ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.84 by root, Thu Jul 7 22:39:56 2011 UTC vs.
Revision 1.139 by root, Thu Jun 25 18:14:19 2015 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>	4	* Copyright (c) 2007,2008,2009,2010,2011,2012,2013 Marc Alexander Lehmann <libeio@schmorp.de>
5	* All rights reserved.	5	* All rights reserved.
6	*	6	*
7	* Redistribution and use in source and binary forms, with or without modifica-	7	* Redistribution and use in source and binary forms, with or without modifica-
8	* tion, are permitted provided that the following conditions are met:	8	* tion, are permitted provided that the following conditions are met:
9	*	9	*
10	* 1. Redistributions of source code must retain the above copyright notice,	10	* 1. Redistributions of source code must retain the above copyright notice,
11	* this list of conditions and the following disclaimer.	11	* this list of conditions and the following disclaimer.
12	*	12	*
13	* 2. Redistributions in binary form must reproduce the above copyright	13	* 2. Redistributions in binary form must reproduce the above copyright
14	* notice, this list of conditions and the following disclaimer in the	14	* notice, this list of conditions and the following disclaimer in the
15	* documentation and/or other materials provided with the distribution.	15	* documentation and/or other materials provided with the distribution.
16	*	16	*
17	* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED	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-	18	* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19	* CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO	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-	20	* EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21	* CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,	21	* CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
…		…
43		43
44	#include "eio.h"	44	#include "eio.h"
45	#include "ecb.h"	45	#include "ecb.h"
46		46
47	#ifdef EIO_STACKSIZE	47	#ifdef EIO_STACKSIZE
48	# define XTHREAD_STACKSIZE EIO_STACKSIZE	48	# define X_STACKSIZE EIO_STACKSIZE
49	#endif	49	#endif
50	#include "xthread.h"	50	#include "xthread.h"
51		51
52	#include <errno.h>	52	#include <errno.h>
53	#include <stddef.h>	53	#include <stddef.h>
54	#include <stdlib.h>	54	#include <stdlib.h>
55	#include <string.h>	55	#include <string.h>
56	#include <errno.h>	56	#include <errno.h>
57	#include <sys/types.h>	57	#include <sys/types.h>
58	#include <sys/stat.h>	58	#include <sys/stat.h>
59	#include <sys/statvfs.h>
60	#include <limits.h>	59	#include <limits.h>
61	#include <fcntl.h>	60	#include <fcntl.h>
62	#include <assert.h>	61	#include <assert.h>
63		62
64	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */	63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
…		…
68	#endif	67	#endif
69		68
70	#ifndef ECANCELED	69	#ifndef ECANCELED
71	# define ECANCELED EDOM	70	# define ECANCELED EDOM
72	#endif	71	#endif
		72	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
		77	# define ENOTSOCK WSAENOTSOCK
		78	#endif
73		79
74	static void eio_destroy (eio_req *req);	80	static void eio_destroy (eio_req *req);
75		81
76	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
77	# 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
…		…
83		89
84	#ifndef EIO_FEED	90	#ifndef EIO_FEED
85	# 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)
86	#endif	92	#endif
87		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
88	#ifdef _WIN32	105	#ifdef _WIN32
89		106
90	/*doh*/	107	#undef PAGESIZE
		108	#define PAGESIZE 4096 /* GetSystemInfo? */
		109
		110	/* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
		111
		112	#ifdef EIO_STRUCT_STATI64
		113	/* look at perl's non-sloppy stat */
		114	#define stat(path,buf) _stati64 (path,buf)
		115	#define fstat(fd,buf) _fstati64 (fd,buf)
		116	#endif
		117	#define lstat(path,buf) stat (path,buf)
		118	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		119	#define mkdir(path,mode) _mkdir (path)
		120	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		121
		122	#define chmod(path,mode) _chmod (path, mode)
		123	#define dup(fd) _dup (fd)
		124	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		125	#define pipe(fds) _pipe (fds, 4096, O_BINARY)
		126
		127	#define fchmod(fd,mode) EIO_ENOSYS ()
		128	#define chown(path,uid,gid) EIO_ENOSYS ()
		129	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		130	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		131	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		132	#define mknod(path,mode,dev) EIO_ENOSYS ()
		133	#define sync() EIO_ENOSYS ()
		134	#define readlink(path,buf,s) EIO_ENOSYS ()
		135	#define statvfs(path,buf) EIO_ENOSYS ()
		136	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		137
		138	#define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
		139	#define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
		140
		141	#if __GNUC__
		142	typedef long long eio_off_t; /* signed for compatibility to msvc */
		143	#else
		144	typedef __int64 eio_off_t; /* unsigned not supported by msvc */
		145	#endif
		146
		147	static eio_ssize_t
		148	eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
		149	{
		150	OVERLAPPED o = { 0 };
		151	DWORD got;
		152
		153	o.Offset = offset;
		154	o.OffsetHigh = offset >> 32;
		155
		156	return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
		157	? got : -1;
		158	}
		159
		160	static eio_ssize_t
		161	eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
		162	{
		163	OVERLAPPED o = { 0 };
		164	DWORD got;
		165
		166	o.Offset = offset;
		167	o.OffsetHigh = offset >> 32;
		168
		169	return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
		170	? got : -1;
		171	}
		172
		173	/* rename() uses MoveFile, which fails to overwrite */
		174	#define rename(old,neu) eio__rename (old, neu)
		175
		176	static int
		177	eio__rename (const char *old, const char *neu)
		178	{
		179	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		180	return 0;
		181
		182	/* should steal _dosmaperr */
		183	switch (GetLastError ())
		184	{
		185	case ERROR_FILE_NOT_FOUND:
		186	case ERROR_PATH_NOT_FOUND:
		187	case ERROR_INVALID_DRIVE:
		188	case ERROR_NO_MORE_FILES:
		189	case ERROR_BAD_NETPATH:
		190	case ERROR_BAD_NET_NAME:
		191	case ERROR_BAD_PATHNAME:
		192	case ERROR_FILENAME_EXCED_RANGE:
		193	errno = ENOENT;
		194	break;
		195
		196	default:
		197	errno = EACCES;
		198	break;
		199	}
		200
		201	return -1;
		202	}
		203
		204	/* we could even stat and see if it exists */
		205	static int
		206	symlink (const char *old, const char *neu)
		207	{
		208	#if WINVER >= 0x0600
		209	if (CreateSymbolicLink (neu, old, 1))
		210	return 0;
		211
		212	if (CreateSymbolicLink (neu, old, 0))
		213	return 0;
		214	#endif
		215
		216	return EIO_ERRNO (ENOENT, -1);
		217	}
		218
		219	/* POSIX API only, causing trouble for win32 apps */
		220	#define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
		221	#define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
		222
		223	struct statvfs
		224	{
		225	int dummy;
		226	};
		227
		228	#define DT_DIR EIO_DT_DIR
		229	#define DT_REG EIO_DT_REG
		230	#define D_NAME(entp) entp.cFileName
		231	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
		232
91	#else	233	#else
92		234
93	# include <sys/time.h>	235	#include <sys/time.h>
94	# include <sys/select.h>	236	#include <sys/select.h>
95	# include <unistd.h>	237	#include <unistd.h>
		238	#include <signal.h>
		239	#include <dirent.h>
		240
		241	#ifdef ANDROID
		242	#include <sys/vfs.h>
		243	#define statvfs statfs
		244	#define fstatvfs fstatfs
		245	#include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
		246	#else
		247	#include <sys/statvfs.h>
		248	#endif
		249
		250	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		251	#include <sys/mman.h>
		252	#endif
		253
		254	#define D_NAME(entp) entp->d_name
		255
		256	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		257	#if __FreeBSD__ || __NetBSD__ || __OpenBSD__
		258	#define _DIRENT_HAVE_D_TYPE /* sigh */
		259	#define D_INO(de) (de)->d_fileno
		260	#define D_NAMLEN(de) (de)->d_namlen
		261	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		262	#define D_INO(de) (de)->d_ino
		263	#endif
		264
		265	#ifdef _D_EXACT_NAMLEN
		266	#undef D_NAMLEN
		267	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		268	#endif
		269
		270	#ifdef _DIRENT_HAVE_D_TYPE
		271	#define D_TYPE(de) (de)->d_type
		272	#endif
		273
		274	#ifndef EIO_STRUCT_DIRENT
		275	#define EIO_STRUCT_DIRENT struct dirent
		276	#endif
		277
		278	#endif
		279
		280	#if HAVE_UTIMES
96	# include <utime.h>	281	# include <utime.h>
97	# include <signal.h>
98	# include <dirent.h>
99
100	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
101	# include <sys/mman.h>
102	#endif
103
104	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
105	# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
106	# define _DIRENT_HAVE_D_TYPE /* sigh */
107	# define D_INO(de) (de)->d_fileno
108	# define D_NAMLEN(de) (de)->d_namlen
109	# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
110	# define D_INO(de) (de)->d_ino
111	# endif	282	#endif
112		283
113	#ifdef _D_EXACT_NAMLEN	284	#if HAVE_SYS_SYSCALL_H
114	# undef D_NAMLEN	285	# include <sys/syscall.h>
115	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
116	#endif
117
118	# ifdef _DIRENT_HAVE_D_TYPE
119	# define D_TYPE(de) (de)->d_type
120	# endif	286	#endif
121		287
122	# ifndef EIO_STRUCT_DIRENT	288	#if HAVE_SYS_PRCTL_H
123	# define EIO_STRUCT_DIRENT struct dirent	289	# include <sys/prctl.h>
124	# endif
125
126	#endif	290	#endif
127		291
128	#if HAVE_SENDFILE	292	#if HAVE_SENDFILE
129	# if __linux	293	# if __linux
130	# include <sys/sendfile.h>	294	# include <sys/sendfile.h>
…		…
145	#endif	309	#endif
146	#ifndef D_INO	310	#ifndef D_INO
147	# define D_INO(de) 0	311	# define D_INO(de) 0
148	#endif	312	#endif
149	#ifndef D_NAMLEN	313	#ifndef D_NAMLEN
150	# define D_NAMLEN(de) strlen ((de)->d_name)	314	# define D_NAMLEN(entp) strlen (D_NAME (entp))
151	#endif	315	#endif
152		316
153	/* used for struct dirent, AIX doesn't provide it */	317	/* used for struct dirent, AIX doesn't provide it */
154	#ifndef NAME_MAX	318	#ifndef NAME_MAX
155	# define NAME_MAX 4096	319	# define NAME_MAX 4096
…		…
161	#endif	325	#endif
162		326
163	/* buffer size for various temporary buffers */	327	/* buffer size for various temporary buffers */
164	#define EIO_BUFSIZE 65536	328	#define EIO_BUFSIZE 65536
165		329
166	#define dBUF \	330	#define dBUF \
167	char *eio_buf; \
168	ETP_WORKER_LOCK (self); \
169	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	331	char *eio_buf = malloc (EIO_BUFSIZE); \
170	ETP_WORKER_UNLOCK (self); \
171	errno = ENOMEM; \	332	errno = ENOMEM; \
172	if (!eio_buf) \	333	if (!eio_buf) \
173	return -1;	334	return -1
174		335
175	#define EIO_TICKS ((1000000 + 1023) >> 10)	336	#define FUBd \
		337	free (eio_buf)
		338
		339	/*****************************************************************************/
		340
		341	struct etp_tmpbuf;
		342
		343	#if _POSIX_VERSION >= 200809L
		344	#define HAVE_AT 1
		345	#define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
		346	#ifndef O_SEARCH
		347	#define O_SEARCH O_RDONLY
		348	#endif
		349	#else
		350	#define HAVE_AT 0
		351	static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
		352	#endif
		353
		354	struct eio_pwd
		355	{
		356	#if HAVE_AT
		357	int fd;
		358	#endif
		359	int len;
		360	char str[1]; /* actually, a 0-terminated canonical path */
		361	};
		362
		363	/*****************************************************************************/
176		364
177	#define ETP_PRI_MIN EIO_PRI_MIN	365	#define ETP_PRI_MIN EIO_PRI_MIN
178	#define ETP_PRI_MAX EIO_PRI_MAX	366	#define ETP_PRI_MAX EIO_PRI_MAX
179		367
		368	#define ETP_TYPE_QUIT -1
		369	#define ETP_TYPE_GROUP EIO_GROUP
		370
		371	static void eio_nop_callback (void) { }
		372	static void (*eio_want_poll_cb)(void) = eio_nop_callback;
		373	static void (*eio_done_poll_cb)(void) = eio_nop_callback;
		374
		375	#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
		376	#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
		377
180	struct etp_worker;	378	struct etp_worker;
181
182	#define ETP_REQ eio_req	379	#define ETP_REQ eio_req
183	#define ETP_DESTROY(req) eio_destroy (req)	380	#define ETP_DESTROY(req) eio_destroy (req)
184	static int eio_finish (eio_req *req);	381	static int eio_finish (eio_req *req);
185	#define ETP_FINISH(req) eio_finish (req)	382	#define ETP_FINISH(req) eio_finish (req)
186	static void eio_execute (struct etp_worker *self, eio_req *req);	383	static void eio_execute (struct etp_worker *self, eio_req *req);
187	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	384	#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
188		385
189	#define ETP_WORKER_CLEAR(req) \	386	#include "etp.c"
190	if (wrk->dbuf) \
191	{ \
192	free (wrk->dbuf); \
193	wrk->dbuf = 0; \
194	} \
195	\
196	if (wrk->dirp) \
197	{ \
198	closedir (wrk->dirp); \
199	wrk->dirp = 0; \
200	}
201		387
202	#define ETP_WORKER_COMMON \	388	static struct etp_pool eio_pool;
203	void *dbuf; \	389	#define EIO_POOL (&eio_pool)
204	DIR *dirp;
205
206	/*****************************************************************************/
207
208	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209
210	/* calculate time difference in ~1/EIO_TICKS of a second */
211	ecb_inline int
212	tvdiff (struct timeval *tv1, struct timeval *tv2)
213	{
214	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
215	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
216	}
217
218	static unsigned int started, idle, wanted = 4;
219
220	static void (*want_poll_cb) (void);
221	static void (*done_poll_cb) (void);
222
223	static unsigned int max_poll_time; /* reslock */
224	static unsigned int max_poll_reqs; /* reslock */
225
226	static volatile unsigned int nreqs; /* reqlock */
227	static volatile unsigned int nready; /* reqlock */
228	static volatile unsigned int npending; /* reqlock */
229	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
230	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
231
232	static xmutex_t wrklock;
233	static xmutex_t reslock;
234	static xmutex_t reqlock;
235	static xcond_t reqwait;
236
237	#if !HAVE_PREADWRITE
238	/*
239	* make our pread/pwrite emulation safe against themselves, but not against
240	* normal read/write by using a mutex. slows down execution a lot,
241	* but that's your problem, not mine.
242	*/
243	static xmutex_t preadwritelock = X_MUTEX_INIT;
244	#endif
245
246	typedef struct etp_worker
247	{
248	/* locked by wrklock */
249	struct etp_worker *prev, *next;
250
251	xthread_t tid;
252
253	/* locked by reslock, reqlock or wrklock */
254	ETP_REQ *req; /* currently processed request */
255
256	ETP_WORKER_COMMON
257	} etp_worker;
258
259	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
260
261	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
262	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
263
264	/* worker threads management */
265
266	static void ecb_cold
267	etp_worker_clear (etp_worker *wrk)
268	{
269	ETP_WORKER_CLEAR (wrk);
270	}
271
272	static void ecb_cold
273	etp_worker_free (etp_worker *wrk)
274	{
275	wrk->next->prev = wrk->prev;
276	wrk->prev->next = wrk->next;
277
278	free (wrk);
279	}
280
281	static unsigned int
282	etp_nreqs (void)
283	{
284	int retval;
285	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
286	retval = nreqs;
287	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
288	return retval;
289	}
290
291	static unsigned int
292	etp_nready (void)
293	{
294	unsigned int retval;
295
296	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
297	retval = nready;
298	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
299
300	return retval;
301	}
302
303	static unsigned int
304	etp_npending (void)
305	{
306	unsigned int retval;
307
308	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309	retval = npending;
310	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311
312	return retval;
313	}
314
315	static unsigned int
316	etp_nthreads (void)
317	{
318	unsigned int retval;
319
320	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
321	retval = started;
322	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
323
324	return retval;
325	}
326
327	/*
328	* a somewhat faster data structure might be nice, but
329	* with 8 priorities this actually needs <20 insns
330	* per shift, the most expensive operation.
331	*/
332	typedef struct {
333	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
334	int size;
335	} etp_reqq;
336
337	static etp_reqq req_queue;
338	static etp_reqq res_queue;
339
340	static int ecb_noinline
341	reqq_push (etp_reqq *q, ETP_REQ *req)
342	{
343	int pri = req->pri;
344	req->next = 0;
345
346	if (q->qe[pri])
347	{
348	q->qe[pri]->next = req;
349	q->qe[pri] = req;
350	}
351	else
352	q->qe[pri] = q->qs[pri] = req;
353
354	return q->size++;
355	}
356
357	static ETP_REQ * ecb_noinline
358	reqq_shift (etp_reqq *q)
359	{
360	int pri;
361
362	if (!q->size)
363	return 0;
364
365	--q->size;
366
367	for (pri = ETP_NUM_PRI; pri--; )
368	{
369	eio_req *req = q->qs[pri];
370
371	if (req)
372	{
373	if (!(q->qs[pri] = (eio_req *)req->next))
374	q->qe[pri] = 0;
375
376	return req;
377	}
378	}
379
380	abort ();
381	}
382
383	static void ecb_cold
384	etp_thread_init (void)
385	{
386	X_MUTEX_CREATE (wrklock);
387	X_MUTEX_CREATE (reslock);
388	X_MUTEX_CREATE (reqlock);
389	X_COND_CREATE (reqwait);
390	}
391
392	static void ecb_cold
393	etp_atfork_prepare (void)
394	{
395	X_LOCK (wrklock);
396	X_LOCK (reqlock);
397	X_LOCK (reslock);
398	#if !HAVE_PREADWRITE
399	X_LOCK (preadwritelock);
400	#endif
401	}
402
403	static void ecb_cold
404	etp_atfork_parent (void)
405	{
406	#if !HAVE_PREADWRITE
407	X_UNLOCK (preadwritelock);
408	#endif
409	X_UNLOCK (reslock);
410	X_UNLOCK (reqlock);
411	X_UNLOCK (wrklock);
412	}
413
414	static void ecb_cold
415	etp_atfork_child (void)
416	{
417	ETP_REQ *prv;
418
419	while ((prv = reqq_shift (&req_queue)))
420	ETP_DESTROY (prv);
421
422	while ((prv = reqq_shift (&res_queue)))
423	ETP_DESTROY (prv);
424
425	while (wrk_first.next != &wrk_first)
426	{
427	etp_worker *wrk = wrk_first.next;
428
429	if (wrk->req)
430	ETP_DESTROY (wrk->req);
431
432	etp_worker_clear (wrk);
433	etp_worker_free (wrk);
434	}
435
436	started = 0;
437	idle = 0;
438	nreqs = 0;
439	nready = 0;
440	npending = 0;
441
442	etp_thread_init ();
443	}
444
445	static void ecb_cold
446	etp_once_init (void)
447	{
448	etp_thread_init ();
449	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
450	}
451
452	static int ecb_cold
453	etp_init (void (*want_poll)(void), void (*done_poll)(void))
454	{
455	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
456
457	pthread_once (&doinit, etp_once_init);
458
459	want_poll_cb = want_poll;
460	done_poll_cb = done_poll;
461
462	return 0;
463	}
464
465	X_THREAD_PROC (etp_proc);
466
467	static void ecb_cold
468	etp_start_thread (void)
469	{
470	etp_worker *wrk = calloc (1, sizeof (etp_worker));
471
472	/*TODO*/
473	assert (("unable to allocate worker thread data", wrk));
474
475	X_LOCK (wrklock);
476
477	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
478	{
479	wrk->prev = &wrk_first;
480	wrk->next = wrk_first.next;
481	wrk_first.next->prev = wrk;
482	wrk_first.next = wrk;
483	++started;
484	}
485	else
486	free (wrk);
487
488	X_UNLOCK (wrklock);
489	}
490
491	static void
492	etp_maybe_start_thread (void)
493	{
494	if (ecb_expect_true (etp_nthreads () >= wanted))
495	return;
496
497	/* todo: maybe use idle here, but might be less exact */
498	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
499	return;
500
501	etp_start_thread ();
502	}
503
504	static void ecb_cold
505	etp_end_thread (void)
506	{
507	eio_req *req = calloc (1, sizeof (eio_req));
508
509	req->type = -1;
510	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
511
512	X_LOCK (reqlock);
513	reqq_push (&req_queue, req);
514	X_COND_SIGNAL (reqwait);
