ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.98 by root, Sun Jul 24 05:53:34 2011 UTC vs.
Revision 1.144 by root, Tue Dec 27 09:58:44 2016 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) 2007,2008,2009,2010,2011,2012,2013,2016 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,
…		…
42	#endif	42	#endif
43		43
44	#include "eio.h"	44	#include "eio.h"
45	#include "ecb.h"	45	#include "ecb.h"
46		46
47	#ifdef EIO_STACKSIZE
48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
49	#endif
50	#include "xthread.h"
51
52	#include <errno.h>	47	#include <errno.h>
53	#include <stddef.h>	48	#include <stddef.h>
54	#include <stdlib.h>	49	#include <stdlib.h>
55	#include <string.h>	50	#include <string.h>
56	#include <errno.h>	51	#include <errno.h>
…		…
105	#ifdef _WIN32	100	#ifdef _WIN32
106		101
107	#undef PAGESIZE	102	#undef PAGESIZE
108	#define PAGESIZE 4096 /* GetSystemInfo? */	103	#define PAGESIZE 4096 /* GetSystemInfo? */
109		104
		105	/* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
		106
110	#ifdef EIO_STRUCT_STATI64	107	#ifdef EIO_STRUCT_STATI64
		108	/* look at perl's non-sloppy stat */
111	#define stat(path,buf) _stati64 (path,buf)	109	#define stat(path,buf) _stati64 (path,buf)
112	#define fstat(fd,buf) _fstati64 (fd,buf)	110	#define fstat(fd,buf) _fstati64 (fd,buf)
113	#endif	111	#endif
114	#define lstat(path,buf) stat (path,buf)	112	#define lstat(path,buf) stat (path,buf)
115	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))	113	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
…		…
117	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))	115	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
118		116
119	#define chmod(path,mode) _chmod (path, mode)	117	#define chmod(path,mode) _chmod (path, mode)
120	#define dup(fd) _dup (fd)	118	#define dup(fd) _dup (fd)
121	#define dup2(fd1,fd2) _dup2 (fd1, fd2)	119	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		120	#define pipe(fds) _pipe (fds, 4096, O_BINARY)
122		121
		122	#define fcntl(fd,cmd,arg) EIO_ENOSYS ()
		123	#define ioctl(fd,cmd,arg) EIO_ENOSYS ()
123	#define fchmod(fd,mode) EIO_ENOSYS ()	124	#define fchmod(fd,mode) EIO_ENOSYS ()
124	#define chown(path,uid,gid) EIO_ENOSYS ()	125	#define chown(path,uid,gid) EIO_ENOSYS ()
125	#define fchown(fd,uid,gid) EIO_ENOSYS ()	126	#define fchown(fd,uid,gid) EIO_ENOSYS ()
126	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */	127	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
127	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */	128	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
…		…
129	#define sync() EIO_ENOSYS ()	130	#define sync() EIO_ENOSYS ()
130	#define readlink(path,buf,s) EIO_ENOSYS ()	131	#define readlink(path,buf,s) EIO_ENOSYS ()
131	#define statvfs(path,buf) EIO_ENOSYS ()	132	#define statvfs(path,buf) EIO_ENOSYS ()
132	#define fstatvfs(fd,buf) EIO_ENOSYS ()	133	#define fstatvfs(fd,buf) EIO_ENOSYS ()
133		134
		135	#define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
		136	#define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
		137
		138	#if __GNUC__
		139	typedef long long eio_off_t; /* signed for compatibility to msvc */
		140	#else
		141	typedef __int64 eio_off_t; /* unsigned not supported by msvc */
		142	#endif
		143
		144	static eio_ssize_t
		145	eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
		146	{
		147	OVERLAPPED o = { 0 };
		148	DWORD got;
		149
		150	o.Offset = offset;
		151	o.OffsetHigh = offset >> 32;
		152
		153	return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
		154	? got : -1;
		155	}
		156
		157	static eio_ssize_t
		158	eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
		159	{
		160	OVERLAPPED o = { 0 };
		161	DWORD got;
		162
		163	o.Offset = offset;
		164	o.OffsetHigh = offset >> 32;
		165
		166	return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
		167	? got : -1;
		168	}
		169
		170	/* rename() uses MoveFile, which fails to overwrite */
		171	#define rename(old,neu) eio__rename (old, neu)
		172
		173	static int
		174	eio__rename (const char *old, const char *neu)
		175	{
		176	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		177	return 0;
		178
		179	/* should steal _dosmaperr */
		180	switch (GetLastError ())
		181	{
		182	case ERROR_FILE_NOT_FOUND:
		183	case ERROR_PATH_NOT_FOUND:
		184	case ERROR_INVALID_DRIVE:
		185	case ERROR_NO_MORE_FILES:
		186	case ERROR_BAD_NETPATH:
		187	case ERROR_BAD_NET_NAME:
		188	case ERROR_BAD_PATHNAME:
		189	case ERROR_FILENAME_EXCED_RANGE:
		190	errno = ENOENT;
		191	break;
		192
		193	default:
		194	errno = EACCES;
		195	break;
		196	}
		197
		198	return -1;
		199	}
		200
134	/* we could even stat and see if it exists */	201	/* we could even stat and see if it exists */
135	static int	202	static int
136	symlink (const char *old, const char *neu)	203	symlink (const char *old, const char *neu)
137	{	204	{
138	#if WINVER >= 0x0600	205	#if WINVER >= 0x0600
		206	int flags;
		207
		208	/* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
		209	* and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
		210	* with directory first.
		211	*/
		212	for (flags = 3; flags >= 0; --flags)
139	if (CreateSymbolicLink (neu, old, 1))	213	if (CreateSymbolicLink (neu, old, flags))
140	return 0;	214	return 0;
141
142	if (CreateSymbolicLink (neu, old, 0))
143	return 0;
144	#endif	215	#endif
145		216
146	return EIO_ERRNO (ENOENT, -1);	217	return EIO_ERRNO (ENOENT, -1);
147	}	218	}
148		219
149	/* POSIX API only */	220	/* POSIX API only, causing trouble for win32 apps */
150	#define CreateHardLink(neu,old,flags) 0	221	#define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
151	#define CreateSymbolicLink(neu,old,flags) 0	222	#define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
152		223
153	struct statvfs	224	struct statvfs
154	{	225	{
155	int dummy;	226	int dummy;
156	};	227	};
…		…
162		233
163	#else	234	#else
164		235
165	#include <sys/time.h>	236	#include <sys/time.h>
166	#include <sys/select.h>	237	#include <sys/select.h>
167	#include <sys/statvfs.h>
168	#include <unistd.h>	238	#include <unistd.h>
169	#include <signal.h>	239	#include <signal.h>
170	#include <dirent.h>	240	#include <dirent.h>
171		241
		242	#ifdef ANDROID
		243	#include <sys/vfs.h>
		244	#define statvfs statfs
		245	#define fstatvfs fstatfs
		246	#include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
		247	#else
		248	#include <sys/statvfs.h>
		249	#endif
		250
172	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES	251	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
173	#include <sys/mman.h>	252	#include <sys/mman.h>
174	#endif	253	#endif
175		254
176	#define D_NAME(entp) entp->d_name	255	#define D_NAME(entp) entp->d_name
177		256
178	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */	257	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
179	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__	258	#if __FreeBSD__ || __NetBSD__ || __OpenBSD__
180	#define _DIRENT_HAVE_D_TYPE /* sigh */	259	#define _DIRENT_HAVE_D_TYPE /* sigh */
181	#define D_INO(de) (de)->d_fileno	260	#define D_INO(de) (de)->d_fileno
182	#define D_NAMLEN(de) (de)->d_namlen	261	#define D_NAMLEN(de) (de)->d_namlen
183	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600	262	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
184	#define D_INO(de) (de)->d_ino	263	#define D_INO(de) (de)->d_ino
…		…
203	# include <utime.h>	282	# include <utime.h>
204	#endif	283	#endif
205		284
206	#if HAVE_SYS_SYSCALL_H	285	#if HAVE_SYS_SYSCALL_H
207	# include <sys/syscall.h>	286	# include <sys/syscall.h>
208	#endif
209
210	#if HAVE_SYS_PRCTL_H
211	# include <sys/prctl.h>
212	#endif	287	#endif
213		288
214	#if HAVE_SENDFILE	289	#if HAVE_SENDFILE
215	# if __linux	290	# if __linux
216	# include <sys/sendfile.h>	291	# include <sys/sendfile.h>
…		…
241	# define NAME_MAX 4096	316	# define NAME_MAX 4096
242	#endif	317	#endif
243		318
244	/* used for readlink etc. */	319	/* used for readlink etc. */
245	#ifndef PATH_MAX	320	#ifndef PATH_MAX
246	# define PATH_MAX 4096	321	# define PATH_MAX 0
247	#endif	322	#endif
		323
		324	#ifndef EIO_PATH_MIN
		325	# define EIO_PATH_MIN 8160
		326	#endif
		327
		328	#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
248		329
249	/* buffer size for various temporary buffers */	330	/* buffer size for various temporary buffers */
