ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/cvsroot/libeio/eio.c

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.81 by root, Tue Jul 5 20:34:42 2011 UTC vs.
Revision 1.114 by root, Fri Dec 30 07:38:00 2011 UTC

…		…
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
		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	/* rename() uses MoveFile, which fails to overwrite */
		138	#define rename(old,neu) eio__rename (old, neu)
		139
		140	static int
		141	eio__rename (const char *old, const char *neu)
		142	{
		143	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		144	return 0;
		145
		146	/* should steal _dosmaperr */
		147	switch (GetLastError ())
		148	{
		149	case ERROR_FILE_NOT_FOUND:
		150	case ERROR_PATH_NOT_FOUND:
		151	case ERROR_INVALID_DRIVE:
		152	case ERROR_NO_MORE_FILES:
		153	case ERROR_BAD_NETPATH:
		154	case ERROR_BAD_NET_NAME:
		155	case ERROR_BAD_PATHNAME:
		156	case ERROR_FILENAME_EXCED_RANGE:
		157	errno = ENOENT;
		158	break;
		159
		160	default:
		161	errno = EACCES;
		162	break;
		163	}
		164
		165	return -1;
		166	}
		167
		168	/* we could even stat and see if it exists */
		169	static int
		170	symlink (const char *old, const char *neu)
		171	{
		172	#if WINVER >= 0x0600
		173	if (CreateSymbolicLink (neu, old, 1))
		174	return 0;
		175
		176	if (CreateSymbolicLink (neu, old, 0))
		177	return 0;
		178	#endif
		179
		180	return EIO_ERRNO (ENOENT, -1);
		181	}
		182
		183	/* POSIX API only */
		184	#define CreateHardLink(neu,old,flags) 0
		185	#define CreateSymbolicLink(neu,old,flags) 0
		186
		187	struct statvfs
		188	{
		189	int dummy;
		190	};
		191
		192	#define DT_DIR EIO_DT_DIR
		193	#define DT_REG EIO_DT_REG
		194	#define D_NAME(entp) entp.cFileName
		195	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
		196
91	#else	197	#else
92		198
93	# include <sys/time.h>	199	#include <sys/time.h>
94	# include <sys/select.h>	200	#include <sys/select.h>
		201	#include <sys/statvfs.h>
95	# include <unistd.h>	202	#include <unistd.h>
		203	#include <signal.h>
		204	#include <dirent.h>
		205
		206	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		207	#include <sys/mman.h>
		208	#endif
		209
		210	#define D_NAME(entp) entp->d_name
		211
		212	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		213	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		214	#define _DIRENT_HAVE_D_TYPE /* sigh */
		215	#define D_INO(de) (de)->d_fileno
		216	#define D_NAMLEN(de) (de)->d_namlen
		217	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		218	#define D_INO(de) (de)->d_ino
		219	#endif
		220
		221	#ifdef _D_EXACT_NAMLEN
		222	#undef D_NAMLEN
		223	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		224	#endif
		225
		226	#ifdef _DIRENT_HAVE_D_TYPE
		227	#define D_TYPE(de) (de)->d_type
		228	#endif
		229
		230	#ifndef EIO_STRUCT_DIRENT
		231	#define EIO_STRUCT_DIRENT struct dirent
		232	#endif
		233
		234	#endif
		235
		236	#if HAVE_UTIMES
96	# include <utime.h>	237	# 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	238	#endif
112		239
113	#ifdef _D_EXACT_NAMLEN	240	#if HAVE_SYS_SYSCALL_H
114	# undef D_NAMLEN	241	# 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	242	#endif
121		243
122	# ifndef EIO_STRUCT_DIRENT	244	#if HAVE_SYS_PRCTL_H
123	# define EIO_STRUCT_DIRENT struct dirent	245	# include <sys/prctl.h>
124	# endif
125
126	#endif	246	#endif
127		247
128	#if HAVE_SENDFILE	248	#if HAVE_SENDFILE
129	# if __linux	249	# if __linux
130	# include <sys/sendfile.h>	250	# include <sys/sendfile.h>
…		…
145	#endif	265	#endif
146	#ifndef D_INO	266	#ifndef D_INO
147	# define D_INO(de) 0	267	# define D_INO(de) 0
148	#endif	268	#endif
149	#ifndef D_NAMLEN	269	#ifndef D_NAMLEN
150	# define D_NAMLEN(de) strlen ((de)->d_name)	270	# define D_NAMLEN(entp) strlen (D_NAME (entp))
151	#endif	271	#endif
152		272
153	/* used for struct dirent, AIX doesn't provide it */	273	/* used for struct dirent, AIX doesn't provide it */
154	#ifndef NAME_MAX	274	#ifndef NAME_MAX
155	# define NAME_MAX 4096	275	# define NAME_MAX 4096
…		…
162		282
163	/* buffer size for various temporary buffers */	283	/* buffer size for various temporary buffers */
164	#define EIO_BUFSIZE 65536	284	#define EIO_BUFSIZE 65536
165		285
166	#define dBUF \	286	#define dBUF \
167	char *eio_buf; \
168	ETP_WORKER_LOCK (self); \
169	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	287	char *eio_buf = malloc (EIO_BUFSIZE); \
170	ETP_WORKER_UNLOCK (self); \
171	errno = ENOMEM; \	288	errno = ENOMEM; \
172	if (!eio_buf) \	289	if (!eio_buf) \
173	return -1;	290	return -1
		291
		292	#define FUBd \
		293	free (eio_buf)
174		294
175	#define EIO_TICKS ((1000000 + 1023) >> 10)	295	#define EIO_TICKS ((1000000 + 1023) >> 10)
		296
		297	/*****************************************************************************/
		298
		299	struct tmpbuf
		300	{
		301	void *ptr;
		302	int len;
		303	};
		304
		305	static void *
		306	tmpbuf_get (struct tmpbuf *buf, int len)
		307	{
		308	if (buf->len < len)
		309	{
		310	free (buf->ptr);
		311	buf->ptr = malloc (buf->len = len);
		312	}
		313
		314	return buf->ptr;
		315	}
		316
		317	struct tmpbuf;
		318
		319	#if _POSIX_VERSION >= 200809L
		320	#define HAVE_AT 1
		321	#define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
		322	#ifndef O_SEARCH
		323	#define O_SEARCH O_RDONLY
		324	#endif
		325	#else
		326	#define HAVE_AT 0
		327	static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
		328	#endif
		329
		330	struct eio_pwd
		331	{
		332	#if HAVE_AT
		333	int fd;
		334	#endif
		335	int len;
		336	char str[1]; /* actually, a 0-terminated canonical path */
		337	};
		338
		339	/*****************************************************************************/
176		340
177	#define ETP_PRI_MIN EIO_PRI_MIN	341	#define ETP_PRI_MIN EIO_PRI_MIN
178	#define ETP_PRI_MAX EIO_PRI_MAX	342	#define ETP_PRI_MAX EIO_PRI_MAX
179		343
180	struct etp_worker;	344	struct etp_worker;
…		…
184	static int eio_finish (eio_req *req);	348	static int eio_finish (eio_req *req);
185	#define ETP_FINISH(req) eio_finish (req)	349	#define ETP_FINISH(req) eio_finish (req)
186	static void eio_execute (struct etp_worker *self, eio_req *req);	350	static void eio_execute (struct etp_worker *self, eio_req *req);
187	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	351	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
188		352
189	#define ETP_WORKER_CLEAR(req) \
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
202	#define ETP_WORKER_COMMON \
203	void *dbuf; \
204	DIR *dirp;
205
206	/*****************************************************************************/	353	/*****************************************************************************/
207		354
208	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	355	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209		356
210	/* calculate time difference in ~1/EIO_TICKS of a second */	357	/* calculate time difference in ~1/EIO_TICKS of a second */
…		…
221	static void (*done_poll_cb) (void);	368	static void (*done_poll_cb) (void);
222		369
223	static unsigned int max_poll_time; /* reslock */	370	static unsigned int max_poll_time; /* reslock */
224	static unsigned int max_poll_reqs; /* reslock */	371	static unsigned int max_poll_reqs; /* reslock */
225		372
226	static volatile unsigned int nreqs; /* reqlock */	373	static unsigned int nreqs; /* reqlock */
227	static volatile unsigned int nready; /* reqlock */	374	static unsigned int nready; /* reqlock */
228	static volatile unsigned int npending; /* reqlock */	375	static unsigned int npending; /* reqlock */
229	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */	376	static 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 */	377	static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
231		378
232	static xmutex_t wrklock;	379	static xmutex_t wrklock;
233	static xmutex_t reslock;	380	static xmutex_t reslock;
234	static xmutex_t reqlock;	381	static xmutex_t reqlock;
235	static xcond_t reqwait;	382	static xcond_t reqwait;
…		…
238	/*	385	/*
239	* make our pread/pwrite emulation safe against themselves, but not against	386	* make our pread/pwrite emulation safe against themselves, but not against
240	* normal read/write by using a mutex. slows down execution a lot,	387	* normal read/write by using a mutex. slows down execution a lot,
241	* but that's your problem, not mine.	388	* but that's your problem, not mine.
