ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libeio/eio.c
(Generate patch)

Comparing libeio/eio.c (file contents):
Revision 1.30 by root, Wed Jun 3 12:24:49 2009 UTC vs.
Revision 1.98 by root, Sun Jul 24 05:53:34 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"
41 46
42#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.h"
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <limits.h> 59#include <limits.h>
55#include <fcntl.h> 60#include <fcntl.h>
56#include <assert.h> 61#include <assert.h>
57 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
58#ifndef EIO_FINISH 82#ifndef EIO_FINISH
59# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60#endif 84#endif
61 85
62#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
65 89
66#ifndef EIO_FEED 90#ifndef EIO_FEED
67# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68#endif 92#endif
69 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
70#ifdef _WIN32 105#ifdef _WIN32
71 106
72 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 #ifdef EIO_STRUCT_STATI64
111 #define stat(path,buf) _stati64 (path,buf)
112 #define fstat(fd,buf) _fstati64 (fd,buf)
113 #endif
114 #define lstat(path,buf) stat (path,buf)
115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
116 #define mkdir(path,mode) _mkdir (path)
117 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
118
119 #define chmod(path,mode) _chmod (path, mode)
120 #define dup(fd) _dup (fd)
121 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
122
123 #define fchmod(fd,mode) EIO_ENOSYS ()
124 #define chown(path,uid,gid) EIO_ENOSYS ()
125 #define fchown(fd,uid,gid) EIO_ENOSYS ()
126 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
127 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
128 #define mknod(path,mode,dev) EIO_ENOSYS ()
129 #define sync() EIO_ENOSYS ()
130 #define readlink(path,buf,s) EIO_ENOSYS ()
131 #define statvfs(path,buf) EIO_ENOSYS ()
132 #define fstatvfs(fd,buf) EIO_ENOSYS ()
133
134 /* we could even stat and see if it exists */
135 static int
136 symlink (const char *old, const char *neu)
137 {
138 #if WINVER >= 0x0600
139 if (CreateSymbolicLink (neu, old, 1))
140 return 0;
141
142 if (CreateSymbolicLink (neu, old, 0))
143 return 0;
144 #endif
145
146 return EIO_ERRNO (ENOENT, -1);
147 }
148
149 /* POSIX API only */
150 #define CreateHardLink(neu,old,flags) 0
151 #define CreateSymbolicLink(neu,old,flags) 0
152
153 struct statvfs
154 {
155 int dummy;
156 };
157
158 #define DT_DIR EIO_DT_DIR
159 #define DT_REG EIO_DT_REG
160 #define D_NAME(entp) entp.cFileName
161 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
73 162
74#else 163#else
75 164
76# include "config.h"
77# include <sys/time.h> 165 #include <sys/time.h>
78# include <sys/select.h> 166 #include <sys/select.h>
167 #include <sys/statvfs.h>
79# include <unistd.h> 168 #include <unistd.h>
169 #include <signal.h>
170 #include <dirent.h>
171
172 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
173 #include <sys/mman.h>
174 #endif
175
176 #define D_NAME(entp) entp->d_name
177
178 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
179 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
180 #define _DIRENT_HAVE_D_TYPE /* sigh */
181 #define D_INO(de) (de)->d_fileno
182 #define D_NAMLEN(de) (de)->d_namlen
183 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
184 #define D_INO(de) (de)->d_ino
185 #endif
186
187 #ifdef _D_EXACT_NAMLEN
188 #undef D_NAMLEN
189 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
190 #endif
191
192 #ifdef _DIRENT_HAVE_D_TYPE
193 #define D_TYPE(de) (de)->d_type
194 #endif
195
196 #ifndef EIO_STRUCT_DIRENT
197 #define EIO_STRUCT_DIRENT struct dirent
198 #endif
199
200#endif
201
202#if HAVE_UTIMES
80# include <utime.h> 203# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84# ifndef EIO_STRUCT_DIRENT
85# define EIO_STRUCT_DIRENT struct dirent
86# endif 204#endif
87 205
206#if HAVE_SYS_SYSCALL_H
207# include <sys/syscall.h>
208#endif
209
210#if HAVE_SYS_PRCTL_H
211# include <sys/prctl.h>
88#endif 212#endif
89 213
90#if HAVE_SENDFILE 214#if HAVE_SENDFILE
91# if __linux 215# if __linux
92# include <sys/sendfile.h> 216# include <sys/sendfile.h>
93# elif __freebsd 217# elif __FreeBSD__ || defined __APPLE__
94# include <sys/socket.h> 218# include <sys/socket.h>
95# include <sys/uio.h> 219# include <sys/uio.h>
96# elif __hpux 220# elif __hpux
97# include <sys/socket.h> 221# include <sys/socket.h>
98# elif __solaris /* not yet */ 222# elif __solaris
99# include <sys/sendfile.h> 223# include <sys/sendfile.h>
100# else 224# else
101# error sendfile support requested but not available 225# error sendfile support requested but not available
102# endif 226# endif
103#endif 227#endif
104 228
105/* number of seconds after which an idle threads exit */ 229#ifndef D_TYPE
106#define IDLE_TIMEOUT 10 230# define D_TYPE(de) 0
231#endif
232#ifndef D_INO
233# define D_INO(de) 0
234#endif
235#ifndef D_NAMLEN
236# define D_NAMLEN(entp) strlen (D_NAME (entp))
237#endif
107 238
108/* used for struct dirent, AIX doesn't provide it */ 239/* used for struct dirent, AIX doesn't provide it */
109#ifndef NAME_MAX 240#ifndef NAME_MAX
110# define NAME_MAX 4096 241# define NAME_MAX 4096
111#endif 242#endif
112 243
244/* used for readlink etc. */
245#ifndef PATH_MAX
246# define PATH_MAX 4096
247#endif
248
113/* buffer size for various temporary buffers */ 249/* buffer size for various temporary buffers */
114#define EIO_BUFSIZE 65536 250#define EIO_BUFSIZE 65536
115 251
116#define dBUF \ 252#define dBUF \
117 char *eio_buf; \
118 ETP_WORKER_LOCK (self); \
119 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 253 char *eio_buf = malloc (EIO_BUFSIZE); \
120 ETP_WORKER_UNLOCK (self); \
121 errno = ENOMEM; \ 254 errno = ENOMEM; \
122 if (!eio_buf) \ 255 if (!eio_buf) \
123 return -1; 256 return -1
257
258#define FUBd \
259 free (eio_buf)
124 260
125#define EIO_TICKS ((1000000 + 1023) >> 10) 261#define EIO_TICKS ((1000000 + 1023) >> 10)
126
127/*****************************************************************************/
128
129#if __GNUC__ >= 3
130# define expect(expr,value) __builtin_expect ((expr),(value))
131#else
132# define expect(expr,value) (expr)
133#endif
134
135#define expect_false(expr) expect ((expr) != 0, 0)
136#define expect_true(expr) expect ((expr) != 0, 1)
137
138/*****************************************************************************/
139 262
140#define ETP_PRI_MIN EIO_PRI_MIN 263#define ETP_PRI_MIN EIO_PRI_MIN
141#define ETP_PRI_MAX EIO_PRI_MAX 264#define ETP_PRI_MAX EIO_PRI_MAX
142 265
143struct etp_worker; 266struct etp_worker;
147static int eio_finish (eio_req *req); 270static int eio_finish (eio_req *req);
148#define ETP_FINISH(req) eio_finish (req) 271#define ETP_FINISH(req) eio_finish (req)
149static void eio_execute (struct etp_worker *self, eio_req *req); 272static void eio_execute (struct etp_worker *self, eio_req *req);
150#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 273#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
151 274
152#define ETP_WORKER_CLEAR(req) \
153 if (wrk->dbuf) \
154 { \
155 free (wrk->dbuf); \
156 wrk->dbuf = 0; \
157 } \
158 \
159 if (wrk->dirp) \
160 { \
161 closedir (wrk->dirp); \
162 wrk->dirp = 0; \
163 }
164#define ETP_WORKER_COMMON \
165 void *dbuf; \
166 DIR *dirp;
167
168/*****************************************************************************/ 275/*****************************************************************************/
169 276
170#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 277#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
171 278
172/* calculcate time difference in ~1/EIO_TICKS of a second */ 279/* calculate time difference in ~1/EIO_TICKS of a second */
280ecb_inline int
173static int tvdiff (struct timeval *tv1, struct timeval *tv2) 281tvdiff (struct timeval *tv1, struct timeval *tv2)
174{ 282{
175 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 283 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
176 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 284 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
177} 285}
178 286
185static unsigned int max_poll_reqs; /* reslock */ 293static unsigned int max_poll_reqs; /* reslock */
186 294
187static volatile unsigned int nreqs; /* reqlock */ 295static volatile unsigned int nreqs; /* reqlock */
188static volatile unsigned int nready; /* reqlock */ 296static volatile unsigned int nready; /* reqlock */
189static volatile unsigned int npending; /* reqlock */ 297static volatile unsigned int npending; /* reqlock */
190static volatile unsigned int max_idle = 4; 298static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
299static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
191 300
192static mutex_t wrklock = X_MUTEX_INIT; 301static xmutex_t wrklock;
193static mutex_t reslock = X_MUTEX_INIT; 302static xmutex_t reslock;
194static mutex_t reqlock = X_MUTEX_INIT; 303static xmutex_t reqlock;
195static cond_t reqwait = X_COND_INIT; 304static xcond_t reqwait;
196 305
197#if !HAVE_PREADWRITE 306#if !HAVE_PREADWRITE
198/* 307/*
199 * make our pread/pwrite emulation safe against themselves, but not against 308 * make our pread/pwrite emulation safe against themselves, but not against
200 * normal read/write by using a mutex. slows down execution a lot, 309 * normal read/write by using a mutex. slows down execution a lot,
201 * but that's your problem, not mine. 310 * but that's your problem, not mine.
