
722	} *u;
723	EIO_STRUCT_DIRENT *entp;	1724	EIO_STRUCT_DIRENT *entp;
724	char *name, *names;	1725	#endif
725	int memlen = 4096;
726	int memofs = 0;
727	int res = 0;
728		1726
729	X_LOCK (wrklock);	1727	req->result = -1;
730	self->dirp = dirp = opendir (req->ptr1);
731	self->dbuf = u = malloc (sizeof (*u));
732	req->flags |= EIO_FLAG_PTR2_FREE;
733	req->ptr2 = names = malloc (memlen);
734	X_UNLOCK (wrklock);
735		1728
736	if (dirp && u && names)	1729	if (!(flags & EIO_READDIR_DENTS))
737	for (;;)	1730	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
		1731
		1732	#ifdef _WIN32
		1733	{
		1734	int len = strlen ((const char *)req->ptr1);
		1735	char *path = malloc (MAX_PATH);
		1736	const char *fmt;
		1737	const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		1738
		1739	if (!len)
		1740	fmt = "./*";
		1741	else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
		1742	fmt = "%s*";
		1743	else
		1744	fmt = "%s/*";
		1745
		1746	_snprintf (path, MAX_PATH, fmt, reqpath);
		1747	dirp = FindFirstFile (path, &entp);
		1748	free (path);
		1749
		1750	if (dirp == INVALID_HANDLE_VALUE)
		1751	{
		1752	/* should steal _dosmaperr */
		1753	switch (GetLastError ())
		1754	{
		1755	case ERROR_FILE_NOT_FOUND:
		1756	req->result = 0;
		1757	break;
		1758
		1759	case ERROR_INVALID_NAME:
		1760	case ERROR_PATH_NOT_FOUND:
		1761	case ERROR_NO_MORE_FILES:
		1762	errno = ENOENT;
		1763	break;
		1764
		1765	case ERROR_NOT_ENOUGH_MEMORY:
		1766	errno = ENOMEM;
		1767	break;
		1768
		1769	default:
		1770	errno = EINVAL;
		1771	break;
		1772	}
		1773
		1774	return;
		1775	}
		1776	}
		1777	#else
		1778	#if HAVE_AT
		1779	if (req->wd)
738	{	1780	{
		1781	int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1782
		1783	if (fd < 0)
		1784	return;
		1785
		1786	dirp = fdopendir (fd);
		1787
		1788	if (!dirp)
		1789	close (fd);
		1790	}
		1791	else
		1792	dirp = opendir (req->ptr1);
		1793	#else
		1794	dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
		1795	#endif
		1796
		1797	if (!dirp)
		1798	return;
		1799	#endif
		1800
		1801	if (req->flags & EIO_FLAG_PTR1_FREE)
		1802	free (req->ptr1);
		1803
		1804	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1805	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
		1806	req->ptr2 = names = malloc (namesalloc);
		1807
		1808	if (!names || (flags && !dents))
		1809	return;
		1810
		1811	for (;;)
		1812	{
		1813	int done;
		1814
		1815	#ifdef _WIN32
		1816	done = !dirp;
		1817	#else
739	errno = 0;	1818	errno = 0;
740	readdir_r (dirp, &u->d, &entp);	1819	entp = readdir (dirp);
		1820	done = !entp;
		1821	#endif
741		1822
		1823	if (done)
		1824	{
		1825	#ifndef _WIN32
		1826	int old_errno = errno;
		1827	closedir (dirp);
		1828	errno = old_errno;
		1829
742	if (!entp)	1830	if (errno)
		1831	break;
		1832	#endif
		1833
		1834	/* sort etc. */
		1835	req->int1 = flags;
		1836	req->result = dentoffs;
		1837
		1838	if (flags & EIO_READDIR_STAT_ORDER)
		1839	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1840	else if (flags & EIO_READDIR_DIRS_FIRST)
		1841	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1842	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1843	else
		1844	{
		1845	/* in this case, all is known, and we just put dirs first and sort them */
		1846	eio_dirent *oth = dents + dentoffs;
		1847	eio_dirent *dir = dents;
		1848
		1849	/* now partition dirs to the front, and non-dirs to the back */
		1850	/* by walking from both sides and swapping if necessary */
		1851	while (oth > dir)
		1852	{
		1853	if (dir->type == EIO_DT_DIR)
		1854	++dir;
		1855	else if ((--oth)->type == EIO_DT_DIR)
		1856	{
		1857	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1858
		1859	++dir;
		1860	}
		1861	}
		1862
		1863	/* now sort the dirs only (dirs all have the same score) */
		1864	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1865	}
		1866
743	break;	1867	break;
		1868	}
744		1869
745	name = entp->d_name;	1870	/* now add the entry to our list(s) */