242	*/	389	*/
243	static xmutex_t preadwritelock = X_MUTEX_INIT;	390	static xmutex_t preadwritelock;
244	#endif	391	#endif
245		392
246	typedef struct etp_worker	393	typedef struct etp_worker
247	{	394	{
		395	struct tmpbuf tmpbuf;
		396
248	/* locked by wrklock */	397	/* locked by wrklock */
249	struct etp_worker *prev, *next;	398	struct etp_worker *prev, *next;
250		399
251	xthread_t tid;	400	xthread_t tid;
252		401
253	/* locked by reslock, reqlock or wrklock */	402	#ifdef ETP_WORKER_COMMON
254	ETP_REQ *req; /* currently processed request */
255
256	ETP_WORKER_COMMON	403	ETP_WORKER_COMMON
		404	#endif
257	} etp_worker;	405	} etp_worker;
258		406
259	static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */	407	static etp_worker wrk_first; /* NOT etp */
260		408
261	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	409	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
262	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	410	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
263		411
264	/* worker threads management */	412	/* worker threads management */
265		413
266	static void ecb_cold	414	static void
267	etp_worker_clear (etp_worker *wrk)	415	etp_worker_clear (etp_worker *wrk)
268	{	416	{
269	ETP_WORKER_CLEAR (wrk);
270	}	417	}
271		418
272	static void ecb_cold	419	static void ecb_cold
273	etp_worker_free (etp_worker *wrk)	420	etp_worker_free (etp_worker *wrk)
274	{	421	{
		422	free (wrk->tmpbuf.ptr);
		423
275	wrk->next->prev = wrk->prev;	424	wrk->next->prev = wrk->prev;
276	wrk->prev->next = wrk->next;	425	wrk->prev->next = wrk->next;
277		426
278	free (wrk);	427	free (wrk);
279	}	428	}
…		…
335	} etp_reqq;	484	} etp_reqq;
336		485
337	static etp_reqq req_queue;	486	static etp_reqq req_queue;
338	static etp_reqq res_queue;	487	static etp_reqq res_queue;
339		488
		489	static void ecb_noinline ecb_cold
		490	reqq_init (etp_reqq *q)
		491	{
		492	int pri;
		493
		494	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		495	q->qs[pri] = q->qe[pri] = 0;
		496
		497	q->size = 0;
		498	}
		499
340	static int ecb_noinline	500	static int ecb_noinline
341	reqq_push (etp_reqq *q, ETP_REQ *req)	501	reqq_push (etp_reqq *q, ETP_REQ *req)
342	{	502	{
343	int pri = req->pri;	503	int pri = req->pri;
344	req->next = 0;	504	req->next = 0;
…		…
378	}	538	}
379		539
380	abort ();	540	abort ();
381	}	541	}
382		542
383	static void ecb_cold	543	static int ecb_cold
384	etp_thread_init (void)	544	etp_init (void (*want_poll)(void), void (*done_poll)(void))
385	{	545	{
386	X_MUTEX_CREATE (wrklock);	546	X_MUTEX_CREATE (wrklock);
387	X_MUTEX_CREATE (reslock);	547	X_MUTEX_CREATE (reslock);
388	X_MUTEX_CREATE (reqlock);	548	X_MUTEX_CREATE (reqlock);
389	X_COND_CREATE (reqwait);	549	X_COND_CREATE (reqwait);
390	}
391		550
392	static void ecb_cold	551	reqq_init (&req_queue);
393	etp_atfork_prepare (void)	552	reqq_init (&res_queue);
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		553
403	static void ecb_cold	554	wrk_first.next =
404	etp_atfork_parent (void)	555	wrk_first.prev = &wrk_first;
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		556
436	started = 0;	557	started = 0;
437	idle = 0;	558	idle = 0;
438	nreqs = 0;	559	nreqs = 0;
439	nready = 0;	560	nready = 0;
440	npending = 0;	561	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		562
459	want_poll_cb = want_poll;	563	want_poll_cb = want_poll;
460	done_poll_cb = done_poll;	564	done_poll_cb = done_poll;
461		565
462	return 0;	566	return 0;
…		…
502	}	606	}
503		607
504	static void ecb_cold	608	static void ecb_cold
505	etp_end_thread (void)	609	etp_end_thread (void)
506	{	610	{
507	eio_req *req = calloc (1, sizeof (eio_req));	611	eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
508		612
509	req->type = -1;	613	req->type = -1;
510	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	614	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
511		615
512	X_LOCK (reqlock);	616	X_LOCK (reqlock);
…		…
845	# undef pread	949	# undef pread
846	# undef pwrite	950	# undef pwrite
847	# define pread eio__pread	951	# define pread eio__pread
848	# define pwrite eio__pwrite	952	# define pwrite eio__pwrite
849		953
850	static ssize_t	954	static eio_ssize_t
851	eio__pread (int fd, void *buf, size_t count, off_t offset)	955	eio__pread (int fd, void *buf, size_t count, off_t offset)
852	{	956	{
853	ssize_t res;	957	eio_ssize_t res;
854	off_t ooffset;	958	off_t ooffset;
855		959
856	X_LOCK (preadwritelock);	960	X_LOCK (preadwritelock);
857	ooffset = lseek (fd, 0, SEEK_CUR);	961	ooffset = lseek (fd, 0, SEEK_CUR);
858	lseek (fd, offset, SEEK_SET);	962	lseek (fd, offset, SEEK_SET);
…		…
861	X_UNLOCK (preadwritelock);	965	X_UNLOCK (preadwritelock);
862		966
863	return res;	967	return res;
864	}	968	}
865		969
866	static ssize_t	970	static eio_ssize_t
867	eio__pwrite (int fd, void *buf, size_t count, off_t offset)	971	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
868	{	972	{
869	ssize_t res;	973	eio_ssize_t res;
870	off_t ooffset;	974	off_t ooffset;
871		975
872	X_LOCK (preadwritelock);	976	X_LOCK (preadwritelock);
873	ooffset = lseek (fd, 0, SEEK_CUR);	977	ooffset = lseek (fd, 0, SEEK_CUR);
874	lseek (fd, offset, SEEK_SET);	978	lseek (fd, offset, SEEK_SET);
…		…
919		1023
920	#if !HAVE_FDATASYNC	1024	#if !HAVE_FDATASYNC
921	# undef fdatasync	1025	# undef fdatasync
922	# define fdatasync(fd) fsync (fd)	1026	# define fdatasync(fd) fsync (fd)
923	#endif	1027	#endif
		1028
		1029	static int
		1030	eio__syncfs (int fd)
		1031	{
		1032	int res;
		1033
		1034	#if HAVE_SYS_SYNCFS
		1035	res = (int)syscall (__NR_syncfs, (int)(fd));
		1036	#else
		1037	res = -1;
		1038	errno = ENOSYS;
		1039	#endif
		1040
		1041	if (res < 0 && errno == ENOSYS && fd >= 0)
		1042	sync ();
		1043
		1044	return res;
		1045	}
924		1046
925	/* sync_file_range always needs emulation */	1047	/* sync_file_range always needs emulation */
926	static int	1048	static int
927	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)	1049	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
928	{	1050	{
…		…
948	/* even though we could play tricks with the flags, it's better to always	1070	/* even though we could play tricks with the flags, it's better to always
949	* call fdatasync, as that matches the expectation of its users best */	1071	* call fdatasync, as that matches the expectation of its users best */
950	return fdatasync (fd);	1072	return fdatasync (fd);
951	}	1073	}
952		1074
		1075	static int
		1076	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		1077	{
		1078	#if HAVE_FALLOCATE
		1079	return fallocate (fd, mode, offset, len);
		1080	#else
		1081	errno = ENOSYS;
		1082	return -1;
		1083	#endif
		1084	}
		1085
953	#if !HAVE_READAHEAD	1086	#if !HAVE_READAHEAD
954	# undef readahead	1087	# undef readahead
955	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	1088	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
956		1089
957	static ssize_t	1090	static eio_ssize_t
958	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1091	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
959	{	1092	{
960	size_t todo = count;	1093	size_t todo = count;
961	dBUF;	1094	dBUF;
962		1095
…		…
967	pread (fd, eio_buf, len, offset);	1100	pread (fd, eio_buf, len, offset);
968	offset += len;	1101	offset += len;
969	todo -= len;	1102	todo -= len;
970	}	1103	}
971		1104
		1105	FUBd;
		1106
972	errno = 0;	1107	errno = 0;
973	return count;	1108	return count;
974	}	1109	}
975		1110
976	#endif	1111	#endif
977		1112
978	/* sendfile always needs emulation */	1113	/* sendfile always needs emulation */