202 */ 311 */
203static mutex_t preadwritelock = X_MUTEX_INIT; 312static xmutex_t preadwritelock;
204#endif 313#endif
205 314
206typedef struct etp_worker 315typedef struct etp_worker
207{ 316{
208 /* locked by wrklock */ 317 /* locked by wrklock */
209 struct etp_worker *prev, *next; 318 struct etp_worker *prev, *next;
210 319
211 thread_t tid; 320 xthread_t tid;
212 321
213 /* locked by reslock, reqlock or wrklock */ 322 /* locked by reslock, reqlock or wrklock */
214 ETP_REQ *req; /* currently processed request */ 323 ETP_REQ *req; /* currently processed request */
215 324
325#ifdef ETP_WORKER_COMMON
216 ETP_WORKER_COMMON 326 ETP_WORKER_COMMON
327#endif
217} etp_worker; 328} etp_worker;
218 329
219static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 330static etp_worker wrk_first; /* NOT etp */
220 331
221#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 332#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
222#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 333#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
223 334
224/* worker threads management */ 335/* worker threads management */
225 336
337static void ecb_cold
226static void etp_worker_clear (etp_worker *wrk) 338etp_worker_clear (etp_worker *wrk)
227{ 339{
228 ETP_WORKER_CLEAR (wrk);
229} 340}
230 341
342static void ecb_cold
231static void etp_worker_free (etp_worker *wrk) 343etp_worker_free (etp_worker *wrk)
232{ 344{
233 wrk->next->prev = wrk->prev; 345 wrk->next->prev = wrk->prev;
234 wrk->prev->next = wrk->next; 346 wrk->prev->next = wrk->next;
235 347
236 free (wrk); 348 free (wrk);
237} 349}
238 350
239static unsigned int etp_nreqs (void) 351static unsigned int
352etp_nreqs (void)
240{ 353{
241 int retval; 354 int retval;
242 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 355 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
243 retval = nreqs; 356 retval = nreqs;
244 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 357 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
245 return retval; 358 return retval;
246} 359}
247 360
248static unsigned int etp_nready (void) 361static unsigned int
362etp_nready (void)
249{ 363{
250 unsigned int retval; 364 unsigned int retval;
251 365
252 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 366 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
253 retval = nready; 367 retval = nready;
254 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 368 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
255 369
256 return retval; 370 return retval;
257} 371}
258 372
259static unsigned int etp_npending (void) 373static unsigned int
374etp_npending (void)
260{ 375{
261 unsigned int retval; 376 unsigned int retval;
262 377
263 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 378 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
264 retval = npending; 379 retval = npending;
265 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 380 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
266 381
267 return retval; 382 return retval;
268} 383}
269 384
270static unsigned int etp_nthreads (void) 385static unsigned int
386etp_nthreads (void)
271{ 387{
272 unsigned int retval; 388 unsigned int retval;
273 389
274 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 390 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
275 retval = started; 391 retval = started;
289} etp_reqq; 405} etp_reqq;
290 406
291static etp_reqq req_queue; 407static etp_reqq req_queue;
292static etp_reqq res_queue; 408static etp_reqq res_queue;
293 409
410static void ecb_noinline ecb_cold
411reqq_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
421static int ecb_noinline
294static int reqq_push (etp_reqq *q, ETP_REQ *req) 422reqq_push (etp_reqq *q, ETP_REQ *req)
295{ 423{
296 int pri = req->pri; 424 int pri = req->pri;
297 req->next = 0; 425 req->next = 0;
298 426
299 if (q->qe[pri]) 427 if (q->qe[pri])
305 q->qe[pri] = q->qs[pri] = req; 433 q->qe[pri] = q->qs[pri] = req;
306 434
307 return q->size++; 435 return q->size++;
308} 436}
309 437
438static ETP_REQ * ecb_noinline
310static ETP_REQ *reqq_shift (etp_reqq *q) 439reqq_shift (etp_reqq *q)
311{ 440{
312 int pri; 441 int pri;
313 442
314 if (!q->size) 443 if (!q->size)
315 return 0; 444 return 0;
330 } 459 }
331 460
332 abort (); 461 abort ();
333} 462}
334 463
335static void etp_atfork_prepare (void) 464static int ecb_cold
465etp_init (void (*want_poll)(void), void (*done_poll)(void))
336{ 466{
337 X_LOCK (wrklock); 467 X_MUTEX_CREATE (wrklock);
338 X_LOCK (reqlock); 468 X_MUTEX_CREATE (reslock);
339 X_LOCK (reslock); 469 X_MUTEX_CREATE (reqlock);
340#if !HAVE_PREADWRITE 470 X_COND_CREATE (reqwait);
341 X_LOCK (preadwritelock);
342#endif
343}
344 471
345static void etp_atfork_parent (void) 472 reqq_init (&req_queue);
346{ 473 reqq_init (&res_queue);
347#if !HAVE_PREADWRITE
348 X_UNLOCK (preadwritelock);
349#endif
350 X_UNLOCK (reslock);
351 X_UNLOCK (reqlock);
352 X_UNLOCK (wrklock);
353}
354 474
355static void etp_atfork_child (void) 475 wrk_first.next =
356{ 476 wrk_first.prev = &wrk_first;
357 ETP_REQ *prv;
358
359 while ((prv = reqq_shift (&req_queue)))
360 ETP_DESTROY (prv);
361
362 while ((prv = reqq_shift (&res_queue)))
363 ETP_DESTROY (prv);
364
365 while (wrk_first.next != &wrk_first)
366 {
367 etp_worker *wrk = wrk_first.next;
368
369 if (wrk->req)
370 ETP_DESTROY (wrk->req);
371
372 etp_worker_clear (wrk);
373 etp_worker_free (wrk);
374 }
375 477
376 started = 0; 478 started = 0;
377 idle = 0; 479 idle = 0;
378 nreqs = 0; 480 nreqs = 0;
379 nready = 0; 481 nready = 0;
380 npending = 0; 482 npending = 0;
381 483
382 etp_atfork_parent ();
383}
384
385static void
386etp_once_init (void)
387{
388 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
389}
390
391static int
392etp_init (void (*want_poll)(void), void (*done_poll)(void))
393{
394 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
395
396 pthread_once (&doinit, etp_once_init);
397
398 want_poll_cb = want_poll; 484 want_poll_cb = want_poll;
399 done_poll_cb = done_poll; 485 done_poll_cb = done_poll;
400 486
401 return 0; 487 return 0;
402} 488}
403 489
404X_THREAD_PROC (etp_proc); 490X_THREAD_PROC (etp_proc);
405 491
492static void ecb_cold
406static void etp_start_thread (void) 493etp_start_thread (void)
407{ 494{
408 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 495 etp_worker *wrk = calloc (1, sizeof (etp_worker));
409 496
410 /*TODO*/ 497 /*TODO*/
411 assert (("unable to allocate worker thread data", wrk)); 498 assert (("unable to allocate worker thread data", wrk));
424 free (wrk); 511 free (wrk);
425 512
426 X_UNLOCK (wrklock); 513 X_UNLOCK (wrklock);
427} 514}
428 515
516static void
429static void etp_maybe_start_thread (void) 517etp_maybe_start_thread (void)
430{ 518{
431 if (expect_true (etp_nthreads () >= wanted)) 519 if (ecb_expect_true (etp_nthreads () >= wanted))
432 return; 520 return;
433 521
434 /* todo: maybe use idle here, but might be less exact */ 522 /* todo: maybe use idle here, but might be less exact */
435 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 523 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
436 return; 524 return;
437 525
438 etp_start_thread (); 526 etp_start_thread ();
439} 527}
440 528
529static void ecb_cold
441static void etp_end_thread (void) 530etp_end_thread (void)
442{ 531{
443 eio_req *req = calloc (1, sizeof (eio_req)); 532 eio_req *req = calloc (1, sizeof (eio_req));
444 533
445 req->type = -1; 534 req->type = -1;
446 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 535 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
453 X_LOCK (wrklock); 542 X_LOCK (wrklock);
454 --started; 543 --started;
455 X_UNLOCK (wrklock); 544 X_UNLOCK (wrklock);
456} 545}
457 546
458static int etp_poll (void) 547static int
548etp_poll (void)
459{ 549{
460 unsigned int maxreqs; 550 unsigned int maxreqs;
461 unsigned int maxtime; 551 unsigned int maxtime;
462 struct timeval tv_start, tv_now; 552 struct timeval tv_start, tv_now;
463 553
493 583
494 X_LOCK (reqlock); 584 X_LOCK (reqlock);
495 --nreqs; 585 --nreqs;
496 X_UNLOCK (reqlock); 586 X_UNLOCK (reqlock);
497 587
498 if (expect_false (req->type == EIO_GROUP && req->size)) 588 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
499 { 589 {
500 req->int1 = 1; /* mark request as delayed */ 590 req->int1 = 1; /* mark request as delayed */
501 continue; 591 continue;
502 } 592 }
503 else 593 else
504 { 594 {
505 int res = ETP_FINISH (req); 595 int res = ETP_FINISH (req);
506 if (expect_false (res)) 596 if (ecb_expect_false (res))
507 return res; 597 return res;
508 } 598 }
509 599
510 if (expect_false (maxreqs && !--maxreqs)) 600 if (ecb_expect_false (maxreqs && !--maxreqs))
511 break; 601 break;
512 602
513 if (maxtime) 603 if (maxtime)
514 { 604 {
515 gettimeofday (&tv_now, 0); 605 gettimeofday (&tv_now, 0);
521 611
522 errno = EAGAIN; 612 errno = EAGAIN;
523 return -1; 613 return -1;
524} 614}
525 615
616static void
526static void etp_cancel (ETP_REQ *req) 617etp_cancel (ETP_REQ *req)
527{ 618{
528 X_LOCK (wrklock); 619 req->cancelled = 1;
529 req->flags |= EIO_FLAG_CANCELLED;
530 X_UNLOCK (wrklock);
531 620
532 eio_grp_cancel (req); 621 eio_grp_cancel (req);
533} 622}
534 623
624static void
535static void etp_submit (ETP_REQ *req) 625etp_submit (ETP_REQ *req)
536{ 626{
537 req->pri -= ETP_PRI_MIN; 627 req->pri -= ETP_PRI_MIN;
538 628
539 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 629 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
540 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 630 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
541 631
542 if (expect_false (req->type == EIO_GROUP)) 632 if (ecb_expect_false (req->type == EIO_GROUP))
543 { 633 {
544 /* I hope this is worth it :/ */ 634 /* I hope this is worth it :/ */
545 X_LOCK (reqlock); 635 X_LOCK (reqlock);
546 ++nreqs; 636 ++nreqs;
547 X_UNLOCK (reqlock); 637 X_UNLOCK (reqlock);
566 656
567 etp_maybe_start_thread (); 657 etp_maybe_start_thread ();
568 } 658 }
569} 659}
570 660
661static void ecb_cold
571static void etp_set_max_poll_time (double nseconds) 662etp_set_max_poll_time (double nseconds)
572{ 663{
573 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 664 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
574 max_poll_time = nseconds; 665 max_poll_time = nseconds * EIO_TICKS;
575 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 666 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
576} 667}
577 668
669static void ecb_cold
578static void etp_set_max_poll_reqs (unsigned int maxreqs) 670etp_set_max_poll_reqs (unsigned int maxreqs)
579{ 671{
580 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 672 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
581 max_poll_reqs = maxreqs; 673 max_poll_reqs = maxreqs;
582 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 674 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
583} 675}
584 676
677static void ecb_cold
585static void etp_set_max_idle (unsigned int nthreads) 678etp_set_max_idle (unsigned int nthreads)
586{ 679{
587 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 680 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
588 max_idle = nthreads <= 0 ? 1 : nthreads; 681 max_idle = nthreads;
589 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 682 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
590} 683}
591 684
685static void ecb_cold
686etp_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
693static void ecb_cold
592static void etp_set_min_parallel (unsigned int nthreads) 694etp_set_min_parallel (unsigned int nthreads)
593{ 695{
594 if (wanted < nthreads) 696 if (wanted < nthreads)
595 wanted = nthreads; 697 wanted = nthreads;
596} 698}
597 699
700static void ecb_cold
598static void etp_set_max_parallel (unsigned int nthreads) 701etp_set_max_parallel (unsigned int nthreads)
599{ 702{
600 if (wanted > nthreads) 703 if (wanted > nthreads)
601 wanted = nthreads; 704 wanted = nthreads;
602 705
603 while (started > wanted) 706 while (started > wanted)
604 etp_end_thread (); 707 etp_end_thread ();
605} 708}
606 709
607/*****************************************************************************/ 710/*****************************************************************************/
608 711
712static void
609static void grp_try_feed (eio_req *grp) 713grp_try_feed (eio_req *grp)
610{ 714{
611 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 715 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
612 { 716 {
613 grp->flags &= ~EIO_FLAG_GROUPADD; 717 grp->flags &= ~EIO_FLAG_GROUPADD;
614 718
621 break; 725 break;
622 } 726 }
623 } 727 }
624} 728}
625 729
730static int
626static int grp_dec (eio_req *grp) 731grp_dec (eio_req *grp)
627{ 732{
628 --grp->size; 733 --grp->size;
629 734
630 /* call feeder, if applicable */ 735 /* call feeder, if applicable */
631 grp_try_feed (grp); 736 grp_try_feed (grp);
635 return eio_finish (grp); 740 return eio_finish (grp);
636 else 741 else
637 return 0; 742 return 0;
638} 743}
639 744
745static void
640void eio_destroy (eio_req *req) 746eio_destroy (eio_req *req)
641{ 747{
642 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 748 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
643 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 749 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
644 750
645 EIO_DESTROY (req); 751 EIO_DESTROY (req);
646} 752}
647 753
754static int
648static int eio_finish (eio_req *req) 755eio_finish (eio_req *req)
649{ 756{
650 int res = EIO_FINISH (req); 757 int res = EIO_FINISH (req);
651 758
652 if (req->grp) 759 if (req->grp)
653 { 760 {
661 if (grp->grp_first == req) 768 if (grp->grp_first == req)
662 grp->grp_first = req->grp_next; 769 grp->grp_first = req->grp_next;
663 770
664 res2 = grp_dec (grp); 771 res2 = grp_dec (grp);
665 772
666 if (!res && res2) 773 if (!res)
667 res = res2; 774 res = res2;
668 } 775 }
669 776
670 eio_destroy (req); 777 eio_destroy (req);
671 778
672 return res; 779 return res;
673} 780}
674 781
782void
675void eio_grp_cancel (eio_req *grp) 783eio_grp_cancel (eio_req *grp)
676{ 784{
677 for (grp = grp->grp_first; grp; grp = grp->grp_next) 785 for (grp = grp->grp_first; grp; grp = grp->grp_next)
678 eio_cancel (grp); 786 eio_cancel (grp);
679} 787}
680 788
789void
681void eio_cancel (eio_req *req) 790eio_cancel (eio_req *req)
682{ 791{
683 etp_cancel (req); 792 etp_cancel (req);
684} 793}
685 794
795void
686void eio_submit (eio_req *req) 796eio_submit (eio_req *req)
687{ 797{
688 etp_submit (req); 798 etp_submit (req);
689} 799}
690 800
691unsigned int eio_nreqs (void) 801unsigned int
802eio_nreqs (void)
692{ 803{
693 return etp_nreqs (); 804 return etp_nreqs ();
694} 805}
695 806
696unsigned int eio_nready (void) 807unsigned int
808eio_nready (void)
697{ 809{
698 return etp_nready (); 810 return etp_nready ();
699} 811}
700 812
701unsigned int eio_npending (void) 813unsigned int
814eio_npending (void)
702{ 815{
703 return etp_npending (); 816 return etp_npending ();
704} 817}
705 818
706unsigned int eio_nthreads (void) 819unsigned int ecb_cold
820eio_nthreads (void)
707{ 821{
708 return etp_nthreads (); 822 return etp_nthreads ();
709} 823}
710 824
825void ecb_cold
711void eio_set_max_poll_time (double nseconds) 826eio_set_max_poll_time (double nseconds)
712{ 827{
713 etp_set_max_poll_time (nseconds); 828 etp_set_max_poll_time (nseconds);
714} 829}
715 830
831void ecb_cold
716void eio_set_max_poll_reqs (unsigned int maxreqs) 832eio_set_max_poll_reqs (unsigned int maxreqs)
717{ 833{
718 etp_set_max_poll_reqs (maxreqs); 834 etp_set_max_poll_reqs (maxreqs);
719} 835}
720 836
837void ecb_cold
721void eio_set_max_idle (unsigned int nthreads) 838eio_set_max_idle (unsigned int nthreads)
722{ 839{
723 etp_set_max_idle (nthreads); 840 etp_set_max_idle (nthreads);
724} 841}
725 842
843void ecb_cold
844eio_set_idle_timeout (unsigned int seconds)
845{
846 etp_set_idle_timeout (seconds);
847}
848
849void ecb_cold
726void eio_set_min_parallel (unsigned int nthreads) 850eio_set_min_parallel (unsigned int nthreads)
727{ 851{
728 etp_set_min_parallel (nthreads); 852 etp_set_min_parallel (nthreads);
729} 853}
730 854
855void ecb_cold
731void eio_set_max_parallel (unsigned int nthreads) 856eio_set_max_parallel (unsigned int nthreads)
732{ 857{
733 etp_set_max_parallel (nthreads); 858 etp_set_max_parallel (nthreads);
734} 859}
735 860
736int eio_poll (void) 861int eio_poll (void)
745# undef pread 870# undef pread
746# undef pwrite 871# undef pwrite
747# define pread eio__pread 872# define pread eio__pread
748# define pwrite eio__pwrite 873# define pwrite eio__pwrite
749 874
750static ssize_t 875static eio_ssize_t
751eio__pread (int fd, void *buf, size_t count, off_t offset) 876eio__pread (int fd, void *buf, size_t count, off_t offset)
752{ 877{
753 ssize_t res; 878 eio_ssize_t res;
754 off_t ooffset; 879 off_t ooffset;
755 880
756 X_LOCK (preadwritelock); 881 X_LOCK (preadwritelock);
757 ooffset = lseek (fd, 0, SEEK_CUR); 882 ooffset = lseek (fd, 0, SEEK_CUR);
758 lseek (fd, offset, SEEK_SET); 883 lseek (fd, offset, SEEK_SET);
761 X_UNLOCK (preadwritelock); 886 X_UNLOCK (preadwritelock);
762 887
763 return res; 888 return res;
764} 889}
765 890
766static ssize_t 891static eio_ssize_t
767eio__pwrite (int fd, void *buf, size_t count, off_t offset) 892eio__pwrite (int fd, void *buf, size_t count, off_t offset)
768{ 893{
769 ssize_t res; 894 eio_ssize_t res;
770 off_t ooffset; 895 off_t ooffset;
771 896
772 X_LOCK (preadwritelock); 897 X_LOCK (preadwritelock);
773 ooffset = lseek (fd, 0, SEEK_CUR); 898 ooffset = lseek (fd, 0, SEEK_CUR);
774 lseek (fd, offset, SEEK_SET); 899 lseek (fd, offset, SEEK_SET);
778 903
779 return res; 904 return res;
780} 905}
781#endif 906#endif
782 907
783#ifndef HAVE_FUTIMES 908#ifndef HAVE_UTIMES
784 909
785# undef utimes 910# undef utimes
786# undef futimes
787# define utimes(path,times) eio__utimes (path, times) 911# define utimes(path,times) eio__utimes (path, times)
788# define futimes(fd,times) eio__futimes (fd, times)
789 912
790static int 913static int
791eio__utimes (const char *filename, const struct timeval times[2]) 914eio__utimes (const char *filename, const struct timeval times[2])
792{ 915{
793 if (times) 916 if (times)
801 } 924 }
802 else 925 else
803 return utime (filename, 0); 926 return utime (filename, 0);
804} 927}
805 928
929#endif
930
931#ifndef HAVE_FUTIMES
932
933# undef futimes
934# define futimes(fd,times) eio__futimes (fd, times)
935
936static int
806static int eio__futimes (int fd, const struct timeval tv[2]) 937eio__futimes (int fd, const struct timeval tv[2])
807{ 938{
808 errno = ENOSYS; 939 errno = ENOSYS;
809 return -1; 940 return -1;
810} 941}
811 942
814#if !HAVE_FDATASYNC 945#if !HAVE_FDATASYNC
815# undef fdatasync 946# undef fdatasync
816# define fdatasync(fd) fsync (fd) 947# define fdatasync(fd) fsync (fd)
817#endif 948#endif
818 949
950static int
951eio__syncfs (int fd)
952{
953 int res;
954
955#if HAVE_SYS_SYNCFS
956 res = (int)syscall (__NR_syncfs, (int)(fd));
957#else
958 res = -1;
959 errno = ENOSYS;
960#endif
961
962 if (res < 0 && errno == ENOSYS && fd >= 0)
963 sync ();
964
965 return res;
966}
967
819/* sync_file_range always needs emulation */ 968/* sync_file_range always needs emulation */
820int 969static int
821eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 970eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
822{ 971{
823#if HAVE_SYNC_FILE_RANGE 972#if HAVE_SYNC_FILE_RANGE
824 int res; 973 int res;
825 974
838 if (!res || errno != ENOSYS) 987 if (!res || errno != ENOSYS)
839 return res; 988 return res;
840#endif 989#endif
841 990
842 /* even though we could play tricks with the flags, it's better to always 991 /* even though we could play tricks with the flags, it's better to always
843 * call fdatasync, as thta matches the expectation of it's users best */ 992 * call fdatasync, as that matches the expectation of its users best */
844 return fdatasync (fd); 993 return fdatasync (fd);
994}
995
996static int
997eio__fallocate (int fd, int mode, off_t offset, size_t len)
998{
999#if HAVE_FALLOCATE
1000 return fallocate (fd, mode, offset, len);
1001#else
1002 errno = ENOSYS;
1003 return -1;
1004#endif
845} 1005}
846 1006
847#if !HAVE_READAHEAD 1007#if !HAVE_READAHEAD
848# undef readahead 1008# undef readahead
849# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1009# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
850 1010
851static ssize_t 1011static eio_ssize_t
852eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1012eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
853{ 1013{
854 size_t todo = count; 1014 size_t todo = count;
855 dBUF; 1015 dBUF;
856 1016
861 pread (fd, eio_buf, len, offset); 1021 pread (fd, eio_buf, len, offset);
862 offset += len; 1022 offset += len;
863 todo -= len; 1023 todo -= len;
864 } 1024 }
865 1025
1026 FUBd;
1027
866 errno = 0; 1028 errno = 0;
867 return count; 1029 return count;
868} 1030}
869 1031
870#endif 1032#endif
871 1033
872/* sendfile always needs emulation */ 1034/* sendfile always needs emulation */
873static ssize_t 1035static eio_ssize_t
874eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1036eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
875{ 1037{
1038 eio_ssize_t written = 0;
876 ssize_t res; 1039 eio_ssize_t res;
877 1040
878 if (!count) 1041 if (!count)
879 return 0; 1042 return 0;
880 1043
1044 for (;;)
1045 {
1046#ifdef __APPLE__
1047# undef HAVE_SENDFILE /* broken, as everything on os x */
1048#endif
881#if HAVE_SENDFILE 1049#if HAVE_SENDFILE
882# if __linux 1050# if __linux
1051 off_t soffset = offset;
883 res = sendfile (ofd, ifd, &offset, count); 1052 res = sendfile (ofd, ifd, &soffset, count);
884 1053
885# elif __freebsd 1054# elif __FreeBSD__
886 /* 1055 /*
887 * Of course, the freebsd sendfile is a dire hack with no thoughts 1056 * Of course, the freebsd sendfile is a dire hack with no thoughts
888 * wasted on making it similar to other I/O functions. 1057 * wasted on making it similar to other I/O functions.
889 */ 1058 */
890 {
891 off_t sbytes; 1059 off_t sbytes;
892 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1060 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
893 1061
894 if (res < 0 && sbytes) 1062 #if 0 /* according to the manpage, this is correct, but broken behaviour */
895 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1063 /* freebsd' sendfile will return 0 on success */
1064 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1065 /* not on e.g. EIO or EPIPE - sounds broken */
1066 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
896 res = sbytes; 1067 res = sbytes;
897 } 1068 #endif
1069
1070 /* according to source inspection, this is correct, and useful behaviour */
1071 if (sbytes)
1072 res = sbytes;
1073
1074# elif defined (__APPLE__)
1075 off_t sbytes = count;
1076 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1077
1078 /* according to the manpage, sbytes is always valid */
1079 if (sbytes)
1080 res = sbytes;
898 1081
899# elif __hpux 1082# elif __hpux
900 res = sendfile (ofd, ifd, offset, count, 0, 0); 1083 res = sendfile (ofd, ifd, offset, count, 0, 0);
901 1084
902# elif __solaris 1085# elif __solaris
903 {
904 struct sendfilevec vec; 1086 struct sendfilevec vec;
905 size_t sbytes; 1087 size_t sbytes;
906 1088
907 vec.sfv_fd = ifd; 1089 vec.sfv_fd = ifd;
908 vec.sfv_flag = 0; 1090 vec.sfv_flag = 0;
909 vec.sfv_off = offset; 1091 vec.sfv_off = offset;
910 vec.sfv_len = count; 1092 vec.sfv_len = count;
911 1093
912 res = sendfilev (ofd, &vec, 1, &sbytes); 1094 res = sendfilev (ofd, &vec, 1, &sbytes);
913 1095
914 if (res < 0 && sbytes) 1096 if (res < 0 && sbytes)
915 res = sbytes; 1097 res = sbytes;
916 }
917 1098
918# endif 1099# endif
1100
1101#elif defined (_WIN32) && 0
1102 /* does not work, just for documentation of what would need to be done */
1103 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1104 /* libeio guarantees that the file offset does not change, and windows */
1105 /* has no way to get an independent handle to the same file description */
1106 HANDLE h = TO_SOCKET (ifd);
1107 SetFilePointer (h, offset, 0, FILE_BEGIN);
1108 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1109
919#else 1110#else
920 res = -1; 1111 res = -1;
921 errno = ENOSYS; 1112 errno = ENOSYS;
922#endif 1113#endif
923 1114
1115 /* we assume sendfile can copy at least 128mb in one go */
1116 if (res <= 128 * 1024 * 1024)
1117 {
1118 if (res > 0)
1119 written += res;
1120
1121 if (written)
1122 return written;
1123
1124 break;
1125 }
1126 else
1127 {
1128 /* if we requested more, then probably the kernel was lazy */
1129 written += res;
1130 offset += res;
1131 count -= res;
1132
1133 if (!count)
1134 return written;
1135 }
1136 }
1137
924 if (res < 0 1138 if (res < 0
925 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1139 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1140 /* BSDs */
1141#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1142 || errno == ENOTSUP
1143#endif
1144#ifdef EOPNOTSUPP /* windows */
1145 || errno == EOPNOTSUPP /* BSDs */
1146#endif
926#if __solaris 1147#if __solaris
927 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1148 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
928#endif 1149#endif
929 ) 1150 )
930 ) 1151 )
934 1155
935 res = 0; 1156 res = 0;
936 1157
937 while (count) 1158 while (count)
938 { 1159 {
939 ssize_t cnt; 1160 eio_ssize_t cnt;
940 1161
941 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1162 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
942 1163
943 if (cnt <= 0) 1164 if (cnt <= 0)
944 { 1165 {
956 1177
957 offset += cnt; 1178 offset += cnt;
958 res += cnt; 1179 res += cnt;
959 count -= cnt; 1180 count -= cnt;
960 } 1181 }
1182
1183 FUBd;
961 } 1184 }
962 1185
963 return res; 1186 return res;
1187}
1188
1189#ifdef PAGESIZE
1190# define eio_pagesize() PAGESIZE
1191#else
1192static intptr_t
1193eio_pagesize (void)
1194{
1195 static intptr_t page;
1196
1197 if (!page)
1198 page = sysconf (_SC_PAGESIZE);
1199
1200 return page;
1201}
1202#endif
1203
1204static void
1205eio_page_align (void **addr, size_t *length)
1206{
1207 intptr_t mask = eio_pagesize () - 1;
1208
1209 /* round down addr */
1210 intptr_t adj = mask & (intptr_t)*addr;
1211
1212 *addr = (void *)((intptr_t)*addr - adj);
1213 *length += adj;
1214
1215 /* round up length */
1216 *length = (*length + mask) & ~mask;
1217}
1218
1219#if !_POSIX_MEMLOCK
1220# define eio__mlockall(a) EIO_ENOSYS ()
1221#else
1222
1223static int
1224eio__mlockall (int flags)
1225{
1226 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1227 extern int mallopt (int, int);
1228 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1229 #endif
1230
1231 if (EIO_MCL_CURRENT != MCL_CURRENT
1232 || EIO_MCL_FUTURE != MCL_FUTURE)
1233 {
1234 flags = 0
1235 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1236 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1237 }
1238
1239 return mlockall (flags);
1240}
1241#endif
1242
1243#if !_POSIX_MEMLOCK_RANGE
1244# define eio__mlock(a,b) EIO_ENOSYS ()
1245#else
1246
1247static int
1248eio__mlock (void *addr, size_t length)
1249{
1250 eio_page_align (&addr, &length);
1251
1252 return mlock (addr, length);
1253}
1254
1255#endif
1256
1257#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1258# define eio__msync(a,b,c) EIO_ENOSYS ()
1259#else
1260
1261static int
1262eio__msync (void *mem, size_t len, int flags)
1263{
1264 eio_page_align (&mem, &len);
1265
1266 if (EIO_MS_ASYNC != MS_SYNC
1267 || EIO_MS_INVALIDATE != MS_INVALIDATE
1268 || EIO_MS_SYNC != MS_SYNC)
1269 {
1270 flags = 0
1271 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1272 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1273 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1274 }
1275
1276 return msync (mem, len, flags);
1277}
1278
1279#endif
1280
1281static int
1282eio__mtouch (eio_req *req)
1283{
1284 void *mem = req->ptr2;
1285 size_t len = req->size;
1286 int flags = req->int1;
1287
1288 eio_page_align (&mem, &len);
1289
1290 {
1291 intptr_t addr = (intptr_t)mem;
1292 intptr_t end = addr + len;
1293 intptr_t page = eio_pagesize ();
1294
1295 if (addr < end)
1296 if (flags & EIO_MT_MODIFY) /* modify */
1297 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1298 else
1299 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1300 }
1301
1302 return 0;
1303}
1304
1305/*****************************************************************************/
1306/* requests implemented outside eio_execute, because they are so large */
1307
1308static void
1309eio__realpath (eio_req *req, etp_worker *self)
1310{
1311 char *rel = req->ptr1;
1312 char *res;
1313 char *tmp1, *tmp2;
1314#if SYMLOOP_MAX > 32
1315 int symlinks = SYMLOOP_MAX;
1316#else
1317 int symlinks = 32;
1318#endif
1319
1320 req->result = -1;
1321
1322 errno = EINVAL;
1323 if (!rel)
1324 return;
1325
1326 errno = ENOENT;
1327 if (!*rel)
1328 return;
1329
1330 if (!req->ptr2)
1331 {
1332 X_LOCK (wrklock);
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;
1340 }
1341
1342 res = req->ptr2;
1343 tmp1 = res + PATH_MAX;
1344 tmp2 = tmp1 + PATH_MAX;
1345
1346#if 0 /* disabled, the musl way to do things is just too racy */
1347#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1348 /* on linux we may be able to ask the kernel */
1349 {
1350 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1351
1352 if (fd >= 0)
1353 {
1354 sprintf (tmp1, "/proc/self/fd/%d", fd);
1355 req->result = readlink (tmp1, res, PATH_MAX);
1356 close (fd);
1357
1358 /* here we should probably stat the open file and the disk file, to make sure they still match */
1359
1360 if (req->result > 0)
1361 goto done;
1362 }
1363 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1364 return;
1365 }
1366#endif
1367#endif
1368
1369 if (*rel != '/')
1370 {
1371 if (!getcwd (res, PATH_MAX))
1372 return;
1373
1374 if (res [1]) /* only use if not / */
1375 res += strlen (res);
1376 }
1377
1378 while (*rel)
1379 {
1380 eio_ssize_t len, linklen;
1381 char *beg = rel;
1382
1383 while (*rel && *rel != '/')
1384 ++rel;
1385
1386 len = rel - beg;
1387
1388 if (!len) /* skip slashes */
1389 {
1390 ++rel;
1391 continue;
1392 }
1393
1394 if (beg [0] == '.')
1395 {
1396 if (len == 1)
1397 continue; /* . - nop */
1398
1399 if (beg [1] == '.' && len == 2)
1400 {
1401 /* .. - back up one component, if possible */
1402
1403 while (res != req->ptr2)
1404 if (*--res == '/')
1405 break;
1406
1407 continue;
1408 }
1409 }
1410
1411 errno = ENAMETOOLONG;
1412 if (res + 1 + len + 1 >= tmp1)
1413 return;
1414
1415 /* copy one component */
1416 *res = '/';
1417 memcpy (res + 1, beg, len);
1418
1419 /* zero-terminate, for readlink */
1420 res [len + 1] = 0;
1421
1422 /* now check if it's a symlink */
1423 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1424
1425 if (linklen < 0)
1426 {
1427 if (errno != EINVAL)
1428 return;
1429
1430 /* it's a normal directory. hopefully */
1431 res += len + 1;
1432 }
1433 else
1434 {
1435 /* yay, it was a symlink - build new path in tmp2 */
1436 int rellen = strlen (rel);
1437
1438 errno = ENAMETOOLONG;
1439 if (linklen + 1 + rellen >= PATH_MAX)
1440 return;
1441
1442 errno = ELOOP;
1443 if (!--symlinks)
1444 return;
1445
1446 if (*tmp1 == '/')
1447 res = req->ptr2; /* symlink resolves to an absolute path */
1448
1449 /* we need to be careful, as rel might point into tmp2 already */
1450 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1451 tmp2 [linklen] = '/';
1452 memcpy (tmp2, tmp1, linklen);
1453
1454 rel = tmp2;
1455 }
1456 }
1457
1458 /* special case for the lone root path */
1459 if (res == req->ptr2)
1460 *res++ = '/';
1461
1462 req->result = res - (char *)req->ptr2;
1463
1464done:
1465 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1466}
1467
1468static signed char
1469eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1470{
1471 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1472 : a->inode < b->inode ? -1
1473 : a->inode > b->inode ? 1
1474 : 0;
1475}
1476
1477#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1478
1479#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1480#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1481
1482static void
1483eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1484{
1485 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1486 unsigned char *bit = bits;
1487
1488 assert (CHAR_BIT == 8);
1489 assert (sizeof (eio_dirent) * 8 < 256);
1490 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1491 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1492
1493 if (size <= EIO_SORT_FAST)
1494 return;
1495
1496 /* first prepare an array of bits to test in our radix sort */
1497 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1498 /* inode_bits must contain all inodes ORed together */
1499 /* which is used to skip bits that are 0 everywhere, which is very common */
1500 {
1501 eio_ino_t endianness;
1502 int i, j;
1503
1504 /* we store the byte offset of byte n into byte n of "endianness" */
1505 for (i = 0; i < sizeof (eio_ino_t); ++i)
1506 ((unsigned char *)&endianness)[i] = i;
1507
1508 *bit++ = 0;
1509
1510 for (i = 0; i < sizeof (eio_ino_t); ++i)
1511 {
1512 /* shifting off the byte offsets out of "endianness" */
1513 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1514 endianness >>= 8;
1515
1516 for (j = 0; j < 8; ++j)
1517 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1518 *bit++ = offs + j;
1519 }
1520
1521 for (j = 0; j < 8; ++j)
1522 if (score_bits & (1 << j))
1523 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1524 }
1525
1526 /* now actually do the sorting (a variant of MSD radix sort) */
1527 {
1528 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1529 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1530 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1531 int stk_idx = 0;
1532
1533 base_stk [stk_idx] = dents;
1534 end_stk [stk_idx] = dents + size;
1535 bit_stk [stk_idx] = bit - 1;
1536
1537 do
1538 {
1539 base = base_stk [stk_idx];
1540 end = end_stk [stk_idx];
1541 bit = bit_stk [stk_idx];
1542
1543 for (;;)
1544 {
1545 unsigned char O = *bit >> 3;
1546 unsigned char M = 1 << (*bit & 7);
1547
1548 eio_dirent *a = base;
1549 eio_dirent *b = end;
1550
1551 if (b - a < EIO_SORT_CUTOFF)
1552 break;
1553
1554 /* now bit-partition the array on the bit */
1555 /* this ugly asymmetric loop seems to perform much better than typical */
1556 /* partition algos found in the literature */
1557 do
1558 if (!(((unsigned char *)a)[O] & M))
1559 ++a;
1560 else if (!(((unsigned char *)--b)[O] & M))
1561 {
1562 eio_dirent tmp = *a; *a = *b; *b = tmp;
1563 ++a;
1564 }
1565 while (b > a);
1566
1567 /* next bit, or stop, if no bits left in this path */
1568 if (!*--bit)
1569 break;
1570
1571 base_stk [stk_idx] = a;
1572 end_stk [stk_idx] = end;
1573 bit_stk [stk_idx] = bit;
1574 ++stk_idx;
1575
1576 end = a;
1577 }
1578 }
1579 while (stk_idx--);
1580 }
1581}
1582
1583static void
1584eio_dent_insertion_sort (eio_dirent *dents, int size)
1585{
1586 /* first move the smallest element to the front, to act as a sentinel */
1587 {
1588 int i;
1589 eio_dirent *min = dents;
1590
1591 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1592 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1593 if (EIO_DENT_CMP (dents [i], <, *min))
1594 min = &dents [i];
1595
1596 /* swap elements 0 and j (minimum) */
1597 {
1598 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1599 }
1600 }
1601
1602 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1603 {
1604 eio_dirent *i, *j;
1605
1606 for (i = dents + 1; i < dents + size; ++i)
1607 {
1608 eio_dirent value = *i;
1609
1610 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1611 j [1] = j [0];
1612
1613 j [1] = value;
1614 }
1615 }
1616}
1617
1618static void
1619eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1620{
1621 if (size <= 1)
1622 return; /* our insertion sort relies on size > 0 */
1623
1624 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1625 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1626 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1627
1628 /* use an insertion sort at the end, or for small arrays, */
1629 /* as insertion sort is more efficient for small partitions */
1630 eio_dent_insertion_sort (dents, size);
964} 1631}
965 1632
966/* read a full directory */ 1633/* read a full directory */
967static void 1634static void
968eio__scandir (eio_req *req, etp_worker *self) 1635eio__scandir (eio_req *req, etp_worker *self)
969{ 1636{
1637 char *name, *names;
1638 int namesalloc = 4096 - sizeof (void *) * 4;
1639 int namesoffs = 0;
1640 int flags = req->int1;
1641 eio_dirent *dents = 0;
1642 int dentalloc = 128;
1643 int dentoffs = 0;
1644 eio_ino_t inode_bits = 0;
1645#ifdef _WIN32
1646 HANDLE dirp;
1647 WIN32_FIND_DATA entp;
1648#else
970 DIR *dirp; 1649 DIR *dirp;
971 EIO_STRUCT_DIRENT *entp; 1650 EIO_STRUCT_DIRENT *entp;
972 char *name, *names; 1651#endif
973 int memlen = 4096;
974 int memofs = 0;
975 int res = 0;
976 1652
977 X_LOCK (wrklock); 1653 req->result = -1;
978 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1654
1655 if (!(flags & EIO_READDIR_DENTS))
1656 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1657
1658#ifdef _WIN32
1659 {
1660 int len = strlen ((const char *)req->ptr1);
1661 char *path = malloc (MAX_PATH);
1662 const char *fmt;
1663
1664 if (!len)
1665 fmt = "./*";
1666 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1667 fmt = "%s*";
1668 else
1669 fmt = "%s/*";
1670
1671 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1672 dirp = FindFirstFile (path, &entp);
1673 free (path);
1674
1675 if (dirp == INVALID_HANDLE_VALUE)
1676 {
1677 dirp = 0;
1678
1679 switch (GetLastError ())
1680 {
1681 case ERROR_FILE_NOT_FOUND:
1682 req->result = 0;
1683 break;
1684
1685 case ERROR_INVALID_NAME:
1686 case ERROR_PATH_NOT_FOUND:
1687 case ERROR_NO_MORE_FILES:
1688 errno = ENOENT;
1689 break;
1690
1691 case ERROR_NOT_ENOUGH_MEMORY:
1692 errno = ENOMEM;
1693 break;
1694
1695 default:
1696 errno = EINVAL;
1697 break;
1698 }
1699 }
1700 }
1701#else
979 self->dirp = dirp = opendir (req->ptr1); 1702 dirp = opendir (req->ptr1);
1703#endif
1704
980 req->flags |= EIO_FLAG_PTR2_FREE; 1705 if (req->flags & EIO_FLAG_PTR1_FREE)
1706 free (req->ptr1);
1707
1708 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1709 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
981 req->ptr2 = names = malloc (memlen); 1710 req->ptr2 = names = malloc (namesalloc);
982 X_UNLOCK (wrklock);
983 1711
984 if (dirp && names) 1712 if (dirp && names && (!flags || dents))
985 for (;;) 1713 for (;;)
986 { 1714 {
1715 int done;
1716
1717#ifdef _WIN32
1718 done = !dirp;
1719#else
987 errno = 0; 1720 errno = 0;
988 entp = readdir (dirp); 1721 entp = readdir (dirp);
1722 done = !entp;
1723#endif
989 1724
990 if (!entp) 1725 if (done)
1726 {
1727#ifndef _WIN32
1728 int old_errno = errno;
1729 closedir (dirp);
1730 errno = old_errno;
1731
1732 if (errno)
1733 break;
1734#endif
1735
1736 /* sort etc. */
1737 req->int1 = flags;
1738 req->result = dentoffs;
1739
1740 if (flags & EIO_READDIR_STAT_ORDER)
1741 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1742 else if (flags & EIO_READDIR_DIRS_FIRST)
1743 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1744 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1745 else
1746 {
1747 /* in this case, all is known, and we just put dirs first and sort them */
1748 eio_dirent *oth = dents + dentoffs;
1749 eio_dirent *dir = dents;
1750
1751 /* now partition dirs to the front, and non-dirs to the back */
1752 /* by walking from both sides and swapping if necessary */
1753 while (oth > dir)
1754 {
1755 if (dir->type == EIO_DT_DIR)
1756 ++dir;
1757 else if ((--oth)->type == EIO_DT_DIR)
1758 {
1759 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1760
1761 ++dir;
1762 }
1763 }
1764
1765 /* now sort the dirs only (dirs all have the same score) */
1766 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1767 }
1768
991 break; 1769 break;
1770 }
992 1771
993 name = entp->d_name; 1772 /* now add the entry to our list(s) */
1773 name = D_NAME (entp);
994 1774
1775 /* skip . and .. entries */
995 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1776 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
996 { 1777 {
997 int len = strlen (name) + 1; 1778 int len = D_NAMLEN (entp) + 1;
998 1779
999 res++; 1780 while (ecb_expect_false (namesoffs + len > namesalloc))
1000
1001 while (memofs + len > memlen)
1002 { 1781 {
1003 memlen *= 2; 1782 namesalloc *= 2;
1004 X_LOCK (wrklock);
1005 req->ptr2 = names = realloc (names, memlen); 1783 req->ptr2 = names = realloc (names, namesalloc);
1006 X_UNLOCK (wrklock);
1007 1784
1008 if (!names) 1785 if (!names)
1009 break; 1786 break;
1010 } 1787 }
1011 1788
1012 memcpy (names + memofs, name, len); 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
1013 memofs += len; 1879 namesoffs += len;
1880 ++dentoffs;
1014 } 1881 }
1882
1883 if (EIO_CANCELLED (req))
1884 {
1885 errno = ECANCELED;
1886 break;
1887 }
1888
1889#ifdef _WIN32
1890 if (!FindNextFile (dirp, &entp))
1891 {
1892 FindClose (dirp);
1893 dirp = 0;
1894 }
1895#endif
1015 } 1896 }
1016
1017 if (errno)
1018 res = -1;
1019
1020 req->result = res;
1021}
1022
1023#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1024# undef msync
1025# define msync(a,b,c) ((errno = ENOSYS), -1)
1026#endif
1027
1028int
1029eio__mtouch (void *mem, size_t len, int flags)
1030{
1031 intptr_t addr = (intptr_t)mem;
1032 intptr_t end = addr + len;
1033#ifdef PAGESIZE
1034 const intptr_t page = PAGESIZE;
1035#else
1036 static intptr_t page;
1037
1038 if (!page)
1039 page = sysconf (_SC_PAGESIZE);
1040#endif
1041
1042 addr &= ~(page - 1); /* assume page size is always a power of two */
1043
1044 if (addr < end)
1045 if (flags) /* modify */
1046 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1047 else
1048 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1049
1050 return 0;
1051} 1897}
1052 1898
1053/*****************************************************************************/ 1899/*****************************************************************************/
1054 1900
1055#define ALLOC(len) \ 1901#define ALLOC(len) \
1071{ 1917{
1072 ETP_REQ *req; 1918 ETP_REQ *req;
1073 struct timespec ts; 1919 struct timespec ts;
1074 etp_worker *self = (etp_worker *)thr_arg; 1920 etp_worker *self = (etp_worker *)thr_arg;
1075 1921
1922#if HAVE_PRCTL_SET_NAME
1923 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1924#endif
1925
1076 /* try to distribute timeouts somewhat randomly */ 1926 /* try to distribute timeouts somewhat evenly */
1077 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1927 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1078 1928
1079 for (;;) 1929 for (;;)
1080 { 1930 {
1931 ts.tv_sec = 0;
1932
1081 X_LOCK (reqlock); 1933 X_LOCK (reqlock);
1082 1934
1083 for (;;) 1935 for (;;)
1084 { 1936 {
1085 self->req = req = reqq_shift (&req_queue); 1937 self->req = req = reqq_shift (&req_queue);
1086 1938
1087 if (req) 1939 if (req)
1088 break; 1940 break;
1089 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
1090 ++idle; 1951 ++idle;
1091 1952
1092 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1953 if (idle <= max_idle)
1093 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1954 /* we are allowed to idle, so do so without any timeout */
1955 X_COND_WAIT (reqwait, reqlock);
1956 else
1094 { 1957 {
1095 if (idle > max_idle) 1958 /* initialise timeout once */
1096 { 1959 if (!ts.tv_sec)
1097 --idle; 1960 ts.tv_sec = time (0) + idle_timeout;
1098 X_UNLOCK (reqlock);
1099 X_LOCK (wrklock);
1100 --started;
1101 X_UNLOCK (wrklock);
1102 goto quit;
1103 }
1104 1961
1105 /* we are allowed to idle, so do so without any timeout */
1106 X_COND_WAIT (reqwait, reqlock); 1962 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1963 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1107 } 1964 }
1108 1965
1109 --idle; 1966 --idle;
1110 } 1967 }
1111 1968
1114 X_UNLOCK (reqlock); 1971 X_UNLOCK (reqlock);
1115 1972
1116 if (req->type < 0) 1973 if (req->type < 0)
1117 goto quit; 1974 goto quit;
1118 1975
1119 if (!EIO_CANCELLED (req))
1120 ETP_EXECUTE (self, req); 1976 ETP_EXECUTE (self, req);
1121 1977
1122 X_LOCK (reslock); 1978 X_LOCK (reslock);
1123 1979
1124 ++npending; 1980 ++npending;
1125 1981
1140 return 0; 1996 return 0;
1141} 1997}
1142 1998
1143/*****************************************************************************/ 1999/*****************************************************************************/
1144 2000
2001int ecb_cold
1145int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2002eio_init (void (*want_poll)(void), void (*done_poll)(void))
1146{ 2003{
2004#if !HAVE_PREADWRITE
2005 X_MUTEX_CREATE (preadwritelock);
2006#endif
2007
1147 return etp_init (want_poll, done_poll); 2008 return etp_init (want_poll, done_poll);
1148} 2009}
1149 2010
2011ecb_inline void
1150static void eio_api_destroy (eio_req *req) 2012eio_api_destroy (eio_req *req)
1151{ 2013{
1152 free (req); 2014 free (req);
1153} 2015}
1154 2016
1155#define REQ(rtype) \ 2017#define REQ(rtype) \
1174 { \ 2036 { \
1175 eio_api_destroy (req); \ 2037 eio_api_destroy (req); \
1176 return 0; \ 2038 return 0; \
1177 } 2039 }
1178 2040
2041static void
1179static void eio_execute (etp_worker *self, eio_req *req) 2042eio_execute (etp_worker *self, eio_req *req)
1180{ 2043{
1181 errno = 0; 2044 if (ecb_expect_false (EIO_CANCELLED (req)))
2045 {
2046 req->result = -1;
2047 req->errorno = ECANCELED;
2048 return;
2049 }
1182 2050
1183 switch (req->type) 2051 switch (req->type)
1184 { 2052 {
1185 case EIO_READ: ALLOC (req->size); 2053 case EIO_READ: ALLOC (req->size);
1186 req->result = req->offs >= 0 2054 req->result = req->offs >= 0
1189 case EIO_WRITE: req->result = req->offs >= 0 2057 case EIO_WRITE: req->result = req->offs >= 0
1190 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2058 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1191 : write (req->int1, req->ptr2, req->size); break; 2059 : write (req->int1, req->ptr2, req->size); break;
1192 2060
1193 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2061 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1194 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2062 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1195 2063
1196 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2064 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2065 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1198 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2066 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1199 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2067 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1200 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2068 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1201 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2069 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2070
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));
2074 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1202 2075
1203 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2076 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1204 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2077 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1205 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2078 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1206 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2079 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1214 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2087 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1215 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2088 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1216 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2089 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1217 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2090 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1218 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2091 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1219 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2092 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1220 2093
2094 case EIO_REALPATH: eio__realpath (req, self); break;
2095
1221 case EIO_READLINK: ALLOC (NAME_MAX); 2096 case EIO_READLINK: ALLOC (PATH_MAX);
1222 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2097 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1223 2098
1224 case EIO_SYNC: req->result = 0; sync (); break; 2099 case EIO_SYNC: req->result = 0; sync (); break;
1225 case EIO_FSYNC: req->result = fsync (req->int1); break; 2100 case EIO_FSYNC: req->result = fsync (req->int1); break;
1226 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2101 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1227 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 2102 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
1228 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1229 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2103 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2104 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2105 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2106 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2107 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2108 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1230 2109
1231 case EIO_READDIR: eio__scandir (req, self); break; 2110 case EIO_READDIR: eio__scandir (req, self); break;
1232 2111
1233 case EIO_BUSY: 2112 case EIO_BUSY:
1234#ifdef _WIN32 2113#ifdef _WIN32
1235 Sleep (req->nv1 * 1000.); 2114 Sleep (req->nv1 * 1e3);
1236#else 2115#else
1237 { 2116 {
1238 struct timeval tv; 2117 struct timeval tv;
1239 2118
1240 tv.tv_sec = req->nv1; 2119 tv.tv_sec = req->nv1;
1241 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2120 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1242 2121
1243 req->result = select (0, 0, 0, 0, &tv); 2122 req->result = select (0, 0, 0, 0, &tv);
1244 } 2123 }
1245#endif 2124#endif
1246 break; 2125 break;
1261 times = tv; 2140 times = tv;
1262 } 2141 }
1263 else 2142 else
1264 times = 0; 2143 times = 0;
1265 2144
1266
1267 req->result = req->type == EIO_FUTIME 2145 req->result = req->type == EIO_FUTIME
1268 ? futimes (req->int1, times) 2146 ? futimes (req->int1, times)
1269 : utimes (req->ptr1, times); 2147 : utimes (req->ptr1, times);
1270 } 2148 }
1271 break; 2149 break;
1276 case EIO_NOP: 2154 case EIO_NOP:
1277 req->result = 0; 2155 req->result = 0;
1278 break; 2156 break;
1279 2157
1280 case EIO_CUSTOM: 2158 case EIO_CUSTOM:
1281 ((void (*)(eio_req *))req->feed) (req); 2159 req->feed (req);
1282 break; 2160 break;
1283 2161
1284 default: 2162 default:
2163 errno = ENOSYS;
1285 req->result = -1; 2164 req->result = -1;
1286 break; 2165 break;
1287 } 2166 }
1288 2167
1289 req->errorno = errno; 2168 req->errorno = errno;
1314eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2193eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1315{ 2194{
1316 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2195 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1317} 2196}
1318 2197
2198eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2199{
2200 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2201}
2202
2203eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2204{
2205 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2206}
2207
2208eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2209{
2210 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2211}
2212
1319eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2213eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1320{ 2214{
1321 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2215 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1322} 2216}
1323 2217
1324eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2218eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1325{ 2219{
1326 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2220 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1327} 2221}
1328 2222
1329eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2223eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1330{ 2224{
1331 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2225 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2226}
2227
2228eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2229{
2230 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1332} 2231}
1333 2232
1334eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2233eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1335{ 2234{
1336 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2235 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1354eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2253eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1355{ 2254{
1356 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2255 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1357} 2256}
1358 2257
2258eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2259{
2260 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2261}
2262
1359eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2263eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1360{ 2264{
1361 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2265 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1362} 2266}
1363 2267
1369eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2273eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1370{ 2274{
1371 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2275 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1372} 2276}
1373 2277
1374eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2278eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1375{ 2279{
1376 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2280 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1377} 2281}
1378 2282
1379eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2283eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1399eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2303eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1400{ 2304{
1401 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2305 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1402} 2306}
1403 2307
1404eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2308eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1405{ 2309{
1406 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2310 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1407} 2311}
1408 2312
1409eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2313eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1425eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2329eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1426{ 2330{
1427 return eio__1path (EIO_READLINK, path, pri, cb, data); 2331 return eio__1path (EIO_READLINK, path, pri, cb, data);
1428} 2332}
1429 2333
2334eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2335{
2336 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2337}
2338
1430eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2339eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1431{ 2340{
1432 return eio__1path (EIO_STAT, path, pri, cb, data); 2341 return eio__1path (EIO_STAT, path, pri, cb, data);
1433} 2342}
1434 2343
1435eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2344eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1436{ 2345{
1437 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2346 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1438} 2347}
1439 2348
2349eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2350{
2351 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2352}
2353
1440eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2354eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1441{ 2355{
1442 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2356 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1443} 2357}
1444 2358
1445eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2359eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1446{ 2360{
1447 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2361 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1448} 2362}
1449 2363
1450eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2364eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1451{ 2365{
1452 return eio__1path (EIO_READDIR, path, pri, cb, data); 2366 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1453} 2367}
1454 2368
1455eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2369eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1456{ 2370{
1457 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2371 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1458} 2372}
1459 2373
1460static eio_req * 2374static eio_req *
1461eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2375eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1462{ 2376{
1486eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2400eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1487{ 2401{
1488 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2402 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1489} 2403}
1490 2404
1491eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2405eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1492{ 2406{
1493 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2407 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1494} 2408}
1495 2409
1496#endif 2410#endif
1497 2411
1498eio_req *eio_grp (eio_cb cb, void *data) 2412eio_req *eio_grp (eio_cb cb, void *data)
1507#undef SEND 2421#undef SEND
1508 2422
1509/*****************************************************************************/ 2423/*****************************************************************************/
1510/* grp functions */ 2424/* grp functions */
1511 2425
2426void
1512void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2427eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1513{ 2428{
1514 grp->int2 = limit; 2429 grp->int2 = limit;
1515 grp->feed = feed; 2430 grp->feed = feed;
1516 2431
1517 grp_try_feed (grp); 2432 grp_try_feed (grp);
1518} 2433}
1519 2434
2435void
1520void eio_grp_limit (eio_req *grp, int limit) 2436eio_grp_limit (eio_req *grp, int limit)
1521{ 2437{
1522 grp->int2 = limit; 2438 grp->int2 = limit;
1523 2439
1524 grp_try_feed (grp); 2440 grp_try_feed (grp);
1525} 2441}
1526 2442
2443void
1527void eio_grp_add (eio_req *grp, eio_req *req) 2444eio_grp_add (eio_req *grp, eio_req *req)
1528{ 2445{
1529 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2446 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1530 2447
1531 grp->flags |= EIO_FLAG_GROUPADD; 2448 grp->flags |= EIO_FLAG_GROUPADD;
1532 2449
1543} 2460}
1544 2461
1545/*****************************************************************************/ 2462/*****************************************************************************/
1546/* misc garbage */ 2463/* misc garbage */
1547 2464
2465eio_ssize_t
1548ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2466eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1549{ 2467{
1550 etp_worker wrk;
1551
1552 wrk.dbuf = 0;
1553
1554 eio__sendfile (ofd, ifd, offset, count, &wrk); 2468 return eio__sendfile (ofd, ifd, offset, count);
1555
1556 if (wrk.dbuf)
1557 free (wrk.dbuf);
1558} 2469}
1559 2470

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines