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 1.15 by root, Sat May 17 12:26:57 2008 UTC vs.
Revision 1.54 by root, Tue Feb 23 14:32:45 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
99#endif 122#endif
100 123
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)
132#endif
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
104/* used for struct dirent, AIX doesn't provide it */ 137/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 138#ifndef NAME_MAX
106# define NAME_MAX 4096 139# define NAME_MAX 4096
140#endif
141
142/* used for readlink etc. */
143#ifndef PATH_MAX
144# define PATH_MAX 4096
107#endif 145#endif
108 146
109/* buffer size for various temporary buffers */ 147/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 148#define EIO_BUFSIZE 65536
111 149
… …
117 errno = ENOMEM; \ 155 errno = ENOMEM; \
118 if (!eio_buf) \ 156 if (!eio_buf) \
119 return -1; 157 return -1;
120 158
121#define EIO_TICKS ((1000000 + 1023) >> 10) 159#define EIO_TICKS ((1000000 + 1023) >> 10)
160
161/*****************************************************************************/
162
163#if __GNUC__ >= 3
164# define expect(expr,value) __builtin_expect ((expr),(value))
165#else
166# define expect(expr,value) (expr)
167#endif
168
169#define expect_false(expr) expect ((expr) != 0, 0)
170#define expect_true(expr) expect ((expr) != 0, 1)
122 171
123/*****************************************************************************/ 172/*****************************************************************************/
124 173
125#define ETP_PRI_MIN EIO_PRI_MIN 174#define ETP_PRI_MIN EIO_PRI_MIN
126#define ETP_PRI_MAX EIO_PRI_MAX 175#define ETP_PRI_MAX EIO_PRI_MAX
… …
144 if (wrk->dirp) \ 193 if (wrk->dirp) \
145 { \ 194 { \
146 closedir (wrk->dirp); \ 195 closedir (wrk->dirp); \
147 wrk->dirp = 0; \ 196 wrk->dirp = 0; \
148 } 197 }
198
149#define ETP_WORKER_COMMON \ 199#define ETP_WORKER_COMMON \
150 void *dbuf; \ 200 void *dbuf; \
151 DIR *dirp; 201 DIR *dirp;
152 202
153/*****************************************************************************/ 203/*****************************************************************************/
154 204
155#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 205#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
156 206
157/* calculcate time difference in ~1/EIO_TICKS of a second */ 207/* calculate time difference in ~1/EIO_TICKS of a second */
158static int tvdiff (struct timeval *tv1, struct timeval *tv2) 208static int tvdiff (struct timeval *tv1, struct timeval *tv2)
159{ 209{
160 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 210 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
161 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 211 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
162} 212}
… …
176 226
177static mutex_t wrklock = X_MUTEX_INIT; 227static mutex_t wrklock = X_MUTEX_INIT;
178static mutex_t reslock = X_MUTEX_INIT; 228static mutex_t reslock = X_MUTEX_INIT;
179static mutex_t reqlock = X_MUTEX_INIT; 229static mutex_t reqlock = X_MUTEX_INIT;
180static cond_t reqwait = X_COND_INIT; 230static cond_t reqwait = X_COND_INIT;
231
232#if !HAVE_PREADWRITE
233/*
234 * make our pread/pwrite emulation safe against themselves, but not against
235 * normal read/write by using a mutex. slows down execution a lot,
236 * but that's your problem, not mine.