979	static ssize_t	1114	static eio_ssize_t
980	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1115	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
981	{	1116	{
982	ssize_t written = 0;	1117	eio_ssize_t written = 0;
983	ssize_t res;	1118	eio_ssize_t res;
984		1119
985	if (!count)	1120	if (!count)
986	return 0;	1121	return 0;
987		1122
988	for (;;)	1123	for (;;)
989	{	1124	{
		1125	#ifdef __APPLE__
		1126	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1127	#endif
990	#if HAVE_SENDFILE	1128	#if HAVE_SENDFILE
991	# if __linux	1129	# if __linux
992	off_t soffset = offset;	1130	off_t soffset = offset;
993	res = sendfile (ofd, ifd, &soffset, count);	1131	res = sendfile (ofd, ifd, &soffset, count);
994		1132
…		…
1010		1148
1011	/* according to source inspection, this is correct, and useful behaviour */	1149	/* according to source inspection, this is correct, and useful behaviour */
1012	if (sbytes)	1150	if (sbytes)
1013	res = sbytes;	1151	res = sbytes;
1014		1152
1015	# elif defined (__APPLE__) && 0 /* broken, as everything on os x */	1153	# elif defined (__APPLE__)
1016	off_t sbytes = count;	1154	off_t sbytes = count;
1017	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	1155	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1018		1156
1019	/* according to the manpage, sbytes is always valid */	1157	/* according to the manpage, sbytes is always valid */
1020	if (sbytes)	1158	if (sbytes)
…		…
1037	if (res < 0 && sbytes)	1175	if (res < 0 && sbytes)
1038	res = sbytes;	1176	res = sbytes;
1039		1177
1040	# endif	1178	# endif
1041		1179
1042	#elif defined (_WIN32)	1180	#elif defined (_WIN32) && 0
1043	/* does not work, just for documentation of what would need to be done */	1181	/* does not work, just for documentation of what would need to be done */
1044	/* actually, cannot be done like this, as TransmitFile changes the file offset, */	1182	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
1045	/* libeio guarantees that the file offset does not change, and windows */	1183	/* libeio guarantees that the file offset does not change, and windows */
1046	/* has no way to get an independent handle to the same file description */	1184	/* has no way to get an independent handle to the same file description */
1047	HANDLE h = TO_SOCKET (ifd);	1185	HANDLE h = TO_SOCKET (ifd);
…		…
1080	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1218	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1081	/* BSDs */	1219	/* BSDs */
1082	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */	1220	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1083	|| errno == ENOTSUP	1221	|| errno == ENOTSUP
1084	#endif	1222	#endif
		1223	#ifdef EOPNOTSUPP /* windows */
1085	|| errno == EOPNOTSUPP /* BSDs */	1224	|| errno == EOPNOTSUPP /* BSDs */
		1225	#endif
1086	#if __solaris	1226	#if __solaris
1087	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1227	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
1088	#endif	1228	#endif
1089	)	1229	)
1090	)	1230	)
…		…
1094		1234
1095	res = 0;	1235	res = 0;
1096		1236
1097	while (count)	1237	while (count)
1098	{	1238	{
1099	ssize_t cnt;	1239	eio_ssize_t cnt;
1100		1240
1101	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1241	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1102		1242
1103	if (cnt <= 0)	1243	if (cnt <= 0)
1104	{	1244	{
…		…
1116		1256
1117	offset += cnt;	1257	offset += cnt;
1118	res += cnt;	1258	res += cnt;
1119	count -= cnt;	1259	count -= cnt;
1120	}	1260	}
		1261
		1262	FUBd;
1121	}	1263	}
1122		1264
1123	return res;	1265	return res;
1124	}	1266	}
1125		1267
…		…
1152	/* round up length */	1294	/* round up length */
1153	*length = (*length + mask) & ~mask;	1295	*length = (*length + mask) & ~mask;
1154	}	1296	}
1155		1297
1156	#if !_POSIX_MEMLOCK	1298	#if !_POSIX_MEMLOCK
1157	# define eio__mlockall(a) ((errno = ENOSYS), -1)	1299	# define eio__mlockall(a) EIO_ENOSYS ()
1158	#else	1300	#else
1159		1301
1160	static int	1302	static int
1161	eio__mlockall (int flags)	1303	eio__mlockall (int flags)
1162	{	1304	{
…		…
1176	return mlockall (flags);	1318	return mlockall (flags);
1177	}	1319	}
1178	#endif	1320	#endif
1179		1321
1180	#if !_POSIX_MEMLOCK_RANGE	1322	#if !_POSIX_MEMLOCK_RANGE
1181	# define eio__mlock(a,b) ((errno = ENOSYS), -1)	1323	# define eio__mlock(a,b) EIO_ENOSYS ()
1182	#else	1324	#else
1183		1325
1184	static int	1326	static int
1185	eio__mlock (void *addr, size_t length)	1327	eio__mlock (void *addr, size_t length)
1186	{	1328	{
…		…
1190	}	1332	}
1191		1333
1192	#endif	1334	#endif
1193		1335
1194	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)	1336	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1195	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)	1337	# define eio__msync(a,b,c) EIO_ENOSYS ()
1196	#else	1338	#else
1197		1339
1198	static int	1340	static int
1199	eio__msync (void *mem, size_t len, int flags)	1341	eio__msync (void *mem, size_t len, int flags)
1200	{	1342	{
…		…
1240	}	1382	}
1241		1383
1242	/*****************************************************************************/	1384	/*****************************************************************************/
1243	/* requests implemented outside eio_execute, because they are so large */	1385	/* requests implemented outside eio_execute, because they are so large */
1244		1386
		1387	/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1245	static void	1388	static int
1246	eio__realpath (eio_req *req, etp_worker *self)	1389	eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1247	{	1390	{
1248	char *rel = req->ptr1;	1391	const char *rel = path;
1249	char *res;	1392	char *res;
1250	char *tmp1, *tmp2;	1393	char *tmp1, *tmp2;
1251	#if SYMLOOP_MAX > 32	1394	#if SYMLOOP_MAX > 32
1252	int symlinks = SYMLOOP_MAX;	1395	int symlinks = SYMLOOP_MAX;
1253	#else	1396	#else
1254	int symlinks = 32;	1397	int symlinks = 32;
1255	#endif	1398	#endif
1256		1399
1257	req->result = -1;
1258
1259	errno = EINVAL;	1400	errno = EINVAL;
1260	if (!rel)	1401	if (!rel)
1261	return;	1402	return -1;
1262		1403
1263	errno = ENOENT;	1404	errno = ENOENT;
1264	if (!*rel)	1405	if (!*rel)
1265	return;	1406	return -1;
1266		1407
1267	if (!req->ptr2)	1408	res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1268	{
1269	X_LOCK (wrklock);
1270	req->flags |= EIO_FLAG_PTR2_FREE;
1271	X_UNLOCK (wrklock);
1272	req->ptr2 = malloc (PATH_MAX * 3);
1273
1274	errno = ENOMEM;
1275	if (!req->ptr2)
1276	return;
1277	}
1278
1279	res = req->ptr2;
1280	tmp1 = res + PATH_MAX;	1409	tmp1 = res + PATH_MAX;
1281	tmp2 = tmp1 + PATH_MAX;	1410	tmp2 = tmp1 + PATH_MAX;
1282		1411
1283	#if 0 /* disabled, the musl way to do things is just too racy */	1412	#if 0 /* disabled, the musl way to do things is just too racy */
1284	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)	1413	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
…		…
1303	#endif	1432	#endif
1304	#endif	1433	#endif
1305		1434
1306	if (*rel != '/')	1435	if (*rel != '/')
1307	{	1436	{
		1437	int len;
		1438
		1439	errno = ENOENT;
		1440	if (wd == EIO_INVALID_WD)
		1441	return -1;
		1442
		1443	if (wd == EIO_CWD)
		1444	{
1308	if (!getcwd (res, PATH_MAX))	1445	if (!getcwd (res, PATH_MAX))
1309	return;	1446	return -1;
		1447
		1448	len = strlen (res);
		1449	}
		1450	else
		1451	memcpy (res, wd->str, len = wd->len);
1310		1452
1311	if (res [1]) /* only use if not / */	1453	if (res [1]) /* only use if not / */
1312	res += strlen (res);	1454	res += len;
1313	}	1455	}
1314		1456
1315	while (*rel)	1457	while (*rel)
1316	{	1458	{