		1871	name = D_NAME (entp);
746		1872
		1873	/* skip . and .. entries */
747	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1874	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
748	{	1875	{
749	int len = strlen (name) + 1;	1876	int len = D_NAMLEN (entp) + 1;
750		1877
751	res++;	1878	while (ecb_expect_false (namesoffs + len > namesalloc))
752
753	while (memofs + len > memlen)
754	{	1879	{
755	memlen *= 2;	1880	namesalloc *= 2;
756	X_LOCK (wrklock);
757	req->ptr2 = names = realloc (names, memlen);	1881	req->ptr2 = names = realloc (names, namesalloc);
758	X_UNLOCK (wrklock);
759		1882
760	if (!names)	1883	if (!names)
		1884	break;
		1885	}
		1886
		1887	memcpy (names + namesoffs, name, len);
		1888
		1889	if (dents)
		1890	{
		1891	struct eio_dirent *ent;
		1892
		1893	if (ecb_expect_false (dentoffs == dentalloc))
		1894	{
		1895	dentalloc *= 2;
		1896	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1897
		1898	if (!dents)
761	break;	1899	break;
762	}	1900	}
763		1901
764	memcpy (names + memofs, name, len);	1902	ent = dents + dentoffs;
765	memofs += len;	1903
		1904	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1905	ent->namelen = len - 1;
		1906	ent->inode = D_INO (entp);
		1907
		1908	inode_bits |= ent->inode;
		1909
		1910	switch (D_TYPE (entp))
		1911	{
		1912	default:
		1913	ent->type = EIO_DT_UNKNOWN;
		1914	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1915	break;
		1916
		1917	#ifdef DT_FIFO
		1918	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1919	#endif
		1920	#ifdef DT_CHR
		1921	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1922	#endif
		1923	#ifdef DT_MPC
		1924	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1925	#endif
		1926	#ifdef DT_DIR
		1927	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1928	#endif
		1929	#ifdef DT_NAM
		1930	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1931	#endif
		1932	#ifdef DT_BLK
		1933	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1934	#endif
		1935	#ifdef DT_MPB
		1936	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1937	#endif
		1938	#ifdef DT_REG
		1939	case DT_REG: ent->type = EIO_DT_REG; break;
		1940	#endif
		1941	#ifdef DT_NWK
		1942	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1943	#endif
		1944	#ifdef DT_CMP
		1945	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1946	#endif
		1947	#ifdef DT_LNK
		1948	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1949	#endif
		1950	#ifdef DT_SOCK
		1951	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1952	#endif
		1953	#ifdef DT_DOOR
		1954	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1955	#endif
		1956	#ifdef DT_WHT
		1957	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1958	#endif
		1959	}
		1960
		1961	ent->score = 7;
		1962
		1963	if (flags & EIO_READDIR_DIRS_FIRST)
		1964	{
		1965	if (ent->type == EIO_DT_UNKNOWN)
		1966	{
		1967	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1968	ent->score = 1;
		1969	else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
		1970	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1971	}
		1972	else if (ent->type == EIO_DT_DIR)
		1973	ent->score = 0;
		1974	}
766	}	1975	}
		1976
		1977	namesoffs += len;
		1978	++dentoffs;
767	}	1979	}
768		1980
769	if (errno)	1981	if (EIO_CANCELLED (req))
770	res = -1;	1982	{
		1983	errno = ECANCELED;
		1984	break;
		1985	}
		1986
		1987	#ifdef _WIN32
		1988	if (!FindNextFile (dirp, &entp))
		1989	{
		1990	FindClose (dirp);
		1991	dirp = 0;
		1992	}
		1993	#endif
		1994	}
		1995	}
		1996
		1997	/*****************************************************************************/
		1998	/* working directory stuff */
		1999	/* various deficiencies in the posix 2008 api force us to */
		2000	/* keep the absolute path in string form at all times */
		2001	/* fuck yeah. */
		2002
		2003	#if !HAVE_AT
		2004
		2005	/* a bit like realpath, but usually faster because it doesn'T have to return */
		2006	/* an absolute or canonical path */
		2007	static const char *
