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

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.33 by root, Sat Jun 6 19:44:17 2009 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 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
76static void eio_destroy (eio_req *req);
77
58#ifndef EIO_FINISH 78#ifndef EIO_FINISH
59# 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
60#endif 80#endif
61 81
62#ifndef EIO_DESTROY 82#ifndef EIO_DESTROY
65 85
66#ifndef EIO_FEED 86#ifndef EIO_FEED
67# 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)
68#endif 88#endif
69 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
70#ifdef _WIN32 101#ifdef _WIN32
71 102
72 /*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
73#else 151#else
74 152
75# include "config.h"
76# include <sys/time.h> 153 #include <sys/time.h>
77# include <sys/select.h> 154 #include <sys/select.h>
155 #include <sys/statvfs.h>
78# include <unistd.h> 156 #include <unistd.h>
79# include <utime.h> 157 #include <utime.h>
80# include <signal.h> 158 #include <signal.h>
81# include <dirent.h> 159 #include <dirent.h>
82 160
161 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
162 #include <sys/mman.h>
163 #endif
164
83/* 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 */
84# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) 166 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
85# define _DIRENT_HAVE_D_TYPE /* sigh */ 167 #define _DIRENT_HAVE_D_TYPE /* sigh */
86# define D_INO(de) (de)->d_fileno 168 #define D_INO(de) (de)->d_fileno
87# define D_NAMLEN(de) (de)->d_namlen 169 #define D_NAMLEN(de) (de)->d_namlen
88# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 170 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
89# define D_INO(de) (de)->d_ino 171 #define D_INO(de) (de)->d_ino
90# endif 172 #endif
91 173
92#ifdef _D_EXACT_NAMLEN 174 #ifdef _D_EXACT_NAMLEN
93# undef D_NAMLEN 175 #undef D_NAMLEN
94# define D_NAMLEN(de) _D_EXACT_NAMLEN (de) 176 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
95#endif 177 #endif
96 178
97# ifdef _DIRENT_HAVE_D_TYPE 179 #ifdef _DIRENT_HAVE_D_TYPE
98# define D_TYPE(de) (de)->d_type 180 #define D_TYPE(de) (de)->d_type
99# endif 181 #endif
100 182
101# ifndef EIO_STRUCT_DIRENT 183 #ifndef EIO_STRUCT_DIRENT
102# define EIO_STRUCT_DIRENT struct dirent 184 #define EIO_STRUCT_DIRENT struct dirent
103# endif 185 #endif
104 186
105#endif 187#endif
106 188
107#if HAVE_SENDFILE 189#if HAVE_SENDFILE
108# if __linux 190# if __linux
109# include <sys/sendfile.h> 191# include <sys/sendfile.h>
110# elif __freebsd 192# elif __FreeBSD__ || defined __APPLE__
111# include <sys/socket.h> 193# include <sys/socket.h>
112# include <sys/uio.h> 194# include <sys/uio.h>
113# elif __hpux 195# elif __hpux
114# include <sys/socket.h> 196# include <sys/socket.h>
115# elif __solaris /* not yet */ 197# elif __solaris
116# include <sys/sendfile.h> 198# include <sys/sendfile.h>
117# else 199# else
118# error sendfile support requested but not available 200# error sendfile support requested but not available
119# endif 201# endif
120#endif 202#endif
127#endif 209#endif
128#ifndef D_NAMLEN 210#ifndef D_NAMLEN
129# define D_NAMLEN(de) strlen ((de)->d_name) 211# define D_NAMLEN(de) strlen ((de)->d_name)
130#endif 212#endif
131 213
132/* number of seconds after which an idle threads exit */
133#define IDLE_TIMEOUT 10
134
135/* used for struct dirent, AIX doesn't provide it */ 214/* used for struct dirent, AIX doesn't provide it */
136#ifndef NAME_MAX 215#ifndef NAME_MAX
137# 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
138#endif 222#endif
139 223
140/* buffer size for various temporary buffers */ 224/* buffer size for various temporary buffers */
141#define EIO_BUFSIZE 65536 225#define EIO_BUFSIZE 65536
142 226
148 errno = ENOMEM; \ 232 errno = ENOMEM; \
149 if (!eio_buf) \ 233 if (!eio_buf) \
150 return -1; 234 return -1;
151 235
152#define EIO_TICKS ((1000000 + 1023) >> 10) 236#define EIO_TICKS ((1000000 + 1023) >> 10)
153
154/*****************************************************************************/
155
156#if __GNUC__ >= 3
157# define expect(expr,value) __builtin_expect ((expr),(value))
158#else
159# define expect(expr,value) (expr)
160#endif
161
162#define expect_false(expr) expect ((expr) != 0, 0)
163#define expect_true(expr) expect ((expr) != 0, 1)
164
165/*****************************************************************************/
166 237
167#define ETP_PRI_MIN EIO_PRI_MIN 238#define ETP_PRI_MIN EIO_PRI_MIN
168#define ETP_PRI_MAX EIO_PRI_MAX 239#define ETP_PRI_MAX EIO_PRI_MAX
169 240
170struct etp_worker; 241struct etp_worker;
195 266
196/*****************************************************************************/ 267/*****************************************************************************/
197 268
198#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 269#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
199 270
200/* calculcate time difference in ~1/EIO_TICKS of a second */ 271/* calculate time difference in ~1/EIO_TICKS of a second */
272ecb_inline int
201static int tvdiff (struct timeval *tv1, struct timeval *tv2) 273tvdiff (struct timeval *tv1, struct timeval *tv2)
202{ 274{
203 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 275 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
204 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 276 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
205} 277}
206 278
213static unsigned int max_poll_reqs; /* reslock */ 285static unsigned int max_poll_reqs; /* reslock */
214 286
215static volatile unsigned int nreqs; /* reqlock */ 287static volatile unsigned int nreqs; /* reqlock */
216static volatile unsigned int nready; /* reqlock */ 288static volatile unsigned int nready; /* reqlock */
217static volatile unsigned int npending; /* reqlock */ 289static volatile unsigned int npending; /* reqlock */
218static 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 */
219 292
220static mutex_t wrklock = X_MUTEX_INIT; 293static xmutex_t wrklock;
221static mutex_t reslock = X_MUTEX_INIT; 294static xmutex_t reslock;
222static mutex_t reqlock = X_MUTEX_INIT; 295static xmutex_t reqlock;
223static cond_t reqwait = X_COND_INIT; 296static xcond_t reqwait;
224 297
225#if !HAVE_PREADWRITE 298#if !HAVE_PREADWRITE
226/* 299/*
227 * make our pread/pwrite emulation safe against themselves, but not against 300 * make our pread/pwrite emulation safe against themselves, but not against
228 * 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,
229 * but that's your problem, not mine. 302 * but that's your problem, not mine.
