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 1.61 by root, Fri Feb 11 00:53:24 2011 UTC vs.
Revision 1.81 by root, Tue Jul 5 20:34:42 2011 UTC

… …
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"
… …
54#include <sys/statvfs.h> 59#include <sys/statvfs.h>
55#include <limits.h> 60#include <limits.h>
56#include <fcntl.h> 61#include <fcntl.h>
57#include <assert.h> 62#include <assert.h>
58 63
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H
67# include <stdint.h>
68#endif
69
70#ifndef ECANCELED
71# define ECANCELED EDOM
72#endif
73
74static void eio_destroy (eio_req *req);
75
59#ifndef EIO_FINISH 76#ifndef EIO_FINISH
60# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 77# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
61#endif 78#endif
62 79
63#ifndef EIO_DESTROY 80#ifndef EIO_DESTROY
… …
71#ifdef _WIN32 88#ifdef _WIN32
72 89
73 /*doh*/ 90 /*doh*/
74#else 91#else
75 92
76# include "config.h"
77# include <sys/time.h> 93# include <sys/time.h>
78# include <sys/select.h> 94# include <sys/select.h>
79# include <unistd.h> 95# include <unistd.h>
80# include <utime.h> 96# include <utime.h>
81# include <signal.h> 97# include <signal.h>
… …
132#endif 148#endif
133#ifndef D_NAMLEN 149#ifndef D_NAMLEN
134# define D_NAMLEN(de) strlen ((de)->d_name) 150# define D_NAMLEN(de) strlen ((de)->d_name)
135#endif 151#endif
136 152
137/* number of seconds after which an idle threads exit */
138#define IDLE_TIMEOUT 10
139
140/* used for struct dirent, AIX doesn't provide it */ 153/* used for struct dirent, AIX doesn't provide it */
141#ifndef NAME_MAX 154#ifndef NAME_MAX
142# define NAME_MAX 4096 155# define NAME_MAX 4096
143#endif 156#endif
144 157
… …
158 errno = ENOMEM; \ 171 errno = ENOMEM; \
159 if (!eio_buf) \ 172 if (!eio_buf) \
160 return -1; 173 return -1;
161 174
162#define EIO_TICKS ((1000000 + 1023) >> 10) 175#define EIO_TICKS ((1000000 + 1023) >> 10)
163
164/*****************************************************************************/
165
166#if __GNUC__ >= 3
167# define expect(expr,value) __builtin_expect ((expr),(value))
168#else
169# define expect(expr,value) (expr)
170#endif
171
172#define expect_false(expr) expect ((expr) != 0, 0)
173#define expect_true(expr) expect ((expr) != 0, 1)
174
175/*****************************************************************************/
176 176
177#define ETP_PRI_MIN EIO_PRI_MIN 177#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 178#define ETP_PRI_MAX EIO_PRI_MAX
179 179
180struct etp_worker; 180struct etp_worker;
… …
206/*****************************************************************************/ 206/*****************************************************************************/
207 207
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209 209
210/* calculate time difference in ~1/EIO_TICKS of a second */ 210/* calculate time difference in ~1/EIO_TICKS of a second */
211ecb_inline int
211static int tvdiff (struct timeval *tv1, struct timeval *tv2) 212tvdiff (struct timeval *tv1, struct timeval *tv2)
212{ 213{
213 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 214 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
214 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 215 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
215} 216}
216 217
… …
223static unsigned int max_poll_reqs; /* reslock */ 224static unsigned int max_poll_reqs; /* reslock */
224 225
225static volatile unsigned int nreqs; /* reqlock */ 226static volatile unsigned int nreqs; /* reqlock */
226static volatile unsigned int nready; /* reqlock */ 227static volatile unsigned int nready; /* reqlock */
227static volatile unsigned int npending; /* reqlock */ 228static volatile unsigned int npending; /* reqlock */
228static volatile unsigned int max_idle = 4; 229static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
230static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
229 231
230static xmutex_t wrklock = X_MUTEX_INIT; 232static xmutex_t wrklock;
231static xmutex_t reslock = X_MUTEX_INIT; 233static xmutex_t reslock;
232static xmutex_t reqlock = X_MUTEX_INIT; 234static xmutex_t reqlock;
233static xcond_t reqwait = X_COND_INIT; 235static xcond_t reqwait;
234 236
235#if !HAVE_PREADWRITE 237#if !HAVE_PREADWRITE
236/* 238/*
237 * make our pread/pwrite emulation safe against themselves, but not against 239 * make our pread/pwrite emulation safe against themselves, but not against
238 * normal read/write by using a mutex. slows down execution a lot, 240 * normal read/write by using a mutex. slows down execution a lot,
… …
259#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 261#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
260#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 262#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
261 263
262/* worker threads management */ 264/* worker threads management */
263 265
266static void ecb_cold
264static void etp_worker_clear (etp_worker *wrk) 267etp_worker_clear (etp_worker *wrk)
265{ 268{
266 ETP_WORKER_CLEAR (wrk); 269 ETP_WORKER_CLEAR (wrk);
267} 270}
268 271
272static void ecb_cold
269static void etp_worker_free (etp_worker *wrk) 273etp_worker_free (etp_worker *wrk)
270{ 274{
271 wrk->next->prev = wrk->prev; 275 wrk->next->prev = wrk->prev;
272 wrk->prev->next = wrk->next; 276 wrk->prev->next = wrk->next;
273 277
274 free (wrk); 278 free (wrk);
275} 279}
276 280
277static unsigned int etp_nreqs (void) 281static unsigned int
282etp_nreqs (void)
278{ 283{
279 int retval; 284 int retval;
280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 285 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281 retval = nreqs; 286 retval = nreqs;
282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 287 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283 return retval; 288 return retval;
284} 289}
285 290
286static unsigned int etp_nready (void) 291static unsigned int
292etp_nready (void)
287{ 293{
288 unsigned int retval; 294 unsigned int retval;
289 295
290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 296 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
291 retval = nready; 297 retval = nready;
292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 298 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
293 299
294 return retval; 300 return retval;
295} 301}
296 302
297static unsigned int etp_npending (void) 303static unsigned int
304etp_npending (void)
298{ 305{
299 unsigned int retval; 306 unsigned int retval;
300 307
301 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 308 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
302 retval = npending; 309 retval = npending;
303 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 310 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
304 311
305 return retval; 312 return retval;
306} 313}
307 314
308static unsigned int etp_nthreads (void) 315static unsigned int
316etp_nthreads (void)
309{ 317{
310 unsigned int retval; 318 unsigned int retval;
311 319
312 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 320 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
313 retval = started; 321 retval = started;
… …
327} etp_reqq; 335} etp_reqq;
328 336
329static etp_reqq req_queue; 337static etp_reqq req_queue;
330static etp_reqq res_queue; 338static etp_reqq res_queue;
331 339
340static int ecb_noinline
332static int reqq_push (etp_reqq *q, ETP_REQ *req) 341reqq_push (etp_reqq *q, ETP_REQ *req)
333{ 342{
334 int pri = req->pri; 343 int pri = req->pri;
335 req->next = 0; 344 req->next = 0;
336 345
337 if (q->qe[pri]) 346 if (q->qe[pri])
… …
343 q->qe[pri] = q->qs[pri] = req; 352 q->qe[pri] = q->qs[pri] = req;
344 353
345 return q->size++; 354 return q->size++;
346} 355}
347 356
357static ETP_REQ * ecb_noinline
348static ETP_REQ *reqq_shift (etp_reqq *q) 358reqq_shift (etp_reqq *q)
349{ 359{
350 int pri; 360 int pri;
351 361
352 if (!q->size) 362 if (!q->size)
353 return 0; 363 return 0;
… …
368 } 378 }
369 379
370 abort (); 380 abort ();
371} 381}
372 382
383static void ecb_cold
384etp_thread_init (void)
385{
386 X_MUTEX_CREATE (wrklock);
387 X_MUTEX_CREATE (reslock);
388 X_MUTEX_CREATE (reqlock);
389 X_COND_CREATE (reqwait);
390}
391
392static void ecb_cold
373static void etp_atfork_prepare (void) 393etp_atfork_prepare (void)
374{ 394{
375 X_LOCK (wrklock); 395 X_LOCK (wrklock);
376 X_LOCK (reqlock); 396 X_LOCK (reqlock);
377 X_LOCK (reslock); 397 X_LOCK (reslock);
378#if !HAVE_PREADWRITE 398#if !HAVE_PREADWRITE
379 X_LOCK (preadwritelock); 399 X_LOCK (preadwritelock);
380#endif 400#endif
381} 401}
382 402
403static void ecb_cold
383static void etp_atfork_parent (void) 404etp_atfork_parent (void)
384{ 405{
385#if !HAVE_PREADWRITE 406#if !HAVE_PREADWRITE
386 X_UNLOCK (preadwritelock); 407 X_UNLOCK (preadwritelock);
387#endif 408#endif
388 X_UNLOCK (reslock); 409 X_UNLOCK (reslock);
389 X_UNLOCK (reqlock); 410 X_UNLOCK (reqlock);
390 X_UNLOCK (wrklock); 411 X_UNLOCK (wrklock);
391} 412}
392 413
414static void ecb_cold
393static void etp_atfork_child (void) 415etp_atfork_child (void)
394{ 416{
395 ETP_REQ *prv; 417 ETP_REQ *prv;
396 418
397 while ((prv = reqq_shift (&req_queue))) 419 while ((prv = reqq_shift (&req_queue)))
398 ETP_DESTROY (prv); 420 ETP_DESTROY (prv);
… …
415 idle = 0; 437 idle = 0;
416 nreqs = 0; 438 nreqs = 0;
417 nready = 0; 439 nready = 0;
418 npending = 0; 440 npending = 0;
419 441
420 etp_atfork_parent (); 442 etp_thread_init ();
421} 443}
422 444
423static void 445static void ecb_cold
424etp_once_init (void) 446etp_once_init (void)
425{ 447{
448 etp_thread_init ();
426 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 449 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
427} 450}
428 451
429static int 452static int ecb_cold
430etp_init (void (*want_poll)(void), void (*done_poll)(void)) 453etp_init (void (*want_poll)(void), void (*done_poll)(void))
431{ 454{
432 static pthread_once_t doinit = PTHREAD_ONCE_INIT; 455 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
433 456
434 pthread_once (&doinit, etp_once_init); 457 pthread_once (&doinit, etp_once_init);
… …
439 return 0; 462 return 0;
440} 463}
441 464
442X_THREAD_PROC (etp_proc); 465X_THREAD_PROC (etp_proc);
443 466
467static void ecb_cold
444static void etp_start_thread (void) 468etp_start_thread (void)
445{ 469{
446 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 470 etp_worker *wrk = calloc (1, sizeof (etp_worker));
447 471
448 /*TODO*/ 472 /*TODO*/
449 assert (("unable to allocate worker thread data", wrk)); 473 assert (("unable to allocate worker thread data", wrk));
… …
462 free (wrk); 486 free (wrk);
463 487
464 X_UNLOCK (wrklock); 488 X_UNLOCK (wrklock);
465} 489}
466 490
491static void
467static void etp_maybe_start_thread (void) 492etp_maybe_start_thread (void)
468{ 493{
469 if (expect_true (etp_nthreads () >= wanted)) 494 if (ecb_expect_true (etp_nthreads () >= wanted))
470 return; 495 return;
471 496
472 /* todo: maybe use idle here, but might be less exact */ 497 /* todo: maybe use idle here, but might be less exact */
473 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 498 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
474 return; 499 return;
475 500
476 etp_start_thread (); 501 etp_start_thread ();
477} 502}
478 503
504static void ecb_cold
479static void etp_end_thread (void) 505etp_end_thread (void)
480{ 506{
481 eio_req *req = calloc (1, sizeof (eio_req)); 507 eio_req *req = calloc (1, sizeof (eio_req));
482 508
483 req->type = -1; 509 req->type = -1;
484 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 510 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
… …
491 X_LOCK (wrklock); 517 X_LOCK (wrklock);
492 --started; 518 --started;
493 X_UNLOCK (wrklock); 519 X_UNLOCK (wrklock);
494} 520}
495 521
496static int etp_poll (void) 522static int
523etp_poll (void)
497{ 524{
498 unsigned int maxreqs; 525 unsigned int maxreqs;
499 unsigned int maxtime; 526 unsigned int maxtime;
500 struct timeval tv_start, tv_now; 527 struct timeval tv_start, tv_now;
501 528
… …
531 558
532 X_LOCK (reqlock); 559 X_LOCK (reqlock);
533 --nreqs; 560 --nreqs;
534 X_UNLOCK (reqlock); 561 X_UNLOCK (reqlock);
535 562
536 if (expect_false (req->type == EIO_GROUP && req->size)) 563 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
537 { 564 {
538 req->int1 = 1; /* mark request as delayed */ 565 req->int1 = 1; /* mark request as delayed */
539 continue; 566 continue;
540 } 567 }
541 else 568 else
542 { 569 {
543 int res = ETP_FINISH (req); 570 int res = ETP_FINISH (req);
544 if (expect_false (res)) 571 if (ecb_expect_false (res))
545 return res; 572 return res;
546 } 573 }
547 574
548 if (expect_false (maxreqs && !--maxreqs)) 575 if (ecb_expect_false (maxreqs && !--maxreqs))
549 break; 576 break;
550 577
551 if (maxtime) 578 if (maxtime)
552 { 579 {
553 gettimeofday (&tv_now, 0); 580 gettimeofday (&tv_now, 0);
… …
559 586
560 errno = EAGAIN; 587 errno = EAGAIN;
561 return -1; 588 return -1;
562} 589}
563 590
591static void
564static void etp_cancel (ETP_REQ *req) 592etp_cancel (ETP_REQ *req)
565{ 593{
566 X_LOCK (wrklock); 594 req->cancelled = 1;
567 req->flags |= EIO_FLAG_CANCELLED;
568 X_UNLOCK (wrklock);
569 595
570 eio_grp_cancel (req); 596 eio_grp_cancel (req);
571} 597}
572 598
599static void
573static void etp_submit (ETP_REQ *req) 600etp_submit (ETP_REQ *req)
574{ 601{
575 req->pri -= ETP_PRI_MIN; 602 req->pri -= ETP_PRI_MIN;
576 603
577 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 604 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
578 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 605 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
579 606
580 if (expect_false (req->type == EIO_GROUP)) 607 if (ecb_expect_false (req->type == EIO_GROUP))
581 { 608 {
582 /* I hope this is worth it :/ */ 609 /* I hope this is worth it :/ */
583 X_LOCK (reqlock); 610 X_LOCK (reqlock);
584 ++nreqs; 611 ++nreqs;
585 X_UNLOCK (reqlock); 612 X_UNLOCK (reqlock);
… …
604 631
605 etp_maybe_start_thread (); 632 etp_maybe_start_thread ();
606 } 633 }
607} 634}
608 635
636static void ecb_cold
609static void etp_set_max_poll_time (double nseconds) 637etp_set_max_poll_time (double nseconds)
610{ 638{
611 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 639 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
612 max_poll_time = nseconds * EIO_TICKS; 640 max_poll_time = nseconds * EIO_TICKS;
613 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 641 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
614} 642}
615 643
644static void ecb_cold
616static void etp_set_max_poll_reqs (unsigned int maxreqs) 645etp_set_max_poll_reqs (unsigned int maxreqs)
617{ 646{
618 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 647 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
619 max_poll_reqs = maxreqs; 648 max_poll_reqs = maxreqs;
620 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 649 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
621} 650}
622 651
652static void ecb_cold
623static void etp_set_max_idle (unsigned int nthreads) 653etp_set_max_idle (unsigned int nthreads)
624{ 654{
625 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 655 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
626 max_idle = nthreads <= 0 ? 1 : nthreads; 656 max_idle = nthreads;
627 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 657 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
628} 658}
629 659
660static void ecb_cold
661etp_set_idle_timeout (unsigned int seconds)
662{
663 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
664 idle_timeout = seconds;
665 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
666}
667
668static void ecb_cold
630static void etp_set_min_parallel (unsigned int nthreads) 669etp_set_min_parallel (unsigned int nthreads)
631{ 670{
632 if (wanted < nthreads) 671 if (wanted < nthreads)
633 wanted = nthreads; 672 wanted = nthreads;
634} 673}
635 674
675static void ecb_cold
636static void etp_set_max_parallel (unsigned int nthreads) 676etp_set_max_parallel (unsigned int nthreads)
637{ 677{
638 if (wanted > nthreads) 678 if (wanted > nthreads)
639 wanted = nthreads; 679 wanted = nthreads;
640 680
641 while (started > wanted) 681 while (started > wanted)
642 etp_end_thread (); 682 etp_end_thread ();
643} 683}
644 684
645/*****************************************************************************/ 685/*****************************************************************************/
646 686
687static void
647static void grp_try_feed (eio_req *grp) 688grp_try_feed (eio_req *grp)
648{ 689{
649 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 690 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
650 { 691 {
651 grp->flags &= ~EIO_FLAG_GROUPADD; 692 grp->flags &= ~EIO_FLAG_GROUPADD;
652 693
… …
659 break; 700 break;
660 } 701 }
661 } 702 }
662} 703}
663 704
705static int
664static int grp_dec (eio_req *grp) 706grp_dec (eio_req *grp)
665{ 707{
666 --grp->size; 708 --grp->size;
667 709
668 /* call feeder, if applicable */ 710 /* call feeder, if applicable */
669 grp_try_feed (grp); 711 grp_try_feed (grp);
… …
673 return eio_finish (grp); 715 return eio_finish (grp);
674 else 716 else
675 return 0; 717 return 0;
676} 718}
677 719
720static void
678void eio_destroy (eio_req *req) 721eio_destroy (eio_req *req)
679{ 722{
680 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 723 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
681 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 724 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
682 725
683 EIO_DESTROY (req); 726 EIO_DESTROY (req);
684} 727}
685 728
729static int
686static int eio_finish (eio_req *req) 730eio_finish (eio_req *req)
687{ 731{
688 int res = EIO_FINISH (req); 732 int res = EIO_FINISH (req);
689 733
690 if (req->grp) 734 if (req->grp)
691 { 735 {
… …
699 if (grp->grp_first == req) 743 if (grp->grp_first == req)
700 grp->grp_first = req->grp_next; 744 grp->grp_first = req->grp_next;
701 745
702 res2 = grp_dec (grp); 746 res2 = grp_dec (grp);
703 747
704 if (!res && res2) 748 if (!res)
705 res = res2; 749 res = res2;
706 } 750 }
707 751
708 eio_destroy (req); 752 eio_destroy (req);
709 753
710 return res; 754 return res;
711} 755}
712 756
757void
713void eio_grp_cancel (eio_req *grp) 758eio_grp_cancel (eio_req *grp)
714{ 759{
715 for (grp = grp->grp_first; grp; grp = grp->grp_next) 760 for (grp = grp->grp_first; grp; grp = grp->grp_next)
716 eio_cancel (grp); 761 eio_cancel (grp);
717} 762}
718 763
764void
719void eio_cancel (eio_req *req) 765eio_cancel (eio_req *req)
720{ 766{
721 etp_cancel (req); 767 etp_cancel (req);
722} 768}
723 769
770void
724void eio_submit (eio_req *req) 771eio_submit (eio_req *req)
725{ 772{
726 etp_submit (req); 773 etp_submit (req);
727} 774}
728 775
729unsigned int eio_nreqs (void) 776unsigned int
777eio_nreqs (void)
730{ 778{
731 return etp_nreqs (); 779 return etp_nreqs ();
732} 780}
733 781
734unsigned int eio_nready (void) 782unsigned int
783eio_nready (void)
735{ 784{
736 return etp_nready (); 785 return etp_nready ();
737} 786}
738 787
739unsigned int eio_npending (void) 788unsigned int
789eio_npending (void)
740{ 790{
741 return etp_npending (); 791 return etp_npending ();
742} 792}
743 793
744unsigned int eio_nthreads (void) 794unsigned int ecb_cold
795eio_nthreads (void)
745{ 796{
746 return etp_nthreads (); 797 return etp_nthreads ();
747} 798}
748 799
800void ecb_cold
749void eio_set_max_poll_time (double nseconds) 801eio_set_max_poll_time (double nseconds)
750{ 802{
751 etp_set_max_poll_time (nseconds); 803 etp_set_max_poll_time (nseconds);
752} 804}
753 805
806void ecb_cold
754void eio_set_max_poll_reqs (unsigned int maxreqs) 807eio_set_max_poll_reqs (unsigned int maxreqs)
755{ 808{
756 etp_set_max_poll_reqs (maxreqs); 809 etp_set_max_poll_reqs (maxreqs);
757} 810}
758 811
812void ecb_cold
759void eio_set_max_idle (unsigned int nthreads) 813eio_set_max_idle (unsigned int nthreads)
760{ 814{
761 etp_set_max_idle (nthreads); 815 etp_set_max_idle (nthreads);
762} 816}
763 817
818void ecb_cold
819eio_set_idle_timeout (unsigned int seconds)
820{
821 etp_set_idle_timeout (seconds);
822}
823
824void ecb_cold
764void eio_set_min_parallel (unsigned int nthreads) 825eio_set_min_parallel (unsigned int nthreads)
765{ 826{
766 etp_set_min_parallel (nthreads); 827 etp_set_min_parallel (nthreads);
767} 828}
768 829
830void ecb_cold
769void eio_set_max_parallel (unsigned int nthreads) 831eio_set_max_parallel (unsigned int nthreads)
770{ 832{
771 etp_set_max_parallel (nthreads); 833 etp_set_max_parallel (nthreads);
772} 834}
773 835
774int eio_poll (void) 836int eio_poll (void)
… …
844#ifndef HAVE_FUTIMES 906#ifndef HAVE_FUTIMES
845 907
846# undef futimes 908# undef futimes
847# define futimes(fd,times) eio__futimes (fd, times) 909# define futimes(fd,times) eio__futimes (fd, times)
848 910
911static int
849static int eio__futimes (int fd, const struct timeval tv[2]) 912eio__futimes (int fd, const struct timeval tv[2])
850{ 913{
851 errno = ENOSYS; 914 errno = ENOSYS;
852 return -1; 915 return -1;
853} 916}
854 917
… …
858# undef fdatasync 921# undef fdatasync
859# define fdatasync(fd) fsync (fd) 922# define fdatasync(fd) fsync (fd)
860#endif 923#endif
861 924
862/* sync_file_range always needs emulation */ 925/* sync_file_range always needs emulation */
863int 926static int
864eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 927eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
865{ 928{
866#if HAVE_SYNC_FILE_RANGE 929#if HAVE_SYNC_FILE_RANGE
867 int res; 930 int res;
868 931
… …
914 977
915/* sendfile always needs emulation */ 978/* sendfile always needs emulation */
916static ssize_t 979static ssize_t
917eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 980eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
918{ 981{
982 ssize_t written = 0;
919 ssize_t res; 983 ssize_t res;
920 984
921 if (!count) 985 if (!count)
922 return 0; 986 return 0;
923 987
988 for (;;)
989 {
924#if HAVE_SENDFILE 990#if HAVE_SENDFILE
925# if __linux 991# if __linux
992 off_t soffset = offset;
926 res = sendfile (ofd, ifd, &offset, count); 993 res = sendfile (ofd, ifd, &soffset, count);
927 994
928# elif __FreeBSD__ 995# elif __FreeBSD__
929 /* 996 /*
930 * Of course, the freebsd sendfile is a dire hack with no thoughts 997 * Of course, the freebsd sendfile is a dire hack with no thoughts
931 * wasted on making it similar to other I/O functions. 998 * wasted on making it similar to other I/O functions.
932 */ 999 */
933 {
934 off_t sbytes; 1000 off_t sbytes;
935 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1001 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
936 1002
937 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1003 #if 0 /* according to the manpage, this is correct, but broken behaviour */
938 /* freebsd' sendfile will return 0 on success */ 1004 /* freebsd' sendfile will return 0 on success */
939 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1005 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
940 /* not on e.g. EIO or EPIPE - sounds broken */ 1006 /* not on e.g. EIO or EPIPE - sounds broken */
941 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1007 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
942 res = sbytes; 1008 res = sbytes;
943 #endif 1009 #endif
944 1010
945 /* according to source inspection, this is correct, and useful behaviour */ 1011 /* according to source inspection, this is correct, and useful behaviour */
946 if (sbytes) 1012 if (sbytes)
947 res = sbytes; 1013 res = sbytes;
948 }
949 1014
950# elif defined (__APPLE__) 1015# elif defined (__APPLE__) && 0 /* broken, as everything on os x */
951
952 {
953 off_t sbytes = count; 1016 off_t sbytes = count;
954 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1017 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
955 1018
956 /* according to the manpage, sbytes is always valid */ 1019 /* according to the manpage, sbytes is always valid */
957 if (sbytes) 1020 if (sbytes)
958 res = sbytes; 1021 res = sbytes;
959 }
960 1022
961# elif __hpux 1023# elif __hpux
962 res = sendfile (ofd, ifd, offset, count, 0, 0); 1024 res = sendfile (ofd, ifd, offset, count, 0, 0);
963 1025
964# elif __solaris 1026# elif __solaris
965 {
966 struct sendfilevec vec; 1027 struct sendfilevec vec;
967 size_t sbytes; 1028 size_t sbytes;
968 1029
969 vec.sfv_fd = ifd; 1030 vec.sfv_fd = ifd;
970 vec.sfv_flag = 0; 1031 vec.sfv_flag = 0;
971 vec.sfv_off = offset; 1032 vec.sfv_off = offset;
972 vec.sfv_len = count; 1033 vec.sfv_len = count;
973 1034
974 res = sendfilev (ofd, &vec, 1, &sbytes); 1035 res = sendfilev (ofd, &vec, 1, &sbytes);
975 1036
976 if (res < 0 && sbytes) 1037 if (res < 0 && sbytes)
977 res = sbytes; 1038 res = sbytes;
978 }
979 1039
980# endif 1040# endif
981 1041
982#elif defined (_WIN32) 1042#elif defined (_WIN32)
983
984 /* does not work, just for documentation of what would need to be done */ 1043 /* does not work, just for documentation of what would need to be done */
985 { 1044 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1045 /* libeio guarantees that the file offset does not change, and windows */
1046 /* has no way to get an independent handle to the same file description */
986 HANDLE h = TO_SOCKET (ifd); 1047 HANDLE h = TO_SOCKET (ifd);
987 SetFilePointer (h, offset, 0, FILE_BEGIN); 1048 SetFilePointer (h, offset, 0, FILE_BEGIN);
988 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1049 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
989 }
990 1050
991#else 1051#else
992 res = -1; 1052 res = -1;
993 errno = ENOSYS; 1053 errno = ENOSYS;
994#endif 1054#endif
995 1055
1056 /* we assume sendfile can copy at least 128mb in one go */
1057 if (res <= 128 * 1024 * 1024)
1058 {
1059 if (res > 0)
1060 written += res;
1061
1062 if (written)
1063 return written;
1064
1065 break;
1066 }
1067 else
1068 {
1069 /* if we requested more, then probably the kernel was lazy */
1070 written += res;
1071 offset += res;
1072 count -= res;
1073
1074 if (!count)
1075 return written;
1076 }
1077 }
1078
996 if (res < 0 1079 if (res < 0
997 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1080 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
998 /* BSDs */ 1081 /* BSDs */
999#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1082#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1000 || errno == ENOTSUP 1083 || errno == ENOTSUP
1001#endif 1084#endif
… …
1038 } 1121 }
1039 1122
1040 return res; 1123 return res;
1041} 1124}
1042 1125
1126#ifdef PAGESIZE
1127# define eio_pagesize() PAGESIZE
1128#else
1129static intptr_t
1130eio_pagesize (void)
1131{
1132 static intptr_t page;
1133
1134 if (!page)
1135 page = sysconf (_SC_PAGESIZE);
1136
1137 return page;
1138}
1139#endif
1140
1141static void
1142eio_page_align (void **addr, size_t *length)
1143{
1144 intptr_t mask = eio_pagesize () - 1;
1145
1146 /* round down addr */
1147 intptr_t adj = mask & (intptr_t)*addr;
1148
1149 *addr = (void *)((intptr_t)*addr - adj);
1150 *length += adj;
1151
1152 /* round up length */
1153 *length = (*length + mask) & ~mask;
1154}
1155
1156#if !_POSIX_MEMLOCK
1157# define eio__mlockall(a) ((errno = ENOSYS), -1)
1158#else
1159
1160static int
1161eio__mlockall (int flags)
1162{
1163 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1164 extern int mallopt (int, int);
1165 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1166 #endif
1167
1168 if (EIO_MCL_CURRENT != MCL_CURRENT
1169 || EIO_MCL_FUTURE != MCL_FUTURE)
1170 {
1171 flags = 0
1172 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1173 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1174 }
1175
1176 return mlockall (flags);
1177}
1178#endif
1179
1180#if !_POSIX_MEMLOCK_RANGE
1181# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1182#else
1183
1184static int
1185eio__mlock (void *addr, size_t length)
1186{
1187 eio_page_align (&addr, &length);
1188
1189 return mlock (addr, length);
1190}
1191
1192#endif
1193
1194#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1195# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1196#else
1197
1198static int
1199eio__msync (void *mem, size_t len, int flags)
1200{
1201 eio_page_align (&mem, &len);
1202
1203 if (EIO_MS_ASYNC != MS_SYNC
1204 || EIO_MS_INVALIDATE != MS_INVALIDATE
1205 || EIO_MS_SYNC != MS_SYNC)
1206 {
1207 flags = 0
1208 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1209 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1210 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1211 }
1212
1213 return msync (mem, len, flags);
1214}
1215
1216#endif
1217
1218static int
1219eio__mtouch (eio_req *req)
1220{
1221 void *mem = req->ptr2;
1222 size_t len = req->size;
1223 int flags = req->int1;
1224
1225 eio_page_align (&mem, &len);
1226
1227 {
1228 intptr_t addr = (intptr_t)mem;
1229 intptr_t end = addr + len;
1230 intptr_t page = eio_pagesize ();
1231
1232 if (addr < end)
1233 if (flags & EIO_MT_MODIFY) /* modify */
1234 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1235 else
1236 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1237 }
1238
1239 return 0;
1240}
1241
1242/*****************************************************************************/
1243/* requests implemented outside eio_execute, because they are so large */
1244
1245static void
1246eio__realpath (eio_req *req, etp_worker *self)
1247{
1248 char *rel = req->ptr1;
1249 char *res;
1250 char *tmp1, *tmp2;
1251#if SYMLOOP_MAX > 32
1252 int symlinks = SYMLOOP_MAX;
1253#else
1254 int symlinks = 32;
1255#endif
1256
1257 req->result = -1;
1258
1259 errno = EINVAL;
1260 if (!rel)
1261 return;
1262
1263 errno = ENOENT;
1264 if (!*rel)
1265 return;
1266
1267 if (!req->ptr2)
1268 {
1269 X_LOCK (wrklock);
1270 req->flags |= EIO_FLAG_PTR2_FREE;
1271 X_UNLOCK (wrklock);
1272 req->ptr2 = malloc (PATH_MAX * 3);
1273
1274 errno = ENOMEM;
1275 if (!req->ptr2)
1276 return;
1277 }
1278
1279 res = req->ptr2;
1280 tmp1 = res + PATH_MAX;
1281 tmp2 = tmp1 + PATH_MAX;
1282
1283#if 0 /* disabled, the musl way to do things is just too racy */
1284#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1285 /* on linux we may be able to ask the kernel */
1286 {
1287 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1288
1289 if (fd >= 0)
1290 {
1291 sprintf (tmp1, "/proc/self/fd/%d", fd);
1292 req->result = readlink (tmp1, res, PATH_MAX);
1293 close (fd);
1294
1295 /* here we should probably stat the open file and the disk file, to make sure they still match */
1296
1297 if (req->result > 0)
1298 goto done;
1299 }
1300 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1301 return;
1302 }
1303#endif
1304#endif
1305
1306 if (*rel != '/')
1307 {
1308 if (!getcwd (res, PATH_MAX))
1309 return;
1310
1311 if (res [1]) /* only use if not / */
1312 res += strlen (res);
1313 }
1314
1315 while (*rel)
1316 {
1317 ssize_t len, linklen;
1318 char *beg = rel;
1319
1320 while (*rel && *rel != '/')
1321 ++rel;
1322
1323 len = rel - beg;
1324
1325 if (!len) /* skip slashes */
1326 {
1327 ++rel;
1328 continue;
1329 }
1330
1331 if (beg [0] == '.')
1332 {
1333 if (len == 1)
1334 continue; /* . - nop */
1335
1336 if (beg [1] == '.' && len == 2)
1337 {
1338 /* .. - back up one component, if possible */
1339
1340 while (res != req->ptr2)
1341 if (*--res == '/')
1342 break;
1343
1344 continue;
1345 }
1346 }
1347
1348 errno = ENAMETOOLONG;
1349 if (res + 1 + len + 1 >= tmp1)
1350 return;
1351
1352 /* copy one component */
1353 *res = '/';
1354 memcpy (res + 1, beg, len);
1355
1356 /* zero-terminate, for readlink */
1357 res [len + 1] = 0;
1358
1359 /* now check if it's a symlink */
1360 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1361
1362 if (linklen < 0)
1363 {
1364 if (errno != EINVAL)
1365 return;
1366
1367 /* it's a normal directory. hopefully */
1368 res += len + 1;
1369 }
1370 else
1371 {
1372 /* yay, it was a symlink - build new path in tmp2 */
1373 int rellen = strlen (rel);
1374
1375 errno = ENAMETOOLONG;
1376 if (linklen + 1 + rellen >= PATH_MAX)
1377 return;
1378
1379 errno = ELOOP;
1380 if (!--symlinks)
1381 return;
1382
1383 if (*tmp1 == '/')
1384 res = req->ptr2; /* symlink resolves to an absolute path */
1385
1386 /* we need to be careful, as rel might point into tmp2 already */
1387 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1388 tmp2 [linklen] = '/';
1389 memcpy (tmp2, tmp1, linklen);
1390
1391 rel = tmp2;
1392 }
1393 }
1394
1395 /* special case for the lone root path */
1396 if (res == req->ptr2)
1397 *res++ = '/';
1398
1399 req->result = res - (char *)req->ptr2;
1400
1401done:
1402 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1403}
1404
1043static signed char 1405static signed char
1044eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1406eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1045{ 1407{
1046 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1408 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1047 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1409 : a->inode < b->inode ? -1
1410 : a->inode > b->inode ? 1
1411 : 0;
1048} 1412}
1049 1413
1050#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1414#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1051 1415
1052#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1416#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
… …
1058 unsigned char bits [9 + sizeof (ino_t) * 8]; 1422 unsigned char bits [9 + sizeof (ino_t) * 8];
1059 unsigned char *bit = bits; 1423 unsigned char *bit = bits;
1060 1424
1061 assert (CHAR_BIT == 8); 1425 assert (CHAR_BIT == 8);
1062 assert (sizeof (eio_dirent) * 8 < 256); 1426 assert (sizeof (eio_dirent) * 8 < 256);
1063 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1427 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1064 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1428 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1065 1429
1066 if (size <= EIO_SORT_FAST) 1430 if (size <= EIO_SORT_FAST)
1067 return; 1431 return;
1068 1432
1069 /* first prepare an array of bits to test in our radix sort */ 1433 /* first prepare an array of bits to test in our radix sort */
… …
1224 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1588 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1225 1589
1226 X_LOCK (wrklock); 1590 X_LOCK (wrklock);
1227 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1591 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1228 self->dirp = dirp = opendir (req->ptr1); 1592 self->dirp = dirp = opendir (req->ptr1);
1593
1594 if (req->flags & EIO_FLAG_PTR1_FREE)
1595 free (req->ptr1);
1596
1229 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1597 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1230 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1598 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1231 req->ptr2 = names = malloc (namesalloc); 1599 req->ptr2 = names = malloc (namesalloc);
1232 X_UNLOCK (wrklock); 1600 X_UNLOCK (wrklock);
1233 1601
… …
1245 /* sort etc. */ 1613 /* sort etc. */
1246 req->int1 = flags; 1614 req->int1 = flags;
1247 req->result = dentoffs; 1615 req->result = dentoffs;
1248 1616
1249 if (flags & EIO_READDIR_STAT_ORDER) 1617 if (flags & EIO_READDIR_STAT_ORDER)
1250 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1618 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1251 else if (flags & EIO_READDIR_DIRS_FIRST) 1619 else if (flags & EIO_READDIR_DIRS_FIRST)
1252 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1620 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1253 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1621 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1254 else 1622 else
1255 { 1623 {
… …
1257 eio_dirent *oth = dents + dentoffs; 1625 eio_dirent *oth = dents + dentoffs;
1258 eio_dirent *dir = dents; 1626 eio_dirent *dir = dents;
1259 1627
1260 /* now partition dirs to the front, and non-dirs to the back */ 1628 /* now partition dirs to the front, and non-dirs to the back */
1261 /* by walking from both sides and swapping if necessary */ 1629 /* by walking from both sides and swapping if necessary */
1262 /* also clear score, so it doesn't influence sorting */
1263 while (oth > dir) 1630 while (oth > dir)
1264 { 1631 {
1265 if (dir->type == EIO_DT_DIR) 1632 if (dir->type == EIO_DT_DIR)
1266 ++dir; 1633 ++dir;
1267 else if ((--oth)->type == EIO_DT_DIR) 1634 else if ((--oth)->type == EIO_DT_DIR)
… …
1270 1637
1271 ++dir; 1638 ++dir;
1272 } 1639 }
1273 } 1640 }
1274 1641
1275 /* now sort the dirs only */ 1642 /* now sort the dirs only (dirs all have the same score) */
1276 eio_dent_sort (dents, dir - dents, 0, inode_bits); 1643 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1277 } 1644 }
1278 1645
1279 break; 1646 break;
1280 } 1647 }
… …
1285 /* skip . and .. entries */ 1652 /* skip . and .. entries */
1286 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1653 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1287 { 1654 {
1288 int len = D_NAMLEN (entp) + 1; 1655 int len = D_NAMLEN (entp) + 1;
1289 1656
1290 while (expect_false (namesoffs + len > namesalloc)) 1657 while (ecb_expect_false (namesoffs + len > namesalloc))
1291 { 1658 {
1292 namesalloc *= 2; 1659 namesalloc *= 2;
1293 X_LOCK (wrklock); 1660 X_LOCK (wrklock);
1294 req->ptr2 = names = realloc (names, namesalloc); 1661 req->ptr2 = names = realloc (names, namesalloc);
1295 X_UNLOCK (wrklock); 1662 X_UNLOCK (wrklock);
… …
1302 1669
1303 if (dents) 1670 if (dents)
1304 { 1671 {
1305 struct eio_dirent *ent; 1672 struct eio_dirent *ent;
1306 1673
1307 if (expect_false (dentoffs == dentalloc)) 1674 if (ecb_expect_false (dentoffs == dentalloc))
1308 { 1675 {
1309 dentalloc *= 2; 1676 dentalloc *= 2;
1310 X_LOCK (wrklock); 1677 X_LOCK (wrklock);
1311 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1678 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1312 X_UNLOCK (wrklock); 1679 X_UNLOCK (wrklock);
… …
1400 break; 1767 break;
1401 } 1768 }
1402 } 1769 }
1403} 1770}
1404 1771
1405#ifdef PAGESIZE
1406# define eio_pagesize() PAGESIZE
1407#else
1408static intptr_t
1409eio_pagesize (void)
1410{
1411 static intptr_t page;
1412
1413 if (!page)
1414 page = sysconf (_SC_PAGESIZE);
1415
1416 return page;
1417}
1418#endif
1419
1420static void
1421eio_page_align (void **addr, size_t *length)
1422{
1423 intptr_t mask = eio_pagesize () - 1;
1424
1425 /* round down addr */
1426 intptr_t adj = mask & (intptr_t)*addr;
1427
1428 *addr = (void *)((intptr_t)*addr - adj);
1429 *length += adj;
1430
1431 /* round up length */
1432 *length = (*length + mask) & ~mask;
1433}
1434
1435#if !_POSIX_MEMLOCK
1436# define eio__mlockall(a) ((errno = ENOSYS), -1)
1437#else
1438
1439static int
1440eio__mlockall (int flags)
1441{
1442 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1443 extern int mallopt (int, int);
1444 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1445 #endif
1446
1447 if (EIO_MCL_CURRENT != MCL_CURRENT
1448 || EIO_MCL_FUTURE != MCL_FUTURE)
1449 {
1450 flags = 0
1451 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1452 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1453 }
1454
1455 return mlockall (flags);
1456}
1457#endif
1458
1459#if !_POSIX_MEMLOCK_RANGE
1460# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1461#else
1462
1463static int
1464eio__mlock (void *addr, size_t length)
1465{
1466 eio_page_align (&addr, &length);
1467
1468 return mlock (addr, length);
1469}
1470
1471#endif
1472
1473#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1474# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1475#else
1476
1477int
1478eio__msync (void *mem, size_t len, int flags)
1479{
1480 eio_page_align (&mem, &len);
1481
1482 if (EIO_MS_ASYNC != MS_SYNC
1483 || EIO_MS_INVALIDATE != MS_INVALIDATE
1484 || EIO_MS_SYNC != MS_SYNC)
1485 {
1486 flags = 0
1487 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1488 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1489 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1490 }
1491
1492 return msync (mem, len, flags);
1493}
1494
1495#endif
1496
1497int
1498eio__mtouch (void *mem, size_t len, int flags)
1499{
1500 eio_page_align (&mem, &len);
1501
1502 {
1503 intptr_t addr = (intptr_t)mem;
1504 intptr_t end = addr + len;
1505 intptr_t page = eio_pagesize ();
1506
1507 if (addr < end)
1508 if (flags & EIO_MT_MODIFY) /* modify */
1509 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1510 else
1511 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1512 }
1513
1514 return 0;
1515}
1516
1517/*****************************************************************************/ 1772/*****************************************************************************/
1518 1773
1519#define ALLOC(len) \ 1774#define ALLOC(len) \
1520 if (!req->ptr2) \ 1775 if (!req->ptr2) \
1521 { \ 1776 { \
… …
1551 if (req) 1806 if (req)
1552 break; 1807 break;
1553 1808
1554 ++idle; 1809 ++idle;
1555 1810
1556 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1811 ts.tv_sec = time (0) + idle_timeout;
1557 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1812 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1558 { 1813 {
1559 if (idle > max_idle) 1814 if (idle > max_idle)
1560 { 1815 {
1561 --idle; 1816 --idle;
… …
1578 X_UNLOCK (reqlock); 1833 X_UNLOCK (reqlock);
1579 1834
1580 if (req->type < 0) 1835 if (req->type < 0)
1581 goto quit; 1836 goto quit;
1582 1837
1583 if (!EIO_CANCELLED (req))
1584 ETP_EXECUTE (self, req); 1838 ETP_EXECUTE (self, req);
1585 1839
1586 X_LOCK (reslock); 1840 X_LOCK (reslock);
1587 1841
1588 ++npending; 1842 ++npending;
1589 1843
… …
1604 return 0; 1858 return 0;
1605} 1859}
1606 1860
1607/*****************************************************************************/ 1861/*****************************************************************************/
1608 1862
1863int ecb_cold
1609int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1864eio_init (void (*want_poll)(void), void (*done_poll)(void))
1610{ 1865{
1611 return etp_init (want_poll, done_poll); 1866 return etp_init (want_poll, done_poll);
1612} 1867}
1613 1868
1869ecb_inline void
1614static void eio_api_destroy (eio_req *req) 1870eio_api_destroy (eio_req *req)
1615{ 1871{
1616 free (req); 1872 free (req);
1617} 1873}
1618 1874
1619#define REQ(rtype) \ 1875#define REQ(rtype) \
… …
1638 { \ 1894 { \
1639 eio_api_destroy (req); \ 1895 eio_api_destroy (req); \
1640 return 0; \ 1896 return 0; \
1641 } 1897 }
1642 1898
1899static void
1643static void eio_execute (etp_worker *self, eio_req *req) 1900eio_execute (etp_worker *self, eio_req *req)
1644{ 1901{
1902 if (ecb_expect_false (EIO_CANCELLED (req)))
1903 {
1904 req->result = -1;
1905 req->errorno = ECANCELED;
1906 return;
1907 }
1908
1645 switch (req->type) 1909 switch (req->type)
1646 { 1910 {
1647 case EIO_READ: ALLOC (req->size); 1911 case EIO_READ: ALLOC (req->size);
1648 req->result = req->offs >= 0 1912 req->result = req->offs >= 0
1649 ? pread (req->int1, req->ptr2, req->size, req->offs) 1913 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1683 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1947 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1684 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1948 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1685 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1949 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1686 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; 1950 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1687 1951
1952 case EIO_REALPATH: eio__realpath (req, self); break;
1953
1688 case EIO_READLINK: ALLOC (PATH_MAX); 1954 case EIO_READLINK: ALLOC (PATH_MAX);
1689 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; 1955 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1690 1956
1691 case EIO_SYNC: req->result = 0; sync (); break; 1957 case EIO_SYNC: req->result = 0; sync (); break;
1692 case EIO_FSYNC: req->result = fsync (req->int1); break; 1958 case EIO_FSYNC: req->result = fsync (req->int1); break;
1693 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1959 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1694 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 1960 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1695 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1961 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1696 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 1962 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1697 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 1963 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1698 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 1964 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1699 1965
1700 case EIO_READDIR: eio__scandir (req, self); break; 1966 case EIO_READDIR: eio__scandir (req, self); break;
… …
1744 case EIO_NOP: 2010 case EIO_NOP:
1745 req->result = 0; 2011 req->result = 0;
1746 break; 2012 break;
1747 2013
1748 case EIO_CUSTOM: 2014 case EIO_CUSTOM:
1749 ((void (*)(eio_req *))req->feed) (req); 2015 req->feed (req);
1750 break; 2016 break;
1751 2017
1752 default: 2018 default:
1753 errno = ENOSYS; 2019 errno = ENOSYS;
1754 req->result = -1; 2020 req->result = -1;
… …
1907} 2173}
1908 2174
1909eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2175eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1910{ 2176{
1911 return eio__1path (EIO_READLINK, path, pri, cb, data); 2177 return eio__1path (EIO_READLINK, path, pri, cb, data);
2178}
2179
2180eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2181{
2182 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1912} 2183}
1913 2184
1914eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2185eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1915{ 2186{
1916 return eio__1path (EIO_STAT, path, pri, cb, data); 2187 return eio__1path (EIO_STAT, path, pri, cb, data);
… …
1975eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2246eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1976{ 2247{
1977 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2248 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1978} 2249}
1979 2250
1980eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2251eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1981{ 2252{
1982 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2253 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1983} 2254}
1984 2255
1985#endif 2256#endif
1986 2257
1987eio_req *eio_grp (eio_cb cb, void *data) 2258eio_req *eio_grp (eio_cb cb, void *data)
… …
1996#undef SEND 2267#undef SEND
1997 2268
1998/*****************************************************************************/ 2269/*****************************************************************************/
1999/* grp functions */ 2270/* grp functions */
2000 2271
2272void
2001void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2273eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2002{ 2274{
2003 grp->int2 = limit; 2275 grp->int2 = limit;
2004 grp->feed = feed; 2276 grp->feed = feed;
2005 2277
2006 grp_try_feed (grp); 2278 grp_try_feed (grp);
2007} 2279}
2008 2280
2281void
2009void eio_grp_limit (eio_req *grp, int limit) 2282eio_grp_limit (eio_req *grp, int limit)
2010{ 2283{
2011 grp->int2 = limit; 2284 grp->int2 = limit;
2012 2285
2013 grp_try_feed (grp); 2286 grp_try_feed (grp);
2014} 2287}
2015 2288
2289void
2016void eio_grp_add (eio_req *grp, eio_req *req) 2290eio_grp_add (eio_req *grp, eio_req *req)
2017{ 2291{
2018 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2292 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2019 2293
2020 grp->flags |= EIO_FLAG_GROUPADD; 2294 grp->flags |= EIO_FLAG_GROUPADD;
2021 2295
… …
2032} 2306}
2033 2307
2034/*****************************************************************************/ 2308/*****************************************************************************/
2035/* misc garbage */ 2309/* misc garbage */
2036 2310
2311ssize_t
2037ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2312eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2038{ 2313{
2039 etp_worker wrk; 2314 etp_worker wrk;
2040 ssize_t ret; 2315 ssize_t ret;
2041 2316
2042 wrk.dbuf = 0; 2317 wrk.dbuf = 0;

Diff Legend

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