ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libeio/eio.c
(Generate patch)

Comparing libeio/eio.c (file contents):
Revision 1.27 by root, Wed Oct 22 18:15:36 2008 UTC vs.
Revision 1.70 by root, Fri Jun 10 06:55:10 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
… …
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45
46#ifdef EIO_STACKSIZE
47# define XTHREAD_STACKSIZE EIO_STACKSIZE
48#endif
41#include "xthread.h" 49#include "xthread.h"
42 50
43#include <errno.h> 51#include <errno.h>
44#include <stddef.h> 52#include <stddef.h>
45#include <stdlib.h> 53#include <stdlib.h>
46#include <string.h> 54#include <string.h>
47#include <errno.h> 55#include <errno.h>
48#include <sys/types.h> 56#include <sys/types.h>
49#include <sys/stat.h> 57#include <sys/stat.h>
58#include <sys/statvfs.h>
50#include <limits.h> 59#include <limits.h>
51#include <fcntl.h> 60#include <fcntl.h>
52#include <assert.h> 61#include <assert.h>
53 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes form stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
54#ifndef EIO_FINISH 69#ifndef EIO_FINISH
55# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 70# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
56#endif 71#endif
57 72
58#ifndef EIO_DESTROY 73#ifndef EIO_DESTROY
… …
64#endif 79#endif
65 80
66#ifdef _WIN32 81#ifdef _WIN32
67 82
68 /*doh*/ 83 /*doh*/
69
70#else 84#else
71 85
72# include "config.h"
73# include <sys/time.h> 86# include <sys/time.h>
74# include <sys/select.h> 87# include <sys/select.h>
75# include <unistd.h> 88# include <unistd.h>
76# include <utime.h> 89# include <utime.h>
77# include <signal.h> 90# include <signal.h>
78# include <dirent.h> 91# include <dirent.h>
79 92
93#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94# include <sys/mman.h>
95#endif
96
97/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99# define _DIRENT_HAVE_D_TYPE /* sigh */
100# define D_INO(de) (de)->d_fileno
101# define D_NAMLEN(de) (de)->d_namlen
102# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103# define D_INO(de) (de)->d_ino
104# endif
105
106#ifdef _D_EXACT_NAMLEN
107# undef D_NAMLEN
108# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109#endif
110
111# ifdef _DIRENT_HAVE_D_TYPE
112# define D_TYPE(de) (de)->d_type
113# endif
114
80# ifndef EIO_STRUCT_DIRENT 115# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 116# define EIO_STRUCT_DIRENT struct dirent
82# endif 117# endif
83 118
84#endif 119#endif
85 120
86#if HAVE_SENDFILE 121#if HAVE_SENDFILE
87# if __linux 122# if __linux
88# include <sys/sendfile.h> 123# include <sys/sendfile.h>
89# elif __freebsd 124# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 125# include <sys/socket.h>
91# include <sys/uio.h> 126# include <sys/uio.h>
92# elif __hpux 127# elif __hpux
93# include <sys/socket.h> 128# include <sys/socket.h>
94# elif __solaris /* not yet */ 129# elif __solaris
95# include <sys/sendfile.h> 130# include <sys/sendfile.h>
96# else 131# else
97# error sendfile support requested but not available 132# error sendfile support requested but not available
98# endif 133# endif
99#endif 134#endif
100 135
101/* number of seconds after which an idle threads exit */ 136#ifndef D_TYPE
102#define IDLE_TIMEOUT 10 137# define D_TYPE(de) 0
138#endif
139#ifndef D_INO
140# define D_INO(de) 0
141#endif
142#ifndef D_NAMLEN
143# define D_NAMLEN(de) strlen ((de)->d_name)
144#endif
103 145
104/* used for struct dirent, AIX doesn't provide it */ 146/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 147#ifndef NAME_MAX
106# define NAME_MAX 4096 148# define NAME_MAX 4096
149#endif
150
151/* used for readlink etc. */
152#ifndef PATH_MAX
153# define PATH_MAX 4096
107#endif 154#endif
108 155
109/* buffer size for various temporary buffers */ 156/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 157#define EIO_BUFSIZE 65536
111 158
… …
155 if (wrk->dirp) \ 202 if (wrk->dirp) \
156 { \ 203 { \
157 closedir (wrk->dirp); \ 204 closedir (wrk->dirp); \
158 wrk->dirp = 0; \ 205 wrk->dirp = 0; \
159 } 206 }
207
160#define ETP_WORKER_COMMON \ 208#define ETP_WORKER_COMMON \
161 void *dbuf; \ 209 void *dbuf; \
162 DIR *dirp; 210 DIR *dirp;
163 211
164/*****************************************************************************/ 212/*****************************************************************************/
165 213
166#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 214#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
167 215
168/* calculcate time difference in ~1/EIO_TICKS of a second */ 216/* calculate time difference in ~1/EIO_TICKS of a second */
169static int tvdiff (struct timeval *tv1, struct timeval *tv2) 217static int tvdiff (struct timeval *tv1, struct timeval *tv2)
170{ 218{
171 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 219 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
172 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 220 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
173} 221}
… …
181static unsigned int max_poll_reqs; /* reslock */ 229static unsigned int max_poll_reqs; /* reslock */
182 230
183static volatile unsigned int nreqs; /* reqlock */ 231static volatile unsigned int nreqs; /* reqlock */
184static volatile unsigned int nready; /* reqlock */ 232static volatile unsigned int nready; /* reqlock */
185static volatile unsigned int npending; /* reqlock */ 233static volatile unsigned int npending; /* reqlock */
186static volatile unsigned int max_idle = 4; 234static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
235static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
187 236
188static mutex_t wrklock = X_MUTEX_INIT; 237static xmutex_t wrklock;
189static mutex_t reslock = X_MUTEX_INIT; 238static xmutex_t reslock;
190static mutex_t reqlock = X_MUTEX_INIT; 239static xmutex_t reqlock;
191static cond_t reqwait = X_COND_INIT; 240static xcond_t reqwait;
192 241
193#if !HAVE_PREADWRITE 242#if !HAVE_PREADWRITE
194/* 243/*
195 * make our pread/pwrite emulation safe against themselves, but not against 244 * make our pread/pwrite emulation safe against themselves, but not against
196 * normal read/write by using a mutex. slows down execution a lot, 245 * normal read/write by using a mutex. slows down execution a lot,
197 * but that's your problem, not mine. 246 * but that's your problem, not mine.
198 */ 247 */
199static mutex_t preadwritelock = X_MUTEX_INIT; 248static xmutex_t preadwritelock = X_MUTEX_INIT;
200#endif 249#endif
201 250
202typedef struct etp_worker 251typedef struct etp_worker
203{ 252{
204 /* locked by wrklock */ 253 /* locked by wrklock */
205 struct etp_worker *prev, *next; 254 struct etp_worker *prev, *next;
206 255
207 thread_t tid; 256 xthread_t tid;
208 257
209 /* locked by reslock, reqlock or wrklock */ 258 /* locked by reslock, reqlock or wrklock */
210 ETP_REQ *req; /* currently processed request */ 259 ETP_REQ *req; /* currently processed request */
211 260
212 ETP_WORKER_COMMON 261 ETP_WORKER_COMMON
… …
326 } 375 }
327 376
328 abort (); 377 abort ();
329} 378}
330 379
380static void etp_thread_init (void)
381{
382 X_MUTEX_CREATE (wrklock);
383 X_MUTEX_CREATE (reslock);
384 X_MUTEX_CREATE (reqlock);
385 X_COND_CREATE (reqwait);
386}
387
331static void etp_atfork_prepare (void) 388static void etp_atfork_prepare (void)
332{ 389{
333 X_LOCK (wrklock); 390 X_LOCK (wrklock);
334 X_LOCK (reqlock); 391 X_LOCK (reqlock);
335 X_LOCK (reslock); 392 X_LOCK (reslock);
… …
373 idle = 0; 430 idle = 0;
374 nreqs = 0; 431 nreqs = 0;
375 nready = 0; 432 nready = 0;
376 npending = 0; 433 npending = 0;
377 434
378 etp_atfork_parent (); 435 etp_thread_init ();
379} 436}
380 437
381static void 438static void
382etp_once_init (void) 439etp_once_init (void)
383{ 440{
441 etp_thread_init ();
384 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 442 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
385} 443}
386 444
387static int 445static int
388etp_init (void (*want_poll)(void), void (*done_poll)(void)) 446etp_init (void (*want_poll)(void), void (*done_poll)(void))
… …
565} 623}
566 624
567static void etp_set_max_poll_time (double nseconds) 625static void etp_set_max_poll_time (double nseconds)
568{ 626{
569 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 627 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
570 max_poll_time = nseconds; 628 max_poll_time = nseconds * EIO_TICKS;
571 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 629 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
572} 630}
573 631
574static void etp_set_max_poll_reqs (unsigned int maxreqs) 632static void etp_set_max_poll_reqs (unsigned int maxreqs)
575{ 633{
… …
579} 637}
580 638
581static void etp_set_max_idle (unsigned int nthreads) 639static void etp_set_max_idle (unsigned int nthreads)
582{ 640{
583 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 641 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
584 max_idle = nthreads <= 0 ? 1 : nthreads; 642 max_idle = nthreads;
643 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644}
645
646static void etp_set_idle_timeout (unsigned int seconds)
647{
648 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649 idle_timeout = seconds;
585 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 650 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
586} 651}
587 652
588static void etp_set_min_parallel (unsigned int nthreads) 653static void etp_set_min_parallel (unsigned int nthreads)
589{ 654{
… …
717void eio_set_max_idle (unsigned int nthreads) 782void eio_set_max_idle (unsigned int nthreads)
718{ 783{
719 etp_set_max_idle (nthreads); 784 etp_set_max_idle (nthreads);
720} 785}
721 786
787void eio_set_idle_timeout (unsigned int seconds)
788{
789 etp_set_idle_timeout (seconds);
790}
791
722void eio_set_min_parallel (unsigned int nthreads) 792void eio_set_min_parallel (unsigned int nthreads)
723{ 793{
724 etp_set_min_parallel (nthreads); 794 etp_set_min_parallel (nthreads);
725} 795}
726 796
… …
767 837
768 X_LOCK (preadwritelock); 838 X_LOCK (preadwritelock);
769 ooffset = lseek (fd, 0, SEEK_CUR); 839 ooffset = lseek (fd, 0, SEEK_CUR);
770 lseek (fd, offset, SEEK_SET); 840 lseek (fd, offset, SEEK_SET);
771 res = write (fd, buf, count); 841 res = write (fd, buf, count);
772 lseek (fd, offset, SEEK_SET); 842 lseek (fd, ooffset, SEEK_SET);
773 X_UNLOCK (preadwritelock); 843 X_UNLOCK (preadwritelock);
774 844
775 return res; 845 return res;
776} 846}
777#endif 847#endif
778 848
779#ifndef HAVE_FUTIMES 849#ifndef HAVE_UTIMES
780 850
781# undef utimes 851# undef utimes
782# undef futimes
783# define utimes(path,times) eio__utimes (path, times) 852# define utimes(path,times) eio__utimes (path, times)
784# define futimes(fd,times) eio__futimes (fd, times)
785 853
786static int 854static int
787eio__utimes (const char *filename, const struct timeval times[2]) 855eio__utimes (const char *filename, const struct timeval times[2])
788{ 856{
789 if (times) 857 if (times)
… …
796 return utime (filename, &buf); 864 return utime (filename, &buf);
797 } 865 }
798 else 866 else
799 return utime (filename, 0); 867 return utime (filename, 0);
800} 868}
869
870#endif
871
872#ifndef HAVE_FUTIMES
873
874# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times)
801 876
802static int eio__futimes (int fd, const struct timeval tv[2]) 877static int eio__futimes (int fd, const struct timeval tv[2])
803{ 878{
804 errno = ENOSYS; 879 errno = ENOSYS;
805 return -1; 880 return -1;
… …
829 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); 904 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
830 } 905 }
831 906
832 res = sync_file_range (fd, offset, nbytes, flags); 907 res = sync_file_range (fd, offset, nbytes, flags);
833 908
834 if (res != ENOSYS) 909 if (!res || errno != ENOSYS)
835 return res; 910 return res;
836#endif 911#endif
837 912
838 /* even though we could play tricks with the flags, it's better to always 913 /* even though we could play tricks with the flags, it's better to always
839 * call fdatasync, as thta matches the expectation of it's users best */ 914 * call fdatasync, as that matches the expectation of its users best */
840 return fdatasync (fd); 915 return fdatasync (fd);
841} 916}
842 917
843#if !HAVE_READAHEAD 918#if !HAVE_READAHEAD
844# undef readahead 919# undef readahead
… …
867 942
868/* sendfile always needs emulation */ 943/* sendfile always needs emulation */
869static ssize_t 944static ssize_t
870eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 945eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
871{ 946{
947 ssize_t written = 0;
872 ssize_t res; 948 ssize_t res;
873 949
874 if (!count) 950 if (!count)
875 return 0; 951 return 0;
876 952
953 for (;;)
954 {
877#if HAVE_SENDFILE 955#if HAVE_SENDFILE
878# if __linux 956# if __linux
957 off_t soffset = offset;
879 res = sendfile (ofd, ifd, &offset, count); 958 res = sendfile (ofd, ifd, &soffset, count);
880 959
881# elif __freebsd 960# elif __FreeBSD__
882 /* 961 /*
883 * Of course, the freebsd sendfile is a dire hack with no thoughts 962 * Of course, the freebsd sendfile is a dire hack with no thoughts
884 * wasted on making it similar to other I/O functions. 963 * wasted on making it similar to other I/O functions.
885 */ 964 */
886 {
887 off_t sbytes; 965 off_t sbytes;
888 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 966 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
889 967
890 if (res < 0 && sbytes) 968 #if 0 /* according to the manpage, this is correct, but broken behaviour */
891 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 969 /* freebsd' sendfile will return 0 on success */
970 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
971 /* not on e.g. EIO or EPIPE - sounds broken */
972 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
892 res = sbytes; 973 res = sbytes;
893 } 974 #endif
975
976 /* according to source inspection, this is correct, and useful behaviour */
977 if (sbytes)
978 res = sbytes;
979
980# elif defined (__APPLE__)
981 off_t sbytes = count;
982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983
984 /* according to the manpage, sbytes is always valid */
985 if (sbytes)
986 res = sbytes;
894 987
895# elif __hpux 988# elif __hpux
896 res = sendfile (ofd, ifd, offset, count, 0, 0); 989 res = sendfile (ofd, ifd, offset, count, 0, 0);
897 990
898# elif __solaris 991# elif __solaris
899 {
900 struct sendfilevec vec; 992 struct sendfilevec vec;
901 size_t sbytes; 993 size_t sbytes;
902 994
903 vec.sfv_fd = ifd; 995 vec.sfv_fd = ifd;
904 vec.sfv_flag = 0; 996 vec.sfv_flag = 0;
905 vec.sfv_off = offset; 997 vec.sfv_off = offset;
906 vec.sfv_len = count; 998 vec.sfv_len = count;
907 999
908 res = sendfilev (ofd, &vec, 1, &sbytes); 1000 res = sendfilev (ofd, &vec, 1, &sbytes);
909 1001
910 if (res < 0 && sbytes) 1002 if (res < 0 && sbytes)
911 res = sbytes; 1003 res = sbytes;
912 }
913 1004
914# endif 1005# endif
1006
1007#elif defined (_WIN32)
1008 /* does not work, just for documentation of what would need to be done */
1009 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1010 /* libeio guarantees that the file offset does not change, and windows */
1011 /* has no way to get an independent handle to the same file description */
1012 HANDLE h = TO_SOCKET (ifd);
1013 SetFilePointer (h, offset, 0, FILE_BEGIN);
1014 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1015
915#else 1016#else
916 res = -1; 1017 res = -1;
917 errno = ENOSYS; 1018 errno = ENOSYS;
918#endif 1019#endif
919 1020
1021 /* we assume sendfile can copy at least 128mb in one go */
1022 if (res <= 128 * 1024 * 1024)
1023 {
1024 if (res > 0)
1025 written += res;
1026
1027 if (written)
1028 return written;
1029
1030 break;
1031 }
1032 else
1033 {
1034 /* if we requested more, then probably the kernel was lazy */
1035 written += res;
1036 offset += res;
1037 count -= res;
1038
1039 if (!count)
1040 return written;
1041 }
1042 }
1043
920 if (res < 0 1044 if (res < 0
921 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1045 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1046 /* BSDs */
1047#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1048 || errno == ENOTSUP
1049#endif
1050 || errno == EOPNOTSUPP /* BSDs */
922#if __solaris 1051#if __solaris
923 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1052 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
924#endif 1053#endif
925 ) 1054 )
926 ) 1055 )
… …
957 } 1086 }
958 1087
959 return res; 1088 return res;
960} 1089}
961 1090
1091static signed char
1092eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1093{
1094 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1095 : a->inode < b->inode ? -1
1096 : a->inode > b->inode ? 1
1097 : 0;
1098}
1099
1100#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1101
1102#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1103#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1104
1105static void
1106eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1107{
1108 unsigned char bits [9 + sizeof (ino_t) * 8];
1109 unsigned char *bit = bits;
1110
1111 assert (CHAR_BIT == 8);
1112 assert (sizeof (eio_dirent) * 8 < 256);
1113 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1114 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1115
1116 if (size <= EIO_SORT_FAST)
1117 return;
1118
1119 /* first prepare an array of bits to test in our radix sort */
1120 /* try to take endianness into account, as well as differences in ino_t sizes */
1121 /* inode_bits must contain all inodes ORed together */
1122 /* which is used to skip bits that are 0 everywhere, which is very common */
1123 {
1124 ino_t endianness;
1125 int i, j;
1126
1127 /* we store the byte offset of byte n into byte n of "endianness" */
1128 for (i = 0; i < sizeof (ino_t); ++i)
1129 ((unsigned char *)&endianness)[i] = i;
1130
1131 *bit++ = 0;
1132
1133 for (i = 0; i < sizeof (ino_t); ++i)
1134 {
1135 /* shifting off the byte offsets out of "endianness" */
1136 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1137 endianness >>= 8;
1138
1139 for (j = 0; j < 8; ++j)
1140 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1141 *bit++ = offs + j;
1142 }
1143
1144 for (j = 0; j < 8; ++j)
1145 if (score_bits & (1 << j))
1146 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1147 }
1148
1149 /* now actually do the sorting (a variant of MSD radix sort) */
1150 {
1151 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1152 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1153 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1154 int stk_idx = 0;
1155
1156 base_stk [stk_idx] = dents;
1157 end_stk [stk_idx] = dents + size;
1158 bit_stk [stk_idx] = bit - 1;
1159
1160 do
1161 {
1162 base = base_stk [stk_idx];
1163 end = end_stk [stk_idx];
1164 bit = bit_stk [stk_idx];
1165
1166 for (;;)
1167 {
1168 unsigned char O = *bit >> 3;
1169 unsigned char M = 1 << (*bit & 7);
1170
1171 eio_dirent *a = base;
1172 eio_dirent *b = end;
1173
1174 if (b - a < EIO_SORT_CUTOFF)
1175 break;
1176
1177 /* now bit-partition the array on the bit */
1178 /* this ugly asymmetric loop seems to perform much better than typical */
1179 /* partition algos found in the literature */
1180 do
1181 if (!(((unsigned char *)a)[O] & M))
1182 ++a;
1183 else if (!(((unsigned char *)--b)[O] & M))
1184 {
1185 eio_dirent tmp = *a; *a = *b; *b = tmp;
1186 ++a;
1187 }
1188 while (b > a);
1189
1190 /* next bit, or stop, if no bits left in this path */
1191 if (!*--bit)
1192 break;
1193
1194 base_stk [stk_idx] = a;
1195 end_stk [stk_idx] = end;
1196 bit_stk [stk_idx] = bit;
1197 ++stk_idx;
1198
1199 end = a;
1200 }
1201 }
1202 while (stk_idx--);
1203 }
1204}
1205
1206static void
1207eio_dent_insertion_sort (eio_dirent *dents, int size)
1208{
1209 /* first move the smallest element to the front, to act as a sentinel */
1210 {
1211 int i;
1212 eio_dirent *min = dents;
1213
1214 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1215 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1216 if (EIO_DENT_CMP (dents [i], <, *min))
1217 min = &dents [i];
1218
1219 /* swap elements 0 and j (minimum) */
1220 {
1221 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1222 }
1223 }
1224
1225 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1226 {
1227 eio_dirent *i, *j;
1228
1229 for (i = dents + 1; i < dents + size; ++i)
1230 {
1231 eio_dirent value = *i;
1232
1233 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1234 j [1] = j [0];
1235
1236 j [1] = value;
1237 }
1238 }
1239}
1240
1241static void
1242eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1243{
1244 if (size <= 1)
1245 return; /* our insertion sort relies on size > 0 */
1246
1247 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1248 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1249 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1250
1251 /* use an insertion sort at the end, or for small arrays, */
1252 /* as insertion sort is more efficient for small partitions */
1253 eio_dent_insertion_sort (dents, size);
1254}
1255
962/* read a full directory */ 1256/* read a full directory */
963static void 1257static void
964eio__scandir (eio_req *req, etp_worker *self) 1258eio__scandir (eio_req *req, etp_worker *self)
965{ 1259{
966 DIR *dirp; 1260 DIR *dirp;
967 EIO_STRUCT_DIRENT *entp; 1261 EIO_STRUCT_DIRENT *entp;
968 char *name, *names; 1262 char *name, *names;
969 int memlen = 4096; 1263 int namesalloc = 4096;
970 int memofs = 0; 1264 int namesoffs = 0;
1265 int flags = req->int1;
1266 eio_dirent *dents = 0;
1267 int dentalloc = 128;
971 int res = 0; 1268 int dentoffs = 0;
1269 ino_t inode_bits = 0;
1270
1271 req->result = -1;
1272
1273 if (!(flags & EIO_READDIR_DENTS))
1274 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
972 1275
973 X_LOCK (wrklock); 1276 X_LOCK (wrklock);
974 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1277 /* the corresponding closedir is in ETP_WORKER_CLEAR */
975 self->dirp = dirp = opendir (req->ptr1); 1278 self->dirp = dirp = opendir (req->ptr1);
1279
976 req->flags |= EIO_FLAG_PTR2_FREE; 1280 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1281 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
977 req->ptr2 = names = malloc (memlen); 1282 req->ptr2 = names = malloc (namesalloc);
978 X_UNLOCK (wrklock); 1283 X_UNLOCK (wrklock);
979 1284
980 if (dirp && names) 1285 if (dirp && names && (!flags || dents))
981 for (;;) 1286 for (;;)
982 { 1287 {
983 errno = 0; 1288 errno = 0;
984 entp = readdir (dirp); 1289 entp = readdir (dirp);
985 1290
986 if (!entp) 1291 if (!entp)
1292 {
1293 if (errno)
1294 break;
1295
1296 /* sort etc. */
1297 req->int1 = flags;
1298 req->result = dentoffs;
1299
1300 if (flags & EIO_READDIR_STAT_ORDER)
1301 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1302 else if (flags & EIO_READDIR_DIRS_FIRST)
1303 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1304 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1305 else
1306 {
1307 /* in this case, all is known, and we just put dirs first and sort them */
1308 eio_dirent *oth = dents + dentoffs;
1309 eio_dirent *dir = dents;
1310
1311 /* now partition dirs to the front, and non-dirs to the back */
1312 /* by walking from both sides and swapping if necessary */
1313 while (oth > dir)
1314 {
1315 if (dir->type == EIO_DT_DIR)
1316 ++dir;
1317 else if ((--oth)->type == EIO_DT_DIR)
1318 {
1319 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1320
1321 ++dir;
1322 }
1323 }
1324
1325 /* now sort the dirs only (dirs all have the same score) */
1326 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1327 }
1328
987 break; 1329 break;
1330 }
988 1331
1332 /* now add the entry to our list(s) */
989 name = entp->d_name; 1333 name = entp->d_name;
990 1334
1335 /* skip . and .. entries */
991 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1336 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
992 { 1337 {
993 int len = strlen (name) + 1; 1338 int len = D_NAMLEN (entp) + 1;
994 1339
995 res++; 1340 while (expect_false (namesoffs + len > namesalloc))
996
997 while (memofs + len > memlen)
998 { 1341 {
999 memlen *= 2; 1342 namesalloc *= 2;
1000 X_LOCK (wrklock); 1343 X_LOCK (wrklock);
1001 req->ptr2 = names = realloc (names, memlen); 1344 req->ptr2 = names = realloc (names, namesalloc);
1002 X_UNLOCK (wrklock); 1345 X_UNLOCK (wrklock);
1003 1346
1004 if (!names) 1347 if (!names)
1005 break; 1348 break;
1006 } 1349 }
1007 1350
1008 memcpy (names + memofs, name, len); 1351 memcpy (names + namesoffs, name, len);
1352
1353 if (dents)
1354 {
1355 struct eio_dirent *ent;
1356
1357 if (expect_false (dentoffs == dentalloc))
1358 {
1359 dentalloc *= 2;
1360 X_LOCK (wrklock);
1361 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1362 X_UNLOCK (wrklock);
1363
1364 if (!dents)
1365 break;
1366 }
1367
1368 ent = dents + dentoffs;
1369
1370 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1371 ent->namelen = len - 1;
1372 ent->inode = D_INO (entp);
1373
1374 inode_bits |= ent->inode;
1375
1376 switch (D_TYPE (entp))
1377 {
1378 default:
1379 ent->type = EIO_DT_UNKNOWN;
1380 flags |= EIO_READDIR_FOUND_UNKNOWN;
1381 break;
1382
1383 #ifdef DT_FIFO
1384 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1385 #endif
1386 #ifdef DT_CHR
1387 case DT_CHR: ent->type = EIO_DT_CHR; break;
1388 #endif
1389 #ifdef DT_MPC
1390 case DT_MPC: ent->type = EIO_DT_MPC; break;
1391 #endif
1392 #ifdef DT_DIR
1393 case DT_DIR: ent->type = EIO_DT_DIR; break;
1394 #endif
1395 #ifdef DT_NAM
1396 case DT_NAM: ent->type = EIO_DT_NAM; break;
1397 #endif
1398 #ifdef DT_BLK
1399 case DT_BLK: ent->type = EIO_DT_BLK; break;
1400 #endif
1401 #ifdef DT_MPB
1402 case DT_MPB: ent->type = EIO_DT_MPB; break;
1403 #endif
1404 #ifdef DT_REG
1405 case DT_REG: ent->type = EIO_DT_REG; break;
1406 #endif
1407 #ifdef DT_NWK
1408 case DT_NWK: ent->type = EIO_DT_NWK; break;
1409 #endif
1410 #ifdef DT_CMP
1411 case DT_CMP: ent->type = EIO_DT_CMP; break;
1412 #endif
1413 #ifdef DT_LNK
1414 case DT_LNK: ent->type = EIO_DT_LNK; break;
1415 #endif
1416 #ifdef DT_SOCK
1417 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1418 #endif
1419 #ifdef DT_DOOR
1420 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1421 #endif
1422 #ifdef DT_WHT
1423 case DT_WHT: ent->type = EIO_DT_WHT; break;
1424 #endif
1425 }
1426
1427 ent->score = 7;
1428
1429 if (flags & EIO_READDIR_DIRS_FIRST)
1430 {
1431 if (ent->type == EIO_DT_UNKNOWN)
1432 {
1433 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1434 ent->score = 1;
1435 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1436 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1437 }
1438 else if (ent->type == EIO_DT_DIR)
1439 ent->score = 0;
1440 }
1441 }
1442
1009 memofs += len; 1443 namesoffs += len;
1444 ++dentoffs;
1445 }
1446
1447 if (EIO_CANCELLED (req))
1448 {
1449 errno = ECANCELED;
1450 break;
1010 } 1451 }
1011 } 1452 }
1012
1013 if (errno)
1014 res = -1;
1015
1016 req->result = res;
1017} 1453}
1018 1454
1019#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1020# undef msync
1021# define msync(a,b,c) ENOSYS
1022#endif
1023
1024int
1025eio__mtouch (void *mem, size_t len, int flags)
1026{
1027 intptr_t addr = (intptr_t)mem;
1028 intptr_t end = addr + len;
1029#ifdef PAGESIZE 1455#ifdef PAGESIZE
1030 const intptr_t page = PAGESIZE; 1456# define eio_pagesize() PAGESIZE
1031#else 1457#else
1458static intptr_t
1459eio_pagesize (void)
1460{
1032 static intptr_t page; 1461 static intptr_t page;
1033 1462
1034 if (!page) 1463 if (!page)
1035 page = sysconf (_SC_PAGESIZE); 1464 page = sysconf (_SC_PAGESIZE);
1465
1466 return page;
1467}
1468#endif
1469
1470static void
1471eio_page_align (void **addr, size_t *length)
1472{
1473 intptr_t mask = eio_pagesize () - 1;
1474
1475 /* round down addr */
1476 intptr_t adj = mask & (intptr_t)*addr;
1477
1478 *addr = (void *)((intptr_t)*addr - adj);
1479 *length += adj;
1480
1481 /* round up length */
1482 *length = (*length + mask) & ~mask;
1483}
1484
1485#if !_POSIX_MEMLOCK
1486# define eio__mlockall(a) ((errno = ENOSYS), -1)
1487#else
1488
1489static int
1490eio__mlockall (int flags)
1491{
1492 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1493 extern int mallopt (int, int);
1494 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1036#endif 1495 #endif
1037 1496
1038 addr &= ~(page - 1); /* assume page size is always a power of two */ 1497 if (EIO_MCL_CURRENT != MCL_CURRENT
1498 || EIO_MCL_FUTURE != MCL_FUTURE)
1499 {
1500 flags = 0
1501 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1502 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1503 }
1039 1504
1505 return mlockall (flags);
1506}
1507#endif
1508
1509#if !_POSIX_MEMLOCK_RANGE
1510# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1511#else
1512
1513static int
1514eio__mlock (void *addr, size_t length)
1515{
1516 eio_page_align (&addr, &length);
1517
1518 return mlock (addr, length);
1519}
1520
1521#endif
1522
1523#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1524# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1525#else
1526
1527int
1528eio__msync (void *mem, size_t len, int flags)
1529{
1530 eio_page_align (&mem, &len);
1531
1532 if (EIO_MS_ASYNC != MS_SYNC
1533 || EIO_MS_INVALIDATE != MS_INVALIDATE
1534 || EIO_MS_SYNC != MS_SYNC)
1535 {
1536 flags = 0
1537 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1538 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1539 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1540 }
1541
1542 return msync (mem, len, flags);
1543}
1544
1545#endif
1546
1547int
1548eio__mtouch (eio_req *req)
1549{
1550 void *mem = req->ptr2;
1551 size_t len = req->size;
1552 int flags = req->int1;
1553
1554 eio_page_align (&mem, &len);
1555
1556 {
1557 intptr_t addr = (intptr_t)mem;
1558 intptr_t end = addr + len;
1559 intptr_t page = eio_pagesize ();
1560
1040 if (addr < end) 1561 if (addr < end)
1041 if (flags) /* modify */ 1562 if (flags & EIO_MT_MODIFY) /* modify */
1042 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1563 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1043 else 1564 else
1044 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1565 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1566 }
1045 1567
1046 return 0; 1568 return 0;
1047} 1569}
1048 1570
1049/*****************************************************************************/ 1571/*****************************************************************************/
… …
1083 if (req) 1605 if (req)
1084 break; 1606 break;
1085 1607
1086 ++idle; 1608 ++idle;
1087 1609
1088 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1610 ts.tv_sec = time (0) + idle_timeout;
1089 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1611 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1090 { 1612 {
1091 if (idle > max_idle) 1613 if (idle > max_idle)
1092 { 1614 {
1093 --idle; 1615 --idle;
… …
1172 return 0; \ 1694 return 0; \
1173 } 1695 }
1174 1696
1175static void eio_execute (etp_worker *self, eio_req *req) 1697static void eio_execute (etp_worker *self, eio_req *req)
1176{ 1698{
1177 errno = 0;
1178
1179 switch (req->type) 1699 switch (req->type)
1180 { 1700 {
1181 case EIO_READ: ALLOC (req->size); 1701 case EIO_READ: ALLOC (req->size);
1182 req->result = req->offs >= 0 1702 req->result = req->offs >= 0
1183 ? pread (req->int1, req->ptr2, req->size, req->offs) 1703 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1194 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1714 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1195 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1715 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1196 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1716 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1717 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1198 1718
1719 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1720 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1721 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1722 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1723
1199 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1724 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1200 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1725 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1201 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1726 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1202 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1727 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1203 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1728 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
… …
1210 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1735 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1211 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1736 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1212 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1737 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1213 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1738 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1214 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1739 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1215 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1740 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1216 1741
1217 case EIO_READLINK: ALLOC (NAME_MAX); 1742 case EIO_READLINK: ALLOC (PATH_MAX);
1218 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1743 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1219 1744
1220 case EIO_SYNC: req->result = 0; sync (); break; 1745 case EIO_SYNC: req->result = 0; sync (); break;
1221 case EIO_FSYNC: req->result = fsync (req->int1); break; 1746 case EIO_FSYNC: req->result = fsync (req->int1); break;
1222 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1747 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1223 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1748 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1749 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1224 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1750 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1751 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1225 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 1752 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1226 1753
1227 case EIO_READDIR: eio__scandir (req, self); break; 1754 case EIO_READDIR: eio__scandir (req, self); break;
1228 1755
1229 case EIO_BUSY: 1756 case EIO_BUSY:
1230#ifdef _WIN32 1757#ifdef _WIN32
1231 Sleep (req->nv1 * 1000.); 1758 Sleep (req->nv1 * 1e3);
1232#else 1759#else
1233 { 1760 {
1234 struct timeval tv; 1761 struct timeval tv;
1235 1762
1236 tv.tv_sec = req->nv1; 1763 tv.tv_sec = req->nv1;
1237 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1764 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1238 1765
1239 req->result = select (0, 0, 0, 0, &tv); 1766 req->result = select (0, 0, 0, 0, &tv);
1240 } 1767 }
1241#endif 1768#endif
1242 break; 1769 break;
… …
1257 times = tv; 1784 times = tv;
1258 } 1785 }
1259 else 1786 else
1260 times = 0; 1787 times = 0;
1261 1788
1262
1263 req->result = req->type == EIO_FUTIME 1789 req->result = req->type == EIO_FUTIME
1264 ? futimes (req->int1, times) 1790 ? futimes (req->int1, times)
1265 : utimes (req->ptr1, times); 1791 : utimes (req->ptr1, times);
1266 } 1792 }
1267 break; 1793 break;
… …
1272 case EIO_NOP: 1798 case EIO_NOP:
1273 req->result = 0; 1799 req->result = 0;
1274 break; 1800 break;
1275 1801
1276 case EIO_CUSTOM: 1802 case EIO_CUSTOM:
1277 ((void (*)(eio_req *))req->feed) (req); 1803 req->feed (req);
1278 break; 1804 break;
1279 1805
1280 default: 1806 default:
1807 errno = ENOSYS;
1281 req->result = -1; 1808 req->result = -1;
1282 break; 1809 break;
1283 } 1810 }
1284 1811
1285 req->errorno = errno; 1812 req->errorno = errno;
… …
1315eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 1842eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1316{ 1843{
1317 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 1844 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1318} 1845}
1319 1846
1847eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1848{
1849 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1850}
1851
1852eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1853{
1854 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1855}
1856
1320eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 1857eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1321{ 1858{
1322 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 1859 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1323} 1860}
1324 1861
… …
1348} 1885}
1349 1886
1350eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1887eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1351{ 1888{
1352 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1889 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1890}
1891
1892eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1893{
1894 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1353} 1895}
1354 1896
1355eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1897eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1356{ 1898{
1357 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1899 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
… …
1431eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 1973eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1432{ 1974{
1433 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1975 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1434} 1976}
1435 1977
1978eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1979{
1980 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1981}
1982
1436eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1983eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1437{ 1984{
1438 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1985 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1439} 1986}
1440 1987
1441eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1988eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1442{ 1989{
1443 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1990 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1444} 1991}
1445 1992
1446eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1993eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1447{ 1994{
1448 return eio__1path (EIO_READDIR, path, pri, cb, data); 1995 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1449} 1996}
1450 1997
1451eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1998eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1452{ 1999{
1453 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2000 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1454} 2001}
1455 2002
1456static eio_req * 2003static eio_req *
1457eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2004eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1458{ 2005{
… …
1482eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2029eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1483{ 2030{
1484 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2031 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1485} 2032}
1486 2033
1487eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2034eio_req *eio_custom (void (*)(eio_req *) execute, int pri, eio_cb cb, void *data);
1488{ 2035{
1489 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2036 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1490} 2037}
1491 2038
1492#endif 2039#endif
1493 2040
1494eio_req *eio_grp (eio_cb cb, void *data) 2041eio_req *eio_grp (eio_cb cb, void *data)
… …
1542/* misc garbage */ 2089/* misc garbage */
1543 2090
1544ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2091ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1545{ 2092{
1546 etp_worker wrk; 2093 etp_worker wrk;
2094 ssize_t ret;
1547 2095
1548 wrk.dbuf = 0; 2096 wrk.dbuf = 0;
1549 2097
1550 eio__sendfile (ofd, ifd, offset, count, &wrk); 2098 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1551 2099
1552 if (wrk.dbuf) 2100 if (wrk.dbuf)
1553 free (wrk.dbuf); 2101 free (wrk.dbuf);
1554}
1555 2102
2103 return ret;
2104}
2105

Diff Legend

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