250	#define EIO_BUFSIZE 65536	331	#define EIO_BUFSIZE 65536
251		332
252	#define dBUF \	333	#define dBUF \
253	char *eio_buf = malloc (EIO_BUFSIZE); \	334	char *eio_buf = malloc (EIO_BUFSIZE); \
254	errno = ENOMEM; \	335	errno = ENOMEM; \
255	if (!eio_buf) \	336	if (!eio_buf) \
256	return -1	337	return -1
257		338
258	#define FUBd \	339	#define FUBd \
259	free (eio_buf)	340	free (eio_buf)
260		341
261	#define EIO_TICKS ((1000000 + 1023) >> 10)	342	/*****************************************************************************/
		343
		344	struct etp_tmpbuf;
		345
		346	#if _POSIX_VERSION >= 200809L
		347	#define HAVE_AT 1
		348	#define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
		349	#ifndef O_SEARCH
		350	#define O_SEARCH O_RDONLY
		351	#endif
		352	#else
		353	#define HAVE_AT 0
		354	static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
		355	#endif
		356
		357	struct eio_pwd
		358	{
		359	#if HAVE_AT
		360	int fd;
		361	#endif
		362	int len;
		363	char str[1]; /* actually, a 0-terminated canonical path */
		364	};
		365
		366	/*****************************************************************************/
262		367
263	#define ETP_PRI_MIN EIO_PRI_MIN	368	#define ETP_PRI_MIN EIO_PRI_MIN
264	#define ETP_PRI_MAX EIO_PRI_MAX	369	#define ETP_PRI_MAX EIO_PRI_MAX
265		370
		371	#define ETP_TYPE_QUIT -1
		372	#define ETP_TYPE_GROUP EIO_GROUP
		373
		374	static void eio_nop_callback (void) { }
		375	static void (*eio_want_poll_cb)(void) = eio_nop_callback;
		376	static void (*eio_done_poll_cb)(void) = eio_nop_callback;
		377
		378	#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
		379	#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
		380
266	struct etp_worker;	381	struct etp_worker;
267
268	#define ETP_REQ eio_req	382	#define ETP_REQ eio_req
269	#define ETP_DESTROY(req) eio_destroy (req)	383	#define ETP_DESTROY(req) eio_destroy (req)
270	static int eio_finish (eio_req *req);	384	static int eio_finish (eio_req *req);
271	#define ETP_FINISH(req) eio_finish (req)	385	#define ETP_FINISH(req) eio_finish (req)
272	static void eio_execute (struct etp_worker *self, eio_req *req);	386	static void eio_execute (struct etp_worker *self, eio_req *req);
273	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)	387	#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
274		388
275	/*****************************************************************************/	389	#include "etp.c"
276		390
277	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	391	static struct etp_pool eio_pool;
278		392	#define EIO_POOL (&eio_pool)
279	/* calculate time difference in ~1/EIO_TICKS of a second */
280	ecb_inline int
281	tvdiff (struct timeval *tv1, struct timeval *tv2)
282	{
283	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
284	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
285	}
286
287	static unsigned int started, idle, wanted = 4;
288
289	static void (*want_poll_cb) (void);
290	static void (*done_poll_cb) (void);
291
292	static unsigned int max_poll_time; /* reslock */
293	static unsigned int max_poll_reqs; /* reslock */
294
295	static volatile unsigned int nreqs; /* reqlock */
296	static volatile unsigned int nready; /* reqlock */
297	static volatile unsigned int npending; /* reqlock */
298	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
299	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
300
301	static xmutex_t wrklock;
302	static xmutex_t reslock;
303	static xmutex_t reqlock;
304	static xcond_t reqwait;
305
306	#if !HAVE_PREADWRITE
307	/*
308	* make our pread/pwrite emulation safe against themselves, but not against
309	* normal read/write by using a mutex. slows down execution a lot,
310	* but that's your problem, not mine.
311	*/
312	static xmutex_t preadwritelock;
313	#endif
314
315	typedef struct etp_worker
316	{
317	/* locked by wrklock */
318	struct etp_worker *prev, *next;
319
320	xthread_t tid;
321
322	/* locked by reslock, reqlock or wrklock */
323	ETP_REQ *req; /* currently processed request */
324
325	#ifdef ETP_WORKER_COMMON
326	ETP_WORKER_COMMON
327	#endif
328	} etp_worker;
329
330	static etp_worker wrk_first; /* NOT etp */
331
332	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
333	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
334
335	/* worker threads management */
336
337	static void ecb_cold
338	etp_worker_clear (etp_worker *wrk)
339	{
340	}
341
342	static void ecb_cold
343	etp_worker_free (etp_worker *wrk)
344	{
345	wrk->next->prev = wrk->prev;
346	wrk->prev->next = wrk->next;
347
348	free (wrk);
349	}
350
351	static unsigned int
352	etp_nreqs (void)
353	{
354	int retval;
355	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
356	retval = nreqs;
357	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
358	return retval;
359	}
360
361	static unsigned int
362	etp_nready (void)
363	{
364	unsigned int retval;
365
366	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
367	retval = nready;
368	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
369
370	return retval;
371	}
372
373	static unsigned int
374	etp_npending (void)
375	{
376	unsigned int retval;
377
378	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
379	retval = npending;
380	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
381
382	return retval;
383	}
384
385	static unsigned int
386	etp_nthreads (void)
387	{
388	unsigned int retval;
389
390	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
391	retval = started;
392	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
393
394	return retval;
395	}
396
397	/*
398	* a somewhat faster data structure might be nice, but
399	* with 8 priorities this actually needs <20 insns
400	* per shift, the most expensive operation.
401	*/
402	typedef struct {
403	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
404	int size;
405	} etp_reqq;
406
407	static etp_reqq req_queue;
408	static etp_reqq res_queue;
409
410	static void ecb_noinline ecb_cold
411	reqq_init (etp_reqq *q)
412	{
413	int pri;
414
415	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
416	q->qs[pri] = q->qe[pri] = 0;
417
418	q->size = 0;
419	}
420
421	static int ecb_noinline
422	reqq_push (etp_reqq *q, ETP_REQ *req)
423	{
424	int pri = req->pri;
425	req->next = 0;
426
427	if (q->qe[pri])
428	{
429	q->qe[pri]->next = req;
430	q->qe[pri] = req;
431	}
432	else
433	q->qe[pri] = q->qs[pri] = req;
434
435	return q->size++;
436	}
437
438	static ETP_REQ * ecb_noinline
439	reqq_shift (etp_reqq *q)
440	{
441	int pri;
442
443	if (!q->size)
444	return 0;
445
446	--q->size;
447
448	for (pri = ETP_NUM_PRI; pri--; )
449	{
450	eio_req *req = q->qs[pri];
451
452	if (req)
453	{
454	if (!(q->qs[pri] = (eio_req *)req->next))
455	q->qe[pri] = 0;
456
457	return req;
458	}
459	}
460
461	abort ();
462	}
463
464	static int ecb_cold
465	etp_init (void (*want_poll)(void), void (*done_poll)(void))
466	{
467	X_MUTEX_CREATE (wrklock);
468	X_MUTEX_CREATE (reslock);
469	X_MUTEX_CREATE (reqlock);
470	X_COND_CREATE (reqwait);
471
472	reqq_init (&req_queue);
473	reqq_init (&res_queue);
474
475	wrk_first.next =
476	wrk_first.prev = &wrk_first;
477
478	started = 0;
479	idle = 0;
480	nreqs = 0;
481	nready = 0;
482	npending = 0;
483
484	want_poll_cb = want_poll;
485	done_poll_cb = done_poll;
486
487	return 0;
488	}
489
490	X_THREAD_PROC (etp_proc);
491
492	static void ecb_cold
493	etp_start_thread (void)
494	{
495	etp_worker *wrk = calloc (1, sizeof (etp_worker));
496
497	/*TODO*/
498	assert (("unable to allocate worker thread data", wrk));
499
500	X_LOCK (wrklock);
501
502	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
503	{
504	wrk->prev = &wrk_first;
505	wrk->next = wrk_first.next;
506	wrk_first.next->prev = wrk;
507	wrk_first.next = wrk;
508	++started;
509	}
510	else
511	free (wrk);
512
513	X_UNLOCK (wrklock);
514	}
515
516	static void
517	etp_maybe_start_thread (void)
518	{
519	if (ecb_expect_true (etp_nthreads () >= wanted))
520	return;
521
522	/* todo: maybe use idle here, but might be less exact */
523	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
524	return;
525
526	etp_start_thread ();
527	}
528
529	static void ecb_cold
530	etp_end_thread (void)
531	{
532	eio_req *req = calloc (1, sizeof (eio_req));
533
534	req->type = -1;
535	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
536
537	X_LOCK (reqlock);
538	reqq_push (&req_queue, req);
539	X_COND_SIGNAL (reqwait);
540	X_UNLOCK (reqlock);
541
