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

Comparing libeio/eio.c (file contents):
Revision 1.53 by root, Sat Jan 9 10:03:09 2010 UTC vs.
Revision 1.90 by root, Sat Jul 16 16:46:10 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010 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"
49#include <stdlib.h> 54#include <stdlib.h>
50#include <string.h> 55#include <string.h>
51#include <errno.h> 56#include <errno.h>
52#include <sys/types.h> 57#include <sys/types.h>
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <sys/statvfs.h>
55#include <limits.h> 59#include <limits.h>
56#include <fcntl.h> 60#include <fcntl.h>
57#include <assert.h> 61#include <assert.h>
58 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
76static void eio_destroy (eio_req *req);
77
59#ifndef EIO_FINISH 78#ifndef EIO_FINISH
60# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 79# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
61#endif 80#endif
62 81
63#ifndef EIO_DESTROY 82#ifndef EIO_DESTROY
66 85
67#ifndef EIO_FEED 86#ifndef EIO_FEED
68# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 87# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
69#endif 88#endif
70 89
90#ifndef EIO_FD_TO_WIN32_HANDLE
91# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
92#endif
93#ifndef EIO_WIN32_HANDLE_TO_FD
94# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
95#endif
96
97#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
98
99#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
100
71#ifdef _WIN32 101#ifdef _WIN32
72 102
73 /*doh*/ 103 #undef PAGESIZE
104 #define PAGESIZE 4096 /* GetSystemInfo? */
105
106 #ifdef EIO_STRUCT_STATI64
107 #define stat(path,buf) _stati64 (path,buf)
108 #define fstat(fd,buf) _fstati64 (path,buf)
109 #endif
110 #define lstat(path,buf) stat (path,buf)
111 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
112 #define mkdir(path,mode) _mkdir (path)
113 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
114
115 #define chmod(path,mode) _chmod (path, mode)
116 #define fchmod(fd,mode) EIO_ENOSYS ()
117 #define chown(path,uid,gid) EIO_ENOSYS ()
118 #define fchown(fd,uid,gid) EIO_ENOSYS ()
119 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
120 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
121 #define mknod(path,mode,dev) EIO_ENOSYS ()
122 #define sync() EIO_ENOSYS ()
123 #define readlink(path,buf,s) EIO_ENOSYS ()
124 #define statvfs(path,buf) EIO_ENOSYS ()
125 #define fstatvfs(fd,buf) EIO_ENOSYS ()
126
127 /* we could even stat and see if it exists */
128 static int
129 symlink (const char *old, const char *neu)
130 {
131 #if WINVER >= 0x0600
132 if (CreateSymbolicLink (neu, old, 1))
133 return 0;
134
135 if (CreateSymbolicLink (neu, old, 0))
136 return 0;
137 #endif
138
139 return EIO_ERRNO (ENOENT, -1);
140 }
141
142 /* POSIX API only */
143 #define CreateHardLink(neu,old) 0
144 #define CreateSymbolicLink(neu,old,flags) 0
145
146 struct statvfs
147 {
148 int dummy;
149 };
150
74#else 151#else
75 152
76# include "config.h"
77# include <sys/time.h> 153 #include <sys/time.h>
78# include <sys/select.h> 154 #include <sys/select.h>
79# include <sys/mman.h> 155 #include <sys/statvfs.h>
80# include <unistd.h> 156 #include <unistd.h>
81# include <utime.h> 157 #include <utime.h>
82# include <signal.h> 158 #include <signal.h>
83# include <dirent.h> 159 #include <dirent.h>
84 160
161 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
162 #include <sys/mman.h>
163 #endif
164
85/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 165 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
86# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 166 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
87# define _DIRENT_HAVE_D_TYPE /* sigh */ 167 #define _DIRENT_HAVE_D_TYPE /* sigh */
88# define D_INO(de) (de)->d_fileno 168 #define D_INO(de) (de)->d_fileno
89# define D_NAMLEN(de) (de)->d_namlen 169 #define D_NAMLEN(de) (de)->d_namlen
90# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 170 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
91# define D_INO(de) (de)->d_ino 171 #define D_INO(de) (de)->d_ino
92# endif 172 #endif
93 173
94#ifdef _D_EXACT_NAMLEN 174 #ifdef _D_EXACT_NAMLEN
95# undef D_NAMLEN 175 #undef D_NAMLEN
96# define D_NAMLEN(de) _D_EXACT_NAMLEN (de) 176 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
97#endif 177 #endif
98 178
99# ifdef _DIRENT_HAVE_D_TYPE 179 #ifdef _DIRENT_HAVE_D_TYPE
100# define D_TYPE(de) (de)->d_type 180 #define D_TYPE(de) (de)->d_type
101# endif 181 #endif
102 182
103# ifndef EIO_STRUCT_DIRENT 183 #ifndef EIO_STRUCT_DIRENT
104# define EIO_STRUCT_DIRENT struct dirent 184 #define EIO_STRUCT_DIRENT struct dirent
105# endif 185 #endif
106 186
107#endif 187#endif
108 188
109#if HAVE_SENDFILE 189#if HAVE_SENDFILE
110# if __linux 190# if __linux
129#endif 209#endif
130#ifndef D_NAMLEN 210#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name) 211# define D_NAMLEN(de) strlen ((de)->d_name)
132#endif 212#endif
133 213
134/* number of seconds after which an idle threads exit */
135#define IDLE_TIMEOUT 10
136
137/* used for struct dirent, AIX doesn't provide it */ 214/* used for struct dirent, AIX doesn't provide it */
138#ifndef NAME_MAX 215#ifndef NAME_MAX
139# define NAME_MAX 4096 216# define NAME_MAX 4096
217#endif
218
219/* used for readlink etc. */
220#ifndef PATH_MAX
221# define PATH_MAX 4096
140#endif 222#endif
141 223
142/* buffer size for various temporary buffers */ 224/* buffer size for various temporary buffers */
143#define EIO_BUFSIZE 65536 225#define EIO_BUFSIZE 65536
144 226
150 errno = ENOMEM; \ 232 errno = ENOMEM; \
151 if (!eio_buf) \ 233 if (!eio_buf) \
152 return -1; 234 return -1;
153 235
154#define EIO_TICKS ((1000000 + 1023) >> 10) 236#define EIO_TICKS ((1000000 + 1023) >> 10)
155
156/*****************************************************************************/
157
158#if __GNUC__ >= 3
159# define expect(expr,value) __builtin_expect ((expr),(value))
160#else
161# define expect(expr,value) (expr)
162#endif
163
164#define expect_false(expr) expect ((expr) != 0, 0)
165#define expect_true(expr) expect ((expr) != 0, 1)
166
167/*****************************************************************************/
168 237
169#define ETP_PRI_MIN EIO_PRI_MIN 238#define ETP_PRI_MIN EIO_PRI_MIN
170#define ETP_PRI_MAX EIO_PRI_MAX 239#define ETP_PRI_MAX EIO_PRI_MAX
171 240
172struct etp_worker; 241struct etp_worker;
198/*****************************************************************************/ 267/*****************************************************************************/
199 268
200#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 269#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
201 270
202/* calculate time difference in ~1/EIO_TICKS of a second */ 271/* calculate time difference in ~1/EIO_TICKS of a second */
272ecb_inline int
203static int tvdiff (struct timeval *tv1, struct timeval *tv2) 273tvdiff (struct timeval *tv1, struct timeval *tv2)
204{ 274{
205 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 275 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
206 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 276 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
207} 277}
208 278
215static unsigned int max_poll_reqs; /* reslock */ 285static unsigned int max_poll_reqs; /* reslock */
216 286
217static volatile unsigned int nreqs; /* reqlock */ 287static volatile unsigned int nreqs; /* reqlock */
218static volatile unsigned int nready; /* reqlock */ 288static volatile unsigned int nready; /* reqlock */
219static volatile unsigned int npending; /* reqlock */ 289static volatile unsigned int npending; /* reqlock */
220static volatile unsigned int max_idle = 4; 290static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
291static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
221 292
222static mutex_t wrklock = X_MUTEX_INIT; 293static xmutex_t wrklock;
223static mutex_t reslock = X_MUTEX_INIT; 294static xmutex_t reslock;
224static mutex_t reqlock = X_MUTEX_INIT; 295static xmutex_t reqlock;
225static cond_t reqwait = X_COND_INIT; 296static xcond_t reqwait;
226 297
227#if !HAVE_PREADWRITE 298#if !HAVE_PREADWRITE
228/* 299/*
229 * make our pread/pwrite emulation safe against themselves, but not against 300 * make our pread/pwrite emulation safe against themselves, but not against
230 * normal read/write by using a mutex. slows down execution a lot, 301 * normal read/write by using a mutex. slows down execution a lot,
231 * but that's your problem, not mine. 302 * but that's your problem, not mine.
