ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.18 by root, Mon Jun 16 12:19:55 2008 UTC vs.
Revision 1.86 by root, Thu Jul 14 18:30:10 2011 UTC

1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de>	4	* Copyright (c) 2007,2008,2009,2010,2011 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	*
…		…
35	* and other provisions required by the GPL. If you do not delete the	35	* and other provisions required by the GPL. If you do not delete the
36	* provisions above, a recipient may use your version of this file under	36	* provisions above, a recipient may use your version of this file under
37	* either the BSD or the GPL.	37	* either the BSD or the GPL.
38	*/	38	*/
39		39
		40	#ifndef _WIN32
		41	# include "config.h"
		42	#endif
		43
40	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
		46
		47	#ifdef EIO_STACKSIZE
		48	# define XTHREAD_STACKSIZE EIO_STACKSIZE
		49	#endif
41	#include "xthread.h"	50	#include "xthread.h"
42		51
43	#include <errno.h>	52	#include <errno.h>
44	#include <stddef.h>	53	#include <stddef.h>
45	#include <stdlib.h>	54	#include <stdlib.h>
…		…
49	#include <sys/stat.h>	58	#include <sys/stat.h>
50	#include <limits.h>	59	#include <limits.h>
51	#include <fcntl.h>	60	#include <fcntl.h>
52	#include <assert.h>	61	#include <assert.h>
53		62
		63	#include <sys/statvfs.h>
		64	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
		65	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
		66	#if HAVE_STDINT_H
		67	# include <stdint.h>
		68	#endif
		69
		70	#ifndef ECANCELED
		71	# define ECANCELED EDOM
		72	#endif
		73
		74	static void eio_destroy (eio_req *req);
		75
54	#ifndef EIO_FINISH	76	#ifndef EIO_FINISH
55	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	77	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
56	#endif	78	#endif
57		79
58	#ifndef EIO_DESTROY	80	#ifndef EIO_DESTROY
…		…
61		83
62	#ifndef EIO_FEED	84	#ifndef EIO_FEED
63	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	85	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
64	#endif	86	#endif
65		87
		88	#ifndef EIO_FD_TO_WIN32_HANDLE
		89	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		90	#endif
		91	#ifndef EIO_WIN32_HANDLE_TO_FD
		92	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		93	#endif
		94
		95	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		96
		97	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		98
66	#ifdef _WIN32	99	#ifdef _WIN32
67		100
68	/*doh*/	101	#define PAGESIZE 4096 /* GetSystemInfo? */
		102
		103	#define stat(path,buf) _stati64 (path,buf)
		104	#define lstat(path,buf) stat (path,buf)
		105	#define fstat(fd,buf) _fstati64 (path,buf)
		106	#define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		107	#define mkdir(path,mode) _mkdir (path)
		108	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		109
		110	#define chown(path,uid,gid) EIO_ENOSYS ()
		111	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		112	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		113	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		114	#define mknod(path,mode,dev) EIO_ENOSYS ()
		115	#define sync() EIO_ENOSYS ()
		116
		117	/* we could even stat and see if it exists */
		118	static int
		119	symlink (const char *old, const char *neu)
		120	{
		121	if (CreateSymbolicLink (neu, old, 1))
		122	return 0;
		123
		124	if (CreateSymbolicLink (neu, old, 0))
		125	return 0;
		126
		127	return EIO_ERRNO (ENOENT, -1);
		128	}
69		129
70	#else	130	#else
71		131
72	# include "config.h"
73	# include <sys/time.h>	132	#include <sys/time.h>
74	# include <sys/select.h>	133	#include <sys/select.h>
		134	#include <sys/statvfs.h>
75	# include <unistd.h>	135	#include <unistd.h>
76	# include <utime.h>	136	#include <utime.h>
77	# include <signal.h>	137	#include <signal.h>
78	# include <dirent.h>	138	#include <dirent.h>
79		139
		140	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		141	#include <sys/mman.h>
		142	#endif
		143
		144	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		145	#if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
		146	#define _DIRENT_HAVE_D_TYPE /* sigh */
		147	#define D_INO(de) (de)->d_fileno
		148	#define D_NAMLEN(de) (de)->d_namlen
		149	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		150	#define D_INO(de) (de)->d_ino
		151	#endif
		152
		153	#ifdef _D_EXACT_NAMLEN
		154	#undef D_NAMLEN
		155	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		156	#endif
		157
		158	#ifdef _DIRENT_HAVE_D_TYPE
		159	#define D_TYPE(de) (de)->d_type
		160	#endif
		161
80	# ifndef EIO_STRUCT_DIRENT	162	#ifndef EIO_STRUCT_DIRENT
81	# define EIO_STRUCT_DIRENT struct dirent	163	#define EIO_STRUCT_DIRENT struct dirent
82	# endif	164	#endif
83		165
84	#endif	166	#endif
85		167
86	#if HAVE_SENDFILE	168	#if HAVE_SENDFILE
87	# if __linux	169	# if __linux
88	# include <sys/sendfile.h>	170	# include <sys/sendfile.h>
89	# elif __freebsd	171	# elif __FreeBSD__ || defined __APPLE__
90	# include <sys/socket.h>	172	# include <sys/socket.h>
91	# include <sys/uio.h>	173	# include <sys/uio.h>
92	# elif __hpux	174	# elif __hpux
93	# include <sys/socket.h>	175	# include <sys/socket.h>
94	# elif __solaris /* not yet */	176	# elif __solaris
95	# include <sys/sendfile.h>	177	# include <sys/sendfile.h>
96	# else	178	# else
97	# error sendfile support requested but not available	179	# error sendfile support requested but not available
98	# endif	180	# endif
99	#endif	181	#endif
100		182
101	/* number of seconds after which an idle threads exit */	183	#ifndef D_TYPE
102	#define IDLE_TIMEOUT 10	184	# define D_TYPE(de) 0
		185	#endif
		186	#ifndef D_INO
		187	# define D_INO(de) 0
		188	#endif
		189	#ifndef D_NAMLEN
		190	# define D_NAMLEN(de) strlen ((de)->d_name)
		191	#endif
103		192
104	/* used for struct dirent, AIX doesn't provide it */	193	/* used for struct dirent, AIX doesn't provide it */
105	#ifndef NAME_MAX	194	#ifndef NAME_MAX
106	# define NAME_MAX 4096	195	# define NAME_MAX 4096
		196	#endif
		197
		198	/* used for readlink etc. */
		199	#ifndef PATH_MAX
		200	# define PATH_MAX 4096
107	#endif	201	#endif
108		202
109	/* buffer size for various temporary buffers */	203	/* buffer size for various temporary buffers */
110	#define EIO_BUFSIZE 65536	204	#define EIO_BUFSIZE 65536
111		205
…		…
117	errno = ENOMEM; \	211	errno = ENOMEM; \
118	if (!eio_buf) \	212	if (!eio_buf) \
119	return -1;	213	return -1;
120		214
121	#define EIO_TICKS ((1000000 + 1023) >> 10)	215	#define EIO_TICKS ((1000000 + 1023) >> 10)
122
123	/*****************************************************************************/
124		216
125	#define ETP_PRI_MIN EIO_PRI_MIN	217	#define ETP_PRI_MIN EIO_PRI_MIN
126	#define ETP_PRI_MAX EIO_PRI_MAX	218	#define ETP_PRI_MAX EIO_PRI_MAX
127		219
128	struct etp_worker;	220	struct etp_worker;
…		…
144	if (wrk->dirp) \	236	if (wrk->dirp) \
145	{ \	237	{ \
146	closedir (wrk->dirp); \	238	closedir (wrk->dirp); \
147	wrk->dirp = 0; \	239	wrk->dirp = 0; \
148	}	240	}
		241
149	#define ETP_WORKER_COMMON \	242	#define ETP_WORKER_COMMON \
150	void *dbuf; \	243	void *dbuf; \
151	DIR *dirp;	244	DIR *dirp;
152		245
153	/*****************************************************************************/	246	/*****************************************************************************/
154		247
155	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	248	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
156		249
157	/* calculcate time difference in ~1/EIO_TICKS of a second */	250	/* calculate time difference in ~1/EIO_TICKS of a second */
		251	ecb_inline int
158	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	252	tvdiff (struct timeval *tv1, struct timeval *tv2)
159	{	253	{
160	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	254	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
161	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	255	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
162	}	256	}
163		257
…		…
170	static unsigned int max_poll_reqs; /* reslock */	264	static unsigned int max_poll_reqs; /* reslock */
171		265
172	static volatile unsigned int nreqs; /* reqlock */	266	static volatile unsigned int nreqs; /* reqlock */
173	static volatile unsigned int nready; /* reqlock */	267	static volatile unsigned int nready; /* reqlock */
174	static volatile unsigned int npending; /* reqlock */	268	static volatile unsigned int npending; /* reqlock */
175	static volatile unsigned int max_idle = 4;	269	static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
		270	static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
176		271
		272	static xmutex_t wrklock;
		273	static xmutex_t reslock;
		274	static xmutex_t reqlock;
		275	static xcond_t reqwait;
		276
		277	#if !HAVE_PREADWRITE
		278	/*
		279	* make our pread/pwrite emulation safe against themselves, but not against
		280	* normal read/write by using a mutex. slows down execution a lot,
		281	* but that's your problem, not mine.
		282	*/
177	static mutex_t wrklock = X_MUTEX_INIT;	283	static xmutex_t preadwritelock = X_MUTEX_INIT;
178	static mutex_t reslock = X_MUTEX_INIT;	284	#endif
179	static mutex_t reqlock = X_MUTEX_INIT;
180	static cond_t reqwait = X_COND_INIT;
181		285
182	typedef struct etp_worker	286	typedef struct etp_worker
183	{	287	{
184	/* locked by wrklock */	288	/* locked by wrklock */
185	struct etp_worker *prev, *next;	289	struct etp_worker *prev, *next;
186		290
187	thread_t tid;	291	xthread_t tid;
188		292
189	/* locked by reslock, reqlock or wrklock */	293	/* locked by reslock, reqlock or wrklock */
190	ETP_REQ *req; /* currently processed request */	294	ETP_REQ *req; /* currently processed request */
191		295
192	ETP_WORKER_COMMON	296	ETP_WORKER_COMMON
…		…
197	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)	301	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
198	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)	302	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
199		303
200	/* worker threads management */	304	/* worker threads management */
201		305
		306	static void ecb_cold
202	static void etp_worker_clear (etp_worker *wrk)	307	etp_worker_clear (etp_worker *wrk)
203	{	308	{
204	ETP_WORKER_CLEAR (wrk);	309	ETP_WORKER_CLEAR (wrk);
205	}	310	}
206		311
		312	static void ecb_cold
207	static void etp_worker_free (etp_worker *wrk)	313	etp_worker_free (etp_worker *wrk)
208	{	314	{
209	wrk->next->prev = wrk->prev;	315	wrk->next->prev = wrk->prev;
210	wrk->prev->next = wrk->next;	316	wrk->prev->next = wrk->next;
211		317
212	free (wrk);	318	free (wrk);
213	}	319	}
214		320
215	static unsigned int etp_nreqs (void)	321	static unsigned int
		322	etp_nreqs (void)
216	{	323	{
217	int retval;	324	int retval;
218	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	325	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
219	retval = nreqs;	326	retval = nreqs;
220	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	327	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
221	return retval;	328	return retval;
222	}	329	}
223		330
224	static unsigned int etp_nready (void)	331	static unsigned int
		332	etp_nready (void)
225	{	333	{
226	unsigned int retval;	334	unsigned int retval;
227		335
228	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	336	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
229	retval = nready;	337	retval = nready;
230	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	338	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
231		339
232	return retval;	340	return retval;
233	}	341	}
234		342
235	static unsigned int etp_npending (void)	343	static unsigned int
		344	etp_npending (void)
236	{	345	{
237	unsigned int retval;	346	unsigned int retval;
238		347
239	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	348	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
240	retval = npending;	349	retval = npending;
241	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	350	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
242		351
243	return retval;	352	return retval;
244	}	353	}
245		354
246	static unsigned int etp_nthreads (void)	355	static unsigned int
		356	etp_nthreads (void)
247	{	357	{
248	unsigned int retval;	358	unsigned int retval;
249		359
250	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	360	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
251	retval = started;	361	retval = started;
…		…
265	} etp_reqq;	375	} etp_reqq;
266		376
267	static etp_reqq req_queue;	377	static etp_reqq req_queue;
268	static etp_reqq res_queue;	378	static etp_reqq res_queue;
269		379
		380	static int ecb_noinline
270	static int reqq_push (etp_reqq *q, ETP_REQ *req)	381	reqq_push (etp_reqq *q, ETP_REQ *req)
271	{	382	{
272	int pri = req->pri;	383	int pri = req->pri;
273	req->next = 0;	384	req->next = 0;
274		385
275	if (q->qe[pri])	386	if (q->qe[pri])
…		…
281	q->qe[pri] = q->qs[pri] = req;	392	q->qe[pri] = q->qs[pri] = req;
282		393
283	return q->size++;	394	return q->size++;
284	}	395	}
285		396
		397	static ETP_REQ * ecb_noinline
286	static ETP_REQ *reqq_shift (etp_reqq *q)	398	reqq_shift (etp_reqq *q)
287	{	399	{
288	int pri;	400	int pri;
289		401
290	if (!q->size)	402	if (!q->size)
291	return 0;	403	return 0;
…		…
306	}	418	}
307		419
308	abort ();	420	abort ();
309	}	421	}
310		422
311	static void etp_atfork_prepare (void)	423	static void ecb_cold
		424	etp_thread_init (void)
312	{	425	{
313	X_LOCK (wrklock);
314	X_LOCK (reqlock);
315	X_LOCK (reslock);
316	#if !HAVE_PREADWRITE	426	#if !HAVE_PREADWRITE
317	X_LOCK (preadwritelock);	427	X_MUTEX_CREATE (preadwritelock);
318	#endif	428	#endif
		429	X_MUTEX_CREATE (wrklock);
		430	X_MUTEX_CREATE (reslock);
		431	X_MUTEX_CREATE (reqlock);
		432	X_COND_CREATE (reqwait);
319	}	433	}
320		434
		435	static void ecb_cold
		436	etp_atfork_prepare (void)
		437	{
		438	}
		439
		440	static void ecb_cold
321	static void etp_atfork_parent (void)	441	etp_atfork_parent (void)
322	{	442	{
323	#if !HAVE_PREADWRITE
324	X_UNLOCK (preadwritelock);
325	#endif
326	X_UNLOCK (reslock);
327	X_UNLOCK (reqlock);
328	X_UNLOCK (wrklock);
329	}	443	}
330		444
		445	static void ecb_cold
331	static void etp_atfork_child (void)	446	etp_atfork_child (void)
332	{	447	{
333	ETP_REQ *prv;	448	ETP_REQ *prv;
334		449
335	while (prv = reqq_shift (&req_queue))	450	while ((prv = reqq_shift (&req_queue)))
336	ETP_DESTROY (prv);	451	ETP_DESTROY (prv);
337		452
338	while (prv = reqq_shift (&res_queue))	453	while ((prv = reqq_shift (&res_queue)))
339	ETP_DESTROY (prv);	454	ETP_DESTROY (prv);
340		455
341	while (wrk_first.next != &wrk_first)	456	while (wrk_first.next != &wrk_first)
342	{	457	{
343	etp_worker *wrk = wrk_first.next;	458	etp_worker *wrk = wrk_first.next;
…		…
353	idle = 0;	468	idle = 0;
354	nreqs = 0;	469	nreqs = 0;
355	nready = 0;	470	nready = 0;
356	npending = 0;	471	npending = 0;
357		472
358	etp_atfork_parent ();	473	etp_thread_init ();
359	}	474	}
360		475
361	static void	476	static void ecb_cold
362	etp_once_init (void)	477	etp_once_init (void)
363	{	478	{
		479	etp_thread_init ();
364	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);	480	X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
365	}	481	}
366		482
367	static int	483	static int ecb_cold
368	etp_init (void (*want_poll)(void), void (*done_poll)(void))	484	etp_init (void (*want_poll)(void), void (*done_poll)(void))
369	{	485	{
370	static pthread_once_t doinit = PTHREAD_ONCE_INIT;	486	static pthread_once_t doinit = PTHREAD_ONCE_INIT;
371		487
372	pthread_once (&doinit, etp_once_init);	488	pthread_once (&doinit, etp_once_init);
373		489
374	want_poll_cb = want_poll;	490	want_poll_cb = want_poll;
375	done_poll_cb = done_poll;	491	done_poll_cb = done_poll;
		492
		493	return 0;
376	}	494	}
377		495
378	X_THREAD_PROC (etp_proc);	496	X_THREAD_PROC (etp_proc);
379		497
		498	static void ecb_cold
380	static void etp_start_thread (void)	499	etp_start_thread (void)
381	{	500	{
382	etp_worker *wrk = calloc (1, sizeof (etp_worker));	501	etp_worker *wrk = calloc (1, sizeof (etp_worker));
383		502
384	/*TODO*/	503	/*TODO*/
385	assert (("unable to allocate worker thread data", wrk));	504	assert (("unable to allocate worker thread data", wrk));
…		…
398	free (wrk);	517	free (wrk);
399		518
400	X_UNLOCK (wrklock);	519	X_UNLOCK (wrklock);
401	}	520	}
402		521
		522	static void
403	static void etp_maybe_start_thread (void)	523	etp_maybe_start_thread (void)
404	{	524	{
405	if (etp_nthreads () >= wanted)	525	if (ecb_expect_true (etp_nthreads () >= wanted))
406	return;	526	return;
407		527
408	/* todo: maybe use idle here, but might be less exact */	528	/* todo: maybe use idle here, but might be less exact */
409	if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())	529	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
410	return;	530	return;
411		531
412	etp_start_thread ();	532	etp_start_thread ();
413	}	533	}
414		534
		535	static void ecb_cold
415	static void etp_end_thread (void)	536	etp_end_thread (void)
416	{	537	{
417	eio_req *req = calloc (1, sizeof (eio_req));	538	eio_req *req = calloc (1, sizeof (eio_req));
418		539
419	req->type = -1;	540	req->type = -1;
420	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	541	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
…		…
427	X_LOCK (wrklock);	548	X_LOCK (wrklock);
428	--started;	549	--started;
429	X_UNLOCK (wrklock);	550	X_UNLOCK (wrklock);
430	}	551	}
431		552
432	static int etp_poll (void)	553	static int
		554	etp_poll (void)
433	{	555	{
434	unsigned int maxreqs;	556	unsigned int maxreqs;
435	unsigned int maxtime;	557	unsigned int maxtime;
436	struct timeval tv_start, tv_now;	558	struct timeval tv_start, tv_now;
437		559
…		…
467		589
468	X_LOCK (reqlock);	590	X_LOCK (reqlock);
469	--nreqs;	591	--nreqs;
470	X_UNLOCK (reqlock);	592	X_UNLOCK (reqlock);
471		593
472	if (req->type == EIO_GROUP && req->size)	594	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
473	{	595	{
474	req->int1 = 1; /* mark request as delayed */	596	req->int1 = 1; /* mark request as delayed */
475	continue;	597	continue;
476	}	598	}
477	else	599	else
478	{	600	{
479	int res = ETP_FINISH (req);	601	int res = ETP_FINISH (req);
480	if (res)	602	if (ecb_expect_false (res))
481	return res;	603	return res;
482	}	604	}
483		605
484	if (maxreqs && !--maxreqs)	606	if (ecb_expect_false (maxreqs && !--maxreqs))
485	break;	607	break;
486		608
487	if (maxtime)	609	if (maxtime)
488	{	610	{
489	gettimeofday (&tv_now, 0);	611	gettimeofday (&tv_now, 0);
…		…
495		617
496	errno = EAGAIN;	618	errno = EAGAIN;
497	return -1;	619	return -1;
498	}	620	}
499		621
		622	static void
500	static void etp_cancel (ETP_REQ *req)	623	etp_cancel (ETP_REQ *req)
501	{	624	{
502	X_LOCK (wrklock);	625	req->cancelled = 1;
503	req->flags |= EIO_FLAG_CANCELLED;
504	X_UNLOCK (wrklock);
505		626
506	eio_grp_cancel (req);	627	eio_grp_cancel (req);
507	}	628	}
508		629
		630	static void
509	static void etp_submit (ETP_REQ *req)	631	etp_submit (ETP_REQ *req)
510	{	632	{
511	req->pri -= ETP_PRI_MIN;	633	req->pri -= ETP_PRI_MIN;
512		634
513	if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;	635	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
514	if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;	636	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
515		637
		638	if (ecb_expect_false (req->type == EIO_GROUP))
		639	{
		640	/* I hope this is worth it :/ */
516	X_LOCK (reqlock);	641	X_LOCK (reqlock);
517	++nreqs;	642	++nreqs;
		643	X_UNLOCK (reqlock);
		644
		645	X_LOCK (reslock);
		646
		647	++npending;
		648
		649	if (!reqq_push (&res_queue, req) && want_poll_cb)
		650	want_poll_cb ();
		651
		652	X_UNLOCK (reslock);
		653	}
		654	else
		655	{
		656	X_LOCK (reqlock);
		657	++nreqs;
518	++nready;	658	++nready;
519	reqq_push (&req_queue, req);	659	reqq_push (&req_queue, req);
520	X_COND_SIGNAL (reqwait);	660	X_COND_SIGNAL (reqwait);
521	X_UNLOCK (reqlock);	661	X_UNLOCK (reqlock);
522		662
523	etp_maybe_start_thread ();	663	etp_maybe_start_thread ();
		664	}
524	}	665	}
525		666
		667	static void ecb_cold
526	static void etp_set_max_poll_time (double nseconds)	668	etp_set_max_poll_time (double nseconds)
527	{	669	{
528	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	670	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
529	max_poll_time = nseconds;	671	max_poll_time = nseconds * EIO_TICKS;
530	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	672	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
531	}	673	}
532		674
		675	static void ecb_cold
533	static void etp_set_max_poll_reqs (unsigned int maxreqs)	676	etp_set_max_poll_reqs (unsigned int maxreqs)
534	{	677	{
535	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	678	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
536	max_poll_reqs = maxreqs;	679	max_poll_reqs = maxreqs;
537	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	680	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
538	}	681	}
539		682
		683	static void ecb_cold
540	static void etp_set_max_idle (unsigned int nthreads)	684	etp_set_max_idle (unsigned int nthreads)
541	{	685	{
542	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	686	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
543	max_idle = nthreads <= 0 ? 1 : nthreads;	687	max_idle = nthreads;
544	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	688	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
545	}	689	}
546		690
		691	static void ecb_cold
		692	etp_set_idle_timeout (unsigned int seconds)
		693	{
		694	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		695	idle_timeout = seconds;
		696	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		697	}
		698
		699	static void ecb_cold
547	static void etp_set_min_parallel (unsigned int nthreads)	700	etp_set_min_parallel (unsigned int nthreads)
548	{	701	{
549	if (wanted < nthreads)	702	if (wanted < nthreads)
550	wanted = nthreads;	703	wanted = nthreads;
551	}	704	}
552		705
		706	static void ecb_cold
553	static void etp_set_max_parallel (unsigned int nthreads)	707	etp_set_max_parallel (unsigned int nthreads)
554	{	708	{
555	if (wanted > nthreads)	709	if (wanted > nthreads)
556	wanted = nthreads;	710	wanted = nthreads;
557		711
558	while (started > wanted)	712	while (started > wanted)
559	etp_end_thread ();	713	etp_end_thread ();
560	}	714	}
561		715
562	/*****************************************************************************/	716	/*****************************************************************************/
563		717
		718	static void
564	static void grp_try_feed (eio_req *grp)	719	grp_try_feed (eio_req *grp)
565	{	720	{
566	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	721	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
567	{	722	{
568	int old_len = grp->size;	723	grp->flags &= ~EIO_FLAG_GROUPADD;
569		724
570	EIO_FEED (grp);	725	EIO_FEED (grp);
571		726
572	/* stop if no progress has been made */	727	/* stop if no progress has been made */
573	if (old_len == grp->size)	728	if (!(grp->flags & EIO_FLAG_GROUPADD))
574	{	729	{
575	grp->feed = 0;	730	grp->feed = 0;
576	break;	731	break;
577	}	732	}
578	}	733	}
579	}	734	}
580		735
		736	static int
581	static int grp_dec (eio_req *grp)	737	grp_dec (eio_req *grp)
582	{	738	{
583	--grp->size;	739	--grp->size;
584		740
585	/* call feeder, if applicable */	741	/* call feeder, if applicable */
586	grp_try_feed (grp);	742	grp_try_feed (grp);
…		…
590	return eio_finish (grp);	746	return eio_finish (grp);
591	else	747	else
592	return 0;	748	return 0;
593	}	749	}
594		750
		751	static void
595	void eio_destroy (eio_req *req)	752	eio_destroy (eio_req *req)
596	{	753	{
597	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);	754	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
598	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);	755	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
599		756
600	EIO_DESTROY (req);	757	EIO_DESTROY (req);
601	}	758	}
602		759
		760	static int
603	static int eio_finish (eio_req *req)	761	eio_finish (eio_req *req)
604	{	762	{
605	int res = EIO_FINISH (req);	763	int res = EIO_FINISH (req);
606		764
607	if (req->grp)	765	if (req->grp)
608	{	766	{
…		…
616	if (grp->grp_first == req)	774	if (grp->grp_first == req)
617	grp->grp_first = req->grp_next;	775	grp->grp_first = req->grp_next;
618		776
619	res2 = grp_dec (grp);	777	res2 = grp_dec (grp);
620		778
621	if (!res && res2)	779	if (!res)
622	res = res2;	780	res = res2;
623	}	781	}
624		782
625	eio_destroy (req);	783	eio_destroy (req);
626		784
627	return res;	785	return res;
628	}	786	}
629		787
		788	void
630	void eio_grp_cancel (eio_req *grp)	789	eio_grp_cancel (eio_req *grp)
631	{	790	{
632	for (grp = grp->grp_first; grp; grp = grp->grp_next)	791	for (grp = grp->grp_first; grp; grp = grp->grp_next)
633	eio_cancel (grp);	792	eio_cancel (grp);
634	}	793	}
635		794
		795	void
636	void eio_cancel (eio_req *req)	796	eio_cancel (eio_req *req)
637	{	797	{
638	etp_cancel (req);	798	etp_cancel (req);
639	}	799	}
640		800
		801	void
641	void eio_submit (eio_req *req)	802	eio_submit (eio_req *req)
642	{	803	{
643	etp_submit (req);	804	etp_submit (req);
644	}	805	}
645		806
646	unsigned int eio_nreqs (void)	807	unsigned int
		808	eio_nreqs (void)
647	{	809	{
648	return etp_nreqs ();	810	return etp_nreqs ();
649	}	811	}
650		812
651	unsigned int eio_nready (void)	813	unsigned int
		814	eio_nready (void)
652	{	815	{
653	return etp_nready ();	816	return etp_nready ();
654	}	817	}
655		818
656	unsigned int eio_npending (void)	819	unsigned int
		820	eio_npending (void)
657	{	821	{
658	return etp_npending ();	822	return etp_npending ();
659	}	823	}
660		824
661	unsigned int eio_nthreads (void)	825	unsigned int ecb_cold
		826	eio_nthreads (void)
662	{	827	{
663	return etp_nthreads ();	828	return etp_nthreads ();
664	}	829	}
665		830
		831	void ecb_cold
666	void eio_set_max_poll_time (double nseconds)	832	eio_set_max_poll_time (double nseconds)
667	{	833	{
668	etp_set_max_poll_time (nseconds);	834	etp_set_max_poll_time (nseconds);
669	}	835	}
670		836
		837	void ecb_cold
671	void eio_set_max_poll_reqs (unsigned int maxreqs)	838	eio_set_max_poll_reqs (unsigned int maxreqs)
672	{	839	{
673	etp_set_max_poll_reqs (maxreqs);	840	etp_set_max_poll_reqs (maxreqs);
674	}	841	}
675		842
		843	void ecb_cold
676	void eio_set_max_idle (unsigned int nthreads)	844	eio_set_max_idle (unsigned int nthreads)
677	{	845	{
678	etp_set_max_idle (nthreads);	846	etp_set_max_idle (nthreads);
679	}	847	}
680		848
		849	void ecb_cold
		850	eio_set_idle_timeout (unsigned int seconds)
		851	{
		852	etp_set_idle_timeout (seconds);
		853	}
		854
		855	void ecb_cold
681	void eio_set_min_parallel (unsigned int nthreads)	856	eio_set_min_parallel (unsigned int nthreads)
682	{	857	{
683	etp_set_min_parallel (nthreads);	858	etp_set_min_parallel (nthreads);
684	}	859	}
685		860
		861	void ecb_cold
686	void eio_set_max_parallel (unsigned int nthreads)	862	eio_set_max_parallel (unsigned int nthreads)
687	{	863	{
688	etp_set_max_parallel (nthreads);	864	etp_set_max_parallel (nthreads);
689	}	865	}
690		866
691	int eio_poll (void)	867	int eio_poll (void)
…		…
695		871
696	/*****************************************************************************/	872	/*****************************************************************************/
697	/* work around various missing functions */	873	/* work around various missing functions */
698		874
699	#if !HAVE_PREADWRITE	875	#if !HAVE_PREADWRITE
		876	# undef pread
		877	# undef pwrite
700	# define pread eio__pread	878	# define pread eio__pread
701	# define pwrite eio__pwrite	879	# define pwrite eio__pwrite
702
703	/*
704	* make our pread/pwrite safe against themselves, but not against
705	* normal read/write by using a mutex. slows down execution a lot,
706	* but that's your problem, not mine.
707	*/
708	static mutex_t preadwritelock = X_MUTEX_INIT;
709		880
710	static ssize_t	881	static ssize_t
711	eio__pread (int fd, void *buf, size_t count, off_t offset)	882	eio__pread (int fd, void *buf, size_t count, off_t offset)
712	{	883	{
713	ssize_t res;	884	ssize_t res;
…		…
731		902
732	X_LOCK (preadwritelock);	903	X_LOCK (preadwritelock);
733	ooffset = lseek (fd, 0, SEEK_CUR);	904	ooffset = lseek (fd, 0, SEEK_CUR);
734	lseek (fd, offset, SEEK_SET);	905	lseek (fd, offset, SEEK_SET);
735	res = write (fd, buf, count);	906	res = write (fd, buf, count);
736	lseek (fd, offset, SEEK_SET);	907	lseek (fd, ooffset, SEEK_SET);
737	X_UNLOCK (preadwritelock);	908	X_UNLOCK (preadwritelock);
738		909
739	return res;	910	return res;
740	}	911	}
741	#endif	912	#endif
742		913
743	#ifndef HAVE_FUTIMES	914	#ifndef HAVE_UTIMES
744		915
		916	# undef utimes
745	# define utimes(path,times) eio__utimes (path, times)	917	# define utimes(path,times) eio__utimes (path, times)
746	# define futimes(fd,times) eio__futimes (fd, times)
747		918
748	static int	919	static int
749	eio__utimes (const char *filename, const struct timeval times[2])	920	eio__utimes (const char *filename, const struct timeval times[2])
750	{	921	{
751	if (times)	922	if (times)
…		…
759	}	930	}
760	else	931	else
761	return utime (filename, 0);	932	return utime (filename, 0);
762	}	933	}
763		934
		935	#endif
		936
		937	#ifndef HAVE_FUTIMES
		938
		939	# undef futimes
		940	# define futimes(fd,times) eio__futimes (fd, times)
		941
		942	static int
764	static int eio__futimes (int fd, const struct timeval tv[2])	943	eio__futimes (int fd, const struct timeval tv[2])
765	{	944	{
766	errno = ENOSYS;	945	errno = ENOSYS;
767	return -1;	946	return -1;
768	}	947	}
769		948
770	#endif	949	#endif
771		950
772	#if !HAVE_FDATASYNC	951	#if !HAVE_FDATASYNC
		952	# undef fdatasync
773	# define fdatasync fsync	953	# define fdatasync(fd) fsync (fd)
774	#endif	954	#endif
		955
		956	/* sync_file_range always needs emulation */
		957	static int
		958	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		959	{
		960	#if HAVE_SYNC_FILE_RANGE
		961	int res;
		962
		963	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		964	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		965	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		966	{
		967	flags = 0
		968	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		969	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		970	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		971	}
		972
		973	res = sync_file_range (fd, offset, nbytes, flags);
		974
		975	if (!res || errno != ENOSYS)
		976	return res;
		977	#endif
		978
		979	/* even though we could play tricks with the flags, it's better to always
		980	* call fdatasync, as that matches the expectation of its users best */
		981	return fdatasync (fd);
		982	}
		983
		984	static int
		985	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		986	{
		987	#if HAVE_FALLOCATE
		988	return fallocate (fd, mode, offset, len);
		989	#else
		990	errno = ENOSYS;
		991	return -1;
		992	#endif
		993	}
775		994
776	#if !HAVE_READAHEAD	995	#if !HAVE_READAHEAD
		996	# undef readahead
777	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	997	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
778		998
779	static ssize_t	999	static ssize_t
780	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	1000	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
781	{	1001	{
…		…
799		1019
800	/* sendfile always needs emulation */	1020	/* sendfile always needs emulation */
801	static ssize_t	1021	static ssize_t
802	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)	1022	eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
803	{	1023	{
		1024	ssize_t written = 0;
804	ssize_t res;	1025	ssize_t res;
805		1026
806	if (!count)	1027	if (!count)
807	return 0;	1028	return 0;
808		1029
		1030	for (;;)
		1031	{
		1032	#ifdef __APPLE__
		1033	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1034	#endif
809	#if HAVE_SENDFILE	1035	#if HAVE_SENDFILE
810	# if __linux	1036	# if __linux
		1037	off_t soffset = offset;
811	res = sendfile (ofd, ifd, &offset, count);	1038	res = sendfile (ofd, ifd, &soffset, count);
812		1039
813	# elif __freebsd	1040	# elif __FreeBSD__
814	/*	1041	/*
815	* Of course, the freebsd sendfile is a dire hack with no thoughts	1042	* Of course, the freebsd sendfile is a dire hack with no thoughts
816	* wasted on making it similar to other I/O functions.	1043	* wasted on making it similar to other I/O functions.
817	*/	1044	*/
818	{
819	off_t sbytes;	1045	off_t sbytes;
820	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1046	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
821		1047
822	if (res < 0 && sbytes)	1048	#if 0 /* according to the manpage, this is correct, but broken behaviour */
823	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1049	/* freebsd' sendfile will return 0 on success */
		1050	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1051	/* not on e.g. EIO or EPIPE - sounds broken */
		1052	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
824	res = sbytes;	1053	res = sbytes;
825	}	1054	#endif
		1055
		1056	/* according to source inspection, this is correct, and useful behaviour */
		1057	if (sbytes)
		1058	res = sbytes;
		1059
		1060	# elif defined (__APPLE__)
		1061	off_t sbytes = count;
		1062	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1063
		1064	/* according to the manpage, sbytes is always valid */
		1065	if (sbytes)
		1066	res = sbytes;
826		1067
827	# elif __hpux	1068	# elif __hpux
828	res = sendfile (ofd, ifd, offset, count, 0, 0);	1069	res = sendfile (ofd, ifd, offset, count, 0, 0);
829		1070
830	# elif __solaris	1071	# elif __solaris
831	{
832	struct sendfilevec vec;	1072	struct sendfilevec vec;
833	size_t sbytes;	1073	size_t sbytes;
834		1074
835	vec.sfv_fd = ifd;	1075	vec.sfv_fd = ifd;
836	vec.sfv_flag = 0;	1076	vec.sfv_flag = 0;
837	vec.sfv_off = offset;	1077	vec.sfv_off = offset;
838	vec.sfv_len = count;	1078	vec.sfv_len = count;
839		1079
840	res = sendfilev (ofd, &vec, 1, &sbytes);	1080	res = sendfilev (ofd, &vec, 1, &sbytes);
841		1081
842	if (res < 0 && sbytes)	1082	if (res < 0 && sbytes)
843	res = sbytes;	1083	res = sbytes;
844	}
845		1084
846	# endif	1085	# endif
		1086
		1087	#elif defined (_WIN32) && 0
		1088	/* does not work, just for documentation of what would need to be done */
		1089	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1090	/* libeio guarantees that the file offset does not change, and windows */
		1091	/* has no way to get an independent handle to the same file description */
		1092	HANDLE h = TO_SOCKET (ifd);
		1093	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1094	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1095
847	#else	1096	#else
848	res = -1;	1097	res = -1;
849	errno = ENOSYS;	1098	errno = ENOSYS;
850	#endif	1099	#endif
851		1100
		1101	/* we assume sendfile can copy at least 128mb in one go */
		1102	if (res <= 128 * 1024 * 1024)
		1103	{
		1104	if (res > 0)
		1105	written += res;
		1106
		1107	if (written)
		1108	return written;
		1109
		1110	break;
		1111	}
		1112	else
		1113	{
		1114	/* if we requested more, then probably the kernel was lazy */
		1115	written += res;
		1116	offset += res;
		1117	count -= res;
		1118
		1119	if (!count)
		1120	return written;
		1121	}
		1122	}
		1123
852	if (res < 0	1124	if (res < 0
853	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1125	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1126	/* BSDs */
		1127	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1128	|| errno == ENOTSUP
		1129	#endif
		1130	|| errno == EOPNOTSUPP /* BSDs */
854	#if __solaris	1131	#if __solaris
855	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1132	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
856	#endif	1133	#endif
857	)	1134	)
858	)	1135	)
…		…
889	}	1166	}
890		1167
891	return res;	1168	return res;
892	}	1169	}
893		1170
		1171	#ifdef PAGESIZE
		1172	# define eio_pagesize() PAGESIZE
		1173	#else
		1174	static intptr_t
		1175	eio_pagesize (void)
		1176	{
		1177	static intptr_t page;
		1178
		1179	if (!page)
		1180	page = sysconf (_SC_PAGESIZE);
		1181
		1182	return page;
		1183	}
		1184	#endif
		1185
		1186	static void
		1187	eio_page_align (void **addr, size_t *length)
		1188	{
		1189	intptr_t mask = eio_pagesize () - 1;
		1190
		1191	/* round down addr */
		1192	intptr_t adj = mask & (intptr_t)*addr;
		1193
		1194	*addr = (void *)((intptr_t)*addr - adj);
		1195	*length += adj;
		1196
		1197	/* round up length */
		1198	*length = (*length + mask) & ~mask;
		1199	}
		1200
		1201	#if !_POSIX_MEMLOCK
		1202	# define eio__mlockall(a) eio_nosyscall()
		1203	#else
		1204
		1205	static int
		1206	eio__mlockall (int flags)
		1207	{
		1208	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1209	extern int mallopt (int, int);
		1210	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1211	#endif
		1212
		1213	if (EIO_MCL_CURRENT != MCL_CURRENT
		1214	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1215	{
		1216	flags = 0
		1217	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1218	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1219	}
		1220
		1221	return mlockall (flags);
		1222	}
		1223	#endif
		1224
		1225	#if !_POSIX_MEMLOCK_RANGE
		1226	# define eio__mlock(a,b) EIO_ENOSYS ()
		1227	#else
		1228
		1229	static int
		1230	eio__mlock (void *addr, size_t length)
		1231	{
		1232	eio_page_align (&addr, &length);
		1233
		1234	return mlock (addr, length);
		1235	}
		1236
		1237	#endif
		1238
		1239	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1240	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1241	#else
		1242
		1243	static int
		1244	eio__msync (void *mem, size_t len, int flags)
		1245	{
		1246	eio_page_align (&mem, &len);
		1247
		1248	if (EIO_MS_ASYNC != MS_SYNC
		1249	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1250	|| EIO_MS_SYNC != MS_SYNC)
		1251	{
		1252	flags = 0
		1253	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1254	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1255	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1256	}
		1257
		1258	return msync (mem, len, flags);
		1259	}
		1260
		1261	#endif
		1262
		1263	static int
		1264	eio__mtouch (eio_req *req)
		1265	{
		1266	void *mem = req->ptr2;
		1267	size_t len = req->size;
		1268	int flags = req->int1;
		1269
		1270	eio_page_align (&mem, &len);
		1271
		1272	{
		1273	intptr_t addr = (intptr_t)mem;
		1274	intptr_t end = addr + len;
		1275	intptr_t page = eio_pagesize ();
		1276
		1277	if (addr < end)
		1278	if (flags & EIO_MT_MODIFY) /* modify */
		1279	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1280	else
		1281	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1282	}
		1283
		1284	return 0;
		1285	}
		1286
		1287	/*****************************************************************************/
		1288	/* requests implemented outside eio_execute, because they are so large */
		1289
		1290	static void
		1291	eio__realpath (eio_req *req, etp_worker *self)
		1292	{
		1293	char *rel = req->ptr1;
		1294	char *res;
		1295	char *tmp1, *tmp2;
		1296	#if SYMLOOP_MAX > 32
		1297	int symlinks = SYMLOOP_MAX;
		1298	#else
		1299	int symlinks = 32;
		1300	#endif
		1301
		1302	req->result = -1;
		1303
		1304	errno = EINVAL;
		1305	if (!rel)
		1306	return;
		1307
		1308	errno = ENOENT;
		1309	if (!*rel)
		1310	return;
		1311
		1312	if (!req->ptr2)
		1313	{
		1314	X_LOCK (wrklock);
		1315	req->flags |= EIO_FLAG_PTR2_FREE;
		1316	X_UNLOCK (wrklock);
		1317	req->ptr2 = malloc (PATH_MAX * 3);
		1318
		1319	errno = ENOMEM;
		1320	if (!req->ptr2)
		1321	return;
		1322	}
		1323
		1324	res = req->ptr2;
		1325	tmp1 = res + PATH_MAX;
		1326	tmp2 = tmp1 + PATH_MAX;
		1327
		1328	#if 0 /* disabled, the musl way to do things is just too racy */
		1329	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1330	/* on linux we may be able to ask the kernel */
		1331	{
		1332	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1333
		1334	if (fd >= 0)
		1335	{
		1336	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1337	req->result = readlink (tmp1, res, PATH_MAX);
		1338	close (fd);
		1339
		1340	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1341
		1342	if (req->result > 0)
		1343	goto done;
		1344	}
		1345	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1346	return;
		1347	}
		1348	#endif
		1349	#endif
		1350
		1351	if (*rel != '/')
		1352	{
		1353	if (!getcwd (res, PATH_MAX))
		1354	return;
		1355
		1356	if (res [1]) /* only use if not / */
		1357	res += strlen (res);
		1358	}
		1359
		1360	while (*rel)
		1361	{
		1362	ssize_t len, linklen;
		1363	char *beg = rel;
		1364
		1365	while (*rel && *rel != '/')
		1366	++rel;
		1367
		1368	len = rel - beg;
		1369
		1370	if (!len) /* skip slashes */
		1371	{
		1372	++rel;
		1373	continue;
		1374	}
		1375
		1376	if (beg [0] == '.')
		1377	{
		1378	if (len == 1)
		1379	continue; /* . - nop */
		1380
		1381	if (beg [1] == '.' && len == 2)
		1382	{
		1383	/* .. - back up one component, if possible */
		1384
		1385	while (res != req->ptr2)
		1386	if (*--res == '/')
		1387	break;
		1388
		1389	continue;
		1390	}
		1391	}
		1392
		1393	errno = ENAMETOOLONG;
		1394	if (res + 1 + len + 1 >= tmp1)
		1395	return;
		1396
		1397	/* copy one component */
		1398	*res = '/';
		1399	memcpy (res + 1, beg, len);
		1400
		1401	/* zero-terminate, for readlink */
		1402	res [len + 1] = 0;
		1403
		1404	/* now check if it's a symlink */
		1405	linklen = readlink (req->ptr2, tmp1, PATH_MAX);
		1406
		1407	if (linklen < 0)
		1408	{
		1409	if (errno != EINVAL)
		1410	return;
		1411
		1412	/* it's a normal directory. hopefully */
		1413	res += len + 1;
		1414	}
		1415	else
		1416	{
		1417	/* yay, it was a symlink - build new path in tmp2 */
		1418	int rellen = strlen (rel);
		1419
		1420	errno = ENAMETOOLONG;
		1421	if (linklen + 1 + rellen >= PATH_MAX)
		1422	return;
		1423
		1424	errno = ELOOP;
		1425	if (!--symlinks)
		1426	return;
		1427
		1428	if (*tmp1 == '/')
		1429	res = req->ptr2; /* symlink resolves to an absolute path */
		1430
		1431	/* we need to be careful, as rel might point into tmp2 already */
		1432	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1433	tmp2 [linklen] = '/';
		1434	memcpy (tmp2, tmp1, linklen);
		1435
		1436	rel = tmp2;
		1437	}
		1438	}
		1439
		1440	/* special case for the lone root path */
		1441	if (res == req->ptr2)
		1442	*res++ = '/';
		1443
		1444	req->result = res - (char *)req->ptr2;
		1445
		1446	done:
		1447	req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
		1448	}
		1449
		1450	static signed char
		1451	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1452	{
		1453	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1454	: a->inode < b->inode ? -1
		1455	: a->inode > b->inode ? 1
		1456	: 0;
		1457	}
		1458
		1459	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1460
		1461	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1462	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1463
		1464	static void
		1465	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1466	{
		1467	unsigned char bits [9 + sizeof (ino_t) * 8];
		1468	unsigned char *bit = bits;
		1469
		1470	assert (CHAR_BIT == 8);
		1471	assert (sizeof (eio_dirent) * 8 < 256);
		1472	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1473	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1474
		1475	if (size <= EIO_SORT_FAST)
		1476	return;
		1477
		1478	/* first prepare an array of bits to test in our radix sort */
		1479	/* try to take endianness into account, as well as differences in ino_t sizes */
		1480	/* inode_bits must contain all inodes ORed together */
		1481	/* which is used to skip bits that are 0 everywhere, which is very common */
		1482	{
		1483	ino_t endianness;
		1484	int i, j;
		1485
		1486	/* we store the byte offset of byte n into byte n of "endianness" */
		1487	for (i = 0; i < sizeof (ino_t); ++i)
		1488	((unsigned char *)&endianness)[i] = i;
		1489
		1490	*bit++ = 0;
		1491
		1492	for (i = 0; i < sizeof (ino_t); ++i)
		1493	{
		1494	/* shifting off the byte offsets out of "endianness" */
		1495	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1496	endianness >>= 8;
		1497
		1498	for (j = 0; j < 8; ++j)
		1499	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1500	*bit++ = offs + j;
		1501	}
		1502
		1503	for (j = 0; j < 8; ++j)
		1504	if (score_bits & (1 << j))
		1505	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1506	}
		1507
		1508	/* now actually do the sorting (a variant of MSD radix sort) */
		1509	{
		1510	eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
		1511	eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
		1512	unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
		1513	int stk_idx = 0;
		1514
		1515	base_stk [stk_idx] = dents;
		1516	end_stk [stk_idx] = dents + size;
		1517	bit_stk [stk_idx] = bit - 1;
		1518
		1519	do
		1520	{
		1521	base = base_stk [stk_idx];
		1522	end = end_stk [stk_idx];
		1523	bit = bit_stk [stk_idx];
		1524
		1525	for (;;)
		1526	{
		1527	unsigned char O = *bit >> 3;
		1528	unsigned char M = 1 << (*bit & 7);
		1529
		1530	eio_dirent *a = base;
		1531	eio_dirent *b = end;
		1532
		1533	if (b - a < EIO_SORT_CUTOFF)
		1534	break;
		1535
		1536	/* now bit-partition the array on the bit */
		1537	/* this ugly asymmetric loop seems to perform much better than typical */
		1538	/* partition algos found in the literature */
		1539	do
		1540	if (!(((unsigned char *)a)[O] & M))
		1541	++a;
		1542	else if (!(((unsigned char *)--b)[O] & M))
		1543	{
		1544	eio_dirent tmp = *a; *a = *b; *b = tmp;
		1545	++a;
		1546	}
		1547	while (b > a);
		1548
		1549	/* next bit, or stop, if no bits left in this path */
		1550	if (!*--bit)
		1551	break;
		1552
		1553	base_stk [stk_idx] = a;
		1554	end_stk [stk_idx] = end;
		1555	bit_stk [stk_idx] = bit;
		1556	++stk_idx;
		1557
		1558	end = a;
		1559	}
		1560	}
		1561	while (stk_idx--);
		1562	}
		1563	}
		1564
		1565	static void
		1566	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1567	{
		1568	/* first move the smallest element to the front, to act as a sentinel */
		1569	{
		1570	int i;
		1571	eio_dirent *min = dents;
		1572
		1573	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1574	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1575	if (EIO_DENT_CMP (dents [i], <, *min))
		1576	min = &dents [i];
		1577
		1578	/* swap elements 0 and j (minimum) */
		1579	{
		1580	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1581	}
		1582	}
		1583
		1584	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1585	{
		1586	eio_dirent *i, *j;
		1587
		1588	for (i = dents + 1; i < dents + size; ++i)
		1589	{
		1590	eio_dirent value = *i;
		1591
		1592	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1593	j [1] = j [0];
		1594
		1595	j [1] = value;
		1596	}
		1597	}
		1598	}
		1599
		1600	static void
		1601	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1602	{
		1603	if (size <= 1)
		1604	return; /* our insertion sort relies on size > 0 */
		1605
		1606	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1607	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1608	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1609
		1610	/* use an insertion sort at the end, or for small arrays, */
		1611	/* as insertion sort is more efficient for small partitions */
		1612	eio_dent_insertion_sort (dents, size);
		1613	}
		1614
894	/* read a full directory */	1615	/* read a full directory */
895	static void	1616	static void
896	eio__scandir (eio_req *req, etp_worker *self)	1617	eio__scandir (eio_req *req, etp_worker *self)
897	{	1618	{
898	DIR *dirp;	1619	DIR *dirp;
899	EIO_STRUCT_DIRENT *entp;	1620	EIO_STRUCT_DIRENT *entp;
900	char *name, *names;	1621	char *name, *names;
901	int memlen = 4096;	1622	int namesalloc = 4096;
902	int memofs = 0;	1623	int namesoffs = 0;
		1624	int flags = req->int1;
		1625	eio_dirent *dents = 0;
		1626	int dentalloc = 128;
903	int res = 0;	1627	int dentoffs = 0;
		1628	ino_t inode_bits = 0;
		1629
		1630	req->result = -1;
		1631
		1632	if (!(flags & EIO_READDIR_DENTS))
		1633	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
904		1634
905	X_LOCK (wrklock);	1635	X_LOCK (wrklock);
		1636	/* the corresponding closedir is in ETP_WORKER_CLEAR */
906	self->dirp = dirp = opendir (req->ptr1);	1637	self->dirp = dirp = opendir (req->ptr1);
		1638
907	req->flags |= EIO_FLAG_PTR2_FREE;	1639	if (req->flags & EIO_FLAG_PTR1_FREE)
		1640	free (req->ptr1);
		1641
		1642	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1643	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
908	req->ptr2 = names = malloc (memlen);	1644	req->ptr2 = names = malloc (namesalloc);
909	X_UNLOCK (wrklock);	1645	X_UNLOCK (wrklock);
910		1646
911	if (dirp && names)	1647	if (dirp && names && (!flags || dents))
912	for (;;)	1648	for (;;)
913	{	1649	{
914	errno = 0;	1650	errno = 0;
915	entp = readdir (dirp);	1651	entp = readdir (dirp);
916		1652
917	if (!entp)	1653	if (!entp)
		1654	{
		1655	if (errno)
		1656	break;
		1657
		1658	/* sort etc. */
		1659	req->int1 = flags;
		1660	req->result = dentoffs;
		1661
		1662	if (flags & EIO_READDIR_STAT_ORDER)
		1663	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1664	else if (flags & EIO_READDIR_DIRS_FIRST)
		1665	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1666	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1667	else
		1668	{
		1669	/* in this case, all is known, and we just put dirs first and sort them */
		1670	eio_dirent *oth = dents + dentoffs;
		1671	eio_dirent *dir = dents;
		1672
		1673	/* now partition dirs to the front, and non-dirs to the back */
		1674	/* by walking from both sides and swapping if necessary */
		1675	while (oth > dir)
		1676	{
		1677	if (dir->type == EIO_DT_DIR)
		1678	++dir;
		1679	else if ((--oth)->type == EIO_DT_DIR)
		1680	{
		1681	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1682
		1683	++dir;
		1684	}
		1685	}
		1686
		1687	/* now sort the dirs only (dirs all have the same score) */
		1688	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1689	}
		1690
918	break;	1691	break;
		1692	}
919		1693
		1694	/* now add the entry to our list(s) */
920	name = entp->d_name;	1695	name = entp->d_name;
921		1696
		1697	/* skip . and .. entries */
922	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1698	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
923	{	1699	{
924	int len = strlen (name) + 1;	1700	int len = D_NAMLEN (entp) + 1;
925		1701
926	res++;	1702	while (ecb_expect_false (namesoffs + len > namesalloc))
927
928	while (memofs + len > memlen)
929	{	1703	{
930	memlen *= 2;	1704	namesalloc *= 2;
931	X_LOCK (wrklock);	1705	X_LOCK (wrklock);
932	req->ptr2 = names = realloc (names, memlen);	1706	req->ptr2 = names = realloc (names, namesalloc);
933	X_UNLOCK (wrklock);	1707	X_UNLOCK (wrklock);
934		1708
935	if (!names)	1709	if (!names)
936	break;	1710	break;
937	}	1711	}
938		1712
939	memcpy (names + memofs, name, len);	1713	memcpy (names + namesoffs, name, len);
		1714
		1715	if (dents)
		1716	{
		1717	struct eio_dirent *ent;
		1718
		1719	if (ecb_expect_false (dentoffs == dentalloc))
		1720	{
		1721	dentalloc *= 2;
		1722	X_LOCK (wrklock);
		1723	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1724	X_UNLOCK (wrklock);
		1725
		1726	if (!dents)
		1727	break;
		1728	}
		1729
		1730	ent = dents + dentoffs;
		1731
		1732	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1733	ent->namelen = len - 1;
		1734	ent->inode = D_INO (entp);
		1735
		1736	inode_bits |= ent->inode;
		1737
		1738	switch (D_TYPE (entp))
		1739	{
		1740	default:
		1741	ent->type = EIO_DT_UNKNOWN;
		1742	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1743	break;
		1744
		1745	#ifdef DT_FIFO
		1746	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1747	#endif
		1748	#ifdef DT_CHR
		1749	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1750	#endif
		1751	#ifdef DT_MPC
		1752	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1753	#endif
		1754	#ifdef DT_DIR
		1755	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1756	#endif
		1757	#ifdef DT_NAM
		1758	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1759	#endif
		1760	#ifdef DT_BLK
		1761	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1762	#endif
		1763	#ifdef DT_MPB
		1764	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1765	#endif
		1766	#ifdef DT_REG
		1767	case DT_REG: ent->type = EIO_DT_REG; break;
		1768	#endif
		1769	#ifdef DT_NWK
		1770	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1771	#endif
		1772	#ifdef DT_CMP
		1773	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1774	#endif
		1775	#ifdef DT_LNK
		1776	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1777	#endif
		1778	#ifdef DT_SOCK
		1779	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1780	#endif
		1781	#ifdef DT_DOOR
		1782	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1783	#endif
		1784	#ifdef DT_WHT
		1785	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1786	#endif
		1787	}
		1788
		1789	ent->score = 7;
		1790
		1791	if (flags & EIO_READDIR_DIRS_FIRST)
		1792	{
		1793	if (ent->type == EIO_DT_UNKNOWN)
		1794	{
		1795	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1796	ent->score = 1;
		1797	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1798	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1799	}
		1800	else if (ent->type == EIO_DT_DIR)
		1801	ent->score = 0;
		1802	}
		1803	}
		1804
940	memofs += len;	1805	namesoffs += len;
		1806	++dentoffs;
		1807	}
		1808
		1809	if (EIO_CANCELLED (req))
		1810	{
		1811	errno = ECANCELED;
		1812	break;
941	}	1813	}
942	}	1814	}
943
944	if (errno)
945	res = -1;
946
947	req->result = res;
948	}	1815	}
949		1816
950	/*****************************************************************************/	1817	/*****************************************************************************/
951		1818
952	#define ALLOC(len) \	1819	#define ALLOC(len) \
…		…
967	X_THREAD_PROC (etp_proc)	1834	X_THREAD_PROC (etp_proc)
968	{	1835	{
969	ETP_REQ *req;	1836	ETP_REQ *req;
970	struct timespec ts;	1837	struct timespec ts;
971	etp_worker *self = (etp_worker *)thr_arg;	1838	etp_worker *self = (etp_worker *)thr_arg;
		1839	int timeout;
972		1840
973	/* try to distribute timeouts somewhat randomly */	1841	/* try to distribute timeouts somewhat evenly */
974	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	1842	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
975		1843
976	for (;;)	1844	for (;;)
977	{	1845	{
		1846	ts.tv_sec = 0;
		1847
978	X_LOCK (reqlock);	1848	X_LOCK (reqlock);
979		1849
980	for (;;)	1850	for (;;)
981	{	1851	{
982	self->req = req = reqq_shift (&req_queue);	1852	self->req = req = reqq_shift (&req_queue);
983		1853
984	if (req)	1854	if (req)
985	break;	1855	break;
986		1856
		1857	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		1858	{
		1859	X_UNLOCK (reqlock);
		1860	X_LOCK (wrklock);
		1861	--started;
		1862	X_UNLOCK (wrklock);
		1863	goto quit;
		1864	}
		1865
987	++idle;	1866	++idle;
988		1867
989	ts.tv_sec = time (0) + IDLE_TIMEOUT;	1868	if (idle <= max_idle)
990	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	1869	/* we are allowed to idle, so do so without any timeout */
		1870	X_COND_WAIT (reqwait, reqlock);
		1871	else
991	{	1872	{
992	if (idle > max_idle)	1873	/* initialise timeout once */
993	{	1874	if (!ts.tv_sec)
994	--idle;	1875	ts.tv_sec = time (0) + idle_timeout;
995	X_UNLOCK (reqlock);
996	X_LOCK (wrklock);
997	--started;
998	X_UNLOCK (wrklock);
999	goto quit;
1000	}
1001		1876
1002	/* we are allowed to idle, so do so without any timeout */
1003	X_COND_WAIT (reqwait, reqlock);	1877	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
		1878	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1004	}	1879	}
1005		1880
1006	--idle;	1881	--idle;
1007	}	1882	}
1008		1883
…		…
1011	X_UNLOCK (reqlock);	1886	X_UNLOCK (reqlock);
1012		1887
1013	if (req->type < 0)	1888	if (req->type < 0)
1014	goto quit;	1889	goto quit;
1015		1890
1016	if (!EIO_CANCELLED (req))
1017	ETP_EXECUTE (self, req);	1891	ETP_EXECUTE (self, req);
1018		1892
1019	X_LOCK (reslock);	1893	X_LOCK (reslock);
1020		1894
1021	++npending;	1895	++npending;
1022		1896
…		…
1037	return 0;	1911	return 0;
1038	}	1912	}
1039		1913
1040	/*****************************************************************************/	1914	/*****************************************************************************/
1041		1915
		1916	int ecb_cold
1042	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	1917	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1043	{	1918	{
1044	etp_init (want_poll, done_poll);	1919	return etp_init (want_poll, done_poll);
1045	}	1920	}
1046		1921
		1922	ecb_inline void
1047	static void eio_api_destroy (eio_req *req)	1923	eio_api_destroy (eio_req *req)
1048	{	1924	{
1049	free (req);	1925	free (req);
1050	}	1926	}
1051		1927
1052	#define REQ(rtype) \	1928	#define REQ(rtype) \
…		…
1071	{ \	1947	{ \
1072	eio_api_destroy (req); \	1948	eio_api_destroy (req); \
1073	return 0; \	1949	return 0; \
1074	}	1950	}
1075		1951
		1952	static void
1076	static void eio_execute (etp_worker *self, eio_req *req)	1953	eio_execute (etp_worker *self, eio_req *req)
1077	{	1954	{
1078	errno = 0;	1955	if (ecb_expect_false (EIO_CANCELLED (req)))
		1956	{
		1957	req->result = -1;
		1958	req->errorno = ECANCELED;
		1959	return;
		1960	}
1079		1961
1080	switch (req->type)	1962	switch (req->type)
1081	{	1963	{
1082	case EIO_READ: ALLOC (req->size);	1964	case EIO_READ: ALLOC (req->size);
1083	req->result = req->offs >= 0	1965	req->result = req->offs >= 0
…		…
1095	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1977	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1096	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1978	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1097	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1979	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1098	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1980	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1099		1981
		1982	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1983	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1984	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1985	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1986
1100	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	1987	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1101	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	1988	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1102	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	1989	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1103	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	1990	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1104	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;	1991	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
…		…
1111	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;	1998	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1112	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;	1999	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1113	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;	2000	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1114	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;	2001	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1115	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;	2002	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1116	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;	2003	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1117		2004
		2005	case EIO_REALPATH: eio__realpath (req, self); break;
		2006
1118	case EIO_READLINK: ALLOC (NAME_MAX);	2007	case EIO_READLINK: ALLOC (PATH_MAX);
1119	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	2008	req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1120		2009
1121	case EIO_SYNC: req->result = 0; sync (); break;	2010	case EIO_SYNC: req->result = 0; sync (); break;
1122	case EIO_FSYNC: req->result = fsync (req->int1); break;	2011	case EIO_FSYNC: req->result = fsync (req->int1); break;
1123	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	2012	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2013	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2014	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		2015	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2016	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		2017	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2018	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1124		2019
1125	case EIO_READDIR: eio__scandir (req, self); break;	2020	case EIO_READDIR: eio__scandir (req, self); break;
1126		2021
1127	case EIO_BUSY:	2022	case EIO_BUSY:
1128	#ifdef _WIN32	2023	#ifdef _WIN32
1129	Sleep (req->nv1 * 1000.);	2024	Sleep (req->nv1 * 1e3);
1130	#else	2025	#else
1131	{	2026	{
1132	struct timeval tv;	2027	struct timeval tv;
1133		2028
1134	tv.tv_sec = req->nv1;	2029	tv.tv_sec = req->nv1;
1135	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;	2030	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1136		2031
1137	req->result = select (0, 0, 0, 0, &tv);	2032	req->result = select (0, 0, 0, 0, &tv);
1138	}	2033	}
1139	#endif	2034	#endif
1140	break;	2035	break;
…		…
1155	times = tv;	2050	times = tv;
1156	}	2051	}
1157	else	2052	else
1158	times = 0;	2053	times = 0;
1159		2054
1160
1161	req->result = req->type == EIO_FUTIME	2055	req->result = req->type == EIO_FUTIME
1162	? futimes (req->int1, times)	2056	? futimes (req->int1, times)
1163	: utimes (req->ptr1, times);	2057	: utimes (req->ptr1, times);
1164	}	2058	}
		2059	break;
1165		2060
1166	case EIO_GROUP:	2061	case EIO_GROUP:
		2062	abort (); /* handled in eio_request */
		2063
1167	case EIO_NOP:	2064	case EIO_NOP:
1168	req->result = 0;	2065	req->result = 0;
1169	break;	2066	break;
1170		2067
1171	case EIO_CUSTOM:	2068	case EIO_CUSTOM:
1172	req->feed (req);	2069	req->feed (req);
1173	break;	2070	break;
1174		2071
1175	default:	2072	default:
		2073	errno = ENOSYS;
1176	req->result = -1;	2074	req->result = -1;
1177	break;	2075	break;
1178	}	2076	}
1179		2077
1180	req->errorno = errno;	2078	req->errorno = errno;
…		…
1200	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)	2098	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1201	{	2099	{
1202	REQ (EIO_FSYNC); req->int1 = fd; SEND;	2100	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1203	}	2101	}
1204		2102
		2103	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2104	{
		2105	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2106	}
		2107
		2108	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2109	{
		2110	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2111	}
		2112
		2113	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2114	{
		2115	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2116	}
		2117
		2118	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2119	{
		2120	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2121	}
		2122
		2123	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2124	{
		2125	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2126	}
		2127
		2128	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2129	{
		2130	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2131	}
		2132
1205	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)	2133	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1206	{	2134	{
1207	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	2135	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1208	}	2136	}
1209		2137
…		…
1228	}	2156	}
1229		2157
1230	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)	2158	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1231	{	2159	{
1232	REQ (EIO_FSTAT); req->int1 = fd; SEND;	2160	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		2161	}
		2162
		2163	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2164	{
		2165	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1233	}	2166	}
1234		2167
1235	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)	2168	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1236	{	2169	{
1237	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2170	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
…		…
1301	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)	2234	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1302	{	2235	{
1303	return eio__1path (EIO_READLINK, path, pri, cb, data);	2236	return eio__1path (EIO_READLINK, path, pri, cb, data);
1304	}	2237	}
1305		2238
		2239	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2240	{
		2241	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2242	}
		2243
1306	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)	2244	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1307	{	2245	{
1308	return eio__1path (EIO_STAT, path, pri, cb, data);	2246	return eio__1path (EIO_STAT, path, pri, cb, data);
1309	}	2247	}
1310		2248
1311	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)	2249	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1312	{	2250	{
1313	return eio__1path (EIO_LSTAT, path, pri, cb, data);	2251	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1314	}	2252	}
1315		2253
		2254	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2255	{
		2256	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2257	}
		2258
1316	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)	2259	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1317	{	2260	{
1318	return eio__1path (EIO_UNLINK, path, pri, cb, data);	2261	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1319	}	2262	}
1320		2263
1321	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)	2264	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1322	{	2265	{
1323	return eio__1path (EIO_RMDIR, path, pri, cb, data);	2266	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1324	}	2267	}
1325		2268
1326	eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data)	2269	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1327	{	2270	{
1328	return eio__1path (EIO_READDIR, path, pri, cb, data);	2271	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1329	}	2272	}
1330		2273
1331	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	2274	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1332	{	2275	{
1333	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	2276	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1334	}	2277	}
1335		2278
1336	static eio_req *	2279	static eio_req *
1337	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2280	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1338	{	2281	{
…		…
1362	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)	2305	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1363	{	2306	{
1364	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);	2307	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1365	}	2308	}
1366		2309
1367	eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)	2310	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1368	{	2311	{
1369	REQ (EIO_CUSTOM); req->feed = execute; SEND;	2312	REQ (EIO_CUSTOM); req->feed = execute; SEND;
1370	}	2313	}
1371		2314
1372	#endif	2315	#endif
…		…
1383	#undef SEND	2326	#undef SEND
1384		2327
1385	/*****************************************************************************/	2328	/*****************************************************************************/
1386	/* grp functions */	2329	/* grp functions */
1387		2330
		2331	void
1388	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2332	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1389	{	2333	{
1390	grp->int2 = limit;	2334	grp->int2 = limit;
1391	grp->feed = feed;	2335	grp->feed = feed;
1392		2336
1393	grp_try_feed (grp);	2337	grp_try_feed (grp);
1394	}	2338	}
1395		2339
		2340	void
1396	void eio_grp_limit (eio_req *grp, int limit)	2341	eio_grp_limit (eio_req *grp, int limit)
1397	{	2342	{
1398	grp->int2 = limit;	2343	grp->int2 = limit;
1399		2344
1400	grp_try_feed (grp);	2345	grp_try_feed (grp);
1401	}	2346	}
1402		2347
		2348	void
1403	void eio_grp_add (eio_req *grp, eio_req *req)	2349	eio_grp_add (eio_req *grp, eio_req *req)
1404	{	2350	{
1405	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2351	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
		2352
		2353	grp->flags |= EIO_FLAG_GROUPADD;
1406		2354
1407	++grp->size;	2355	++grp->size;
1408	req->grp = grp;	2356	req->grp = grp;
1409		2357
1410	req->grp_prev = 0;	2358	req->grp_prev = 0;
…		…
1417	}	2365	}
1418		2366
1419	/*****************************************************************************/	2367	/*****************************************************************************/
1420	/* misc garbage */	2368	/* misc garbage */
1421		2369
		2370	ssize_t
1422	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	2371	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1423	{	2372	{
1424	etp_worker wrk;	2373	etp_worker wrk;
		2374	ssize_t ret;
1425		2375
1426	wrk.dbuf = 0;	2376	wrk.dbuf = 0;
1427		2377
1428	eio__sendfile (ofd, ifd, offset, count, &wrk);	2378	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1429		2379
1430	if (wrk.dbuf)	2380	if (wrk.dbuf)
1431	free (wrk.dbuf);	2381	free (wrk.dbuf);
1432	}
1433		2382
		2383	return ret;
		2384	}
		2385

Diff Legend

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