1317	ssize_t len, linklen;	1459	eio_ssize_t len, linklen;
1318	char *beg = rel;	1460	const char *beg = rel;
1319		1461
1320	while (*rel && *rel != '/')	1462	while (*rel && *rel != '/')
1321	++rel;	1463	++rel;
1322		1464
1323	len = rel - beg;	1465	len = rel - beg;
…		…
1335		1477
1336	if (beg [1] == '.' && len == 2)	1478	if (beg [1] == '.' && len == 2)
1337	{	1479	{
1338	/* .. - back up one component, if possible */	1480	/* .. - back up one component, if possible */
1339		1481
1340	while (res != req->ptr2)	1482	while (res != tmpbuf->ptr)
1341	if (*--res == '/')	1483	if (*--res == '/')
1342	break;	1484	break;
1343		1485
1344	continue;	1486	continue;
1345	}	1487	}
1346	}	1488	}
1347		1489
1348	errno = ENAMETOOLONG;	1490	errno = ENAMETOOLONG;
1349	if (res + 1 + len + 1 >= tmp1)	1491	if (res + 1 + len + 1 >= tmp1)
1350	return;	1492	return -1;
1351		1493
1352	/* copy one component */	1494	/* copy one component */
1353	*res = '/';	1495	*res = '/';
1354	memcpy (res + 1, beg, len);	1496	memcpy (res + 1, beg, len);
1355		1497
1356	/* zero-terminate, for readlink */	1498	/* zero-terminate, for readlink */
1357	res [len + 1] = 0;	1499	res [len + 1] = 0;
1358		1500
1359	/* now check if it's a symlink */	1501	/* now check if it's a symlink */
1360	linklen = readlink (req->ptr2, tmp1, PATH_MAX);	1502	linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1361		1503
1362	if (linklen < 0)	1504	if (linklen < 0)
1363	{	1505	{
1364	if (errno != EINVAL)	1506	if (errno != EINVAL)
1365	return;	1507	return -1;
1366		1508
1367	/* it's a normal directory. hopefully */	1509	/* it's a normal directory. hopefully */
1368	res += len + 1;	1510	res += len + 1;
1369	}	1511	}
1370	else	1512	else
…		…
1372	/* yay, it was a symlink - build new path in tmp2 */	1514	/* yay, it was a symlink - build new path in tmp2 */
1373	int rellen = strlen (rel);	1515	int rellen = strlen (rel);
1374		1516
1375	errno = ENAMETOOLONG;	1517	errno = ENAMETOOLONG;
1376	if (linklen + 1 + rellen >= PATH_MAX)	1518	if (linklen + 1 + rellen >= PATH_MAX)
1377	return;	1519	return -1;
1378		1520
1379	errno = ELOOP;	1521	errno = ELOOP;
1380	if (!--symlinks)	1522	if (!--symlinks)
1381	return;	1523	return -1;
1382		1524
1383	if (*tmp1 == '/')	1525	if (*tmp1 == '/')
1384	res = req->ptr2; /* symlink resolves to an absolute path */	1526	res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1385		1527
1386	/* we need to be careful, as rel might point into tmp2 already */	1528	/* we need to be careful, as rel might point into tmp2 already */
1387	memmove (tmp2 + linklen + 1, rel, rellen + 1);	1529	memmove (tmp2 + linklen + 1, rel, rellen + 1);
1388	tmp2 [linklen] = '/';	1530	tmp2 [linklen] = '/';
1389	memcpy (tmp2, tmp1, linklen);	1531	memcpy (tmp2, tmp1, linklen);
…		…
1391	rel = tmp2;	1533	rel = tmp2;
1392	}	1534	}
1393	}	1535	}
1394		1536
1395	/* special case for the lone root path */	1537	/* special case for the lone root path */
1396	if (res == req->ptr2)	1538	if (res == tmpbuf->ptr)
1397	*res++ = '/';	1539	*res++ = '/';
1398		1540
1399	req->result = res - (char *)req->ptr2;	1541	return res - (char *)tmpbuf->ptr;
1400
1401	done:
1402	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1403	}	1542	}
1404		1543
1405	static signed char	1544	static signed char
1406	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1545	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1407	{	1546	{
…		…
1415		1554
1416	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */	1555	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1417	#define EIO_SORT_FAST 60 /* when to only use insertion sort */	1556	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1418		1557
1419	static void	1558	static void
1420	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1559	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1421	{	1560	{
1422	unsigned char bits [9 + sizeof (ino_t) * 8];	1561	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1423	unsigned char *bit = bits;	1562	unsigned char *bit = bits;
1424		1563
1425	assert (CHAR_BIT == 8);	1564	assert (CHAR_BIT == 8);
1426	assert (sizeof (eio_dirent) * 8 < 256);	1565	assert (sizeof (eio_dirent) * 8 < 256);
1427	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */	1566	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
…		…
1429		1568
1430	if (size <= EIO_SORT_FAST)	1569	if (size <= EIO_SORT_FAST)
1431	return;	1570	return;
1432		1571
1433	/* first prepare an array of bits to test in our radix sort */	1572	/* first prepare an array of bits to test in our radix sort */
1434	/* try to take endianness into account, as well as differences in ino_t sizes */	1573	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
1435	/* inode_bits must contain all inodes ORed together */	1574	/* inode_bits must contain all inodes ORed together */
1436	/* which is used to skip bits that are 0 everywhere, which is very common */	1575	/* which is used to skip bits that are 0 everywhere, which is very common */
1437	{	1576	{
1438	ino_t endianness;	1577	eio_ino_t endianness;
1439	int i, j;	1578	int i, j;
1440		1579
1441	/* we store the byte offset of byte n into byte n of "endianness" */	1580	/* we store the byte offset of byte n into byte n of "endianness" */
1442	for (i = 0; i < sizeof (ino_t); ++i)	1581	for (i = 0; i < sizeof (eio_ino_t); ++i)
1443	((unsigned char *)&endianness)[i] = i;	1582	((unsigned char *)&endianness)[i] = i;
1444		1583
1445	*bit++ = 0;	1584	*bit++ = 0;
1446		1585
1447	for (i = 0; i < sizeof (ino_t); ++i)	1586	for (i = 0; i < sizeof (eio_ino_t); ++i)
1448	{	1587	{
1449	/* shifting off the byte offsets out of "endianness" */	1588	/* shifting off the byte offsets out of "endianness" */
1450	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;	1589	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1451	endianness >>= 8;	1590	endianness >>= 8;
1452		1591
1453	for (j = 0; j < 8; ++j)	1592	for (j = 0; j < 8; ++j)
1454	if (inode_bits & (((ino_t)1) << (i * 8 + j)))	1593	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1455	*bit++ = offs + j;	1594	*bit++ = offs + j;
1456	}	1595	}
1457		1596
1458	for (j = 0; j < 8; ++j)	1597	for (j = 0; j < 8; ++j)
1459	if (score_bits & (1 << j))	1598	if (score_bits & (1 << j))
1460	*bit++ = offsetof (eio_dirent, score) * 8 + j;	1599	*bit++ = offsetof (eio_dirent, score) * 8 + j;
1461	}	1600	}
1462		1601
1463	/* now actually do the sorting (a variant of MSD radix sort) */	1602	/* now actually do the sorting (a variant of MSD radix sort) */
1464	{	1603	{
1465	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;	1604	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1466	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;	1605	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1467	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];	1606	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1468	int stk_idx = 0;	1607	int stk_idx = 0;
1469		1608
1470	base_stk [stk_idx] = dents;	1609	base_stk [stk_idx] = dents;
1471	end_stk [stk_idx] = dents + size;	1610	end_stk [stk_idx] = dents + size;
1472	bit_stk [stk_idx] = bit - 1;	1611	bit_stk [stk_idx] = bit - 1;
…		…
1551	}	1690	}
1552	}	1691	}
1553	}	1692	}
1554		1693
1555	static void	1694	static void
1556	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)	1695	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1557	{	1696	{
1558	if (size <= 1)	1697	if (size <= 1)
1559	return; /* our insertion sort relies on size > 0 */	1698	return; /* our insertion sort relies on size > 0 */
1560		1699
1561	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */	1700	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
…		…
1569		1708
1570	/* read a full directory */	1709	/* read a full directory */
1571	static void	1710	static void
1572	eio__scandir (eio_req *req, etp_worker *self)	1711	eio__scandir (eio_req *req, etp_worker *self)
1573	{	1712	{
1574	DIR *dirp;
1575	EIO_STRUCT_DIRENT *entp;
1576	char *name, *names;	1713	char *name, *names;
1577	int namesalloc = 4096;	1714	int namesalloc = 4096 - sizeof (void *) * 4;
1578	int namesoffs = 0;	1715	int namesoffs = 0;
1579	int flags = req->int1;	1716	int flags = req->int1;
1580	eio_dirent *dents = 0;	1717	eio_dirent *dents = 0;
1581	int dentalloc = 128;	1718	int dentalloc = 128;
1582	int dentoffs = 0;	1719	int dentoffs = 0;
1583	ino_t inode_bits = 0;	1720	eio_ino_t inode_bits = 0;
		1721	#ifdef _WIN32
		1722	HANDLE dirp;
		1723	WIN32_FIND_DATA entp;
		1724	#else
		1725	DIR *dirp;
		1726	EIO_STRUCT_DIRENT *entp;
		1727	#endif
1584		1728
1585	req->result = -1;	1729	req->result = -1;
1586		1730
1587	if (!(flags & EIO_READDIR_DENTS))	1731	if (!(flags & EIO_READDIR_DENTS))
1588	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);	1732	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1589		1733
1590	X_LOCK (wrklock);	1734	#ifdef _WIN32
1591	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1735	{
		1736	int len = strlen ((const char *)req->ptr1);
		1737	char *path = malloc (MAX_PATH);
		1738	const char *fmt;
		1739	const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		1740
		1741	if (!len)
		1742	fmt = "./*";
		1743	else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
		1744	fmt = "%s*";
		1745	else
		1746	fmt = "%s/*";
		1747
		1748	_snprintf (path, MAX_PATH, fmt, reqpath);
		1749	dirp = FindFirstFile (path, &entp);
		1750	free (path);
		1751
		1752	if (dirp == INVALID_HANDLE_VALUE)
		1753	{
		1754	/* should steal _dosmaperr */
		1755	switch (GetLastError ())
		1756	{
		1757	case ERROR_FILE_NOT_FOUND:
		1758	req->result = 0;
		1759	break;
		1760
		1761	case ERROR_INVALID_NAME:
		1762	case ERROR_PATH_NOT_FOUND:
		1763	case ERROR_NO_MORE_FILES:
		1764	errno = ENOENT;
		1765	break;
		1766
		1767	case ERROR_NOT_ENOUGH_MEMORY:
		1768	errno = ENOMEM;
		1769	break;
		1770
		1771	default:
		1772	errno = EINVAL;
		1773	break;
		1774	}
		1775
		1776	return;
		1777	}
		1778	}
		1779	#else
		1780	#if HAVE_AT
		1781	if (req->wd)
		1782	{
		1783	int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1784
		1785	if (fd < 0)
		1786	return;
		1787
		1788	dirp = fdopendir (fd);
		1789
		1790	if (!dirp)
		1791	close (fd);
		1792	}
		1793	else
1592	self->dirp = dirp = opendir (req->ptr1);	1794	dirp = opendir (req->ptr1);
		1795	#else
		1796	dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
		1797	#endif
		1798
		1799	if (!dirp)
		1800	return;
		1801	#endif
1593		1802
1594	if (req->flags & EIO_FLAG_PTR1_FREE)	1803	if (req->flags & EIO_FLAG_PTR1_FREE)
1595	free (req->ptr1);	1804	free (req->ptr1);
1596		1805
1597	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1806	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1598	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1807	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1599	req->ptr2 = names = malloc (namesalloc);	1808	req->ptr2 = names = malloc (namesalloc);
1600	X_UNLOCK (wrklock);
1601		1809
1602	if (dirp && names && (!flags || dents))	1810	if (!names || (flags && !dents))
		1811	return;
		1812
1603	for (;;)	1813	for (;;)
1604	{	1814	{
		1815	int done;
		1816
		1817	#ifdef _WIN32
		1818	done = !dirp;
		1819	#else
1605	errno = 0;	1820	errno = 0;
1606	entp = readdir (dirp);	1821	entp = readdir (dirp);
		1822	done = !entp;
		1823	#endif
1607		1824
1608	if (!entp)	1825	if (done)
1609	{	1826	{
		1827	#ifndef _WIN32
		1828	int old_errno = errno;
		1829	closedir (dirp);
		1830	errno = old_errno;
		1831
1610	if (errno)	1832	if (errno)
1611	break;	1833	break;
		1834	#endif
1612		1835
1613	/* sort etc. */	1836	/* sort etc. */
1614	req->int1 = flags;	1837	req->int1 = flags;
1615	req->result = dentoffs;	1838	req->result = dentoffs;
1616		1839
1617	if (flags & EIO_READDIR_STAT_ORDER)	1840	if (flags & EIO_READDIR_STAT_ORDER)
1618	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);	1841	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1619	else if (flags & EIO_READDIR_DIRS_FIRST)	1842	else if (flags & EIO_READDIR_DIRS_FIRST)
1620	if (flags & EIO_READDIR_FOUND_UNKNOWN)	1843	if (flags & EIO_READDIR_FOUND_UNKNOWN)
1621	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */	1844	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1622	else	1845	else
		1846	{
		1847	/* in this case, all is known, and we just put dirs first and sort them */
		1848	eio_dirent *oth = dents + dentoffs;
		1849	eio_dirent *dir = dents;
		1850
		1851	/* now partition dirs to the front, and non-dirs to the back */
		1852	/* by walking from both sides and swapping if necessary */
		1853	while (oth > dir)
		1854	{
		1855	if (dir->type == EIO_DT_DIR)
		1856	++dir;
		1857	else if ((--oth)->type == EIO_DT_DIR)
		1858	{
		1859	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1860
		1861	++dir;
		1862	}
		1863	}
		1864
		1865	/* now sort the dirs only (dirs all have the same score) */
		1866	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1867	}
		1868
		1869	break;
		1870	}
		1871
		1872	/* now add the entry to our list(s) */
		1873	name = D_NAME (entp);
		1874
		1875	/* skip . and .. entries */
		1876	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
		1877	{
		1878	int len = D_NAMLEN (entp) + 1;
		1879
		1880	while (ecb_expect_false (namesoffs + len > namesalloc))
		1881	{
		1882	namesalloc *= 2;
		1883	req->ptr2 = names = realloc (names, namesalloc);
		1884
		1885	if (!names)
		1886	break;
		1887	}
		1888
		1889	memcpy (names + namesoffs, name, len);
		1890
		1891	if (dents)
		1892	{
		1893	struct eio_dirent *ent;
		1894
		1895	if (ecb_expect_false (dentoffs == dentalloc))
1623	{	1896	{
1624	/* in this case, all is known, and we just put dirs first and sort them */	1897	dentalloc *= 2;
		1898	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1899
		1900	if (!dents)
		1901	break;
		1902	}
		1903
1625	eio_dirent *oth = dents + dentoffs;	1904	ent = dents + dentoffs;
1626	eio_dirent *dir = dents;
1627		1905
1628	/* now partition dirs to the front, and non-dirs to the back */	1906	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1629	/* by walking from both sides and swapping if necessary */	1907	ent->namelen = len - 1;
1630	while (oth > dir)	1908	ent->inode = D_INO (entp);
		1909
		1910	inode_bits |= ent->inode;
		1911
		1912	switch (D_TYPE (entp))
		1913	{
		1914	default:
		1915	ent->type = EIO_DT_UNKNOWN;
		1916	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1917	break;
		1918
		1919	#ifdef DT_FIFO
		1920	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1921	#endif
		1922	#ifdef DT_CHR
		1923	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1924	#endif
		1925	#ifdef DT_MPC
		1926	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1927	#endif
		1928	#ifdef DT_DIR
		1929	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1930	#endif
		1931	#ifdef DT_NAM
		1932	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1933	#endif
		1934	#ifdef DT_BLK
		1935	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1936	#endif
		1937	#ifdef DT_MPB
		1938	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1939	#endif
		1940	#ifdef DT_REG
		1941	case DT_REG: ent->type = EIO_DT_REG; break;
		1942	#endif
		1943	#ifdef DT_NWK
		1944	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1945	#endif
		1946	#ifdef DT_CMP
		1947	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1948	#endif
		1949	#ifdef DT_LNK
		1950	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1951	#endif
		1952	#ifdef DT_SOCK
		1953	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1954	#endif
		1955	#ifdef DT_DOOR
		1956	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1957	#endif
		1958	#ifdef DT_WHT
		1959	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1960	#endif
		1961	}
		1962
		1963	ent->score = 7;
		1964
		1965	if (flags & EIO_READDIR_DIRS_FIRST)
		1966	{
		1967	if (ent->type == EIO_DT_UNKNOWN)
1631	{	1968	{
1632	if (dir->type == EIO_DT_DIR)	1969	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1633	++dir;	1970	ent->score = 1;
1634	else if ((--oth)->type == EIO_DT_DIR)	1971	else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1635	{	1972	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1636	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1637
1638	++dir;
1639	}
1640	}	1973	}
1641		1974	else if (ent->type == EIO_DT_DIR)
1642	/* now sort the dirs only (dirs all have the same score) */	1975	ent->score = 0;
1643	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1644	}	1976	}
1645
1646	break;
1647	}
1648
1649	/* now add the entry to our list(s) */
1650	name = entp->d_name;
1651
1652	/* skip . and .. entries */
1653	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1654	{
1655	int len = D_NAMLEN (entp) + 1;
1656
1657	while (ecb_expect_false (namesoffs + len > namesalloc))
1658	{
1659	namesalloc *= 2;
1660	X_LOCK (wrklock);
1661	req->ptr2 = names = realloc (names, namesalloc);
1662	X_UNLOCK (wrklock);
1663
1664	if (!names)
1665	break;
1666	}	1977	}
1667		1978
1668	memcpy (names + namesoffs, name, len);
1669
1670	if (dents)
1671	{
1672	struct eio_dirent *ent;
1673
1674	if (ecb_expect_false (dentoffs == dentalloc))
1675	{
1676	dentalloc *= 2;
1677	X_LOCK (wrklock);
1678	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1679	X_UNLOCK (wrklock);
1680
1681	if (!dents)
1682	break;
1683	}
1684
1685	ent = dents + dentoffs;
1686
1687	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1688	ent->namelen = len - 1;
1689	ent->inode = D_INO (entp);
1690
1691	inode_bits |= ent->inode;
1692
1693	switch (D_TYPE (entp))
1694	{
1695	default:
1696	ent->type = EIO_DT_UNKNOWN;
1697	flags |= EIO_READDIR_FOUND_UNKNOWN;
1698	break;
1699
1700	#ifdef DT_FIFO
1701	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1702	#endif
1703	#ifdef DT_CHR
1704	case DT_CHR: ent->type = EIO_DT_CHR; break;
1705	#endif
1706	#ifdef DT_MPC
1707	case DT_MPC: ent->type = EIO_DT_MPC; break;
1708	#endif
1709	#ifdef DT_DIR
1710	case DT_DIR: ent->type = EIO_DT_DIR; break;
1711	#endif
1712	#ifdef DT_NAM
1713	case DT_NAM: ent->type = EIO_DT_NAM; break;
1714	#endif
1715	#ifdef DT_BLK
1716	case DT_BLK: ent->type = EIO_DT_BLK; break;
1717	#endif
1718	#ifdef DT_MPB
1719	case DT_MPB: ent->type = EIO_DT_MPB; break;
1720	#endif
1721	#ifdef DT_REG
1722	case DT_REG: ent->type = EIO_DT_REG; break;
1723	#endif
1724	#ifdef DT_NWK
1725	case DT_NWK: ent->type = EIO_DT_NWK; break;
1726	#endif
1727	#ifdef DT_CMP
1728	case DT_CMP: ent->type = EIO_DT_CMP; break;
1729	#endif
1730	#ifdef DT_LNK
1731	case DT_LNK: ent->type = EIO_DT_LNK; break;
1732	#endif
1733	#ifdef DT_SOCK
1734	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1735	#endif
1736	#ifdef DT_DOOR
1737	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1738	#endif
1739	#ifdef DT_WHT
1740	case DT_WHT: ent->type = EIO_DT_WHT; break;
1741	#endif
1742	}
1743
1744	ent->score = 7;
1745
1746	if (flags & EIO_READDIR_DIRS_FIRST)
1747	{
1748	if (ent->type == EIO_DT_UNKNOWN)
1749	{
1750	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1751	ent->score = 1;
1752	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1753	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1754	}
1755	else if (ent->type == EIO_DT_DIR)
1756	ent->score = 0;
1757	}
1758	}
1759
1760	namesoffs += len;	1979	namesoffs += len;
1761	++dentoffs;	1980	++dentoffs;
1762	}	1981	}
1763		1982
1764	if (EIO_CANCELLED (req))	1983	if (EIO_CANCELLED (req))
1765	{	1984	{
1766	errno = ECANCELED;	1985	errno = ECANCELED;
1767	break;	1986	break;
1768	}	1987	}
		1988
		1989	#ifdef _WIN32
		1990	if (!FindNextFile (dirp, &entp))
		1991	{
		1992	FindClose (dirp);
		1993	dirp = 0;
1769	}	1994	}
		1995	#endif
		1996	}
1770	}	1997	}
		1998
		1999	/*****************************************************************************/
		2000	/* working directory stuff */
		2001	/* various deficiencies in the posix 2008 api force us to */
		2002	/* keep the absolute path in string form at all times */
		2003	/* fuck yeah. */
		2004
		2005	#if !HAVE_AT
		2006
		2007	/* a bit like realpath, but usually faster because it doesn'T have to return */
		2008	/* an absolute or canonical path */
		2009	static const char *
		2010	wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2011	{
		2012	if (!wd || *path == '/')
		2013	return path;
		2014
		2015	if (path [0] == '.' && !path [1])
		2016	return wd->str;
		2017
		2018	{
		2019	int l1 = wd->len;
		2020	int l2 = strlen (path);
		2021
		2022	char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
		2023
		2024	memcpy (res, wd->str, l1);
		2025	res [l1] = '/';
		2026	memcpy (res + l1 + 1, path, l2 + 1);
		2027
		2028	return res;
		2029	}
		2030	}
		2031
		2032	#endif
		2033
		2034	static eio_wd
		2035	eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2036	{
		2037	int fd;
		2038	eio_wd res;
		2039	int len = eio__realpath (tmpbuf, wd, path);
		2040
		2041	if (len < 0)
		2042	return EIO_INVALID_WD;
		2043
		2044	#if HAVE_AT
		2045	fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		2046
		2047	if (fd < 0)
		2048	return EIO_INVALID_WD;
		2049	#endif
		2050
		2051	res = malloc (sizeof (*res) + len); /* one extra 0-byte */
		2052
		2053	#if HAVE_AT
		2054	res->fd = fd;
		2055	#endif
		2056
		2057	res->len = len;
		2058	memcpy (res->str, tmpbuf->ptr, len);
		2059	res->str [len] = 0;
		2060
		2061	return res;
		2062	}
		2063
		2064	eio_wd
		2065	eio_wd_open_sync (eio_wd wd, const char *path)
		2066	{
		2067	struct tmpbuf tmpbuf = { 0 };
		2068	wd = eio__wd_open_sync (&tmpbuf, wd, path);
		2069	free (tmpbuf.ptr);
		2070
		2071	return wd;
		2072	}
		2073
		2074	void
		2075	eio_wd_close_sync (eio_wd wd)
		2076	{
		2077	if (wd != EIO_INVALID_WD && wd != EIO_CWD)
		2078	{
		2079	#if HAVE_AT
		2080	close (wd->fd);
		2081	#endif
		2082	free (wd);
		2083	}
		2084	}
		2085
		2086	#if HAVE_AT
		2087
		2088	/* they forgot these */
		2089
		2090	static int
		2091	eio__truncateat (int dirfd, const char *path, off_t length)
		2092	{
		2093	int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
		2094	int res;
		2095
		2096	if (fd < 0)
		2097	return fd;
		2098
		2099	res = ftruncate (fd, length);
		2100	close (fd);
		2101	return res;
		2102	}
		2103
		2104	static int
		2105	eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
		2106	{
		2107	int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
		2108	int res;
		2109
		2110	if (fd < 0)
		2111	return fd;
		2112
		2113	res = fstatvfs (fd, buf);
		2114	close (fd);
		2115	return res;
		2116
		2117	}
		2118
		2119	#endif
1771		2120
1772	/*****************************************************************************/	2121	/*****************************************************************************/
1773		2122
1774	#define ALLOC(len) \	2123	#define ALLOC(len) \
1775	if (!req->ptr2) \	2124	if (!req->ptr2) \
…		…
1784	req->result = -1; \	2133	req->result = -1; \
1785	break; \	2134	break; \
1786	} \	2135	} \
1787	}	2136	}
1788		2137
		2138	static void ecb_noinline ecb_cold
		2139	etp_proc_init (void)
		2140	{
		2141	#if HAVE_PRCTL_SET_NAME
		2142	/* provide a more sensible "thread name" */
		2143	char name[16 + 1];
		2144	const int namelen = sizeof (name) - 1;
		2145	int len;
		2146
		2147	prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
		2148	name [namelen] = 0;
		2149	len = strlen (name);
		2150	strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
		2151	prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
		2152	#endif
		2153	}
		2154
1789	X_THREAD_PROC (etp_proc)	2155	X_THREAD_PROC (etp_proc)
1790	{	2156	{
1791	ETP_REQ *req;	2157	ETP_REQ *req;
1792	struct timespec ts;	2158	struct timespec ts;
1793	etp_worker *self = (etp_worker *)thr_arg;	2159	etp_worker *self = (etp_worker *)thr_arg;
1794		2160
		2161	etp_proc_init ();
		2162
1795	/* try to distribute timeouts somewhat randomly */	2163	/* try to distribute timeouts somewhat evenly */
1796	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	2164	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1797		2165
1798	for (;;)	2166	for (;;)
1799	{	2167	{
		2168	ts.tv_sec = 0;
		2169
1800	X_LOCK (reqlock);	2170	X_LOCK (reqlock);
1801		2171
1802	for (;;)	2172	for (;;)
1803	{	2173	{
1804	self->req = req = reqq_shift (&req_queue);	2174	req = reqq_shift (&req_queue);
1805		2175
1806	if (req)	2176	if (req)
1807	break;	2177	break;
1808		2178
		2179	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		2180	{
		2181	X_UNLOCK (reqlock);
		2182	X_LOCK (wrklock);
		2183	--started;
		2184	X_UNLOCK (wrklock);
		2185	goto quit;
		2186	}
		2187
1809	++idle;	2188	++idle;
1810		2189
1811	ts.tv_sec = time (0) + idle_timeout;	2190	if (idle <= max_idle)
1812	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	2191	/* we are allowed to idle, so do so without any timeout */
		2192	X_COND_WAIT (reqwait, reqlock);
		2193	else
1813	{	2194	{
1814	if (idle > max_idle)	2195	/* initialise timeout once */
1815	{	2196	if (!ts.tv_sec)
1816	--idle;	2197	ts.tv_sec = time (0) + idle_timeout;
1817	X_UNLOCK (reqlock);
1818	X_LOCK (wrklock);
1819	--started;
1820	X_UNLOCK (wrklock);
1821	goto quit;
1822	}
1823		2198
1824	/* we are allowed to idle, so do so without any timeout */
1825	X_COND_WAIT (reqwait, reqlock);	2199	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		2200	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1826	}	2201	}
1827		2202
1828	--idle;	2203	--idle;
1829	}	2204	}
1830		2205
…		…
1842	++npending;	2217	++npending;
1843		2218
1844	if (!reqq_push (&res_queue, req) && want_poll_cb)	2219	if (!reqq_push (&res_queue, req) && want_poll_cb)
1845	want_poll_cb ();	2220	want_poll_cb ();
1846		2221
1847	self->req = 0;
1848	etp_worker_clear (self);	2222	etp_worker_clear (self);
1849		2223
1850	X_UNLOCK (reslock);	2224	X_UNLOCK (reslock);
1851	}	2225	}
1852		2226
1853	quit:	2227	quit:
		2228	free (req);
		2229
1854	X_LOCK (wrklock);	2230	X_LOCK (wrklock);
1855	etp_worker_free (self);	2231	etp_worker_free (self);
1856	X_UNLOCK (wrklock);	2232	X_UNLOCK (wrklock);
1857		2233
1858	return 0;	2234	return 0;
…		…
1861	/*****************************************************************************/	2237	/*****************************************************************************/
1862		2238
1863	int ecb_cold	2239	int ecb_cold
1864	eio_init (void (*want_poll)(void), void (*done_poll)(void))	2240	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1865	{	2241	{
		2242	#if !HAVE_PREADWRITE
		2243	X_MUTEX_CREATE (preadwritelock);
		2244	#endif
		2245
1866	return etp_init (want_poll, done_poll);	2246	return etp_init (want_poll, done_poll);
1867	}	2247	}
1868		2248
1869	ecb_inline void	2249	ecb_inline void
1870	eio_api_destroy (eio_req *req)	2250	eio_api_destroy (eio_req *req)
…		…
1897	}	2277	}
1898		2278
1899	static void	2279	static void
1900	eio_execute (etp_worker *self, eio_req *req)	2280	eio_execute (etp_worker *self, eio_req *req)
1901	{	2281	{
		2282	#if HAVE_AT
		2283	int dirfd;
		2284	#else
		2285	const char *path;
		2286	#endif
		2287
1902	if (ecb_expect_false (EIO_CANCELLED (req)))	2288	if (ecb_expect_false (EIO_CANCELLED (req)))
1903	{	2289	{
1904	req->result = -1;	2290	req->result = -1;
1905	req->errorno = ECANCELED;	2291	req->errorno = ECANCELED;
1906	return;	2292	return;
1907	}	2293	}
1908		2294
		2295	if (ecb_expect_false (req->wd == EIO_INVALID_WD))
		2296	{
		2297	req->result = -1;
		2298	req->errorno = ENOENT;
		2299	return;
		2300	}
		2301
		2302	if (req->type >= EIO_OPEN)
		2303	{
		2304	#if HAVE_AT
		2305	dirfd = WD2FD (req->wd);
		2306	#else
		2307	path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		2308	#endif
		2309	}
		2310
1909	switch (req->type)	2311	switch (req->type)
1910	{	2312	{
		2313	case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
		2314	req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
		2315	break;
		2316	case EIO_WD_CLOSE: req->result = 0;
		2317	eio_wd_close_sync (req->wd); break;
		2318
1911	case EIO_READ: ALLOC (req->size);	2319	case EIO_READ: ALLOC (req->size);
1912	req->result = req->offs >= 0	2320	req->result = req->offs >= 0
1913	? pread (req->int1, req->ptr2, req->size, req->offs)	2321	? pread (req->int1, req->ptr2, req->size, req->offs)
1914	: read (req->int1, req->ptr2, req->size); break;	2322	: read (req->int1, req->ptr2, req->size); break;
1915	case EIO_WRITE: req->result = req->offs >= 0	2323	case EIO_WRITE: req->result = req->offs >= 0
1916	? pwrite (req->int1, req->ptr2, req->size, req->offs)	2324	? pwrite (req->int1, req->ptr2, req->size, req->offs)
1917	: write (req->int1, req->ptr2, req->size); break;	2325	: write (req->int1, req->ptr2, req->size); break;
1918		2326
1919	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	2327	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1920	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;	2328	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
		2329
		2330	#if HAVE_AT
1921		2331
1922	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2332	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1923	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2333	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1924	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2334	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2335	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
		2336	case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
		2337	case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
		2338	case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
		2339	case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
		2340
		2341	case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
		2342	case EIO_RMDIR: req->result = 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: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
		2345	case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
		2346	case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
		2347	case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2348	case EIO_READLINK: ALLOC (PATH_MAX);
		2349	req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
		2350	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2351	req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2352	case EIO_UTIME:
		2353	case EIO_FUTIME:
		2354	{
		2355	struct timespec ts[2];
		2356	struct timespec *times;
		2357
		2358	if (req->nv1 != -1. || req->nv2 != -1.)
