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 1.79 by root, Tue Jul 5 16:57:41 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 form stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H
67# include <stdint.h>
68#endif
69
59#ifndef EIO_FINISH 70#ifndef EIO_FINISH
60# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 71# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
61#endif 72#endif
62 73
63#ifndef EIO_DESTROY 74#ifndef EIO_DESTROY
… …
71#ifdef _WIN32 82#ifdef _WIN32
72 83
73 /*doh*/ 84 /*doh*/
74#else 85#else
75 86
76# include "config.h"
77# include <sys/time.h> 87# include <sys/time.h>
78# include <sys/select.h> 88# include <sys/select.h>
79# include <unistd.h> 89# include <unistd.h>
80# include <utime.h> 90# include <utime.h>
81# include <signal.h> 91# include <signal.h>
… …
132#endif 142#endif
133#ifndef D_NAMLEN 143#ifndef D_NAMLEN
134# define D_NAMLEN(de) strlen ((de)->d_name) 144# define D_NAMLEN(de) strlen ((de)->d_name)
135#endif 145#endif
136 146
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 */ 147/* used for struct dirent, AIX doesn't provide it */
141#ifndef NAME_MAX 148#ifndef NAME_MAX
142# define NAME_MAX 4096 149# define NAME_MAX 4096
143#endif 150#endif
144 151
… …
158 errno = ENOMEM; \ 165 errno = ENOMEM; \
159 if (!eio_buf) \ 166 if (!eio_buf) \
160 return -1; 167 return -1;
161 168
162#define EIO_TICKS ((1000000 + 1023) >> 10) 169#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/*****************************************************************************/
176 170
177#define ETP_PRI_MIN EIO_PRI_MIN 171#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 172#define ETP_PRI_MAX EIO_PRI_MAX
179 173
180struct etp_worker; 174struct etp_worker;
… …
206/*****************************************************************************/ 200/*****************************************************************************/
207 201
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 202#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209 203
210/* calculate time difference in ~1/EIO_TICKS of a second */ 204/* calculate time difference in ~1/EIO_TICKS of a second */
205ecb_inline int
211static int tvdiff (struct timeval *tv1, struct timeval *tv2) 206tvdiff (struct timeval *tv1, struct timeval *tv2)
212{ 207{
213 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 208 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
214 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 209 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
215} 210}
216 211
… …
223static unsigned int max_poll_reqs; /* reslock */ 218static unsigned int max_poll_reqs; /* reslock */
224 219
225static volatile unsigned int nreqs; /* reqlock */ 220static volatile unsigned int nreqs; /* reqlock */
226static volatile unsigned int nready; /* reqlock */ 221static volatile unsigned int nready; /* reqlock */
227static volatile unsigned int npending; /* reqlock */ 222static volatile unsigned int npending; /* reqlock */
228static volatile unsigned int max_idle = 4; 223static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
224static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
229 225
230static xmutex_t wrklock = X_MUTEX_INIT; 226static xmutex_t wrklock;
231static xmutex_t reslock = X_MUTEX_INIT; 227static xmutex_t reslock;
232static xmutex_t reqlock = X_MUTEX_INIT; 228static xmutex_t reqlock;
233static xcond_t reqwait = X_COND_INIT; 229static xcond_t reqwait;
234 230
235#if !HAVE_PREADWRITE 231#if !HAVE_PREADWRITE
236/* 232/*
237 * make our pread/pwrite emulation safe against themselves, but not against 233 * make our pread/pwrite emulation safe against themselves, but not against
238 * normal read/write by using a mutex. slows down execution a lot, 234 * normal read/write by using a mutex. slows down execution a lot,
… …
259#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 255#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
260#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 256#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
261 257
262/* worker threads management */ 258/* worker threads management */
263 259
260static void ecb_cold
264static void etp_worker_clear (etp_worker *wrk) 261etp_worker_clear (etp_worker *wrk)
265{ 262{
266 ETP_WORKER_CLEAR (wrk); 263 ETP_WORKER_CLEAR (wrk);
267} 264}
268 265
266static void ecb_cold
269static void etp_worker_free (etp_worker *wrk) 267etp_worker_free (etp_worker *wrk)
270{ 268{
271 wrk->next->prev = wrk->prev; 269 wrk->next->prev = wrk->prev;
272 wrk->prev->next = wrk->next; 270 wrk->prev->next = wrk->next;
273 271
274 free (wrk); 272 free (wrk);
275} 273}
276 274
277static unsigned int etp_nreqs (void) 275static unsigned int
276etp_nreqs (void)
278{ 277{
279 int retval; 278 int retval;
280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 279 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281 retval = nreqs; 280 retval = nreqs;
282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 281 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283 return retval; 282 return retval;
284} 283}
285 284
286static unsigned int etp_nready (void) 285static unsigned int
286etp_nready (void)
287{ 287{
288 unsigned int retval; 288 unsigned int retval;
289 289
290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
291 retval = nready; 291 retval = nready;
292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
293 293
294 return retval; 294 return retval;
295} 295}
296 296
297static unsigned int etp_npending (void) 297static unsigned int
298etp_npending (void)
298{ 299{
299 unsigned int retval; 300 unsigned int retval;
300 301
301 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 302 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
302 retval = npending; 303 retval = npending;
303 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 304 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
304 305
305 return retval; 306 return retval;
306} 307}
307 308
308static unsigned int etp_nthreads (void) 309static unsigned int
310etp_nthreads (void)
309{ 311{
310 unsigned int retval; 312 unsigned int retval;
311 313
312 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 314 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
313 retval = started; 315 retval = started;
… …
327} etp_reqq; 329} etp_reqq;
328 330
329static etp_reqq req_queue; 331static etp_reqq req_queue;
330static etp_reqq res_queue; 332static etp_reqq res_queue;
331 333
334static int ecb_noinline
332static int reqq_push (etp_reqq *q, ETP_REQ *req) 335reqq_push (etp_reqq *q, ETP_REQ *req)
333{ 336{
334 int pri = req->pri; 337 int pri = req->pri;
335 req->next = 0; 338 req->next = 0;
336 339
337 if (q->qe[pri]) 340 if (q->qe[pri])
… …
343 q->qe[pri] = q->qs[pri] = req; 346 q->qe[pri] = q->qs[pri] = req;
344 347
345 return q->size++; 348 return q->size++;
346} 349}
347 350
351static ETP_REQ * ecb_noinline
348static ETP_REQ *reqq_shift (etp_reqq *q) 352reqq_shift (etp_reqq *q)
349{ 353{
350 int pri; 354 int pri;
351 355
352 if (!q->size) 356 if (!q->size)
353 return 0; 357 return 0;
… …
368 } 372 }
369 373
370 abort (); 374 abort ();
371} 375}
372 376
377static void ecb_cold
378etp_thread_init (void)
379{
380 X_MUTEX_CREATE (wrklock);
381 X_MUTEX_CREATE (reslock);
382 X_MUTEX_CREATE (reqlock);
383 X_COND_CREATE (reqwait);
384}
385
386static void ecb_cold
373static void etp_atfork_prepare (void) 387etp_atfork_prepare (void)
374{ 388{
375 X_LOCK (wrklock); 389 X_LOCK (wrklock);
376 X_LOCK (reqlock); 390 X_LOCK (reqlock);
377 X_LOCK (reslock); 391 X_LOCK (reslock);
378#if !HAVE_PREADWRITE 392#if !HAVE_PREADWRITE
379 X_LOCK (preadwritelock); 393 X_LOCK (preadwritelock);
380#endif 394#endif
381} 395}
382 396
397static void ecb_cold
383static void etp_atfork_parent (void) 398etp_atfork_parent (void)
384{ 399{
385#if !HAVE_PREADWRITE 400#if !HAVE_PREADWRITE
386 X_UNLOCK (preadwritelock); 401 X_UNLOCK (preadwritelock);
387#endif 402#endif
388 X_UNLOCK (reslock); 403 X_UNLOCK (reslock);
389 X_UNLOCK (reqlock); 404 X_UNLOCK (reqlock);
390 X_UNLOCK (wrklock); 405 X_UNLOCK (wrklock);
391} 406}
392 407
408static void ecb_cold
393static void etp_atfork_child (void) 409etp_atfork_child (void)
394{ 410{
395 ETP_REQ *prv; 411 ETP_REQ *prv;
396 412
397 while ((prv = reqq_shift (&req_queue))) 413 while ((prv = reqq_shift (&req_queue)))
398 ETP_DESTROY (prv); 414 ETP_DESTROY (prv);
… …
415 idle = 0; 431 idle = 0;
416 nreqs = 0; 432 nreqs = 0;
417 nready = 0; 433 nready = 0;
418 npending = 0; 434 npending = 0;
419 435
420 etp_atfork_parent (); 436 etp_thread_init ();
421} 437}
422 438
423static void 439static void ecb_cold
424etp_once_init (void) 440etp_once_init (void)
425{ 441{
442 etp_thread_init ();
426 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 443 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
427} 444}
428 445
429static int 446static int ecb_cold
430etp_init (void (*want_poll)(void), void (*done_poll)(void)) 447etp_init (void (*want_poll)(void), void (*done_poll)(void))
431{ 448{
432 static pthread_once_t doinit = PTHREAD_ONCE_INIT; 449 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
433 450
434 pthread_once (&doinit, etp_once_init); 451 pthread_once (&doinit, etp_once_init);
… …
439 return 0; 456 return 0;
440} 457}
441 458
442X_THREAD_PROC (etp_proc); 459X_THREAD_PROC (etp_proc);
443 460
461static void ecb_cold
444static void etp_start_thread (void) 462etp_start_thread (void)
445{ 463{
446 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 464 etp_worker *wrk = calloc (1, sizeof (etp_worker));
447 465
448 /*TODO*/ 466 /*TODO*/
449 assert (("unable to allocate worker thread data", wrk)); 467 assert (("unable to allocate worker thread data", wrk));
… …
462 free (wrk); 480 free (wrk);
463 481
464 X_UNLOCK (wrklock); 482 X_UNLOCK (wrklock);
465} 483}
466 484
485static void
467static void etp_maybe_start_thread (void) 486etp_maybe_start_thread (void)
468{ 487{
469 if (expect_true (etp_nthreads () >= wanted)) 488 if (ecb_expect_true (etp_nthreads () >= wanted))
470 return; 489 return;
471 490
472 /* todo: maybe use idle here, but might be less exact */ 491 /* todo: maybe use idle here, but might be less exact */
473 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 492 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
474 return; 493 return;
475 494
476 etp_start_thread (); 495 etp_start_thread ();
477} 496}
478 497
498static void ecb_cold
479static void etp_end_thread (void) 499etp_end_thread (void)
480{ 500{
481 eio_req *req = calloc (1, sizeof (eio_req)); 501 eio_req *req = calloc (1, sizeof (eio_req));
482 502
483 req->type = -1; 503 req->type = -1;
484 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 504 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
… …
491 X_LOCK (wrklock); 511 X_LOCK (wrklock);
492 --started; 512 --started;
493 X_UNLOCK (wrklock); 513 X_UNLOCK (wrklock);
494} 514}
495 515
496static int etp_poll (void) 516static int
517etp_poll (void)
497{ 518{
498 unsigned int maxreqs; 519 unsigned int maxreqs;
499 unsigned int maxtime; 520 unsigned int maxtime;
500 struct timeval tv_start, tv_now; 521 struct timeval tv_start, tv_now;
501 522
… …
531 552
532 X_LOCK (reqlock); 553 X_LOCK (reqlock);
533 --nreqs; 554 --nreqs;
534 X_UNLOCK (reqlock); 555 X_UNLOCK (reqlock);
535 556
536 if (expect_false (req->type == EIO_GROUP && req->size)) 557 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
537 { 558 {
538 req->int1 = 1; /* mark request as delayed */ 559 req->int1 = 1; /* mark request as delayed */
539 continue; 560 continue;
540 } 561 }
541 else 562 else
542 { 563 {
543 int res = ETP_FINISH (req); 564 int res = ETP_FINISH (req);
544 if (expect_false (res)) 565 if (ecb_expect_false (res))
545 return res; 566 return res;
546 } 567 }
547 568
548 if (expect_false (maxreqs && !--maxreqs)) 569 if (ecb_expect_false (maxreqs && !--maxreqs))
549 break; 570 break;
550 571
551 if (maxtime) 572 if (maxtime)
552 { 573 {
553 gettimeofday (&tv_now, 0); 574 gettimeofday (&tv_now, 0);
… …
559 580
560 errno = EAGAIN; 581 errno = EAGAIN;
561 return -1; 582 return -1;
562} 583}
563 584
585static void
564static void etp_cancel (ETP_REQ *req) 586etp_cancel (ETP_REQ *req)
565{ 587{
566 X_LOCK (wrklock); 588 X_LOCK (wrklock);
567 req->flags |= EIO_FLAG_CANCELLED; 589 req->flags |= EIO_FLAG_CANCELLED;
568 X_UNLOCK (wrklock); 590 X_UNLOCK (wrklock);
569 591
570 eio_grp_cancel (req); 592 eio_grp_cancel (req);
571} 593}
572 594
595static void
573static void etp_submit (ETP_REQ *req) 596etp_submit (ETP_REQ *req)
574{ 597{
575 req->pri -= ETP_PRI_MIN; 598 req->pri -= ETP_PRI_MIN;
576 599
577 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 600 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; 601 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
579 602
580 if (expect_false (req->type == EIO_GROUP)) 603 if (ecb_expect_false (req->type == EIO_GROUP))
581 { 604 {
582 /* I hope this is worth it :/ */ 605 /* I hope this is worth it :/ */
583 X_LOCK (reqlock); 606 X_LOCK (reqlock);
584 ++nreqs; 607 ++nreqs;
585 X_UNLOCK (reqlock); 608 X_UNLOCK (reqlock);
… …
604 627
605 etp_maybe_start_thread (); 628 etp_maybe_start_thread ();
606 } 629 }
607} 630}
608 631
632static void ecb_cold
609static void etp_set_max_poll_time (double nseconds) 633etp_set_max_poll_time (double nseconds)
610{ 634{
611 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 635 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
612 max_poll_time = nseconds * EIO_TICKS; 636 max_poll_time = nseconds * EIO_TICKS;
613 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 637 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
614} 638}
615 639
640static void ecb_cold
616static void etp_set_max_poll_reqs (unsigned int maxreqs) 641etp_set_max_poll_reqs (unsigned int maxreqs)
617{ 642{
618 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 643 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
619 max_poll_reqs = maxreqs; 644 max_poll_reqs = maxreqs;
620 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 645 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
621} 646}
622 647
648static void ecb_cold
623static void etp_set_max_idle (unsigned int nthreads) 649etp_set_max_idle (unsigned int nthreads)
624{ 650{
625 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 651 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
626 max_idle = nthreads <= 0 ? 1 : nthreads; 652 max_idle = nthreads;
627 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 653 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
628} 654}
629 655
656static void ecb_cold
657etp_set_idle_timeout (unsigned int seconds)
658{
659 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
660 idle_timeout = seconds;
661 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
662}
663
664static void ecb_cold
630static void etp_set_min_parallel (unsigned int nthreads) 665etp_set_min_parallel (unsigned int nthreads)
631{ 666{
632 if (wanted < nthreads) 667 if (wanted < nthreads)
633 wanted = nthreads; 668 wanted = nthreads;
634} 669}
635 670
671static void ecb_cold
636static void etp_set_max_parallel (unsigned int nthreads) 672etp_set_max_parallel (unsigned int nthreads)
637{ 673{
638 if (wanted > nthreads) 674 if (wanted > nthreads)
639 wanted = nthreads; 675 wanted = nthreads;
640 676
641 while (started > wanted) 677 while (started > wanted)
642 etp_end_thread (); 678 etp_end_thread ();
643} 679}
644 680
645/*****************************************************************************/ 681/*****************************************************************************/
646 682
683static void
647static void grp_try_feed (eio_req *grp) 684grp_try_feed (eio_req *grp)
648{ 685{
649 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 686 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
650 { 687 {
651 grp->flags &= ~EIO_FLAG_GROUPADD; 688 grp->flags &= ~EIO_FLAG_GROUPADD;
652 689
… …
659 break; 696 break;
660 } 697 }
661 } 698 }
662} 699}
663 700
701static int
664static int grp_dec (eio_req *grp) 702grp_dec (eio_req *grp)
665{ 703{
666 --grp->size; 704 --grp->size;
667 705
668 /* call feeder, if applicable */ 706 /* call feeder, if applicable */
669 grp_try_feed (grp); 707 grp_try_feed (grp);
… …
673 return eio_finish (grp); 711 return eio_finish (grp);
674 else 712 else
675 return 0; 713 return 0;
676} 714}
677 715
716void
678void eio_destroy (eio_req *req) 717eio_destroy (eio_req *req)
679{ 718{
680 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 719 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
681 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 720 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
682 721
683 EIO_DESTROY (req); 722 EIO_DESTROY (req);
684} 723}
685 724
725static int
686static int eio_finish (eio_req *req) 726eio_finish (eio_req *req)
687{ 727{
688 int res = EIO_FINISH (req); 728 int res = EIO_FINISH (req);
689 729
690 if (req->grp) 730 if (req->grp)
691 { 731 {
… …
699 if (grp->grp_first == req) 739 if (grp->grp_first == req)
700 grp->grp_first = req->grp_next; 740 grp->grp_first = req->grp_next;
701 741
702 res2 = grp_dec (grp); 742 res2 = grp_dec (grp);
703 743
704 if (!res && res2) 744 if (!res)
705 res = res2; 745 res = res2;
706 } 746 }
707 747
708 eio_destroy (req); 748 eio_destroy (req);
709 749
710 return res; 750 return res;
711} 751}
712 752
753void
713void eio_grp_cancel (eio_req *grp) 754eio_grp_cancel (eio_req *grp)
714{ 755{
715 for (grp = grp->grp_first; grp; grp = grp->grp_next) 756 for (grp = grp->grp_first; grp; grp = grp->grp_next)
716 eio_cancel (grp); 757 eio_cancel (grp);
717} 758}
718 759
760void
719void eio_cancel (eio_req *req) 761eio_cancel (eio_req *req)
720{ 762{
721 etp_cancel (req); 763 etp_cancel (req);
722} 764}
723 765
766void
724void eio_submit (eio_req *req) 767eio_submit (eio_req *req)
725{ 768{
726 etp_submit (req); 769 etp_submit (req);
727} 770}
728 771
729unsigned int eio_nreqs (void) 772unsigned int
773eio_nreqs (void)
730{ 774{
731 return etp_nreqs (); 775 return etp_nreqs ();
732} 776}
733 777
734unsigned int eio_nready (void) 778unsigned int
779eio_nready (void)
735{ 780{
736 return etp_nready (); 781 return etp_nready ();
737} 782}
738 783
739unsigned int eio_npending (void) 784unsigned int
785eio_npending (void)
740{ 786{
741 return etp_npending (); 787 return etp_npending ();
742} 788}
743 789
744unsigned int eio_nthreads (void) 790unsigned int ecb_cold
791eio_nthreads (void)
745{ 792{
746 return etp_nthreads (); 793 return etp_nthreads ();
747} 794}
748 795
796void ecb_cold
749void eio_set_max_poll_time (double nseconds) 797eio_set_max_poll_time (double nseconds)
750{ 798{
751 etp_set_max_poll_time (nseconds); 799 etp_set_max_poll_time (nseconds);
752} 800}
753 801
802void ecb_cold
754void eio_set_max_poll_reqs (unsigned int maxreqs) 803eio_set_max_poll_reqs (unsigned int maxreqs)
755{ 804{
756 etp_set_max_poll_reqs (maxreqs); 805 etp_set_max_poll_reqs (maxreqs);
757} 806}
758 807
808void ecb_cold
759void eio_set_max_idle (unsigned int nthreads) 809eio_set_max_idle (unsigned int nthreads)
760{ 810{
761 etp_set_max_idle (nthreads); 811 etp_set_max_idle (nthreads);
762} 812}
763 813
814void ecb_cold
815eio_set_idle_timeout (unsigned int seconds)
816{
817 etp_set_idle_timeout (seconds);
818}
819
820void ecb_cold
764void eio_set_min_parallel (unsigned int nthreads) 821eio_set_min_parallel (unsigned int nthreads)
765{ 822{
766 etp_set_min_parallel (nthreads); 823 etp_set_min_parallel (nthreads);
767} 824}
768 825
826void ecb_cold
769void eio_set_max_parallel (unsigned int nthreads) 827eio_set_max_parallel (unsigned int nthreads)
770{ 828{
771 etp_set_max_parallel (nthreads); 829 etp_set_max_parallel (nthreads);
772} 830}
773 831
774int eio_poll (void) 832int eio_poll (void)
… …
844#ifndef HAVE_FUTIMES 902#ifndef HAVE_FUTIMES
845 903
846# undef futimes 904# undef futimes
847# define futimes(fd,times) eio__futimes (fd, times) 905# define futimes(fd,times) eio__futimes (fd, times)
848 906
907static int
849static int eio__futimes (int fd, const struct timeval tv[2]) 908eio__futimes (int fd, const struct timeval tv[2])
850{ 909{
851 errno = ENOSYS; 910 errno = ENOSYS;
852 return -1; 911 return -1;
853} 912}
854 913
… …
858# undef fdatasync 917# undef fdatasync
859# define fdatasync(fd) fsync (fd) 918# define fdatasync(fd) fsync (fd)
860#endif 919#endif
861 920
862/* sync_file_range always needs emulation */ 921/* sync_file_range always needs emulation */
863int 922static int
864eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 923eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
865{ 924{
866#if HAVE_SYNC_FILE_RANGE 925#if HAVE_SYNC_FILE_RANGE
867 int res; 926 int res;
868 927
… …
914 973
915/* sendfile always needs emulation */ 974/* sendfile always needs emulation */
916static ssize_t 975static ssize_t
917eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 976eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
918{ 977{
978 ssize_t written = 0;
919 ssize_t res; 979 ssize_t res;
920 980
921 if (!count) 981 if (!count)
922 return 0; 982 return 0;
923 983
984 for (;;)
985 {
924#if HAVE_SENDFILE 986#if HAVE_SENDFILE
925# if __linux 987# if __linux
988 off_t soffset = offset;
926 res = sendfile (ofd, ifd, &offset, count); 989 res = sendfile (ofd, ifd, &soffset, count);
927 990
928# elif __FreeBSD__ 991# elif __FreeBSD__
929 /* 992 /*
930 * Of course, the freebsd sendfile is a dire hack with no thoughts 993 * Of course, the freebsd sendfile is a dire hack with no thoughts
931 * wasted on making it similar to other I/O functions. 994 * wasted on making it similar to other I/O functions.
932 */ 995 */
933 {
934 off_t sbytes; 996 off_t sbytes;
935 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 997 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
936 998
937 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 999 #if 0 /* according to the manpage, this is correct, but broken behaviour */
938 /* freebsd' sendfile will return 0 on success */ 1000 /* freebsd' sendfile will return 0 on success */
939 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1001 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
940 /* not on e.g. EIO or EPIPE - sounds broken */ 1002 /* not on e.g. EIO or EPIPE - sounds broken */
941 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1003 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
942 res = sbytes; 1004 res = sbytes;
943 #endif 1005 #endif
944 1006
945 /* according to source inspection, this is correct, and useful behaviour */ 1007 /* according to source inspection, this is correct, and useful behaviour */
946 if (sbytes) 1008 if (sbytes)
947 res = sbytes; 1009 res = sbytes;
948 }
949 1010
950# elif defined (__APPLE__) 1011# elif defined (__APPLE__) && 0 /* broken, as everything on os x */
951
952 {
953 off_t sbytes = count; 1012 off_t sbytes = count;
954 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1013 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
955 1014
956 /* according to the manpage, sbytes is always valid */ 1015 /* according to the manpage, sbytes is always valid */
957 if (sbytes) 1016 if (sbytes)
958 res = sbytes; 1017 res = sbytes;
959 }
960 1018
961# elif __hpux 1019# elif __hpux
962 res = sendfile (ofd, ifd, offset, count, 0, 0); 1020 res = sendfile (ofd, ifd, offset, count, 0, 0);
963 1021
964# elif __solaris 1022# elif __solaris
965 {
966 struct sendfilevec vec; 1023 struct sendfilevec vec;
967 size_t sbytes; 1024 size_t sbytes;
968 1025
969 vec.sfv_fd = ifd; 1026 vec.sfv_fd = ifd;
970 vec.sfv_flag = 0; 1027 vec.sfv_flag = 0;
971 vec.sfv_off = offset; 1028 vec.sfv_off = offset;
972 vec.sfv_len = count; 1029 vec.sfv_len = count;
973 1030
974 res = sendfilev (ofd, &vec, 1, &sbytes); 1031 res = sendfilev (ofd, &vec, 1, &sbytes);
975 1032
976 if (res < 0 && sbytes) 1033 if (res < 0 && sbytes)
977 res = sbytes; 1034 res = sbytes;
978 }
979 1035
980# endif 1036# endif
981 1037
982#elif defined (_WIN32) 1038#elif defined (_WIN32)
983
984 /* does not work, just for documentation of what would need to be done */ 1039 /* does not work, just for documentation of what would need to be done */
985 { 1040 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1041 /* libeio guarantees that the file offset does not change, and windows */
1042 /* has no way to get an independent handle to the same file description */
986 HANDLE h = TO_SOCKET (ifd); 1043 HANDLE h = TO_SOCKET (ifd);
987 SetFilePointer (h, offset, 0, FILE_BEGIN); 1044 SetFilePointer (h, offset, 0, FILE_BEGIN);
988 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1045 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
989 }
990 1046
991#else 1047#else
992 res = -1; 1048 res = -1;
993 errno = ENOSYS; 1049 errno = ENOSYS;
994#endif 1050#endif
995 1051
1052 /* we assume sendfile can copy at least 128mb in one go */
1053 if (res <= 128 * 1024 * 1024)
1054 {
1055 if (res > 0)
1056 written += res;
1057
1058 if (written)
1059 return written;
1060
1061 break;
1062 }
1063 else
1064 {
1065 /* if we requested more, then probably the kernel was lazy */
1066 written += res;
1067 offset += res;
1068 count -= res;
1069
1070 if (!count)
1071 return written;
1072 }
1073 }
1074
996 if (res < 0 1075 if (res < 0
997 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1076 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
998 /* BSDs */ 1077 /* BSDs */
999#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1078#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1000 || errno == ENOTSUP 1079 || errno == ENOTSUP
1001#endif 1080#endif
… …
1038 } 1117 }
1039 1118
1040 return res; 1119 return res;
1041} 1120}
1042 1121
1122#ifdef PAGESIZE
1123# define eio_pagesize() PAGESIZE
1124#else
1125static intptr_t
1126eio_pagesize (void)
1127{
1128 static intptr_t page;
1129
1130 if (!page)
1131 page = sysconf (_SC_PAGESIZE);
1132
1133 return page;
1134}
1135#endif
1136
1137static void
1138eio_page_align (void **addr, size_t *length)
1139{
1140 intptr_t mask = eio_pagesize () - 1;
1141
1142 /* round down addr */
1143 intptr_t adj = mask & (intptr_t)*addr;
1144
1145 *addr = (void *)((intptr_t)*addr - adj);
1146 *length += adj;
1147
1148 /* round up length */
1149 *length = (*length + mask) & ~mask;
1150}
1151
1152#if !_POSIX_MEMLOCK
1153# define eio__mlockall(a) ((errno = ENOSYS), -1)
1154#else
1155
1156static int
1157eio__mlockall (int flags)
1158{
1159 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1160 extern int mallopt (int, int);
1161 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1162 #endif
1163
1164 if (EIO_MCL_CURRENT != MCL_CURRENT
1165 || EIO_MCL_FUTURE != MCL_FUTURE)
1166 {
1167 flags = 0
1168 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1169 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1170 }
1171
1172 return mlockall (flags);
1173}
1174#endif
1175
1176#if !_POSIX_MEMLOCK_RANGE
1177# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1178#else
1179
1180static int
1181eio__mlock (void *addr, size_t length)
1182{
1183 eio_page_align (&addr, &length);
1184
1185 return mlock (addr, length);
1186}
1187
1188#endif
1189
1190#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1191# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1192#else
1193
1194static int
1195eio__msync (void *mem, size_t len, int flags)
1196{
1197 eio_page_align (&mem, &len);
1198
1199 if (EIO_MS_ASYNC != MS_SYNC
1200 || EIO_MS_INVALIDATE != MS_INVALIDATE
1201 || EIO_MS_SYNC != MS_SYNC)
1202 {
1203 flags = 0
1204 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1205 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1206 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1207 }
1208
1209 return msync (mem, len, flags);
1210}
1211
1212#endif
1213
1214static int
1215eio__mtouch (eio_req *req)
1216{
1217 void *mem = req->ptr2;
1218 size_t len = req->size;
1219 int flags = req->int1;
1220
1221 eio_page_align (&mem, &len);
1222
1223 {
1224 intptr_t addr = (intptr_t)mem;
1225 intptr_t end = addr + len;
1226 intptr_t page = eio_pagesize ();
1227
1228 if (addr < end)
1229 if (flags & EIO_MT_MODIFY) /* modify */
1230 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1231 else
1232 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1233 }
1234
1235 return 0;
1236}
1237
1238/*****************************************************************************/
1239/* requests implemented outside eio_execute, because they are so large */
1240
1241static void
1242eio__realpath (eio_req *req, etp_worker *self)
1243{
1244 char *rel = req->ptr1;
1245 char *res;
1246 char *tmp1, *tmp2;
1247#if SYMLOOP_MAX > 32
1248 int symlinks = SYMLOOP_MAX;
1249#else
1250 int symlinks = 32;
1251#endif
1252
1253 req->result = -1;
1254
1255 errno = EINVAL;
1256 if (!rel)
1257 return;
1258
1259 errno = ENOENT;
1260 if (!*rel)
1261 return;
1262
1263 if (!req->ptr2)
1264 {
1265 X_LOCK (wrklock);
1266 req->flags |= EIO_FLAG_PTR2_FREE;
1267 X_UNLOCK (wrklock);
1268 req->ptr2 = malloc (PATH_MAX * 3);
1269
1270 errno = ENOMEM;
1271 if (!req->ptr2)
1272 return;
1273 }
1274
1275 res = req->ptr2;
1276 tmp1 = res + PATH_MAX;
1277 tmp2 = tmp1 + PATH_MAX;
1278
1279#if 0 /* disabled, the musl way to do things is just too racy */
1280#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1281 /* on linux we may be able to ask the kernel */
1282 {
1283 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1284
1285 if (fd >= 0)
1286 {
1287 sprintf (tmp1, "/proc/self/fd/%d", fd);
1288 req->result = readlink (tmp1, res, PATH_MAX);
1289 close (fd);
1290
1291 /* here we should probably stat the open file and the disk file, to make sure they still match */
1292
1293 if (req->result > 0)
1294 goto done;
1295 }
1296 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1297 return;
1298 }
1299#endif
1300#endif
1301
1302 if (*rel != '/')
1303 {
1304 if (!getcwd (res, PATH_MAX))
1305 return;
1306
1307 if (res [1]) /* only use if not / */
1308 res += strlen (res);
1309 }
1310
1311 while (*rel)
1312 {
1313 ssize_t len, linklen;
1314 char *beg = rel;
1315
1316 while (*rel && *rel != '/')
1317 ++rel;
1318
1319 len = rel - beg;
1320
1321 if (!len) /* skip slashes */
1322 {
1323 ++rel;
1324 continue;
1325 }
1326
1327 if (beg [0] == '.')
1328 {
1329 if (len == 1)
1330 continue; /* . - nop */
1331
1332 if (beg [1] == '.' && len == 2)
1333 {
1334 /* .. - back up one component, if possible */
1335
1336 while (res != req->ptr2)
1337 if (*--res == '/')
1338 break;
1339
1340 continue;
1341 }
1342 }
1343
1344 errno = ENAMETOOLONG;
1345 if (res + 1 + len + 1 >= tmp1)
1346 return;
1347
1348 /* copy one component */
1349 *res = '/';
1350 memcpy (res + 1, beg, len);
1351
1352 /* zero-terminate, for readlink */
1353 res [len + 1] = 0;
1354
1355 /* now check if it's a symlink */
1356 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1357
1358 if (linklen < 0)
1359 {
1360 if (errno != EINVAL)
1361 return;
1362
1363 /* it's a normal directory. hopefully */
1364 res += len + 1;
1365 }
1366 else
1367 {
1368 /* yay, it was a symlink - build new path in tmp2 */
1369 int rellen = strlen (rel);
1370
1371 errno = ENAMETOOLONG;
1372 if (linklen + 1 + rellen >= PATH_MAX)
1373 return;
1374
1375 errno = ELOOP;
1376 if (!--symlinks)
1377 return;
1378
1379 if (*tmp1 == '/')
1380 res = req->ptr2; /* symlink resolves to an absolute path */
1381
1382 /* we need to be careful, as rel might point into tmp2 already */
1383 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1384 tmp2 [linklen] = '/';
1385 memcpy (tmp2, tmp1, linklen);
1386
1387 rel = tmp2;
1388 }
1389 }
1390
1391 /* special case for the lone root path */
1392 if (res == req->ptr2)
1393 *res++ = '/';
1394
1395 req->result = res - (char *)req->ptr2;
1396
1397done:
1398 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1399}
1400
1043static signed char 1401static signed char
1044eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1402eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1045{ 1403{
1046 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1404 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; 1405 : a->inode < b->inode ? -1
1406 : a->inode > b->inode ? 1
1407 : 0;
1048} 1408}
1049 1409
1050#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1410#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1051 1411
1052#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1412#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
… …
1058 unsigned char bits [9 + sizeof (ino_t) * 8]; 1418 unsigned char bits [9 + sizeof (ino_t) * 8];
1059 unsigned char *bit = bits; 1419 unsigned char *bit = bits;
1060 1420
1061 assert (CHAR_BIT == 8); 1421 assert (CHAR_BIT == 8);
1062 assert (sizeof (eio_dirent) * 8 < 256); 1422 assert (sizeof (eio_dirent) * 8 < 256);
1063 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1423 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1064 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1424 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1065 1425
1066 if (size <= EIO_SORT_FAST) 1426 if (size <= EIO_SORT_FAST)
1067 return; 1427 return;
1068 1428
1069 /* first prepare an array of bits to test in our radix sort */ 1429 /* first prepare an array of bits to test in our radix sort */
… …
1224 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1584 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1225 1585
1226 X_LOCK (wrklock); 1586 X_LOCK (wrklock);
1227 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1587 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1228 self->dirp = dirp = opendir (req->ptr1); 1588 self->dirp = dirp = opendir (req->ptr1);
1589
1590 if (req->flags & EIO_FLAG_PTR1_FREE)
1591 free (req->ptr1);
1592
1229 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1593 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1230 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1594 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1231 req->ptr2 = names = malloc (namesalloc); 1595 req->ptr2 = names = malloc (namesalloc);
1232 X_UNLOCK (wrklock); 1596 X_UNLOCK (wrklock);
1233 1597
… …
1245 /* sort etc. */ 1609 /* sort etc. */
1246 req->int1 = flags; 1610 req->int1 = flags;
1247 req->result = dentoffs; 1611 req->result = dentoffs;
1248 1612
1249 if (flags & EIO_READDIR_STAT_ORDER) 1613 if (flags & EIO_READDIR_STAT_ORDER)
1250 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1614 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1251 else if (flags & EIO_READDIR_DIRS_FIRST) 1615 else if (flags & EIO_READDIR_DIRS_FIRST)
1252 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1616 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1253 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1617 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1254 else 1618 else
1255 { 1619 {
… …
1257 eio_dirent *oth = dents + dentoffs; 1621 eio_dirent *oth = dents + dentoffs;
1258 eio_dirent *dir = dents; 1622 eio_dirent *dir = dents;
1259 1623
1260 /* now partition dirs to the front, and non-dirs to the back */ 1624 /* now partition dirs to the front, and non-dirs to the back */
1261 /* by walking from both sides and swapping if necessary */ 1625 /* by walking from both sides and swapping if necessary */
1262 /* also clear score, so it doesn't influence sorting */
1263 while (oth > dir) 1626 while (oth > dir)
1264 { 1627 {
1265 if (dir->type == EIO_DT_DIR) 1628 if (dir->type == EIO_DT_DIR)
1266 ++dir; 1629 ++dir;
1267 else if ((--oth)->type == EIO_DT_DIR) 1630 else if ((--oth)->type == EIO_DT_DIR)
… …
1270 1633
1271 ++dir; 1634 ++dir;
1272 } 1635 }
1273 } 1636 }
1274 1637
1275 /* now sort the dirs only */ 1638 /* now sort the dirs only (dirs all have the same score) */
1276 eio_dent_sort (dents, dir - dents, 0, inode_bits); 1639 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1277 } 1640 }
1278 1641
1279 break; 1642 break;
1280 } 1643 }
… …
1285 /* skip . and .. entries */ 1648 /* skip . and .. entries */
1286 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1649 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1287 { 1650 {
1288 int len = D_NAMLEN (entp) + 1; 1651 int len = D_NAMLEN (entp) + 1;
1289 1652
1290 while (expect_false (namesoffs + len > namesalloc)) 1653 while (ecb_expect_false (namesoffs + len > namesalloc))
1291 { 1654 {
1292 namesalloc *= 2; 1655 namesalloc *= 2;
1293 X_LOCK (wrklock); 1656 X_LOCK (wrklock);
1294 req->ptr2 = names = realloc (names, namesalloc); 1657 req->ptr2 = names = realloc (names, namesalloc);
1295 X_UNLOCK (wrklock); 1658 X_UNLOCK (wrklock);
… …
1302 1665
1303 if (dents) 1666 if (dents)
1304 { 1667 {
1305 struct eio_dirent *ent; 1668 struct eio_dirent *ent;
1306 1669
1307 if (expect_false (dentoffs == dentalloc)) 1670 if (ecb_expect_false (dentoffs == dentalloc))
1308 { 1671 {
1309 dentalloc *= 2; 1672 dentalloc *= 2;
1310 X_LOCK (wrklock); 1673 X_LOCK (wrklock);
1311 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1674 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1312 X_UNLOCK (wrklock); 1675 X_UNLOCK (wrklock);
… …
1400 break; 1763 break;
1401 } 1764 }
1402 } 1765 }
1403} 1766}
1404 1767
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/*****************************************************************************/ 1768/*****************************************************************************/
1518 1769
1519#define ALLOC(len) \ 1770#define ALLOC(len) \
1520 if (!req->ptr2) \ 1771 if (!req->ptr2) \
1521 { \ 1772 { \
… …
1551 if (req) 1802 if (req)
1552 break; 1803 break;
1553 1804
1554 ++idle; 1805 ++idle;
1555 1806
1556 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1807 ts.tv_sec = time (0) + idle_timeout;
1557 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1808 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1558 { 1809 {
1559 if (idle > max_idle) 1810 if (idle > max_idle)
1560 { 1811 {
1561 --idle; 1812 --idle;
… …
1604 return 0; 1855 return 0;
1605} 1856}
1606 1857
1607/*****************************************************************************/ 1858/*****************************************************************************/
1608 1859
1860int ecb_cold
1609int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1861eio_init (void (*want_poll)(void), void (*done_poll)(void))
1610{ 1862{
1611 return etp_init (want_poll, done_poll); 1863 return etp_init (want_poll, done_poll);
1612} 1864}
1613 1865
1866ecb_inline void
1614static void eio_api_destroy (eio_req *req) 1867eio_api_destroy (eio_req *req)
1615{ 1868{
1616 free (req); 1869 free (req);
1617} 1870}
1618 1871
1619#define REQ(rtype) \ 1872#define REQ(rtype) \
… …
1638 { \ 1891 { \
1639 eio_api_destroy (req); \ 1892 eio_api_destroy (req); \
1640 return 0; \ 1893 return 0; \
1641 } 1894 }
1642 1895
1896static void
1643static void eio_execute (etp_worker *self, eio_req *req) 1897eio_execute (etp_worker *self, eio_req *req)
1644{ 1898{
1645 switch (req->type) 1899 switch (req->type)
1646 { 1900 {
1647 case EIO_READ: ALLOC (req->size); 1901 case EIO_READ: ALLOC (req->size);
1648 req->result = req->offs >= 0 1902 req->result = req->offs >= 0
… …
1683 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1937 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1684 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1938 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1685 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1939 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; 1940 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1687 1941
1942 case EIO_REALPATH: eio__realpath (req, self); break;
1943
1688 case EIO_READLINK: ALLOC (PATH_MAX); 1944 case EIO_READLINK: ALLOC (PATH_MAX);
1689 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; 1945 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1690 1946
1691 case EIO_SYNC: req->result = 0; sync (); break; 1947 case EIO_SYNC: req->result = 0; sync (); break;
1692 case EIO_FSYNC: req->result = fsync (req->int1); break; 1948 case EIO_FSYNC: req->result = fsync (req->int1); break;
1693 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1949 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1694 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 1950 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; 1951 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1696 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 1952 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1697 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 1953 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; 1954 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1699 1955
1700 case EIO_READDIR: eio__scandir (req, self); break; 1956 case EIO_READDIR: eio__scandir (req, self); break;
… …
1744 case EIO_NOP: 2000 case EIO_NOP:
1745 req->result = 0; 2001 req->result = 0;
1746 break; 2002 break;
1747 2003
1748 case EIO_CUSTOM: 2004 case EIO_CUSTOM:
1749 ((void (*)(eio_req *))req->feed) (req); 2005 req->feed (req);
1750 break; 2006 break;
1751 2007
1752 default: 2008 default:
1753 errno = ENOSYS; 2009 errno = ENOSYS;
1754 req->result = -1; 2010 req->result = -1;
… …
1907} 2163}
1908 2164
1909eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2165eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1910{ 2166{
1911 return eio__1path (EIO_READLINK, path, pri, cb, data); 2167 return eio__1path (EIO_READLINK, path, pri, cb, data);
2168}
2169
2170eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2171{
2172 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1912} 2173}
1913 2174
1914eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2175eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1915{ 2176{
1916 return eio__1path (EIO_STAT, path, pri, cb, data); 2177 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) 2236eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1976{ 2237{
1977 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2238 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1978} 2239}
1979 2240
1980eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2241eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1981{ 2242{
1982 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2243 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1983} 2244}
1984 2245
1985#endif 2246#endif
1986 2247
1987eio_req *eio_grp (eio_cb cb, void *data) 2248eio_req *eio_grp (eio_cb cb, void *data)
… …
1996#undef SEND 2257#undef SEND
1997 2258
1998/*****************************************************************************/ 2259/*****************************************************************************/
1999/* grp functions */ 2260/* grp functions */
2000 2261
2262void
2001void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2263eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2002{ 2264{
2003 grp->int2 = limit; 2265 grp->int2 = limit;
2004 grp->feed = feed; 2266 grp->feed = feed;
2005 2267
2006 grp_try_feed (grp); 2268 grp_try_feed (grp);
2007} 2269}
2008 2270
2271void
2009void eio_grp_limit (eio_req *grp, int limit) 2272eio_grp_limit (eio_req *grp, int limit)
2010{ 2273{
2011 grp->int2 = limit; 2274 grp->int2 = limit;
2012 2275
2013 grp_try_feed (grp); 2276 grp_try_feed (grp);
2014} 2277}
2015 2278
2279void
2016void eio_grp_add (eio_req *grp, eio_req *req) 2280eio_grp_add (eio_req *grp, eio_req *req)
2017{ 2281{
2018 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2282 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2019 2283
2020 grp->flags |= EIO_FLAG_GROUPADD; 2284 grp->flags |= EIO_FLAG_GROUPADD;
2021 2285
… …
2032} 2296}
2033 2297
2034/*****************************************************************************/ 2298/*****************************************************************************/
2035/* misc garbage */ 2299/* misc garbage */
2036 2300
2301ssize_t
2037ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2302eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2038{ 2303{
2039 etp_worker wrk; 2304 etp_worker wrk;
2040 ssize_t ret; 2305 ssize_t ret;
2041 2306
2042 wrk.dbuf = 0; 2307 wrk.dbuf = 0;

Diff Legend

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