		2008	wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2009	{
		2010	if (!wd || *path == '/')
		2011	return path;
		2012
		2013	if (path [0] == '.' && !path [1])
		2014	return wd->str;
		2015
771		2016	{
772	req->result = res;	2017	int l1 = wd->len;
		2018	int l2 = strlen (path);
		2019
		2020	char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
		2021
		2022	memcpy (res, wd->str, l1);
		2023	res [l1] = '/';
		2024	memcpy (res + l1 + 1, path, l2 + 1);
		2025
		2026	return res;
		2027	}
773	}	2028	}
		2029
		2030	#endif
		2031
		2032	static eio_wd
		2033	eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2034	{
		2035	int fd;
		2036	eio_wd res;
		2037	int len = eio__realpath (tmpbuf, wd, path);
		2038
		2039	if (len < 0)
		2040	return EIO_INVALID_WD;
		2041
		2042	#if HAVE_AT
		2043	fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		2044
		2045	if (fd < 0)
		2046	return EIO_INVALID_WD;
		2047	#endif
		2048
		2049	res = malloc (sizeof (*res) + len); /* one extra 0-byte */
		2050
		2051	#if HAVE_AT
		2052	res->fd = fd;
		2053	#endif
		2054
		2055	res->len = len;
		2056	memcpy (res->str, tmpbuf->ptr, len);
		2057	res->str [len] = 0;
		2058
		2059	return res;
		2060	}
		2061
		2062	eio_wd
		2063	eio_wd_open_sync (eio_wd wd, const char *path)
		2064	{
		2065	struct tmpbuf tmpbuf = { 0 };
		2066	wd = eio__wd_open_sync (&tmpbuf, wd, path);
		2067	free (tmpbuf.ptr);
		2068
		2069	return wd;
		2070	}
		2071
		2072	void
		2073	eio_wd_close_sync (eio_wd wd)
		2074	{
		2075	if (wd != EIO_INVALID_WD && wd != EIO_CWD)
		2076	{
		2077	#if HAVE_AT
		2078	close (wd->fd);
		2079	#endif
		2080	free (wd);
		2081	}
		2082	}
		2083
		2084	#if HAVE_AT
		2085
		2086	/* they forgot these */
		2087
		2088	static int
		2089	eio__truncateat (int dirfd, const char *path, off_t length)
		2090	{
		2091	int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
		2092	int res;
		2093
		2094	if (fd < 0)
		2095	return fd;
		2096
		2097	res = ftruncate (fd, length);
		2098	close (fd);
		2099	return res;
		2100	}
		2101
		2102	static int
		2103	eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
		2104	{
		2105	int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
		2106	int res;
		2107
		2108	if (fd < 0)
		2109	return fd;
		2110
		2111	res = fstatvfs (fd, buf);
		2112	close (fd);
		2113	return res;
		2114
		2115	}
		2116
		2117	#endif
774		2118
775	/*****************************************************************************/	2119	/*****************************************************************************/
776		2120
777	#define ALLOC(len) \	2121	#define ALLOC(len) \
778	if (!req->ptr2) \	2122	if (!req->ptr2) \
…		…
787	req->result = -1; \	2131	req->result = -1; \
788	break; \	2132	break; \
789	} \	2133	} \
790	}	2134	}
791		2135
		2136	static void ecb_noinline ecb_cold
		2137	etp_proc_init (void)
		2138	{
		2139	#if HAVE_PRCTL_SET_NAME
		2140	/* provide a more sensible "thread name" */
		2141	char name[16 + 1];
		2142	const int namelen = sizeof (name) - 1;
		2143	int len;
		2144
		2145	prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
		2146	name [namelen] = 0;
		2147	len = strlen (name);
		2148	strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
		2149	prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
		2150	#endif
		2151	}
		2152
792	X_THREAD_PROC (eio_proc)	2153	X_THREAD_PROC (etp_proc)
793	{	2154	{
794	eio_req *req;	2155	ETP_REQ *req;
795	struct timespec ts;	2156	struct timespec ts;
796	worker *self = (worker *)thr_arg;	2157	etp_worker *self = (etp_worker *)thr_arg;
797		2158
		2159	etp_proc_init ();
		2160
798	/* try to distribute timeouts somewhat randomly */	2161	/* try to distribute timeouts somewhat evenly */
799	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	2162	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
800		2163
801	for (;;)	2164	for (;;)
802	{	2165	{
803	ts.tv_sec = time (0) + IDLE_TIMEOUT;	2166	ts.tv_sec = 0;
804		2167
805	X_LOCK (reqlock);	2168	X_LOCK (reqlock);