542	X_LOCK (wrklock);
543	--started;
544	X_UNLOCK (wrklock);
545	}
546
547	static int
548	etp_poll (void)
549	{
550	unsigned int maxreqs;
551	unsigned int maxtime;
552	struct timeval tv_start, tv_now;
553
554	X_LOCK (reslock);
555	maxreqs = max_poll_reqs;
556	maxtime = max_poll_time;
557	X_UNLOCK (reslock);
558
559	if (maxtime)
560	gettimeofday (&tv_start, 0);
561
562	for (;;)
563	{
564	ETP_REQ *req;
565
566	etp_maybe_start_thread ();
567
568	X_LOCK (reslock);
569	req = reqq_shift (&res_queue);
570
571	if (req)
572	{
573	--npending;
574
575	if (!res_queue.size && done_poll_cb)
576	done_poll_cb ();
577	}
578
579	X_UNLOCK (reslock);
580
581	if (!req)
582	return 0;
583
584	X_LOCK (reqlock);
585	--nreqs;
586	X_UNLOCK (reqlock);
587
588	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
589	{
590	req->int1 = 1; /* mark request as delayed */
591	continue;
592	}
593	else
594	{
595	int res = ETP_FINISH (req);
596	if (ecb_expect_false (res))
597	return res;
598	}
599
600	if (ecb_expect_false (maxreqs && !--maxreqs))
601	break;
602
603	if (maxtime)
604	{
605	gettimeofday (&tv_now, 0);
606
607	if (tvdiff (&tv_start, &tv_now) >= maxtime)
608	break;
609	}
610	}
611
612	errno = EAGAIN;
613	return -1;
614	}
615
616	static void
617	etp_cancel (ETP_REQ *req)
618	{
619	req->cancelled = 1;
620
621	eio_grp_cancel (req);
622	}
623
624	static void
625	etp_submit (ETP_REQ *req)
626	{
627	req->pri -= ETP_PRI_MIN;
628
629	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
630	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
631
632	if (ecb_expect_false (req->type == EIO_GROUP))
633	{
634	/* I hope this is worth it :/ */
635	X_LOCK (reqlock);
636	++nreqs;
637	X_UNLOCK (reqlock);
638
639	X_LOCK (reslock);
640
641	++npending;
642
643	if (!reqq_push (&res_queue, req) && want_poll_cb)
644	want_poll_cb ();
645
646	X_UNLOCK (reslock);
647	}
648	else
649	{
650	X_LOCK (reqlock);
651	++nreqs;
652	++nready;
653	reqq_push (&req_queue, req);
654	X_COND_SIGNAL (reqwait);
655	X_UNLOCK (reqlock);
656
657	etp_maybe_start_thread ();
658	}
659	}
660
661	static void ecb_cold
662	etp_set_max_poll_time (double nseconds)
663	{
664	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
665	max_poll_time = nseconds * EIO_TICKS;
666	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
667	}
668
669	static void ecb_cold
670	etp_set_max_poll_reqs (unsigned int maxreqs)
671	{
672	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
673	max_poll_reqs = maxreqs;
674	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
675	}
676
677	static void ecb_cold
678	etp_set_max_idle (unsigned int nthreads)
679	{
680	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
681	max_idle = nthreads;
682	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
683	}
684
685	static void ecb_cold
686	etp_set_idle_timeout (unsigned int seconds)
687	{
688	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
689	idle_timeout = seconds;
690	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
691	}
692
693	static void ecb_cold
694	etp_set_min_parallel (unsigned int nthreads)
695	{
696	if (wanted < nthreads)
697	wanted = nthreads;
698	}
699
700	static void ecb_cold
701	etp_set_max_parallel (unsigned int nthreads)
702	{
703	if (wanted > nthreads)
704	wanted = nthreads;
705
706	while (started > wanted)
707	etp_end_thread ();
708	}
709		393
710	/*****************************************************************************/	394	/*****************************************************************************/
711		395
712	static void	396	static void
713	grp_try_feed (eio_req *grp)	397	grp_try_feed (eio_req *grp)
714	{	398	{
715	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	399	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
716	{	400	{
717	grp->flags &= ~EIO_FLAG_GROUPADD;	401	grp->flags &= ~ETP_FLAG_GROUPADD;
718		402
719	EIO_FEED (grp);	403	EIO_FEED (grp);
720		404
721	/* stop if no progress has been made */	405	/* stop if no progress has been made */
722	if (!(grp->flags & EIO_FLAG_GROUPADD))	406	if (!(grp->flags & ETP_FLAG_GROUPADD))
723	{	407	{
724	grp->feed = 0;	408	grp->feed = 0;
725	break;	409	break;
726	}	410	}
727	}	411	}
…		…
734		418
735	/* call feeder, if applicable */	419	/* call feeder, if applicable */
736	grp_try_feed (grp);	420	grp_try_feed (grp);
737		421
738	/* finish, if done */	422	/* finish, if done */
739	if (!grp->size && grp->int1)	423	if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
740	return eio_finish (grp);	424	return eio_finish (grp);
741	else	425	else
742	return 0;	426	return 0;
743	}	427	}
744		428
…		…
780	}	464	}
781		465
782	void	466	void
783	eio_grp_cancel (eio_req *grp)	467	eio_grp_cancel (eio_req *grp)
784	{	468	{
785	for (grp = grp->grp_first; grp; grp = grp->grp_next)	469	etp_grp_cancel (EIO_POOL, grp);
786	eio_cancel (grp);
787	}	470	}
788		471
789	void	472	void
790	eio_cancel (eio_req *req)	473	eio_cancel (eio_req *req)
791	{	474	{
792	etp_cancel (req);	475	etp_cancel (EIO_POOL, req);
793	}	476	}
794		477
795	void	478	void
796	eio_submit (eio_req *req)	479	eio_submit (eio_req *req)
797	{	480	{
798	etp_submit (req);	481	etp_submit (EIO_POOL, req);
799	}	482	}
800		483
801	unsigned int	484	unsigned int
802	eio_nreqs (void)	485	eio_nreqs (void)
803	{	486	{
804	return etp_nreqs ();	487	return etp_nreqs (EIO_POOL);
805	}	488	}
806		489
807	unsigned int	490	unsigned int
808	eio_nready (void)	491	eio_nready (void)
809	{	492	{
810	return etp_nready ();	493	return etp_nready (EIO_POOL);
811	}	494	}
812		495
813	unsigned int	496	unsigned int
814	eio_npending (void)	497	eio_npending (void)
815	{	498	{
816	return etp_npending ();	499	return etp_npending (EIO_POOL);
817	}	500	}
818		501
819	unsigned int ecb_cold	502	unsigned int ecb_cold
820	eio_nthreads (void)	503	eio_nthreads (void)
821	{	504	{
822	return etp_nthreads ();	505	return etp_nthreads (EIO_POOL);
823	}	506	}
824		507
825	void ecb_cold	508	void ecb_cold
826	eio_set_max_poll_time (double nseconds)	509	eio_set_max_poll_time (double nseconds)
827	{	510	{
828	etp_set_max_poll_time (nseconds);	511	etp_set_max_poll_time (EIO_POOL, nseconds);
829	}	512	}
830		513
831	void ecb_cold	514	void ecb_cold
832	eio_set_max_poll_reqs (unsigned int maxreqs)	515	eio_set_max_poll_reqs (unsigned int maxreqs)
833	{	516	{
834	etp_set_max_poll_reqs (maxreqs);	517	etp_set_max_poll_reqs (EIO_POOL, maxreqs);
835	}	518	}
836		519
837	void ecb_cold	520	void ecb_cold
838	eio_set_max_idle (unsigned int nthreads)	521	eio_set_max_idle (unsigned int nthreads)
839	{	522	{
840	etp_set_max_idle (nthreads);	523	etp_set_max_idle (EIO_POOL, nthreads);
841	}	524	}
842		525
843	void ecb_cold	526	void ecb_cold
844	eio_set_idle_timeout (unsigned int seconds)	527	eio_set_idle_timeout (unsigned int seconds)
845	{	528	{
846	etp_set_idle_timeout (seconds);	529	etp_set_idle_timeout (EIO_POOL, seconds);
847	}	530	}
848		531
849	void ecb_cold	532	void ecb_cold
850	eio_set_min_parallel (unsigned int nthreads)	533	eio_set_min_parallel (unsigned int nthreads)
851	{	534	{
852	etp_set_min_parallel (nthreads);	535	etp_set_min_parallel (EIO_POOL, nthreads);
853	}	536	}
854		537
855	void ecb_cold	538	void ecb_cold
856	eio_set_max_parallel (unsigned int nthreads)	539	eio_set_max_parallel (unsigned int nthreads)
857	{	540	{
858	etp_set_max_parallel (nthreads);	541	etp_set_max_parallel (EIO_POOL, nthreads);
859	}	542	}
860		543
861	int eio_poll (void)	544	int eio_poll (void)
862	{	545	{
863	return etp_poll ();	546	return etp_poll (EIO_POOL);
864	}	547	}
865		548
866	/*****************************************************************************/	549	/*****************************************************************************/
867	/* work around various missing functions */	550	/* work around various missing functions */
868
869	#if !HAVE_PREADWRITE
870	# undef pread
871	# undef pwrite
872	# define pread eio__pread
873	# define pwrite eio__pwrite
874
875	static eio_ssize_t
876	eio__pread (int fd, void *buf, size_t count, off_t offset)
877	{
878	eio_ssize_t res;
879	off_t ooffset;
880
881	X_LOCK (preadwritelock);
882	ooffset = lseek (fd, 0, SEEK_CUR);
883	lseek (fd, offset, SEEK_SET);
884	res = read (fd, buf, count);
885	lseek (fd, ooffset, SEEK_SET);
886	X_UNLOCK (preadwritelock);
887
888	return res;
889	}
890
891	static eio_ssize_t
892	eio__pwrite (int fd, void *buf, size_t count, off_t offset)
893	{
894	eio_ssize_t res;
895	off_t ooffset;
896
897	X_LOCK (preadwritelock);
898	ooffset = lseek (fd, 0, SEEK_CUR);
899	lseek (fd, offset, SEEK_SET);
900	res = write (fd, buf, count);
901	lseek (fd, ooffset, SEEK_SET);
902	X_UNLOCK (preadwritelock);
903
904	return res;
905	}
906	#endif
907		551
908	#ifndef HAVE_UTIMES	552	#ifndef HAVE_UTIMES
909		553
910	# undef utimes	554	# undef utimes
911	# define utimes(path,times) eio__utimes (path, times)	555	# define utimes(path,times) eio__utimes (path, times)
…		…
953	int res;	597	int res;
954		598
955	#if HAVE_SYS_SYNCFS	599	#if HAVE_SYS_SYNCFS
956	res = (int)syscall (__NR_syncfs, (int)(fd));	600	res = (int)syscall (__NR_syncfs, (int)(fd));
957	#else	601	#else
958	res = -1;	602	res = EIO_ENOSYS ();
959	errno = ENOSYS;
960	#endif	603	#endif
961		604
962	if (res < 0 && errno == ENOSYS && fd >= 0)	605	if (res < 0 && errno == ENOSYS && fd >= 0)
963	sync ();	606	sync ();
964		607
…		…
994	}	637	}
995		638
996	static int	639	static int
997	eio__fallocate (int fd, int mode, off_t offset, size_t len)	640	eio__fallocate (int fd, int mode, off_t offset, size_t len)
998	{	641	{
999	#if HAVE_FALLOCATE	642	#if HAVE_LINUX_FALLOCATE
1000	return fallocate (fd, mode, offset, len);	643	return fallocate (fd, mode, offset, len);
1001	#else	644	#else
1002	errno = ENOSYS;	645	return EIO_ENOSYS ();
1003	return -1;
1004	#endif	646	#endif
1005	}	647	}
1006		648
1007	#if !HAVE_READAHEAD	649	#if !HAVE_READAHEAD
1008	# undef readahead	650	# undef readahead
…		…
1023	todo -= len;	665	todo -= len;
1024	}	666	}
1025		667
1026	FUBd;	668	FUBd;
1027		669
1028	errno = 0;	670	/* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
		671	/* but not for e.g. EIO or eof, so we also never fail */
1029	return count;	672	return 0;
1030	}	673	}
1031		674
1032	#endif	675	#endif
1033		676
1034	/* sendfile always needs emulation */	677	/* sendfile always needs emulation */
…		…
1069		712
1070	/* according to source inspection, this is correct, and useful behaviour */	713	/* according to source inspection, this is correct, and useful behaviour */
1071	if (sbytes)	714	if (sbytes)
1072	res = sbytes;	715	res = sbytes;
1073		716
1074	# elif defined (__APPLE__)	717	# elif defined __APPLE__
1075	off_t sbytes = count;	718	off_t sbytes = count;
1076	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);	719	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1077		720
1078	/* according to the manpage, sbytes is always valid */	721	/* according to the manpage, sbytes is always valid */
1079	if (sbytes)	722	if (sbytes)
…		…
1106	HANDLE h = TO_SOCKET (ifd);	749	HANDLE h = TO_SOCKET (ifd);
1107	SetFilePointer (h, offset, 0, FILE_BEGIN);	750	SetFilePointer (h, offset, 0, FILE_BEGIN);
1108	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);	751	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1109		752
1110	#else	753	#else
1111	res = -1;	754	res = EIO_ENOSYS ();
1112	errno = ENOSYS;
1113	#endif	755	#endif
1114		756
1115	/* we assume sendfile can copy at least 128mb in one go */	757	/* we assume sendfile can copy at least 128mb in one go */
1116	if (res <= 128 * 1024 * 1024)	758	if (res <= 128 * 1024 * 1024)
1117	{	759	{
…		…
1304		946
1305	/*****************************************************************************/	947	/*****************************************************************************/
1306	/* requests implemented outside eio_execute, because they are so large */	948	/* requests implemented outside eio_execute, because they are so large */
1307		949
1308	static void	950	static void
1309	eio__realpath (eio_req *req, etp_worker *self)	951	eio__lseek (eio_req *req)
1310	{	952	{
1311	char *rel = req->ptr1;	953	/* this usually gets optimised away completely, or your compiler sucks, */
		954	/* or the whence constants really are not 0, 1, 2 */
		955	int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
		956	: req->int2 == EIO_SEEK_CUR ? SEEK_CUR
		957	: req->int2 == EIO_SEEK_END ? SEEK_END
		958	: req->int2;
		959
		960	req->offs = lseek (req->int1, req->offs, whence);
		961	req->result = req->offs == (off_t)-1 ? -1 : 0;
		962	}
		963
		964	/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
		965	static int
		966	eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
		967	{
1312	char *res;	968	char *res;
		969	const char *rel = path;
1313	char *tmp1, *tmp2;	970	char *tmp1, *tmp2;
1314	#if SYMLOOP_MAX > 32	971	#if SYMLOOP_MAX > 32
1315	int symlinks = SYMLOOP_MAX;	972	int symlinks = SYMLOOP_MAX;
1316	#else	973	#else
1317	int symlinks = 32;	974	int symlinks = 32;
1318	#endif	975	#endif
1319		976
1320	req->result = -1;
1321
1322	errno = EINVAL;	977	errno = EINVAL;
1323	if (!rel)	978	if (!rel)
1324	return;	979	return -1;
1325		980
1326	errno = ENOENT;	981	errno = ENOENT;
1327	if (!*rel)	982	if (!*rel)
1328	return;	983	return -1;
1329		984
1330	if (!req->ptr2)	985	res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1331	{	986	#ifdef _WIN32
1332	X_LOCK (wrklock);	987	if (_access (rel, 4) != 0)
1333	req->flags |= EIO_FLAG_PTR2_FREE;
1334	X_UNLOCK (wrklock);
1335	req->ptr2 = malloc (PATH_MAX * 3);
1336
1337	errno = ENOMEM;
1338	if (!req->ptr2)
1339	return;	988	return -1;
1340	}
1341		989
1342	res = req->ptr2;	990	symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
		991
		992	errno = ENAMETOOLONG;
		993	if (symlinks >= EIO_PATH_MAX * 3)
		994	return -1;
		995
		996	errno = EIO;
		997	if (symlinks <= 0)
		998	return -1;
		999
		1000	return symlinks;
		1001
		1002	#else
1343	tmp1 = res + PATH_MAX;	1003	tmp1 = res + EIO_PATH_MAX;
1344	tmp2 = tmp1 + PATH_MAX;	1004	tmp2 = tmp1 + EIO_PATH_MAX;
1345		1005
1346	#if 0 /* disabled, the musl way to do things is just too racy */	1006	#if 0 /* disabled, the musl way to do things is just too racy */
1347	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)	1007	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1348	/* on linux we may be able to ask the kernel */	1008	/* on linux we may be able to ask the kernel */
1349	{	1009	{
1350	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);	1010	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1351		1011
1352	if (fd >= 0)	1012	if (fd >= 0)
1353	{	1013	{
1354	sprintf (tmp1, "/proc/self/fd/%d", fd);	1014	sprintf (tmp1, "/proc/self/fd/%d", fd);
1355	req->result = readlink (tmp1, res, PATH_MAX);	1015	req->result = readlink (tmp1, res, EIO_PATH_MAX);
		1016	/* here we should probably stat the open file and the disk file, to make sure they still match */
1356	close (fd);	1017	close (fd);
1357
1358	/* here we should probably stat the open file and the disk file, to make sure they still match */
1359		1018
1360	if (req->result > 0)	1019	if (req->result > 0)
1361	goto done;	1020	goto done;
1362	}	1021	}
1363	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)	1022	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1364	return;	1023	return -1;
1365	}	1024	}
1366	#endif	1025	#endif
1367	#endif	1026	#endif
1368		1027
1369	if (*rel != '/')	1028	if (*rel != '/')
1370	{	1029	{
1371	if (!getcwd (res, PATH_MAX))	1030	int len;
		1031
		1032	errno = ENOENT;
		1033	if (wd == EIO_INVALID_WD)
1372	return;	1034	return -1;
		1035
		1036	if (wd == EIO_CWD)
		1037	{
		1038	if (!getcwd (res, EIO_PATH_MAX))
		1039	return -1;
		1040
		1041	len = strlen (res);
		1042	}
		1043	else
		1044	memcpy (res, wd->str, len = wd->len);
1373		1045
1374	if (res [1]) /* only use if not / */	1046	if (res [1]) /* only use if not / */
1375	res += strlen (res);	1047	res += len;
1376	}	1048	}
1377		1049
1378	while (*rel)	1050	while (*rel)
1379	{	1051	{
1380	eio_ssize_t len, linklen;	1052	eio_ssize_t len, linklen;
1381	char *beg = rel;	1053	const char *beg = rel;
1382		1054
1383	while (*rel && *rel != '/')	1055	while (*rel && *rel != '/')
1384	++rel;	1056	++rel;
1385		1057
1386	len = rel - beg;	1058	len = rel - beg;
…		…
1398		1070
1399	if (beg [1] == '.' && len == 2)	1071	if (beg [1] == '.' && len == 2)
1400	{	1072	{
1401	/* .. - back up one component, if possible */	1073	/* .. - back up one component, if possible */
1402		1074
1403	while (res != req->ptr2)	1075	while (res != tmpbuf->ptr)
1404	if (*--res == '/')	1076	if (*--res == '/')
1405	break;	1077	break;
1406		1078
1407	continue;	1079	continue;
1408	}	1080	}
1409	}	1081	}
1410		1082
1411	errno = ENAMETOOLONG;	1083	errno = ENAMETOOLONG;
1412	if (res + 1 + len + 1 >= tmp1)	1084	if (res + 1 + len + 1 >= tmp1)
