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 1.17 by root, Tue Jun 3 05:12:51 2008 UTC vs.
Revision 1.52 by root, Fri Jan 8 09:29:56 2010 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>
79# include <sys/mman.h>
75# include <unistd.h> 80# include <unistd.h>
76# include <utime.h> 81# include <utime.h>
77# include <signal.h> 82# include <signal.h>
78# include <dirent.h> 83# include <dirent.h>
79 84
85/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
86# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
87# define _DIRENT_HAVE_D_TYPE /* sigh */
88# define D_INO(de) (de)->d_fileno
89# define D_NAMLEN(de) (de)->d_namlen
90# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
91# define D_INO(de) (de)->d_ino
92# endif
93
94#ifdef _D_EXACT_NAMLEN
95# undef D_NAMLEN
96# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
97#endif
98
99# ifdef _DIRENT_HAVE_D_TYPE
100# define D_TYPE(de) (de)->d_type
101# endif
102
80# ifndef EIO_STRUCT_DIRENT 103# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 104# define EIO_STRUCT_DIRENT struct dirent
82# endif 105# endif
83 106
84#endif 107#endif
85 108
86#if HAVE_SENDFILE 109#if HAVE_SENDFILE
87# if __linux 110# if __linux
88# include <sys/sendfile.h> 111# include <sys/sendfile.h>
89# elif __freebsd 112# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 113# include <sys/socket.h>
91# include <sys/uio.h> 114# include <sys/uio.h>
92# elif __hpux 115# elif __hpux
93# include <sys/socket.h> 116# include <sys/socket.h>
94# elif __solaris /* not yet */ 117# elif __solaris
95# include <sys/sendfile.h> 118# include <sys/sendfile.h>
96# else 119# else
97# error sendfile support requested but not available 120# error sendfile support requested but not available
98# endif 121# endif
122#endif
123
124#ifndef D_TYPE
125# define D_TYPE(de) 0
126#endif
127#ifndef D_INO
128# define D_INO(de) 0
129#endif
130#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name)
99#endif 132#endif
100 133
101/* number of seconds after which an idle threads exit */ 134/* number of seconds after which an idle threads exit */
102#define IDLE_TIMEOUT 10 135#define IDLE_TIMEOUT 10
103 136
117 errno = ENOMEM; \ 150 errno = ENOMEM; \
118 if (!eio_buf) \ 151 if (!eio_buf) \
119 return -1; 152 return -1;
120 153
121#define EIO_TICKS ((1000000 + 1023) >> 10) 154#define EIO_TICKS ((1000000 + 1023) >> 10)
155
156/*****************************************************************************/
157
158#if __GNUC__ >= 3
159# define expect(expr,value) __builtin_expect ((expr),(value))
160#else
161# define expect(expr,value) (expr)
162#endif
163
164#define expect_false(expr) expect ((expr) != 0, 0)
165#define expect_true(expr) expect ((expr) != 0, 1)
122 166
123/*****************************************************************************/ 167/*****************************************************************************/
124 168
125#define ETP_PRI_MIN EIO_PRI_MIN 169#define ETP_PRI_MIN EIO_PRI_MIN
126#define ETP_PRI_MAX EIO_PRI_MAX 170#define ETP_PRI_MAX EIO_PRI_MAX
144 if (wrk->dirp) \ 188 if (wrk->dirp) \
145 { \ 189 { \
146 closedir (wrk->dirp); \ 190 closedir (wrk->dirp); \
147 wrk->dirp = 0; \ 191 wrk->dirp = 0; \
148 } 192 }
193
149#define ETP_WORKER_COMMON \ 194#define ETP_WORKER_COMMON \
150 void *dbuf; \ 195 void *dbuf; \
151 DIR *dirp; 196 DIR *dirp;
152 197
153/*****************************************************************************/ 198/*****************************************************************************/
154 199
155#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 200#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
156 201
157/* calculcate time difference in ~1/EIO_TICKS of a second */ 202/* calculate time difference in ~1/EIO_TICKS of a second */
158static int tvdiff (struct timeval *tv1, struct timeval *tv2) 203static int tvdiff (struct timeval *tv1, struct timeval *tv2)
159{ 204{
160 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 205 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
161 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 206 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
162} 207}
176 221
177static mutex_t wrklock = X_MUTEX_INIT; 222static mutex_t wrklock = X_MUTEX_INIT;
178static mutex_t reslock = X_MUTEX_INIT; 223static mutex_t reslock = X_MUTEX_INIT;
179static mutex_t reqlock = X_MUTEX_INIT; 224static mutex_t reqlock = X_MUTEX_INIT;
180static cond_t reqwait = X_COND_INIT; 225static cond_t reqwait = X_COND_INIT;
226
227#if !HAVE_PREADWRITE
228/*
229 * make our pread/pwrite emulation safe against themselves, but not against
230 * normal read/write by using a mutex. slows down execution a lot,
231 * but that's your problem, not mine.
232 */
233static mutex_t preadwritelock = X_MUTEX_INIT;
234#endif
181 235
182typedef struct etp_worker 236typedef struct etp_worker
183{ 237{
184 /* locked by wrklock */ 238 /* locked by wrklock */
185 struct etp_worker *prev, *next; 239 struct etp_worker *prev, *next;
330 384
331static void etp_atfork_child (void) 385static void etp_atfork_child (void)
332{ 386{
333 ETP_REQ *prv; 387 ETP_REQ *prv;
334 388
335 while (prv = reqq_shift (&req_queue)) 389 while ((prv = reqq_shift (&req_queue)))
336 ETP_DESTROY (prv); 390 ETP_DESTROY (prv);
337 391
338 while (prv = reqq_shift (&res_queue)) 392 while ((prv = reqq_shift (&res_queue)))
339 ETP_DESTROY (prv); 393 ETP_DESTROY (prv);
340 394
341 while (wrk_first.next != &wrk_first) 395 while (wrk_first.next != &wrk_first)
342 { 396 {
343 etp_worker *wrk = wrk_first.next; 397 etp_worker *wrk = wrk_first.next;
371 425
372 pthread_once (&doinit, etp_once_init); 426 pthread_once (&doinit, etp_once_init);
373 427
374 want_poll_cb = want_poll; 428 want_poll_cb = want_poll;
375 done_poll_cb = done_poll; 429 done_poll_cb = done_poll;
430
431 return 0;
376} 432}
377 433
378X_THREAD_PROC (etp_proc); 434X_THREAD_PROC (etp_proc);
379 435
380static void etp_start_thread (void) 436static void etp_start_thread (void)
400 X_UNLOCK (wrklock); 456 X_UNLOCK (wrklock);
401} 457}
402 458
403static void etp_maybe_start_thread (void) 459static void etp_maybe_start_thread (void)
404{ 460{
405 if (etp_nthreads () >= wanted) 461 if (expect_true (etp_nthreads () >= wanted))
406 return; 462 return;
407 463
408 /* todo: maybe use idle here, but might be less exact */ 464 /* todo: maybe use idle here, but might be less exact */
409 if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) 465 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
410 return; 466 return;
411 467
412 etp_start_thread (); 468 etp_start_thread ();
413} 469}
414 470
467 523
468 X_LOCK (reqlock); 524 X_LOCK (reqlock);
469 --nreqs; 525 --nreqs;
470 X_UNLOCK (reqlock); 526 X_UNLOCK (reqlock);
471 527
472 if (req->type == EIO_GROUP && req->size) 528 if (expect_false (req->type == EIO_GROUP && req->size))
473 { 529 {
474 req->int1 = 1; /* mark request as delayed */ 530 req->int1 = 1; /* mark request as delayed */
475 continue; 531 continue;
476 } 532 }
477 else 533 else
478 { 534 {
479 int res = ETP_FINISH (req); 535 int res = ETP_FINISH (req);
480 if (res) 536 if (expect_false (res))
481 return res; 537 return res;
482 } 538 }
483 539
484 if (maxreqs && !--maxreqs) 540 if (expect_false (maxreqs && !--maxreqs))
485 break; 541 break;
486 542
487 if (maxtime) 543 if (maxtime)
488 { 544 {
489 gettimeofday (&tv_now, 0); 545 gettimeofday (&tv_now, 0);
508 564
509static void etp_submit (ETP_REQ *req) 565static void etp_submit (ETP_REQ *req)
510{ 566{
511 req->pri -= ETP_PRI_MIN; 567 req->pri -= ETP_PRI_MIN;
512 568
513 if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 569 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
514 if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 570 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
515 571
572 if (expect_false (req->type == EIO_GROUP))
573 {
574 /* I hope this is worth it :/ */
516 X_LOCK (reqlock); 575 X_LOCK (reqlock);
517 ++nreqs; 576 ++nreqs;
577 X_UNLOCK (reqlock);
578
579 X_LOCK (reslock);
580
581 ++npending;
582
583 if (!reqq_push (&res_queue, req) && want_poll_cb)
584 want_poll_cb ();
585
586 X_UNLOCK (reslock);
587 }
588 else
589 {
590 X_LOCK (reqlock);
591 ++nreqs;
518 ++nready; 592 ++nready;
519 reqq_push (&req_queue, req); 593 reqq_push (&req_queue, req);
520 X_COND_SIGNAL (reqwait); 594 X_COND_SIGNAL (reqwait);
521 X_UNLOCK (reqlock); 595 X_UNLOCK (reqlock);
522 596
523 etp_maybe_start_thread (); 597 etp_maybe_start_thread ();
598 }
524} 599}
525 600
526static void etp_set_max_poll_time (double nseconds) 601static void etp_set_max_poll_time (double nseconds)
527{ 602{
528 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 603 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
529 max_poll_time = nseconds; 604 max_poll_time = nseconds * EIO_TICKS;
530 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 605 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
531} 606}
532 607
533static void etp_set_max_poll_reqs (unsigned int maxreqs) 608static void etp_set_max_poll_reqs (unsigned int maxreqs)
534{ 609{
563 638
564static void grp_try_feed (eio_req *grp) 639static void grp_try_feed (eio_req *grp)
565{ 640{
566 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 641 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
567 { 642 {
568 int old_len = grp->size; 643 grp->flags &= ~EIO_FLAG_GROUPADD;
569 644
570 EIO_FEED (grp); 645 EIO_FEED (grp);
571 646
572 /* stop if no progress has been made */ 647 /* stop if no progress has been made */
573 if (old_len == grp->size) 648 if (!(grp->flags & EIO_FLAG_GROUPADD))
574 { 649 {
575 grp->feed = 0; 650 grp->feed = 0;
576 break; 651 break;
577 } 652 }
578 } 653 }
695 770
696/*****************************************************************************/ 771/*****************************************************************************/
697/* work around various missing functions */ 772/* work around various missing functions */
698 773
699#if !HAVE_PREADWRITE 774#if !HAVE_PREADWRITE
775# undef pread
776# undef pwrite
700# define pread eio__pread 777# define pread eio__pread
701# define pwrite eio__pwrite 778# define pwrite eio__pwrite
702
703/*
704 * make our pread/pwrite safe against themselves, but not against
705 * normal read/write by using a mutex. slows down execution a lot,
706 * but that's your problem, not mine.
707 */
708static mutex_t preadwritelock = X_MUTEX_INIT;
709 779
710static ssize_t 780static ssize_t
711eio__pread (int fd, void *buf, size_t count, off_t offset) 781eio__pread (int fd, void *buf, size_t count, off_t offset)
712{ 782{
713 ssize_t res; 783 ssize_t res;
731 801
732 X_LOCK (preadwritelock); 802 X_LOCK (preadwritelock);
733 ooffset = lseek (fd, 0, SEEK_CUR); 803 ooffset = lseek (fd, 0, SEEK_CUR);
734 lseek (fd, offset, SEEK_SET); 804 lseek (fd, offset, SEEK_SET);
735 res = write (fd, buf, count); 805 res = write (fd, buf, count);
736 lseek (fd, offset, SEEK_SET); 806 lseek (fd, ooffset, SEEK_SET);
737 X_UNLOCK (preadwritelock); 807 X_UNLOCK (preadwritelock);
738 808
739 return res; 809 return res;
740} 810}
741#endif 811#endif
742 812
743#ifndef HAVE_FUTIMES 813#ifndef HAVE_FUTIMES
744 814
815# undef utimes
816# undef futimes
745# define utimes(path,times) eio__utimes (path, times) 817# define utimes(path,times) eio__utimes (path, times)
746# define futimes(fd,times) eio__futimes (fd, times) 818# define futimes(fd,times) eio__futimes (fd, times)
747 819
748static int 820static int
749eio__utimes (const char *filename, const struct timeval times[2]) 821eio__utimes (const char *filename, const struct timeval times[2])
768} 840}
769 841
770#endif 842#endif
771 843
772#if !HAVE_FDATASYNC 844#if !HAVE_FDATASYNC
845# undef fdatasync
773# define fdatasync fsync 846# define fdatasync(fd) fsync (fd)
774#endif 847#endif
848
849/* sync_file_range always needs emulation */
850int
851eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
852{
853#if HAVE_SYNC_FILE_RANGE
854 int res;
855
856 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
857 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
858 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
859 {
860 flags = 0
861 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
862 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
863 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
864 }
865
866 res = sync_file_range (fd, offset, nbytes, flags);
867
868 if (!res || errno != ENOSYS)
869 return res;
870#endif
871
872 /* even though we could play tricks with the flags, it's better to always
873 * call fdatasync, as that matches the expectation of its users best */
874 return fdatasync (fd);
875}
775 876
776#if !HAVE_READAHEAD 877#if !HAVE_READAHEAD
878# undef readahead
777# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 879# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
778 880
779static ssize_t 881static ssize_t
780eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 882eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
781{ 883{
808 910
809#if HAVE_SENDFILE 911#if HAVE_SENDFILE
810# if __linux 912# if __linux
811 res = sendfile (ofd, ifd, &offset, count); 913 res = sendfile (ofd, ifd, &offset, count);
812 914
813# elif __freebsd 915# elif __FreeBSD__
814 /* 916 /*
815 * Of course, the freebsd sendfile is a dire hack with no thoughts 917 * Of course, the freebsd sendfile is a dire hack with no thoughts
816 * wasted on making it similar to other I/O functions. 918 * wasted on making it similar to other I/O functions.
817 */ 919 */
818 { 920 {
819 off_t sbytes; 921 off_t sbytes;
820 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 922 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
821 923
822 if (res < 0 && sbytes) 924 /* freebsd' sendfile will return 0 on success */
823 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 925 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
926 /* not on e.g. EIO or EPIPE - sounds broken */
927 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
928 res = sbytes;
929 }
930
931# elif defined (__APPLE__)
932
933 {
934 off_t sbytes = count;
935 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
936
937 /* according to the manpage, sbytes is always valid */
938 if (sbytes)
824 res = sbytes; 939 res = sbytes;
825 } 940 }
826 941
827# elif __hpux 942# elif __hpux
828 res = sendfile (ofd, ifd, offset, count, 0, 0); 943 res = sendfile (ofd, ifd, offset, count, 0, 0);
842 if (res < 0 && sbytes) 957 if (res < 0 && sbytes)
843 res = sbytes; 958 res = sbytes;
844 } 959 }
845 960
846# endif 961# endif
962
963#elif defined (_WIN32)
964
965 /* does not work, just for documentation of what would need to be done */
966 {
967 HANDLE h = TO_SOCKET (ifd);
968 SetFilePointer (h, offset, 0, FILE_BEGIN);
969 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
970 }
971
847#else 972#else
848 res = -1; 973 res = -1;
849 errno = ENOSYS; 974 errno = ENOSYS;
850#endif 975#endif
851 976
852 if (res < 0 977 if (res < 0
853 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 978 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
979 /* BSDs */
980#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
981 || errno == ENOTSUP
982#endif
983 || errno == EOPNOTSUPP /* BSDs */
854#if __solaris 984#if __solaris
855 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 985 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
856#endif 986#endif
857 ) 987 )
858 ) 988 )
889 } 1019 }
890 1020
891 return res; 1021 return res;
892} 1022}
893 1023
1024static signed char
1025eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1026{
1027 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1028 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1029}
1030
1031#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1032
1033#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1034#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1035
1036static void
1037eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1038{
1039 unsigned char bits [9 + sizeof (ino_t) * 8];
1040 unsigned char *bit = bits;
1041
1042 assert (CHAR_BIT == 8);
1043 assert (sizeof (eio_dirent) * 8 < 256);
1044 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1045 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1046
1047 if (size <= EIO_SORT_FAST)
1048 return;
1049
1050 /* first prepare an array of bits to test in our radix sort */
1051 /* try to take endianness into account, as well as differences in ino_t sizes */
1052 /* inode_bits must contain all inodes ORed together */
1053 /* which is used to skip bits that are 0 everywhere, which is very common */
1054 {
1055 ino_t endianness;
1056 int i, j;
1057
1058 /* we store the byte offset of byte n into byte n of "endianness" */
1059 for (i = 0; i < sizeof (ino_t); ++i)
1060 ((unsigned char *)&endianness)[i] = i;
1061
1062 *bit++ = 0;
1063
1064 for (i = 0; i < sizeof (ino_t); ++i)
1065 {
1066 /* shifting off the byte offsets out of "endianness" */
1067 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1068 endianness >>= 8;
1069
1070 for (j = 0; j < 8; ++j)
1071 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1072 *bit++ = offs + j;
1073 }
1074
1075 for (j = 0; j < 8; ++j)
1076 if (score_bits & (1 << j))
1077 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1078 }
1079
1080 /* now actually do the sorting (a variant of MSD radix sort) */
1081 {
1082 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1083 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1084 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1085 int stk_idx = 0;
1086
1087 base_stk [stk_idx] = dents;
1088 end_stk [stk_idx] = dents + size;
1089 bit_stk [stk_idx] = bit - 1;
1090
1091 do
1092 {
1093 base = base_stk [stk_idx];
1094 end = end_stk [stk_idx];
1095 bit = bit_stk [stk_idx];
1096
1097 for (;;)
1098 {
1099 unsigned char O = *bit >> 3;
1100 unsigned char M = 1 << (*bit & 7);
1101
1102 eio_dirent *a = base;
1103 eio_dirent *b = end;
1104
1105 if (b - a < EIO_SORT_CUTOFF)
1106 break;
1107
1108 /* now bit-partition the array on the bit */
1109 /* this ugly asymmetric loop seems to perform much better than typical */
1110 /* partition algos found in the literature */
1111 do
1112 if (!(((unsigned char *)a)[O] & M))
1113 ++a;
1114 else if (!(((unsigned char *)--b)[O] & M))
1115 {
1116 eio_dirent tmp = *a; *a = *b; *b = tmp;
1117 ++a;
1118 }
1119 while (b > a);
1120
1121 /* next bit, or stop, if no bits left in this path */
1122 if (!*--bit)
1123 break;
1124
1125 base_stk [stk_idx] = a;
1126 end_stk [stk_idx] = end;
1127 bit_stk [stk_idx] = bit;
1128 ++stk_idx;
1129
1130 end = a;
1131 }
1132 }
1133 while (stk_idx--);
1134 }
1135}
1136
1137static void
1138eio_dent_insertion_sort (eio_dirent *dents, int size)
1139{
1140 /* first move the smallest element to the front, to act as a sentinel */
1141 {
1142 int i;
1143 eio_dirent *min = dents;
1144
1145 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1146 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1147 if (EIO_DENT_CMP (dents [i], <, *min))
1148 min = &dents [i];
1149
1150 /* swap elements 0 and j (minimum) */
1151 {
1152 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1153 }
1154 }
1155
1156 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1157 {
1158 eio_dirent *i, *j;
1159
1160 for (i = dents + 1; i < dents + size; ++i)
1161 {
1162 eio_dirent value = *i;
1163
1164 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1165 j [1] = j [0];
1166
1167 j [1] = value;
1168 }
1169 }
1170}
1171
1172static void
1173eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1174{
1175 if (size <= 1)
1176 return; /* our insertion sort relies on size > 0 */
1177
1178 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1179 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1180 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1181
1182 /* use an insertion sort at the end, or for small arrays, */
1183 /* as insertion sort is more efficient for small partitions */
1184 eio_dent_insertion_sort (dents, size);
1185}
1186
894/* read a full directory */ 1187/* read a full directory */
895static void 1188static void
896eio__scandir (eio_req *req, etp_worker *self) 1189eio__scandir (eio_req *req, etp_worker *self)
897{ 1190{
898 DIR *dirp; 1191 DIR *dirp;
899 EIO_STRUCT_DIRENT *entp; 1192 EIO_STRUCT_DIRENT *entp;
900 char *name, *names; 1193 char *name, *names;
901 int memlen = 4096; 1194 int namesalloc = 4096;
902 int memofs = 0; 1195 int namesoffs = 0;
1196 int flags = req->int1;
1197 eio_dirent *dents = 0;
1198 int dentalloc = 128;
903 int res = 0; 1199 int dentoffs = 0;
1200 ino_t inode_bits = 0;
1201
1202 req->result = -1;
1203
1204 if (!(flags & EIO_READDIR_DENTS))
1205 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
904 1206
905 X_LOCK (wrklock); 1207 X_LOCK (wrklock);
1208 /* the corresponding closedir is in ETP_WORKER_CLEAR */
906 self->dirp = dirp = opendir (req->ptr1); 1209 self->dirp = dirp = opendir (req->ptr1);
907 req->flags |= EIO_FLAG_PTR2_FREE; 1210 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1211 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
908 req->ptr2 = names = malloc (memlen); 1212 req->ptr2 = names = malloc (namesalloc);
909 X_UNLOCK (wrklock); 1213 X_UNLOCK (wrklock);
910 1214
911 if (dirp && names) 1215 if (dirp && names && (!flags || dents))
912 for (;;) 1216 for (;;)
913 { 1217 {
914 errno = 0; 1218 errno = 0;
915 entp = readdir (dirp); 1219 entp = readdir (dirp);
916 1220
917 if (!entp) 1221 if (!entp)
1222 {
1223 if (errno)
1224 break;
1225
1226 /* sort etc. */
1227 req->int1 = flags;
1228 req->result = dentoffs;
1229
1230 if (flags & EIO_READDIR_STAT_ORDER)
1231 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
1232 else if (flags & EIO_READDIR_DIRS_FIRST)
1233 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1234 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1235 else
1236 {
1237 /* in this case, all is known, and we just put dirs first and sort them */
1238 eio_dirent *oth = dents + dentoffs;
1239 eio_dirent *dir = dents;
1240
1241 /* now partition dirs to the front, and non-dirs to the back */
1242 /* by walking from both sides and swapping if necessary */
1243 /* also clear score, so it doesn't influence sorting */
1244 while (oth > dir)
1245 {
1246 if (dir->type == EIO_DT_DIR)
1247 ++dir;
1248 else if ((--oth)->type == EIO_DT_DIR)
1249 {
1250 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1251
1252 ++dir;
1253 }
1254 }
1255
1256 /* now sort the dirs only */
1257 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1258 }
1259
918 break; 1260 break;
1261 }
919 1262
1263 /* now add the entry to our list(s) */
920 name = entp->d_name; 1264 name = entp->d_name;
921 1265
1266 /* skip . and .. entries */
922 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1267 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
923 { 1268 {
924 int len = strlen (name) + 1; 1269 int len = D_NAMLEN (entp) + 1;
925 1270
926 res++; 1271 while (expect_false (namesoffs + len > namesalloc))
927
928 while (memofs + len > memlen)
929 { 1272 {
930 memlen *= 2; 1273 namesalloc *= 2;
931 X_LOCK (wrklock); 1274 X_LOCK (wrklock);
932 req->ptr2 = names = realloc (names, memlen); 1275 req->ptr2 = names = realloc (names, namesalloc);
933 X_UNLOCK (wrklock); 1276 X_UNLOCK (wrklock);
934 1277
935 if (!names) 1278 if (!names)
936 break; 1279 break;
937 } 1280 }
938 1281
939 memcpy (names + memofs, name, len); 1282 memcpy (names + namesoffs, name, len);
1283
1284 if (dents)
1285 {
1286 struct eio_dirent *ent;
1287
1288 if (expect_false (dentoffs == dentalloc))
1289 {
1290 dentalloc *= 2;
1291 X_LOCK (wrklock);
1292 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1293 X_UNLOCK (wrklock);
1294
1295 if (!dents)
1296 break;
1297 }
1298
1299 ent = dents + dentoffs;
1300
1301 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1302 ent->namelen = len - 1;
1303 ent->inode = D_INO (entp);
1304
1305 inode_bits |= ent->inode;
1306
1307 switch (D_TYPE (entp))
1308 {
1309 default:
1310 ent->type = EIO_DT_UNKNOWN;
1311 flags |= EIO_READDIR_FOUND_UNKNOWN;
1312 break;
1313
1314 #ifdef DT_FIFO
1315 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1316 #endif
1317 #ifdef DT_CHR
1318 case DT_CHR: ent->type = EIO_DT_CHR; break;
1319 #endif
1320 #ifdef DT_MPC
1321 case DT_MPC: ent->type = EIO_DT_MPC; break;
1322 #endif
1323 #ifdef DT_DIR
1324 case DT_DIR: ent->type = EIO_DT_DIR; break;
1325 #endif
1326 #ifdef DT_NAM
1327 case DT_NAM: ent->type = EIO_DT_NAM; break;
1328 #endif
1329 #ifdef DT_BLK
1330 case DT_BLK: ent->type = EIO_DT_BLK; break;
1331 #endif
1332 #ifdef DT_MPB
1333 case DT_MPB: ent->type = EIO_DT_MPB; break;
1334 #endif
1335 #ifdef DT_REG
1336 case DT_REG: ent->type = EIO_DT_REG; break;
1337 #endif
1338 #ifdef DT_NWK
1339 case DT_NWK: ent->type = EIO_DT_NWK; break;
1340 #endif
1341 #ifdef DT_CMP
1342 case DT_CMP: ent->type = EIO_DT_CMP; break;
1343 #endif
1344 #ifdef DT_LNK
1345 case DT_LNK: ent->type = EIO_DT_LNK; break;
1346 #endif
1347 #ifdef DT_SOCK
1348 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1349 #endif
1350 #ifdef DT_DOOR
1351 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1352 #endif
1353 #ifdef DT_WHT
1354 case DT_WHT: ent->type = EIO_DT_WHT; break;
1355 #endif
1356 }
1357
1358 ent->score = 7;
1359
1360 if (flags & EIO_READDIR_DIRS_FIRST)
1361 {
1362 if (ent->type == EIO_DT_UNKNOWN)
1363 {
1364 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1365 ent->score = 1;
1366 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1367 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1368 }
1369 else if (ent->type == EIO_DT_DIR)
1370 ent->score = 0;
1371 }
1372 }
1373
940 memofs += len; 1374 namesoffs += len;
1375 ++dentoffs;
1376 }
1377
1378 if (EIO_CANCELLED (req))
1379 {
1380 errno = ECANCELED;
1381 break;
941 } 1382 }
942 } 1383 }
1384}
943 1385
944 if (errno) 1386#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
945 res = -1; 1387# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
946 1388#else
947 req->result = res; 1389
1390int
1391eio__msync (void *mem, size_t len, int flags)
1392{
1393 if (EIO_MS_ASYNC != MS_SYNC
1394 || EIO_MS_INVALIDATE != MS_INVALIDATE
1395 || EIO_MS_SYNC != MS_SYNC)
1396 {
1397 flags = 0
1398 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1399 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1400 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1401 }
1402
1403 return msync (mem, len, flags);
1404}
1405
1406#endif
1407
1408int
1409eio__mtouch (void *mem, size_t len, int flags)
1410{
1411 intptr_t addr = (intptr_t)mem;
1412 intptr_t end = addr + len;
1413#ifdef PAGESIZE
1414 const intptr_t page = PAGESIZE;
1415#else
1416 static intptr_t page;
1417
1418 if (!page)
1419 page = sysconf (_SC_PAGESIZE);
1420#endif
1421
1422 /* round down to start of page, although this is probably useless */
1423 addr &= ~(page - 1); /* assume page size is always a power of two */
1424
1425 if (addr < end)
1426 if (flags & EIO_MT_MODIFY) /* modify */
1427 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1428 else
1429 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1430
1431 return 0;
948} 1432}
949 1433
950/*****************************************************************************/ 1434/*****************************************************************************/
951 1435
952#define ALLOC(len) \ 1436#define ALLOC(len) \
1039 1523
1040/*****************************************************************************/ 1524/*****************************************************************************/
1041 1525
1042int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1526int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1043{ 1527{
1044 etp_init (want_poll, done_poll); 1528 return etp_init (want_poll, done_poll);
1045} 1529}
1046 1530
1047static void eio_api_destroy (eio_req *req) 1531static void eio_api_destroy (eio_req *req)
1048{ 1532{
1049 free (req); 1533 free (req);
1095 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1579 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1096 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1580 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1097 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1581 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1098 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1582 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1099 1583
1584 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1585 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1586 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1587 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1588
1100 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1589 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1101 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1590 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1102 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1591 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1103 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1592 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1104 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1593 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1111 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1600 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1112 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1601 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1113 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1602 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1114 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1603 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1115 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1604 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1116 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; 1605 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1117 1606
1118 case EIO_READLINK: ALLOC (NAME_MAX); 1607 case EIO_READLINK: ALLOC (NAME_MAX);
1119 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1608 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1120 1609
1121 case EIO_SYNC: req->result = 0; sync (); break; 1610 case EIO_SYNC: req->result = 0; sync (); break;
1122 case EIO_FSYNC: req->result = fsync (req->int1); break; 1611 case EIO_FSYNC: req->result = fsync (req->int1); break;
1123 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1612 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1613 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1614 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1615 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1124 1616
1125 case EIO_READDIR: eio__scandir (req, self); break; 1617 case EIO_READDIR: eio__scandir (req, self); break;
1126 1618
1127 case EIO_BUSY: 1619 case EIO_BUSY:
1128#ifdef _WIN32 1620#ifdef _WIN32
1160 1652
1161 req->result = req->type == EIO_FUTIME 1653 req->result = req->type == EIO_FUTIME
1162 ? futimes (req->int1, times) 1654 ? futimes (req->int1, times)
1163 : utimes (req->ptr1, times); 1655 : utimes (req->ptr1, times);
1164 } 1656 }
1657 break;
1165 1658
1166 case EIO_GROUP: 1659 case EIO_GROUP:
1660 abort (); /* handled in eio_request */
1661
1167 case EIO_NOP: 1662 case EIO_NOP:
1168 req->result = 0; 1663 req->result = 0;
1169 break; 1664 break;
1170 1665
1171 case EIO_CUSTOM: 1666 case EIO_CUSTOM:
1172 req->feed (req); 1667 ((void (*)(eio_req *))req->feed) (req);
1173 break; 1668 break;
1174 1669
1175 default: 1670 default:
1176 req->result = -1; 1671 req->result = -1;
1177 break; 1672 break;
1200eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 1695eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1201{ 1696{
1202 REQ (EIO_FSYNC); req->int1 = fd; SEND; 1697 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1203} 1698}
1204 1699
1700eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1701{
1702 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1703}
1704
1705eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1706{
1707 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1708}
1709
1710eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1711{
1712 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1713}
1714
1205eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 1715eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1206{ 1716{
1207 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1717 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1208} 1718}
1209 1719
1228} 1738}
1229 1739
1230eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1740eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1231{ 1741{
1232 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1742 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1743}
1744
1745eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1746{
1747 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1233} 1748}
1234 1749
1235eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1750eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1236{ 1751{
1237 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1752 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1311eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 1826eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1312{ 1827{
1313 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1828 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1314} 1829}
1315 1830
1831eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1832{
1833 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1834}
1835
1316eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1836eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1317{ 1837{
1318 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1838 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1319} 1839}
1320 1840
1321eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1841eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1322{ 1842{
1323 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1843 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1324} 1844}
1325 1845
1326eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1846eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1327{ 1847{
1328 return eio__1path (EIO_READDIR, path, pri, cb, data); 1848 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1329} 1849}
1330 1850
1331eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1851eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1332{ 1852{
1333 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1853 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1334} 1854}
1335 1855
1336static eio_req * 1856static eio_req *
1337eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1857eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1338{ 1858{
1364 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 1884 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1365} 1885}
1366 1886
1367eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 1887eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1368{ 1888{
1369 REQ (EIO_CUSTOM); req->feed = execute; SEND; 1889 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1370} 1890}
1371 1891
1372#endif 1892#endif
1373 1893
1374eio_req *eio_grp (eio_cb cb, void *data) 1894eio_req *eio_grp (eio_cb cb, void *data)
1402 1922
1403void eio_grp_add (eio_req *grp, eio_req *req) 1923void eio_grp_add (eio_req *grp, eio_req *req)
1404{ 1924{
1405 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1925 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1406 1926
1927 grp->flags |= EIO_FLAG_GROUPADD;
1928
1407 ++grp->size; 1929 ++grp->size;
1408 req->grp = grp; 1930 req->grp = grp;
1409 1931
1410 req->grp_prev = 0; 1932 req->grp_prev = 0;
1411 req->grp_next = grp->grp_first; 1933 req->grp_next = grp->grp_first;
1420/* misc garbage */ 1942/* misc garbage */
1421 1943
1422ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 1944ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1423{ 1945{
1424 etp_worker wrk; 1946 etp_worker wrk;
1947 ssize_t ret;
1425 1948
1426 wrk.dbuf = 0; 1949 wrk.dbuf = 0;
1427 1950
1428 eio__sendfile (ofd, ifd, offset, count, &wrk); 1951 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1429 1952
1430 if (wrk.dbuf) 1953 if (wrk.dbuf)
1431 free (wrk.dbuf); 1954 free (wrk.dbuf);
1432}
1433 1955
1956 return ret;
1957}
1958

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines