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 1.80 by root, Tue Jul 5 18:59:28 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 X_LOCK (wrklock);
567 req->flags |= EIO_FLAG_CANCELLED; 595 req->flags |= EIO_FLAG_CANCELLED;
568 X_UNLOCK (wrklock); 596 X_UNLOCK (wrklock);
569 597
570 eio_grp_cancel (req); 598 eio_grp_cancel (req);
571} 599}
572 600
601static void
573static void etp_submit (ETP_REQ *req) 602etp_submit (ETP_REQ *req)
574{ 603{
575 req->pri -= ETP_PRI_MIN; 604 req->pri -= ETP_PRI_MIN;
576 605
577 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 606 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; 607 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
579 608
580 if (expect_false (req->type == EIO_GROUP)) 609 if (ecb_expect_false (req->type == EIO_GROUP))
581 { 610 {
582 /* I hope this is worth it :/ */ 611 /* I hope this is worth it :/ */
583 X_LOCK (reqlock); 612 X_LOCK (reqlock);
584 ++nreqs; 613 ++nreqs;
585 X_UNLOCK (reqlock); 614 X_UNLOCK (reqlock);
… …
604 633
605 etp_maybe_start_thread (); 634 etp_maybe_start_thread ();
606 } 635 }
607} 636}
608 637
638static void ecb_cold
609static void etp_set_max_poll_time (double nseconds) 639etp_set_max_poll_time (double nseconds)
610{ 640{
611 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 641 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
612 max_poll_time = nseconds * EIO_TICKS; 642 max_poll_time = nseconds * EIO_TICKS;
613 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 643 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
614} 644}
615 645
646static void ecb_cold
616static void etp_set_max_poll_reqs (unsigned int maxreqs) 647etp_set_max_poll_reqs (unsigned int maxreqs)
617{ 648{
618 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 649 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
619 max_poll_reqs = maxreqs; 650 max_poll_reqs = maxreqs;
620 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 651 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
621} 652}
622 653
654static void ecb_cold
623static void etp_set_max_idle (unsigned int nthreads) 655etp_set_max_idle (unsigned int nthreads)
624{ 656{
625 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 657 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
626 max_idle = nthreads <= 0 ? 1 : nthreads; 658 max_idle = nthreads;
627 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 659 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
628} 660}
629 661
662static void ecb_cold
663etp_set_idle_timeout (unsigned int seconds)
664{
665 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
666 idle_timeout = seconds;
667 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
668}
669
670static void ecb_cold
630static void etp_set_min_parallel (unsigned int nthreads) 671etp_set_min_parallel (unsigned int nthreads)
631{ 672{
632 if (wanted < nthreads) 673 if (wanted < nthreads)
633 wanted = nthreads; 674 wanted = nthreads;
634} 675}
635 676
677static void ecb_cold
636static void etp_set_max_parallel (unsigned int nthreads) 678etp_set_max_parallel (unsigned int nthreads)
637{ 679{
638 if (wanted > nthreads) 680 if (wanted > nthreads)
639 wanted = nthreads; 681 wanted = nthreads;
640 682
641 while (started > wanted) 683 while (started > wanted)
642 etp_end_thread (); 684 etp_end_thread ();
643} 685}
644 686
645/*****************************************************************************/ 687/*****************************************************************************/
646 688
689static void
647static void grp_try_feed (eio_req *grp) 690grp_try_feed (eio_req *grp)
648{ 691{
649 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 692 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
650 { 693 {
651 grp->flags &= ~EIO_FLAG_GROUPADD; 694 grp->flags &= ~EIO_FLAG_GROUPADD;
652 695
… …
659 break; 702 break;
660 } 703 }
661 } 704 }
662} 705}
663 706
707static int
664static int grp_dec (eio_req *grp) 708grp_dec (eio_req *grp)
665{ 709{
666 --grp->size; 710 --grp->size;
667 711
668 /* call feeder, if applicable */ 712 /* call feeder, if applicable */
669 grp_try_feed (grp); 713 grp_try_feed (grp);
… …
673 return eio_finish (grp); 717 return eio_finish (grp);
674 else 718 else
675 return 0; 719 return 0;
676} 720}
677 721
722static void
678void eio_destroy (eio_req *req) 723eio_destroy (eio_req *req)
679{ 724{
680 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 725 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
681 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 726 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
682 727
683 EIO_DESTROY (req); 728 EIO_DESTROY (req);
684} 729}
685 730
731static int
686static int eio_finish (eio_req *req) 732eio_finish (eio_req *req)
687{ 733{
688 int res = EIO_FINISH (req); 734 int res = EIO_FINISH (req);
689 735
690 if (req->grp) 736 if (req->grp)
691 { 737 {
… …
699 if (grp->grp_first == req) 745 if (grp->grp_first == req)
700 grp->grp_first = req->grp_next; 746 grp->grp_first = req->grp_next;
701 747
702 res2 = grp_dec (grp); 748 res2 = grp_dec (grp);
703 749
704 if (!res && res2) 750 if (!res)
705 res = res2; 751 res = res2;
706 } 752 }
707 753
708 eio_destroy (req); 754 eio_destroy (req);
709 755
710 return res; 756 return res;
711} 757}
712 758
759void
713void eio_grp_cancel (eio_req *grp) 760eio_grp_cancel (eio_req *grp)
714{ 761{
715 for (grp = grp->grp_first; grp; grp = grp->grp_next) 762 for (grp = grp->grp_first; grp; grp = grp->grp_next)
716 eio_cancel (grp); 763 eio_cancel (grp);
717} 764}
718 765
766void
719void eio_cancel (eio_req *req) 767eio_cancel (eio_req *req)
720{ 768{
721 etp_cancel (req); 769 etp_cancel (req);
722} 770}
723 771
772void
724void eio_submit (eio_req *req) 773eio_submit (eio_req *req)
725{ 774{
726 etp_submit (req); 775 etp_submit (req);
727} 776}
728 777
729unsigned int eio_nreqs (void) 778unsigned int
779eio_nreqs (void)
730{ 780{
731 return etp_nreqs (); 781 return etp_nreqs ();
732} 782}
733 783
734unsigned int eio_nready (void) 784unsigned int
785eio_nready (void)
735{ 786{
736 return etp_nready (); 787 return etp_nready ();
737} 788}
738 789
739unsigned int eio_npending (void) 790unsigned int
791eio_npending (void)
740{ 792{
741 return etp_npending (); 793 return etp_npending ();
742} 794}
743 795
744unsigned int eio_nthreads (void) 796unsigned int ecb_cold
797eio_nthreads (void)
745{ 798{
746 return etp_nthreads (); 799 return etp_nthreads ();
747} 800}
748 801
802void ecb_cold
749void eio_set_max_poll_time (double nseconds) 803eio_set_max_poll_time (double nseconds)
750{ 804{
751 etp_set_max_poll_time (nseconds); 805 etp_set_max_poll_time (nseconds);
752} 806}
753 807
808void ecb_cold
754void eio_set_max_poll_reqs (unsigned int maxreqs) 809eio_set_max_poll_reqs (unsigned int maxreqs)
755{ 810{
756 etp_set_max_poll_reqs (maxreqs); 811 etp_set_max_poll_reqs (maxreqs);
757} 812}
758 813
814void ecb_cold
759void eio_set_max_idle (unsigned int nthreads) 815eio_set_max_idle (unsigned int nthreads)
760{ 816{
761 etp_set_max_idle (nthreads); 817 etp_set_max_idle (nthreads);
762} 818}
763 819
820void ecb_cold
821eio_set_idle_timeout (unsigned int seconds)
822{
823 etp_set_idle_timeout (seconds);
824}
825
826void ecb_cold
764void eio_set_min_parallel (unsigned int nthreads) 827eio_set_min_parallel (unsigned int nthreads)
765{ 828{
766 etp_set_min_parallel (nthreads); 829 etp_set_min_parallel (nthreads);
767} 830}
768 831
832void ecb_cold
769void eio_set_max_parallel (unsigned int nthreads) 833eio_set_max_parallel (unsigned int nthreads)
770{ 834{
771 etp_set_max_parallel (nthreads); 835 etp_set_max_parallel (nthreads);
772} 836}
773 837
774int eio_poll (void) 838int eio_poll (void)
… …
844#ifndef HAVE_FUTIMES 908#ifndef HAVE_FUTIMES
845 909
846# undef futimes 910# undef futimes
847# define futimes(fd,times) eio__futimes (fd, times) 911# define futimes(fd,times) eio__futimes (fd, times)
848 912
913static int
849static int eio__futimes (int fd, const struct timeval tv[2]) 914eio__futimes (int fd, const struct timeval tv[2])
850{ 915{
851 errno = ENOSYS; 916 errno = ENOSYS;
852 return -1; 917 return -1;
853} 918}
854 919
… …
858# undef fdatasync 923# undef fdatasync
859# define fdatasync(fd) fsync (fd) 924# define fdatasync(fd) fsync (fd)
860#endif 925#endif
861 926
862/* sync_file_range always needs emulation */ 927/* sync_file_range always needs emulation */
863int 928static int
864eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 929eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
865{ 930{
866#if HAVE_SYNC_FILE_RANGE 931#if HAVE_SYNC_FILE_RANGE
867 int res; 932 int res;
868 933
… …
914 979
915/* sendfile always needs emulation */ 980/* sendfile always needs emulation */
916static ssize_t 981static ssize_t
917eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 982eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
918{ 983{
984 ssize_t written = 0;
919 ssize_t res; 985 ssize_t res;
920 986
921 if (!count) 987 if (!count)
922 return 0; 988 return 0;
923 989
990 for (;;)
991 {
924#if HAVE_SENDFILE 992#if HAVE_SENDFILE
925# if __linux 993# if __linux
994 off_t soffset = offset;
926 res = sendfile (ofd, ifd, &offset, count); 995 res = sendfile (ofd, ifd, &soffset, count);
927 996
928# elif __FreeBSD__ 997# elif __FreeBSD__
929 /* 998 /*
930 * Of course, the freebsd sendfile is a dire hack with no thoughts 999 * Of course, the freebsd sendfile is a dire hack with no thoughts
931 * wasted on making it similar to other I/O functions. 1000 * wasted on making it similar to other I/O functions.
932 */ 1001 */
933 {
934 off_t sbytes; 1002 off_t sbytes;
935 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1003 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
936 1004
937 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1005 #if 0 /* according to the manpage, this is correct, but broken behaviour */
938 /* freebsd' sendfile will return 0 on success */ 1006 /* freebsd' sendfile will return 0 on success */
939 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1007 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
940 /* not on e.g. EIO or EPIPE - sounds broken */ 1008 /* not on e.g. EIO or EPIPE - sounds broken */
941 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1009 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
942 res = sbytes; 1010 res = sbytes;
943 #endif 1011 #endif
944 1012
945 /* according to source inspection, this is correct, and useful behaviour */ 1013 /* according to source inspection, this is correct, and useful behaviour */
946 if (sbytes) 1014 if (sbytes)
947 res = sbytes; 1015 res = sbytes;
948 }
949 1016
950# elif defined (__APPLE__) 1017# elif defined (__APPLE__) && 0 /* broken, as everything on os x */
951
952 {
953 off_t sbytes = count; 1018 off_t sbytes = count;
954 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1019 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
955 1020
956 /* according to the manpage, sbytes is always valid */ 1021 /* according to the manpage, sbytes is always valid */
957 if (sbytes) 1022 if (sbytes)
958 res = sbytes; 1023 res = sbytes;
959 }
960 1024
961# elif __hpux 1025# elif __hpux
962 res = sendfile (ofd, ifd, offset, count, 0, 0); 1026 res = sendfile (ofd, ifd, offset, count, 0, 0);
963 1027
964# elif __solaris 1028# elif __solaris
965 {
966 struct sendfilevec vec; 1029 struct sendfilevec vec;
967 size_t sbytes; 1030 size_t sbytes;
968 1031
969 vec.sfv_fd = ifd; 1032 vec.sfv_fd = ifd;
970 vec.sfv_flag = 0; 1033 vec.sfv_flag = 0;
971 vec.sfv_off = offset; 1034 vec.sfv_off = offset;
972 vec.sfv_len = count; 1035 vec.sfv_len = count;
973 1036
974 res = sendfilev (ofd, &vec, 1, &sbytes); 1037 res = sendfilev (ofd, &vec, 1, &sbytes);
975 1038
976 if (res < 0 && sbytes) 1039 if (res < 0 && sbytes)
977 res = sbytes; 1040 res = sbytes;
978 }
979 1041
980# endif 1042# endif
981 1043
982#elif defined (_WIN32) 1044#elif defined (_WIN32)
983
984 /* does not work, just for documentation of what would need to be done */ 1045 /* does not work, just for documentation of what would need to be done */
985 { 1046 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1047 /* libeio guarantees that the file offset does not change, and windows */
1048 /* has no way to get an independent handle to the same file description */
986 HANDLE h = TO_SOCKET (ifd); 1049 HANDLE h = TO_SOCKET (ifd);
987 SetFilePointer (h, offset, 0, FILE_BEGIN); 1050 SetFilePointer (h, offset, 0, FILE_BEGIN);
988 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1051 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
989 }
990 1052
991#else 1053#else
992 res = -1; 1054 res = -1;
993 errno = ENOSYS; 1055 errno = ENOSYS;
994#endif 1056#endif
995 1057
1058 /* we assume sendfile can copy at least 128mb in one go */
1059 if (res <= 128 * 1024 * 1024)
1060 {
1061 if (res > 0)
1062 written += res;
1063
1064 if (written)
1065 return written;
1066
1067 break;
1068 }
1069 else
1070 {
1071 /* if we requested more, then probably the kernel was lazy */
1072 written += res;
1073 offset += res;
1074 count -= res;
1075
1076 if (!count)
1077 return written;
1078 }
1079 }
1080
996 if (res < 0 1081 if (res < 0
997 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1082 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
998 /* BSDs */ 1083 /* BSDs */
999#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1084#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1000 || errno == ENOTSUP 1085 || errno == ENOTSUP
1001#endif 1086#endif
… …
1038 } 1123 }
1039 1124
1040 return res; 1125 return res;
1041} 1126}
1042 1127
1128#ifdef PAGESIZE
1129# define eio_pagesize() PAGESIZE
1130#else
1131static intptr_t
1132eio_pagesize (void)
1133{
1134 static intptr_t page;
1135
1136 if (!page)
1137 page = sysconf (_SC_PAGESIZE);
1138
1139 return page;
1140}
1141#endif
1142
1143static void
1144eio_page_align (void **addr, size_t *length)
1145{
1146 intptr_t mask = eio_pagesize () - 1;
1147
1148 /* round down addr */
1149 intptr_t adj = mask & (intptr_t)*addr;
1150
1151 *addr = (void *)((intptr_t)*addr - adj);
1152 *length += adj;
1153
1154 /* round up length */
1155 *length = (*length + mask) & ~mask;
1156}
1157
1158#if !_POSIX_MEMLOCK
1159# define eio__mlockall(a) ((errno = ENOSYS), -1)
1160#else
1161
1162static int
1163eio__mlockall (int flags)
1164{
1165 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1166 extern int mallopt (int, int);
1167 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1168 #endif
1169
1170 if (EIO_MCL_CURRENT != MCL_CURRENT
1171 || EIO_MCL_FUTURE != MCL_FUTURE)
1172 {
1173 flags = 0
1174 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1175 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1176 }
1177
1178 return mlockall (flags);
1179}
1180#endif
1181
1182#if !_POSIX_MEMLOCK_RANGE
1183# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1184#else
1185
1186static int
1187eio__mlock (void *addr, size_t length)
1188{
1189 eio_page_align (&addr, &length);
1190
1191 return mlock (addr, length);
1192}
1193
1194#endif
1195
1196#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1197# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1198#else
1199
1200static int
1201eio__msync (void *mem, size_t len, int flags)
1202{
1203 eio_page_align (&mem, &len);
1204
1205 if (EIO_MS_ASYNC != MS_SYNC
1206 || EIO_MS_INVALIDATE != MS_INVALIDATE
1207 || EIO_MS_SYNC != MS_SYNC)
1208 {
1209 flags = 0
1210 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1211 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1212 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1213 }
1214
1215 return msync (mem, len, flags);
1216}
1217
1218#endif
1219
1220static int
1221eio__mtouch (eio_req *req)
1222{
1223 void *mem = req->ptr2;
1224 size_t len = req->size;
1225 int flags = req->int1;
1226
1227 eio_page_align (&mem, &len);
1228
1229 {
1230 intptr_t addr = (intptr_t)mem;
1231 intptr_t end = addr + len;
1232 intptr_t page = eio_pagesize ();
1233
1234 if (addr < end)
1235 if (flags & EIO_MT_MODIFY) /* modify */
1236 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1237 else
1238 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1239 }
1240
1241 return 0;
1242}
1243
1244/*****************************************************************************/
1245/* requests implemented outside eio_execute, because they are so large */
1246
1247static void
1248eio__realpath (eio_req *req, etp_worker *self)
1249{
1250 char *rel = req->ptr1;
1251 char *res;
1252 char *tmp1, *tmp2;
1253#if SYMLOOP_MAX > 32
1254 int symlinks = SYMLOOP_MAX;
1255#else
1256 int symlinks = 32;
1257#endif
1258
1259 req->result = -1;
1260
1261 errno = EINVAL;
1262 if (!rel)
1263 return;
1264
1265 errno = ENOENT;
1266 if (!*rel)
1267 return;
1268
1269 if (!req->ptr2)
1270 {
1271 X_LOCK (wrklock);
1272 req->flags |= EIO_FLAG_PTR2_FREE;
1273 X_UNLOCK (wrklock);
1274 req->ptr2 = malloc (PATH_MAX * 3);
1275
1276 errno = ENOMEM;
1277 if (!req->ptr2)
1278 return;
1279 }
1280
1281 res = req->ptr2;
1282 tmp1 = res + PATH_MAX;
1283 tmp2 = tmp1 + PATH_MAX;
1284
1285#if 0 /* disabled, the musl way to do things is just too racy */
1286#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1287 /* on linux we may be able to ask the kernel */
1288 {
1289 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1290
1291 if (fd >= 0)
1292 {
1293 sprintf (tmp1, "/proc/self/fd/%d", fd);
1294 req->result = readlink (tmp1, res, PATH_MAX);
1295 close (fd);
1296
1297 /* here we should probably stat the open file and the disk file, to make sure they still match */
1298
1299 if (req->result > 0)
1300 goto done;
1301 }
1302 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1303 return;
1304 }
1305#endif
1306#endif
1307
1308 if (*rel != '/')
1309 {
1310 if (!getcwd (res, PATH_MAX))
1311 return;
1312
1313 if (res [1]) /* only use if not / */
1314 res += strlen (res);
1315 }
1316
1317 while (*rel)
1318 {
1319 ssize_t len, linklen;
1320 char *beg = rel;
1321
1322 while (*rel && *rel != '/')
1323 ++rel;
1324
1325 len = rel - beg;
1326
1327 if (!len) /* skip slashes */
1328 {
1329 ++rel;
1330 continue;
1331 }
1332
1333 if (beg [0] == '.')
1334 {
1335 if (len == 1)
1336 continue; /* . - nop */
1337
1338 if (beg [1] == '.' && len == 2)
1339 {
1340 /* .. - back up one component, if possible */
1341
1342 while (res != req->ptr2)
1343 if (*--res == '/')
1344 break;
1345
1346 continue;
1347 }
1348 }
1349
1350 errno = ENAMETOOLONG;
1351 if (res + 1 + len + 1 >= tmp1)
1352 return;
1353
1354 /* copy one component */
1355 *res = '/';
1356 memcpy (res + 1, beg, len);
1357
1358 /* zero-terminate, for readlink */
1359 res [len + 1] = 0;
1360
1361 /* now check if it's a symlink */
1362 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1363
1364 if (linklen < 0)
1365 {
1366 if (errno != EINVAL)
1367 return;
1368
1369 /* it's a normal directory. hopefully */
1370 res += len + 1;
1371 }
1372 else
1373 {
1374 /* yay, it was a symlink - build new path in tmp2 */
1375 int rellen = strlen (rel);
1376
1377 errno = ENAMETOOLONG;
1378 if (linklen + 1 + rellen >= PATH_MAX)
1379 return;
1380
1381 errno = ELOOP;
1382 if (!--symlinks)
1383 return;
1384
1385 if (*tmp1 == '/')
1386 res = req->ptr2; /* symlink resolves to an absolute path */
1387
1388 /* we need to be careful, as rel might point into tmp2 already */
1389 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1390 tmp2 [linklen] = '/';
1391 memcpy (tmp2, tmp1, linklen);
1392
1393 rel = tmp2;
1394 }
1395 }
1396
1397 /* special case for the lone root path */
1398 if (res == req->ptr2)
1399 *res++ = '/';
1400
1401 req->result = res - (char *)req->ptr2;
1402
1403done:
1404 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1405}
1406
1043static signed char 1407static signed char
1044eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1408eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1045{ 1409{
1046 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1410 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; 1411 : a->inode < b->inode ? -1
1412 : a->inode > b->inode ? 1
1413 : 0;
1048} 1414}
1049 1415
1050#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1416#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1051 1417
1052#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1418#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
… …
1058 unsigned char bits [9 + sizeof (ino_t) * 8]; 1424 unsigned char bits [9 + sizeof (ino_t) * 8];
1059 unsigned char *bit = bits; 1425 unsigned char *bit = bits;
1060 1426
1061 assert (CHAR_BIT == 8); 1427 assert (CHAR_BIT == 8);
1062 assert (sizeof (eio_dirent) * 8 < 256); 1428 assert (sizeof (eio_dirent) * 8 < 256);
1063 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1429 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1064 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1430 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1065 1431
1066 if (size <= EIO_SORT_FAST) 1432 if (size <= EIO_SORT_FAST)
1067 return; 1433 return;
1068 1434
1069 /* first prepare an array of bits to test in our radix sort */ 1435 /* first prepare an array of bits to test in our radix sort */
… …
1224 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1590 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1225 1591
1226 X_LOCK (wrklock); 1592 X_LOCK (wrklock);
1227 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1593 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1228 self->dirp = dirp = opendir (req->ptr1); 1594 self->dirp = dirp = opendir (req->ptr1);
1595
1596 if (req->flags & EIO_FLAG_PTR1_FREE)
1597 free (req->ptr1);
1598
1229 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1599 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1230 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1600 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1231 req->ptr2 = names = malloc (namesalloc); 1601 req->ptr2 = names = malloc (namesalloc);
1232 X_UNLOCK (wrklock); 1602 X_UNLOCK (wrklock);
1233 1603
… …
1245 /* sort etc. */ 1615 /* sort etc. */
1246 req->int1 = flags; 1616 req->int1 = flags;
1247 req->result = dentoffs; 1617 req->result = dentoffs;
1248 1618
1249 if (flags & EIO_READDIR_STAT_ORDER) 1619 if (flags & EIO_READDIR_STAT_ORDER)
1250 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1620 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1251 else if (flags & EIO_READDIR_DIRS_FIRST) 1621 else if (flags & EIO_READDIR_DIRS_FIRST)
1252 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1622 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1253 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1623 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1254 else 1624 else
1255 { 1625 {
… …
1257 eio_dirent *oth = dents + dentoffs; 1627 eio_dirent *oth = dents + dentoffs;
1258 eio_dirent *dir = dents; 1628 eio_dirent *dir = dents;
1259 1629
1260 /* now partition dirs to the front, and non-dirs to the back */ 1630 /* now partition dirs to the front, and non-dirs to the back */
1261 /* by walking from both sides and swapping if necessary */ 1631 /* by walking from both sides and swapping if necessary */
1262 /* also clear score, so it doesn't influence sorting */
1263 while (oth > dir) 1632 while (oth > dir)
1264 { 1633 {
1265 if (dir->type == EIO_DT_DIR) 1634 if (dir->type == EIO_DT_DIR)
1266 ++dir; 1635 ++dir;
1267 else if ((--oth)->type == EIO_DT_DIR) 1636 else if ((--oth)->type == EIO_DT_DIR)
… …
1270 1639
1271 ++dir; 1640 ++dir;
1272 } 1641 }
1273 } 1642 }
1274 1643
1275 /* now sort the dirs only */ 1644 /* now sort the dirs only (dirs all have the same score) */
1276 eio_dent_sort (dents, dir - dents, 0, inode_bits); 1645 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1277 } 1646 }
1278 1647
1279 break; 1648 break;
1280 } 1649 }
… …
1285 /* skip . and .. entries */ 1654 /* skip . and .. entries */
1286 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1655 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1287 { 1656 {
1288 int len = D_NAMLEN (entp) + 1; 1657 int len = D_NAMLEN (entp) + 1;
1289 1658
1290 while (expect_false (namesoffs + len > namesalloc)) 1659 while (ecb_expect_false (namesoffs + len > namesalloc))
1291 { 1660 {
1292 namesalloc *= 2; 1661 namesalloc *= 2;
1293 X_LOCK (wrklock); 1662 X_LOCK (wrklock);
1294 req->ptr2 = names = realloc (names, namesalloc); 1663 req->ptr2 = names = realloc (names, namesalloc);
1295 X_UNLOCK (wrklock); 1664 X_UNLOCK (wrklock);
… …
1302 1671
1303 if (dents) 1672 if (dents)
1304 { 1673 {
1305 struct eio_dirent *ent; 1674 struct eio_dirent *ent;
1306 1675
1307 if (expect_false (dentoffs == dentalloc)) 1676 if (ecb_expect_false (dentoffs == dentalloc))
1308 { 1677 {
1309 dentalloc *= 2; 1678 dentalloc *= 2;
1310 X_LOCK (wrklock); 1679 X_LOCK (wrklock);
1311 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1680 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1312 X_UNLOCK (wrklock); 1681 X_UNLOCK (wrklock);
… …
1400 break; 1769 break;
1401 } 1770 }
1402 } 1771 }
1403} 1772}
1404 1773
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/*****************************************************************************/ 1774/*****************************************************************************/
1518 1775
1519#define ALLOC(len) \ 1776#define ALLOC(len) \
1520 if (!req->ptr2) \ 1777 if (!req->ptr2) \
1521 { \ 1778 { \
… …
1551 if (req) 1808 if (req)
1552 break; 1809 break;
1553 1810
1554 ++idle; 1811 ++idle;
1555 1812
1556 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1813 ts.tv_sec = time (0) + idle_timeout;
1557 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1814 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1558 { 1815 {
1559 if (idle > max_idle) 1816 if (idle > max_idle)
1560 { 1817 {
1561 --idle; 1818 --idle;
… …
1578 X_UNLOCK (reqlock); 1835 X_UNLOCK (reqlock);
1579 1836
1580 if (req->type < 0) 1837 if (req->type < 0)
1581 goto quit; 1838 goto quit;
1582 1839
1583 if (!EIO_CANCELLED (req))
1584 ETP_EXECUTE (self, req); 1840 ETP_EXECUTE (self, req);
1585 1841
1586 X_LOCK (reslock); 1842 X_LOCK (reslock);
1587 1843
1588 ++npending; 1844 ++npending;
1589 1845
… …
1604 return 0; 1860 return 0;
1605} 1861}
1606 1862
1607/*****************************************************************************/ 1863/*****************************************************************************/
1608 1864
1865int ecb_cold
1609int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1866eio_init (void (*want_poll)(void), void (*done_poll)(void))
1610{ 1867{
1611 return etp_init (want_poll, done_poll); 1868 return etp_init (want_poll, done_poll);
1612} 1869}
1613 1870
1871ecb_inline void
1614static void eio_api_destroy (eio_req *req) 1872eio_api_destroy (eio_req *req)
1615{ 1873{
1616 free (req); 1874 free (req);
1617} 1875}
1618 1876
1619#define REQ(rtype) \ 1877#define REQ(rtype) \
… …
1638 { \ 1896 { \
1639 eio_api_destroy (req); \ 1897 eio_api_destroy (req); \
1640 return 0; \ 1898 return 0; \
1641 } 1899 }
1642 1900
1901static void
1643static void eio_execute (etp_worker *self, eio_req *req) 1902eio_execute (etp_worker *self, eio_req *req)
1644{ 1903{
1904 if (ecb_expect_false (EIO_CANCELLED (req)))
1905 {
1906 req->result = -1;
1907 req->errorno = ECANCELED;
1908 return;
1909 }
1910
1645 switch (req->type) 1911 switch (req->type)
1646 { 1912 {
1647 case EIO_READ: ALLOC (req->size); 1913 case EIO_READ: ALLOC (req->size);
1648 req->result = req->offs >= 0 1914 req->result = req->offs >= 0
1649 ? pread (req->int1, req->ptr2, req->size, req->offs) 1915 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1683 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1949 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1684 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1950 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1685 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1951 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; 1952 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1687 1953
1954 case EIO_REALPATH: eio__realpath (req, self); break;
1955
1688 case EIO_READLINK: ALLOC (PATH_MAX); 1956 case EIO_READLINK: ALLOC (PATH_MAX);
1689 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; 1957 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1690 1958
1691 case EIO_SYNC: req->result = 0; sync (); break; 1959 case EIO_SYNC: req->result = 0; sync (); break;
1692 case EIO_FSYNC: req->result = fsync (req->int1); break; 1960 case EIO_FSYNC: req->result = fsync (req->int1); break;
1693 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1961 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1694 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 1962 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; 1963 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1696 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 1964 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1697 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 1965 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; 1966 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1699 1967
1700 case EIO_READDIR: eio__scandir (req, self); break; 1968 case EIO_READDIR: eio__scandir (req, self); break;
… …
1744 case EIO_NOP: 2012 case EIO_NOP:
1745 req->result = 0; 2013 req->result = 0;
1746 break; 2014 break;
1747 2015
1748 case EIO_CUSTOM: 2016 case EIO_CUSTOM:
1749 ((void (*)(eio_req *))req->feed) (req); 2017 req->feed (req);
1750 break; 2018 break;
1751 2019
1752 default: 2020 default:
1753 errno = ENOSYS; 2021 errno = ENOSYS;
1754 req->result = -1; 2022 req->result = -1;
… …
1907} 2175}
1908 2176
1909eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2177eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1910{ 2178{
1911 return eio__1path (EIO_READLINK, path, pri, cb, data); 2179 return eio__1path (EIO_READLINK, path, pri, cb, data);
2180}
2181
2182eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2183{
2184 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1912} 2185}
1913 2186
1914eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2187eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1915{ 2188{
1916 return eio__1path (EIO_STAT, path, pri, cb, data); 2189 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) 2248eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1976{ 2249{
1977 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2250 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1978} 2251}
1979 2252
1980eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2253eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1981{ 2254{
1982 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2255 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1983} 2256}
1984 2257
1985#endif 2258#endif
1986 2259
1987eio_req *eio_grp (eio_cb cb, void *data) 2260eio_req *eio_grp (eio_cb cb, void *data)
… …
1996#undef SEND 2269#undef SEND
1997 2270
1998/*****************************************************************************/ 2271/*****************************************************************************/
1999/* grp functions */ 2272/* grp functions */
2000 2273
2274void
2001void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2275eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2002{ 2276{
2003 grp->int2 = limit; 2277 grp->int2 = limit;
2004 grp->feed = feed; 2278 grp->feed = feed;
2005 2279
2006 grp_try_feed (grp); 2280 grp_try_feed (grp);
2007} 2281}
2008 2282
2283void
2009void eio_grp_limit (eio_req *grp, int limit) 2284eio_grp_limit (eio_req *grp, int limit)
2010{ 2285{
2011 grp->int2 = limit; 2286 grp->int2 = limit;
2012 2287
2013 grp_try_feed (grp); 2288 grp_try_feed (grp);
2014} 2289}
2015 2290
2291void
2016void eio_grp_add (eio_req *grp, eio_req *req) 2292eio_grp_add (eio_req *grp, eio_req *req)
2017{ 2293{
2018 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2294 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2019 2295
2020 grp->flags |= EIO_FLAG_GROUPADD; 2296 grp->flags |= EIO_FLAG_GROUPADD;
2021 2297
… …
2032} 2308}
2033 2309
2034/*****************************************************************************/ 2310/*****************************************************************************/
2035/* misc garbage */ 2311/* misc garbage */
2036 2312
2313ssize_t
2037ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2314eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2038{ 2315{
2039 etp_worker wrk; 2316 etp_worker wrk;
2040 ssize_t ret; 2317 ssize_t ret;
2041 2318
2042 wrk.dbuf = 0; 2319 wrk.dbuf = 0;

Diff Legend

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