1413	return;	1085	return -1;
1414		1086
1415	/* copy one component */	1087	/* copy one component */
1416	*res = '/';	1088	*res = '/';
1417	memcpy (res + 1, beg, len);	1089	memcpy (res + 1, beg, len);
1418		1090
1419	/* zero-terminate, for readlink */	1091	/* zero-terminate, for readlink */
1420	res [len + 1] = 0;	1092	res [len + 1] = 0;
1421		1093
1422	/* now check if it's a symlink */	1094	/* now check if it's a symlink */
1423	linklen = readlink (req->ptr2, tmp1, PATH_MAX);	1095	linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1424		1096
1425	if (linklen < 0)	1097	if (linklen < 0)
1426	{	1098	{
1427	if (errno != EINVAL)	1099	if (errno != EINVAL)
1428	return;	1100	return -1;
1429		1101
1430	/* it's a normal directory. hopefully */	1102	/* it's a normal directory. hopefully */
1431	res += len + 1;	1103	res += len + 1;
1432	}	1104	}
1433	else	1105	else
1434	{	1106	{
1435	/* yay, it was a symlink - build new path in tmp2 */	1107	/* yay, it was a symlink - build new path in tmp2 */
1436	int rellen = strlen (rel);	1108	int rellen = strlen (rel);
1437		1109
1438	errno = ENAMETOOLONG;	1110	errno = ENAMETOOLONG;
1439	if (linklen + 1 + rellen >= PATH_MAX)	1111	if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1440	return;	1112	return -1;
1441		1113
1442	errno = ELOOP;	1114	errno = ELOOP;
1443	if (!--symlinks)	1115	if (!--symlinks)
1444	return;	1116	return -1;
1445		1117
1446	if (*tmp1 == '/')	1118	if (*tmp1 == '/')
1447	res = req->ptr2; /* symlink resolves to an absolute path */	1119	res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1448		1120
1449	/* we need to be careful, as rel might point into tmp2 already */	1121	/* we need to be careful, as rel might point into tmp2 already */
1450	memmove (tmp2 + linklen + 1, rel, rellen + 1);	1122	memmove (tmp2 + linklen + 1, rel, rellen + 1);
1451	tmp2 [linklen] = '/';	1123	tmp2 [linklen] = '/';
1452	memcpy (tmp2, tmp1, linklen);	1124	memcpy (tmp2, tmp1, linklen);
…		…
1454	rel = tmp2;	1126	rel = tmp2;
1455	}	1127	}
1456	}	1128	}
1457		1129
1458	/* special case for the lone root path */	1130	/* special case for the lone root path */
1459	if (res == req->ptr2)	1131	if (res == tmpbuf->ptr)
1460	*res++ = '/';	1132	*res++ = '/';
1461		1133
1462	req->result = res - (char *)req->ptr2;	1134	return res - (char *)tmpbuf->ptr;
1463		1135	#endif
1464	done:
1465	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1466	}	1136	}
1467		1137
1468	static signed char	1138	static signed char
1469	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)	1139	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1470	{	1140	{
…		…
1658	#ifdef _WIN32	1328	#ifdef _WIN32
1659	{	1329	{
1660	int len = strlen ((const char *)req->ptr1);	1330	int len = strlen ((const char *)req->ptr1);
1661	char *path = malloc (MAX_PATH);	1331	char *path = malloc (MAX_PATH);
1662	const char *fmt;	1332	const char *fmt;
		1333	const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1663		1334
1664	if (!len)	1335	if (!len)
1665	fmt = "./*";	1336	fmt = "./*";
1666	else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')	1337	else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1667	fmt = "%s*";	1338	fmt = "%s*";
1668	else	1339	else
1669	fmt = "%s/*";	1340	fmt = "%s/*";
1670		1341
1671	_snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);	1342	_snprintf (path, MAX_PATH, fmt, reqpath);
1672	dirp = FindFirstFile (path, &entp);	1343	dirp = FindFirstFile (path, &entp);
1673	free (path);	1344	free (path);
1674		1345
1675	if (dirp == INVALID_HANDLE_VALUE)	1346	if (dirp == INVALID_HANDLE_VALUE)
1676	{	1347	{
1677	dirp = 0;	1348	/* should steal _dosmaperr */
1678
1679	switch (GetLastError ())	1349	switch (GetLastError ())
1680	{	1350	{
1681	case ERROR_FILE_NOT_FOUND:	1351	case ERROR_FILE_NOT_FOUND:
1682	req->result = 0;	1352	req->result = 0;
1683	break;	1353	break;
1684		1354
1685	case ERROR_INVALID_NAME:	1355	case ERROR_INVALID_NAME:
1686	case ERROR_PATH_NOT_FOUND:	1356	case ERROR_PATH_NOT_FOUND:
1687	case ERROR_NO_MORE_FILES:	1357	case ERROR_NO_MORE_FILES:
1688	errno = ENOENT;	1358	errno = ENOENT;
1689	break;	1359	break;
1690		1360
1691	case ERROR_NOT_ENOUGH_MEMORY:	1361	case ERROR_NOT_ENOUGH_MEMORY:
1692	errno = ENOMEM;	1362	errno = ENOMEM;
1693	break;	1363	break;
1694		1364
1695	default:	1365	default:
1696	errno = EINVAL;	1366	errno = EINVAL;
1697	break;	1367	break;
1698	}	1368	}
		1369
		1370	return;
1699	}	1371	}
1700	}	1372	}
1701	#else	1373	#else
		1374	#if HAVE_AT
		1375	if (req->wd)
		1376	{
		1377	int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1378
		1379	if (fd < 0)
		1380	return;
		1381
		1382	dirp = fdopendir (fd);
		1383
		1384	if (!dirp)
		1385	close (fd);
		1386	}
		1387	else
1702	dirp = opendir (req->ptr1);	1388	dirp = opendir (req->ptr1);
		1389	#else
		1390	dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
		1391	#endif
		1392
		1393	if (!dirp)
		1394	return;
1703	#endif	1395	#endif
1704		1396
1705	if (req->flags & EIO_FLAG_PTR1_FREE)	1397	if (req->flags & EIO_FLAG_PTR1_FREE)
1706	free (req->ptr1);	1398	free (req->ptr1);
1707		1399
1708	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;	1400	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1709	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;	1401	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1710	req->ptr2 = names = malloc (namesalloc);	1402	req->ptr2 = names = malloc (namesalloc);
1711		1403
1712	if (dirp && names && (!flags || dents))	1404	if (!names || (flags && !dents))
		1405	return;
		1406
1713	for (;;)	1407	for (;;)
1714	{	1408	{
1715	int done;	1409	int done;
1716		1410
1717	#ifdef _WIN32	1411	#ifdef _WIN32
1718	done = !dirp;	1412	done = !dirp;
1719	#else	1413	#else
1720	errno = 0;	1414	errno = 0;
1721	entp = readdir (dirp);	1415	entp = readdir (dirp);
1722	done = !entp;	1416	done = !entp;
1723	#endif	1417	#endif
1724		1418
1725	if (done)	1419	if (done)
1726	{	1420	{
1727	#ifndef _WIN32	1421	#ifndef _WIN32
1728	int old_errno = errno;	1422	int old_errno = errno;
1729	closedir (dirp);	1423	closedir (dirp);
1730	errno = old_errno;	1424	errno = old_errno;
1731		1425
1732	if (errno)	1426	if (errno)
1733	break;	1427	break;
1734	#endif	1428	#endif
1735		1429
1736	/* sort etc. */	1430	/* sort etc. */
1737	req->int1 = flags;	1431	req->int1 = flags;
1738	req->result = dentoffs;	1432	req->result = dentoffs;
1739		1433
1740	if (flags & EIO_READDIR_STAT_ORDER)	1434	if (flags & EIO_READDIR_STAT_ORDER)
1741	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);	1435	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1742	else if (flags & EIO_READDIR_DIRS_FIRST)	1436	else if (flags & EIO_READDIR_DIRS_FIRST)
1743	if (flags & EIO_READDIR_FOUND_UNKNOWN)	1437	if (flags & EIO_READDIR_FOUND_UNKNOWN)
1744	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */	1438	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1745	else	1439	else
		1440	{
		1441	/* in this case, all is known, and we just put dirs first and sort them */
		1442	eio_dirent *oth = dents + dentoffs;
		1443	eio_dirent *dir = dents;
		1444
		1445	/* now partition dirs to the front, and non-dirs to the back */
		1446	/* by walking from both sides and swapping if necessary */
		1447	while (oth > dir)
		1448	{
		1449	if (dir->type == EIO_DT_DIR)
		1450	++dir;
		1451	else if ((--oth)->type == EIO_DT_DIR)
		1452	{
		1453	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1454
		1455	++dir;
		1456	}
		1457	}
		1458
		1459	/* now sort the dirs only (dirs all have the same score) */
		1460	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1461	}
		1462
		1463	break;
		1464	}
		1465
		1466	/* now add the entry to our list(s) */
		1467	name = D_NAME (entp);
		1468
		1469	/* skip . and .. entries */
		1470	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
		1471	{
		1472	int len = D_NAMLEN (entp) + 1;
		1473
		1474	while (ecb_expect_false (namesoffs + len > namesalloc))
		1475	{
		1476	namesalloc *= 2;
		1477	req->ptr2 = names = realloc (names, namesalloc);
		1478
		1479	if (!names)
		1480	break;
		1481	}
		1482
		1483	memcpy (names + namesoffs, name, len);
		1484
		1485	if (dents)
		1486	{
		1487	struct eio_dirent *ent;
		1488
		1489	if (ecb_expect_false (dentoffs == dentalloc))
1746	{	1490	{
1747	/* in this case, all is known, and we just put dirs first and sort them */	1491	dentalloc *= 2;
		1492	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1493
		1494	if (!dents)
		1495	break;
		1496	}
		1497
1748	eio_dirent *oth = dents + dentoffs;	1498	ent = dents + dentoffs;
1749	eio_dirent *dir = dents;
1750		1499
1751	/* now partition dirs to the front, and non-dirs to the back */	1500	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1752	/* by walking from both sides and swapping if necessary */	1501	ent->namelen = len - 1;
1753	while (oth > dir)	1502	ent->inode = D_INO (entp);
		1503
		1504	inode_bits |= ent->inode;
		1505
		1506	switch (D_TYPE (entp))
		1507	{
		1508	default:
		1509	ent->type = EIO_DT_UNKNOWN;
		1510	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1511	break;
		1512
		1513	#ifdef DT_FIFO
		1514	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1515	#endif
		1516	#ifdef DT_CHR
		1517	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1518	#endif
		1519	#ifdef DT_MPC
		1520	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1521	#endif
		1522	#ifdef DT_DIR
		1523	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1524	#endif
		1525	#ifdef DT_NAM
		1526	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1527	#endif
		1528	#ifdef DT_BLK
		1529	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1530	#endif
		1531	#ifdef DT_MPB
		1532	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1533	#endif
		1534	#ifdef DT_REG
		1535	case DT_REG: ent->type = EIO_DT_REG; break;
		1536	#endif
		1537	#ifdef DT_NWK
		1538	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1539	#endif
		1540	#ifdef DT_CMP
		1541	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1542	#endif
		1543	#ifdef DT_LNK
		1544	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1545	#endif
		1546	#ifdef DT_SOCK
		1547	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1548	#endif
		1549	#ifdef DT_DOOR
		1550	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1551	#endif
		1552	#ifdef DT_WHT
		1553	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1554	#endif
		1555	}
		1556
		1557	ent->score = 7;
		1558
		1559	if (flags & EIO_READDIR_DIRS_FIRST)
		1560	{
		1561	if (ent->type == EIO_DT_UNKNOWN)
1754	{	1562	{
1755	if (dir->type == EIO_DT_DIR)	1563	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1756	++dir;	1564	ent->score = 1;
1757	else if ((--oth)->type == EIO_DT_DIR)	1565	else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1758	{	1566	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1759	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1760
1761	++dir;
1762	}
1763	}	1567	}
1764		1568	else if (ent->type == EIO_DT_DIR)
1765	/* now sort the dirs only (dirs all have the same score) */	1569	ent->score = 0;
1766	eio_dent_sort (dents, dir - dents, 0, inode_bits);
1767	}	1570	}
1768
1769	break;
1770	}
1771
1772	/* now add the entry to our list(s) */
1773	name = D_NAME (entp);
1774
1775	/* skip . and .. entries */
1776	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1777	{
1778	int len = D_NAMLEN (entp) + 1;
1779
1780	while (ecb_expect_false (namesoffs + len > namesalloc))
1781	{
1782	namesalloc *= 2;
1783	req->ptr2 = names = realloc (names, namesalloc);
1784
1785	if (!names)
1786	break;
1787	}	1571	}
1788		1572
1789	memcpy (names + namesoffs, name, len);
1790
1791	if (dents)
1792	{
1793	struct eio_dirent *ent;
1794
1795	if (ecb_expect_false (dentoffs == dentalloc))
1796	{
1797	dentalloc *= 2;
1798	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1799
1800	if (!dents)
1801	break;
1802	}
1803
1804	ent = dents + dentoffs;
1805
1806	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1807	ent->namelen = len - 1;
1808	ent->inode = D_INO (entp);
1809
1810	inode_bits |= ent->inode;
1811
1812	switch (D_TYPE (entp))
1813	{
1814	default:
1815	ent->type = EIO_DT_UNKNOWN;
1816	flags |= EIO_READDIR_FOUND_UNKNOWN;
1817	break;
1818
1819	#ifdef DT_FIFO
1820	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1821	#endif
1822	#ifdef DT_CHR
1823	case DT_CHR: ent->type = EIO_DT_CHR; break;
1824	#endif
1825	#ifdef DT_MPC
1826	case DT_MPC: ent->type = EIO_DT_MPC; break;
1827	#endif
1828	#ifdef DT_DIR
1829	case DT_DIR: ent->type = EIO_DT_DIR; break;
1830	#endif
1831	#ifdef DT_NAM
1832	case DT_NAM: ent->type = EIO_DT_NAM; break;
1833	#endif
1834	#ifdef DT_BLK
1835	case DT_BLK: ent->type = EIO_DT_BLK; break;
1836	#endif
1837	#ifdef DT_MPB
1838	case DT_MPB: ent->type = EIO_DT_MPB; break;
1839	#endif
1840	#ifdef DT_REG
1841	case DT_REG: ent->type = EIO_DT_REG; break;
1842	#endif
1843	#ifdef DT_NWK
1844	case DT_NWK: ent->type = EIO_DT_NWK; break;
1845	#endif
1846	#ifdef DT_CMP
1847	case DT_CMP: ent->type = EIO_DT_CMP; break;
1848	#endif
1849	#ifdef DT_LNK
1850	case DT_LNK: ent->type = EIO_DT_LNK; break;
1851	#endif
1852	#ifdef DT_SOCK
1853	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1854	#endif
1855	#ifdef DT_DOOR
1856	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1857	#endif
1858	#ifdef DT_WHT
1859	case DT_WHT: ent->type = EIO_DT_WHT; break;
1860	#endif
1861	}
1862
1863	ent->score = 7;
1864
1865	if (flags & EIO_READDIR_DIRS_FIRST)
1866	{
1867	if (ent->type == EIO_DT_UNKNOWN)
1868	{
1869	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1870	ent->score = 1;
1871	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1872	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1873	}
1874	else if (ent->type == EIO_DT_DIR)
1875	ent->score = 0;
1876	}
1877	}
1878
1879	namesoffs += len;	1573	namesoffs += len;
1880	++dentoffs;	1574	++dentoffs;
1881	}	1575	}
1882		1576
1883	if (EIO_CANCELLED (req))	1577	if (EIO_CANCELLED (req))
1884	{	1578	{
1885	errno = ECANCELED;	1579	errno = ECANCELED;
1886	break;	1580	break;
1887	}	1581	}
1888		1582
1889	#ifdef _WIN32	1583	#ifdef _WIN32
1890	if (!FindNextFile (dirp, &entp))	1584	if (!FindNextFile (dirp, &entp))
1891	{	1585	{
1892	FindClose (dirp);	1586	FindClose (dirp);
1893	dirp = 0;	1587	dirp = 0;
1894	}	1588	}
1895	#endif	1589	#endif
1896	}	1590	}
1897	}	1591	}
		1592
		1593	/*****************************************************************************/
		1594	/* working directory stuff */
		1595	/* various deficiencies in the posix 2008 api force us to */
		1596	/* keep the absolute path in string form at all times */
		1597	/* fuck yeah. */
		1598
		1599	#if !HAVE_AT
		1600
		1601	/* a bit like realpath, but usually faster because it doesn'T have to return */
		1602	/* an absolute or canonical path */
		1603	static const char *
		1604	wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1605	{
		1606	if (!wd || *path == '/')
		1607	return path;
		1608
		1609	if (path [0] == '.' && !path [1])
		1610	return wd->str;
		1611
		1612	{
		1613	int l1 = wd->len;
		1614	int l2 = strlen (path);
		1615
		1616	char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
		1617
		1618	memcpy (res, wd->str, l1);
		1619	res [l1] = '/';
		1620	memcpy (res + l1 + 1, path, l2 + 1);
		1621
		1622	return res;
		1623	}
		1624	}
		1625
		1626	#endif
		1627
		1628	static eio_wd
		1629	eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1630	{
		1631	int fd;
		1632	eio_wd res;
		1633	int len = eio__realpath (tmpbuf, wd, path);
		1634
		1635	if (len < 0)
		1636	return EIO_INVALID_WD;
		1637
		1638	#if HAVE_AT
		1639	fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1640
		1641	if (fd < 0)
		1642	return EIO_INVALID_WD;
		1643	#endif
		1644
		1645	res = malloc (sizeof (*res) + len); /* one extra 0-byte */
		1646
		1647	#if HAVE_AT
		1648	res->fd = fd;
		1649	#endif
		1650
		1651	res->len = len;
		1652	memcpy (res->str, tmpbuf->ptr, len);
		1653	res->str [len] = 0;
		1654
		1655	return res;
		1656	}
		1657
		1658	eio_wd
		1659	eio_wd_open_sync (eio_wd wd, const char *path)
		1660	{
		1661	struct etp_tmpbuf tmpbuf = { };
		1662	wd = eio__wd_open_sync (&tmpbuf, wd, path);
		1663	free (tmpbuf.ptr);
		1664
		1665	return wd;
		1666	}
		1667
		1668	void
		1669	eio_wd_close_sync (eio_wd wd)
		1670	{
		1671	if (wd != EIO_INVALID_WD && wd != EIO_CWD)
		1672	{
		1673	#if HAVE_AT
		1674	close (wd->fd);
		1675	#endif
		1676	free (wd);
		1677	}
		1678	}
		1679
		1680	#if HAVE_AT
		1681
		1682	/* they forgot these */
		1683
		1684	static int
		1685	eio__truncateat (int dirfd, const char *path, off_t length)
		1686	{
		1687	int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
		1688	int res;
		1689
		1690	if (fd < 0)
		1691	return fd;
		1692
		1693	res = ftruncate (fd, length);
		1694	close (fd);
		1695	return res;
		1696	}