806		2169
807	for (;;)	2170	for (;;)
808	{	2171	{
809	self->req = req = reqq_shift (&req_queue);	2172	req = reqq_shift (&req_queue);
810		2173
811	if (req)	2174	if (req)
812	break;	2175	break;
813		2176
		2177	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		2178	{
		2179	X_UNLOCK (reqlock);
		2180	X_LOCK (wrklock);
		2181	--started;
		2182	X_UNLOCK (wrklock);
		2183	goto quit;
		2184	}
		2185
814	++idle;	2186	++idle;
815		2187
816	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	2188	if (idle <= max_idle)
		2189	/* we are allowed to idle, so do so without any timeout */
		2190	X_COND_WAIT (reqwait, reqlock);
		2191	else
817	{	2192	{
818	if (idle > max_idle)	2193	/* initialise timeout once */
819	{	2194	if (!ts.tv_sec)
820	--idle;	2195	ts.tv_sec = time (0) + idle_timeout;
821	X_UNLOCK (reqlock);
822	X_LOCK (wrklock);
823	--started;
824	X_UNLOCK (wrklock);
825	goto quit;
826	}
827		2196
828	/* we are allowed to idle, so do so without any timeout */
829	X_COND_WAIT (reqwait, reqlock);	2197	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
830	ts.tv_sec = time (0) + IDLE_TIMEOUT;	2198	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
831	}	2199	}
832		2200
833	--idle;	2201	--idle;
834	}	2202	}
835		2203
836	--nready;	2204	--nready;
837		2205
838	X_UNLOCK (reqlock);	2206	X_UNLOCK (reqlock);
839		2207
840	errno = 0; /* strictly unnecessary */
841
842	if (!EIO_CANCELLED (req))
843	switch (req->type)	2208	if (req->type < 0)
844	{
845	case EIO_READ: ALLOC (req->size);
846	req->result = req->offs >= 0
847	? pread (req->int1, req->ptr2, req->size, req->offs)
848	: read (req->int1, req->ptr2, req->size); break;
849	case EIO_WRITE: req->result = req->offs >= 0
850	? pwrite (req->int1, req->ptr2, req->size, req->offs)
851	: write (req->int1, req->ptr2, req->size); break;
852
853	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
854	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
855
856	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
857	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
858	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
859	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
860	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
861	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
862
863	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
864	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
865	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
866	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
867	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
868	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
869
870	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
871	case EIO_CLOSE: req->result = close (req->int1); break;
872	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
873	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
874	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
875	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
876	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
877	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
878	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
879	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
880
881	case EIO_READLINK: ALLOC (NAME_MAX);
882	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
883
884	case EIO_SYNC: req->result = 0; sync (); break;
885	case EIO_FSYNC: req->result = fsync (req->int1); break;
886	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
887
888	case EIO_READDIR: eio__scandir (req, self); break;
889
890	case EIO_BUSY:
891	#ifdef _WIN32
892	Sleep (req->nv1 * 1000.);
893	#else
894	{
895	struct timeval tv;
896
897	tv.tv_sec = req->nv1;
898	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
899
900	req->result = select (0, 0, 0, 0, &tv);
901	}
902	#endif
903	break;
904
905	case EIO_UTIME:
906	case EIO_FUTIME:
907	{
908	struct timeval tv[2];
909	struct timeval *times;
910
911	if (req->nv1 != -1. || req->nv2 != -1.)