237 */
238static mutex_t preadwritelock = X_MUTEX_INIT;
239#endif
181 240
182typedef struct etp_worker 241typedef struct etp_worker
183{ 242{
184 /* locked by wrklock */ 243 /* locked by wrklock */
185 struct etp_worker *prev, *next; 244 struct etp_worker *prev, *next;
… …
314 X_LOCK (reqlock); 373 X_LOCK (reqlock);
315 X_LOCK (reslock); 374 X_LOCK (reslock);
316#if !HAVE_PREADWRITE 375#if !HAVE_PREADWRITE
317 X_LOCK (preadwritelock); 376 X_LOCK (preadwritelock);
318#endif 377#endif
319#if !HAVE_READDIR_R
320 X_LOCK (readdirlock);
321#endif
322} 378}
323 379
324static void etp_atfork_parent (void) 380static void etp_atfork_parent (void)
325{ 381{
326#if !HAVE_READDIR_R
327 X_UNLOCK (readdirlock);
328#endif
329#if !HAVE_PREADWRITE 382#if !HAVE_PREADWRITE
330 X_UNLOCK (preadwritelock); 383 X_UNLOCK (preadwritelock);
331#endif 384#endif
332 X_UNLOCK (reslock); 385 X_UNLOCK (reslock);
333 X_UNLOCK (reqlock); 386 X_UNLOCK (reqlock);
… …
336 389
337static void etp_atfork_child (void) 390static void etp_atfork_child (void)
338{ 391{
339 ETP_REQ *prv; 392 ETP_REQ *prv;
340 393
341 while (prv = reqq_shift (&req_queue)) 394 while ((prv = reqq_shift (&req_queue)))
342 ETP_DESTROY (prv); 395 ETP_DESTROY (prv);
343 396
344 while (prv = reqq_shift (&res_queue)) 397 while ((prv = reqq_shift (&res_queue)))
345 ETP_DESTROY (prv); 398 ETP_DESTROY (prv);
346 399
347 while (wrk_first.next != &wrk_first) 400 while (wrk_first.next != &wrk_first)
348 { 401 {
349 etp_worker *wrk = wrk_first.next; 402 etp_worker *wrk = wrk_first.next;
… …
377 430
378 pthread_once (&doinit, etp_once_init); 431 pthread_once (&doinit, etp_once_init);
379 432
380 want_poll_cb = want_poll; 433 want_poll_cb = want_poll;
381 done_poll_cb = done_poll; 434 done_poll_cb = done_poll;
435
436 return 0;
382} 437}
383 438
384X_THREAD_PROC (etp_proc); 439X_THREAD_PROC (etp_proc);
385 440
386static void etp_start_thread (void) 441static void etp_start_thread (void)
… …
406 X_UNLOCK (wrklock); 461 X_UNLOCK (wrklock);
407} 462}
408 463
409static void etp_maybe_start_thread (void) 464static void etp_maybe_start_thread (void)
410{ 465{
411 if (etp_nthreads () >= wanted) 466 if (expect_true (etp_nthreads () >= wanted))
412 return; 467 return;
413 468
414 /* todo: maybe use idle here, but might be less exact */ 469 /* todo: maybe use idle here, but might be less exact */
415 if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) 470 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
416 return; 471 return;
417 472
418 etp_start_thread (); 473 etp_start_thread ();
419} 474}
420 475
… …
473 528
474 X_LOCK (reqlock); 529 X_LOCK (reqlock);
475 --nreqs; 530 --nreqs;
476 X_UNLOCK (reqlock); 531 X_UNLOCK (reqlock);
477 532
478 if (req->type == EIO_GROUP && req->size) 533 if (expect_false (req->type == EIO_GROUP && req->size))
479 { 534 {
480 req->int1 = 1; /* mark request as delayed */ 535 req->int1 = 1; /* mark request as delayed */
481 continue; 536 continue;
482 } 537 }
483 else 538 else
484 { 539 {
485 int res = ETP_FINISH (req); 540 int res = ETP_FINISH (req);
486 if (res) 541 if (expect_false (res))
487 return res; 542 return res;
488 } 543 }
489 544
490 if (maxreqs && !--maxreqs) 545 if (expect_false (maxreqs && !--maxreqs))
491 break; 546 break;
492 547
493 if (maxtime) 548 if (maxtime)
494 { 549 {
495 gettimeofday (&tv_now, 0); 550 gettimeofday (&tv_now, 0);
… …
514 569
515static void etp_submit (ETP_REQ *req) 570static void etp_submit (ETP_REQ *req)
516{ 571{
517 req->pri -= ETP_PRI_MIN; 572 req->pri -= ETP_PRI_MIN;
518 573
519 if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 574 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
520 if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 575 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
521 576
577 if (expect_false (req->type == EIO_GROUP))
578 {
579 /* I hope this is worth it :/ */
522 X_LOCK (reqlock); 580 X_LOCK (reqlock);
523 ++nreqs; 581 ++nreqs;
582 X_UNLOCK (reqlock);
583
584 X_LOCK (reslock);
585
586 ++npending;
587
588 if (!reqq_push (&res_queue, req) && want_poll_cb)
589 want_poll_cb ();
590
591 X_UNLOCK (reslock);
592 }
593 else
594 {
595 X_LOCK (reqlock);
596 ++nreqs;
524 ++nready; 597 ++nready;
525 reqq_push (&req_queue, req); 598 reqq_push (&req_queue, req);
526 X_COND_SIGNAL (reqwait); 599 X_COND_SIGNAL (reqwait);
527 X_UNLOCK (reqlock); 600 X_UNLOCK (reqlock);
528 601
529 etp_maybe_start_thread (); 602 etp_maybe_start_thread ();
603 }
530} 604}
531 605
532static void etp_set_max_poll_time (double nseconds) 606static void etp_set_max_poll_time (double nseconds)
533{ 607{
534 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 608 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
535 max_poll_time = nseconds; 609 max_poll_time = nseconds * EIO_TICKS;
536 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 610 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
537} 611}
538 612
539static void etp_set_max_poll_reqs (unsigned int maxreqs) 613static void etp_set_max_poll_reqs (unsigned int maxreqs)
540{ 614{
… …
569 643
570static void grp_try_feed (eio_req *grp) 644static void grp_try_feed (eio_req *grp)
571{ 645{
572 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 646 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
573 { 647 {
574 int old_len = grp->size; 648 grp->flags &= ~EIO_FLAG_GROUPADD;
575 649
576 EIO_FEED (grp); 650 EIO_FEED (grp);
577 651
578 /* stop if no progress has been made */ 652 /* stop if no progress has been made */
579 if (old_len == grp->size) 653 if (!(grp->flags & EIO_FLAG_GROUPADD))
580 { 654 {
581 grp->feed = 0; 655 grp->feed = 0;
582 break; 656 break;
583 } 657 }
584 } 658 }
… …
701 775
702/*****************************************************************************/ 776/*****************************************************************************/
703/* work around various missing functions */ 777/* work around various missing functions */
704 778
705#if !HAVE_PREADWRITE 779#if !HAVE_PREADWRITE
780# undef pread
781# undef pwrite
706# define pread eio__pread 782# define pread eio__pread
707# define pwrite eio__pwrite 783# define pwrite eio__pwrite
708
709/*
710 * make our pread/pwrite safe against themselves, but not against
711 * normal read/write by using a mutex. slows down execution a lot,
712 * but that's your problem, not mine.
713 */
714static mutex_t preadwritelock = X_MUTEX_INIT;
715 784
716static ssize_t 785static ssize_t
717eio__pread (int fd, void *buf, size_t count, off_t offset) 786eio__pread (int fd, void *buf, size_t count, off_t offset)
718{ 787{
719 ssize_t res; 788 ssize_t res;
… …
737 806
738 X_LOCK (preadwritelock); 807 X_LOCK (preadwritelock);
739 ooffset = lseek (fd, 0, SEEK_CUR); 808 ooffset = lseek (fd, 0, SEEK_CUR);
740 lseek (fd, offset, SEEK_SET); 809 lseek (fd, offset, SEEK_SET);
741 res = write (fd, buf, count); 810 res = write (fd, buf, count);
742 lseek (fd, offset, SEEK_SET); 811 lseek (fd, ooffset, SEEK_SET);
743 X_UNLOCK (preadwritelock); 812 X_UNLOCK (preadwritelock);
744 813
745 return res; 814 return res;
746} 815}
747#endif 816#endif
748 817
749#ifndef HAVE_FUTIMES 818#ifndef HAVE_FUTIMES
750 819
820# undef utimes
821# undef futimes
751# define utimes(path,times) eio__utimes (path, times) 822# define utimes(path,times) eio__utimes (path, times)
752# define futimes(fd,times) eio__futimes (fd, times) 823# define futimes(fd,times) eio__futimes (fd, times)
753 824
754static int 825static int
755eio__utimes (const char *filename, const struct timeval times[2]) 826eio__utimes (const char *filename, const struct timeval times[2])
… …
774} 845}
775 846
776#endif 847#endif
777 848
778#if !HAVE_FDATASYNC 849#if !HAVE_FDATASYNC
850# undef fdatasync
779# define fdatasync fsync 851# define fdatasync(fd) fsync (fd)
780#endif 852#endif
853
854/* sync_file_range always needs emulation */
855int
856eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
857{
858#if HAVE_SYNC_FILE_RANGE
859 int res;
860
861 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
862 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
863 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
864 {
865 flags = 0
866 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
867 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
868 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
869 }
870
871 res = sync_file_range (fd, offset, nbytes, flags);
872
873 if (!res || errno != ENOSYS)
874 return res;
875#endif
876
877 /* even though we could play tricks with the flags, it's better to always
878 * call fdatasync, as that matches the expectation of its users best */
879 return fdatasync (fd);
880}
781 881
782#if !HAVE_READAHEAD 882#if !HAVE_READAHEAD
883# undef readahead
783# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 884# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
784 885
785static ssize_t 886static ssize_t
786eio__readahead (int fd, off_t offset, size_t count, worker *self) 887eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
787{ 888{
788 size_t todo = count; 889 size_t todo = count;
789 dBUF; 890 dBUF;
790 891
791 while (todo > 0) 892 while (todo > 0)
… …
801 return count; 902 return count;
802} 903}
803 904
804#endif 905#endif
805 906
806#if !HAVE_READDIR_R
807# define readdir_r eio__readdir_r
808
809static mutex_t readdirlock = X_MUTEX_INIT;
810
811static int
812eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
813{
814 EIO_STRUCT_DIRENT *e;
815 int errorno;
816
817 X_LOCK (readdirlock);
818
819 e = readdir (dirp);
820 errorno = errno;
821
822 if (e)
823 {
824 *res = ent;
825 strcpy (ent->d_name, e->d_name);
826 }
827 else
828 *res = 0;
829
830 X_UNLOCK (readdirlock);
831
832 errno = errorno;
833 return e ? 0 : -1;
834}
835#endif
836
837/* sendfile always needs emulation */ 907/* sendfile always needs emulation */
838static ssize_t 908static ssize_t
839eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 909eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
840{ 910{
841 ssize_t res; 911 ssize_t res;
… …
845 915
846#if HAVE_SENDFILE 916#if HAVE_SENDFILE
847# if __linux 917# if __linux
848 res = sendfile (ofd, ifd, &offset, count); 918 res = sendfile (ofd, ifd, &offset, count);
849 919
850# elif __freebsd 920# elif __FreeBSD__
851 /* 921 /*
852 * Of course, the freebsd sendfile is a dire hack with no thoughts 922 * Of course, the freebsd sendfile is a dire hack with no thoughts
853 * wasted on making it similar to other I/O functions. 923 * wasted on making it similar to other I/O functions.
854 */ 924 */
855 { 925 {
856 off_t sbytes; 926 off_t sbytes;
857 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 927 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
858 928
859 if (res < 0 && sbytes) 929 #if 0 /* according to the manpage, this is correct, but broken behaviour */
860 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 930 /* freebsd' sendfile will return 0 on success */
931 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
932 /* not on e.g. EIO or EPIPE - sounds broken */
933 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
934 res = sbytes;
935 #endif
936
937 /* according to source inspection, this is correct, and useful behaviour */
938 if (sbytes)
939 res = sbytes;
940 }
941
942# elif defined (__APPLE__)
943
944 {
945 off_t sbytes = count;
946 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
947
948 /* according to the manpage, sbytes is always valid */
949 if (sbytes)
861 res = sbytes; 950 res = sbytes;
862 } 951 }
863 952
864# elif __hpux 953# elif __hpux
865 res = sendfile (ofd, ifd, offset, count, 0, 0); 954 res = sendfile (ofd, ifd, offset, count, 0, 0);
… …
879 if (res < 0 && sbytes) 968 if (res < 0 && sbytes)
880 res = sbytes; 969 res = sbytes;
881 } 970 }
882 971
883# endif 972# endif
973
974#elif defined (_WIN32)
975
976 /* does not work, just for documentation of what would need to be done */
977 {
978 HANDLE h = TO_SOCKET (ifd);
979 SetFilePointer (h, offset, 0, FILE_BEGIN);
980 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
981 }
982
884#else 983#else
885 res = -1; 984 res = -1;
886 errno = ENOSYS; 985 errno = ENOSYS;
887#endif 986#endif
888 987
889 if (res < 0 988 if (res < 0
890 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 989 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
990 /* BSDs */
991#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
992 || errno == ENOTSUP
993#endif
994 || errno == EOPNOTSUPP /* BSDs */
891#if __solaris 995#if __solaris
892 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 996 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
893#endif 997#endif
894 ) 998 )
895 ) 999 )
… …
926 } 1030 }
927 1031
928 return res; 1032 return res;
929} 1033}
930 1034
1035static signed char
1036eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1037{
1038 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1039 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1040}
1041
1042#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1043
1044#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1045#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1046
1047static void
1048eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1049{
1050 unsigned char bits [9 + sizeof (ino_t) * 8];
1051 unsigned char *bit = bits;
1052
1053 assert (CHAR_BIT == 8);
1054 assert (sizeof (eio_dirent) * 8 < 256);
1055 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1056 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1057
1058 if (size <= EIO_SORT_FAST)
1059 return;
1060
1061 /* first prepare an array of bits to test in our radix sort */
1062 /* try to take endianness into account, as well as differences in ino_t sizes */
1063 /* inode_bits must contain all inodes ORed together */
1064 /* which is used to skip bits that are 0 everywhere, which is very common */
1065 {
1066 ino_t endianness;
1067 int i, j;
1068
1069 /* we store the byte offset of byte n into byte n of "endianness" */
1070 for (i = 0; i < sizeof (ino_t); ++i)
1071 ((unsigned char *)&endianness)[i] = i;
1072
1073 *bit++ = 0;
1074
1075 for (i = 0; i < sizeof (ino_t); ++i)
1076 {
1077 /* shifting off the byte offsets out of "endianness" */
1078 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1079 endianness >>= 8;
1080
1081 for (j = 0; j < 8; ++j)
1082 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1083 *bit++ = offs + j;
1084 }
1085
1086 for (j = 0; j < 8; ++j)
1087 if (score_bits & (1 << j))
1088 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1089 }
1090
1091 /* now actually do the sorting (a variant of MSD radix sort) */
1092 {
1093 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1094 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1095 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1096 int stk_idx = 0;
1097
1098 base_stk [stk_idx] = dents;
1099 end_stk [stk_idx] = dents + size;
1100 bit_stk [stk_idx] = bit - 1;
1101
1102 do
1103 {
1104 base = base_stk [stk_idx];
1105 end = end_stk [stk_idx];
1106 bit = bit_stk [stk_idx];
1107
1108 for (;;)
1109 {
1110 unsigned char O = *bit >> 3;
1111 unsigned char M = 1 << (*bit & 7);
1112
1113 eio_dirent *a = base;
1114 eio_dirent *b = end;
1115
1116 if (b - a < EIO_SORT_CUTOFF)
1117 break;
1118
1119 /* now bit-partition the array on the bit */
1120 /* this ugly asymmetric loop seems to perform much better than typical */
1121 /* partition algos found in the literature */
1122 do
1123 if (!(((unsigned char *)a)[O] & M))
1124 ++a;
1125 else if (!(((unsigned char *)--b)[O] & M))
1126 {
1127 eio_dirent tmp = *a; *a = *b; *b = tmp;
1128 ++a;
1129 }
1130 while (b > a);
1131
1132 /* next bit, or stop, if no bits left in this path */
1133 if (!*--bit)
1134 break;
1135
1136 base_stk [stk_idx] = a;
1137 end_stk [stk_idx] = end;
1138 bit_stk [stk_idx] = bit;
1139 ++stk_idx;
1140
1141 end = a;
1142 }
1143 }
1144 while (stk_idx--);
1145 }
1146}
1147
1148static void
1149eio_dent_insertion_sort (eio_dirent *dents, int size)
1150{
1151 /* first move the smallest element to the front, to act as a sentinel */
1152 {
1153 int i;
1154 eio_dirent *min = dents;
1155
1156 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1157 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1158 if (EIO_DENT_CMP (dents [i], <, *min))
1159 min = &dents [i];
1160
1161 /* swap elements 0 and j (minimum) */
1162 {
1163 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1164 }
1165 }
1166
1167 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1168 {
1169 eio_dirent *i, *j;
1170
1171 for (i = dents + 1; i < dents + size; ++i)
1172 {
1173 eio_dirent value = *i;
1174
1175 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1176 j [1] = j [0];
1177
1178 j [1] = value;
1179 }
1180 }
1181}
1182
1183static void
1184eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1185{
1186 if (size <= 1)
1187 return; /* our insertion sort relies on size > 0 */
1188
1189 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1190 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1191 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1192
1193 /* use an insertion sort at the end, or for small arrays, */
1194 /* as insertion sort is more efficient for small partitions */
1195 eio_dent_insertion_sort (dents, size);
1196}
1197
931/* read a full directory */ 1198/* read a full directory */
932static void 1199static void
933eio__scandir (eio_req *req, etp_worker *self) 1200eio__scandir (eio_req *req, etp_worker *self)
934{ 1201{
935 DIR *dirp; 1202 DIR *dirp;
936 union
937 {
938 EIO_STRUCT_DIRENT d;
939 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
940 } *u;
941 EIO_STRUCT_DIRENT *entp; 1203 EIO_STRUCT_DIRENT *entp;
942 char *name, *names; 1204 char *name, *names;
943 int memlen = 4096; 1205 int namesalloc = 4096;
944 int memofs = 0; 1206 int namesoffs = 0;
1207 int flags = req->int1;
1208 eio_dirent *dents = 0;
1209 int dentalloc = 128;
945 int res = 0; 1210 int dentoffs = 0;
1211 ino_t inode_bits = 0;
1212
1213 req->result = -1;
1214
1215 if (!(flags & EIO_READDIR_DENTS))
1216 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
946 1217
947 X_LOCK (wrklock); 1218 X_LOCK (wrklock);
1219 /* the corresponding closedir is in ETP_WORKER_CLEAR */
948 self->dirp = dirp = opendir (req->ptr1); 1220 self->dirp = dirp = opendir (req->ptr1);
949 self->dbuf = u = malloc (sizeof (*u));
950 req->flags |= EIO_FLAG_PTR2_FREE; 1221 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1222 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
951 req->ptr2 = names = malloc (memlen); 1223 req->ptr2 = names = malloc (namesalloc);
952 X_UNLOCK (wrklock); 1224 X_UNLOCK (wrklock);
953 1225
954 if (dirp && u && names) 1226 if (dirp && names && (!flags || dents))
955 for (;;) 1227 for (;;)
956 { 1228 {
957 errno = 0; 1229 errno = 0;
958 readdir_r (dirp, &u->d, &entp); 1230 entp = readdir (dirp);
959 1231
960 if (!entp) 1232 if (!entp)
1233 {
1234 if (errno)
1235 break;
1236
1237 /* sort etc. */
1238 req->int1 = flags;
1239 req->result = dentoffs;
1240
1241 if (flags & EIO_READDIR_STAT_ORDER)
1242 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
1243 else if (flags & EIO_READDIR_DIRS_FIRST)
1244 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1245 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1246 else
1247 {
1248 /* in this case, all is known, and we just put dirs first and sort them */
1249 eio_dirent *oth = dents + dentoffs;
1250 eio_dirent *dir = dents;
1251
1252 /* now partition dirs to the front, and non-dirs to the back */
1253 /* by walking from both sides and swapping if necessary */
1254 /* also clear score, so it doesn't influence sorting */
1255 while (oth > dir)
1256 {
1257 if (dir->type == EIO_DT_DIR)
1258 ++dir;
1259 else if ((--oth)->type == EIO_DT_DIR)
1260 {
1261 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1262
1263 ++dir;
1264 }
1265 }
1266
1267 /* now sort the dirs only */
1268 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1269 }
1270
961 break; 1271 break;
1272 }
962 1273
1274 /* now add the entry to our list(s) */
963 name = entp->d_name; 1275 name = entp->d_name;
964 1276
1277 /* skip . and .. entries */
965 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1278 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
966 { 1279 {
967 int len = strlen (name) + 1; 1280 int len = D_NAMLEN (entp) + 1;
968 1281
969 res++; 1282 while (expect_false (namesoffs + len > namesalloc))
970
971 while (memofs + len > memlen)
972 { 1283 {
973 memlen *= 2; 1284 namesalloc *= 2;
974 X_LOCK (wrklock); 1285 X_LOCK (wrklock);
975 req->ptr2 = names = realloc (names, memlen); 1286 req->ptr2 = names = realloc (names, namesalloc);
976 X_UNLOCK (wrklock); 1287 X_UNLOCK (wrklock);
977 1288
978 if (!names) 1289 if (!names)
979 break; 1290 break;
980 } 1291 }
981 1292
982 memcpy (names + memofs, name, len); 1293 memcpy (names + namesoffs, name, len);
1294
1295 if (dents)
1296 {
1297 struct eio_dirent *ent;
1298
1299 if (expect_false (dentoffs == dentalloc))
1300 {
1301 dentalloc *= 2;
1302 X_LOCK (wrklock);
1303 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1304 X_UNLOCK (wrklock);
1305
1306 if (!dents)
1307 break;
1308 }
1309
1310 ent = dents + dentoffs;
1311
1312 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1313 ent->namelen = len - 1;
1314 ent->inode = D_INO (entp);
1315
1316 inode_bits |= ent->inode;
1317
1318 switch (D_TYPE (entp))
1319 {
1320 default:
1321 ent->type = EIO_DT_UNKNOWN;
1322 flags |= EIO_READDIR_FOUND_UNKNOWN;
1323 break;
1324
1325 #ifdef DT_FIFO
1326 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1327 #endif
1328 #ifdef DT_CHR
1329 case DT_CHR: ent->type = EIO_DT_CHR; break;
1330 #endif
1331 #ifdef DT_MPC
1332 case DT_MPC: ent->type = EIO_DT_MPC; break;
1333 #endif
1334 #ifdef DT_DIR
1335 case DT_DIR: ent->type = EIO_DT_DIR; break;
1336 #endif
1337 #ifdef DT_NAM
1338 case DT_NAM: ent->type = EIO_DT_NAM; break;
1339 #endif
1340 #ifdef DT_BLK
1341 case DT_BLK: ent->type = EIO_DT_BLK; break;
1342 #endif
1343 #ifdef DT_MPB
1344 case DT_MPB: ent->type = EIO_DT_MPB; break;
1345 #endif
1346 #ifdef DT_REG
1347 case DT_REG: ent->type = EIO_DT_REG; break;
1348 #endif
1349 #ifdef DT_NWK
1350 case DT_NWK: ent->type = EIO_DT_NWK; break;
1351 #endif
1352 #ifdef DT_CMP
1353 case DT_CMP: ent->type = EIO_DT_CMP; break;
1354 #endif
1355 #ifdef DT_LNK
1356 case DT_LNK: ent->type = EIO_DT_LNK; break;
1357 #endif
1358 #ifdef DT_SOCK
1359 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1360 #endif
1361 #ifdef DT_DOOR
1362 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1363 #endif
1364 #ifdef DT_WHT
1365 case DT_WHT: ent->type = EIO_DT_WHT; break;
1366 #endif
1367 }
1368
1369 ent->score = 7;
1370
1371 if (flags & EIO_READDIR_DIRS_FIRST)
1372 {
1373 if (ent->type == EIO_DT_UNKNOWN)
1374 {
1375 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1376 ent->score = 1;
1377 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1378 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1379 }
1380 else if (ent->type == EIO_DT_DIR)
1381 ent->score = 0;
1382 }
1383 }
1384
983 memofs += len; 1385 namesoffs += len;
1386 ++dentoffs;
1387 }
1388
1389 if (EIO_CANCELLED (req))
1390 {
1391 errno = ECANCELED;
1392 break;
984 } 1393 }
985 } 1394 }
1395}
986 1396
987 if (errno) 1397#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
988 res = -1; 1398# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
989 1399#else
990 req->result = res; 1400
1401int
1402eio__msync (void *mem, size_t len, int flags)
1403{
1404 if (EIO_MS_ASYNC != MS_SYNC
1405 || EIO_MS_INVALIDATE != MS_INVALIDATE
1406 || EIO_MS_SYNC != MS_SYNC)
1407 {
1408 flags = 0
1409 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1410 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1411 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1412 }
1413
1414 return msync (mem, len, flags);
1415}
1416
1417#endif
1418
1419int
1420eio__mtouch (void *mem, size_t len, int flags)
1421{
1422 intptr_t addr = (intptr_t)mem;
1423 intptr_t end = addr + len;
1424#ifdef PAGESIZE
1425 const intptr_t page = PAGESIZE;
1426#else
1427 static intptr_t page;
1428
1429 if (!page)
1430 page = sysconf (_SC_PAGESIZE);
1431#endif
1432
1433 /* round down to start of page, although this is probably useless */
1434 addr &= ~(page - 1); /* assume page size is always a power of two */
1435
1436 if (addr < end)
1437 if (flags & EIO_MT_MODIFY) /* modify */
1438 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1439 else
1440 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1441
1442 return 0;
991} 1443}
992 1444
993/*****************************************************************************/ 1445/*****************************************************************************/
994 1446
995#define ALLOC(len) \ 1447#define ALLOC(len) \
… …
1082 1534
1083/*****************************************************************************/ 1535/*****************************************************************************/
1084 1536
1085int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1537int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1086{ 1538{
1087 etp_init (want_poll, done_poll); 1539 return etp_init (want_poll, done_poll);
1088} 1540}
1089 1541
1090static void eio_api_destroy (eio_req *req) 1542static void eio_api_destroy (eio_req *req)
1091{ 1543{
1092 free (req); 1544 free (req);
… …
1138 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1590 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1139 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1591 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1140 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1592 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1141 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1593 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1142 1594
1595 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1596 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1597 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1598 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1599
1143 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1600 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1144 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1601 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1145 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1602 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1146 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1603 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1147 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1604 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
… …
1154 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1611 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1155 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1612 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1156 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1613 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1157 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1614 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1158 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1615 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1159 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; 1616 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1160 1617
1161 case EIO_READLINK: ALLOC (NAME_MAX); 1618 case EIO_READLINK: ALLOC (PATH_MAX);
1162 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1619 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1163 1620
1164 case EIO_SYNC: req->result = 0; sync (); break; 1621 case EIO_SYNC: req->result = 0; sync (); break;
1165 case EIO_FSYNC: req->result = fsync (req->int1); break; 1622 case EIO_FSYNC: req->result = fsync (req->int1); break;
1166 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1623 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1624 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1625 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1626 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1167 1627
1168 case EIO_READDIR: eio__scandir (req, self); break; 1628 case EIO_READDIR: eio__scandir (req, self); break;
1169 1629
1170 case EIO_BUSY: 1630 case EIO_BUSY:
1171#ifdef _WIN32 1631#ifdef _WIN32
… …
1203 1663
1204 req->result = req->type == EIO_FUTIME 1664 req->result = req->type == EIO_FUTIME
1205 ? futimes (req->int1, times) 1665 ? futimes (req->int1, times)
1206 : utimes (req->ptr1, times); 1666 : utimes (req->ptr1, times);
1207 } 1667 }
1668 break;
1208 1669
1209 case EIO_GROUP: 1670 case EIO_GROUP:
1671 abort (); /* handled in eio_request */
1672
1210 case EIO_NOP: 1673 case EIO_NOP:
1211 req->result = 0; 1674 req->result = 0;
1212 break; 1675 break;
1213 1676
1214 case EIO_CUSTOM: 1677 case EIO_CUSTOM:
1215 req->feed (req); 1678 ((void (*)(eio_req *))req->feed) (req);
1216 break; 1679 break;
1217 1680
1218 default: 1681 default:
1219 req->result = -1; 1682 req->result = -1;
1220 break; 1683 break;
… …
1243eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 1706eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1244{ 1707{
1245 REQ (EIO_FSYNC); req->int1 = fd; SEND; 1708 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1246} 1709}
1247 1710
1711eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1712{
1713 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1714}
1715
1716eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1717{
1718 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1719}
1720
1721eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1722{
1723 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1724}
1725
1248eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 1726eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1249{ 1727{
1250 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1728 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1251} 1729}
1252 1730
… …
1271} 1749}
1272 1750
1273eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1751eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1274{ 1752{
1275 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1753 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1754}
1755
1756eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1757{
1758 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1276} 1759}
1277 1760
1278eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1761eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1279{ 1762{
1280 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1763 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
… …
1354eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 1837eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1355{ 1838{
1356 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1839 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1357} 1840}
1358 1841
1842eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1843{
1844 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1845}
1846
1359eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1847eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1360{ 1848{
1361 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1849 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1362} 1850}
1363 1851
1364eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1852eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1365{ 1853{
1366 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1854 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1367} 1855}
1368 1856
1369eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1857eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1370{ 1858{
1371 return eio__1path (EIO_READDIR, path, pri, cb, data); 1859 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1372} 1860}
1373 1861
1374eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1862eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1375{ 1863{
1376 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1864 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1377} 1865}
1378 1866
1379static eio_req * 1867static eio_req *
1380eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1868eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1381{ 1869{
… …
1407 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 1895 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1408} 1896}
1409 1897
1410eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 1898eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1411{ 1899{
1412 REQ (EIO_CUSTOM); req->feed = execute; SEND; 1900 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1413} 1901}
1414 1902
1415#endif 1903#endif
1416 1904
1417eio_req *eio_grp (eio_cb cb, void *data) 1905eio_req *eio_grp (eio_cb cb, void *data)
… …
1445 1933
1446void eio_grp_add (eio_req *grp, eio_req *req) 1934void eio_grp_add (eio_req *grp, eio_req *req)
1447{ 1935{
1448 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1936 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1449 1937
1938 grp->flags |= EIO_FLAG_GROUPADD;
1939
1450 ++grp->size; 1940 ++grp->size;
1451 req->grp = grp; 1941 req->grp = grp;
1452 1942
1453 req->grp_prev = 0; 1943 req->grp_prev = 0;
1454 req->grp_next = grp->grp_first; 1944 req->grp_next = grp->grp_first;
… …
1463/* misc garbage */ 1953/* misc garbage */
1464 1954
1465ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 1955ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1466{ 1956{
1467 etp_worker wrk; 1957 etp_worker wrk;
1958 ssize_t ret;
1468 1959
1469 wrk.dbuf = 0; 1960 wrk.dbuf = 0;
1470 1961
1471 eio__sendfile (ofd, ifd, offset, count, &wrk); 1962 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1472 1963
1473 if (wrk.dbuf) 1964 if (wrk.dbuf)
1474 free (wrk.dbuf); 1965 free (wrk.dbuf);
1475}
1476 1966
1967 return ret;
1968}
1969

Diff Legend

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