		2359	{
		2360	ts[0].tv_sec = req->nv1;
		2361	ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
		2362	ts[1].tv_sec = req->nv2;
		2363	ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
		2364
		2365	times = ts;
		2366	}
		2367	else
		2368	times = 0;
		2369
		2370	req->result = req->type == EIO_FUTIME
		2371	? futimens (req->int1, times)
		2372	: utimensat (dirfd, req->ptr1, times, 0);
		2373	}
		2374	break;
		2375
		2376	#else
		2377
		2378	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2379	req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2380	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1925	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	2381	req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2382	case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
		2383	case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
		2384	case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
		2385	case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
		2386
		2387	case EIO_UNLINK: req->result = unlink (path ); break;
		2388	case EIO_RMDIR: req->result = rmdir (path ); break;
		2389	case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
		2390	case EIO_RENAME: req->result = rename (path , req->ptr2); break;
		2391	case EIO_LINK: req->result = link (path , req->ptr2); break;
		2392	case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
		2393	case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
		2394	case EIO_READLINK: ALLOC (PATH_MAX);
		2395	req->result = readlink (path, req->ptr2, PATH_MAX); break;
		2396	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2397	req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2398
		2399	case EIO_UTIME:
		2400	case EIO_FUTIME:
		2401	{
		2402	struct timeval tv[2];
		2403	struct timeval *times;
		2404
		2405	if (req->nv1 != -1. || req->nv2 != -1.)
		2406	{
		2407	tv[0].tv_sec = req->nv1;
		2408	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
		2409	tv[1].tv_sec = req->nv2;
		2410	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
		2411
		2412	times = tv;
		2413	}
		2414	else
		2415	times = 0;
		2416
		2417	req->result = req->type == EIO_FUTIME
		2418	? futimes (req->int1, times)
		2419	: utimes (req->ptr1, times);
		2420	}
		2421	break;
		2422
		2423	#endif
		2424
		2425	case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
		2426	{
		2427	ALLOC (req->result);
		2428	memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
		2429	}
		2430	break;
		2431
1926	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	2432	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1927	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	2433	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1928		2434
1929	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1930	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1931	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));	2435	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1932	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;	2436	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1933		2437
1934	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1935	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	2438	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1936	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1937	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	2439	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1938	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1939	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;	2440	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1940		2441
1941	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1942	case EIO_CLOSE: req->result = close (req->int1); break;	2442	case EIO_CLOSE: req->result = close (req->int1); break;
1943	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;	2443	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1944	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1945	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1946	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1947	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1948	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1949	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1950	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1951
1952	case EIO_REALPATH: eio__realpath (req, self); break;
1953
1954	case EIO_READLINK: ALLOC (PATH_MAX);
1955	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1956
1957	case EIO_SYNC: req->result = 0; sync (); break;	2444	case EIO_SYNC: req->result = 0; sync (); break;
1958	case EIO_FSYNC: req->result = fsync (req->int1); break;	2445	case EIO_FSYNC: req->result = fsync (req->int1); break;
1959	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2446	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2447	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2448	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1960	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;	2449	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1961	case EIO_MTOUCH: req->result = eio__mtouch (req); break;	2450	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1962	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;	2451	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1963	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;	2452	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1964	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	2453	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1965		2454
1966	case EIO_READDIR: eio__scandir (req, self); break;	2455	case EIO_READDIR: eio__scandir (req, self); break;
1967		2456
1968	case EIO_BUSY:	2457	case EIO_BUSY:
1969	#ifdef _WIN32	2458	#ifdef _WIN32
…		…
1978	req->result = select (0, 0, 0, 0, &tv);	2467	req->result = select (0, 0, 0, 0, &tv);
1979	}	2468	}
1980	#endif	2469	#endif
1981	break;	2470	break;
1982		2471
1983	case EIO_UTIME:
1984	case EIO_FUTIME:
1985	{
1986	struct timeval tv[2];
1987	struct timeval *times;
1988
1989	if (req->nv1 != -1. || req->nv2 != -1.)
1990	{
1991	tv[0].tv_sec = req->nv1;
1992	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1993	tv[1].tv_sec = req->nv2;
1994	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1995
1996	times = tv;
1997	}
1998	else
1999	times = 0;
2000
2001	req->result = req->type == EIO_FUTIME
2002	? futimes (req->int1, times)
2003	: utimes (req->ptr1, times);
2004	}
2005	break;
2006
2007	case EIO_GROUP:	2472	case EIO_GROUP:
2008	abort (); /* handled in eio_request */	2473	abort (); /* handled in eio_request */
2009		2474
2010	case EIO_NOP:	2475	case EIO_NOP:
2011	req->result = 0;	2476	req->result = 0;
…		…
2024	req->errorno = errno;	2489	req->errorno = errno;
2025	}	2490	}
2026		2491
2027	#ifndef EIO_NO_WRAPPERS	2492	#ifndef EIO_NO_WRAPPERS
2028		2493
		2494	eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
		2495	{
		2496	REQ (EIO_WD_OPEN); PATH; SEND;
		2497	}
		2498
		2499	eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
		2500	{
		2501	REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
		2502	}
		2503
2029	eio_req *eio_nop (int pri, eio_cb cb, void *data)	2504	eio_req *eio_nop (int pri, eio_cb cb, void *data)
2030	{	2505	{
2031	REQ (EIO_NOP); SEND;	2506	REQ (EIO_NOP); SEND;
2032	}	2507	}
2033		2508
…		…
2049	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2524	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2050	{	2525	{
2051	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2526	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2052	}	2527	}
2053		2528
		2529	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
		2530	{
		2531	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
		2532	}
		2533
		2534	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2535	{
		2536	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2537	}
		2538
		2539	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2540	{
		2541	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2542	}
		2543
2054	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)	2544	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2055	{	2545	{
2056	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	2546	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2057	}	2547	}
2058		2548
…		…
2064	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)	2554	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2065	{	2555	{
2066	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;	2556	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2067	}	2557	}
2068		2558
2069	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)	2559	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2070	{	2560	{
2071	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;	2561	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2072	}
2073
2074	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2075	{
2076	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2077	}	2562	}
2078		2563
2079	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)	2564	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2080	{	2565	{
2081	REQ (EIO_CLOSE); req->int1 = fd; SEND;	2566	REQ (EIO_CLOSE); req->int1 = fd; SEND;
…		…
2119	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)	2604	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2120	{	2605	{
2121	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2606	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2122	}	2607	}
2123		2608
2124	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2609	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2125	{	2610	{
2126	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2611	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2127	}	2612	}
2128		2613
2129	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)	2614	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
…		…
2149	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)	2634	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2150	{	2635	{
2151	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;	2636	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2152	}	2637	}
2153		2638
2154	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)	2639	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2155	{	2640	{
2156	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2641	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2157	}	2642	}
2158		2643
2159	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)	2644	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
…		…
2306	}	2791	}
2307		2792
2308	/*****************************************************************************/	2793	/*****************************************************************************/
2309	/* misc garbage */	2794	/* misc garbage */
2310		2795
2311	ssize_t	2796	eio_ssize_t
2312	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2797	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2313	{	2798	{
2314	etp_worker wrk;
2315	ssize_t ret;
2316
2317	wrk.dbuf = 0;
2318
2319	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);	2799	return eio__sendfile (ofd, ifd, offset, count);
2320
2321	if (wrk.dbuf)
2322	free (wrk.dbuf);
2323
2324	return ret;
2325	}	2800	}
2326		2801

Diff Legend

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