515	X_UNLOCK (reqlock);
516
517	X_LOCK (wrklock);
518	--started;
519	X_UNLOCK (wrklock);
520	}
521
522	static int
523	etp_poll (void)
524	{
525	unsigned int maxreqs;
526	unsigned int maxtime;
527	struct timeval tv_start, tv_now;
528
529	X_LOCK (reslock);
530	maxreqs = max_poll_reqs;
531	maxtime = max_poll_time;
532	X_UNLOCK (reslock);
533
534	if (maxtime)
535	gettimeofday (&tv_start, 0);
536
537	for (;;)
538	{
539	ETP_REQ *req;
540
541	etp_maybe_start_thread ();
542
543	X_LOCK (reslock);
544	req = reqq_shift (&res_queue);
545
546	if (req)
547	{
548	--npending;
549
550	if (!res_queue.size && done_poll_cb)
551	done_poll_cb ();
552	}
553
554	X_UNLOCK (reslock);
555
556	if (!req)
557	return 0;
558
559	X_LOCK (reqlock);
560	--nreqs;
561	X_UNLOCK (reqlock);
562
563	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
564	{
565	req->int1 = 1; /* mark request as delayed */
566	continue;
567	}
568	else
569	{
570	int res = ETP_FINISH (req);
571	if (ecb_expect_false (res))
572	return res;
573	}
574
575	if (ecb_expect_false (maxreqs && !--maxreqs))
576	break;
577
578	if (maxtime)
579	{
580	gettimeofday (&tv_now, 0);
581
582	if (tvdiff (&tv_start, &tv_now) >= maxtime)
583	break;
584	}
585	}
586
587	errno = EAGAIN;
588	return -1;
589	}
590
591	static void
592	etp_cancel (ETP_REQ *req)
593	{
594	req->cancelled = 1;
595
596	eio_grp_cancel (req);
597	}
598
599	static void
600	etp_submit (ETP_REQ *req)
601	{
602	req->pri -= ETP_PRI_MIN;
603
604	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
605	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
606
607	if (ecb_expect_false (req->type == EIO_GROUP))
608	{
609	/* I hope this is worth it :/ */
610	X_LOCK (reqlock);
611	++nreqs;
612	X_UNLOCK (reqlock);
613
614	X_LOCK (reslock);
615
616	++npending;
617
618	if (!reqq_push (&res_queue, req) && want_poll_cb)
619	want_poll_cb ();
620
621	X_UNLOCK (reslock);
622	}
623	else
624	{
625	X_LOCK (reqlock);
626	++nreqs;
627	++nready;
628	reqq_push (&req_queue, req);
629	X_COND_SIGNAL (reqwait);
630	X_UNLOCK (reqlock);
631
632	etp_maybe_start_thread ();
633	}
634	}
635
636	static void ecb_cold
637	etp_set_max_poll_time (double nseconds)
638	{
639	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
640	max_poll_time = nseconds * EIO_TICKS;
641	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
642	}
643
644	static void ecb_cold
645	etp_set_max_poll_reqs (unsigned int maxreqs)
646	{
647	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
648	max_poll_reqs = maxreqs;
649	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
650	}
651
652	static void ecb_cold
653	etp_set_max_idle (unsigned int nthreads)
654	{
655	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
656	max_idle = nthreads;
657	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
658	}
659
660	static void ecb_cold
661	etp_set_idle_timeout (unsigned int seconds)
662	{
663	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
664	idle_timeout = seconds;
665	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
666	}
667
668	static void ecb_cold
669	etp_set_min_parallel (unsigned int nthreads)
670	{
671	if (wanted < nthreads)
672	wanted = nthreads;
673	}
674
675	static void ecb_cold
676	etp_set_max_parallel (unsigned int nthreads)
677	{
678	if (wanted > nthreads)
679	wanted = nthreads;
680
681	while (started > wanted)
682	etp_end_thread ();
683	}
684		390
685	/*****************************************************************************/	391	/*****************************************************************************/
686		392
687	static void	393	static void
688	grp_try_feed (eio_req *grp)	394	grp_try_feed (eio_req *grp)
689	{	395	{
690	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	396	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
691	{	397	{
692	grp->flags &= ~EIO_FLAG_GROUPADD;	398	grp->flags &= ~ETP_FLAG_GROUPADD;
693		399
694	EIO_FEED (grp);	400	EIO_FEED (grp);
695		401
696	/* stop if no progress has been made */	402	/* stop if no progress has been made */
697	if (!(grp->flags & EIO_FLAG_GROUPADD))	403	if (!(grp->flags & ETP_FLAG_GROUPADD))
698	{	404	{
699	grp->feed = 0;	405	grp->feed = 0;
700	break;	406	break;
701	}	407	}
702	}	408	}
…		…
709		415
710	/* call feeder, if applicable */	416	/* call feeder, if applicable */
711	grp_try_feed (grp);	417	grp_try_feed (grp);
712		418
713	/* finish, if done */	419	/* finish, if done */
714	if (!grp->size && grp->int1)	420	if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
715	return eio_finish (grp);	421	return eio_finish (grp);
716	else	422	else
717	return 0;	423	return 0;
718	}	424	}
719		425
…		…
755	}	461	}
756		462
757	void	463	void
758	eio_grp_cancel (eio_req *grp)	464	eio_grp_cancel (eio_req *grp)
759	{	465	{
760	for (grp = grp->grp_first; grp; grp = grp->grp_next)	466	etp_grp_cancel (EIO_POOL, grp);
761	eio_cancel (grp);
762	}	467	}
763		468
764	void	469	void
765	eio_cancel (eio_req *req)	470	eio_cancel (eio_req *req)
766	{	471	{
767	etp_cancel (req);	472	etp_cancel (EIO_POOL, req);
768	}	473	}
769		474
770	void	475	void
771	eio_submit (eio_req *req)	476	eio_submit (eio_req *req)
772	{	477	{
773	etp_submit (req);	478	etp_submit (EIO_POOL, req);
774	}	479	}
775		480
776	unsigned int	481	unsigned int
777	eio_nreqs (void)	482	eio_nreqs (void)
778	{	483	{
779	return etp_nreqs ();	484	return etp_nreqs (EIO_POOL);
780	}	485	}
781		486
782	unsigned int	487	unsigned int
783	eio_nready (void)	488	eio_nready (void)
784	{	489	{
785	return etp_nready ();	490	return etp_nready (EIO_POOL);
786	}	491	}
787		492
788	unsigned int	493	unsigned int
789	eio_npending (void)	494	eio_npending (void)
790	{	495	{
791	return etp_npending ();	496	return etp_npending (EIO_POOL);
792	}	497	}
793		498
794	unsigned int ecb_cold	499	unsigned int ecb_cold
795	eio_nthreads (void)	500	eio_nthreads (void)
796	{	501	{
797	return etp_nthreads ();	502	return etp_nthreads (EIO_POOL);
798	}	503	}
799		504
800	void ecb_cold	505	void ecb_cold
801	eio_set_max_poll_time (double nseconds)	506	eio_set_max_poll_time (double nseconds)
802	{	507	{
803	etp_set_max_poll_time (nseconds);	508	etp_set_max_poll_time (EIO_POOL, nseconds);
804	}	509	}
805		510
806	void ecb_cold	511	void ecb_cold
807	eio_set_max_poll_reqs (unsigned int maxreqs)	512	eio_set_max_poll_reqs (unsigned int maxreqs)
808	{	513	{
809	etp_set_max_poll_reqs (maxreqs);	514	etp_set_max_poll_reqs (EIO_POOL, maxreqs);
810	}	515	}
811		516
812	void ecb_cold	517	void ecb_cold
813	eio_set_max_idle (unsigned int nthreads)	518	eio_set_max_idle (unsigned int nthreads)
814	{	519	{
815	etp_set_max_idle (nthreads);	520	etp_set_max_idle (EIO_POOL, nthreads);
816	}	521	}
817		522
818	void ecb_cold	523	void ecb_cold
819	eio_set_idle_timeout (unsigned int seconds)	524	eio_set_idle_timeout (unsigned int seconds)
820	{	525	{
821	etp_set_idle_timeout (seconds);	526	etp_set_idle_timeout (EIO_POOL, seconds);
822	}	527	}
823		528
824	void ecb_cold	529	void ecb_cold
825	eio_set_min_parallel (unsigned int nthreads)	530	eio_set_min_parallel (unsigned int nthreads)
826	{	531	{
827	etp_set_min_parallel (nthreads);	532	etp_set_min_parallel (EIO_POOL, nthreads);
828	}	533	}
829		534
830	void ecb_cold	535	void ecb_cold
831	eio_set_max_parallel (unsigned int nthreads)	536	eio_set_max_parallel (unsigned int nthreads)
832	{	537	{
833	etp_set_max_parallel (nthreads);	538	etp_set_max_parallel (EIO_POOL, nthreads);
834	}	539	}
835		540
836	int eio_poll (void)	541	int eio_poll (void)
837	{	542	{
838	return etp_poll ();	543	return etp_poll (EIO_POOL);
839	}	544	}
840		545
841	/*****************************************************************************/	546	/*****************************************************************************/
842	/* work around various missing functions */	547	/* work around various missing functions */
843
844	#if !HAVE_PREADWRITE
845	# undef pread
846	# undef pwrite
847	# define pread eio__pread
848	# define pwrite eio__pwrite
849
850	static ssize_t
851	eio__pread (int fd, void *buf, size_t count, off_t offset)
852	{
853	ssize_t res;
854	off_t ooffset;
855
856	X_LOCK (preadwritelock);
857	ooffset = lseek (fd, 0, SEEK_CUR);
858	lseek (fd, offset, SEEK_SET);
859	res = read (fd, buf, count);
860	lseek (fd, ooffset, SEEK_SET);
861	X_UNLOCK (preadwritelock);
862
863	return res;
864	}
865
866	static ssize_t
867	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
868	{
869	ssize_t res;
870	off_t ooffset;
871
872	X_LOCK (preadwritelock);
873	ooffset = lseek (fd, 0, SEEK_CUR);
874	lseek (fd, offset, SEEK_SET);
875	res = write (fd, buf, count);
876	lseek (fd, ooffset, SEEK_SET);
877	X_UNLOCK (preadwritelock);
878
879	return res;
880	}
881	#endif
882		548
883	#ifndef HAVE_UTIMES	549	#ifndef HAVE_UTIMES
884		550
885	# undef utimes	551	# undef utimes
886	# define utimes(path,times) eio__utimes (path, times)	552	# define utimes(path,times) eio__utimes (path, times)
…		…
919		585
920	#if !HAVE_FDATASYNC	586	#if !HAVE_FDATASYNC
921	# undef fdatasync	587	# undef fdatasync
922	# define fdatasync(fd) fsync (fd)	588	# define fdatasync(fd) fsync (fd)
923	#endif	589	#endif
		590
		591	static int
		592	eio__syncfs (int fd)
		593	{
		594	int res;
		595
		596	#if HAVE_SYS_SYNCFS
		597	res = (int)syscall (__NR_syncfs, (int)(fd));
		598	#else
		599	res = EIO_ENOSYS ();
		600	#endif
		601
		602	if (res < 0 && errno == ENOSYS && fd >= 0)
		603	sync ();
		604
		605	return res;
		606	}
924		607
925	/* sync_file_range always needs emulation */	608	/* sync_file_range always needs emulation */
926	static int	609	static int
927	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	610	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
928	{	611	{
…		…
951	}	634	}
952		635
953	static int	636	static int
954	eio__fallocate (int fd, int mode, off_t offset, size_t len)	637	eio__fallocate (int fd, int mode, off_t offset, size_t len)
955	{	638	{
956	#if HAVE_FALLOCATE	639	#if HAVE_LINUX_FALLOCATE
957	return fallocate (fd, mode, offset, len);	640	return fallocate (fd, mode, offset, len);