		1697
		1698	static int
		1699	eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
		1700	{
		1701	int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
		1702	int res;
		1703
		1704	if (fd < 0)
		1705	return fd;
		1706
		1707	res = fstatvfs (fd, buf);
		1708	close (fd);
		1709	return res;
		1710
		1711	}
		1712
		1713	#endif
1898		1714
1899	/*****************************************************************************/	1715	/*****************************************************************************/
1900		1716
1901	#define ALLOC(len) \	1717	#define ALLOC(len) \
1902	if (!req->ptr2) \	1718	if (!req->ptr2) \
1903	{ \	1719	{ \
1904	X_LOCK (wrklock); \	1720	X_LOCK (EIO_POOL->wrklock); \
1905	req->flags |= EIO_FLAG_PTR2_FREE; \	1721	req->flags |= EIO_FLAG_PTR2_FREE; \
1906	X_UNLOCK (wrklock); \	1722	X_UNLOCK (EIO_POOL->wrklock); \
1907	req->ptr2 = malloc (len); \	1723	req->ptr2 = malloc (len); \
1908	if (!req->ptr2) \	1724	if (!req->ptr2) \
1909	{ \	1725	{ \
1910	errno = ENOMEM; \	1726	errno = ENOMEM; \
1911	req->result = -1; \	1727	req->result = -1; \
1912	break; \	1728	break; \
1913	} \	1729	} \
1914	}	1730	}
1915		1731
1916	X_THREAD_PROC (etp_proc)
1917	{
1918	ETP_REQ *req;
1919	struct timespec ts;
1920	etp_worker *self = (etp_worker *)thr_arg;
1921
1922	#if HAVE_PRCTL_SET_NAME
1923	prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1924	#endif
1925
1926	/* try to distribute timeouts somewhat evenly */
1927	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1928
1929	for (;;)
1930	{
1931	ts.tv_sec = 0;
1932
1933	X_LOCK (reqlock);
1934
1935	for (;;)
1936	{
1937	self->req = req = reqq_shift (&req_queue);
1938
1939	if (req)
1940	break;
1941
1942	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1943	{
1944	X_UNLOCK (reqlock);
1945	X_LOCK (wrklock);
1946	--started;
1947	X_UNLOCK (wrklock);
1948	goto quit;
1949	}
1950
1951	++idle;
1952
1953	if (idle <= max_idle)
1954	/* we are allowed to idle, so do so without any timeout */
1955	X_COND_WAIT (reqwait, reqlock);
1956	else
1957	{
1958	/* initialise timeout once */
1959	if (!ts.tv_sec)
1960	ts.tv_sec = time (0) + idle_timeout;
1961
1962	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1963	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1964	}
1965
1966	--idle;
1967	}
1968
1969	--nready;
1970
1971	X_UNLOCK (reqlock);
1972
1973	if (req->type < 0)
1974	goto quit;
1975
1976	ETP_EXECUTE (self, req);
1977
1978	X_LOCK (reslock);
1979
1980	++npending;
1981
1982	if (!reqq_push (&res_queue, req) && want_poll_cb)
1983	want_poll_cb ();
1984
1985	self->req = 0;
1986	etp_worker_clear (self);
1987
1988	X_UNLOCK (reslock);
1989	}
1990
1991	quit:
1992	X_LOCK (wrklock);
1993	etp_worker_free (self);
1994	X_UNLOCK (wrklock);
1995
1996	return 0;
1997	}
1998
1999	/*****************************************************************************/	1732	/*****************************************************************************/
2000		1733
2001	int ecb_cold	1734	int ecb_cold
2002	eio_init (void (*want_poll)(void), void (*done_poll)(void))	1735	eio_init (void (*want_poll)(void), void (*done_poll)(void))
2003	{	1736	{
2004	#if !HAVE_PREADWRITE	1737	eio_want_poll_cb = want_poll;
2005	X_MUTEX_CREATE (preadwritelock);	1738	eio_done_poll_cb = done_poll;
2006	#endif
2007		1739
2008	return etp_init (want_poll, done_poll);	1740	return etp_init (EIO_POOL, 0, 0, 0);
2009	}	1741	}
2010		1742
2011	ecb_inline void	1743	ecb_inline void
2012	eio_api_destroy (eio_req *req)	1744	eio_api_destroy (eio_req *req)
2013	{	1745	{
2014	free (req);	1746	free (req);
2015	}	1747	}
2016		1748
2017	#define REQ(rtype) \	1749	#define REQ(rtype) \
2018	eio_req *req; \	1750	eio_req *req; \
2019	\	1751	\
2020	req = (eio_req *)calloc (1, sizeof *req); \	1752	req = (eio_req *)calloc (1, sizeof *req); \
2021	if (!req) \	1753	if (!req) \
2022	return 0; \	1754	return 0; \
…		…
2036	{ \	1768	{ \
2037	eio_api_destroy (req); \	1769	eio_api_destroy (req); \
2038	return 0; \	1770	return 0; \
2039	}	1771	}
2040		1772
		1773	#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
		1774
2041	static void	1775	static void
2042	eio_execute (etp_worker *self, eio_req *req)	1776	eio_execute (etp_worker *self, eio_req *req)
2043	{	1777	{
		1778	#if HAVE_AT
		1779	int dirfd;
		1780	#else
		1781	const char *path;
		1782	#endif
		1783
2044	if (ecb_expect_false (EIO_CANCELLED (req)))	1784	if (ecb_expect_false (EIO_CANCELLED (req)))
2045	{	1785	{
2046	req->result = -1;	1786	req->result = -1;
2047	req->errorno = ECANCELED;	1787	req->errorno = ECANCELED;
2048	return;	1788	return;
2049	}	1789	}
2050		1790
		1791	if (ecb_expect_false (req->wd == EIO_INVALID_WD))
		1792	{
		1793	req->result = -1;
		1794	req->errorno = ENOENT;
		1795	return;
		1796	}
		1797
		1798	if (req->type >= EIO_OPEN)
		1799	{
		1800	#if HAVE_AT
		1801	dirfd = WD2FD (req->wd);
		1802	#else
		1803	path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		1804	#endif
		1805	}
		1806
2051	switch (req->type)	1807	switch (req->type)
2052	{	1808	{
		1809	case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
		1810	req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
		1811	break;
		1812	case EIO_WD_CLOSE: req->result = 0;
		1813	eio_wd_close_sync (req->wd); break;
		1814
		1815	case EIO_SEEK: eio__lseek (req); break;
2053	case EIO_READ: ALLOC (req->size);	1816	case EIO_READ: ALLOC (req->size);
2054	req->result = req->offs >= 0	1817	req->result = req->offs >= 0
2055	? pread (req->int1, req->ptr2, req->size, req->offs)	1818	? pread (req->int1, req->ptr2, req->size, req->offs)
2056	: read (req->int1, req->ptr2, req->size); break;	1819	: read (req->int1, req->ptr2, req->size); break;
2057	case EIO_WRITE: req->result = req->offs >= 0	1820	case EIO_WRITE: req->result = req->offs >= 0
2058	? pwrite (req->int1, req->ptr2, req->size, req->offs)	1821	? pwrite (req->int1, req->ptr2, req->size, req->offs)
2059	: write (req->int1, req->ptr2, req->size); break;	1822	: write (req->int1, req->ptr2, req->size); break;
2060		1823
		1824	case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
		1825	case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
		1826
2061	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;	1827	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2062	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;	1828	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2063		1829
		1830	#if HAVE_AT
		1831
2064	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1832	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2065	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1833	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
2066	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1834	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1835	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
		1836	case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
		1837	case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
		1838	case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
		1839	case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
		1840
		1841	case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
		1842	case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
		1843	req->result = req->wd && SINGLEDOT (req->ptr1)
		1844	? rmdir (req->wd->str)
		1845	: unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
		1846	case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
		1847	case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
		1848	req->result = req->wd && SINGLEDOT (req->ptr1)
		1849	? rename (req->wd->str, req->ptr2)
		1850	: renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
		1851	case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
		1852	case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
		1853	case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		1854	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1855	req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1856	case EIO_READLINK: ALLOC (EIO_PATH_MAX);
		1857	req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
		1858	if (req->result == EIO_PATH_MAX)
		1859	{
		1860	req->result = -1;
		1861	errno = ENAMETOOLONG;
		1862	}
		1863	break;
		1864	case EIO_UTIME:
		1865	case EIO_FUTIME:
		1866	{
		1867	struct timespec ts[2];
		1868	struct timespec *times;
		1869
		1870	if (req->nv1 != -1. || req->nv2 != -1.)
		1871	{
		1872	ts[0].tv_sec = req->nv1;
		1873	ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
		1874	ts[1].tv_sec = req->nv2;
		1875	ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
		1876
		1877	times = ts;
		1878	}
		1879	else
		1880	times = 0;
		1881
		1882	req->result = req->type == EIO_FUTIME
		1883	? futimens (req->int1, times)
		1884	: utimensat (dirfd, req->ptr1, times, 0);
		1885	}
		1886	break;
		1887
		1888	#else
		1889
		1890	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		1891	req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		1892	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2067	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1893	req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		1894	case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
		1895	case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
		1896	case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
		1897	case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
		1898
		1899	case EIO_UNLINK: req->result = unlink (path ); break;
		1900	case EIO_RMDIR: req->result = rmdir (path ); break;
		1901	case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
		1902	case EIO_RENAME: req->result = rename (path , req->ptr2); break;
		1903	case EIO_LINK: req->result = link (path , req->ptr2); break;
		1904	case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
		1905	case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
		1906	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1907	req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1908	case EIO_READLINK: ALLOC (EIO_PATH_MAX);
		1909	req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
		1910	if (req->result == EIO_PATH_MAX)
		1911	{
		1912	req->result = -1;
		1913	errno = ENAMETOOLONG;
		1914	}
		1915	break;
		1916
		1917	case EIO_UTIME:
		1918	case EIO_FUTIME:
		1919	{
		1920	struct timeval tv[2];
		1921	struct timeval *times;
		1922
		1923	if (req->nv1 != -1. || req->nv2 != -1.)
		1924	{
		1925	tv[0].tv_sec = req->nv1;
		1926	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
		1927	tv[1].tv_sec = req->nv2;
		1928	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
		1929
		1930	times = tv;
		1931	}
		1932	else
		1933	times = 0;
		1934
		1935	req->result = req->type == EIO_FUTIME
		1936	? futimes (req->int1, times)
		1937	: utimes (req->ptr1, times);
		1938	}
		1939	break;
		1940
		1941	#endif
		1942
		1943	case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
		1944	{
		1945	ALLOC (req->result);
		1946	memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
		1947	}
		1948	break;
		1949
2068	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1950	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2069	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1951	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2070		1952
2071	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2072	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2073	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));	1953	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2074	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;	1954	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2075		1955
2076	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
2077	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	1956	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2078	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
2079	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	1957	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2080	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
2081	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;	1958	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2082		1959
2083	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
2084	case EIO_CLOSE: req->result = close (req->int1); break;	1960	case EIO_CLOSE: req->result = close (req->int1); break;
2085	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;	1961	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2086	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
2087	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2088	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2089	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2090	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2091	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2092	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2093
2094	case EIO_REALPATH: eio__realpath (req, self); break;
2095
2096	case EIO_READLINK: ALLOC (PATH_MAX);
2097	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2098
2099	case EIO_SYNC: req->result = 0; sync (); break;	1962	case EIO_SYNC: req->result = 0; sync (); break;
2100	case EIO_FSYNC: req->result = fsync (req->int1); break;	1963	case EIO_FSYNC: req->result = fsync (req->int1); break;
2101	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	1964	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2102	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;	1965	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2103	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;	1966	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
…		…
2122	req->result = select (0, 0, 0, 0, &tv);	1985	req->result = select (0, 0, 0, 0, &tv);
2123	}	1986	}
2124	#endif	1987	#endif
2125	break;	1988	break;
2126		1989
2127	case EIO_UTIME:	1990	#if 0
2128	case EIO_FUTIME:
2129	{
2130	struct timeval tv[2];
2131	struct timeval *times;
2132
2133	if (req->nv1 != -1. || req->nv2 != -1.)
2134	{
2135	tv[0].tv_sec = req->nv1;
2136	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2137	tv[1].tv_sec = req->nv2;
2138	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2139
2140	times = tv;
2141	}
2142	else
2143	times = 0;
2144
2145	req->result = req->type == EIO_FUTIME
2146	? futimes (req->int1, times)
2147	: utimes (req->ptr1, times);
2148	}
2149	break;
2150
2151	case EIO_GROUP:	1991	case EIO_GROUP:
2152	abort (); /* handled in eio_request */	1992	abort (); /* handled in eio_request */
		1993	#endif
2153		1994
2154	case EIO_NOP:	1995	case EIO_NOP:
2155	req->result = 0;	1996	req->result = 0;
2156	break;	1997	break;
2157		1998
2158	case EIO_CUSTOM:	1999	case EIO_CUSTOM:
2159	req->feed (req);	2000	req->feed (req);
2160	break;	2001	break;
2161		2002
2162	default:	2003	default:
2163	errno = ENOSYS;
2164	req->result = -1;	2004	req->result = EIO_ENOSYS ();
2165	break;	2005	break;
2166	}	2006	}
2167		2007
2168	req->errorno = errno;	2008	req->errorno = errno;
2169	}	2009	}
2170		2010
2171	#ifndef EIO_NO_WRAPPERS	2011	#ifndef EIO_NO_WRAPPERS
2172		2012
		2013	eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
		2014	{
		2015	REQ (EIO_WD_OPEN); PATH; SEND;
		2016	}
		2017
		2018	eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
		2019	{
		2020	REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
		2021	}
		2022
2173	eio_req *eio_nop (int pri, eio_cb cb, void *data)	2023	eio_req *eio_nop (int pri, eio_cb cb, void *data)
2174	{	2024	{
2175	REQ (EIO_NOP); SEND;	2025	REQ (EIO_NOP); SEND;
2176	}	2026	}
2177		2027
…		…
2238	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)	2088	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2239	{	2089	{
2240	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;	2090	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2241	}	2091	}
2242		2092
		2093	eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
		2094	{
		2095	REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
		2096	}
		2097
2243	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)	2098	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2244	{	2099	{
2245	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;	2100	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2246	}	2101	}
2247		2102
2248	eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)	2103	eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2249	{	2104	{
2250	REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;	2105	REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
		2106	}
		2107
		2108	eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
		2109	{
		2110	REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
		2111	}
		2112
		2113	eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
		2114	{
		2115	REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2251	}	2116	}
2252		2117
2253	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2118	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2254	{	2119	{
2255	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2120	REQ (EIO_FSTAT); req->int1 = fd; SEND;
…		…
2443	void	2308	void
2444	eio_grp_add (eio_req *grp, eio_req *req)	2309	eio_grp_add (eio_req *grp, eio_req *req)
2445	{	2310	{
2446	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2311	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2447		2312
2448	grp->flags |= EIO_FLAG_GROUPADD;	2313	grp->flags |= ETP_FLAG_GROUPADD;
2449		2314
2450	++grp->size;	2315	++grp->size;
2451	req->grp = grp;	2316	req->grp = grp;
2452		2317
2453	req->grp_prev = 0;	2318	req->grp_prev = 0;

Diff Legend

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