912	{
913	tv[0].tv_sec = req->nv1;
914	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
915	tv[1].tv_sec = req->nv2;
916	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
917
918	times = tv;
919	}
920	else
921	times = 0;
922
923
924	req->result = req->type == EIO_FUTIME
925	? futimes (req->int1, times)
926	: utimes (req->ptr1, times);
927	}
928
929	case EIO_GROUP:
930	case EIO_NOP:
931	req->result = 0;
932	break;
933
934	case EIO_QUIT:
935	goto quit;	2209	goto quit;
936		2210
937	default:	2211	ETP_EXECUTE (self, req);
938	req->result = -1;
939	break;
940	}
941
942	req->errorno = errno;
943		2212
944	X_LOCK (reslock);	2213	X_LOCK (reslock);
945		2214
946	++npending;	2215	++npending;
947		2216
948	if (!reqq_push (&res_queue, req) && want_poll_cb)	2217	if (!reqq_push (&res_queue, req) && want_poll_cb)
949	want_poll_cb ();	2218	want_poll_cb ();
950		2219
951	self->req = 0;
952	worker_clear (self);	2220	etp_worker_clear (self);
953		2221
954	X_UNLOCK (reslock);	2222	X_UNLOCK (reslock);
955	}	2223	}
956		2224
957	quit:	2225	quit:
		2226	free (req);
		2227
958	X_LOCK (wrklock);	2228	X_LOCK (wrklock);
959	worker_free (self);	2229	etp_worker_free (self);
960	X_UNLOCK (wrklock);	2230	X_UNLOCK (wrklock);
961		2231
962	return 0;	2232	return 0;
963	}	2233	}
964		2234
965	/*****************************************************************************/	2235	/*****************************************************************************/
966		2236
967	static void eio_atfork_prepare (void)	2237	int ecb_cold
968	{
969	X_LOCK (wrklock);
970	X_LOCK (reqlock);
971	X_LOCK (reslock);
972	#if !HAVE_PREADWRITE
973	X_LOCK (preadwritelock);
974	#endif
975	#if !HAVE_READDIR_R
976	X_LOCK (readdirlock);
977	#endif
978	}
979
980	static void eio_atfork_parent (void)
981	{
982	#if !HAVE_READDIR_R
983	X_UNLOCK (readdirlock);
984	#endif
985	#if !HAVE_PREADWRITE
986	X_UNLOCK (preadwritelock);
987	#endif
988	X_UNLOCK (reslock);
989	X_UNLOCK (reqlock);
990	X_UNLOCK (wrklock);
991	}
992
993	static void eio_atfork_child (void)
994	{
995	eio_req *prv;
996
997	while (prv = reqq_shift (&req_queue))
998	eio_destroy (prv);
999
1000	while (prv = reqq_shift (&res_queue))
1001	eio_destroy (prv);
1002
1003	while (wrk_first.next != &wrk_first)
1004	{
1005	worker *wrk = wrk_first.next;
1006
1007	if (wrk->req)
1008	eio_destroy (wrk->req);
1009
1010	worker_clear (wrk);
1011	worker_free (wrk);
1012	}
1013
1014	started = 0;
1015	idle = 0;
1016	nreqs = 0;
1017	nready = 0;
1018	npending = 0;
1019
1020	eio_atfork_parent ();
1021	}
1022
1023	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	2238	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1024	{	2239	{
1025	want_poll_cb = want_poll;	2240	return etp_init (want_poll, done_poll);
1026	done_poll_cb = done_poll;
1027
1028	#ifdef _WIN32
1029	X_MUTEX_CHECK (wrklock);
1030	X_MUTEX_CHECK (reslock);
1031	X_MUTEX_CHECK (reqlock);
1032	X_MUTEX_CHECK (reqwait);
1033	X_MUTEX_CHECK (preadwritelock);
1034	X_MUTEX_CHECK (readdirlock);
1035
1036	X_COND_CHECK (reqwait);
1037	#endif
1038
1039	X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1040	}	2241	}
1041		2242
		2243	ecb_inline void
1042	static void eio_api_destroy (eio_req *req)	2244	eio_api_destroy (eio_req *req)
1043	{	2245	{
1044	free (req);	2246	free (req);
1045	}	2247	}
1046		2248
1047	#define REQ(rtype) \	2249	#define REQ(rtype) \
1048	eio_req *req; \	2250	eio_req *req; \
1049	\	2251	\
1050	req = (eio_req *)calloc (1, sizeof *req); \	2252	req = (eio_req *)calloc (1, sizeof *req); \
1051	if (!req) \	2253	if (!req) \
1052	return 0; \	2254	return 0; \
1053	\	2255	\
1054	req->type = rtype; \	2256	req->type = rtype; \
1055	req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \	2257	req->pri = pri; \
1056	req->finish = cb; \	2258	req->finish = cb; \
		2259	req->data = data; \
1057	req->destroy = eio_api_destroy;	2260	req->destroy = eio_api_destroy;
1058		2261
1059	#define SEND eio_submit (req); return req	2262	#define SEND eio_submit (req); return req
1060		2263
1061	#define PATH \	2264	#define PATH \
…		…
1065	{ \	2268	{ \
1066	eio_api_destroy (req); \	2269	eio_api_destroy (req); \
1067	return 0; \	2270	return 0; \
1068	}	2271	}
1069		2272
		2273	#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
		2274
		2275	static void
		2276	eio_execute (etp_worker *self, eio_req *req)
		2277	{
		2278	#if HAVE_AT
		2279	int dirfd;
		2280	#else
		2281	const char *path;
		2282	#endif
		2283
		2284	if (ecb_expect_false (EIO_CANCELLED (req)))
		2285	{
		2286	req->result = -1;
		2287	req->errorno = ECANCELED;
		2288	return;
		2289	}
		2290
		2291	if (ecb_expect_false (req->wd == EIO_INVALID_WD))
		2292	{
		2293	req->result = -1;
		2294	req->errorno = ENOENT;
		2295	return;
		2296	}
		2297
		2298	if (req->type >= EIO_OPEN)
		2299	{
		2300	#if HAVE_AT
		2301	dirfd = WD2FD (req->wd);
		2302	#else
		2303	path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		2304	#endif
		2305	}
		2306
		2307	switch (req->type)
		2308	{
		2309	case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
		2310	req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
		2311	break;
		2312	case EIO_WD_CLOSE: req->result = 0;
		2313	eio_wd_close_sync (req->wd); break;
		2314
		2315	case EIO_SEEK: eio__lseek (req); break;
		2316	case EIO_READ: ALLOC (req->size);
		2317	req->result = req->offs >= 0
		2318	? pread (req->int1, req->ptr2, req->size, req->offs)
		2319	: read (req->int1, req->ptr2, req->size); break;
		2320	case EIO_WRITE: req->result = req->offs >= 0
		2321	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		2322	: write (req->int1, req->ptr2, req->size); break;
		2323
		2324	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		2325	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
		2326
		2327	#if HAVE_AT
		2328
		2329	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2330	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
		2331	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2332	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
		2333	case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
		2334	case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
		2335	case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
		2336	case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
		2337
		2338	case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
		2339	case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
		2340	req->result = req->wd && SINGLEDOT (req->ptr1)
		2341	? rmdir (req->wd->str)
		2342	: unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
		2343	case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
		2344	case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
		2345	req->result = req->wd && SINGLEDOT (req->ptr1)
		2346	? rename (req->wd->str, req->ptr2)
		2347	: renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
		2348	case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
		2349	case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
		2350	case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2351	case EIO_READLINK: ALLOC (PATH_MAX);
		2352	req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
		2353	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2354	req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2355	case EIO_UTIME:
		2356	case EIO_FUTIME:
		2357	{
		2358	struct timespec ts[2];
		2359	struct timespec *times;
		2360
		2361	if (req->nv1 != -1. || req->nv2 != -1.)
		2362	{
		2363	ts[0].tv_sec = req->nv1;
		2364	ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
		2365	ts[1].tv_sec = req->nv2;
		2366	ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
		2367
		2368	times = ts;
		2369	}
		2370	else
		2371	times = 0;
		2372
		2373	req->result = req->type == EIO_FUTIME
		2374	? futimens (req->int1, times)
		2375	: utimensat (dirfd, req->ptr1, times, 0);
		2376	}
		2377	break;
		2378
		2379	#else
		2380
		2381	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2382	req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2383	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2384	req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2385	case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
		2386	case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
		2387	case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
		2388	case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
		2389
		2390	case EIO_UNLINK: req->result = unlink (path ); break;
		2391	case EIO_RMDIR: req->result = rmdir (path ); break;
		2392	case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
		2393	case EIO_RENAME: req->result = rename (path , req->ptr2); break;
		2394	case EIO_LINK: req->result = link (path , req->ptr2); break;
		2395	case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
		2396	case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
		2397	case EIO_READLINK: ALLOC (PATH_MAX);
		2398	req->result = readlink (path, req->ptr2, PATH_MAX); break;
		2399	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2400	req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2401
		2402	case EIO_UTIME:
		2403	case EIO_FUTIME:
		2404	{
		2405	struct timeval tv[2];
		2406	struct timeval *times;
		2407
		2408	if (req->nv1 != -1. || req->nv2 != -1.)
		2409	{
		2410	tv[0].tv_sec = req->nv1;
		2411	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
		2412	tv[1].tv_sec = req->nv2;
		2413	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
		2414
		2415	times = tv;
		2416	}
		2417	else
		2418	times = 0;
		2419
		2420	req->result = req->type == EIO_FUTIME
		2421	? futimes (req->int1, times)
		2422	: utimes (req->ptr1, times);
		2423	}
		2424	break;
		2425
		2426	#endif
		2427
		2428	case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
		2429	{
		2430	ALLOC (req->result);
		2431	memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
		2432	}
		2433	break;
		2434
		2435	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2436	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2437
		2438	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2439	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2440
		2441	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		2442	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		2443	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		2444
		2445	case EIO_CLOSE: req->result = close (req->int1); break;
		2446	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		2447	case EIO_SYNC: req->result = 0; sync (); break;
		2448	case EIO_FSYNC: req->result = fsync (req->int1); break;
		2449	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2450	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2451	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2452	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2453	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		2454	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2455	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		2456	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
		2457
		2458	case EIO_READDIR: eio__scandir (req, self); break;
		2459
		2460	case EIO_BUSY:
		2461	#ifdef _WIN32
		2462	Sleep (req->nv1 * 1e3);
		2463	#else
		2464	{
		2465	struct timeval tv;
		2466
		2467	tv.tv_sec = req->nv1;
		2468	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
		2469
		2470	req->result = select (0, 0, 0, 0, &tv);
		2471	}
		2472	#endif
		2473	break;
		2474
		2475	case EIO_GROUP:
		2476	abort (); /* handled in eio_request */
		2477
		2478	case EIO_NOP:
		2479	req->result = 0;
		2480	break;
		2481
		2482	case EIO_CUSTOM:
		2483	req->feed (req);
		2484	break;
		2485
		2486	default:
		2487	req->result = EIO_ENOSYS ();
		2488	break;
		2489	}
		2490
		2491	req->errorno = errno;
		2492	}
		2493
		2494	#ifndef EIO_NO_WRAPPERS
		2495
		2496	eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
		2497	{
		2498	REQ (EIO_WD_OPEN); PATH; SEND;
		2499	}
		2500
		2501	eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
		2502	{
		2503	REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
		2504	}
		2505
1070	eio_req *eio_nop (int pri, eio_cb cb, void *data)	2506	eio_req *eio_nop (int pri, eio_cb cb, void *data)
1071	{	2507	{
1072	REQ (EIO_NOP); SEND;	2508	REQ (EIO_NOP); SEND;
1073	}	2509	}
1074		2510
…		…
1085	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	2521	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1086	{	2522	{
1087	REQ (EIO_FSYNC); req->int1 = fd; SEND;	2523	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1088	}	2524	}
1089		2525
		2526	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2527	{
		2528	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2529	}
		2530
1090	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2531	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1091	{	2532	{
1092	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2533	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1093	}	2534	}
1094		2535
		2536	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2537	{
		2538	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2539	}
		2540
		2541	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2542	{
		2543	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2544	}
		2545
		2546	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2547	{
		2548	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2549	}
		2550
		2551	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2552	{
		2553	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2554	}
		2555
		2556	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2557	{
		2558	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2559	}
		2560
		2561	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2562	{
		2563	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2564	}
		2565
1095	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2566	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1096	{	2567	{
1097	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2568	REQ (EIO_CLOSE); req->int1 = fd; SEND;
1098	}	2569	}
1099		2570
1100	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)	2571	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1101	{	2572	{
1102	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;	2573	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1103	}	2574	}
1104		2575
		2576	eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
		2577	{
		2578	REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
		2579	}
		2580
1105	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)	2581	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1106	{	2582	{
1107	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;	2583	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1108	}	2584	}
1109		2585
…		…
1115	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2591	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1116	{	2592	{
1117	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2593	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1118	}	2594	}
1119		2595
		2596	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2597	{
		2598	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2599	}
		2600
1120	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2601	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1121	{	2602	{
1122	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2603	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1123	}	2604	}
1124		2605
…		…
1130	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2611	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1131	{	2612	{
1132	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2613	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1133	}	2614	}
1134		2615
1135	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2616	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1136	{	2617	{
1137	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2618	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1138	}	2619	}
1139		2620
1140	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2621	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
1160	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2641	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1161	{	2642	{
1162	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2643	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1163	}	2644	}
1164		2645
1165	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2646	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1166	{	2647	{
1167	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2648	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1168	}	2649	}
1169		2650
1170	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2651	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
1186	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2667	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1187	{	2668	{
1188	return eio__1path (EIO_READLINK, path, pri, cb, data);	2669	return eio__1path (EIO_READLINK, path, pri, cb, data);
1189	}	2670	}
1190		2671
		2672	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2673	{
		2674	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2675	}
		2676
1191	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2677	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1192	{	2678	{
1193	return eio__1path (EIO_STAT, path, pri, cb, data);	2679	return eio__1path (EIO_STAT, path, pri, cb, data);
1194	}	2680	}
1195		2681
1196	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2682	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1197	{	2683	{
1198	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2684	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1199	}	2685	}
1200		2686
		2687	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2688	{
		2689	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2690	}
		2691
1201	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2692	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1202	{	2693	{
1203	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2694	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1204	}	2695	}
1205		2696
1206	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	2697	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1207	{	2698	{
1208	return eio__1path (EIO_RMDIR, path, pri, cb, data);	2699	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1209	}	2700	}
1210		2701
1211	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	2702	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1212	{	2703	{
1213	return eio__1path (EIO_READDIR, path, pri, cb, data);	2704	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1214	}	2705	}
1215		2706
1216	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2707	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1217	{	2708	{
1218	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	2709	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1219	}	2710	}
1220		2711
1221	static eio_req *	2712	static eio_req *
1222	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2713	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1223	{	2714	{
…		…
1247	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2738	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1248	{	2739	{
1249	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2740	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1250	}	2741	}
1251		2742
		2743	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
		2744	{
		2745	REQ (EIO_CUSTOM); req->feed = execute; SEND;
		2746	}
		2747
		2748	#endif
		2749
1252	eio_req *eio_grp (eio_cb cb, void *data)	2750	eio_req *eio_grp (eio_cb cb, void *data)
1253	{	2751	{
1254	const int pri = EIO_PRI_MAX;	2752	const int pri = EIO_PRI_MAX;
1255		2753
1256	REQ (EIO_GROUP); SEND;	2754	REQ (EIO_GROUP); SEND;
…		…
1261	#undef SEND	2759	#undef SEND
1262		2760
1263	/*****************************************************************************/	2761	/*****************************************************************************/
1264	/* grp functions */	2762	/* grp functions */
1265		2763
		2764	void
1266	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2765	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1267	{	2766	{
1268	grp->int2 = limit;	2767	grp->int2 = limit;
1269	grp->feed = feed;	2768	grp->feed = feed;
1270		2769
1271	grp_try_feed (grp);	2770	grp_try_feed (grp);
1272	}	2771	}
1273		2772
		2773	void
1274	void eio_grp_limit (eio_req *grp, int limit)	2774	eio_grp_limit (eio_req *grp, int limit)
1275	{	2775	{
1276	grp->int2 = limit;	2776	grp->int2 = limit;
1277		2777
1278	grp_try_feed (grp);	2778	grp_try_feed (grp);
1279	}	2779	}
1280		2780
		2781	void
1281	void eio_grp_add (eio_req *grp, eio_req *req)	2782	eio_grp_add (eio_req *grp, eio_req *req)
1282	{	2783	{
1283	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2784	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
		2785
		2786	grp->flags |= EIO_FLAG_GROUPADD;
1284		2787
1285	++grp->size;	2788	++grp->size;
1286	req->grp = grp;	2789	req->grp = grp;
1287		2790
1288	req->grp_prev = 0;	2791	req->grp_prev = 0;
…		…
1295	}	2798	}
1296		2799
1297	/*****************************************************************************/	2800	/*****************************************************************************/
1298	/* misc garbage */	2801	/* misc garbage */
1299		2802
		2803	eio_ssize_t
1300	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2804	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1301	{	2805	{
1302	worker wrk;
1303
1304	wrk.dbuf = 0;
1305
1306	eio__sendfile (ofd, ifd, offset, count, &wrk);	2806	return eio__sendfile (ofd, ifd, offset, count);
1307
1308	if (wrk.dbuf)
1309	free (wrk.dbuf);
1310	}	2807	}
1311		2808

Diff Legend

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