958	#else	641	#else
959	errno = ENOSYS;	642	return EIO_ENOSYS ();
960	return -1;
961	#endif	643	#endif
962	}	644	}
963		645
964	#if !HAVE_READAHEAD	646	#if !HAVE_READAHEAD
965	# undef readahead	647	# undef readahead
966	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	648	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
967		649
968	static ssize_t	650	static eio_ssize_t
969	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	651	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
970	{	652	{
971	size_t todo = count;	653	size_t todo = count;
972	dBUF;	654	dBUF;
973		655
…		…
978	pread (fd, eio_buf, len, offset);	660	pread (fd, eio_buf, len, offset);
979	offset += len;	661	offset += len;
980	todo -= len;	662	todo -= len;
981	}	663	}
982		664
983	errno = 0;	665	FUBd;
		666
		667	/* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
		668	/* but not for e.g. EIO or eof, so we also never fail */
984	return count;	669	return 0;
985	}	670	}
986		671
987	#endif	672	#endif
988		673
989	/* sendfile always needs emulation */	674	/* sendfile always needs emulation */
990	static ssize_t	675	static eio_ssize_t
991	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	676	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
992	{	677	{
993	ssize_t written = 0;	678	eio_ssize_t written = 0;
994	ssize_t res;	679	eio_ssize_t res;
995		680
996	if (!count)	681	if (!count)
997	return 0;	682	return 0;
998		683
999	for (;;)	684	for (;;)
…		…
1024		709
1025	/* according to source inspection, this is correct, and useful behaviour */	710	/* according to source inspection, this is correct, and useful behaviour */
1026	if (sbytes)	711	if (sbytes)
1027	res = sbytes;	712	res = sbytes;
1028		713
1029	# elif defined (__APPLE__)	714	# elif defined __APPLE__
1030	off_t sbytes = count;	715	off_t sbytes = count;
1031	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	716	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1032		717
1033	/* according to the manpage, sbytes is always valid */	718	/* according to the manpage, sbytes is always valid */
1034	if (sbytes)	719	if (sbytes)
…		…
1051	if (res < 0 && sbytes)	736	if (res < 0 && sbytes)
1052	res = sbytes;	737	res = sbytes;
1053		738
1054	# endif	739	# endif
1055		740
1056	#elif defined (_WIN32)	741	#elif defined (_WIN32) && 0
1057	/* does not work, just for documentation of what would need to be done */	742	/* does not work, just for documentation of what would need to be done */
1058	/* actually, cannot be done like this, as TransmitFile changes the file offset, */	743	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
1059	/* libeio guarantees that the file offset does not change, and windows */	744	/* libeio guarantees that the file offset does not change, and windows */
1060	/* has no way to get an independent handle to the same file description */	745	/* has no way to get an independent handle to the same file description */
1061	HANDLE h = TO_SOCKET (ifd);	746	HANDLE h = TO_SOCKET (ifd);
1062	SetFilePointer (h, offset, 0, FILE_BEGIN);	747	SetFilePointer (h, offset, 0, FILE_BEGIN);
1063	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);	748	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1064		749
1065	#else	750	#else
1066	res = -1;	751	res = EIO_ENOSYS ();
1067	errno = ENOSYS;
1068	#endif	752	#endif
1069		753
1070	/* we assume sendfile can copy at least 128mb in one go */	754	/* we assume sendfile can copy at least 128mb in one go */
1071	if (res <= 128 * 1024 * 1024)	755	if (res <= 128 * 1024 * 1024)
1072	{	756	{
…		…
1094	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	778	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1095	/* BSDs */	779	/* BSDs */
1096	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	780	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1097	|| errno == ENOTSUP	781	|| errno == ENOTSUP
1098	#endif	782	#endif
		783	#ifdef EOPNOTSUPP /* windows */
1099	|| errno == EOPNOTSUPP /* BSDs */	784	|| errno == EOPNOTSUPP /* BSDs */
		785	#endif
1100	#if __solaris	786	#if __solaris
1101	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	787	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1102	#endif	788	#endif
1103	)	789	)
1104	)	790	)
…		…
1108		794
1109	res = 0;	795	res = 0;
1110		796
1111	while (count)	797	while (count)
1112	{	798	{
1113	ssize_t cnt;	799	eio_ssize_t cnt;
1114		800
1115	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	801	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1116		802
1117	if (cnt <= 0)	803	if (cnt <= 0)
1118	{	804	{
…		…
1130		816
1131	offset += cnt;	817	offset += cnt;
1132	res += cnt;	818	res += cnt;
1133	count -= cnt;	819	count -= cnt;
1134	}	820	}
		821
		822	FUBd;
1135	}	823	}
1136		824
1137	return res;	825	return res;
1138	}	826	}
1139		827
…		…
1166	/* round up length */	854	/* round up length */
1167	*length = (*length + mask) & ~mask;	855	*length = (*length + mask) & ~mask;
1168	}	856	}
1169		857
1170	#if !_POSIX_MEMLOCK	858	#if !_POSIX_MEMLOCK
1171	# define eio__mlockall(a) ((errno = ENOSYS), -1)	859	# define eio__mlockall(a) EIO_ENOSYS ()
1172	#else	860	#else
1173		861
1174	static int	862	static int
1175	eio__mlockall (int flags)	863	eio__mlockall (int flags)
1176	{	864	{
…		…
1190	return mlockall (flags);	878	return mlockall (flags);
1191	}	879	}
1192	#endif	880	#endif
1193		881
1194	#if !_POSIX_MEMLOCK_RANGE	882	#if !_POSIX_MEMLOCK_RANGE
1195	# define eio__mlock(a,b) ((errno = ENOSYS), -1)	883	# define eio__mlock(a,b) EIO_ENOSYS ()
1196	#else	884	#else
1197		885
1198	static int	886	static int
1199	eio__mlock (void *addr, size_t length)	887	eio__mlock (void *addr, size_t length)
1200	{	888	{
…		…
1204	}	892	}
1205		893
1206	#endif	894	#endif
1207		895
1208	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)	896	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1209	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)	897	# define eio__msync(a,b,c) EIO_ENOSYS ()
1210	#else	898	#else
1211		899
1212	static int	900	static int
1213	eio__msync (void *mem, size_t len, int flags)	901	eio__msync (void *mem, size_t len, int flags)
1214	{	902	{
…		…
1255		943
1256	/*****************************************************************************/	944	/*****************************************************************************/
1257	/* requests implemented outside eio_execute, because they are so large */	945	/* requests implemented outside eio_execute, because they are so large */
1258		946
1259	static void	947	static void
1260	eio__realpath (eio_req *req, etp_worker *self)	948	eio__lseek (eio_req *req)
1261	{	949	{
1262	char *rel = req->ptr1;	950	/* this usually gets optimised away completely, or your compiler sucks, */
		951	/* or the whence constants really are not 0, 1, 2 */
		952	int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
		953	: req->int2 == EIO_SEEK_CUR ? SEEK_CUR
		954	: req->int2 == EIO_SEEK_END ? SEEK_END
		955	: req->int2;
		956
		957	req->offs = lseek (req->int1, req->offs, whence);
		958	req->result = req->offs == (off_t)-1 ? -1 : 0;
		959	}
		960
		961	/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
		962	static int
		963	eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
		964	{
1263	char *res;	965	char *res;
		966	const char *rel = path;
1264	char *tmp1, *tmp2;	967	char *tmp1, *tmp2;
1265	#if SYMLOOP_MAX > 32	968	#if SYMLOOP_MAX > 32
1266	int symlinks = SYMLOOP_MAX;	969	int symlinks = SYMLOOP_MAX;
1267	#else	970	#else
1268	int symlinks = 32;	971	int symlinks = 32;
1269	#endif	972	#endif
1270		973
1271	req->result = -1;
1272
1273	errno = EINVAL;	974	errno = EINVAL;
1274	if (!rel)	975	if (!rel)
1275	return;	976	return -1;
1276		977
1277	errno = ENOENT;	978	errno = ENOENT;
1278	if (!*rel)	979	if (!*rel)
1279	return;	980	return -1;
1280		981
1281	if (!req->ptr2)	982	res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
1282	{	983	#ifdef _WIN32
1283	X_LOCK (wrklock);	984	if (_access (rel, 4) != 0)
1284	req->flags |= EIO_FLAG_PTR2_FREE;
1285	X_UNLOCK (wrklock);
1286	req->ptr2 = malloc (PATH_MAX * 3);
1287
1288	errno = ENOMEM;
1289	if (!req->ptr2)
1290	return;	985	return -1;
1291	}
1292		986
1293	res = req->ptr2;	987	symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
		988
		989	errno = ENAMETOOLONG;
		990	if (symlinks >= PATH_MAX * 3)
		991	return -1;
		992
		993	errno = EIO;
		994	if (symlinks <= 0)
		995	return -1;
		996
		997	return symlinks;
		998
		999	#else
1294	tmp1 = res + PATH_MAX;	1000	tmp1 = res + PATH_MAX;
1295	tmp2 = tmp1 + PATH_MAX;	1001	tmp2 = tmp1 + PATH_MAX;
1296		1002
1297	#if 0 /* disabled, the musl way to do things is just too racy */	1003	#if 0 /* disabled, the musl way to do things is just too racy */
1298	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)	1004	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
…		…
1302		1008
1303	if (fd >= 0)	1009	if (fd >= 0)
1304	{	1010	{
1305	sprintf (tmp1, "/proc/self/fd/%d", fd);	1011	sprintf (tmp1, "/proc/self/fd/%d", fd);
1306	req->result = readlink (tmp1, res, PATH_MAX);	1012	req->result = readlink (tmp1, res, PATH_MAX);
		1013	/* here we should probably stat the open file and the disk file, to make sure they still match */
1307	close (fd);	1014	close (fd);
1308
1309	/* here we should probably stat the open file and the disk file, to make sure they still match */
1310		1015
1311	if (req->result > 0)	1016	if (req->result > 0)
1312	goto done;	1017	goto done;
1313	}	1018	}
1314	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)	1019	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1315	return;	1020	return -1;
1316	}	1021	}
1317	#endif	1022	#endif
1318	#endif	1023	#endif
1319		1024
1320	if (*rel != '/')	1025	if (*rel != '/')
1321	{	1026	{
		1027	int len;
		1028
		1029	errno = ENOENT;
		1030	if (wd == EIO_INVALID_WD)
		1031	return -1;
		1032
		1033	if (wd == EIO_CWD)
		1034	{
1322	if (!getcwd (res, PATH_MAX))	1035	if (!getcwd (res, PATH_MAX))
1323	return;	1036	return -1;
		1037
		1038	len = strlen (res);
		1039	}
		1040	else
		1041	memcpy (res, wd->str, len = wd->len);
1324		1042
1325	if (res [1]) /* only use if not / */	1043	if (res [1]) /* only use if not / */
1326	res += strlen (res);	1044	res += len;
1327	}	1045	}
1328		1046
1329	while (*rel)	1047	while (*rel)
1330	{	1048	{
1331	ssize_t len, linklen;	1049	eio_ssize_t len, linklen;
1332	char *beg = rel;	1050	const char *beg = rel;
1333		1051
1334	while (*rel && *rel != '/')	1052	while (*rel && *rel != '/')
1335	++rel;	1053	++rel;
1336		1054
1337	len = rel - beg;	1055	len = rel - beg;
…		…
1349		1067
1350	if (beg [1] == '.' && len == 2)	1068	if (beg [1] == '.' && len == 2)
1351	{	1069	{
1352	/* .. - back up one component, if possible */	1070	/* .. - back up one component, if possible */
1353		1071
1354	while (res != req->ptr2)	1072	while (res != tmpbuf->ptr)
1355	if (*--res == '/')	1073	if (*--res == '/')
1356	break;	1074	break;
1357		1075
1358	continue;	1076	continue;
1359	}	1077	}
1360	}	1078	}
1361		1079
1362	errno = ENAMETOOLONG;	1080	errno = ENAMETOOLONG;
1363	if (res + 1 + len + 1 >= tmp1)	1081	if (res + 1 + len + 1 >= tmp1)
1364	return;	1082	return -1;
1365		1083
1366	/* copy one component */	1084	/* copy one component */
1367	*res = '/';	1085	*res = '/';
1368	memcpy (res + 1, beg, len);	1086	memcpy (res + 1, beg, len);
1369		1087
1370	/* zero-terminate, for readlink */	1088	/* zero-terminate, for readlink */
1371	res [len + 1] = 0;	1089	res [len + 1] = 0;
1372		1090
1373	/* now check if it's a symlink */	1091	/* now check if it's a symlink */
1374	linklen = readlink (req->ptr2, tmp1, PATH_MAX);	1092	linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1375		1093
1376	if (linklen < 0)	1094	if (linklen < 0)
1377	{	1095	{
1378	if (errno != EINVAL)	1096	if (errno != EINVAL)
1379	return;	1097	return -1;
1380		1098
1381	/* it's a normal directory. hopefully */	1099	/* it's a normal directory. hopefully */
1382	res += len + 1;	1100	res += len + 1;
1383	}	1101	}
1384	else	1102	else
…		…
1386	/* yay, it was a symlink - build new path in tmp2 */	1104	/* yay, it was a symlink - build new path in tmp2 */
1387	int rellen = strlen (rel);	1105	int rellen = strlen (rel);
1388		1106
1389	errno = ENAMETOOLONG;	1107	errno = ENAMETOOLONG;
1390	if (linklen + 1 + rellen >= PATH_MAX)	1108	if (linklen + 1 + rellen >= PATH_MAX)
1391	return;	1109	return -1;
1392		1110
1393	errno = ELOOP;	1111	errno = ELOOP;
1394	if (!--symlinks)	1112	if (!--symlinks)
1395	return;	1113	return -1;
1396		1114
1397	if (*tmp1 == '/')	1115	if (*tmp1 == '/')
1398	res = req->ptr2; /* symlink resolves to an absolute path */	1116	res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1399		1117
1400	/* we need to be careful, as rel might point into tmp2 already */	1118	/* we need to be careful, as rel might point into tmp2 already */
1401	memmove (tmp2 + linklen + 1, rel, rellen + 1);	1119	memmove (tmp2 + linklen + 1, rel, rellen + 1);
1402	tmp2 [linklen] = '/';	1120	tmp2 [linklen] = '/';
1403	memcpy (tmp2, tmp1, linklen);	1121	memcpy (tmp2, tmp1, linklen);
…		…
1405	rel = tmp2;	1123	rel = tmp2;
1406	}	1124	}
1407	}	1125	}
1408		1126
1409	/* special case for the lone root path */	1127	/* special case for the lone root path */
1410	if (res == req->ptr2)	1128	if (res == tmpbuf->ptr)
1411	*res++ = '/';	1129	*res++ = '/';
1412		1130
1413	req->result = res - (char *)req->ptr2;	1131	return res - (char *)tmpbuf->ptr;
1414		1132	#endif
1415	done:
1416	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1417	}	1133	}
1418		1134
1419	static signed char	1135	static signed char
1420	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1136	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1421	{	1137	{
…		…
1429		1145
1430	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1146	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1431	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1147	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1432		1148
1433	static void	1149	static void
1434	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1150	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1435	{	1151	{
1436	unsigned char bits [9 + sizeof (ino_t) * 8];	1152	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1437	unsigned char *bit = bits;	1153	unsigned char *bit = bits;
1438		1154
1439	assert (CHAR_BIT == 8);	1155	assert (CHAR_BIT == 8);
1440	assert (sizeof (eio_dirent) * 8 < 256);	1156	assert (sizeof (eio_dirent) * 8 < 256);
1441	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1157	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1443		1159
1444	if (size <= EIO_SORT_FAST)	1160	if (size <= EIO_SORT_FAST)
1445	return;	1161	return;
1446		1162
1447	/* first prepare an array of bits to test in our radix sort */	1163	/* first prepare an array of bits to test in our radix sort */
1448	/* try to take endianness into account, as well as differences in ino_t sizes */	1164	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1449	/* inode_bits must contain all inodes ORed together */	1165	/* inode_bits must contain all inodes ORed together */
1450	/* which is used to skip bits that are 0 everywhere, which is very common */	1166	/* which is used to skip bits that are 0 everywhere, which is very common */
1451	{	1167	{
1452	ino_t endianness;	1168	eio_ino_t endianness;
1453	int i, j;	1169	int i, j;
1454		1170
1455	/* we store the byte offset of byte n into byte n of "endianness" */	1171	/* we store the byte offset of byte n into byte n of "endianness" */
1456	for (i = 0; i < sizeof (ino_t); ++i)	1172	for (i = 0; i < sizeof (eio_ino_t); ++i)
1457	((unsigned char *)&endianness)[i] = i;	1173	((unsigned char *)&endianness)[i] = i;
1458		1174
1459	*bit++ = 0;	1175	*bit++ = 0;
1460		1176
1461	for (i = 0; i < sizeof (ino_t); ++i)	1177	for (i = 0; i < sizeof (eio_ino_t); ++i)
1462	{	1178	{
1463	/* shifting off the byte offsets out of "endianness" */	1179	/* shifting off the byte offsets out of "endianness" */
1464	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1180	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1465	endianness >>= 8;	1181	endianness >>= 8;
1466		1182
1467	for (j = 0; j < 8; ++j)	1183	for (j = 0; j < 8; ++j)
1468	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1184	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1469	*bit++ = offs + j;	1185	*bit++ = offs + j;
1470	}	1186	}
1471		1187
1472	for (j = 0; j < 8; ++j)	1188	for (j = 0; j < 8; ++j)
1473	if (score_bits & (1 << j))	1189	if (score_bits & (1 << j))
1474	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1190	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1475	}	1191	}
1476		1192
1477	/* now actually do the sorting (a variant of MSD radix sort) */	1193	/* now actually do the sorting (a variant of MSD radix sort) */
1478	{	1194	{
1479	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1195	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1480	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1196	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1481	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1197	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1482	int stk_idx = 0;	1198	int stk_idx = 0;
1483		1199
1484	base_stk [stk_idx] = dents;	1200	base_stk [stk_idx] = dents;
1485	end_stk [stk_idx] = dents + size;	1201	end_stk [stk_idx] = dents + size;
1486	bit_stk [stk_idx] = bit - 1;	1202	bit_stk [stk_idx] = bit - 1;
…		…
1565	}	1281	}
1566	}	1282	}
1567	}	1283	}
1568		1284
1569	static void	1285	static void
1570	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1286	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1571	{	1287	{
1572	if (size <= 1)	1288	if (size <= 1)
1573	return; /* our insertion sort relies on size > 0 */	1289	return; /* our insertion sort relies on size > 0 */
1574		1290
1575	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1291	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1583		1299
1584	/* read a full directory */	1300	/* read a full directory */
1585	static void	1301	static void
1586	eio__scandir (eio_req *req, etp_worker *self)	1302	eio__scandir (eio_req *req, etp_worker *self)
1587	{	1303	{
1588	DIR *dirp;
1589	EIO_STRUCT_DIRENT *entp;
1590	char *name, *names;	1304	char *name, *names;
1591	int namesalloc = 4096;	1305	int namesalloc = 4096 - sizeof (void *) * 4;
1592	int namesoffs = 0;	1306	int namesoffs = 0;
1593	int flags = req->int1;	1307	int flags = req->int1;
1594	eio_dirent *dents = 0;	1308	eio_dirent *dents = 0;
1595	int dentalloc = 128;	1309	int dentalloc = 128;
1596	int dentoffs = 0;	1310	int dentoffs = 0;
1597	ino_t inode_bits = 0;	1311	eio_ino_t inode_bits = 0;
		1312	#ifdef _WIN32
		1313	HANDLE dirp;
		1314	WIN32_FIND_DATA entp;
		1315	#else
		1316	DIR *dirp;
		1317	EIO_STRUCT_DIRENT *entp;
		1318	#endif
1598		1319
1599	req->result = -1;	1320	req->result = -1;
1600		1321
1601	if (!(flags & EIO_READDIR_DENTS))	1322	if (!(flags & EIO_READDIR_DENTS))
1602	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1323	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1603		1324
1604	X_LOCK (wrklock);	1325	#ifdef _WIN32
1605	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1326	{
		1327	int len = strlen ((const char *)req->ptr1);
		1328	char *path = malloc (MAX_PATH);
		1329	const char *fmt;
		1330	const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		1331
		1332	if (!len)
		1333	fmt = "./*";
		1334	else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
		1335	fmt = "%s*";
		1336	else
		1337	fmt = "%s/*";
		1338
		1339	_snprintf (path, MAX_PATH, fmt, reqpath);
		1340	dirp = FindFirstFile (path, &entp);
		1341	free (path);
		1342
		1343	if (dirp == INVALID_HANDLE_VALUE)
		1344	{
		1345	/* should steal _dosmaperr */
		1346	switch (GetLastError ())
		1347	{
		1348	case ERROR_FILE_NOT_FOUND:
		1349	req->result = 0;
		1350	break;
		1351
		1352	case ERROR_INVALID_NAME:
		1353	case ERROR_PATH_NOT_FOUND:
		1354	case ERROR_NO_MORE_FILES:
		1355	errno = ENOENT;
		1356	break;
		1357
		1358	case ERROR_NOT_ENOUGH_MEMORY:
		1359	errno = ENOMEM;
		1360	break;
		1361
		1362	default:
		1363	errno = EINVAL;
		1364	break;
		1365	}
		1366
		1367	return;
		1368	}
		1369	}
		1370	#else
		1371	#if HAVE_AT
		1372	if (req->wd)
		1373	{
		1374	int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1375
		1376	if (fd < 0)
		1377	return;
		1378
		1379	dirp = fdopendir (fd);
		1380
		1381	if (!dirp)
		1382	close (fd);
		1383	}
		1384	else
1606	self->dirp = dirp = opendir (req->ptr1);	1385	dirp = opendir (req->ptr1);
		1386	#else
		1387	dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
		1388	#endif
		1389
		1390	if (!dirp)
		1391	return;
		1392	#endif
1607		1393
1608	if (req->flags & EIO_FLAG_PTR1_FREE)	1394	if (req->flags & EIO_FLAG_PTR1_FREE)
1609	free (req->ptr1);	1395	free (req->ptr1);
1610		1396
1611	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1397	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1612	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1398	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1613	req->ptr2 = names = malloc (namesalloc);	1399	req->ptr2 = names = malloc (namesalloc);
1614	X_UNLOCK (wrklock);
1615		1400
1616	if (dirp && names && (!flags || dents))	1401	if (!names || (flags && !dents))
		1402	return;
		1403
1617	for (;;)	1404	for (;;)
1618	{	1405	{
		1406	int done;
		1407
		1408	#ifdef _WIN32
		1409	done = !dirp;
		1410	#else
1619	errno = 0;	1411	errno = 0;
1620	entp = readdir (dirp);	1412	entp = readdir (dirp);
		1413	done = !entp;
		1414	#endif
1621		1415
1622	if (!entp)	1416	if (done)
1623	{	1417	{
		1418	#ifndef _WIN32
		1419	int old_errno = errno;
		1420	closedir (dirp);
		1421	errno = old_errno;
		1422
1624	if (errno)	1423	if (errno)
1625	break;	1424	break;
		1425	#endif
1626		1426
1627	/* sort etc. */	1427	/* sort etc. */
1628	req->int1 = flags;	1428	req->int1 = flags;
1629	req->result = dentoffs;	1429	req->result = dentoffs;
1630		1430
1631	if (flags & EIO_READDIR_STAT_ORDER)	1431	if (flags & EIO_READDIR_STAT_ORDER)
1632	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);	1432	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1633	else if (flags & EIO_READDIR_DIRS_FIRST)	1433	else if (flags & EIO_READDIR_DIRS_FIRST)
1634	if (flags & EIO_READDIR_FOUND_UNKNOWN)	1434	if (flags & EIO_READDIR_FOUND_UNKNOWN)
1635	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */	1435	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1636	else	1436	else
		1437	{
		1438	/* in this case, all is known, and we just put dirs first and sort them */
		1439	eio_dirent *oth = dents + dentoffs;
		1440	eio_dirent *dir = dents;
		1441
		1442	/* now partition dirs to the front, and non-dirs to the back */
		1443	/* by walking from both sides and swapping if necessary */
		1444	while (oth > dir)
		1445	{
		1446	if (dir->type == EIO_DT_DIR)
		1447	++dir;
		1448	else if ((--oth)->type == EIO_DT_DIR)
		1449	{
		1450	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1451
		1452	++dir;
		1453	}
		1454	}
		1455
		1456	/* now sort the dirs only (dirs all have the same score) */
		1457	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1458	}
		1459
		1460	break;
		1461	}
		1462
		1463	/* now add the entry to our list(s) */
		1464	name = D_NAME (entp);
		1465
		1466	/* skip . and .. entries */
		1467	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
		1468	{
		1469	int len = D_NAMLEN (entp) + 1;
		1470
		1471	while (ecb_expect_false (namesoffs + len > namesalloc))
		1472	{
		1473	namesalloc *= 2;
		1474	req->ptr2 = names = realloc (names, namesalloc);
		1475
		1476	if (!names)
		1477	break;
		1478	}
		1479
		1480	memcpy (names + namesoffs, name, len);
		1481
		1482	if (dents)
		1483	{
		1484	struct eio_dirent *ent;
		1485
		1486	if (ecb_expect_false (dentoffs == dentalloc))
1637	{	1487	{
1638	/* in this case, all is known, and we just put dirs first and sort them */	1488	dentalloc *= 2;
		1489	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1490
		1491	if (!dents)
		1492	break;
		1493	}
		1494
1639	eio_dirent *oth = dents + dentoffs;	1495	ent = dents + dentoffs;
1640	eio_dirent *dir = dents;
1641		1496
1642	/* now partition dirs to the front, and non-dirs to the back */	1497	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1643	/* by walking from both sides and swapping if necessary */	1498	ent->namelen = len - 1;
1644	while (oth > dir)	1499	ent->inode = D_INO (entp);
		1500
		1501	inode_bits |= ent->inode;
		1502
		1503	switch (D_TYPE (entp))
		1504	{
		1505	default:
		1506	ent->type = EIO_DT_UNKNOWN;
		1507	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1508	break;
		1509
		1510	#ifdef DT_FIFO
		1511	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1512	#endif
		1513	#ifdef DT_CHR
		1514	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1515	#endif
		1516	#ifdef DT_MPC
		1517	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1518	#endif
		1519	#ifdef DT_DIR
		1520	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1521	#endif
		1522	#ifdef DT_NAM
		1523	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1524	#endif
		1525	#ifdef DT_BLK
		1526	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1527	#endif
		1528	#ifdef DT_MPB
		1529	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1530	#endif
		1531	#ifdef DT_REG
		1532	case DT_REG: ent->type = EIO_DT_REG; break;
		1533	#endif
		1534	#ifdef DT_NWK
		1535	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1536	#endif
		1537	#ifdef DT_CMP
		1538	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1539	#endif
		1540	#ifdef DT_LNK
		1541	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1542	#endif
		1543	#ifdef DT_SOCK
		1544	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1545	#endif
		1546	#ifdef DT_DOOR
		1547	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1548	#endif
		1549	#ifdef DT_WHT
		1550	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1551	#endif
		1552	}
		1553
		1554	ent->score = 7;
		1555
		1556	if (flags & EIO_READDIR_DIRS_FIRST)
		1557	{
		1558	if (ent->type == EIO_DT_UNKNOWN)
1645	{	1559	{
1646	if (dir->type == EIO_DT_DIR)	1560	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1647	++dir;	1561	ent->score = 1;
1648	else if ((--oth)->type == EIO_DT_DIR)	1562	else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1649	{	1563	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1650	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1651
1652	++dir;
1653	}
1654	}	1564	}
1655		1565	else if (ent->type == EIO_DT_DIR)
1656	/* now sort the dirs only (dirs all have the same score) */	1566	ent->score = 0;
1657	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1658	}	1567	}
1659
1660	break;
1661	}
1662
1663	/* now add the entry to our list(s) */
1664	name = entp->d_name;
1665
1666	/* skip . and .. entries */
1667	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1668	{
1669	int len = D_NAMLEN (entp) + 1;
1670
1671	while (ecb_expect_false (namesoffs + len > namesalloc))
1672	{
1673	namesalloc *= 2;
1674	X_LOCK (wrklock);
1675	req->ptr2 = names = realloc (names, namesalloc);
1676	X_UNLOCK (wrklock);
1677
1678	if (!names)
1679	break;
1680	}	1568	}
1681		1569
1682	memcpy (names + namesoffs, name, len);
1683
1684	if (dents)
1685	{
1686	struct eio_dirent *ent;
1687
1688	if (ecb_expect_false (dentoffs == dentalloc))
1689	{
1690	dentalloc *= 2;
1691	X_LOCK (wrklock);
1692	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1693	X_UNLOCK (wrklock);
1694
1695	if (!dents)
1696	break;
1697	}
1698
1699	ent = dents + dentoffs;
1700
1701	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1702	ent->namelen = len - 1;
1703	ent->inode = D_INO (entp);
1704
1705	inode_bits |= ent->inode;
1706
1707	switch (D_TYPE (entp))
1708	{
1709	default:
1710	ent->type = EIO_DT_UNKNOWN;
1711	flags |= EIO_READDIR_FOUND_UNKNOWN;
1712	break;
1713
1714	#ifdef DT_FIFO
1715	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1716	#endif
1717	#ifdef DT_CHR
1718	case DT_CHR: ent->type = EIO_DT_CHR; break;
1719	#endif
1720	#ifdef DT_MPC
1721	case DT_MPC: ent->type = EIO_DT_MPC; break;
1722	#endif
1723	#ifdef DT_DIR
1724	case DT_DIR: ent->type = EIO_DT_DIR; break;
1725	#endif
1726	#ifdef DT_NAM
1727	case DT_NAM: ent->type = EIO_DT_NAM; break;
1728	#endif
1729	#ifdef DT_BLK
1730	case DT_BLK: ent->type = EIO_DT_BLK; break;
1731	#endif
1732	#ifdef DT_MPB
1733	case DT_MPB: ent->type = EIO_DT_MPB; break;
1734	#endif
1735	#ifdef DT_REG
1736	case DT_REG: ent->type = EIO_DT_REG; break;
1737	#endif
1738	#ifdef DT_NWK
1739	case DT_NWK: ent->type = EIO_DT_NWK; break;
1740	#endif
1741	#ifdef DT_CMP
1742	case DT_CMP: ent->type = EIO_DT_CMP; break;
1743	#endif
1744	#ifdef DT_LNK
1745	case DT_LNK: ent->type = EIO_DT_LNK; break;
1746	#endif
1747	#ifdef DT_SOCK
1748	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1749	#endif
1750	#ifdef DT_DOOR
1751	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1752	#endif
1753	#ifdef DT_WHT
1754	case DT_WHT: ent->type = EIO_DT_WHT; break;
1755	#endif
1756	}
1757
1758	ent->score = 7;
1759
1760	if (flags & EIO_READDIR_DIRS_FIRST)
1761	{
1762	if (ent->type == EIO_DT_UNKNOWN)
1763	{
1764	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1765	ent->score = 1;
1766	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1767	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1768	}
1769	else if (ent->type == EIO_DT_DIR)
1770	ent->score = 0;
1771	}
1772	}
1773
1774	namesoffs += len;	1570	namesoffs += len;
1775	++dentoffs;	1571	++dentoffs;
1776	}	1572	}
1777		1573
1778	if (EIO_CANCELLED (req))	1574	if (EIO_CANCELLED (req))
1779	{	1575	{
1780	errno = ECANCELED;	1576	errno = ECANCELED;
1781	break;	1577	break;
1782	}	1578	}
		1579
		1580	#ifdef _WIN32
		1581	if (!FindNextFile (dirp, &entp))
		1582	{
		1583	FindClose (dirp);
		1584	dirp = 0;
1783	}	1585	}
		1586	#endif
		1587	}
1784	}	1588	}
		1589
		1590	/*****************************************************************************/
		1591	/* working directory stuff */
		1592	/* various deficiencies in the posix 2008 api force us to */
		1593	/* keep the absolute path in string form at all times */
		1594	/* fuck yeah. */
		1595
		1596	#if !HAVE_AT
		1597
		1598	/* a bit like realpath, but usually faster because it doesn'T have to return */
		1599	/* an absolute or canonical path */
		1600	static const char *
		1601	wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1602	{
		1603	if (!wd || *path == '/')
		1604	return path;
		1605
		1606	if (path [0] == '.' && !path [1])
		1607	return wd->str;
		1608
		1609	{
		1610	int l1 = wd->len;
		1611	int l2 = strlen (path);
		1612
		1613	char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
		1614
		1615	memcpy (res, wd->str, l1);
		1616	res [l1] = '/';
		1617	memcpy (res + l1 + 1, path, l2 + 1);
		1618
		1619	return res;
		1620	}
		1621	}
		1622
		1623	#endif
		1624
		1625	static eio_wd
		1626	eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1627	{
		1628	int fd;
		1629	eio_wd res;
		1630	int len = eio__realpath (tmpbuf, wd, path);
		1631
		1632	if (len < 0)
		1633	return EIO_INVALID_WD;
		1634
		1635	#if HAVE_AT
		1636	fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1637
		1638	if (fd < 0)
		1639	return EIO_INVALID_WD;
		1640	#endif
		1641
		1642	res = malloc (sizeof (*res) + len); /* one extra 0-byte */
		1643
		1644	#if HAVE_AT
		1645	res->fd = fd;
		1646	#endif
		1647
		1648	res->len = len;
		1649	memcpy (res->str, tmpbuf->ptr, len);
		1650	res->str [len] = 0;
		1651
		1652	return res;
		1653	}
		1654
		1655	eio_wd
		1656	eio_wd_open_sync (eio_wd wd, const char *path)
		1657	{
		1658	struct etp_tmpbuf tmpbuf = { };
		1659	wd = eio__wd_open_sync (&tmpbuf, wd, path);
		1660	free (tmpbuf.ptr);
		1661
		1662	return wd;
		1663	}
		1664
		1665	void
		1666	eio_wd_close_sync (eio_wd wd)
		1667	{
		1668	if (wd != EIO_INVALID_WD && wd != EIO_CWD)
		1669	{
		1670	#if HAVE_AT
		1671	close (wd->fd);
		1672	#endif
		1673	free (wd);
		1674	}
		1675	}
		1676
		1677	#if HAVE_AT
		1678
		1679	/* they forgot these */
		1680
		1681	static int
		1682	eio__truncateat (int dirfd, const char *path, off_t length)
		1683	{
		1684	int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
		1685	int res;
		1686
		1687	if (fd < 0)
		1688	return fd;
		1689
		1690	res = ftruncate (fd, length);
		1691	close (fd);
		1692	return res;
		1693	}
		1694
		1695	static int
		1696	eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
		1697	{
		1698	int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
		1699	int res;
		1700
		1701	if (fd < 0)
		1702	return fd;
		1703
		1704	res = fstatvfs (fd, buf);
		1705	close (fd);
		1706	return res;
		1707
		1708	}
		1709
		1710	#endif
1785		1711
1786	/*****************************************************************************/	1712	/*****************************************************************************/
1787		1713
1788	#define ALLOC(len) \	1714	#define ALLOC(len) \
1789	if (!req->ptr2) \	1715	if (!req->ptr2) \
1790	{ \	1716	{ \
1791	X_LOCK (wrklock); \	1717	X_LOCK (EIO_POOL->wrklock); \
1792	req->flags |= EIO_FLAG_PTR2_FREE; \	1718	req->flags |= EIO_FLAG_PTR2_FREE; \
1793	X_UNLOCK (wrklock); \	1719	X_UNLOCK (EIO_POOL->wrklock); \
1794	req->ptr2 = malloc (len); \	1720	req->ptr2 = malloc (len); \
1795	if (!req->ptr2) \	1721	if (!req->ptr2) \
1796	{ \	1722	{ \
1797	errno = ENOMEM; \	1723	errno = ENOMEM; \
1798	req->result = -1; \	1724	req->result = -1; \
1799	break; \	1725	break; \
1800	} \	1726	} \
1801	}	1727	}
1802		1728
1803	X_THREAD_PROC (etp_proc)
1804	{
1805	ETP_REQ *req;
1806	struct timespec ts;
1807	etp_worker *self = (etp_worker *)thr_arg;
1808
1809	/* try to distribute timeouts somewhat randomly */
1810	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1811
1812	for (;;)
1813	{
1814	X_LOCK (reqlock);
1815
1816	for (;;)
1817	{
1818	self->req = req = reqq_shift (&req_queue);
1819
1820	if (req)
1821	break;
1822
1823	++idle;
1824
1825	ts.tv_sec = time (0) + idle_timeout;
1826	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1827	{
1828	if (idle > max_idle)
1829	{
1830	--idle;
1831	X_UNLOCK (reqlock);
1832	X_LOCK (wrklock);
1833	--started;
1834	X_UNLOCK (wrklock);
1835	goto quit;
1836	}
1837
1838	/* we are allowed to idle, so do so without any timeout */
1839	X_COND_WAIT (reqwait, reqlock);
1840	}
1841
1842	--idle;
1843	}
1844
1845	--nready;
1846
1847	X_UNLOCK (reqlock);
1848
1849	if (req->type < 0)
1850	goto quit;
1851
1852	ETP_EXECUTE (self, req);
1853
1854	X_LOCK (reslock);
1855
1856	++npending;
1857
1858	if (!reqq_push (&res_queue, req) && want_poll_cb)
1859	want_poll_cb ();
1860
1861	self->req = 0;
1862	etp_worker_clear (self);
1863
1864	X_UNLOCK (reslock);
1865	}
1866
1867	quit:
1868	X_LOCK (wrklock);
1869	etp_worker_free (self);
1870	X_UNLOCK (wrklock);
1871
1872	return 0;
1873	}
1874
1875	/*****************************************************************************/	1729	/*****************************************************************************/
1876		1730
1877	int ecb_cold	1731	int ecb_cold
1878	eio_init (void (*want_poll)(void), void (*done_poll)(void))	1732	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1879	{	1733	{
1880	return etp_init (want_poll, done_poll);	1734	eio_want_poll_cb = want_poll;
		1735	eio_done_poll_cb = done_poll;
		1736
		1737	return etp_init (EIO_POOL, 0, 0, 0);
1881	}	1738	}
1882		1739
1883	ecb_inline void	1740	ecb_inline void
1884	eio_api_destroy (eio_req *req)	1741	eio_api_destroy (eio_req *req)
1885	{	1742	{
1886	free (req);	1743	free (req);
1887	}	1744	}
1888		1745
1889	#define REQ(rtype) \	1746	#define REQ(rtype) \
1890	eio_req *req; \	1747	eio_req *req; \
1891	\	1748	\
1892	req = (eio_req *)calloc (1, sizeof *req); \	1749	req = (eio_req *)calloc (1, sizeof *req); \
1893	if (!req) \	1750	if (!req) \
1894	return 0; \	1751	return 0; \
…		…
1908	{ \	1765	{ \
1909	eio_api_destroy (req); \	1766	eio_api_destroy (req); \
1910	return 0; \	1767	return 0; \
1911	}	1768	}
1912		1769
		1770	#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
		1771
1913	static void	1772	static void
1914	eio_execute (etp_worker *self, eio_req *req)	1773	eio_execute (etp_worker *self, eio_req *req)
1915	{	1774	{
		1775	#if HAVE_AT
		1776	int dirfd;
		1777	#else
		1778	const char *path;
		1779	#endif
		1780
1916	if (ecb_expect_false (EIO_CANCELLED (req)))	1781	if (ecb_expect_false (EIO_CANCELLED (req)))
1917	{	1782	{
1918	req->result = -1;	1783	req->result = -1;
1919	req->errorno = ECANCELED;	1784	req->errorno = ECANCELED;
1920	return;	1785	return;
1921	}	1786	}
1922		1787
		1788	if (ecb_expect_false (req->wd == EIO_INVALID_WD))
		1789	{
		1790	req->result = -1;
		1791	req->errorno = ENOENT;
		1792	return;
		1793	}
		1794
		1795	if (req->type >= EIO_OPEN)
		1796	{
		1797	#if HAVE_AT
		1798	dirfd = WD2FD (req->wd);
		1799	#else
		1800	path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		1801	#endif
		1802	}
		1803
1923	switch (req->type)	1804	switch (req->type)
1924	{	1805	{
		1806	case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
		1807	req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
		1808	break;
		1809	case EIO_WD_CLOSE: req->result = 0;
		1810	eio_wd_close_sync (req->wd); break;
		1811
		1812	case EIO_SEEK: eio__lseek (req); break;
1925	case EIO_READ: ALLOC (req->size);	1813	case EIO_READ: ALLOC (req->size);
1926	req->result = req->offs >= 0	1814	req->result = req->offs >= 0
1927	? pread (req->int1, req->ptr2, req->size, req->offs)	1815	? pread (req->int1, req->ptr2, req->size, req->offs)
1928	: read (req->int1, req->ptr2, req->size); break;	1816	: read (req->int1, req->ptr2, req->size); break;
1929	case EIO_WRITE: req->result = req->offs >= 0	1817	case EIO_WRITE: req->result = req->offs >= 0
1930	? pwrite (req->int1, req->ptr2, req->size, req->offs)	1818	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1931	: write (req->int1, req->ptr2, req->size); break;	1819	: write (req->int1, req->ptr2, req->size); break;
1932		1820
1933	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	1821	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1934	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	1822	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
		1823
		1824	#if HAVE_AT
1935		1825
1936	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1826	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1937	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1827	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1938	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1828	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1829	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
		1830	case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
		1831	case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
		1832	case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
		1833	case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
		1834
		1835	case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
		1836	case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
		1837	req->result = req->wd && SINGLEDOT (req->ptr1)
		1838	? rmdir (req->wd->str)
		1839	: unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
		1840	case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
		1841	case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
		1842	req->result = req->wd && SINGLEDOT (req->ptr1)
		1843	? rename (req->wd->str, req->ptr2)
		1844	: renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
		1845	case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
		1846	case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
		1847	case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		1848	case EIO_READLINK: ALLOC (PATH_MAX);
		1849	req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
		1850	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1851	req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1852	case EIO_UTIME:
		1853	case EIO_FUTIME:
		1854	{
		1855	struct timespec ts[2];
		1856	struct timespec *times;
		1857
		1858	if (req->nv1 != -1. || req->nv2 != -1.)
		1859	{
		1860	ts[0].tv_sec = req->nv1;
		1861	ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
		1862	ts[1].tv_sec = req->nv2;
		1863	ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
		1864
		1865	times = ts;
		1866	}
		1867	else
		1868	times = 0;
		1869
		1870	req->result = req->type == EIO_FUTIME
		1871	? futimens (req->int1, times)
		1872	: utimensat (dirfd, req->ptr1, times, 0);
		1873	}
		1874	break;
		1875
		1876	#else
		1877
		1878	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1879	req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		1880	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1939	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1881	req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		1882	case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
		1883	case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
		1884	case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
		1885	case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
		1886
		1887	case EIO_UNLINK: req->result = unlink (path ); break;
		1888	case EIO_RMDIR: req->result = rmdir (path ); break;
		1889	case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
		1890	case EIO_RENAME: req->result = rename (path , req->ptr2); break;
		1891	case EIO_LINK: req->result = link (path , req->ptr2); break;
		1892	case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
		1893	case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
		1894	case EIO_READLINK: ALLOC (PATH_MAX);
		1895	req->result = readlink (path, req->ptr2, PATH_MAX); break;
		1896	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1897	req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1898
		1899	case EIO_UTIME:
		1900	case EIO_FUTIME:
		1901	{
		1902	struct timeval tv[2];
		1903	struct timeval *times;
		1904
		1905	if (req->nv1 != -1. || req->nv2 != -1.)
		1906	{
		1907	tv[0].tv_sec = req->nv1;
		1908	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
		1909	tv[1].tv_sec = req->nv2;
		1910	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
		1911
		1912	times = tv;
		1913	}
		1914	else
		1915	times = 0;
		1916
		1917	req->result = req->type == EIO_FUTIME
		1918	? futimes (req->int1, times)
		1919	: utimes (req->ptr1, times);
		1920	}
		1921	break;
		1922
		1923	#endif
		1924
		1925	case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
		1926	{
		1927	ALLOC (req->result);
		1928	memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
		1929	}
		1930	break;
		1931
1940	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1932	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1941	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1933	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1942		1934
1943	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1944	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1945	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));	1935	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1946	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;	1936	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1947		1937
1948	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1949	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	1938	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1950	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1951	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	1939	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1952	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1953	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;	1940	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1954		1941
1955	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1956	case EIO_CLOSE: req->result = close (req->int1); break;	1942	case EIO_CLOSE: req->result = close (req->int1); break;
1957	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;	1943	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1958	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1959	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1960	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1961	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1962	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1963	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1964	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1965
1966	case EIO_REALPATH: eio__realpath (req, self); break;
1967
1968	case EIO_READLINK: ALLOC (PATH_MAX);
1969	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1970
1971	case EIO_SYNC: req->result = 0; sync (); break;	1944	case EIO_SYNC: req->result = 0; sync (); break;
1972	case EIO_FSYNC: req->result = fsync (req->int1); break;	1945	case EIO_FSYNC: req->result = fsync (req->int1); break;
1973	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	1946	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		1947	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		1948	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1974	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	1949	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1975	case EIO_MTOUCH: req->result = eio__mtouch (req); break;	1950	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1976	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	1951	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1977	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	1952	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1978	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1979	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;	1953	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1980		1954
1981	case EIO_READDIR: eio__scandir (req, self); break;	1955	case EIO_READDIR: eio__scandir (req, self); break;
1982		1956
1983	case EIO_BUSY:	1957	case EIO_BUSY:
…		…
1993	req->result = select (0, 0, 0, 0, &tv);	1967	req->result = select (0, 0, 0, 0, &tv);
1994	}	1968	}
1995	#endif	1969	#endif
1996	break;	1970	break;
1997		1971
1998	case EIO_UTIME:	1972	#if 0
1999	case EIO_FUTIME:
2000	{
2001	struct timeval tv[2];
2002	struct timeval *times;
2003
2004	if (req->nv1 != -1. || req->nv2 != -1.)
2005	{
2006	tv[0].tv_sec = req->nv1;
2007	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2008	tv[1].tv_sec = req->nv2;
2009	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2010
2011	times = tv;
2012	}
2013	else
2014	times = 0;
2015
2016	req->result = req->type == EIO_FUTIME
2017	? futimes (req->int1, times)
2018	: utimes (req->ptr1, times);
2019	}
2020	break;
2021
2022	case EIO_GROUP:	1973	case EIO_GROUP:
2023	abort (); /* handled in eio_request */	1974	abort (); /* handled in eio_request */
		1975	#endif
2024		1976
2025	case EIO_NOP:	1977	case EIO_NOP:
2026	req->result = 0;	1978	req->result = 0;
2027	break;	1979	break;
2028		1980
2029	case EIO_CUSTOM:	1981	case EIO_CUSTOM:
2030	req->feed (req);	1982	req->feed (req);
2031	break;	1983	break;
2032		1984
2033	default:	1985	default:
2034	errno = ENOSYS;
2035	req->result = -1;	1986	req->result = EIO_ENOSYS ();
2036	break;	1987	break;
2037	}	1988	}
2038		1989
2039	req->errorno = errno;	1990	req->errorno = errno;
2040	}	1991	}
2041		1992
2042	#ifndef EIO_NO_WRAPPERS	1993	#ifndef EIO_NO_WRAPPERS
2043		1994
		1995	eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
		1996	{
		1997	REQ (EIO_WD_OPEN); PATH; SEND;
		1998	}
		1999
		2000	eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
		2001	{
		2002	REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
		2003	}
		2004
2044	eio_req *eio_nop (int pri, eio_cb cb, void *data)	2005	eio_req *eio_nop (int pri, eio_cb cb, void *data)
2045	{	2006	{
2046	REQ (EIO_NOP); SEND;	2007	REQ (EIO_NOP); SEND;
2047	}	2008	}
2048		2009
…		…
2064	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2025	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2065	{	2026	{
2066	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2027	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2067	}	2028	}
2068		2029
		2030	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		2031	{
		2032	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2033	}
		2034
		2035	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2036	{
		2037	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2038	}
		2039
		2040	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2041	{
		2042	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2043	}
		2044
2069	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2045	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2070	{	2046	{
2071	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2047	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2072	}	2048	}
2073		2049
…		…
2079	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)	2055	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2080	{	2056	{
2081	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;	2057	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2082	}	2058	}
2083		2059
2084	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2085	{
2086	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2087	}
2088
2089	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)	2060	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2090	{	2061	{
2091	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;	2062	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2092	}	2063	}
2093		2064
2094	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2095	{
2096	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2097	}
2098
2099	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2065	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2100	{	2066	{
2101	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2067	REQ (EIO_CLOSE); req->int1 = fd; SEND;
2102	}	2068	}
2103		2069
2104	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)	2070	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2105	{	2071	{
2106	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;	2072	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2107	}	2073	}
2108		2074
		2075	eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
		2076	{
		2077	REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
		2078	}
		2079
2109	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)	2080	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2110	{	2081	{
2111	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;	2082	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2112	}	2083	}
2113		2084
…		…
2139	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2110	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2140	{	2111	{
2141	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2112	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2142	}	2113	}
2143		2114
2144	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2115	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2145	{	2116	{
2146	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2117	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2147	}	2118	}
2148		2119
2149	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2120	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
2169	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2140	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2170	{	2141	{
2171	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2142	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2172	}	2143	}
2173		2144
2174	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2145	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2175	{	2146	{
2176	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2147	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2177	}	2148	}
2178		2149
2179	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2150	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
2309	void	2280	void
2310	eio_grp_add (eio_req *grp, eio_req *req)	2281	eio_grp_add (eio_req *grp, eio_req *req)
2311	{	2282	{
2312	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2283	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2313		2284
2314	grp->flags |= EIO_FLAG_GROUPADD;	2285	grp->flags |= ETP_FLAG_GROUPADD;
2315		2286
2316	++grp->size;	2287	++grp->size;
2317	req->grp = grp;	2288	req->grp = grp;
2318		2289
2319	req->grp_prev = 0;	2290	req->grp_prev = 0;
…		…
2326	}	2297	}
2327		2298
2328	/*****************************************************************************/	2299	/*****************************************************************************/
2329	/* misc garbage */	2300	/* misc garbage */
2330		2301
2331	ssize_t	2302	eio_ssize_t
2332	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2303	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2333	{	2304	{
2334	etp_worker wrk;
2335	ssize_t ret;
2336
2337	wrk.dbuf = 0;
2338
2339	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2305	return eio__sendfile (ofd, ifd, offset, count);
2340
2341	if (wrk.dbuf)
2342	free (wrk.dbuf);
2343
2344	return ret;
2345	}	2306	}
2346		2307

Diff Legend

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