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 1.20 by root, Thu Jun 19 09:05:43 2008 UTC vs.
Revision 1.60 by root, Fri Jan 7 21:45:42 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 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 *
… …
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#include "eio.h" 40#include "eio.h"
41
42#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif
41#include "xthread.h" 45#include "xthread.h"
42 46
43#include <errno.h> 47#include <errno.h>
44#include <stddef.h> 48#include <stddef.h>
45#include <stdlib.h> 49#include <stdlib.h>
46#include <string.h> 50#include <string.h>
47#include <errno.h> 51#include <errno.h>
48#include <sys/types.h> 52#include <sys/types.h>
49#include <sys/stat.h> 53#include <sys/stat.h>
54#include <sys/statvfs.h>
50#include <limits.h> 55#include <limits.h>
51#include <fcntl.h> 56#include <fcntl.h>
52#include <assert.h> 57#include <assert.h>
53 58
54#ifndef EIO_FINISH 59#ifndef EIO_FINISH
… …
64#endif 69#endif
65 70
66#ifdef _WIN32 71#ifdef _WIN32
67 72
68 /*doh*/ 73 /*doh*/
69
70#else 74#else
71 75
72# include "config.h" 76# include "config.h"
73# include <sys/time.h> 77# include <sys/time.h>
74# include <sys/select.h> 78# include <sys/select.h>
75# include <unistd.h> 79# include <unistd.h>
76# include <utime.h> 80# include <utime.h>
77# include <signal.h> 81# include <signal.h>
78# include <dirent.h> 82# include <dirent.h>
79 83
84#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
85# include <sys/mman.h>
86#endif
87
88/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
89# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
90# define _DIRENT_HAVE_D_TYPE /* sigh */
91# define D_INO(de) (de)->d_fileno
92# define D_NAMLEN(de) (de)->d_namlen
93# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
94# define D_INO(de) (de)->d_ino
95# endif
96
97#ifdef _D_EXACT_NAMLEN
98# undef D_NAMLEN
99# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
100#endif
101
102# ifdef _DIRENT_HAVE_D_TYPE
103# define D_TYPE(de) (de)->d_type
104# endif
105
80# ifndef EIO_STRUCT_DIRENT 106# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 107# define EIO_STRUCT_DIRENT struct dirent
82# endif 108# endif
83 109
84#endif 110#endif
85 111
86#if HAVE_SENDFILE 112#if HAVE_SENDFILE
87# if __linux 113# if __linux
88# include <sys/sendfile.h> 114# include <sys/sendfile.h>
89# elif __freebsd 115# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 116# include <sys/socket.h>
91# include <sys/uio.h> 117# include <sys/uio.h>
92# elif __hpux 118# elif __hpux
93# include <sys/socket.h> 119# include <sys/socket.h>
94# elif __solaris /* not yet */ 120# elif __solaris
95# include <sys/sendfile.h> 121# include <sys/sendfile.h>
96# else 122# else
97# error sendfile support requested but not available 123# error sendfile support requested but not available
98# endif 124# endif
99#endif 125#endif
100 126
127#ifndef D_TYPE
128# define D_TYPE(de) 0
129#endif
130#ifndef D_INO
131# define D_INO(de) 0
132#endif
133#ifndef D_NAMLEN
134# define D_NAMLEN(de) strlen ((de)->d_name)
135#endif
136
101/* number of seconds after which an idle threads exit */ 137/* number of seconds after which an idle threads exit */
102#define IDLE_TIMEOUT 10 138#define IDLE_TIMEOUT 10
103 139
104/* used for struct dirent, AIX doesn't provide it */ 140/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 141#ifndef NAME_MAX
106# define NAME_MAX 4096 142# define NAME_MAX 4096
143#endif
144
145/* used for readlink etc. */
146#ifndef PATH_MAX
147# define PATH_MAX 4096
107#endif 148#endif
108 149
109/* buffer size for various temporary buffers */ 150/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 151#define EIO_BUFSIZE 65536
111 152
… …
117 errno = ENOMEM; \ 158 errno = ENOMEM; \
118 if (!eio_buf) \ 159 if (!eio_buf) \
119 return -1; 160 return -1;
120 161
121#define EIO_TICKS ((1000000 + 1023) >> 10) 162#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)
122 174
123/*****************************************************************************/ 175/*****************************************************************************/
124 176
125#define ETP_PRI_MIN EIO_PRI_MIN 177#define ETP_PRI_MIN EIO_PRI_MIN
126#define ETP_PRI_MAX EIO_PRI_MAX 178#define ETP_PRI_MAX EIO_PRI_MAX
… …
144 if (wrk->dirp) \ 196 if (wrk->dirp) \
145 { \ 197 { \
146 closedir (wrk->dirp); \ 198 closedir (wrk->dirp); \
147 wrk->dirp = 0; \ 199 wrk->dirp = 0; \
148 } 200 }
201
149#define ETP_WORKER_COMMON \ 202#define ETP_WORKER_COMMON \
150 void *dbuf; \ 203 void *dbuf; \
151 DIR *dirp; 204 DIR *dirp;
152 205
153/*****************************************************************************/ 206/*****************************************************************************/
154 207
155#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
156 209
157/* calculcate time difference in ~1/EIO_TICKS of a second */ 210/* calculate time difference in ~1/EIO_TICKS of a second */
158static int tvdiff (struct timeval *tv1, struct timeval *tv2) 211static int tvdiff (struct timeval *tv1, struct timeval *tv2)
159{ 212{
160 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 213 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
161 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 214 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
162} 215}
… …
172static volatile unsigned int nreqs; /* reqlock */ 225static volatile unsigned int nreqs; /* reqlock */
173static volatile unsigned int nready; /* reqlock */ 226static volatile unsigned int nready; /* reqlock */
174static volatile unsigned int npending; /* reqlock */ 227static volatile unsigned int npending; /* reqlock */
175static volatile unsigned int max_idle = 4; 228static volatile unsigned int max_idle = 4;
176 229
177static mutex_t wrklock = X_MUTEX_INIT; 230static xmutex_t wrklock = X_MUTEX_INIT;
178static mutex_t reslock = X_MUTEX_INIT; 231static xmutex_t reslock = X_MUTEX_INIT;
179static mutex_t reqlock = X_MUTEX_INIT; 232static xmutex_t reqlock = X_MUTEX_INIT;
180static cond_t reqwait = X_COND_INIT; 233static xcond_t reqwait = X_COND_INIT;
234
235#if !HAVE_PREADWRITE
236/*
237 * make our pread/pwrite emulation safe against themselves, but not against
238 * normal read/write by using a mutex. slows down execution a lot,
239 * but that's your problem, not mine.
240 */
241static xmutex_t preadwritelock = X_MUTEX_INIT;
242#endif
181 243
182typedef struct etp_worker 244typedef struct etp_worker
183{ 245{
184 /* locked by wrklock */ 246 /* locked by wrklock */
185 struct etp_worker *prev, *next; 247 struct etp_worker *prev, *next;
186 248
187 thread_t tid; 249 xthread_t tid;
188 250
189 /* locked by reslock, reqlock or wrklock */ 251 /* locked by reslock, reqlock or wrklock */
190 ETP_REQ *req; /* currently processed request */ 252 ETP_REQ *req; /* currently processed request */
191 253
192 ETP_WORKER_COMMON 254 ETP_WORKER_COMMON
… …
330 392
331static void etp_atfork_child (void) 393static void etp_atfork_child (void)
332{ 394{
333 ETP_REQ *prv; 395 ETP_REQ *prv;
334 396
335 while (prv = reqq_shift (&req_queue)) 397 while ((prv = reqq_shift (&req_queue)))
336 ETP_DESTROY (prv); 398 ETP_DESTROY (prv);
337 399
338 while (prv = reqq_shift (&res_queue)) 400 while ((prv = reqq_shift (&res_queue)))
339 ETP_DESTROY (prv); 401 ETP_DESTROY (prv);
340 402
341 while (wrk_first.next != &wrk_first) 403 while (wrk_first.next != &wrk_first)
342 { 404 {
343 etp_worker *wrk = wrk_first.next; 405 etp_worker *wrk = wrk_first.next;
… …
402 X_UNLOCK (wrklock); 464 X_UNLOCK (wrklock);
403} 465}
404 466
405static void etp_maybe_start_thread (void) 467static void etp_maybe_start_thread (void)
406{ 468{
407 if (etp_nthreads () >= wanted) 469 if (expect_true (etp_nthreads () >= wanted))
408 return; 470 return;
409 471
410 /* todo: maybe use idle here, but might be less exact */ 472 /* todo: maybe use idle here, but might be less exact */
411 if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) 473 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
412 return; 474 return;
413 475
414 etp_start_thread (); 476 etp_start_thread ();
415} 477}
416 478
… …
469 531
470 X_LOCK (reqlock); 532 X_LOCK (reqlock);
471 --nreqs; 533 --nreqs;
472 X_UNLOCK (reqlock); 534 X_UNLOCK (reqlock);
473 535
474 if (req->type == EIO_GROUP && req->size) 536 if (expect_false (req->type == EIO_GROUP && req->size))
475 { 537 {
476 req->int1 = 1; /* mark request as delayed */ 538 req->int1 = 1; /* mark request as delayed */
477 continue; 539 continue;
478 } 540 }
479 else 541 else
480 { 542 {
481 int res = ETP_FINISH (req); 543 int res = ETP_FINISH (req);
482 if (res) 544 if (expect_false (res))
483 return res; 545 return res;
484 } 546 }
485 547
486 if (maxreqs && !--maxreqs) 548 if (expect_false (maxreqs && !--maxreqs))
487 break; 549 break;
488 550
489 if (maxtime) 551 if (maxtime)
490 { 552 {
491 gettimeofday (&tv_now, 0); 553 gettimeofday (&tv_now, 0);
… …
510 572
511static void etp_submit (ETP_REQ *req) 573static void etp_submit (ETP_REQ *req)
512{ 574{
513 req->pri -= ETP_PRI_MIN; 575 req->pri -= ETP_PRI_MIN;
514 576
515 if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 577 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
516 if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 578 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
517 579
580 if (expect_false (req->type == EIO_GROUP))
581 {
582 /* I hope this is worth it :/ */
518 X_LOCK (reqlock); 583 X_LOCK (reqlock);
519 ++nreqs; 584 ++nreqs;
585 X_UNLOCK (reqlock);
586
587 X_LOCK (reslock);
588
589 ++npending;
590
591 if (!reqq_push (&res_queue, req) && want_poll_cb)
592 want_poll_cb ();
593
594 X_UNLOCK (reslock);
595 }
596 else
597 {
598 X_LOCK (reqlock);
599 ++nreqs;
520 ++nready; 600 ++nready;
521 reqq_push (&req_queue, req); 601 reqq_push (&req_queue, req);
522 X_COND_SIGNAL (reqwait); 602 X_COND_SIGNAL (reqwait);
523 X_UNLOCK (reqlock); 603 X_UNLOCK (reqlock);
524 604
525 etp_maybe_start_thread (); 605 etp_maybe_start_thread ();
606 }
526} 607}
527 608
528static void etp_set_max_poll_time (double nseconds) 609static void etp_set_max_poll_time (double nseconds)
529{ 610{
530 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 611 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
531 max_poll_time = nseconds; 612 max_poll_time = nseconds * EIO_TICKS;
532 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 613 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
533} 614}
534 615
535static void etp_set_max_poll_reqs (unsigned int maxreqs) 616static void etp_set_max_poll_reqs (unsigned int maxreqs)
536{ 617{
… …
565 646
566static void grp_try_feed (eio_req *grp) 647static void grp_try_feed (eio_req *grp)
567{ 648{
568 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 649 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
569 { 650 {
570 int old_len = grp->size; 651 grp->flags &= ~EIO_FLAG_GROUPADD;
571 652
572 EIO_FEED (grp); 653 EIO_FEED (grp);
573 654
574 /* stop if no progress has been made */ 655 /* stop if no progress has been made */
575 if (old_len == grp->size) 656 if (!(grp->flags & EIO_FLAG_GROUPADD))
576 { 657 {
577 grp->feed = 0; 658 grp->feed = 0;
578 break; 659 break;
579 } 660 }
580 } 661 }
… …
697 778
698/*****************************************************************************/ 779/*****************************************************************************/
699/* work around various missing functions */ 780/* work around various missing functions */
700 781
701#if !HAVE_PREADWRITE 782#if !HAVE_PREADWRITE
783# undef pread
784# undef pwrite
702# define pread eio__pread 785# define pread eio__pread
703# define pwrite eio__pwrite 786# define pwrite eio__pwrite
704
705/*
706 * make our pread/pwrite safe against themselves, but not against
707 * normal read/write by using a mutex. slows down execution a lot,
708 * but that's your problem, not mine.
709 */
710static mutex_t preadwritelock = X_MUTEX_INIT;
711 787
712static ssize_t 788static ssize_t
713eio__pread (int fd, void *buf, size_t count, off_t offset) 789eio__pread (int fd, void *buf, size_t count, off_t offset)
714{ 790{
715 ssize_t res; 791 ssize_t res;
… …
733 809
734 X_LOCK (preadwritelock); 810 X_LOCK (preadwritelock);
735 ooffset = lseek (fd, 0, SEEK_CUR); 811 ooffset = lseek (fd, 0, SEEK_CUR);
736 lseek (fd, offset, SEEK_SET); 812 lseek (fd, offset, SEEK_SET);
737 res = write (fd, buf, count); 813 res = write (fd, buf, count);
738 lseek (fd, offset, SEEK_SET); 814 lseek (fd, ooffset, SEEK_SET);
739 X_UNLOCK (preadwritelock); 815 X_UNLOCK (preadwritelock);
740 816
741 return res; 817 return res;
742} 818}
743#endif 819#endif
744 820
745#ifndef HAVE_FUTIMES 821#ifndef HAVE_UTIMES
746 822
823# undef utimes
747# define utimes(path,times) eio__utimes (path, times) 824# define utimes(path,times) eio__utimes (path, times)
748# define futimes(fd,times) eio__futimes (fd, times)
749 825
750static int 826static int
751eio__utimes (const char *filename, const struct timeval times[2]) 827eio__utimes (const char *filename, const struct timeval times[2])
752{ 828{
753 if (times) 829 if (times)
… …
761 } 837 }
762 else 838 else
763 return utime (filename, 0); 839 return utime (filename, 0);
764} 840}
765 841
842#endif
843
844#ifndef HAVE_FUTIMES
845
846# undef futimes
847# define futimes(fd,times) eio__futimes (fd, times)
848
766static int eio__futimes (int fd, const struct timeval tv[2]) 849static int eio__futimes (int fd, const struct timeval tv[2])
767{ 850{
768 errno = ENOSYS; 851 errno = ENOSYS;
769 return -1; 852 return -1;
770} 853}
771 854
772#endif 855#endif
773 856
774#if !HAVE_FDATASYNC 857#if !HAVE_FDATASYNC
858# undef fdatasync
775# define fdatasync fsync 859# define fdatasync(fd) fsync (fd)
776#endif 860#endif
861
862/* sync_file_range always needs emulation */
863int
864eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
865{
866#if HAVE_SYNC_FILE_RANGE
867 int res;
868
869 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
870 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
871 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
872 {
873 flags = 0
874 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
875 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
876 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
877 }
878
879 res = sync_file_range (fd, offset, nbytes, flags);
880
881 if (!res || errno != ENOSYS)
882 return res;
883#endif
884
885 /* even though we could play tricks with the flags, it's better to always
886 * call fdatasync, as that matches the expectation of its users best */
887 return fdatasync (fd);
888}
777 889
778#if !HAVE_READAHEAD 890#if !HAVE_READAHEAD
891# undef readahead
779# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 892# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
780 893
781static ssize_t 894static ssize_t
782eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 895eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
783{ 896{
… …
810 923
811#if HAVE_SENDFILE 924#if HAVE_SENDFILE
812# if __linux 925# if __linux
813 res = sendfile (ofd, ifd, &offset, count); 926 res = sendfile (ofd, ifd, &offset, count);
814 927
815# elif __freebsd 928# elif __FreeBSD__
816 /* 929 /*
817 * Of course, the freebsd sendfile is a dire hack with no thoughts 930 * Of course, the freebsd sendfile is a dire hack with no thoughts
818 * wasted on making it similar to other I/O functions. 931 * wasted on making it similar to other I/O functions.
819 */ 932 */
820 { 933 {
821 off_t sbytes; 934 off_t sbytes;
822 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 935 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
823 936
824 if (res < 0 && sbytes) 937 #if 0 /* according to the manpage, this is correct, but broken behaviour */
825 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 938 /* freebsd' sendfile will return 0 on success */
939 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
940 /* not on e.g. EIO or EPIPE - sounds broken */
941 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
942 res = sbytes;
943 #endif
944
945 /* according to source inspection, this is correct, and useful behaviour */
946 if (sbytes)
947 res = sbytes;
948 }
949
950# elif defined (__APPLE__)
951
952 {
953 off_t sbytes = count;
954 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
955
956 /* according to the manpage, sbytes is always valid */
957 if (sbytes)
826 res = sbytes; 958 res = sbytes;
827 } 959 }
828 960
829# elif __hpux 961# elif __hpux
830 res = sendfile (ofd, ifd, offset, count, 0, 0); 962 res = sendfile (ofd, ifd, offset, count, 0, 0);
… …
844 if (res < 0 && sbytes) 976 if (res < 0 && sbytes)
845 res = sbytes; 977 res = sbytes;
846 } 978 }
847 979
848# endif 980# endif
981
982#elif defined (_WIN32)
983
984 /* does not work, just for documentation of what would need to be done */
985 {
986 HANDLE h = TO_SOCKET (ifd);
987 SetFilePointer (h, offset, 0, FILE_BEGIN);
988 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
989 }
990
849#else 991#else
850 res = -1; 992 res = -1;
851 errno = ENOSYS; 993 errno = ENOSYS;
852#endif 994#endif
853 995
854 if (res < 0 996 if (res < 0
855 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 997 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
998 /* BSDs */
999#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1000 || errno == ENOTSUP
1001#endif
1002 || errno == EOPNOTSUPP /* BSDs */
856#if __solaris 1003#if __solaris
857 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1004 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
858#endif 1005#endif
859 ) 1006 )
860 ) 1007 )
… …
891 } 1038 }
892 1039
893 return res; 1040 return res;
894} 1041}
895 1042
1043static signed char
1044eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1045{
1046 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;
1048}
1049
1050#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1051
1052#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1053#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1054
1055static void
1056eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1057{
1058 unsigned char bits [9 + sizeof (ino_t) * 8];
1059 unsigned char *bit = bits;
1060
1061 assert (CHAR_BIT == 8);
1062 assert (sizeof (eio_dirent) * 8 < 256);
1063 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1064 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1065
1066 if (size <= EIO_SORT_FAST)
1067 return;
1068
1069 /* first prepare an array of bits to test in our radix sort */
1070 /* try to take endianness into account, as well as differences in ino_t sizes */
1071 /* inode_bits must contain all inodes ORed together */
1072 /* which is used to skip bits that are 0 everywhere, which is very common */
1073 {
1074 ino_t endianness;
1075 int i, j;
1076
1077 /* we store the byte offset of byte n into byte n of "endianness" */
1078 for (i = 0; i < sizeof (ino_t); ++i)
1079 ((unsigned char *)&endianness)[i] = i;
1080
1081 *bit++ = 0;
1082
1083 for (i = 0; i < sizeof (ino_t); ++i)
1084 {
1085 /* shifting off the byte offsets out of "endianness" */
1086 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1087 endianness >>= 8;
1088
1089 for (j = 0; j < 8; ++j)
1090 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1091 *bit++ = offs + j;
1092 }
1093
1094 for (j = 0; j < 8; ++j)
1095 if (score_bits & (1 << j))
1096 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1097 }
1098
1099 /* now actually do the sorting (a variant of MSD radix sort) */
1100 {
1101 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1102 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1103 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1104 int stk_idx = 0;
1105
1106 base_stk [stk_idx] = dents;
1107 end_stk [stk_idx] = dents + size;
1108 bit_stk [stk_idx] = bit - 1;
1109
1110 do
1111 {
1112 base = base_stk [stk_idx];
1113 end = end_stk [stk_idx];
1114 bit = bit_stk [stk_idx];
1115
1116 for (;;)
1117 {
1118 unsigned char O = *bit >> 3;
1119 unsigned char M = 1 << (*bit & 7);
1120
1121 eio_dirent *a = base;
1122 eio_dirent *b = end;
1123
1124 if (b - a < EIO_SORT_CUTOFF)
1125 break;
1126
1127 /* now bit-partition the array on the bit */
1128 /* this ugly asymmetric loop seems to perform much better than typical */
1129 /* partition algos found in the literature */
1130 do
1131 if (!(((unsigned char *)a)[O] & M))
1132 ++a;
1133 else if (!(((unsigned char *)--b)[O] & M))
1134 {
1135 eio_dirent tmp = *a; *a = *b; *b = tmp;
1136 ++a;
1137 }
1138 while (b > a);
1139
1140 /* next bit, or stop, if no bits left in this path */
1141 if (!*--bit)
1142 break;
1143
1144 base_stk [stk_idx] = a;
1145 end_stk [stk_idx] = end;
1146 bit_stk [stk_idx] = bit;
1147 ++stk_idx;
1148
1149 end = a;
1150 }
1151 }
1152 while (stk_idx--);
1153 }
1154}
1155
1156static void
1157eio_dent_insertion_sort (eio_dirent *dents, int size)
1158{
1159 /* first move the smallest element to the front, to act as a sentinel */
1160 {
1161 int i;
1162 eio_dirent *min = dents;
1163
1164 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1165 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1166 if (EIO_DENT_CMP (dents [i], <, *min))
1167 min = &dents [i];
1168
1169 /* swap elements 0 and j (minimum) */
1170 {
1171 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1172 }
1173 }
1174
1175 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1176 {
1177 eio_dirent *i, *j;
1178
1179 for (i = dents + 1; i < dents + size; ++i)
1180 {
1181 eio_dirent value = *i;
1182
1183 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1184 j [1] = j [0];
1185
1186 j [1] = value;
1187 }
1188 }
1189}
1190
1191static void
1192eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1193{
1194 if (size <= 1)
1195 return; /* our insertion sort relies on size > 0 */
1196
1197 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1198 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1199 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1200
1201 /* use an insertion sort at the end, or for small arrays, */
1202 /* as insertion sort is more efficient for small partitions */
1203 eio_dent_insertion_sort (dents, size);
1204}
1205
896/* read a full directory */ 1206/* read a full directory */
897static void 1207static void
898eio__scandir (eio_req *req, etp_worker *self) 1208eio__scandir (eio_req *req, etp_worker *self)
899{ 1209{
900 DIR *dirp; 1210 DIR *dirp;
901 EIO_STRUCT_DIRENT *entp; 1211 EIO_STRUCT_DIRENT *entp;
902 char *name, *names; 1212 char *name, *names;
903 int memlen = 4096; 1213 int namesalloc = 4096;
904 int memofs = 0; 1214 int namesoffs = 0;
1215 int flags = req->int1;
1216 eio_dirent *dents = 0;
1217 int dentalloc = 128;
905 int res = 0; 1218 int dentoffs = 0;
1219 ino_t inode_bits = 0;
1220
1221 req->result = -1;
1222
1223 if (!(flags & EIO_READDIR_DENTS))
1224 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
906 1225
907 X_LOCK (wrklock); 1226 X_LOCK (wrklock);
908 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1227 /* the corresponding closedir is in ETP_WORKER_CLEAR */
909 self->dirp = dirp = opendir (req->ptr1); 1228 self->dirp = dirp = opendir (req->ptr1);
910 req->flags |= EIO_FLAG_PTR2_FREE; 1229 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1230 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
911 req->ptr2 = names = malloc (memlen); 1231 req->ptr2 = names = malloc (namesalloc);
912 X_UNLOCK (wrklock); 1232 X_UNLOCK (wrklock);
913 1233
914 if (dirp && names) 1234 if (dirp && names && (!flags || dents))
915 for (;;) 1235 for (;;)
916 { 1236 {
917 errno = 0; 1237 errno = 0;
918 entp = readdir (dirp); 1238 entp = readdir (dirp);
919 1239
920 if (!entp) 1240 if (!entp)
1241 {
1242 if (errno)
1243 break;
1244
1245 /* sort etc. */
1246 req->int1 = flags;
1247 req->result = dentoffs;
1248
1249 if (flags & EIO_READDIR_STAT_ORDER)
1250 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
1251 else if (flags & EIO_READDIR_DIRS_FIRST)
1252 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1253 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1254 else
1255 {
1256 /* in this case, all is known, and we just put dirs first and sort them */
1257 eio_dirent *oth = dents + dentoffs;
1258 eio_dirent *dir = dents;
1259
1260 /* now partition dirs to the front, and non-dirs to the back */
1261 /* by walking from both sides and swapping if necessary */
1262 /* also clear score, so it doesn't influence sorting */
1263 while (oth > dir)
1264 {
1265 if (dir->type == EIO_DT_DIR)
1266 ++dir;
1267 else if ((--oth)->type == EIO_DT_DIR)
1268 {
1269 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1270
1271 ++dir;
1272 }
1273 }
1274
1275 /* now sort the dirs only */
1276 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1277 }
1278
921 break; 1279 break;
1280 }
922 1281
1282 /* now add the entry to our list(s) */
923 name = entp->d_name; 1283 name = entp->d_name;
924 1284
1285 /* skip . and .. entries */
925 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1286 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
926 { 1287 {
927 int len = strlen (name) + 1; 1288 int len = D_NAMLEN (entp) + 1;
928 1289
929 res++; 1290 while (expect_false (namesoffs + len > namesalloc))
930
931 while (memofs + len > memlen)
932 { 1291 {
933 memlen *= 2; 1292 namesalloc *= 2;
934 X_LOCK (wrklock); 1293 X_LOCK (wrklock);
935 req->ptr2 = names = realloc (names, memlen); 1294 req->ptr2 = names = realloc (names, namesalloc);
936 X_UNLOCK (wrklock); 1295 X_UNLOCK (wrklock);
937 1296
938 if (!names) 1297 if (!names)
939 break; 1298 break;
940 } 1299 }
941 1300
942 memcpy (names + memofs, name, len); 1301 memcpy (names + namesoffs, name, len);
1302
1303 if (dents)
1304 {
1305 struct eio_dirent *ent;
1306
1307 if (expect_false (dentoffs == dentalloc))
1308 {
1309 dentalloc *= 2;
1310 X_LOCK (wrklock);
1311 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1312 X_UNLOCK (wrklock);
1313
1314 if (!dents)
1315 break;
1316 }
1317
1318 ent = dents + dentoffs;
1319
1320 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1321 ent->namelen = len - 1;
1322 ent->inode = D_INO (entp);
1323
1324 inode_bits |= ent->inode;
1325
1326 switch (D_TYPE (entp))
1327 {
1328 default:
1329 ent->type = EIO_DT_UNKNOWN;
1330 flags |= EIO_READDIR_FOUND_UNKNOWN;
1331 break;
1332
1333 #ifdef DT_FIFO
1334 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1335 #endif
1336 #ifdef DT_CHR
1337 case DT_CHR: ent->type = EIO_DT_CHR; break;
1338 #endif
1339 #ifdef DT_MPC
1340 case DT_MPC: ent->type = EIO_DT_MPC; break;
1341 #endif
1342 #ifdef DT_DIR
1343 case DT_DIR: ent->type = EIO_DT_DIR; break;
1344 #endif
1345 #ifdef DT_NAM
1346 case DT_NAM: ent->type = EIO_DT_NAM; break;
1347 #endif
1348 #ifdef DT_BLK
1349 case DT_BLK: ent->type = EIO_DT_BLK; break;
1350 #endif
1351 #ifdef DT_MPB
1352 case DT_MPB: ent->type = EIO_DT_MPB; break;
1353 #endif
1354 #ifdef DT_REG
1355 case DT_REG: ent->type = EIO_DT_REG; break;
1356 #endif
1357 #ifdef DT_NWK
1358 case DT_NWK: ent->type = EIO_DT_NWK; break;
1359 #endif
1360 #ifdef DT_CMP
1361 case DT_CMP: ent->type = EIO_DT_CMP; break;
1362 #endif
1363 #ifdef DT_LNK
1364 case DT_LNK: ent->type = EIO_DT_LNK; break;
1365 #endif
1366 #ifdef DT_SOCK
1367 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1368 #endif
1369 #ifdef DT_DOOR
1370 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1371 #endif
1372 #ifdef DT_WHT
1373 case DT_WHT: ent->type = EIO_DT_WHT; break;
1374 #endif
1375 }
1376
1377 ent->score = 7;
1378
1379 if (flags & EIO_READDIR_DIRS_FIRST)
1380 {
1381 if (ent->type == EIO_DT_UNKNOWN)
1382 {
1383 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1384 ent->score = 1;
1385 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1386 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1387 }
1388 else if (ent->type == EIO_DT_DIR)
1389 ent->score = 0;
1390 }
1391 }
1392
943 memofs += len; 1393 namesoffs += len;
1394 ++dentoffs;
1395 }
1396
1397 if (EIO_CANCELLED (req))
1398 {
1399 errno = ECANCELED;
1400 break;
944 } 1401 }
945 } 1402 }
1403}
946 1404
947 if (errno) 1405#ifdef PAGESIZE
948 res = -1; 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
949 1502 {
950 req->result = res; 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;
951} 1515}
952 1516
953/*****************************************************************************/ 1517/*****************************************************************************/
954 1518
955#define ALLOC(len) \ 1519#define ALLOC(len) \
… …
1076 return 0; \ 1640 return 0; \
1077 } 1641 }
1078 1642
1079static void eio_execute (etp_worker *self, eio_req *req) 1643static void eio_execute (etp_worker *self, eio_req *req)
1080{ 1644{
1081 errno = 0;
1082
1083 switch (req->type) 1645 switch (req->type)
1084 { 1646 {
1085 case EIO_READ: ALLOC (req->size); 1647 case EIO_READ: ALLOC (req->size);
1086 req->result = req->offs >= 0 1648 req->result = req->offs >= 0
1087 ? pread (req->int1, req->ptr2, req->size, req->offs) 1649 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1097 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1659 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1098 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1660 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1099 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1661 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1100 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1662 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1101 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1663 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1664
1665 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1666 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1667 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1668 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1102 1669
1103 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1670 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1104 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1671 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1105 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1672 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1106 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1673 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
… …
1116 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1683 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1117 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1684 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1118 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1685 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1119 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1686 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1120 1687
1121 case EIO_READLINK: ALLOC (NAME_MAX); 1688 case EIO_READLINK: ALLOC (PATH_MAX);
1122 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1689 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1123 1690
1124 case EIO_SYNC: req->result = 0; sync (); break; 1691 case EIO_SYNC: req->result = 0; sync (); break;
1125 case EIO_FSYNC: req->result = fsync (req->int1); break; 1692 case EIO_FSYNC: req->result = fsync (req->int1); break;
1126 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1693 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1694 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;
1696 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1697 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;
1127 1699
1128 case EIO_READDIR: eio__scandir (req, self); break; 1700 case EIO_READDIR: eio__scandir (req, self); break;
1129 1701
1130 case EIO_BUSY: 1702 case EIO_BUSY:
1131#ifdef _WIN32 1703#ifdef _WIN32
1132 Sleep (req->nv1 * 1000.); 1704 Sleep (req->nv1 * 1e3);
1133#else 1705#else
1134 { 1706 {
1135 struct timeval tv; 1707 struct timeval tv;
1136 1708
1137 tv.tv_sec = req->nv1; 1709 tv.tv_sec = req->nv1;
1138 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1710 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1139 1711
1140 req->result = select (0, 0, 0, 0, &tv); 1712 req->result = select (0, 0, 0, 0, &tv);
1141 } 1713 }
1142#endif 1714#endif
1143 break; 1715 break;
… …
1158 times = tv; 1730 times = tv;
1159 } 1731 }
1160 else 1732 else
1161 times = 0; 1733 times = 0;
1162 1734
1163
1164 req->result = req->type == EIO_FUTIME 1735 req->result = req->type == EIO_FUTIME
1165 ? futimes (req->int1, times) 1736 ? futimes (req->int1, times)
1166 : utimes (req->ptr1, times); 1737 : utimes (req->ptr1, times);
1167 } 1738 }
1739 break;
1168 1740
1169 case EIO_GROUP: 1741 case EIO_GROUP:
1742 abort (); /* handled in eio_request */
1743
1170 case EIO_NOP: 1744 case EIO_NOP:
1171 req->result = 0; 1745 req->result = 0;
1172 break; 1746 break;
1173 1747
1174 case EIO_CUSTOM: 1748 case EIO_CUSTOM:
1175 req->feed (req); 1749 ((void (*)(eio_req *))req->feed) (req);
1176 break; 1750 break;
1177 1751
1178 default: 1752 default:
1753 errno = ENOSYS;
1179 req->result = -1; 1754 req->result = -1;
1180 break; 1755 break;
1181 } 1756 }
1182 1757
1183 req->errorno = errno; 1758 req->errorno = errno;
… …
1203eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 1778eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1204{ 1779{
1205 REQ (EIO_FSYNC); req->int1 = fd; SEND; 1780 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1206} 1781}
1207 1782
1783eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1784{
1785 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1786}
1787
1788eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1789{
1790 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1791}
1792
1793eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1794{
1795 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1796}
1797
1798eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1799{
1800 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1801}
1802
1803eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1804{
1805 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1806}
1807
1208eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 1808eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1209{ 1809{
1210 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1810 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1211} 1811}
1212 1812
… …
1231} 1831}
1232 1832
1233eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1833eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1234{ 1834{
1235 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1835 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1836}
1837
1838eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1839{
1840 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1236} 1841}
1237 1842
1238eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1843eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1239{ 1844{
1240 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1845 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
… …
1314eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 1919eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1315{ 1920{
1316 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1921 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1317} 1922}
1318 1923
1924eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1925{
1926 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1927}
1928
1319eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1929eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1320{ 1930{
1321 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1931 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1322} 1932}
1323 1933
1324eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1934eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1325{ 1935{
1326 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1936 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1327} 1937}
1328 1938
1329eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1939eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1330{ 1940{
1331 return eio__1path (EIO_READDIR, path, pri, cb, data); 1941 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1332} 1942}
1333 1943
1334eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1944eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1335{ 1945{
1336 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1946 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
… …
1367 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 1977 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1368} 1978}
1369 1979
1370eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 1980eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1371{ 1981{
1372 REQ (EIO_CUSTOM); req->feed = execute; SEND; 1982 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1373} 1983}
1374 1984
1375#endif 1985#endif
1376 1986
1377eio_req *eio_grp (eio_cb cb, void *data) 1987eio_req *eio_grp (eio_cb cb, void *data)
… …
1405 2015
1406void eio_grp_add (eio_req *grp, eio_req *req) 2016void eio_grp_add (eio_req *grp, eio_req *req)
1407{ 2017{
1408 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2018 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1409 2019
2020 grp->flags |= EIO_FLAG_GROUPADD;
2021
1410 ++grp->size; 2022 ++grp->size;
1411 req->grp = grp; 2023 req->grp = grp;
1412 2024
1413 req->grp_prev = 0; 2025 req->grp_prev = 0;
1414 req->grp_next = grp->grp_first; 2026 req->grp_next = grp->grp_first;
… …
1423/* misc garbage */ 2035/* misc garbage */
1424 2036
1425ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2037ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1426{ 2038{
1427 etp_worker wrk; 2039 etp_worker wrk;
2040 ssize_t ret;
1428 2041
1429 wrk.dbuf = 0; 2042 wrk.dbuf = 0;
1430 2043
1431 eio__sendfile (ofd, ifd, offset, count, &wrk); 2044 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1432 2045
1433 if (wrk.dbuf) 2046 if (wrk.dbuf)
1434 free (wrk.dbuf); 2047 free (wrk.dbuf);
1435}
1436 2048
2049 return ret;
2050}
2051

Diff Legend

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