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

Comparing libeio/eio.c (file contents):
Revision 1.32 by root, Sat Jun 6 18:06:55 2009 UTC vs.
Revision 1.71 by root, Fri Jun 10 12:35:18 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009 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"
41 45
42#ifdef EIO_STACKSIZE 46#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 47# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif 48#endif
… …
49#include <stdlib.h> 53#include <stdlib.h>
50#include <string.h> 54#include <string.h>
51#include <errno.h> 55#include <errno.h>
52#include <sys/types.h> 56#include <sys/types.h>
53#include <sys/stat.h> 57#include <sys/stat.h>
58#include <sys/statvfs.h>
54#include <limits.h> 59#include <limits.h>
55#include <fcntl.h> 60#include <fcntl.h>
56#include <assert.h> 61#include <assert.h>
57 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
58#ifndef EIO_FINISH 69#ifndef EIO_FINISH
59# 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
60#endif 71#endif
61 72
62#ifndef EIO_DESTROY 73#ifndef EIO_DESTROY
… …
70#ifdef _WIN32 81#ifdef _WIN32
71 82
72 /*doh*/ 83 /*doh*/
73#else 84#else
74 85
75# include "config.h"
76# include <sys/time.h> 86# include <sys/time.h>
77# include <sys/select.h> 87# include <sys/select.h>
78# include <unistd.h> 88# include <unistd.h>
79# include <utime.h> 89# include <utime.h>
80# include <signal.h> 90# include <signal.h>
81# include <dirent.h> 91# include <dirent.h>
82 92
93#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94# include <sys/mman.h>
95#endif
96
83/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 97/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) 98# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99# define _DIRENT_HAVE_D_TYPE /* sigh */
85# define D_INO(de) (de)->d_fileno 100# define D_INO(de) (de)->d_fileno
86# define _DIRENT_HAVE_D_TYPE /* sigh */ 101# define D_NAMLEN(de) (de)->d_namlen
87# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 102# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
88# define D_INO(de) (de)->d_ino 103# define D_INO(de) (de)->d_ino
89# endif 104# endif
105
106#ifdef _D_EXACT_NAMLEN
107# undef D_NAMLEN
108# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109#endif
90 110
91# ifdef _DIRENT_HAVE_D_TYPE 111# ifdef _DIRENT_HAVE_D_TYPE
92# define D_TYPE(de) (de)->d_type 112# define D_TYPE(de) (de)->d_type
93# endif 113# endif
94 114
… …
99#endif 119#endif
100 120
101#if HAVE_SENDFILE 121#if HAVE_SENDFILE
102# if __linux 122# if __linux
103# include <sys/sendfile.h> 123# include <sys/sendfile.h>
104# elif __freebsd 124# elif __FreeBSD__ || defined __APPLE__
105# include <sys/socket.h> 125# include <sys/socket.h>
106# include <sys/uio.h> 126# include <sys/uio.h>
107# elif __hpux 127# elif __hpux
108# include <sys/socket.h> 128# include <sys/socket.h>
109# elif __solaris /* not yet */ 129# elif __solaris
110# include <sys/sendfile.h> 130# include <sys/sendfile.h>
111# else 131# else
112# error sendfile support requested but not available 132# error sendfile support requested but not available
113# endif 133# endif
114#endif 134#endif
115 135
116#ifndef D_TYPE 136#ifndef D_TYPE
117# define D_TYPE(de) 0 137# define D_TYPE(de) 0
118#endif 138#endif
119
120#ifndef D_INO 139#ifndef D_INO
121# define D_INO(de) 0 140# define D_INO(de) 0
122#endif 141#endif
123 142#ifndef D_NAMLEN
124/* number of seconds after which an idle threads exit */ 143# define D_NAMLEN(de) strlen ((de)->d_name)
125#define IDLE_TIMEOUT 10 144#endif
126 145
127/* used for struct dirent, AIX doesn't provide it */ 146/* used for struct dirent, AIX doesn't provide it */
128#ifndef NAME_MAX 147#ifndef NAME_MAX
129# 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
130#endif 154#endif
131 155
132/* buffer size for various temporary buffers */ 156/* buffer size for various temporary buffers */
133#define EIO_BUFSIZE 65536 157#define EIO_BUFSIZE 65536
134 158
… …
187 211
188/*****************************************************************************/ 212/*****************************************************************************/
189 213
190#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 214#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
191 215
192/* calculcate time difference in ~1/EIO_TICKS of a second */ 216/* calculate time difference in ~1/EIO_TICKS of a second */
193static int tvdiff (struct timeval *tv1, struct timeval *tv2) 217static int tvdiff (struct timeval *tv1, struct timeval *tv2)
194{ 218{
195 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 219 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
196 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 220 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
197} 221}
… …
205static unsigned int max_poll_reqs; /* reslock */ 229static unsigned int max_poll_reqs; /* reslock */
206 230
207static volatile unsigned int nreqs; /* reqlock */ 231static volatile unsigned int nreqs; /* reqlock */
208static volatile unsigned int nready; /* reqlock */ 232static volatile unsigned int nready; /* reqlock */
209static volatile unsigned int npending; /* reqlock */ 233static volatile unsigned int npending; /* reqlock */
210static 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 */
211 236
212static mutex_t wrklock = X_MUTEX_INIT; 237static xmutex_t wrklock;
213static mutex_t reslock = X_MUTEX_INIT; 238static xmutex_t reslock;
214static mutex_t reqlock = X_MUTEX_INIT; 239static xmutex_t reqlock;
215static cond_t reqwait = X_COND_INIT; 240static xcond_t reqwait;
216 241
217#if !HAVE_PREADWRITE 242#if !HAVE_PREADWRITE
218/* 243/*
219 * make our pread/pwrite emulation safe against themselves, but not against 244 * make our pread/pwrite emulation safe against themselves, but not against
220 * 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,
221 * but that's your problem, not mine. 246 * but that's your problem, not mine.
222 */ 247 */
223static mutex_t preadwritelock = X_MUTEX_INIT; 248static xmutex_t preadwritelock = X_MUTEX_INIT;
224#endif 249#endif
225 250
226typedef struct etp_worker 251typedef struct etp_worker
227{ 252{
228 /* locked by wrklock */ 253 /* locked by wrklock */
229 struct etp_worker *prev, *next; 254 struct etp_worker *prev, *next;
230 255
231 thread_t tid; 256 xthread_t tid;
232 257
233 /* locked by reslock, reqlock or wrklock */ 258 /* locked by reslock, reqlock or wrklock */
234 ETP_REQ *req; /* currently processed request */ 259 ETP_REQ *req; /* currently processed request */
235 260
236 ETP_WORKER_COMMON 261 ETP_WORKER_COMMON
… …
350 } 375 }
351 376
352 abort (); 377 abort ();
353} 378}
354 379
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
355static void etp_atfork_prepare (void) 388static void etp_atfork_prepare (void)
356{ 389{
357 X_LOCK (wrklock); 390 X_LOCK (wrklock);
358 X_LOCK (reqlock); 391 X_LOCK (reqlock);
359 X_LOCK (reslock); 392 X_LOCK (reslock);
… …
397 idle = 0; 430 idle = 0;
398 nreqs = 0; 431 nreqs = 0;
399 nready = 0; 432 nready = 0;
400 npending = 0; 433 npending = 0;
401 434
402 etp_atfork_parent (); 435 etp_thread_init ();
403} 436}
404 437
405static void 438static void
406etp_once_init (void) 439etp_once_init (void)
407{ 440{
441 etp_thread_init ();
408 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);
409} 443}
410 444
411static int 445static int
412etp_init (void (*want_poll)(void), void (*done_poll)(void)) 446etp_init (void (*want_poll)(void), void (*done_poll)(void))
… …
589} 623}
590 624
591static void etp_set_max_poll_time (double nseconds) 625static void etp_set_max_poll_time (double nseconds)
592{ 626{
593 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 627 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
594 max_poll_time = nseconds; 628 max_poll_time = nseconds * EIO_TICKS;
595 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 629 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
596} 630}
597 631
598static void etp_set_max_poll_reqs (unsigned int maxreqs) 632static void etp_set_max_poll_reqs (unsigned int maxreqs)
599{ 633{
… …
603} 637}
604 638
605static void etp_set_max_idle (unsigned int nthreads) 639static void etp_set_max_idle (unsigned int nthreads)
606{ 640{
607 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 641 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
608 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;
609 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 650 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
610} 651}
611 652
612static void etp_set_min_parallel (unsigned int nthreads) 653static void etp_set_min_parallel (unsigned int nthreads)
613{ 654{
… …
741void eio_set_max_idle (unsigned int nthreads) 782void eio_set_max_idle (unsigned int nthreads)
742{ 783{
743 etp_set_max_idle (nthreads); 784 etp_set_max_idle (nthreads);
744} 785}
745 786
787void eio_set_idle_timeout (unsigned int seconds)
788{
789 etp_set_idle_timeout (seconds);
790}
791
746void eio_set_min_parallel (unsigned int nthreads) 792void eio_set_min_parallel (unsigned int nthreads)
747{ 793{
748 etp_set_min_parallel (nthreads); 794 etp_set_min_parallel (nthreads);
749} 795}
750 796
… …
798 844
799 return res; 845 return res;
800} 846}
801#endif 847#endif
802 848
803#ifndef HAVE_FUTIMES 849#ifndef HAVE_UTIMES
804 850
805# undef utimes 851# undef utimes
806# undef futimes
807# define utimes(path,times) eio__utimes (path, times) 852# define utimes(path,times) eio__utimes (path, times)
808# define futimes(fd,times) eio__futimes (fd, times)
809 853
810static int 854static int
811eio__utimes (const char *filename, const struct timeval times[2]) 855eio__utimes (const char *filename, const struct timeval times[2])
812{ 856{
813 if (times) 857 if (times)
… …
820 return utime (filename, &buf); 864 return utime (filename, &buf);
821 } 865 }
822 else 866 else
823 return utime (filename, 0); 867 return utime (filename, 0);
824} 868}
869
870#endif
871
872#ifndef HAVE_FUTIMES
873
874# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times)
825 876
826static int eio__futimes (int fd, const struct timeval tv[2]) 877static int eio__futimes (int fd, const struct timeval tv[2])
827{ 878{
828 errno = ENOSYS; 879 errno = ENOSYS;
829 return -1; 880 return -1;
… …
858 if (!res || errno != ENOSYS) 909 if (!res || errno != ENOSYS)
859 return res; 910 return res;
860#endif 911#endif
861 912
862 /* 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
863 * call fdatasync, as thta matches the expectation of it's users best */ 914 * call fdatasync, as that matches the expectation of its users best */
864 return fdatasync (fd); 915 return fdatasync (fd);
865} 916}
866 917
867#if !HAVE_READAHEAD 918#if !HAVE_READAHEAD
868# undef readahead 919# undef readahead
… …
891 942
892/* sendfile always needs emulation */ 943/* sendfile always needs emulation */
893static ssize_t 944static ssize_t
894eio__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)
895{ 946{
947 ssize_t written = 0;
896 ssize_t res; 948 ssize_t res;
897 949
898 if (!count) 950 if (!count)
899 return 0; 951 return 0;
900 952
953 for (;;)
954 {
901#if HAVE_SENDFILE 955#if HAVE_SENDFILE
902# if __linux 956# if __linux
957 off_t soffset = offset;
903 res = sendfile (ofd, ifd, &offset, count); 958 res = sendfile (ofd, ifd, &soffset, count);
904 959
905# elif __freebsd 960# elif __FreeBSD__
906 /* 961 /*
907 * 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
908 * wasted on making it similar to other I/O functions. 963 * wasted on making it similar to other I/O functions.
909 */ 964 */
910 {
911 off_t sbytes; 965 off_t sbytes;
912 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 966 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
913 967
914 if (res < 0 && sbytes) 968 #if 0 /* according to the manpage, this is correct, but broken behaviour */
915 /* 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)
916 res = sbytes; 973 res = sbytes;
917 } 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;
918 987
919# elif __hpux 988# elif __hpux
920 res = sendfile (ofd, ifd, offset, count, 0, 0); 989 res = sendfile (ofd, ifd, offset, count, 0, 0);
921 990
922# elif __solaris 991# elif __solaris
923 {
924 struct sendfilevec vec; 992 struct sendfilevec vec;
925 size_t sbytes; 993 size_t sbytes;
926 994
927 vec.sfv_fd = ifd; 995 vec.sfv_fd = ifd;
928 vec.sfv_flag = 0; 996 vec.sfv_flag = 0;
929 vec.sfv_off = offset; 997 vec.sfv_off = offset;
930 vec.sfv_len = count; 998 vec.sfv_len = count;
931 999
932 res = sendfilev (ofd, &vec, 1, &sbytes); 1000 res = sendfilev (ofd, &vec, 1, &sbytes);
933 1001
934 if (res < 0 && sbytes) 1002 if (res < 0 && sbytes)
935 res = sbytes; 1003 res = sbytes;
936 }
937 1004
938# 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
939#else 1016#else
940 res = -1; 1017 res = -1;
941 errno = ENOSYS; 1018 errno = ENOSYS;
942#endif 1019#endif
943 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
944 if (res < 0 1044 if (res < 0
945 && (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 */
946#if __solaris 1051#if __solaris
947 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1052 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
948#endif 1053#endif
949 ) 1054 )
950 ) 1055 )
… …
981 } 1086 }
982 1087
983 return res; 1088 return res;
984} 1089}
985 1090
986static int 1091static signed char
987eio_dent_cmp (const void *a_, const void *b_) 1092eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
988{ 1093{
989 const eio_dirent *a = (const eio_dirent *)a_; 1094 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
990 const eio_dirent *b = (const eio_dirent *)b_; 1095 : a->inode < b->inode ? -1
1096 : a->inode > b->inode ? 1
1097 : 0;
1098}
991 1099
992 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score 1100#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
993 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */ 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);
994} 1254}
995 1255
996/* read a full directory */ 1256/* read a full directory */
997static void 1257static void
998eio__scandir (eio_req *req, etp_worker *self) 1258eio__scandir (eio_req *req, etp_worker *self)
999{ 1259{
1000 DIR *dirp; 1260 DIR *dirp;
1001 EIO_STRUCT_DIRENT *entp; 1261 EIO_STRUCT_DIRENT *entp;
1002 unsigned char *name, *names; 1262 char *name, *names;
1003 int namesalloc = 4096; 1263 int namesalloc = 4096;
1004 int namesoffs = 0; 1264 int namesoffs = 0;
1005 int flags = req->int1; 1265 int flags = req->int1;
1006 eio_dirent *dents = 0; 1266 eio_dirent *dents = 0;
1007 int dentalloc = 128; 1267 int dentalloc = 128;
1008 int dentoffs = 0; 1268 int dentoffs = 0;
1269 ino_t inode_bits = 0;
1009 1270
1010 req->result = -1; 1271 req->result = -1;
1011 1272
1012 if (!(flags & EIO_READDIR_DENTS)) 1273 if (!(flags & EIO_READDIR_DENTS))
1013 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1274 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1014 1275
1015 X_LOCK (wrklock); 1276 X_LOCK (wrklock);
1016 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1277 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1017 self->dirp = dirp = opendir (req->ptr1); 1278 self->dirp = dirp = opendir (req->ptr1);
1279
1018 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1280 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1019 req->ptr1 = names = malloc (namesalloc);
1020 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1281 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1282 req->ptr2 = names = malloc (namesalloc);
1021 X_UNLOCK (wrklock); 1283 X_UNLOCK (wrklock);
1022 1284
1023 if (dirp && names && (!flags || dents)) 1285 if (dirp && names && (!flags || dents))
1024 for (;;) 1286 for (;;)
1025 { 1287 {
… …
1033 1295
1034 /* sort etc. */ 1296 /* sort etc. */
1035 req->int1 = flags; 1297 req->int1 = flags;
1036 req->result = dentoffs; 1298 req->result = dentoffs;
1037 1299
1038 if (dents)
1039 {
1040 eio_dirent *ent = dents + dentoffs;
1041
1042 while (ent > dents)
1043 (--ent)->name = names + (size_t)ent->name;
1044 }
1045
1046 if (flags & EIO_READDIR_STAT_ORDER 1300 if (flags & EIO_READDIR_STAT_ORDER)
1047 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1301 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1048 {
1049 /* pray your qsort doesn't use quicksort */
1050 qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1051 }
1052 else if (flags & EIO_READDIR_DIRS_FIRST) 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
1053 { 1306 {
1054 /* in this case, all is known, and we just put dirs first and sort them */ 1307 /* in this case, all is known, and we just put dirs first and sort them */
1055 eio_dirent *ent = dents + dentoffs; 1308 eio_dirent *oth = dents + dentoffs;
1056 eio_dirent *dir = dents; 1309 eio_dirent *dir = dents;
1057 1310
1311 /* now partition dirs to the front, and non-dirs to the back */
1312 /* by walking from both sides and swapping if necessary */
1058 while (ent > dir) 1313 while (oth > dir)
1059 { 1314 {
1060 if (dir->type == DT_DIR) 1315 if (dir->type == EIO_DT_DIR)
1061 ++dir; 1316 ++dir;
1062 else
1063 {
1064 --ent;
1065
1066 if (ent->type == DT_DIR) 1317 else if ((--oth)->type == EIO_DT_DIR)
1067 { 1318 {
1068 eio_dirent tmp = *dir; 1319 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1069 *dir = *ent;
1070 *ent = tmp;
1071 1320
1072 ++dir; 1321 ++dir;
1073 } 1322 }
1074 } 1323 }
1324
1325 /* now sort the dirs only (dirs all have the same score) */
1326 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1075 } 1327 }
1076
1077 /* now sort the dirs only */
1078 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1079 }
1080
1081 /* only provide the names array unless DENTS is specified */
1082 if (!(flags & EIO_READDIR_DENTS))
1083 {
1084 X_LOCK (wrklock);
1085 assert (!dents);
1086 req->ptr1 = 0;
1087 req->ptr2 = names;
1088 X_UNLOCK (wrklock);
1089 }
1090 1328
1091 break; 1329 break;
1092 } 1330 }
1093 1331
1094 /* now add the entry to our list(s) */ 1332 /* now add the entry to our list(s) */
1095 name = entp->d_name; 1333 name = entp->d_name;
1096 1334
1097 /* skip . and .. entries */ 1335 /* skip . and .. entries */
1098 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1336 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1099 { 1337 {
1100 int len = strlen (name) + 1; 1338 int len = D_NAMLEN (entp) + 1;
1101 1339
1102 while (expect_false (namesoffs + len > namesalloc)) 1340 while (expect_false (namesoffs + len > namesalloc))
1103 { 1341 {
1104 namesalloc *= 2; 1342 namesalloc *= 2;
1105 X_LOCK (wrklock); 1343 X_LOCK (wrklock);
1106 req->ptr1 = names = realloc (names, namesalloc); 1344 req->ptr2 = names = realloc (names, namesalloc);
1107 X_UNLOCK (wrklock); 1345 X_UNLOCK (wrklock);
1108 1346
1109 if (!names) 1347 if (!names)
1110 break; 1348 break;
1111 } 1349 }
… …
1118 1356
1119 if (expect_false (dentoffs == dentalloc)) 1357 if (expect_false (dentoffs == dentalloc))
1120 { 1358 {
1121 dentalloc *= 2; 1359 dentalloc *= 2;
1122 X_LOCK (wrklock); 1360 X_LOCK (wrklock);
1123 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1361 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1124 X_UNLOCK (wrklock); 1362 X_UNLOCK (wrklock);
1125 1363
1126 if (!dents) 1364 if (!dents)
1127 break; 1365 break;
1128 } 1366 }
1129 1367
1130 ent = dents + dentoffs; 1368 ent = dents + dentoffs;
1131 1369
1132 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1370 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1133 ent->namelen = len - 1; 1371 ent->namelen = len - 1;
1134 ent->inode = D_INO (entp); 1372 ent->inode = D_INO (entp);
1373
1374 inode_bits |= ent->inode;
1135 1375
1136 switch (D_TYPE (entp)) 1376 switch (D_TYPE (entp))
1137 { 1377 {
1138 default: 1378 default:
1139 ent->type = EIO_DT_UNKNOWN; 1379 ent->type = EIO_DT_UNKNOWN;
… …
1144 case DT_FIFO: ent->type = EIO_DT_FIFO; break; 1384 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1145 #endif 1385 #endif
1146 #ifdef DT_CHR 1386 #ifdef DT_CHR
1147 case DT_CHR: ent->type = EIO_DT_CHR; break; 1387 case DT_CHR: ent->type = EIO_DT_CHR; break;
1148 #endif 1388 #endif
1389 #ifdef DT_MPC
1390 case DT_MPC: ent->type = EIO_DT_MPC; break;
1391 #endif
1149 #ifdef DT_DIR 1392 #ifdef DT_DIR
1150 case DT_DIR: ent->type = EIO_DT_DIR; break; 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;
1151 #endif 1397 #endif
1152 #ifdef DT_BLK 1398 #ifdef DT_BLK
1153 case DT_BLK: ent->type = EIO_DT_BLK; break; 1399 case DT_BLK: ent->type = EIO_DT_BLK; break;
1154 #endif 1400 #endif
1401 #ifdef DT_MPB
1402 case DT_MPB: ent->type = EIO_DT_MPB; break;
1403 #endif
1155 #ifdef DT_REG 1404 #ifdef DT_REG
1156 case DT_REG: ent->type = EIO_DT_REG; break; 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;
1157 #endif 1412 #endif
1158 #ifdef DT_LNK 1413 #ifdef DT_LNK
1159 case DT_LNK: ent->type = EIO_DT_LNK; break; 1414 case DT_LNK: ent->type = EIO_DT_LNK; break;
1160 #endif 1415 #endif
1161 #ifdef DT_SOCK 1416 #ifdef DT_SOCK
1162 case DT_SOCK: ent->type = EIO_DT_SOCK; break; 1417 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1163 #endif 1418 #endif
1419 #ifdef DT_DOOR
1420 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1421 #endif
1164 #ifdef DT_WHT 1422 #ifdef DT_WHT
1165 case DT_WHT: ent->type = EIO_DT_WHT; break; 1423 case DT_WHT: ent->type = EIO_DT_WHT; break;
1166 #endif 1424 #endif
1167 } 1425 }
1168 1426
1169 ent->score = 0; 1427 ent->score = 7;
1170 1428
1171 if (flags & EIO_READDIR_DIRS_FIRST) 1429 if (flags & EIO_READDIR_DIRS_FIRST)
1172 { 1430 {
1173 if (ent->type == EIO_DT_UNKNOWN) 1431 if (ent->type == EIO_DT_UNKNOWN)
1174 { 1432 {
1175 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1433 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1176 ent->score = 98; 1434 ent->score = 1;
1177 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1435 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1178 ent->score = len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ 1436 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1179 } 1437 }
1180 else if (ent->type == DT_DIR) 1438 else if (ent->type == EIO_DT_DIR)
1181 ent->score = 100; 1439 ent->score = 0;
1182 } 1440 }
1183 } 1441 }
1184 1442
1185 namesoffs += len; 1443 namesoffs += len;
1186 ++dentoffs; 1444 ++dentoffs;
1187 } 1445 }
1446
1447 if (EIO_CANCELLED (req))
1448 {
1449 errno = ECANCELED;
1450 break;
1451 }
1188 } 1452 }
1189 else
1190 req->result = -1;
1191} 1453}
1192 1454
1193#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1194# undef msync
1195# define msync(a,b,c) ((errno = ENOSYS), -1)
1196#endif
1197
1198int
1199eio__mtouch (void *mem, size_t len, int flags)
1200{
1201 intptr_t addr = (intptr_t)mem;
1202 intptr_t end = addr + len;
1203#ifdef PAGESIZE 1455#ifdef PAGESIZE
1204 const intptr_t page = PAGESIZE; 1456# define eio_pagesize() PAGESIZE
1205#else 1457#else
1458static intptr_t
1459eio_pagesize (void)
1460{
1206 static intptr_t page; 1461 static intptr_t page;
1207 1462
1208 if (!page) 1463 if (!page)
1209 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 */
1210#endif 1495 #endif
1211 1496
1212 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 }
1213 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
1214 if (addr < end) 1561 if (addr < end)
1215 if (flags) /* modify */ 1562 if (flags & EIO_MT_MODIFY) /* modify */
1216 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));
1217 else 1564 else
1218 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 }
1219 1567
1220 return 0; 1568 return 0;
1221} 1569}
1222 1570
1223/*****************************************************************************/ 1571/*****************************************************************************/
… …
1257 if (req) 1605 if (req)
1258 break; 1606 break;
1259 1607
1260 ++idle; 1608 ++idle;
1261 1609
1262 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1610 ts.tv_sec = time (0) + idle_timeout;
1263 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1611 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1264 { 1612 {
1265 if (idle > max_idle) 1613 if (idle > max_idle)
1266 { 1614 {
1267 --idle; 1615 --idle;
… …
1346 return 0; \ 1694 return 0; \
1347 } 1695 }
1348 1696
1349static void eio_execute (etp_worker *self, eio_req *req) 1697static void eio_execute (etp_worker *self, eio_req *req)
1350{ 1698{
1351 errno = 0;
1352
1353 switch (req->type) 1699 switch (req->type)
1354 { 1700 {
1355 case EIO_READ: ALLOC (req->size); 1701 case EIO_READ: ALLOC (req->size);
1356 req->result = req->offs >= 0 1702 req->result = req->offs >= 0
1357 ? pread (req->int1, req->ptr2, req->size, req->offs) 1703 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1368 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1714 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1369 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1715 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1370 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1716 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1371 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1717 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1372 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
1373 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;
1374 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;
1375 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;
1376 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;
1377 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1728 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
… …
1384 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1735 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1385 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;
1386 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1737 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1387 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1738 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1388 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1739 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1389 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;
1390 1741
1391 case EIO_READLINK: ALLOC (NAME_MAX); 1742 case EIO_READLINK: ALLOC (PATH_MAX);
1392 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1743 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1393 1744
1394 case EIO_SYNC: req->result = 0; sync (); break; 1745 case EIO_SYNC: req->result = 0; sync (); break;
1395 case EIO_FSYNC: req->result = fsync (req->int1); break; 1746 case EIO_FSYNC: req->result = fsync (req->int1); break;
1396 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1747 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1397 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;
1398 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;
1399 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;
1400 1753
1401 case EIO_READDIR: eio__scandir (req, self); break; 1754 case EIO_READDIR: eio__scandir (req, self); break;
1402 1755
1403 case EIO_BUSY: 1756 case EIO_BUSY:
1404#ifdef _WIN32 1757#ifdef _WIN32
1405 Sleep (req->nv1 * 1000.); 1758 Sleep (req->nv1 * 1e3);
1406#else 1759#else
1407 { 1760 {
1408 struct timeval tv; 1761 struct timeval tv;
1409 1762
1410 tv.tv_sec = req->nv1; 1763 tv.tv_sec = req->nv1;
1411 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1764 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1412 1765
1413 req->result = select (0, 0, 0, 0, &tv); 1766 req->result = select (0, 0, 0, 0, &tv);
1414 } 1767 }
1415#endif 1768#endif
1416 break; 1769 break;
… …
1431 times = tv; 1784 times = tv;
1432 } 1785 }
1433 else 1786 else
1434 times = 0; 1787 times = 0;
1435 1788
1436
1437 req->result = req->type == EIO_FUTIME 1789 req->result = req->type == EIO_FUTIME
1438 ? futimes (req->int1, times) 1790 ? futimes (req->int1, times)
1439 : utimes (req->ptr1, times); 1791 : utimes (req->ptr1, times);
1440 } 1792 }
1441 break; 1793 break;
… …
1446 case EIO_NOP: 1798 case EIO_NOP:
1447 req->result = 0; 1799 req->result = 0;
1448 break; 1800 break;
1449 1801
1450 case EIO_CUSTOM: 1802 case EIO_CUSTOM:
1451 ((void (*)(eio_req *))req->feed) (req); 1803 req->feed (req);
1452 break; 1804 break;
1453 1805
1454 default: 1806 default:
1807 errno = ENOSYS;
1455 req->result = -1; 1808 req->result = -1;
1456 break; 1809 break;
1457 } 1810 }
1458 1811
1459 req->errorno = errno; 1812 req->errorno = errno;
… …
1489eio_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)
1490{ 1843{
1491 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;
1492} 1845}
1493 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
1494eio_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)
1495{ 1858{
1496 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;
1497} 1860}
1498 1861
… …
1522} 1885}
1523 1886
1524eio_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)
1525{ 1888{
1526 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;
1527} 1895}
1528 1896
1529eio_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)
1530{ 1898{
1531 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;
… …
1605eio_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)
1606{ 1974{
1607 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1975 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1608} 1976}
1609 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
1610eio_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)
1611{ 1984{
1612 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1985 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1613} 1986}
1614 1987
… …
1622 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 1995 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1623} 1996}
1624 1997
1625eio_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)
1626{ 1999{
1627 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;
1628} 2001}
1629 2002
1630static eio_req * 2003static eio_req *
1631eio__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)
1632{ 2005{
… …
1656eio_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)
1657{ 2030{
1658 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2031 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1659} 2032}
1660 2033
1661eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2034eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1662{ 2035{
1663 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2036 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1664} 2037}
1665 2038
1666#endif 2039#endif
1667 2040
1668eio_req *eio_grp (eio_cb cb, void *data) 2041eio_req *eio_grp (eio_cb cb, void *data)
… …
1716/* misc garbage */ 2089/* misc garbage */
1717 2090
1718ssize_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)
1719{ 2092{
1720 etp_worker wrk; 2093 etp_worker wrk;
2094 ssize_t ret;
1721 2095
1722 wrk.dbuf = 0; 2096 wrk.dbuf = 0;
1723 2097
1724 eio__sendfile (ofd, ifd, offset, count, &wrk); 2098 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1725 2099
1726 if (wrk.dbuf) 2100 if (wrk.dbuf)
1727 free (wrk.dbuf); 2101 free (wrk.dbuf);
1728}
1729 2102
2103 return ret;
2104}
2105

Diff Legend

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