232 */ 303 */
233static mutex_t preadwritelock = X_MUTEX_INIT; 304static xmutex_t preadwritelock = X_MUTEX_INIT;
234#endif 305#endif
235 306
236typedef struct etp_worker 307typedef struct etp_worker
237{ 308{
238 /* locked by wrklock */ 309 /* locked by wrklock */
239 struct etp_worker *prev, *next; 310 struct etp_worker *prev, *next;
240 311
241 thread_t tid; 312 xthread_t tid;
242 313
243 /* locked by reslock, reqlock or wrklock */ 314 /* locked by reslock, reqlock or wrklock */
244 ETP_REQ *req; /* currently processed request */ 315 ETP_REQ *req; /* currently processed request */
245 316
246 ETP_WORKER_COMMON 317 ETP_WORKER_COMMON
251#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 322#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
252#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 323#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
253 324
254/* worker threads management */ 325/* worker threads management */
255 326
327static void ecb_cold
256static void etp_worker_clear (etp_worker *wrk) 328etp_worker_clear (etp_worker *wrk)
257{ 329{
258 ETP_WORKER_CLEAR (wrk); 330 ETP_WORKER_CLEAR (wrk);
259} 331}
260 332
333static void ecb_cold
261static void etp_worker_free (etp_worker *wrk) 334etp_worker_free (etp_worker *wrk)
262{ 335{
263 wrk->next->prev = wrk->prev; 336 wrk->next->prev = wrk->prev;
264 wrk->prev->next = wrk->next; 337 wrk->prev->next = wrk->next;
265 338
266 free (wrk); 339 free (wrk);
267} 340}
268 341
269static unsigned int etp_nreqs (void) 342static unsigned int
343etp_nreqs (void)
270{ 344{
271 int retval; 345 int retval;
272 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 346 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
273 retval = nreqs; 347 retval = nreqs;
274 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 348 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
275 return retval; 349 return retval;
276} 350}
277 351
278static unsigned int etp_nready (void) 352static unsigned int
353etp_nready (void)
279{ 354{
280 unsigned int retval; 355 unsigned int retval;
281 356
282 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 357 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
283 retval = nready; 358 retval = nready;
284 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 359 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
285 360
286 return retval; 361 return retval;
287} 362}
288 363
289static unsigned int etp_npending (void) 364static unsigned int
365etp_npending (void)
290{ 366{
291 unsigned int retval; 367 unsigned int retval;
292 368
293 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 369 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
294 retval = npending; 370 retval = npending;
295 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 371 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
296 372
297 return retval; 373 return retval;
298} 374}
299 375
300static unsigned int etp_nthreads (void) 376static unsigned int
377etp_nthreads (void)
301{ 378{
302 unsigned int retval; 379 unsigned int retval;
303 380
304 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 381 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
305 retval = started; 382 retval = started;
319} etp_reqq; 396} etp_reqq;
320 397
321static etp_reqq req_queue; 398static etp_reqq req_queue;
322static etp_reqq res_queue; 399static etp_reqq res_queue;
323 400
401static int ecb_noinline
324static int reqq_push (etp_reqq *q, ETP_REQ *req) 402reqq_push (etp_reqq *q, ETP_REQ *req)
325{ 403{
326 int pri = req->pri; 404 int pri = req->pri;
327 req->next = 0; 405 req->next = 0;
328 406
329 if (q->qe[pri]) 407 if (q->qe[pri])
335 q->qe[pri] = q->qs[pri] = req; 413 q->qe[pri] = q->qs[pri] = req;
336 414
337 return q->size++; 415 return q->size++;
338} 416}
339 417
418static ETP_REQ * ecb_noinline
340static ETP_REQ *reqq_shift (etp_reqq *q) 419reqq_shift (etp_reqq *q)
341{ 420{
342 int pri; 421 int pri;
343 422
344 if (!q->size) 423 if (!q->size)
345 return 0; 424 return 0;
360 } 439 }
361 440
362 abort (); 441 abort ();
363} 442}
364 443
365static void etp_atfork_prepare (void) 444static void ecb_cold
445etp_thread_init (void)
366{ 446{
367 X_LOCK (wrklock);
368 X_LOCK (reqlock);
369 X_LOCK (reslock);
370#if !HAVE_PREADWRITE 447#if !HAVE_PREADWRITE
371 X_LOCK (preadwritelock); 448 X_MUTEX_CREATE (preadwritelock);
372#endif 449#endif
450 X_MUTEX_CREATE (wrklock);
451 X_MUTEX_CREATE (reslock);
452 X_MUTEX_CREATE (reqlock);
453 X_COND_CREATE (reqwait);
373} 454}
374 455
456static void ecb_cold
457etp_atfork_prepare (void)
458{
459}
460
461static void ecb_cold
375static void etp_atfork_parent (void) 462etp_atfork_parent (void)
376{ 463{
377#if !HAVE_PREADWRITE
378 X_UNLOCK (preadwritelock);
379#endif
380 X_UNLOCK (reslock);
381 X_UNLOCK (reqlock);
382 X_UNLOCK (wrklock);
383} 464}
384 465
466static void ecb_cold
385static void etp_atfork_child (void) 467etp_atfork_child (void)
386{ 468{
387 ETP_REQ *prv; 469 ETP_REQ *prv;
388 470
389 while ((prv = reqq_shift (&req_queue))) 471 while ((prv = reqq_shift (&req_queue)))
390 ETP_DESTROY (prv); 472 ETP_DESTROY (prv);
407 idle = 0; 489 idle = 0;
408 nreqs = 0; 490 nreqs = 0;
409 nready = 0; 491 nready = 0;
410 npending = 0; 492 npending = 0;
411 493
412 etp_atfork_parent (); 494 etp_thread_init ();
413} 495}
414 496
415static void 497static void ecb_cold
416etp_once_init (void) 498etp_once_init (void)
417{ 499{
500 etp_thread_init ();
418 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 501 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
419} 502}
420 503
421static int 504static int ecb_cold
422etp_init (void (*want_poll)(void), void (*done_poll)(void)) 505etp_init (void (*want_poll)(void), void (*done_poll)(void))
423{ 506{
424 static pthread_once_t doinit = PTHREAD_ONCE_INIT; 507 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
425 508
426 pthread_once (&doinit, etp_once_init); 509 pthread_once (&doinit, etp_once_init);
431 return 0; 514 return 0;
432} 515}
433 516
434X_THREAD_PROC (etp_proc); 517X_THREAD_PROC (etp_proc);
435 518
519static void ecb_cold
436static void etp_start_thread (void) 520etp_start_thread (void)
437{ 521{
438 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 522 etp_worker *wrk = calloc (1, sizeof (etp_worker));
439 523
440 /*TODO*/ 524 /*TODO*/
441 assert (("unable to allocate worker thread data", wrk)); 525 assert (("unable to allocate worker thread data", wrk));
454 free (wrk); 538 free (wrk);
455 539
456 X_UNLOCK (wrklock); 540 X_UNLOCK (wrklock);
457} 541}
458 542
543static void
459static void etp_maybe_start_thread (void) 544etp_maybe_start_thread (void)
460{ 545{
461 if (expect_true (etp_nthreads () >= wanted)) 546 if (ecb_expect_true (etp_nthreads () >= wanted))
462 return; 547 return;
463 548
464 /* todo: maybe use idle here, but might be less exact */ 549 /* todo: maybe use idle here, but might be less exact */
465 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 550 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
466 return; 551 return;
467 552
468 etp_start_thread (); 553 etp_start_thread ();
469} 554}
470 555
556static void ecb_cold
471static void etp_end_thread (void) 557etp_end_thread (void)
472{ 558{
473 eio_req *req = calloc (1, sizeof (eio_req)); 559 eio_req *req = calloc (1, sizeof (eio_req));
474 560
475 req->type = -1; 561 req->type = -1;
476 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 562 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
483 X_LOCK (wrklock); 569 X_LOCK (wrklock);
484 --started; 570 --started;
485 X_UNLOCK (wrklock); 571 X_UNLOCK (wrklock);
486} 572}
487 573
488static int etp_poll (void) 574static int
575etp_poll (void)
489{ 576{
490 unsigned int maxreqs; 577 unsigned int maxreqs;
491 unsigned int maxtime; 578 unsigned int maxtime;
492 struct timeval tv_start, tv_now; 579 struct timeval tv_start, tv_now;
493 580
523 610
524 X_LOCK (reqlock); 611 X_LOCK (reqlock);
525 --nreqs; 612 --nreqs;
526 X_UNLOCK (reqlock); 613 X_UNLOCK (reqlock);
527 614
528 if (expect_false (req->type == EIO_GROUP && req->size)) 615 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
529 { 616 {
530 req->int1 = 1; /* mark request as delayed */ 617 req->int1 = 1; /* mark request as delayed */
531 continue; 618 continue;
532 } 619 }
533 else 620 else
534 { 621 {
535 int res = ETP_FINISH (req); 622 int res = ETP_FINISH (req);
536 if (expect_false (res)) 623 if (ecb_expect_false (res))
537 return res; 624 return res;
538 } 625 }
539 626
540 if (expect_false (maxreqs && !--maxreqs)) 627 if (ecb_expect_false (maxreqs && !--maxreqs))
541 break; 628 break;
542 629
543 if (maxtime) 630 if (maxtime)
544 { 631 {
545 gettimeofday (&tv_now, 0); 632 gettimeofday (&tv_now, 0);
551 638
552 errno = EAGAIN; 639 errno = EAGAIN;
553 return -1; 640 return -1;
554} 641}
555 642
643static void
556static void etp_cancel (ETP_REQ *req) 644etp_cancel (ETP_REQ *req)
557{ 645{
558 X_LOCK (wrklock); 646 req->cancelled = 1;
559 req->flags |= EIO_FLAG_CANCELLED;
560 X_UNLOCK (wrklock);
561 647
562 eio_grp_cancel (req); 648 eio_grp_cancel (req);
563} 649}
564 650
651static void
565static void etp_submit (ETP_REQ *req) 652etp_submit (ETP_REQ *req)
566{ 653{
567 req->pri -= ETP_PRI_MIN; 654 req->pri -= ETP_PRI_MIN;
568 655
569 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 656 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
570 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 657 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
571 658
572 if (expect_false (req->type == EIO_GROUP)) 659 if (ecb_expect_false (req->type == EIO_GROUP))
573 { 660 {
574 /* I hope this is worth it :/ */ 661 /* I hope this is worth it :/ */
575 X_LOCK (reqlock); 662 X_LOCK (reqlock);
576 ++nreqs; 663 ++nreqs;
577 X_UNLOCK (reqlock); 664 X_UNLOCK (reqlock);
596 683
597 etp_maybe_start_thread (); 684 etp_maybe_start_thread ();
598 } 685 }
599} 686}
600 687
688static void ecb_cold
601static void etp_set_max_poll_time (double nseconds) 689etp_set_max_poll_time (double nseconds)
602{ 690{
603 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 691 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
604 max_poll_time = nseconds * EIO_TICKS; 692 max_poll_time = nseconds * EIO_TICKS;
605 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 693 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
606} 694}
607 695
696static void ecb_cold
608static void etp_set_max_poll_reqs (unsigned int maxreqs) 697etp_set_max_poll_reqs (unsigned int maxreqs)
609{ 698{
610 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 699 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
611 max_poll_reqs = maxreqs; 700 max_poll_reqs = maxreqs;
612 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 701 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
613} 702}
614 703
704static void ecb_cold
615static void etp_set_max_idle (unsigned int nthreads) 705etp_set_max_idle (unsigned int nthreads)
616{ 706{
617 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 707 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
618 max_idle = nthreads <= 0 ? 1 : nthreads; 708 max_idle = nthreads;
619 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 709 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
620} 710}
621 711
712static void ecb_cold
713etp_set_idle_timeout (unsigned int seconds)
714{
715 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
716 idle_timeout = seconds;
717 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
718}
719
720static void ecb_cold
622static void etp_set_min_parallel (unsigned int nthreads) 721etp_set_min_parallel (unsigned int nthreads)
623{ 722{
624 if (wanted < nthreads) 723 if (wanted < nthreads)
625 wanted = nthreads; 724 wanted = nthreads;
626} 725}
627 726
727static void ecb_cold
628static void etp_set_max_parallel (unsigned int nthreads) 728etp_set_max_parallel (unsigned int nthreads)
629{ 729{
630 if (wanted > nthreads) 730 if (wanted > nthreads)
631 wanted = nthreads; 731 wanted = nthreads;
632 732
633 while (started > wanted) 733 while (started > wanted)
634 etp_end_thread (); 734 etp_end_thread ();
635} 735}
636 736
637/*****************************************************************************/ 737/*****************************************************************************/
638 738
739static void
639static void grp_try_feed (eio_req *grp) 740grp_try_feed (eio_req *grp)
640{ 741{
641 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 742 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
642 { 743 {
643 grp->flags &= ~EIO_FLAG_GROUPADD; 744 grp->flags &= ~EIO_FLAG_GROUPADD;
644 745
651 break; 752 break;
652 } 753 }
653 } 754 }
654} 755}
655 756
757static int
656static int grp_dec (eio_req *grp) 758grp_dec (eio_req *grp)
657{ 759{
658 --grp->size; 760 --grp->size;
659 761
660 /* call feeder, if applicable */ 762 /* call feeder, if applicable */
661 grp_try_feed (grp); 763 grp_try_feed (grp);
665 return eio_finish (grp); 767 return eio_finish (grp);
666 else 768 else
667 return 0; 769 return 0;
668} 770}
669 771
772static void
670void eio_destroy (eio_req *req) 773eio_destroy (eio_req *req)
671{ 774{
672 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 775 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
673 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 776 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
674 777
675 EIO_DESTROY (req); 778 EIO_DESTROY (req);
676} 779}
677 780
781static int
678static int eio_finish (eio_req *req) 782eio_finish (eio_req *req)
679{ 783{
680 int res = EIO_FINISH (req); 784 int res = EIO_FINISH (req);
681 785
682 if (req->grp) 786 if (req->grp)
683 { 787 {
691 if (grp->grp_first == req) 795 if (grp->grp_first == req)
692 grp->grp_first = req->grp_next; 796 grp->grp_first = req->grp_next;
693 797
694 res2 = grp_dec (grp); 798 res2 = grp_dec (grp);
695 799
696 if (!res && res2) 800 if (!res)
697 res = res2; 801 res = res2;
698 } 802 }
699 803
700 eio_destroy (req); 804 eio_destroy (req);
701 805
702 return res; 806 return res;
703} 807}
704 808
809void
705void eio_grp_cancel (eio_req *grp) 810eio_grp_cancel (eio_req *grp)
706{ 811{
707 for (grp = grp->grp_first; grp; grp = grp->grp_next) 812 for (grp = grp->grp_first; grp; grp = grp->grp_next)
708 eio_cancel (grp); 813 eio_cancel (grp);
709} 814}
710 815
816void
711void eio_cancel (eio_req *req) 817eio_cancel (eio_req *req)
712{ 818{
713 etp_cancel (req); 819 etp_cancel (req);
714} 820}
715 821
822void
716void eio_submit (eio_req *req) 823eio_submit (eio_req *req)
717{ 824{
718 etp_submit (req); 825 etp_submit (req);
719} 826}
720 827
721unsigned int eio_nreqs (void) 828unsigned int
829eio_nreqs (void)
722{ 830{
723 return etp_nreqs (); 831 return etp_nreqs ();
724} 832}
725 833
726unsigned int eio_nready (void) 834unsigned int
835eio_nready (void)
727{ 836{
728 return etp_nready (); 837 return etp_nready ();
729} 838}
730 839
731unsigned int eio_npending (void) 840unsigned int
841eio_npending (void)
732{ 842{
733 return etp_npending (); 843 return etp_npending ();
734} 844}
735 845
736unsigned int eio_nthreads (void) 846unsigned int ecb_cold
847eio_nthreads (void)
737{ 848{
738 return etp_nthreads (); 849 return etp_nthreads ();
739} 850}
740 851
852void ecb_cold
741void eio_set_max_poll_time (double nseconds) 853eio_set_max_poll_time (double nseconds)
742{ 854{
743 etp_set_max_poll_time (nseconds); 855 etp_set_max_poll_time (nseconds);
744} 856}
745 857
858void ecb_cold
746void eio_set_max_poll_reqs (unsigned int maxreqs) 859eio_set_max_poll_reqs (unsigned int maxreqs)
747{ 860{
748 etp_set_max_poll_reqs (maxreqs); 861 etp_set_max_poll_reqs (maxreqs);
749} 862}
750 863
864void ecb_cold
751void eio_set_max_idle (unsigned int nthreads) 865eio_set_max_idle (unsigned int nthreads)
752{ 866{
753 etp_set_max_idle (nthreads); 867 etp_set_max_idle (nthreads);
754} 868}
755 869
870void ecb_cold
871eio_set_idle_timeout (unsigned int seconds)
872{
873 etp_set_idle_timeout (seconds);
874}
875
876void ecb_cold
756void eio_set_min_parallel (unsigned int nthreads) 877eio_set_min_parallel (unsigned int nthreads)
757{ 878{
758 etp_set_min_parallel (nthreads); 879 etp_set_min_parallel (nthreads);
759} 880}
760 881
882void ecb_cold
761void eio_set_max_parallel (unsigned int nthreads) 883eio_set_max_parallel (unsigned int nthreads)
762{ 884{
763 etp_set_max_parallel (nthreads); 885 etp_set_max_parallel (nthreads);
764} 886}
765 887
766int eio_poll (void) 888int eio_poll (void)
775# undef pread 897# undef pread
776# undef pwrite 898# undef pwrite
777# define pread eio__pread 899# define pread eio__pread
778# define pwrite eio__pwrite 900# define pwrite eio__pwrite
779 901
780static ssize_t 902static eio_ssize_t
781eio__pread (int fd, void *buf, size_t count, off_t offset) 903eio__pread (int fd, void *buf, size_t count, off_t offset)
782{ 904{
783 ssize_t res; 905 eio_ssize_t res;
784 off_t ooffset; 906 off_t ooffset;
785 907
786 X_LOCK (preadwritelock); 908 X_LOCK (preadwritelock);
787 ooffset = lseek (fd, 0, SEEK_CUR); 909 ooffset = lseek (fd, 0, SEEK_CUR);
788 lseek (fd, offset, SEEK_SET); 910 lseek (fd, offset, SEEK_SET);
791 X_UNLOCK (preadwritelock); 913 X_UNLOCK (preadwritelock);
792 914
793 return res; 915 return res;
794} 916}
795 917
796static ssize_t 918static eio_ssize_t
797eio__pwrite (int fd, void *buf, size_t count, off_t offset) 919eio__pwrite (int fd, void *buf, size_t count, off_t offset)
798{ 920{
799 ssize_t res; 921 eio_ssize_t res;
800 off_t ooffset; 922 off_t ooffset;
801 923
802 X_LOCK (preadwritelock); 924 X_LOCK (preadwritelock);
803 ooffset = lseek (fd, 0, SEEK_CUR); 925 ooffset = lseek (fd, 0, SEEK_CUR);
804 lseek (fd, offset, SEEK_SET); 926 lseek (fd, offset, SEEK_SET);
808 930
809 return res; 931 return res;
810} 932}
811#endif 933#endif
812 934
813#ifndef HAVE_FUTIMES 935#ifndef HAVE_UTIMES
814 936
815# undef utimes 937# undef utimes
816# undef futimes
817# define utimes(path,times) eio__utimes (path, times) 938# define utimes(path,times) eio__utimes (path, times)
818# define futimes(fd,times) eio__futimes (fd, times)
819 939
820static int 940static int
821eio__utimes (const char *filename, const struct timeval times[2]) 941eio__utimes (const char *filename, const struct timeval times[2])
822{ 942{
823 if (times) 943 if (times)
831 } 951 }
832 else 952 else
833 return utime (filename, 0); 953 return utime (filename, 0);
834} 954}
835 955
956#endif
957
958#ifndef HAVE_FUTIMES
959
960# undef futimes
961# define futimes(fd,times) eio__futimes (fd, times)
962
963static int
836static int eio__futimes (int fd, const struct timeval tv[2]) 964eio__futimes (int fd, const struct timeval tv[2])
837{ 965{
838 errno = ENOSYS; 966 errno = ENOSYS;
839 return -1; 967 return -1;
840} 968}
841 969
845# undef fdatasync 973# undef fdatasync
846# define fdatasync(fd) fsync (fd) 974# define fdatasync(fd) fsync (fd)
847#endif 975#endif
848 976
849/* sync_file_range always needs emulation */ 977/* sync_file_range always needs emulation */
850int 978static int
851eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 979eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
852{ 980{
853#if HAVE_SYNC_FILE_RANGE 981#if HAVE_SYNC_FILE_RANGE
854 int res; 982 int res;
855 983
872 /* even though we could play tricks with the flags, it's better to always 1000 /* even though we could play tricks with the flags, it's better to always
873 * call fdatasync, as that matches the expectation of its users best */ 1001 * call fdatasync, as that matches the expectation of its users best */
874 return fdatasync (fd); 1002 return fdatasync (fd);
875} 1003}
876 1004
1005static int
1006eio__fallocate (int fd, int mode, off_t offset, size_t len)
1007{
1008#if HAVE_FALLOCATE
1009 return fallocate (fd, mode, offset, len);
1010#else
1011 errno = ENOSYS;
1012 return -1;
1013#endif
1014}
1015
877#if !HAVE_READAHEAD 1016#if !HAVE_READAHEAD
878# undef readahead 1017# undef readahead
879# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1018# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
880 1019
881static ssize_t 1020static eio_ssize_t
882eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1021eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
883{ 1022{
884 size_t todo = count; 1023 size_t todo = count;
885 dBUF; 1024 dBUF;
886 1025
898} 1037}
899 1038
900#endif 1039#endif
901 1040
902/* sendfile always needs emulation */ 1041/* sendfile always needs emulation */
903static ssize_t 1042static eio_ssize_t
904eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1043eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
905{ 1044{
1045 eio_ssize_t written = 0;
906 ssize_t res; 1046 eio_ssize_t res;
907 1047
908 if (!count) 1048 if (!count)
909 return 0; 1049 return 0;
910 1050
1051 for (;;)
1052 {
1053#ifdef __APPLE__
1054# undef HAVE_SENDFILE /* broken, as everything on os x */
1055#endif
911#if HAVE_SENDFILE 1056#if HAVE_SENDFILE
912# if __linux 1057# if __linux
1058 off_t soffset = offset;
913 res = sendfile (ofd, ifd, &offset, count); 1059 res = sendfile (ofd, ifd, &soffset, count);
914 1060
915# elif __FreeBSD__ 1061# elif __FreeBSD__
916 /* 1062 /*
917 * Of course, the freebsd sendfile is a dire hack with no thoughts 1063 * Of course, the freebsd sendfile is a dire hack with no thoughts
918 * wasted on making it similar to other I/O functions. 1064 * wasted on making it similar to other I/O functions.
919 */ 1065 */
920 {
921 off_t sbytes; 1066 off_t sbytes;
922 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1067 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
923 1068
924 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1069 #if 0 /* according to the manpage, this is correct, but broken behaviour */
925 /* freebsd' sendfile will return 0 on success */ 1070 /* freebsd' sendfile will return 0 on success */
926 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1071 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
927 /* not on e.g. EIO or EPIPE - sounds broken */ 1072 /* not on e.g. EIO or EPIPE - sounds broken */
928 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1073 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
929 res = sbytes; 1074 res = sbytes;
930 #endif 1075 #endif
931 1076
932 /* according to source inspection, this is correct, and useful behaviour */ 1077 /* according to source inspection, this is correct, and useful behaviour */
933 if (sbytes) 1078 if (sbytes)
934 res = sbytes; 1079 res = sbytes;
935 }
936 1080
937# elif defined (__APPLE__) 1081# elif defined (__APPLE__)
938
939 {
940 off_t sbytes = count; 1082 off_t sbytes = count;
941 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1083 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
942 1084
943 /* according to the manpage, sbytes is always valid */ 1085 /* according to the manpage, sbytes is always valid */
944 if (sbytes) 1086 if (sbytes)
945 res = sbytes; 1087 res = sbytes;
946 }
947 1088
948# elif __hpux 1089# elif __hpux
949 res = sendfile (ofd, ifd, offset, count, 0, 0); 1090 res = sendfile (ofd, ifd, offset, count, 0, 0);
950 1091
951# elif __solaris 1092# elif __solaris
952 {
953 struct sendfilevec vec; 1093 struct sendfilevec vec;
954 size_t sbytes; 1094 size_t sbytes;
955 1095
956 vec.sfv_fd = ifd; 1096 vec.sfv_fd = ifd;
957 vec.sfv_flag = 0; 1097 vec.sfv_flag = 0;
958 vec.sfv_off = offset; 1098 vec.sfv_off = offset;
959 vec.sfv_len = count; 1099 vec.sfv_len = count;
960 1100
961 res = sendfilev (ofd, &vec, 1, &sbytes); 1101 res = sendfilev (ofd, &vec, 1, &sbytes);
962 1102
963 if (res < 0 && sbytes) 1103 if (res < 0 && sbytes)
964 res = sbytes; 1104 res = sbytes;
965 }
966 1105
967# endif 1106# endif
968 1107
969#elif defined (_WIN32) 1108#elif defined (_WIN32) && 0
970
971 /* does not work, just for documentation of what would need to be done */ 1109 /* does not work, just for documentation of what would need to be done */
972 { 1110 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1111 /* libeio guarantees that the file offset does not change, and windows */
1112 /* has no way to get an independent handle to the same file description */
973 HANDLE h = TO_SOCKET (ifd); 1113 HANDLE h = TO_SOCKET (ifd);
974 SetFilePointer (h, offset, 0, FILE_BEGIN); 1114 SetFilePointer (h, offset, 0, FILE_BEGIN);
975 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1115 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
976 }
977 1116
978#else 1117#else
979 res = -1; 1118 res = -1;
980 errno = ENOSYS; 1119 errno = ENOSYS;
981#endif 1120#endif
982 1121
1122 /* we assume sendfile can copy at least 128mb in one go */
1123 if (res <= 128 * 1024 * 1024)
1124 {
1125 if (res > 0)
1126 written += res;
1127
1128 if (written)
1129 return written;
1130
1131 break;
1132 }
1133 else
1134 {
1135 /* if we requested more, then probably the kernel was lazy */
1136 written += res;
1137 offset += res;
1138 count -= res;
1139
1140 if (!count)
1141 return written;
1142 }
1143 }
1144
983 if (res < 0 1145 if (res < 0
984 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1146 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
985 /* BSDs */ 1147 /* BSDs */
986#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1148#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
987 || errno == ENOTSUP 1149 || errno == ENOTSUP
988#endif 1150#endif
1151#ifdef EOPNOTSUPP /* windows */
989 || errno == EOPNOTSUPP /* BSDs */ 1152 || errno == EOPNOTSUPP /* BSDs */
1153#endif
990#if __solaris 1154#if __solaris
991 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1155 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
992#endif 1156#endif
993 ) 1157 )
994 ) 1158 )
998 1162
999 res = 0; 1163 res = 0;
1000 1164
1001 while (count) 1165 while (count)
1002 { 1166 {
1003 ssize_t cnt; 1167 eio_ssize_t cnt;
1004 1168
1005 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1169 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1006 1170
1007 if (cnt <= 0) 1171 if (cnt <= 0)
1008 { 1172 {
1025 } 1189 }
1026 1190
1027 return res; 1191 return res;
1028} 1192}
1029 1193
1194#ifdef PAGESIZE
1195# define eio_pagesize() PAGESIZE
1196#else
1197static intptr_t
1198eio_pagesize (void)
1199{
1200 static intptr_t page;
1201
1202 if (!page)
1203 page = sysconf (_SC_PAGESIZE);
1204
1205 return page;
1206}
1207#endif
1208
1209static void
1210eio_page_align (void **addr, size_t *length)
1211{
1212 intptr_t mask = eio_pagesize () - 1;
1213
1214 /* round down addr */
1215 intptr_t adj = mask & (intptr_t)*addr;
1216
1217 *addr = (void *)((intptr_t)*addr - adj);
1218 *length += adj;
1219
1220 /* round up length */
1221 *length = (*length + mask) & ~mask;
1222}
1223
1224#if !_POSIX_MEMLOCK
1225# define eio__mlockall(a) EIO_ENOSYS ()
1226#else
1227
1228static int
1229eio__mlockall (int flags)
1230{
1231 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1232 extern int mallopt (int, int);
1233 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1234 #endif
1235
1236 if (EIO_MCL_CURRENT != MCL_CURRENT
1237 || EIO_MCL_FUTURE != MCL_FUTURE)
1238 {
1239 flags = 0
1240 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1241 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1242 }
1243
1244 return mlockall (flags);
1245}
1246#endif
1247
1248#if !_POSIX_MEMLOCK_RANGE
1249# define eio__mlock(a,b) EIO_ENOSYS ()
1250#else
1251
1252static int
1253eio__mlock (void *addr, size_t length)
1254{
1255 eio_page_align (&addr, &length);
1256
1257 return mlock (addr, length);
1258}
1259
1260#endif
1261
1262#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1263# define eio__msync(a,b,c) EIO_ENOSYS ()
1264#else
1265
1266static int
1267eio__msync (void *mem, size_t len, int flags)
1268{
1269 eio_page_align (&mem, &len);
1270
1271 if (EIO_MS_ASYNC != MS_SYNC
1272 || EIO_MS_INVALIDATE != MS_INVALIDATE
1273 || EIO_MS_SYNC != MS_SYNC)
1274 {
1275 flags = 0
1276 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1277 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1278 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1279 }
1280
1281 return msync (mem, len, flags);
1282}
1283
1284#endif
1285
1286static int
1287eio__mtouch (eio_req *req)
1288{
1289 void *mem = req->ptr2;
1290 size_t len = req->size;
1291 int flags = req->int1;
1292
1293 eio_page_align (&mem, &len);
1294
1295 {
1296 intptr_t addr = (intptr_t)mem;
1297 intptr_t end = addr + len;
1298 intptr_t page = eio_pagesize ();
1299
1300 if (addr < end)
1301 if (flags & EIO_MT_MODIFY) /* modify */
1302 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1303 else
1304 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1305 }
1306
1307 return 0;
1308}
1309
1310/*****************************************************************************/
1311/* requests implemented outside eio_execute, because they are so large */
1312
1313static void
1314eio__realpath (eio_req *req, etp_worker *self)
1315{
1316 char *rel = req->ptr1;
1317 char *res;
1318 char *tmp1, *tmp2;
1319#if SYMLOOP_MAX > 32
1320 int symlinks = SYMLOOP_MAX;
1321#else
1322 int symlinks = 32;
1323#endif
1324
1325 req->result = -1;
1326
1327 errno = EINVAL;
1328 if (!rel)
1329 return;
1330
1331 errno = ENOENT;
1332 if (!*rel)
1333 return;
1334
1335 if (!req->ptr2)
1336 {
1337 X_LOCK (wrklock);
1338 req->flags |= EIO_FLAG_PTR2_FREE;
1339 X_UNLOCK (wrklock);
1340 req->ptr2 = malloc (PATH_MAX * 3);
1341
1342 errno = ENOMEM;
1343 if (!req->ptr2)
1344 return;
1345 }
1346
1347 res = req->ptr2;
1348 tmp1 = res + PATH_MAX;
1349 tmp2 = tmp1 + PATH_MAX;
1350
1351#if 0 /* disabled, the musl way to do things is just too racy */
1352#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1353 /* on linux we may be able to ask the kernel */
1354 {
1355 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1356
1357 if (fd >= 0)
1358 {
1359 sprintf (tmp1, "/proc/self/fd/%d", fd);
1360 req->result = readlink (tmp1, res, PATH_MAX);
1361 close (fd);
1362
1363 /* here we should probably stat the open file and the disk file, to make sure they still match */
1364
1365 if (req->result > 0)
1366 goto done;
1367 }
1368 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1369 return;
1370 }
1371#endif
1372#endif
1373
1374 if (*rel != '/')
1375 {
1376 if (!getcwd (res, PATH_MAX))
1377 return;
1378
1379 if (res [1]) /* only use if not / */
1380 res += strlen (res);
1381 }
1382
1383 while (*rel)
1384 {
1385 eio_ssize_t len, linklen;
1386 char *beg = rel;
1387
1388 while (*rel && *rel != '/')
1389 ++rel;
1390
1391 len = rel - beg;
1392
1393 if (!len) /* skip slashes */
1394 {
1395 ++rel;
1396 continue;
1397 }
1398
1399 if (beg [0] == '.')
1400 {
1401 if (len == 1)
1402 continue; /* . - nop */
1403
1404 if (beg [1] == '.' && len == 2)
1405 {
1406 /* .. - back up one component, if possible */
1407
1408 while (res != req->ptr2)
1409 if (*--res == '/')
1410 break;
1411
1412 continue;
1413 }
1414 }
1415
1416 errno = ENAMETOOLONG;
1417 if (res + 1 + len + 1 >= tmp1)
1418 return;
1419
1420 /* copy one component */
1421 *res = '/';
1422 memcpy (res + 1, beg, len);
1423
1424 /* zero-terminate, for readlink */
1425 res [len + 1] = 0;
1426
1427 /* now check if it's a symlink */
1428 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1429
1430 if (linklen < 0)
1431 {
1432 if (errno != EINVAL)
1433 return;
1434
1435 /* it's a normal directory. hopefully */
1436 res += len + 1;
1437 }
1438 else
1439 {
1440 /* yay, it was a symlink - build new path in tmp2 */
1441 int rellen = strlen (rel);
1442
1443 errno = ENAMETOOLONG;
1444 if (linklen + 1 + rellen >= PATH_MAX)
1445 return;
1446
1447 errno = ELOOP;
1448 if (!--symlinks)
1449 return;
1450
1451 if (*tmp1 == '/')
1452 res = req->ptr2; /* symlink resolves to an absolute path */
1453
1454 /* we need to be careful, as rel might point into tmp2 already */
1455 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1456 tmp2 [linklen] = '/';
1457 memcpy (tmp2, tmp1, linklen);
1458
1459 rel = tmp2;
1460 }
1461 }
1462
1463 /* special case for the lone root path */
1464 if (res == req->ptr2)
1465 *res++ = '/';
1466
1467 req->result = res - (char *)req->ptr2;
1468
1469done:
1470 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1471}
1472
1030static signed char 1473static signed char
1031eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1474eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1032{ 1475{
1033 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1476 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1034 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1477 : a->inode < b->inode ? -1
1478 : a->inode > b->inode ? 1
1479 : 0;
1035} 1480}
1036 1481
1037#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1482#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1038 1483
1039#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1484#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1045 unsigned char bits [9 + sizeof (ino_t) * 8]; 1490 unsigned char bits [9 + sizeof (ino_t) * 8];
1046 unsigned char *bit = bits; 1491 unsigned char *bit = bits;
1047 1492
1048 assert (CHAR_BIT == 8); 1493 assert (CHAR_BIT == 8);
1049 assert (sizeof (eio_dirent) * 8 < 256); 1494 assert (sizeof (eio_dirent) * 8 < 256);
1050 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1495 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1051 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1496 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1052 1497
1053 if (size <= EIO_SORT_FAST) 1498 if (size <= EIO_SORT_FAST)
1054 return; 1499 return;
1055 1500
1056 /* first prepare an array of bits to test in our radix sort */ 1501 /* first prepare an array of bits to test in our radix sort */
1211 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1656 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1212 1657
1213 X_LOCK (wrklock); 1658 X_LOCK (wrklock);
1214 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1659 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1215 self->dirp = dirp = opendir (req->ptr1); 1660 self->dirp = dirp = opendir (req->ptr1);
1661
1662 if (req->flags & EIO_FLAG_PTR1_FREE)
1663 free (req->ptr1);
1664
1216 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1665 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1217 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1666 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1218 req->ptr2 = names = malloc (namesalloc); 1667 req->ptr2 = names = malloc (namesalloc);
1219 X_UNLOCK (wrklock); 1668 X_UNLOCK (wrklock);
1220 1669
1232 /* sort etc. */ 1681 /* sort etc. */
1233 req->int1 = flags; 1682 req->int1 = flags;
1234 req->result = dentoffs; 1683 req->result = dentoffs;
1235 1684
1236 if (flags & EIO_READDIR_STAT_ORDER) 1685 if (flags & EIO_READDIR_STAT_ORDER)
1237 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1686 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1238 else if (flags & EIO_READDIR_DIRS_FIRST) 1687 else if (flags & EIO_READDIR_DIRS_FIRST)
1239 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1688 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1240 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1689 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1241 else 1690 else
1242 { 1691 {
1244 eio_dirent *oth = dents + dentoffs; 1693 eio_dirent *oth = dents + dentoffs;
1245 eio_dirent *dir = dents; 1694 eio_dirent *dir = dents;
1246 1695
1247 /* now partition dirs to the front, and non-dirs to the back */ 1696 /* now partition dirs to the front, and non-dirs to the back */
1248 /* by walking from both sides and swapping if necessary */ 1697 /* by walking from both sides and swapping if necessary */
1249 /* also clear score, so it doesn't influence sorting */
1250 while (oth > dir) 1698 while (oth > dir)
1251 { 1699 {
1252 if (dir->type == EIO_DT_DIR) 1700 if (dir->type == EIO_DT_DIR)
1253 ++dir; 1701 ++dir;
1254 else if ((--oth)->type == EIO_DT_DIR) 1702 else if ((--oth)->type == EIO_DT_DIR)
1257 1705
1258 ++dir; 1706 ++dir;
1259 } 1707 }
1260 } 1708 }
1261 1709
1262 /* now sort the dirs only */ 1710 /* now sort the dirs only (dirs all have the same score) */
1263 eio_dent_sort (dents, dir - dents, 0, inode_bits); 1711 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1264 } 1712 }
1265 1713
1266 break; 1714 break;
1267 } 1715 }
1272 /* skip . and .. entries */ 1720 /* skip . and .. entries */
1273 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1721 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1274 { 1722 {
1275 int len = D_NAMLEN (entp) + 1; 1723 int len = D_NAMLEN (entp) + 1;
1276 1724
1277 while (expect_false (namesoffs + len > namesalloc)) 1725 while (ecb_expect_false (namesoffs + len > namesalloc))
1278 { 1726 {
1279 namesalloc *= 2; 1727 namesalloc *= 2;
1280 X_LOCK (wrklock); 1728 X_LOCK (wrklock);
1281 req->ptr2 = names = realloc (names, namesalloc); 1729 req->ptr2 = names = realloc (names, namesalloc);
1282 X_UNLOCK (wrklock); 1730 X_UNLOCK (wrklock);
1289 1737
1290 if (dents) 1738 if (dents)
1291 { 1739 {
1292 struct eio_dirent *ent; 1740 struct eio_dirent *ent;
1293 1741
1294 if (expect_false (dentoffs == dentalloc)) 1742 if (ecb_expect_false (dentoffs == dentalloc))
1295 { 1743 {
1296 dentalloc *= 2; 1744 dentalloc *= 2;
1297 X_LOCK (wrklock); 1745 X_LOCK (wrklock);
1298 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1746 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1299 X_UNLOCK (wrklock); 1747 X_UNLOCK (wrklock);
1387 break; 1835 break;
1388 } 1836 }
1389 } 1837 }
1390} 1838}
1391 1839
1392#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1393# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1394#else
1395
1396int
1397eio__msync (void *mem, size_t len, int flags)
1398{
1399 if (EIO_MS_ASYNC != MS_SYNC
1400 || EIO_MS_INVALIDATE != MS_INVALIDATE
1401 || EIO_MS_SYNC != MS_SYNC)
1402 {
1403 flags = 0
1404 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1405 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1406 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1407 }
1408
1409 return msync (mem, len, flags);
1410}
1411
1412#endif
1413
1414int
1415eio__mtouch (void *mem, size_t len, int flags)
1416{
1417 intptr_t addr = (intptr_t)mem;
1418 intptr_t end = addr + len;
1419#ifdef PAGESIZE
1420 const intptr_t page = PAGESIZE;
1421#else
1422 static intptr_t page;
1423
1424 if (!page)
1425 page = sysconf (_SC_PAGESIZE);
1426#endif
1427
1428 /* round down to start of page, although this is probably useless */
1429 addr &= ~(page - 1); /* assume page size is always a power of two */
1430
1431 if (addr < end)
1432 if (flags & EIO_MT_MODIFY) /* modify */
1433 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1434 else
1435 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1436
1437 return 0;
1438}
1439
1440/*****************************************************************************/ 1840/*****************************************************************************/
1441 1841
1442#define ALLOC(len) \ 1842#define ALLOC(len) \
1443 if (!req->ptr2) \ 1843 if (!req->ptr2) \
1444 { \ 1844 { \
1457X_THREAD_PROC (etp_proc) 1857X_THREAD_PROC (etp_proc)
1458{ 1858{
1459 ETP_REQ *req; 1859 ETP_REQ *req;
1460 struct timespec ts; 1860 struct timespec ts;
1461 etp_worker *self = (etp_worker *)thr_arg; 1861 etp_worker *self = (etp_worker *)thr_arg;
1862 int timeout;
1462 1863
1463 /* try to distribute timeouts somewhat randomly */ 1864 /* try to distribute timeouts somewhat evenly */
1464 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1865 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1465 1866
1466 for (;;) 1867 for (;;)
1467 { 1868 {
1869 ts.tv_sec = 0;
1870
1468 X_LOCK (reqlock); 1871 X_LOCK (reqlock);
1469 1872
1470 for (;;) 1873 for (;;)
1471 { 1874 {
1472 self->req = req = reqq_shift (&req_queue); 1875 self->req = req = reqq_shift (&req_queue);
1473 1876
1474 if (req) 1877 if (req)
1475 break; 1878 break;
1476 1879
1880 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1881 {
1882 X_UNLOCK (reqlock);
1883 X_LOCK (wrklock);
1884 --started;
1885 X_UNLOCK (wrklock);
1886 goto quit;
1887 }
1888
1477 ++idle; 1889 ++idle;
1478 1890
1479 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1891 if (idle <= max_idle)
1480 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1892 /* we are allowed to idle, so do so without any timeout */
1893 X_COND_WAIT (reqwait, reqlock);
1894 else
1481 { 1895 {
1482 if (idle > max_idle) 1896 /* initialise timeout once */
1483 { 1897 if (!ts.tv_sec)
1484 --idle; 1898 ts.tv_sec = time (0) + idle_timeout;
1485 X_UNLOCK (reqlock);
1486 X_LOCK (wrklock);
1487 --started;
1488 X_UNLOCK (wrklock);
1489 goto quit;
1490 }
1491 1899
1492 /* we are allowed to idle, so do so without any timeout */
1493 X_COND_WAIT (reqwait, reqlock); 1900 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1901 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1494 } 1902 }
1495 1903
1496 --idle; 1904 --idle;
1497 } 1905 }
1498 1906
1501 X_UNLOCK (reqlock); 1909 X_UNLOCK (reqlock);
1502 1910
1503 if (req->type < 0) 1911 if (req->type < 0)
1504 goto quit; 1912 goto quit;
1505 1913
1506 if (!EIO_CANCELLED (req))
1507 ETP_EXECUTE (self, req); 1914 ETP_EXECUTE (self, req);
1508 1915
1509 X_LOCK (reslock); 1916 X_LOCK (reslock);
1510 1917
1511 ++npending; 1918 ++npending;
1512 1919
1527 return 0; 1934 return 0;
1528} 1935}
1529 1936
1530/*****************************************************************************/ 1937/*****************************************************************************/
1531 1938
1939int ecb_cold
1532int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1940eio_init (void (*want_poll)(void), void (*done_poll)(void))
1533{ 1941{
1534 return etp_init (want_poll, done_poll); 1942 return etp_init (want_poll, done_poll);
1535} 1943}
1536 1944
1945ecb_inline void
1537static void eio_api_destroy (eio_req *req) 1946eio_api_destroy (eio_req *req)
1538{ 1947{
1539 free (req); 1948 free (req);
1540} 1949}
1541 1950
1542#define REQ(rtype) \ 1951#define REQ(rtype) \
1561 { \ 1970 { \
1562 eio_api_destroy (req); \ 1971 eio_api_destroy (req); \
1563 return 0; \ 1972 return 0; \
1564 } 1973 }
1565 1974
1975static void
1566static void eio_execute (etp_worker *self, eio_req *req) 1976eio_execute (etp_worker *self, eio_req *req)
1567{ 1977{
1568 errno = 0; 1978 if (ecb_expect_false (EIO_CANCELLED (req)))
1979 {
1980 req->result = -1;
1981 req->errorno = ECANCELED;
1982 return;
1983 }
1569 1984
1570 switch (req->type) 1985 switch (req->type)
1571 { 1986 {
1572 case EIO_READ: ALLOC (req->size); 1987 case EIO_READ: ALLOC (req->size);
1573 req->result = req->offs >= 0 1988 req->result = req->offs >= 0
1606 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2021 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1607 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2022 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1608 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2023 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1609 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2024 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1610 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2025 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1611 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2026 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1612 2027
2028 case EIO_REALPATH: eio__realpath (req, self); break;
2029
1613 case EIO_READLINK: ALLOC (NAME_MAX); 2030 case EIO_READLINK: ALLOC (PATH_MAX);
1614 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2031 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1615 2032
1616 case EIO_SYNC: req->result = 0; sync (); break; 2033 case EIO_SYNC: req->result = 0; sync (); break;
1617 case EIO_FSYNC: req->result = fsync (req->int1); break; 2034 case EIO_FSYNC: req->result = fsync (req->int1); break;
1618 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2035 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1619 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2036 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2037 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1620 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2038 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2039 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1621 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2040 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2041 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1622 2042
1623 case EIO_READDIR: eio__scandir (req, self); break; 2043 case EIO_READDIR: eio__scandir (req, self); break;
1624 2044
1625 case EIO_BUSY: 2045 case EIO_BUSY:
1626#ifdef _WIN32 2046#ifdef _WIN32
1627 Sleep (req->nv1 * 1000.); 2047 Sleep (req->nv1 * 1e3);
1628#else 2048#else
1629 { 2049 {
1630 struct timeval tv; 2050 struct timeval tv;
1631 2051
1632 tv.tv_sec = req->nv1; 2052 tv.tv_sec = req->nv1;
1633 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2053 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1634 2054
1635 req->result = select (0, 0, 0, 0, &tv); 2055 req->result = select (0, 0, 0, 0, &tv);
1636 } 2056 }
1637#endif 2057#endif
1638 break; 2058 break;
1653 times = tv; 2073 times = tv;
1654 } 2074 }
1655 else 2075 else
1656 times = 0; 2076 times = 0;
1657 2077
1658
1659 req->result = req->type == EIO_FUTIME 2078 req->result = req->type == EIO_FUTIME
1660 ? futimes (req->int1, times) 2079 ? futimes (req->int1, times)
1661 : utimes (req->ptr1, times); 2080 : utimes (req->ptr1, times);
1662 } 2081 }
1663 break; 2082 break;
1668 case EIO_NOP: 2087 case EIO_NOP:
1669 req->result = 0; 2088 req->result = 0;
1670 break; 2089 break;
1671 2090
1672 case EIO_CUSTOM: 2091 case EIO_CUSTOM:
1673 ((void (*)(eio_req *))req->feed) (req); 2092 req->feed (req);
1674 break; 2093 break;
1675 2094
1676 default: 2095 default:
2096 errno = ENOSYS;
1677 req->result = -1; 2097 req->result = -1;
1678 break; 2098 break;
1679 } 2099 }
1680 2100
1681 req->errorno = errno; 2101 req->errorno = errno;
1711eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2131eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1712{ 2132{
1713 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2133 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1714} 2134}
1715 2135
2136eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2137{
2138 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2139}
2140
2141eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2142{
2143 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2144}
2145
1716eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2146eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1717{ 2147{
1718 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2148 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1719} 2149}
1720 2150
2151eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2152{
2153 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2154}
2155
1721eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2156eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1722{ 2157{
1723 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2158 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1724} 2159}
1725 2160
1766eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2201eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1767{ 2202{
1768 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2203 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1769} 2204}
1770 2205
1771eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2206eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1772{ 2207{
1773 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2208 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1774} 2209}
1775 2210
1776eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2211eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1796eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2231eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1797{ 2232{
1798 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2233 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1799} 2234}
1800 2235
1801eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2236eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1802{ 2237{
1803 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2238 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1804} 2239}
1805 2240
1806eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2241eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1822eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2257eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1823{ 2258{
1824 return eio__1path (EIO_READLINK, path, pri, cb, data); 2259 return eio__1path (EIO_READLINK, path, pri, cb, data);
1825} 2260}
1826 2261
2262eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2263{
2264 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2265}
2266
1827eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2267eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1828{ 2268{
1829 return eio__1path (EIO_STAT, path, pri, cb, data); 2269 return eio__1path (EIO_STAT, path, pri, cb, data);
1830} 2270}
1831 2271
1854 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2294 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1855} 2295}
1856 2296
1857eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2297eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1858{ 2298{
1859 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2299 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1860} 2300}
1861 2301
1862static eio_req * 2302static eio_req *
1863eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2303eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1864{ 2304{
1888eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2328eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1889{ 2329{
1890 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2330 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1891} 2331}
1892 2332
1893eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2333eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1894{ 2334{
1895 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2335 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1896} 2336}
1897 2337
1898#endif 2338#endif
1899 2339
1900eio_req *eio_grp (eio_cb cb, void *data) 2340eio_req *eio_grp (eio_cb cb, void *data)
1909#undef SEND 2349#undef SEND
1910 2350
1911/*****************************************************************************/ 2351/*****************************************************************************/
1912/* grp functions */ 2352/* grp functions */
1913 2353
2354void
1914void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2355eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1915{ 2356{
1916 grp->int2 = limit; 2357 grp->int2 = limit;
1917 grp->feed = feed; 2358 grp->feed = feed;
1918 2359
1919 grp_try_feed (grp); 2360 grp_try_feed (grp);
1920} 2361}
1921 2362
2363void
1922void eio_grp_limit (eio_req *grp, int limit) 2364eio_grp_limit (eio_req *grp, int limit)
1923{ 2365{
1924 grp->int2 = limit; 2366 grp->int2 = limit;
1925 2367
1926 grp_try_feed (grp); 2368 grp_try_feed (grp);
1927} 2369}
1928 2370
2371void
1929void eio_grp_add (eio_req *grp, eio_req *req) 2372eio_grp_add (eio_req *grp, eio_req *req)
1930{ 2373{
1931 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2374 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1932 2375
1933 grp->flags |= EIO_FLAG_GROUPADD; 2376 grp->flags |= EIO_FLAG_GROUPADD;
1934 2377
1945} 2388}
1946 2389
1947/*****************************************************************************/ 2390/*****************************************************************************/
1948/* misc garbage */ 2391/* misc garbage */
1949 2392
2393eio_ssize_t
1950ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2394eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1951{ 2395{
1952 etp_worker wrk; 2396 etp_worker wrk;
1953 ssize_t ret; 2397 eio_ssize_t ret;
1954 2398
1955 wrk.dbuf = 0; 2399 wrk.dbuf = 0;
1956 2400
1957 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2401 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1958 2402

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines