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 1.23 by root, Sun Sep 21 00:23:45 2008 UTC vs.
Revision 1.66 by root, Sun Jun 5 19:58:37 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45
46#ifdef EIO_STACKSIZE
47# define XTHREAD_STACKSIZE EIO_STACKSIZE
48#endif
41#include "xthread.h" 49#include "xthread.h"
42 50
43#include <errno.h> 51#include <errno.h>
44#include <stddef.h> 52#include <stddef.h>
45#include <stdlib.h> 53#include <stdlib.h>
46#include <string.h> 54#include <string.h>
47#include <errno.h> 55#include <errno.h>
48#include <sys/types.h> 56#include <sys/types.h>
49#include <sys/stat.h> 57#include <sys/stat.h>
58#include <sys/statvfs.h>
50#include <limits.h> 59#include <limits.h>
51#include <fcntl.h> 60#include <fcntl.h>
52#include <assert.h> 61#include <assert.h>
53 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes form stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
54#ifndef EIO_FINISH 69#ifndef EIO_FINISH
55# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 70# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
56#endif 71#endif
57 72
58#ifndef EIO_DESTROY 73#ifndef EIO_DESTROY
… …
64#endif 79#endif
65 80
66#ifdef _WIN32 81#ifdef _WIN32
67 82
68 /*doh*/ 83 /*doh*/
69
70#else 84#else
71 85
72# include "config.h"
73# include <sys/time.h> 86# include <sys/time.h>
74# include <sys/select.h> 87# include <sys/select.h>
75# include <unistd.h> 88# include <unistd.h>
76# include <utime.h> 89# include <utime.h>
77# include <signal.h> 90# include <signal.h>
78# include <dirent.h> 91# include <dirent.h>
79 92
93#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94# include <sys/mman.h>
95#endif
96
97/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99# define _DIRENT_HAVE_D_TYPE /* sigh */
100# define D_INO(de) (de)->d_fileno
101# define D_NAMLEN(de) (de)->d_namlen
102# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103# define D_INO(de) (de)->d_ino
104# endif
105
106#ifdef _D_EXACT_NAMLEN
107# undef D_NAMLEN
108# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109#endif
110
111# ifdef _DIRENT_HAVE_D_TYPE
112# define D_TYPE(de) (de)->d_type
113# endif
114
80# ifndef EIO_STRUCT_DIRENT 115# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 116# define EIO_STRUCT_DIRENT struct dirent
82# endif 117# endif
83 118
84#endif 119#endif
85 120
86#if HAVE_SENDFILE 121#if HAVE_SENDFILE
87# if __linux 122# if __linux
88# include <sys/sendfile.h> 123# include <sys/sendfile.h>
89# elif __freebsd 124# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 125# include <sys/socket.h>
91# include <sys/uio.h> 126# include <sys/uio.h>
92# elif __hpux 127# elif __hpux
93# include <sys/socket.h> 128# include <sys/socket.h>
94# elif __solaris /* not yet */ 129# elif __solaris
95# include <sys/sendfile.h> 130# include <sys/sendfile.h>
96# else 131# else
97# error sendfile support requested but not available 132# error sendfile support requested but not available
98# endif 133# endif
99#endif 134#endif
100 135
101/* number of seconds after which an idle threads exit */ 136#ifndef D_TYPE
102#define IDLE_TIMEOUT 10 137# define D_TYPE(de) 0
138#endif
139#ifndef D_INO
140# define D_INO(de) 0
141#endif
142#ifndef D_NAMLEN
143# define D_NAMLEN(de) strlen ((de)->d_name)
144#endif
103 145
104/* used for struct dirent, AIX doesn't provide it */ 146/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 147#ifndef NAME_MAX
106# define NAME_MAX 4096 148# define NAME_MAX 4096
149#endif
150
151/* used for readlink etc. */
152#ifndef PATH_MAX
153# define PATH_MAX 4096
107#endif 154#endif
108 155
109/* buffer size for various temporary buffers */ 156/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 157#define EIO_BUFSIZE 65536
111 158
… …
117 errno = ENOMEM; \ 164 errno = ENOMEM; \
118 if (!eio_buf) \ 165 if (!eio_buf) \
119 return -1; 166 return -1;
120 167
121#define EIO_TICKS ((1000000 + 1023) >> 10) 168#define EIO_TICKS ((1000000 + 1023) >> 10)
169
170/*****************************************************************************/
171
172#if __GNUC__ >= 3
173# define expect(expr,value) __builtin_expect ((expr),(value))
174#else
175# define expect(expr,value) (expr)
176#endif
177
178#define expect_false(expr) expect ((expr) != 0, 0)
179#define expect_true(expr) expect ((expr) != 0, 1)
122 180
123/*****************************************************************************/ 181/*****************************************************************************/
124 182
125#define ETP_PRI_MIN EIO_PRI_MIN 183#define ETP_PRI_MIN EIO_PRI_MIN
126#define ETP_PRI_MAX EIO_PRI_MAX 184#define ETP_PRI_MAX EIO_PRI_MAX
… …
144 if (wrk->dirp) \ 202 if (wrk->dirp) \
145 { \ 203 { \
146 closedir (wrk->dirp); \ 204 closedir (wrk->dirp); \
147 wrk->dirp = 0; \ 205 wrk->dirp = 0; \
148 } 206 }
207
149#define ETP_WORKER_COMMON \ 208#define ETP_WORKER_COMMON \
150 void *dbuf; \ 209 void *dbuf; \
151 DIR *dirp; 210 DIR *dirp;
152 211
153/*****************************************************************************/ 212/*****************************************************************************/
154 213
155#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 214#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
156 215
157/* calculcate time difference in ~1/EIO_TICKS of a second */ 216/* calculate time difference in ~1/EIO_TICKS of a second */
158static int tvdiff (struct timeval *tv1, struct timeval *tv2) 217static int tvdiff (struct timeval *tv1, struct timeval *tv2)
159{ 218{
160 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 219 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
161 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 220 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
162} 221}
… …
170static unsigned int max_poll_reqs; /* reslock */ 229static unsigned int max_poll_reqs; /* reslock */
171 230
172static volatile unsigned int nreqs; /* reqlock */ 231static volatile unsigned int nreqs; /* reqlock */
173static volatile unsigned int nready; /* reqlock */ 232static volatile unsigned int nready; /* reqlock */
174static volatile unsigned int npending; /* reqlock */ 233static volatile unsigned int npending; /* reqlock */
175static volatile unsigned int max_idle = 4; 234static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
235static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
176 236
177static mutex_t wrklock = X_MUTEX_INIT; 237static xmutex_t wrklock;
178static mutex_t reslock = X_MUTEX_INIT; 238static xmutex_t reslock;
179static mutex_t reqlock = X_MUTEX_INIT; 239static xmutex_t reqlock;
180static cond_t reqwait = X_COND_INIT; 240static xcond_t reqwait;
181 241
182#if !HAVE_PREADWRITE 242#if !HAVE_PREADWRITE
183/* 243/*
184 * make our pread/pwrite emulation safe against themselves, but not against 244 * make our pread/pwrite emulation safe against themselves, but not against
185 * normal read/write by using a mutex. slows down execution a lot, 245 * normal read/write by using a mutex. slows down execution a lot,
186 * but that's your problem, not mine. 246 * but that's your problem, not mine.
187 */ 247 */
188static mutex_t preadwritelock = X_MUTEX_INIT; 248static xmutex_t preadwritelock = X_MUTEX_INIT;
189#endif 249#endif
190 250
191typedef struct etp_worker 251typedef struct etp_worker
192{ 252{
193 /* locked by wrklock */ 253 /* locked by wrklock */
194 struct etp_worker *prev, *next; 254 struct etp_worker *prev, *next;
195 255
196 thread_t tid; 256 xthread_t tid;
197 257
198 /* locked by reslock, reqlock or wrklock */ 258 /* locked by reslock, reqlock or wrklock */
199 ETP_REQ *req; /* currently processed request */ 259 ETP_REQ *req; /* currently processed request */
200 260
201 ETP_WORKER_COMMON 261 ETP_WORKER_COMMON
… …
315 } 375 }
316 376
317 abort (); 377 abort ();
318} 378}
319 379
380static void etp_thread_init (void)
381{
382 X_MUTEX_CREATE (wrklock);
383 X_MUTEX_CREATE (reslock);
384 X_MUTEX_CREATE (reqlock);
385 X_COND_CREATE (reqwait);
386}
387
320static void etp_atfork_prepare (void) 388static void etp_atfork_prepare (void)
321{ 389{
322 X_LOCK (wrklock); 390 X_LOCK (wrklock);
323 X_LOCK (reqlock); 391 X_LOCK (reqlock);
324 X_LOCK (reslock); 392 X_LOCK (reslock);
… …
362 idle = 0; 430 idle = 0;
363 nreqs = 0; 431 nreqs = 0;
364 nready = 0; 432 nready = 0;
365 npending = 0; 433 npending = 0;
366 434
367 etp_atfork_parent (); 435 etp_thread_init ();
368} 436}
369 437
370static void 438static void
371etp_once_init (void) 439etp_once_init (void)
372{ 440{
441 etp_thread_init ();
373 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 442 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
374} 443}
375 444
376static int 445static int
377etp_init (void (*want_poll)(void), void (*done_poll)(void)) 446etp_init (void (*want_poll)(void), void (*done_poll)(void))
… …
411 X_UNLOCK (wrklock); 480 X_UNLOCK (wrklock);
412} 481}
413 482
414static void etp_maybe_start_thread (void) 483static void etp_maybe_start_thread (void)
415{ 484{
416 if (etp_nthreads () >= wanted) 485 if (expect_true (etp_nthreads () >= wanted))
417 return; 486 return;
418 487
419 /* todo: maybe use idle here, but might be less exact */ 488 /* todo: maybe use idle here, but might be less exact */
420 if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) 489 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
421 return; 490 return;
422 491
423 etp_start_thread (); 492 etp_start_thread ();
424} 493}
425 494
… …
478 547
479 X_LOCK (reqlock); 548 X_LOCK (reqlock);
480 --nreqs; 549 --nreqs;
481 X_UNLOCK (reqlock); 550 X_UNLOCK (reqlock);
482 551
483 if (req->type == EIO_GROUP && req->size) 552 if (expect_false (req->type == EIO_GROUP && req->size))
484 { 553 {
485 req->int1 = 1; /* mark request as delayed */ 554 req->int1 = 1; /* mark request as delayed */
486 continue; 555 continue;
487 } 556 }
488 else 557 else
489 { 558 {
490 int res = ETP_FINISH (req); 559 int res = ETP_FINISH (req);
491 if (res) 560 if (expect_false (res))
492 return res; 561 return res;
493 } 562 }
494 563
495 if (maxreqs && !--maxreqs) 564 if (expect_false (maxreqs && !--maxreqs))
496 break; 565 break;
497 566
498 if (maxtime) 567 if (maxtime)
499 { 568 {
500 gettimeofday (&tv_now, 0); 569 gettimeofday (&tv_now, 0);
… …
519 588
520static void etp_submit (ETP_REQ *req) 589static void etp_submit (ETP_REQ *req)
521{ 590{
522 req->pri -= ETP_PRI_MIN; 591 req->pri -= ETP_PRI_MIN;
523 592
524 if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 593 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
525 if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 594 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
526 595
596 if (expect_false (req->type == EIO_GROUP))
597 {
598 /* I hope this is worth it :/ */
527 X_LOCK (reqlock); 599 X_LOCK (reqlock);
528 ++nreqs; 600 ++nreqs;
601 X_UNLOCK (reqlock);
602
603 X_LOCK (reslock);
604
605 ++npending;
606
607 if (!reqq_push (&res_queue, req) && want_poll_cb)
608 want_poll_cb ();
609
610 X_UNLOCK (reslock);
611 }
612 else
613 {
614 X_LOCK (reqlock);
615 ++nreqs;
529 ++nready; 616 ++nready;
530 reqq_push (&req_queue, req); 617 reqq_push (&req_queue, req);
531 X_COND_SIGNAL (reqwait); 618 X_COND_SIGNAL (reqwait);
532 X_UNLOCK (reqlock); 619 X_UNLOCK (reqlock);
533 620
534 etp_maybe_start_thread (); 621 etp_maybe_start_thread ();
622 }
535} 623}
536 624
537static void etp_set_max_poll_time (double nseconds) 625static void etp_set_max_poll_time (double nseconds)
538{ 626{
539 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 627 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
540 max_poll_time = nseconds; 628 max_poll_time = nseconds * EIO_TICKS;
541 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 629 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
542} 630}
543 631
544static void etp_set_max_poll_reqs (unsigned int maxreqs) 632static void etp_set_max_poll_reqs (unsigned int maxreqs)
545{ 633{
… …
549} 637}
550 638
551static void etp_set_max_idle (unsigned int nthreads) 639static void etp_set_max_idle (unsigned int nthreads)
552{ 640{
553 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 641 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
554 max_idle = nthreads <= 0 ? 1 : nthreads; 642 max_idle = nthreads;
643 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644}
645
646static void etp_set_idle_timeout (unsigned int seconds)
647{
648 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649 idle_timeout = seconds;
555 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 650 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
556} 651}
557 652
558static void etp_set_min_parallel (unsigned int nthreads) 653static void etp_set_min_parallel (unsigned int nthreads)
559{ 654{
… …
574 669
575static void grp_try_feed (eio_req *grp) 670static void grp_try_feed (eio_req *grp)
576{ 671{
577 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 672 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
578 { 673 {
579 int old_len = grp->size; 674 grp->flags &= ~EIO_FLAG_GROUPADD;
580 675
581 EIO_FEED (grp); 676 EIO_FEED (grp);
582 677
583 /* stop if no progress has been made */ 678 /* stop if no progress has been made */
584 if (old_len == grp->size) 679 if (!(grp->flags & EIO_FLAG_GROUPADD))
585 { 680 {
586 grp->feed = 0; 681 grp->feed = 0;
587 break; 682 break;
588 } 683 }
589 } 684 }
… …
687void eio_set_max_idle (unsigned int nthreads) 782void eio_set_max_idle (unsigned int nthreads)
688{ 783{
689 etp_set_max_idle (nthreads); 784 etp_set_max_idle (nthreads);
690} 785}
691 786
787void eio_set_idle_timeout (unsigned int seconds)
788{
789 etp_set_idle_timeout (seconds);
790}
791
692void eio_set_min_parallel (unsigned int nthreads) 792void eio_set_min_parallel (unsigned int nthreads)
693{ 793{
694 etp_set_min_parallel (nthreads); 794 etp_set_min_parallel (nthreads);
695} 795}
696 796
… …
706 806
707/*****************************************************************************/ 807/*****************************************************************************/
708/* work around various missing functions */ 808/* work around various missing functions */
709 809
710#if !HAVE_PREADWRITE 810#if !HAVE_PREADWRITE
811# undef pread
812# undef pwrite
711# define pread eio__pread 813# define pread eio__pread
712# define pwrite eio__pwrite 814# define pwrite eio__pwrite
713 815
714static ssize_t 816static ssize_t
715eio__pread (int fd, void *buf, size_t count, off_t offset) 817eio__pread (int fd, void *buf, size_t count, off_t offset)
… …
735 837
736 X_LOCK (preadwritelock); 838 X_LOCK (preadwritelock);
737 ooffset = lseek (fd, 0, SEEK_CUR); 839 ooffset = lseek (fd, 0, SEEK_CUR);
738 lseek (fd, offset, SEEK_SET); 840 lseek (fd, offset, SEEK_SET);
739 res = write (fd, buf, count); 841 res = write (fd, buf, count);
740 lseek (fd, offset, SEEK_SET); 842 lseek (fd, ooffset, SEEK_SET);
741 X_UNLOCK (preadwritelock); 843 X_UNLOCK (preadwritelock);
742 844
743 return res; 845 return res;
744} 846}
745#endif 847#endif
746 848
747#ifndef HAVE_FUTIMES 849#ifndef HAVE_UTIMES
748 850
851# undef utimes
749# define utimes(path,times) eio__utimes (path, times) 852# define utimes(path,times) eio__utimes (path, times)
750# define futimes(fd,times) eio__futimes (fd, times)
751 853
752static int 854static int
753eio__utimes (const char *filename, const struct timeval times[2]) 855eio__utimes (const char *filename, const struct timeval times[2])
754{ 856{
755 if (times) 857 if (times)
… …
763 } 865 }
764 else 866 else
765 return utime (filename, 0); 867 return utime (filename, 0);
766} 868}
767 869
870#endif
871
872#ifndef HAVE_FUTIMES
873
874# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times)
876
768static int eio__futimes (int fd, const struct timeval tv[2]) 877static int eio__futimes (int fd, const struct timeval tv[2])
769{ 878{
770 errno = ENOSYS; 879 errno = ENOSYS;
771 return -1; 880 return -1;
772} 881}
773 882
774#endif 883#endif
775 884
776#if !HAVE_FDATASYNC 885#if !HAVE_FDATASYNC
886# undef fdatasync
777# define fdatasync fsync 887# define fdatasync(fd) fsync (fd)
778#endif 888#endif
889
890/* sync_file_range always needs emulation */
891int
892eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893{
894#if HAVE_SYNC_FILE_RANGE
895 int res;
896
897 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
898 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
899 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
900 {
901 flags = 0
902 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
903 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
904 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
905 }
906
907 res = sync_file_range (fd, offset, nbytes, flags);
908
909 if (!res || errno != ENOSYS)
910 return res;
911#endif
912
913 /* even though we could play tricks with the flags, it's better to always
914 * call fdatasync, as that matches the expectation of its users best */
915 return fdatasync (fd);
916}
779 917
780#if !HAVE_READAHEAD 918#if !HAVE_READAHEAD
919# undef readahead
781# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 920# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
782 921
783static ssize_t 922static ssize_t
784eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 923eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
785{ 924{
… …
812 951
813#if HAVE_SENDFILE 952#if HAVE_SENDFILE
814# if __linux 953# if __linux
815 res = sendfile (ofd, ifd, &offset, count); 954 res = sendfile (ofd, ifd, &offset, count);
816 955
817# elif __freebsd 956# elif __FreeBSD__
818 /* 957 /*
819 * Of course, the freebsd sendfile is a dire hack with no thoughts 958 * Of course, the freebsd sendfile is a dire hack with no thoughts
820 * wasted on making it similar to other I/O functions. 959 * wasted on making it similar to other I/O functions.
821 */ 960 */
822 { 961 {
823 off_t sbytes; 962 off_t sbytes;
824 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 963 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
825 964
826 if (res < 0 && sbytes) 965 #if 0 /* according to the manpage, this is correct, but broken behaviour */
827 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 966 /* freebsd' sendfile will return 0 on success */
967 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
968 /* not on e.g. EIO or EPIPE - sounds broken */
969 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
970 res = sbytes;
971 #endif
972
973 /* according to source inspection, this is correct, and useful behaviour */
974 if (sbytes)
975 res = sbytes;
976 }
977
978# elif defined (__APPLE__)
979
980 {
981 off_t sbytes = count;
982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983
984 /* according to the manpage, sbytes is always valid */
985 if (sbytes)
828 res = sbytes; 986 res = sbytes;
829 } 987 }
830 988
831# elif __hpux 989# elif __hpux
832 res = sendfile (ofd, ifd, offset, count, 0, 0); 990 res = sendfile (ofd, ifd, offset, count, 0, 0);
… …
846 if (res < 0 && sbytes) 1004 if (res < 0 && sbytes)
847 res = sbytes; 1005 res = sbytes;
848 } 1006 }
849 1007
850# endif 1008# endif
1009
1010#elif defined (_WIN32)
1011
1012 /* does not work, just for documentation of what would need to be done */
1013 {
1014 HANDLE h = TO_SOCKET (ifd);
1015 SetFilePointer (h, offset, 0, FILE_BEGIN);
1016 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1017 }
1018
851#else 1019#else
852 res = -1; 1020 res = -1;
853 errno = ENOSYS; 1021 errno = ENOSYS;
854#endif 1022#endif
855 1023
856 if (res < 0 1024 if (res < 0
857 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1025 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1026 /* BSDs */
1027#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1028 || errno == ENOTSUP
1029#endif
1030 || errno == EOPNOTSUPP /* BSDs */
858#if __solaris 1031#if __solaris
859 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1032 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
860#endif 1033#endif
861 ) 1034 )
862 ) 1035 )
… …
893 } 1066 }
894 1067
895 return res; 1068 return res;
896} 1069}
897 1070
1071static signed char
1072eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1073{
1074 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1075 : a->inode < b->inode ? -1
1076 : a->inode > b->inode ? 1
1077 : 0;
1078}
1079
1080#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1081
1082#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1083#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1084
1085static void
1086eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1087{
1088 unsigned char bits [9 + sizeof (ino_t) * 8];
1089 unsigned char *bit = bits;
1090
1091 assert (CHAR_BIT == 8);
1092 assert (sizeof (eio_dirent) * 8 < 256);
1093 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1094 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1095
1096 if (size <= EIO_SORT_FAST)
1097 return;
1098
1099 /* first prepare an array of bits to test in our radix sort */
1100 /* try to take endianness into account, as well as differences in ino_t sizes */
1101 /* inode_bits must contain all inodes ORed together */
1102 /* which is used to skip bits that are 0 everywhere, which is very common */
1103 {
1104 ino_t endianness;
1105 int i, j;
1106
1107 /* we store the byte offset of byte n into byte n of "endianness" */
1108 for (i = 0; i < sizeof (ino_t); ++i)
1109 ((unsigned char *)&endianness)[i] = i;
1110
1111 *bit++ = 0;
1112
1113 for (i = 0; i < sizeof (ino_t); ++i)
1114 {
1115 /* shifting off the byte offsets out of "endianness" */
1116 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1117 endianness >>= 8;
1118
1119 for (j = 0; j < 8; ++j)
1120 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1121 *bit++ = offs + j;
1122 }
1123
1124 for (j = 0; j < 8; ++j)
1125 if (score_bits & (1 << j))
1126 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1127 }
1128
1129 /* now actually do the sorting (a variant of MSD radix sort) */
1130 {
1131 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1132 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1133 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1134 int stk_idx = 0;
1135
1136 base_stk [stk_idx] = dents;
1137 end_stk [stk_idx] = dents + size;
1138 bit_stk [stk_idx] = bit - 1;
1139
1140 do
1141 {
1142 base = base_stk [stk_idx];
1143 end = end_stk [stk_idx];
1144 bit = bit_stk [stk_idx];
1145
1146 for (;;)
1147 {
1148 unsigned char O = *bit >> 3;
1149 unsigned char M = 1 << (*bit & 7);
1150
1151 eio_dirent *a = base;
1152 eio_dirent *b = end;
1153
1154 if (b - a < EIO_SORT_CUTOFF)
1155 break;
1156
1157 /* now bit-partition the array on the bit */
1158 /* this ugly asymmetric loop seems to perform much better than typical */
1159 /* partition algos found in the literature */
1160 do
1161 if (!(((unsigned char *)a)[O] & M))
1162 ++a;
1163 else if (!(((unsigned char *)--b)[O] & M))
1164 {
1165 eio_dirent tmp = *a; *a = *b; *b = tmp;
1166 ++a;
1167 }
1168 while (b > a);
1169
1170 /* next bit, or stop, if no bits left in this path */
1171 if (!*--bit)
1172 break;
1173
1174 base_stk [stk_idx] = a;
1175 end_stk [stk_idx] = end;
1176 bit_stk [stk_idx] = bit;
1177 ++stk_idx;
1178
1179 end = a;
1180 }
1181 }
1182 while (stk_idx--);
1183 }
1184}
1185
1186static void
1187eio_dent_insertion_sort (eio_dirent *dents, int size)
1188{
1189 /* first move the smallest element to the front, to act as a sentinel */
1190 {
1191 int i;
1192 eio_dirent *min = dents;
1193
1194 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1195 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1196 if (EIO_DENT_CMP (dents [i], <, *min))
1197 min = &dents [i];
1198
1199 /* swap elements 0 and j (minimum) */
1200 {
1201 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1202 }
1203 }
1204
1205 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1206 {
1207 eio_dirent *i, *j;
1208
1209 for (i = dents + 1; i < dents + size; ++i)
1210 {
1211 eio_dirent value = *i;
1212
1213 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1214 j [1] = j [0];
1215
1216 j [1] = value;
1217 }
1218 }
1219}
1220
1221static void
1222eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1223{
1224 if (size <= 1)
1225 return; /* our insertion sort relies on size > 0 */
1226
1227 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1228 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1229 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1230
1231 /* use an insertion sort at the end, or for small arrays, */
1232 /* as insertion sort is more efficient for small partitions */
1233 eio_dent_insertion_sort (dents, size);
1234}
1235
898/* read a full directory */ 1236/* read a full directory */
899static void 1237static void
900eio__scandir (eio_req *req, etp_worker *self) 1238eio__scandir (eio_req *req, etp_worker *self)
901{ 1239{
902 DIR *dirp; 1240 DIR *dirp;
903 EIO_STRUCT_DIRENT *entp; 1241 EIO_STRUCT_DIRENT *entp;
904 char *name, *names; 1242 char *name, *names;
905 int memlen = 4096; 1243 int namesalloc = 4096;
906 int memofs = 0; 1244 int namesoffs = 0;
1245 int flags = req->int1;
1246 eio_dirent *dents = 0;
1247 int dentalloc = 128;
907 int res = 0; 1248 int dentoffs = 0;
1249 ino_t inode_bits = 0;
1250
1251 req->result = -1;
1252
1253 if (!(flags & EIO_READDIR_DENTS))
1254 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
908 1255
909 X_LOCK (wrklock); 1256 X_LOCK (wrklock);
910 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1257 /* the corresponding closedir is in ETP_WORKER_CLEAR */
911 self->dirp = dirp = opendir (req->ptr1); 1258 self->dirp = dirp = opendir (req->ptr1);
1259
912 req->flags |= EIO_FLAG_PTR2_FREE; 1260 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1261 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
913 req->ptr2 = names = malloc (memlen); 1262 req->ptr2 = names = malloc (namesalloc);
914 X_UNLOCK (wrklock); 1263 X_UNLOCK (wrklock);
915 1264
916 if (dirp && names) 1265 if (dirp && names && (!flags || dents))
917 for (;;) 1266 for (;;)
918 { 1267 {
919 errno = 0; 1268 errno = 0;
920 entp = readdir (dirp); 1269 entp = readdir (dirp);
921 1270
922 if (!entp) 1271 if (!entp)
1272 {
1273 if (errno)
1274 break;
1275
1276 /* sort etc. */
1277 req->int1 = flags;
1278 req->result = dentoffs;
1279
1280 if (flags & EIO_READDIR_STAT_ORDER)
1281 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1282 else if (flags & EIO_READDIR_DIRS_FIRST)
1283 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1284 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1285 else
1286 {
1287 /* in this case, all is known, and we just put dirs first and sort them */
1288 eio_dirent *oth = dents + dentoffs;
1289 eio_dirent *dir = dents;
1290
1291 /* now partition dirs to the front, and non-dirs to the back */
1292 /* by walking from both sides and swapping if necessary */
1293 while (oth > dir)
1294 {
1295 if (dir->type == EIO_DT_DIR)
1296 ++dir;
1297 else if ((--oth)->type == EIO_DT_DIR)
1298 {
1299 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1300
1301 ++dir;
1302 }
1303 }
1304
1305 /* now sort the dirs only (dirs all have the same score) */
1306 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1307 }
1308
923 break; 1309 break;
1310 }
924 1311
1312 /* now add the entry to our list(s) */
925 name = entp->d_name; 1313 name = entp->d_name;
926 1314
1315 /* skip . and .. entries */
927 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1316 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
928 { 1317 {
929 int len = strlen (name) + 1; 1318 int len = D_NAMLEN (entp) + 1;
930 1319
931 res++; 1320 while (expect_false (namesoffs + len > namesalloc))
932
933 while (memofs + len > memlen)
934 { 1321 {
935 memlen *= 2; 1322 namesalloc *= 2;
936 X_LOCK (wrklock); 1323 X_LOCK (wrklock);
937 req->ptr2 = names = realloc (names, memlen); 1324 req->ptr2 = names = realloc (names, namesalloc);
938 X_UNLOCK (wrklock); 1325 X_UNLOCK (wrklock);
939 1326
940 if (!names) 1327 if (!names)
941 break; 1328 break;
942 } 1329 }
943 1330
944 memcpy (names + memofs, name, len); 1331 memcpy (names + namesoffs, name, len);
1332
1333 if (dents)
1334 {
1335 struct eio_dirent *ent;
1336
1337 if (expect_false (dentoffs == dentalloc))
1338 {
1339 dentalloc *= 2;
1340 X_LOCK (wrklock);
1341 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1342 X_UNLOCK (wrklock);
1343
1344 if (!dents)
1345 break;
1346 }
1347
1348 ent = dents + dentoffs;
1349
1350 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1351 ent->namelen = len - 1;
1352 ent->inode = D_INO (entp);
1353
1354 inode_bits |= ent->inode;
1355
1356 switch (D_TYPE (entp))
1357 {
1358 default:
1359 ent->type = EIO_DT_UNKNOWN;
1360 flags |= EIO_READDIR_FOUND_UNKNOWN;
1361 break;
1362
1363 #ifdef DT_FIFO
1364 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1365 #endif
1366 #ifdef DT_CHR
1367 case DT_CHR: ent->type = EIO_DT_CHR; break;
1368 #endif
1369 #ifdef DT_MPC
1370 case DT_MPC: ent->type = EIO_DT_MPC; break;
1371 #endif
1372 #ifdef DT_DIR
1373 case DT_DIR: ent->type = EIO_DT_DIR; break;
1374 #endif
1375 #ifdef DT_NAM
1376 case DT_NAM: ent->type = EIO_DT_NAM; break;
1377 #endif
1378 #ifdef DT_BLK
1379 case DT_BLK: ent->type = EIO_DT_BLK; break;
1380 #endif
1381 #ifdef DT_MPB
1382 case DT_MPB: ent->type = EIO_DT_MPB; break;
1383 #endif
1384 #ifdef DT_REG
1385 case DT_REG: ent->type = EIO_DT_REG; break;
1386 #endif
1387 #ifdef DT_NWK
1388 case DT_NWK: ent->type = EIO_DT_NWK; break;
1389 #endif
1390 #ifdef DT_CMP
1391 case DT_CMP: ent->type = EIO_DT_CMP; break;
1392 #endif
1393 #ifdef DT_LNK
1394 case DT_LNK: ent->type = EIO_DT_LNK; break;
1395 #endif
1396 #ifdef DT_SOCK
1397 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1398 #endif
1399 #ifdef DT_DOOR
1400 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1401 #endif
1402 #ifdef DT_WHT
1403 case DT_WHT: ent->type = EIO_DT_WHT; break;
1404 #endif
1405 }
1406
1407 ent->score = 7;
1408
1409 if (flags & EIO_READDIR_DIRS_FIRST)
1410 {
1411 if (ent->type == EIO_DT_UNKNOWN)
1412 {
1413 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1414 ent->score = 1;
1415 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1416 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1417 }
1418 else if (ent->type == EIO_DT_DIR)
1419 ent->score = 0;
1420 }
1421 }
1422
945 memofs += len; 1423 namesoffs += len;
1424 ++dentoffs;
1425 }
1426
1427 if (EIO_CANCELLED (req))
1428 {
1429 errno = ECANCELED;
1430 break;
946 } 1431 }
947 } 1432 }
948
949 if (errno)
950 res = -1;
951
952 req->result = res;
953} 1433}
1434
1435#ifdef PAGESIZE
1436# define eio_pagesize() PAGESIZE
1437#else
1438static intptr_t
1439eio_pagesize (void)
1440{
1441 static intptr_t page;
1442
1443 if (!page)
1444 page = sysconf (_SC_PAGESIZE);
1445
1446 return page;
1447}
1448#endif
1449
1450static void
1451eio_page_align (void **addr, size_t *length)
1452{
1453 intptr_t mask = eio_pagesize () - 1;
1454
1455 /* round down addr */
1456 intptr_t adj = mask & (intptr_t)*addr;
1457
1458 *addr = (void *)((intptr_t)*addr - adj);
1459 *length += adj;
1460
1461 /* round up length */
1462 *length = (*length + mask) & ~mask;
1463}
1464
1465#if !_POSIX_MEMLOCK
1466# define eio__mlockall(a) ((errno = ENOSYS), -1)
1467#else
1468
1469static int
1470eio__mlockall (int flags)
1471{
1472 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1473 extern int mallopt (int, int);
1474 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1475 #endif
1476
1477 if (EIO_MCL_CURRENT != MCL_CURRENT
1478 || EIO_MCL_FUTURE != MCL_FUTURE)
1479 {
1480 flags = 0
1481 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1482 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1483 }
1484
1485 return mlockall (flags);
1486}
1487#endif
1488
1489#if !_POSIX_MEMLOCK_RANGE
1490# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1491#else
1492
1493static int
1494eio__mlock (void *addr, size_t length)
1495{
1496 eio_page_align (&addr, &length);
1497
1498 return mlock (addr, length);
1499}
1500
1501#endif
954 1502
955#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1503#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
956# define msync(a,b,c) ENOSYS 1504# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1505#else
1506
1507int
1508eio__msync (void *mem, size_t len, int flags)
1509{
1510 eio_page_align (&mem, &len);
1511
1512 if (EIO_MS_ASYNC != MS_SYNC
1513 || EIO_MS_INVALIDATE != MS_INVALIDATE
1514 || EIO_MS_SYNC != MS_SYNC)
1515 {
1516 flags = 0
1517 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1518 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1519 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1520 }
1521
1522 return msync (mem, len, flags);
1523}
1524
957#endif 1525#endif
958 1526
959int 1527int
960eio__mtouch (void *mem, size_t len, int flags) 1528eio__mtouch (void *mem, size_t len, int flags)
961{ 1529{
1530 eio_page_align (&mem, &len);
1531
1532 {
962 intptr_t addr = (intptr_t)mem; 1533 intptr_t addr = (intptr_t)mem;
963 intptr_t end = addr + len; 1534 intptr_t end = addr + len;
964#ifdef PAGESIZE 1535 intptr_t page = eio_pagesize ();
965 const intptr_t page = PAGESIZE;
966#else
967 static intptr_t page;
968 1536
969 if (!page)
970 page = sysconf (_SC_PAGESIZE);
971#endif
972
973 addr &= ~(page - 1); /* assume page size is always a power of two */
974
975 if (addr < end) 1537 if (addr < end)
976 if (flags) /* modify */ 1538 if (flags & EIO_MT_MODIFY) /* modify */
977 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1539 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
978 else 1540 else
979 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1541 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1542 }
980 1543
981 return 0; 1544 return 0;
982} 1545}
983 1546
984/*****************************************************************************/ 1547/*****************************************************************************/
… …
1018 if (req) 1581 if (req)
1019 break; 1582 break;
1020 1583
1021 ++idle; 1584 ++idle;
1022 1585
1023 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1586 ts.tv_sec = time (0) + idle_timeout;
1024 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1587 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1025 { 1588 {
1026 if (idle > max_idle) 1589 if (idle > max_idle)
1027 { 1590 {
1028 --idle; 1591 --idle;
… …
1107 return 0; \ 1670 return 0; \
1108 } 1671 }
1109 1672
1110static void eio_execute (etp_worker *self, eio_req *req) 1673static void eio_execute (etp_worker *self, eio_req *req)
1111{ 1674{
1112 errno = 0;
1113
1114 switch (req->type) 1675 switch (req->type)
1115 { 1676 {
1116 case EIO_READ: ALLOC (req->size); 1677 case EIO_READ: ALLOC (req->size);
1117 req->result = req->offs >= 0 1678 req->result = req->offs >= 0
1118 ? pread (req->int1, req->ptr2, req->size, req->offs) 1679 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1129 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1690 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1130 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1691 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1131 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1692 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1132 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1693 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1133 1694
1695 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1696 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1697 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1698 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1699
1134 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1700 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1135 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1701 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1136 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1702 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1137 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1703 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1138 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1704 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
… …
1145 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1711 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1146 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1712 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1147 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1713 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1148 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1714 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1149 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1715 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1150 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1716 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1151 1717
1152 case EIO_READLINK: ALLOC (NAME_MAX); 1718 case EIO_READLINK: ALLOC (PATH_MAX);
1153 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1719 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1154 1720
1155 case EIO_SYNC: req->result = 0; sync (); break; 1721 case EIO_SYNC: req->result = 0; sync (); break;
1156 case EIO_FSYNC: req->result = fsync (req->int1); break; 1722 case EIO_FSYNC: req->result = fsync (req->int1); break;
1157 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1723 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1158 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1724 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1159 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1725 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1726 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1727 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1728 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1160 1729
1161 case EIO_READDIR: eio__scandir (req, self); break; 1730 case EIO_READDIR: eio__scandir (req, self); break;
1162 1731
1163 case EIO_BUSY: 1732 case EIO_BUSY:
1164#ifdef _WIN32 1733#ifdef _WIN32
1165 Sleep (req->nv1 * 1000.); 1734 Sleep (req->nv1 * 1e3);
1166#else 1735#else
1167 { 1736 {
1168 struct timeval tv; 1737 struct timeval tv;
1169 1738
1170 tv.tv_sec = req->nv1; 1739 tv.tv_sec = req->nv1;
1171 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1740 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1172 1741
1173 req->result = select (0, 0, 0, 0, &tv); 1742 req->result = select (0, 0, 0, 0, &tv);
1174 } 1743 }
1175#endif 1744#endif
1176 break; 1745 break;
… …
1191 times = tv; 1760 times = tv;
1192 } 1761 }
1193 else 1762 else
1194 times = 0; 1763 times = 0;
1195 1764
1196
1197 req->result = req->type == EIO_FUTIME 1765 req->result = req->type == EIO_FUTIME
1198 ? futimes (req->int1, times) 1766 ? futimes (req->int1, times)
1199 : utimes (req->ptr1, times); 1767 : utimes (req->ptr1, times);
1200 } 1768 }
1769 break;
1201 1770
1202 case EIO_GROUP: 1771 case EIO_GROUP:
1772 abort (); /* handled in eio_request */
1773
1203 case EIO_NOP: 1774 case EIO_NOP:
1204 req->result = 0; 1775 req->result = 0;
1205 break; 1776 break;
1206 1777
1207 case EIO_CUSTOM: 1778 case EIO_CUSTOM:
1208 ((void (*)(eio_req *))req->feed) (req); 1779 ((void (*)(eio_req *))req->feed) (req);
1209 break; 1780 break;
1210 1781
1211 default: 1782 default:
1783 errno = ENOSYS;
1212 req->result = -1; 1784 req->result = -1;
1213 break; 1785 break;
1214 } 1786 }
1215 1787
1216 req->errorno = errno; 1788 req->errorno = errno;
… …
1246eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 1818eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1247{ 1819{
1248 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 1820 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1249} 1821}
1250 1822
1823eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1824{
1825 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1826}
1827
1828eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1829{
1830 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1831}
1832
1833eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1834{
1835 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1836}
1837
1251eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 1838eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1252{ 1839{
1253 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1840 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1254} 1841}
1255 1842
… …
1274} 1861}
1275 1862
1276eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1863eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1277{ 1864{
1278 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1865 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1866}
1867
1868eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1869{
1870 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1279} 1871}
1280 1872
1281eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1873eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1282{ 1874{
1283 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1875 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
… …
1357eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 1949eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1358{ 1950{
1359 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1951 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1360} 1952}
1361 1953
1954eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1955{
1956 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1957}
1958
1362eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1959eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1363{ 1960{
1364 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1961 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1365} 1962}
1366 1963
1367eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1964eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1368{ 1965{
1369 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1966 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1370} 1967}
1371 1968
1372eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1969eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1373{ 1970{
1374 return eio__1path (EIO_READDIR, path, pri, cb, data); 1971 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1375} 1972}
1376 1973
1377eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1974eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1378{ 1975{
1379 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1976 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1380} 1977}
1381 1978
1382static eio_req * 1979static eio_req *
1383eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1980eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1384{ 1981{
… …
1448 2045
1449void eio_grp_add (eio_req *grp, eio_req *req) 2046void eio_grp_add (eio_req *grp, eio_req *req)
1450{ 2047{
1451 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2048 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1452 2049
2050 grp->flags |= EIO_FLAG_GROUPADD;
2051
1453 ++grp->size; 2052 ++grp->size;
1454 req->grp = grp; 2053 req->grp = grp;
1455 2054
1456 req->grp_prev = 0; 2055 req->grp_prev = 0;
1457 req->grp_next = grp->grp_first; 2056 req->grp_next = grp->grp_first;
… …
1466/* misc garbage */ 2065/* misc garbage */
1467 2066
1468ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2067ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1469{ 2068{
1470 etp_worker wrk; 2069 etp_worker wrk;
2070 ssize_t ret;
1471 2071
1472 wrk.dbuf = 0; 2072 wrk.dbuf = 0;
1473 2073
1474 eio__sendfile (ofd, ifd, offset, count, &wrk); 2074 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1475 2075
1476 if (wrk.dbuf) 2076 if (wrk.dbuf)
1477 free (wrk.dbuf); 2077 free (wrk.dbuf);
1478}
1479 2078
2079 return ret;
2080}
2081

Diff Legend

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