230 */ 303 */
231static mutex_t preadwritelock = X_MUTEX_INIT; 304static xmutex_t preadwritelock = X_MUTEX_INIT;
232#endif 305#endif
233 306
234typedef struct etp_worker 307typedef struct etp_worker
235{ 308{
236 /* locked by wrklock */ 309 /* locked by wrklock */
237 struct etp_worker *prev, *next; 310 struct etp_worker *prev, *next;
238 311
239 thread_t tid; 312 xthread_t tid;
240 313
241 /* locked by reslock, reqlock or wrklock */ 314 /* locked by reslock, reqlock or wrklock */
242 ETP_REQ *req; /* currently processed request */ 315 ETP_REQ *req; /* currently processed request */
243 316
244 ETP_WORKER_COMMON 317 ETP_WORKER_COMMON
249#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 322#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
250#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 323#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
251 324
252/* worker threads management */ 325/* worker threads management */
253 326
327static void ecb_cold
254static void etp_worker_clear (etp_worker *wrk) 328etp_worker_clear (etp_worker *wrk)
255{ 329{
256 ETP_WORKER_CLEAR (wrk); 330 ETP_WORKER_CLEAR (wrk);
257} 331}
258 332
333static void ecb_cold
259static void etp_worker_free (etp_worker *wrk) 334etp_worker_free (etp_worker *wrk)
260{ 335{
261 wrk->next->prev = wrk->prev; 336 wrk->next->prev = wrk->prev;
262 wrk->prev->next = wrk->next; 337 wrk->prev->next = wrk->next;
263 338
264 free (wrk); 339 free (wrk);
265} 340}
266 341
267static unsigned int etp_nreqs (void) 342static unsigned int
343etp_nreqs (void)
268{ 344{
269 int retval; 345 int retval;
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 346 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = nreqs; 347 retval = nreqs;
272 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 348 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273 return retval; 349 return retval;
274} 350}
275 351
276static unsigned int etp_nready (void) 352static unsigned int
353etp_nready (void)
277{ 354{
278 unsigned int retval; 355 unsigned int retval;
279 356
280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 357 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281 retval = nready; 358 retval = nready;
282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 359 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283 360
284 return retval; 361 return retval;
285} 362}
286 363
287static unsigned int etp_npending (void) 364static unsigned int
365etp_npending (void)
288{ 366{
289 unsigned int retval; 367 unsigned int retval;
290 368
291 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 369 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
292 retval = npending; 370 retval = npending;
293 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 371 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
294 372
295 return retval; 373 return retval;
296} 374}
297 375
298static unsigned int etp_nthreads (void) 376static unsigned int
377etp_nthreads (void)
299{ 378{
300 unsigned int retval; 379 unsigned int retval;
301 380
302 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 381 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303 retval = started; 382 retval = started;
317} etp_reqq; 396} etp_reqq;
318 397
319static etp_reqq req_queue; 398static etp_reqq req_queue;
320static etp_reqq res_queue; 399static etp_reqq res_queue;
321 400
401static int ecb_noinline
322static int reqq_push (etp_reqq *q, ETP_REQ *req) 402reqq_push (etp_reqq *q, ETP_REQ *req)
323{ 403{
324 int pri = req->pri; 404 int pri = req->pri;
325 req->next = 0; 405 req->next = 0;
326 406
327 if (q->qe[pri]) 407 if (q->qe[pri])
333 q->qe[pri] = q->qs[pri] = req; 413 q->qe[pri] = q->qs[pri] = req;
334 414
335 return q->size++; 415 return q->size++;
336} 416}
337 417
418static ETP_REQ * ecb_noinline
338static ETP_REQ *reqq_shift (etp_reqq *q) 419reqq_shift (etp_reqq *q)
339{ 420{
340 int pri; 421 int pri;
341 422
342 if (!q->size) 423 if (!q->size)
343 return 0; 424 return 0;
358 } 439 }
359 440
360 abort (); 441 abort ();
361} 442}
362 443
363static void etp_atfork_prepare (void) 444static void ecb_cold
445etp_thread_init (void)
364{ 446{
365 X_LOCK (wrklock);
366 X_LOCK (reqlock);
367 X_LOCK (reslock);
368#if !HAVE_PREADWRITE 447#if !HAVE_PREADWRITE
369 X_LOCK (preadwritelock); 448 X_MUTEX_CREATE (preadwritelock);
370#endif 449#endif
450 X_MUTEX_CREATE (wrklock);
451 X_MUTEX_CREATE (reslock);
452 X_MUTEX_CREATE (reqlock);
453 X_COND_CREATE (reqwait);
371} 454}
372 455
456static void ecb_cold
457etp_atfork_prepare (void)
458{
459}
460
461static void ecb_cold
373static void etp_atfork_parent (void) 462etp_atfork_parent (void)
374{ 463{
375#if !HAVE_PREADWRITE
376 X_UNLOCK (preadwritelock);
377#endif
378 X_UNLOCK (reslock);
379 X_UNLOCK (reqlock);
380 X_UNLOCK (wrklock);
381} 464}
382 465
466static void ecb_cold
383static void etp_atfork_child (void) 467etp_atfork_child (void)
384{ 468{
385 ETP_REQ *prv; 469 ETP_REQ *prv;
386 470
387 while ((prv = reqq_shift (&req_queue))) 471 while ((prv = reqq_shift (&req_queue)))
388 ETP_DESTROY (prv); 472 ETP_DESTROY (prv);
405 idle = 0; 489 idle = 0;
406 nreqs = 0; 490 nreqs = 0;
407 nready = 0; 491 nready = 0;
408 npending = 0; 492 npending = 0;
409 493
410 etp_atfork_parent (); 494 etp_thread_init ();
411} 495}
412 496
413static void 497static void ecb_cold
414etp_once_init (void) 498etp_once_init (void)
415{ 499{
500 etp_thread_init ();
416 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);
417} 502}
418 503
419static int 504static int ecb_cold
420etp_init (void (*want_poll)(void), void (*done_poll)(void)) 505etp_init (void (*want_poll)(void), void (*done_poll)(void))
421{ 506{
422 static pthread_once_t doinit = PTHREAD_ONCE_INIT; 507 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
423 508
424 pthread_once (&doinit, etp_once_init); 509 pthread_once (&doinit, etp_once_init);
429 return 0; 514 return 0;
430} 515}
431 516
432X_THREAD_PROC (etp_proc); 517X_THREAD_PROC (etp_proc);
433 518
519static void ecb_cold
434static void etp_start_thread (void) 520etp_start_thread (void)
435{ 521{
436 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 522 etp_worker *wrk = calloc (1, sizeof (etp_worker));
437 523
438 /*TODO*/ 524 /*TODO*/
439 assert (("unable to allocate worker thread data", wrk)); 525 assert (("unable to allocate worker thread data", wrk));
452 free (wrk); 538 free (wrk);
453 539
454 X_UNLOCK (wrklock); 540 X_UNLOCK (wrklock);
455} 541}
456 542
543static void
457static void etp_maybe_start_thread (void) 544etp_maybe_start_thread (void)
458{ 545{
459 if (expect_true (etp_nthreads () >= wanted)) 546 if (ecb_expect_true (etp_nthreads () >= wanted))
460 return; 547 return;
461 548
462 /* todo: maybe use idle here, but might be less exact */ 549 /* todo: maybe use idle here, but might be less exact */
463 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 ()))
464 return; 551 return;
465 552
466 etp_start_thread (); 553 etp_start_thread ();
467} 554}
468 555
556static void ecb_cold
469static void etp_end_thread (void) 557etp_end_thread (void)
470{ 558{
471 eio_req *req = calloc (1, sizeof (eio_req)); 559 eio_req *req = calloc (1, sizeof (eio_req));
472 560
473 req->type = -1; 561 req->type = -1;
474 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 562 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
481 X_LOCK (wrklock); 569 X_LOCK (wrklock);
482 --started; 570 --started;
483 X_UNLOCK (wrklock); 571 X_UNLOCK (wrklock);
484} 572}
485 573
486static int etp_poll (void) 574static int
575etp_poll (void)
487{ 576{
488 unsigned int maxreqs; 577 unsigned int maxreqs;
489 unsigned int maxtime; 578 unsigned int maxtime;
490 struct timeval tv_start, tv_now; 579 struct timeval tv_start, tv_now;
491 580
521 610
522 X_LOCK (reqlock); 611 X_LOCK (reqlock);
523 --nreqs; 612 --nreqs;
524 X_UNLOCK (reqlock); 613 X_UNLOCK (reqlock);
525 614
526 if (expect_false (req->type == EIO_GROUP && req->size)) 615 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
527 { 616 {
528 req->int1 = 1; /* mark request as delayed */ 617 req->int1 = 1; /* mark request as delayed */
529 continue; 618 continue;
530 } 619 }
531 else 620 else
532 { 621 {
533 int res = ETP_FINISH (req); 622 int res = ETP_FINISH (req);
534 if (expect_false (res)) 623 if (ecb_expect_false (res))
535 return res; 624 return res;
536 } 625 }
537 626
538 if (expect_false (maxreqs && !--maxreqs)) 627 if (ecb_expect_false (maxreqs && !--maxreqs))
539 break; 628 break;
540 629
541 if (maxtime) 630 if (maxtime)
542 { 631 {
543 gettimeofday (&tv_now, 0); 632 gettimeofday (&tv_now, 0);
549 638
550 errno = EAGAIN; 639 errno = EAGAIN;
551 return -1; 640 return -1;
552} 641}
553 642
643static void
554static void etp_cancel (ETP_REQ *req) 644etp_cancel (ETP_REQ *req)
555{ 645{
556 X_LOCK (wrklock); 646 req->cancelled = 1;
557 req->flags |= EIO_FLAG_CANCELLED;
558 X_UNLOCK (wrklock);
559 647
560 eio_grp_cancel (req); 648 eio_grp_cancel (req);
561} 649}
562 650
651static void
563static void etp_submit (ETP_REQ *req) 652etp_submit (ETP_REQ *req)
564{ 653{
565 req->pri -= ETP_PRI_MIN; 654 req->pri -= ETP_PRI_MIN;
566 655
567 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;
568 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;
569 658
570 if (expect_false (req->type == EIO_GROUP)) 659 if (ecb_expect_false (req->type == EIO_GROUP))
571 { 660 {
572 /* I hope this is worth it :/ */ 661 /* I hope this is worth it :/ */
573 X_LOCK (reqlock); 662 X_LOCK (reqlock);
574 ++nreqs; 663 ++nreqs;
575 X_UNLOCK (reqlock); 664 X_UNLOCK (reqlock);
594 683
595 etp_maybe_start_thread (); 684 etp_maybe_start_thread ();
596 } 685 }
597} 686}
598 687
688static void ecb_cold
599static void etp_set_max_poll_time (double nseconds) 689etp_set_max_poll_time (double nseconds)
600{ 690{
601 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 691 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
602 max_poll_time = nseconds; 692 max_poll_time = nseconds * EIO_TICKS;
603 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 693 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
604} 694}
605 695
696static void ecb_cold
606static void etp_set_max_poll_reqs (unsigned int maxreqs) 697etp_set_max_poll_reqs (unsigned int maxreqs)
607{ 698{
608 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 699 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609 max_poll_reqs = maxreqs; 700 max_poll_reqs = maxreqs;
610 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 701 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611} 702}
612 703
704static void ecb_cold
613static void etp_set_max_idle (unsigned int nthreads) 705etp_set_max_idle (unsigned int nthreads)
614{ 706{
615 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 707 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
616 max_idle = nthreads <= 0 ? 1 : nthreads; 708 max_idle = nthreads;
617 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 709 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
618} 710}
619 711
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
620static void etp_set_min_parallel (unsigned int nthreads) 721etp_set_min_parallel (unsigned int nthreads)
621{ 722{
622 if (wanted < nthreads) 723 if (wanted < nthreads)
623 wanted = nthreads; 724 wanted = nthreads;
624} 725}
625 726
727static void ecb_cold
626static void etp_set_max_parallel (unsigned int nthreads) 728etp_set_max_parallel (unsigned int nthreads)
627{ 729{
628 if (wanted > nthreads) 730 if (wanted > nthreads)
629 wanted = nthreads; 731 wanted = nthreads;
630 732
631 while (started > wanted) 733 while (started > wanted)
632 etp_end_thread (); 734 etp_end_thread ();
633} 735}
634 736
635/*****************************************************************************/ 737/*****************************************************************************/
636 738
739static void
637static void grp_try_feed (eio_req *grp) 740grp_try_feed (eio_req *grp)
638{ 741{
639 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 742 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
640 { 743 {
641 grp->flags &= ~EIO_FLAG_GROUPADD; 744 grp->flags &= ~EIO_FLAG_GROUPADD;
642 745
649 break; 752 break;
650 } 753 }
651 } 754 }
652} 755}
653 756
757static int
654static int grp_dec (eio_req *grp) 758grp_dec (eio_req *grp)
655{ 759{
656 --grp->size; 760 --grp->size;
657 761
658 /* call feeder, if applicable */ 762 /* call feeder, if applicable */
659 grp_try_feed (grp); 763 grp_try_feed (grp);
663 return eio_finish (grp); 767 return eio_finish (grp);
664 else 768 else
665 return 0; 769 return 0;
666} 770}
667 771
772static void
668void eio_destroy (eio_req *req) 773eio_destroy (eio_req *req)
669{ 774{
670 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 775 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
671 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 776 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
672 777
673 EIO_DESTROY (req); 778 EIO_DESTROY (req);
674} 779}
675 780
781static int
676static int eio_finish (eio_req *req) 782eio_finish (eio_req *req)
677{ 783{
678 int res = EIO_FINISH (req); 784 int res = EIO_FINISH (req);
679 785
680 if (req->grp) 786 if (req->grp)
681 { 787 {
689 if (grp->grp_first == req) 795 if (grp->grp_first == req)
690 grp->grp_first = req->grp_next; 796 grp->grp_first = req->grp_next;
691 797
692 res2 = grp_dec (grp); 798 res2 = grp_dec (grp);
693 799
694 if (!res && res2) 800 if (!res)
695 res = res2; 801 res = res2;
696 } 802 }
697 803
698 eio_destroy (req); 804 eio_destroy (req);
699 805
700 return res; 806 return res;
701} 807}
702 808
809void
703void eio_grp_cancel (eio_req *grp) 810eio_grp_cancel (eio_req *grp)
704{ 811{
705 for (grp = grp->grp_first; grp; grp = grp->grp_next) 812 for (grp = grp->grp_first; grp; grp = grp->grp_next)
706 eio_cancel (grp); 813 eio_cancel (grp);
707} 814}
708 815
816void
709void eio_cancel (eio_req *req) 817eio_cancel (eio_req *req)
710{ 818{
711 etp_cancel (req); 819 etp_cancel (req);
712} 820}
713 821
822void
714void eio_submit (eio_req *req) 823eio_submit (eio_req *req)
715{ 824{
716 etp_submit (req); 825 etp_submit (req);
717} 826}
718 827
719unsigned int eio_nreqs (void) 828unsigned int
829eio_nreqs (void)
720{ 830{
721 return etp_nreqs (); 831 return etp_nreqs ();
722} 832}
723 833
724unsigned int eio_nready (void) 834unsigned int
835eio_nready (void)
725{ 836{
726 return etp_nready (); 837 return etp_nready ();
727} 838}
728 839
729unsigned int eio_npending (void) 840unsigned int
841eio_npending (void)
730{ 842{
731 return etp_npending (); 843 return etp_npending ();
732} 844}
733 845
734unsigned int eio_nthreads (void) 846unsigned int ecb_cold
847eio_nthreads (void)
735{ 848{
736 return etp_nthreads (); 849 return etp_nthreads ();
737} 850}
738 851
852void ecb_cold
739void eio_set_max_poll_time (double nseconds) 853eio_set_max_poll_time (double nseconds)
740{ 854{
741 etp_set_max_poll_time (nseconds); 855 etp_set_max_poll_time (nseconds);
742} 856}
743 857
858void ecb_cold
744void eio_set_max_poll_reqs (unsigned int maxreqs) 859eio_set_max_poll_reqs (unsigned int maxreqs)
745{ 860{
746 etp_set_max_poll_reqs (maxreqs); 861 etp_set_max_poll_reqs (maxreqs);
747} 862}
748 863
864void ecb_cold
749void eio_set_max_idle (unsigned int nthreads) 865eio_set_max_idle (unsigned int nthreads)
750{ 866{
751 etp_set_max_idle (nthreads); 867 etp_set_max_idle (nthreads);
752} 868}
753 869
870void ecb_cold
871eio_set_idle_timeout (unsigned int seconds)
872{
873 etp_set_idle_timeout (seconds);
874}
875
876void ecb_cold
754void eio_set_min_parallel (unsigned int nthreads) 877eio_set_min_parallel (unsigned int nthreads)
755{ 878{
756 etp_set_min_parallel (nthreads); 879 etp_set_min_parallel (nthreads);
757} 880}
758 881
882void ecb_cold
759void eio_set_max_parallel (unsigned int nthreads) 883eio_set_max_parallel (unsigned int nthreads)
760{ 884{
761 etp_set_max_parallel (nthreads); 885 etp_set_max_parallel (nthreads);
762} 886}
763 887
764int eio_poll (void) 888int eio_poll (void)
773# undef pread 897# undef pread
774# undef pwrite 898# undef pwrite
775# define pread eio__pread 899# define pread eio__pread
776# define pwrite eio__pwrite 900# define pwrite eio__pwrite
777 901
778static ssize_t 902static eio_ssize_t
779eio__pread (int fd, void *buf, size_t count, off_t offset) 903eio__pread (int fd, void *buf, size_t count, off_t offset)
780{ 904{
781 ssize_t res; 905 eio_ssize_t res;
782 off_t ooffset; 906 off_t ooffset;
783 907
784 X_LOCK (preadwritelock); 908 X_LOCK (preadwritelock);
785 ooffset = lseek (fd, 0, SEEK_CUR); 909 ooffset = lseek (fd, 0, SEEK_CUR);
786 lseek (fd, offset, SEEK_SET); 910 lseek (fd, offset, SEEK_SET);
789 X_UNLOCK (preadwritelock); 913 X_UNLOCK (preadwritelock);
790 914
791 return res; 915 return res;
792} 916}
793 917
794static ssize_t 918static eio_ssize_t
795eio__pwrite (int fd, void *buf, size_t count, off_t offset) 919eio__pwrite (int fd, void *buf, size_t count, off_t offset)
796{ 920{
797 ssize_t res; 921 eio_ssize_t res;
798 off_t ooffset; 922 off_t ooffset;
799 923
800 X_LOCK (preadwritelock); 924 X_LOCK (preadwritelock);
801 ooffset = lseek (fd, 0, SEEK_CUR); 925 ooffset = lseek (fd, 0, SEEK_CUR);
802 lseek (fd, offset, SEEK_SET); 926 lseek (fd, offset, SEEK_SET);
806 930
807 return res; 931 return res;
808} 932}
809#endif 933#endif
810 934
811#ifndef HAVE_FUTIMES 935#ifndef HAVE_UTIMES
812 936
813# undef utimes 937# undef utimes
814# undef futimes
815# define utimes(path,times) eio__utimes (path, times) 938# define utimes(path,times) eio__utimes (path, times)
816# define futimes(fd,times) eio__futimes (fd, times)
817 939
818static int 940static int
819eio__utimes (const char *filename, const struct timeval times[2]) 941eio__utimes (const char *filename, const struct timeval times[2])
820{ 942{
821 if (times) 943 if (times)
829 } 951 }
830 else 952 else
831 return utime (filename, 0); 953 return utime (filename, 0);
832} 954}
833 955
956#endif
957
958#ifndef HAVE_FUTIMES
959
960# undef futimes
961# define futimes(fd,times) eio__futimes (fd, times)
962
963static int
834static int eio__futimes (int fd, const struct timeval tv[2]) 964eio__futimes (int fd, const struct timeval tv[2])
835{ 965{
836 errno = ENOSYS; 966 errno = ENOSYS;
837 return -1; 967 return -1;
838} 968}
839 969
843# undef fdatasync 973# undef fdatasync
844# define fdatasync(fd) fsync (fd) 974# define fdatasync(fd) fsync (fd)
845#endif 975#endif
846 976
847/* sync_file_range always needs emulation */ 977/* sync_file_range always needs emulation */
848int 978static int
849eio__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)
850{ 980{
851#if HAVE_SYNC_FILE_RANGE 981#if HAVE_SYNC_FILE_RANGE
852 int res; 982 int res;
853 983
866 if (!res || errno != ENOSYS) 996 if (!res || errno != ENOSYS)
867 return res; 997 return res;
868#endif 998#endif
869 999
870 /* 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
871 * call fdatasync, as thta matches the expectation of it's users best */ 1001 * call fdatasync, as that matches the expectation of its users best */
872 return fdatasync (fd); 1002 return fdatasync (fd);
1003}
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
873} 1014}
874 1015
875#if !HAVE_READAHEAD 1016#if !HAVE_READAHEAD
876# undef readahead 1017# undef readahead
877# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1018# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
878 1019
879static ssize_t 1020static eio_ssize_t
880eio__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)
881{ 1022{
882 size_t todo = count; 1023 size_t todo = count;
883 dBUF; 1024 dBUF;
884 1025
896} 1037}
897 1038
898#endif 1039#endif
899 1040
900/* sendfile always needs emulation */ 1041/* sendfile always needs emulation */
901static ssize_t 1042static eio_ssize_t
902eio__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)
903{ 1044{
1045 eio_ssize_t written = 0;
904 ssize_t res; 1046 eio_ssize_t res;
905 1047
906 if (!count) 1048 if (!count)
907 return 0; 1049 return 0;
908 1050
1051 for (;;)
1052 {
1053#ifdef __APPLE__
1054# undef HAVE_SENDFILE /* broken, as everything on os x */
1055#endif
909#if HAVE_SENDFILE 1056#if HAVE_SENDFILE
910# if __linux 1057# if __linux
1058 off_t soffset = offset;
911 res = sendfile (ofd, ifd, &offset, count); 1059 res = sendfile (ofd, ifd, &soffset, count);
912 1060
913# elif __freebsd 1061# elif __FreeBSD__
914 /* 1062 /*
915 * 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
916 * wasted on making it similar to other I/O functions. 1064 * wasted on making it similar to other I/O functions.
917 */ 1065 */
918 {
919 off_t sbytes; 1066 off_t sbytes;
920 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1067 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
921 1068
922 if (res < 0 && sbytes) 1069 #if 0 /* according to the manpage, this is correct, but broken behaviour */
923 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1070 /* freebsd' sendfile will return 0 on success */
1071 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1072 /* not on e.g. EIO or EPIPE - sounds broken */
1073 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
924 res = sbytes; 1074 res = sbytes;
925 } 1075 #endif
1076
1077 /* according to source inspection, this is correct, and useful behaviour */
1078 if (sbytes)
1079 res = sbytes;
1080
1081# elif defined (__APPLE__)
1082 off_t sbytes = count;
1083 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1084
1085 /* according to the manpage, sbytes is always valid */
1086 if (sbytes)
1087 res = sbytes;
926 1088
927# elif __hpux 1089# elif __hpux
928 res = sendfile (ofd, ifd, offset, count, 0, 0); 1090 res = sendfile (ofd, ifd, offset, count, 0, 0);
929 1091
930# elif __solaris 1092# elif __solaris
931 {
932 struct sendfilevec vec; 1093 struct sendfilevec vec;
933 size_t sbytes; 1094 size_t sbytes;
934 1095
935 vec.sfv_fd = ifd; 1096 vec.sfv_fd = ifd;
936 vec.sfv_flag = 0; 1097 vec.sfv_flag = 0;
937 vec.sfv_off = offset; 1098 vec.sfv_off = offset;
938 vec.sfv_len = count; 1099 vec.sfv_len = count;
939 1100
940 res = sendfilev (ofd, &vec, 1, &sbytes); 1101 res = sendfilev (ofd, &vec, 1, &sbytes);
941 1102
942 if (res < 0 && sbytes) 1103 if (res < 0 && sbytes)
943 res = sbytes; 1104 res = sbytes;
944 }
945 1105
946# endif 1106# endif
1107
1108#elif defined (_WIN32) && 0
1109 /* does not work, just for documentation of what would need to be done */
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 */
1113 HANDLE h = TO_SOCKET (ifd);
1114 SetFilePointer (h, offset, 0, FILE_BEGIN);
1115 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1116
947#else 1117#else
948 res = -1; 1118 res = -1;
949 errno = ENOSYS; 1119 errno = ENOSYS;
950#endif 1120#endif
951 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
952 if (res < 0 1145 if (res < 0
953 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1146 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1147 /* BSDs */
1148#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1149 || errno == ENOTSUP
1150#endif
1151#ifdef EOPNOTSUPP /* windows */
1152 || errno == EOPNOTSUPP /* BSDs */
1153#endif
954#if __solaris 1154#if __solaris
955 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1155 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
956#endif 1156#endif
957 ) 1157 )
958 ) 1158 )
962 1162
963 res = 0; 1163 res = 0;
964 1164
965 while (count) 1165 while (count)
966 { 1166 {
967 ssize_t cnt; 1167 eio_ssize_t cnt;
968 1168
969 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);
970 1170
971 if (cnt <= 0) 1171 if (cnt <= 0)
972 { 1172 {
989 } 1189 }
990 1190
991 return res; 1191 return res;
992} 1192}
993 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
994static int 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
1473static signed char
995eio_dent_cmp (const void *a_, const void *b_) 1474eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
996{ 1475{
997 const eio_dirent *a = (const eio_dirent *)a_; 1476 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
998 const eio_dirent *b = (const eio_dirent *)b_; 1477 : a->inode < b->inode ? -1
1478 : a->inode > b->inode ? 1
1479 : 0;
1480}
999 1481
1000 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score 1482#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1001 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */ 1483
1484#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1485#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1486
1487static void
1488eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1489{
1490 unsigned char bits [9 + sizeof (ino_t) * 8];
1491 unsigned char *bit = bits;
1492
1493 assert (CHAR_BIT == 8);
1494 assert (sizeof (eio_dirent) * 8 < 256);
1495 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1496 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1497
1498 if (size <= EIO_SORT_FAST)
1499 return;
1500
1501 /* first prepare an array of bits to test in our radix sort */
1502 /* try to take endianness into account, as well as differences in ino_t sizes */
1503 /* inode_bits must contain all inodes ORed together */
1504 /* which is used to skip bits that are 0 everywhere, which is very common */
1505 {
1506 ino_t endianness;
1507 int i, j;
1508
1509 /* we store the byte offset of byte n into byte n of "endianness" */
1510 for (i = 0; i < sizeof (ino_t); ++i)
1511 ((unsigned char *)&endianness)[i] = i;
1512
1513 *bit++ = 0;
1514
1515 for (i = 0; i < sizeof (ino_t); ++i)
1516 {
1517 /* shifting off the byte offsets out of "endianness" */
1518 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1519 endianness >>= 8;
1520
1521 for (j = 0; j < 8; ++j)
1522 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1523 *bit++ = offs + j;
1524 }
1525
1526 for (j = 0; j < 8; ++j)
1527 if (score_bits & (1 << j))
1528 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1529 }
1530
1531 /* now actually do the sorting (a variant of MSD radix sort) */
1532 {
1533 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1534 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1535 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1536 int stk_idx = 0;
1537
1538 base_stk [stk_idx] = dents;
1539 end_stk [stk_idx] = dents + size;
1540 bit_stk [stk_idx] = bit - 1;
1541
1542 do
1543 {
1544 base = base_stk [stk_idx];
1545 end = end_stk [stk_idx];
1546 bit = bit_stk [stk_idx];
1547
1548 for (;;)
1549 {
1550 unsigned char O = *bit >> 3;
1551 unsigned char M = 1 << (*bit & 7);
1552
1553 eio_dirent *a = base;
1554 eio_dirent *b = end;
1555
1556 if (b - a < EIO_SORT_CUTOFF)
1557 break;
1558
1559 /* now bit-partition the array on the bit */
1560 /* this ugly asymmetric loop seems to perform much better than typical */
1561 /* partition algos found in the literature */
1562 do
1563 if (!(((unsigned char *)a)[O] & M))
1564 ++a;
1565 else if (!(((unsigned char *)--b)[O] & M))
1566 {
1567 eio_dirent tmp = *a; *a = *b; *b = tmp;
1568 ++a;
1569 }
1570 while (b > a);
1571
1572 /* next bit, or stop, if no bits left in this path */
1573 if (!*--bit)
1574 break;
1575
1576 base_stk [stk_idx] = a;
1577 end_stk [stk_idx] = end;
1578 bit_stk [stk_idx] = bit;
1579 ++stk_idx;
1580
1581 end = a;
1582 }
1583 }
1584 while (stk_idx--);
1585 }
1586}
1587
1588static void
1589eio_dent_insertion_sort (eio_dirent *dents, int size)
1590{
1591 /* first move the smallest element to the front, to act as a sentinel */
1592 {
1593 int i;
1594 eio_dirent *min = dents;
1595
1596 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1597 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1598 if (EIO_DENT_CMP (dents [i], <, *min))
1599 min = &dents [i];
1600
1601 /* swap elements 0 and j (minimum) */
1602 {
1603 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1604 }
1605 }
1606
1607 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1608 {
1609 eio_dirent *i, *j;
1610
1611 for (i = dents + 1; i < dents + size; ++i)
1612 {
1613 eio_dirent value = *i;
1614
1615 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1616 j [1] = j [0];
1617
1618 j [1] = value;
1619 }
1620 }
1621}
1622
1623static void
1624eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1625{
1626 if (size <= 1)
1627 return; /* our insertion sort relies on size > 0 */
1628
1629 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1630 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1631 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1632
1633 /* use an insertion sort at the end, or for small arrays, */
1634 /* as insertion sort is more efficient for small partitions */
1635 eio_dent_insertion_sort (dents, size);
1002} 1636}
1003 1637
1004/* read a full directory */ 1638/* read a full directory */
1005static void 1639static void
1006eio__scandir (eio_req *req, etp_worker *self) 1640eio__scandir (eio_req *req, etp_worker *self)
1007{ 1641{
1008 DIR *dirp; 1642 DIR *dirp;
1009 EIO_STRUCT_DIRENT *entp; 1643 EIO_STRUCT_DIRENT *entp;
1010 unsigned char *name, *names; 1644 char *name, *names;
1011 int namesalloc = 4096; 1645 int namesalloc = 4096;
1012 int namesoffs = 0; 1646 int namesoffs = 0;
1013 int flags = req->int1; 1647 int flags = req->int1;
1014 eio_dirent *dents = 0; 1648 eio_dirent *dents = 0;
1015 int dentalloc = 128; 1649 int dentalloc = 128;
1016 int dentoffs = 0; 1650 int dentoffs = 0;
1651 ino_t inode_bits = 0;
1017 1652
1018 req->result = -1; 1653 req->result = -1;
1019 1654
1020 if (!(flags & EIO_READDIR_DENTS)) 1655 if (!(flags & EIO_READDIR_DENTS))
1021 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1656 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1022 1657
1023 X_LOCK (wrklock); 1658 X_LOCK (wrklock);
1024 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1659 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1025 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
1026 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1665 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1027 req->ptr1 = names = malloc (namesalloc);
1028 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1666 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1667 req->ptr2 = names = malloc (namesalloc);
1029 X_UNLOCK (wrklock); 1668 X_UNLOCK (wrklock);
1030 1669
1031 if (dirp && names && (!flags || dents)) 1670 if (dirp && names && (!flags || dents))
1032 for (;;) 1671 for (;;)
1033 { 1672 {
1041 1680
1042 /* sort etc. */ 1681 /* sort etc. */
1043 req->int1 = flags; 1682 req->int1 = flags;
1044 req->result = dentoffs; 1683 req->result = dentoffs;
1045 1684
1046 if (dents)
1047 {
1048 eio_dirent *ent = dents + dentoffs;
1049
1050 while (ent > dents)
1051 (--ent)->name = names + (size_t)ent->name;
1052 }
1053
1054 if (flags & EIO_READDIR_STAT_ORDER 1685 if (flags & EIO_READDIR_STAT_ORDER)
1055 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1686 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1056 {
1057 /* pray your qsort doesn't use quicksort */
1058 qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1059 }
1060 else if (flags & EIO_READDIR_DIRS_FIRST) 1687 else if (flags & EIO_READDIR_DIRS_FIRST)
1688 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1689 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1690 else
1061 { 1691 {
1062 /* in this case, all is known, and we just put dirs first and sort them */ 1692 /* in this case, all is known, and we just put dirs first and sort them */
1063 eio_dirent *ent = dents + dentoffs; 1693 eio_dirent *oth = dents + dentoffs;
1064 eio_dirent *dir = dents; 1694 eio_dirent *dir = dents;
1065 1695
1696 /* now partition dirs to the front, and non-dirs to the back */
1697 /* by walking from both sides and swapping if necessary */
1066 while (ent > dir) 1698 while (oth > dir)
1067 { 1699 {
1068 if (dir->type == DT_DIR) 1700 if (dir->type == EIO_DT_DIR)
1069 ++dir; 1701 ++dir;
1070 else
1071 {
1072 --ent;
1073
1074 if (ent->type == DT_DIR) 1702 else if ((--oth)->type == EIO_DT_DIR)
1075 { 1703 {
1076 eio_dirent tmp = *dir; 1704 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1077 *dir = *ent;
1078 *ent = tmp;
1079 1705
1080 ++dir; 1706 ++dir;
1081 } 1707 }
1082 } 1708 }
1709
1710 /* now sort the dirs only (dirs all have the same score) */
1711 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1083 } 1712 }
1084
1085 /* now sort the dirs only */
1086 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1087 }
1088
1089 /* only provide the names array unless DENTS is specified */
1090 if (!(flags & EIO_READDIR_DENTS))
1091 {
1092 X_LOCK (wrklock);
1093 assert (!dents);
1094 req->ptr1 = 0;
1095 req->ptr2 = names;
1096 X_UNLOCK (wrklock);
1097 }
1098 1713
1099 break; 1714 break;
1100 } 1715 }
1101 1716
1102 /* now add the entry to our list(s) */ 1717 /* now add the entry to our list(s) */
1105 /* skip . and .. entries */ 1720 /* skip . and .. entries */
1106 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1721 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1107 { 1722 {
1108 int len = D_NAMLEN (entp) + 1; 1723 int len = D_NAMLEN (entp) + 1;
1109 1724
1110 while (expect_false (namesoffs + len > namesalloc)) 1725 while (ecb_expect_false (namesoffs + len > namesalloc))
1111 { 1726 {
1112 namesalloc *= 2; 1727 namesalloc *= 2;
1113 X_LOCK (wrklock); 1728 X_LOCK (wrklock);
1114 req->ptr1 = names = realloc (names, namesalloc); 1729 req->ptr2 = names = realloc (names, namesalloc);
1115 X_UNLOCK (wrklock); 1730 X_UNLOCK (wrklock);
1116 1731
1117 if (!names) 1732 if (!names)
1118 break; 1733 break;
1119 } 1734 }
1122 1737
1123 if (dents) 1738 if (dents)
1124 { 1739 {
1125 struct eio_dirent *ent; 1740 struct eio_dirent *ent;
1126 1741
1127 if (expect_false (dentoffs == dentalloc)) 1742 if (ecb_expect_false (dentoffs == dentalloc))
1128 { 1743 {
1129 dentalloc *= 2; 1744 dentalloc *= 2;
1130 X_LOCK (wrklock); 1745 X_LOCK (wrklock);
1131 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1746 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1132 X_UNLOCK (wrklock); 1747 X_UNLOCK (wrklock);
1133 1748
1134 if (!dents) 1749 if (!dents)
1135 break; 1750 break;
1136 } 1751 }
1137 1752
1138 ent = dents + dentoffs; 1753 ent = dents + dentoffs;
1139 1754
1140 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1755 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1141 ent->namelen = len - 1; 1756 ent->namelen = len - 1;
1142 ent->inode = D_INO (entp); 1757 ent->inode = D_INO (entp);
1758
1759 inode_bits |= ent->inode;
1143 1760
1144 switch (D_TYPE (entp)) 1761 switch (D_TYPE (entp))
1145 { 1762 {
1146 default: 1763 default:
1147 ent->type = EIO_DT_UNKNOWN; 1764 ent->type = EIO_DT_UNKNOWN;
1190 #ifdef DT_WHT 1807 #ifdef DT_WHT
1191 case DT_WHT: ent->type = EIO_DT_WHT; break; 1808 case DT_WHT: ent->type = EIO_DT_WHT; break;
1192 #endif 1809 #endif
1193 } 1810 }
1194 1811
1195 ent->score = 0; 1812 ent->score = 7;
1196 1813
1197 if (flags & EIO_READDIR_DIRS_FIRST) 1814 if (flags & EIO_READDIR_DIRS_FIRST)
1198 { 1815 {
1199 if (ent->type == EIO_DT_UNKNOWN) 1816 if (ent->type == EIO_DT_UNKNOWN)
1200 { 1817 {
1201 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1818 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1202 ent->score = 98; 1819 ent->score = 1;
1203 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1820 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1204 ent->score = len <= 2 ? len + 60 : len <= 4 ? 50 : len <= 7 ? 40 : 10; /* shorter == more likely dir, but avoid too many classes */ 1821 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1205 } 1822 }
1206 else if (ent->type == EIO_DT_DIR) 1823 else if (ent->type == EIO_DT_DIR)
1207 ent->score = 100; 1824 ent->score = 0;
1208 } 1825 }
1209 } 1826 }
1210 1827
1211 namesoffs += len; 1828 namesoffs += len;
1212 ++dentoffs; 1829 ++dentoffs;
1213 } 1830 }
1831
1832 if (EIO_CANCELLED (req))
1833 {
1834 errno = ECANCELED;
1835 break;
1836 }
1214 } 1837 }
1215 else
1216 req->result = -1;
1217}
1218
1219#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1220# undef msync
1221# define msync(a,b,c) ((errno = ENOSYS), -1)
1222#endif
1223
1224int
1225eio__mtouch (void *mem, size_t len, int flags)
1226{
1227 intptr_t addr = (intptr_t)mem;
1228 intptr_t end = addr + len;
1229#ifdef PAGESIZE
1230 const intptr_t page = PAGESIZE;
1231#else
1232 static intptr_t page;
1233
1234 if (!page)
1235 page = sysconf (_SC_PAGESIZE);
1236#endif
1237
1238 addr &= ~(page - 1); /* assume page size is always a power of two */
1239
1240 if (addr < end)
1241 if (flags) /* modify */
1242 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1243 else
1244 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1245
1246 return 0;
1247} 1838}
1248 1839
1249/*****************************************************************************/ 1840/*****************************************************************************/
1250 1841
1251#define ALLOC(len) \ 1842#define ALLOC(len) \
1266X_THREAD_PROC (etp_proc) 1857X_THREAD_PROC (etp_proc)
1267{ 1858{
1268 ETP_REQ *req; 1859 ETP_REQ *req;
1269 struct timespec ts; 1860 struct timespec ts;
1270 etp_worker *self = (etp_worker *)thr_arg; 1861 etp_worker *self = (etp_worker *)thr_arg;
1862 int timeout;
1271 1863
1272 /* try to distribute timeouts somewhat randomly */ 1864 /* try to distribute timeouts somewhat evenly */
1273 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1865 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1274 1866
1275 for (;;) 1867 for (;;)
1276 { 1868 {
1869 ts.tv_sec = 0;
1870
1277 X_LOCK (reqlock); 1871 X_LOCK (reqlock);
1278 1872
1279 for (;;) 1873 for (;;)
1280 { 1874 {
1281 self->req = req = reqq_shift (&req_queue); 1875 self->req = req = reqq_shift (&req_queue);
1282 1876
1283 if (req) 1877 if (req)
1284 break; 1878 break;
1285 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
1286 ++idle; 1889 ++idle;
1287 1890
1288 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1891 if (idle <= max_idle)
1289 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
1290 { 1895 {
1291 if (idle > max_idle) 1896 /* initialise timeout once */
1292 { 1897 if (!ts.tv_sec)
1293 --idle; 1898 ts.tv_sec = time (0) + idle_timeout;
1294 X_UNLOCK (reqlock);
1295 X_LOCK (wrklock);
1296 --started;
1297 X_UNLOCK (wrklock);
1298 goto quit;
1299 }
1300 1899
1301 /* we are allowed to idle, so do so without any timeout */
1302 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.,.. */
1303 } 1902 }
1304 1903
1305 --idle; 1904 --idle;
1306 } 1905 }
1307 1906
1310 X_UNLOCK (reqlock); 1909 X_UNLOCK (reqlock);
1311 1910
1312 if (req->type < 0) 1911 if (req->type < 0)
1313 goto quit; 1912 goto quit;
1314 1913
1315 if (!EIO_CANCELLED (req))
1316 ETP_EXECUTE (self, req); 1914 ETP_EXECUTE (self, req);
1317 1915
1318 X_LOCK (reslock); 1916 X_LOCK (reslock);
1319 1917
1320 ++npending; 1918 ++npending;
1321 1919
1336 return 0; 1934 return 0;
1337} 1935}
1338 1936
1339/*****************************************************************************/ 1937/*****************************************************************************/
1340 1938
1939int ecb_cold
1341int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1940eio_init (void (*want_poll)(void), void (*done_poll)(void))
1342{ 1941{
1343 return etp_init (want_poll, done_poll); 1942 return etp_init (want_poll, done_poll);
1344} 1943}
1345 1944
1945ecb_inline void
1346static void eio_api_destroy (eio_req *req) 1946eio_api_destroy (eio_req *req)
1347{ 1947{
1348 free (req); 1948 free (req);
1349} 1949}
1350 1950
1351#define REQ(rtype) \ 1951#define REQ(rtype) \
1370 { \ 1970 { \
1371 eio_api_destroy (req); \ 1971 eio_api_destroy (req); \
1372 return 0; \ 1972 return 0; \
1373 } 1973 }
1374 1974
1975static void
1375static void eio_execute (etp_worker *self, eio_req *req) 1976eio_execute (etp_worker *self, eio_req *req)
1376{ 1977{
1377 errno = 0; 1978 if (ecb_expect_false (EIO_CANCELLED (req)))
1979 {
1980 req->result = -1;
1981 req->errorno = ECANCELED;
1982 return;
1983 }
1378 1984
1379 switch (req->type) 1985 switch (req->type)
1380 { 1986 {
1381 case EIO_READ: ALLOC (req->size); 1987 case EIO_READ: ALLOC (req->size);
1382 req->result = req->offs >= 0 1988 req->result = req->offs >= 0
1394 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2000 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1395 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2001 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1396 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2002 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1397 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2003 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1398 2004
2005 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2006 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2007 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2008 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2009
1399 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2010 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1400 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2011 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1401 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2012 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1402 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2013 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1403 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2014 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1410 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2021 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1411 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;
1412 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2023 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1413 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2024 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1414 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2025 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1415 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;
1416 2027
2028 case EIO_REALPATH: eio__realpath (req, self); break;
2029
1417 case EIO_READLINK: ALLOC (NAME_MAX); 2030 case EIO_READLINK: ALLOC (PATH_MAX);
1418 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2031 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1419 2032
1420 case EIO_SYNC: req->result = 0; sync (); break; 2033 case EIO_SYNC: req->result = 0; sync (); break;
1421 case EIO_FSYNC: req->result = fsync (req->int1); break; 2034 case EIO_FSYNC: req->result = fsync (req->int1); break;
1422 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2035 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1423 case EIO_MSYNC: req->result = 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;
1424 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;
1425 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;
1426 2042
1427 case EIO_READDIR: eio__scandir (req, self); break; 2043 case EIO_READDIR: eio__scandir (req, self); break;
1428 2044
1429 case EIO_BUSY: 2045 case EIO_BUSY:
1430#ifdef _WIN32 2046#ifdef _WIN32
1431 Sleep (req->nv1 * 1000.); 2047 Sleep (req->nv1 * 1e3);
1432#else 2048#else
1433 { 2049 {
1434 struct timeval tv; 2050 struct timeval tv;
1435 2051
1436 tv.tv_sec = req->nv1; 2052 tv.tv_sec = req->nv1;
1437 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2053 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1438 2054
1439 req->result = select (0, 0, 0, 0, &tv); 2055 req->result = select (0, 0, 0, 0, &tv);
1440 } 2056 }
1441#endif 2057#endif
1442 break; 2058 break;
1457 times = tv; 2073 times = tv;
1458 } 2074 }
1459 else 2075 else
1460 times = 0; 2076 times = 0;
1461 2077
1462
1463 req->result = req->type == EIO_FUTIME 2078 req->result = req->type == EIO_FUTIME
1464 ? futimes (req->int1, times) 2079 ? futimes (req->int1, times)
1465 : utimes (req->ptr1, times); 2080 : utimes (req->ptr1, times);
1466 } 2081 }
1467 break; 2082 break;
1472 case EIO_NOP: 2087 case EIO_NOP:
1473 req->result = 0; 2088 req->result = 0;
1474 break; 2089 break;
1475 2090
1476 case EIO_CUSTOM: 2091 case EIO_CUSTOM:
1477 ((void (*)(eio_req *))req->feed) (req); 2092 req->feed (req);
1478 break; 2093 break;
1479 2094
1480 default: 2095 default:
2096 errno = ENOSYS;
1481 req->result = -1; 2097 req->result = -1;
1482 break; 2098 break;
1483 } 2099 }
1484 2100
1485 req->errorno = errno; 2101 req->errorno = errno;
1515eio_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)
1516{ 2132{
1517 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;
1518} 2134}
1519 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
1520eio_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)
1521{ 2147{
1522 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;
1523} 2149}
1524 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
1525eio_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)
1526{ 2157{
1527 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2158 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1528} 2159}
1529 2160
1550eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2181eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1551{ 2182{
1552 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2183 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1553} 2184}
1554 2185
2186eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2187{
2188 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2189}
2190
1555eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2191eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1556{ 2192{
1557 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2193 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1558} 2194}
1559 2195
1565eio_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)
1566{ 2202{
1567 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2203 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1568} 2204}
1569 2205
1570eio_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)
1571{ 2207{
1572 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;
1573} 2209}
1574 2210
1575eio_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)
1595eio_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)
1596{ 2232{
1597 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2233 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1598} 2234}
1599 2235
1600eio_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)
1601{ 2237{
1602 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;
1603} 2239}
1604 2240
1605eio_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)
1621eio_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)
1622{ 2258{
1623 return eio__1path (EIO_READLINK, path, pri, cb, data); 2259 return eio__1path (EIO_READLINK, path, pri, cb, data);
1624} 2260}
1625 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
1626eio_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)
1627{ 2268{
1628 return eio__1path (EIO_STAT, path, pri, cb, data); 2269 return eio__1path (EIO_STAT, path, pri, cb, data);
1629} 2270}
1630 2271
1631eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2272eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1632{ 2273{
1633 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2274 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1634} 2275}
1635 2276
2277eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2278{
2279 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2280}
2281
1636eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2282eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1637{ 2283{
1638 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2284 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1639} 2285}
1640 2286
1648 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2294 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1649} 2295}
1650 2296
1651eio_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)
1652{ 2298{
1653 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;
1654} 2300}
1655 2301
1656static eio_req * 2302static eio_req *
1657eio__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)
1658{ 2304{
1682eio_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)
1683{ 2329{
1684 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2330 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1685} 2331}
1686 2332
1687eio_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)
1688{ 2334{
1689 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2335 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1690} 2336}
1691 2337
1692#endif 2338#endif
1693 2339
1694eio_req *eio_grp (eio_cb cb, void *data) 2340eio_req *eio_grp (eio_cb cb, void *data)
1703#undef SEND 2349#undef SEND
1704 2350
1705/*****************************************************************************/ 2351/*****************************************************************************/
1706/* grp functions */ 2352/* grp functions */
1707 2353
2354void
1708void 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)
1709{ 2356{
1710 grp->int2 = limit; 2357 grp->int2 = limit;
1711 grp->feed = feed; 2358 grp->feed = feed;
1712 2359
1713 grp_try_feed (grp); 2360 grp_try_feed (grp);
1714} 2361}
1715 2362
2363void
1716void eio_grp_limit (eio_req *grp, int limit) 2364eio_grp_limit (eio_req *grp, int limit)
1717{ 2365{
1718 grp->int2 = limit; 2366 grp->int2 = limit;
1719 2367
1720 grp_try_feed (grp); 2368 grp_try_feed (grp);
1721} 2369}
1722 2370
2371void
1723void eio_grp_add (eio_req *grp, eio_req *req) 2372eio_grp_add (eio_req *grp, eio_req *req)
1724{ 2373{
1725 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));
1726 2375
1727 grp->flags |= EIO_FLAG_GROUPADD; 2376 grp->flags |= EIO_FLAG_GROUPADD;
1728 2377
1739} 2388}
1740 2389
1741/*****************************************************************************/ 2390/*****************************************************************************/
1742/* misc garbage */ 2391/* misc garbage */
1743 2392
2393eio_ssize_t
1744ssize_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)
1745{ 2395{
1746 etp_worker wrk; 2396 etp_worker wrk;
2397 eio_ssize_t ret;
1747 2398
1748 wrk.dbuf = 0; 2399 wrk.dbuf = 0;
1749 2400
1750 eio__sendfile (ofd, ifd, offset, count, &wrk); 2401 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1751 2402
1752 if (wrk.dbuf) 2403 if (wrk.dbuf)
1753 free (wrk.dbuf); 2404 free (wrk.dbuf);
1754}
1755 2405
2406 return ret;
2407}
2408

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines