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 1.32 by root, Sat Jun 6 18:06:55 2009 UTC vs.
Revision 1.66 by root, Sun Jun 5 19:58:37 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
900 951
901#if HAVE_SENDFILE 952#if HAVE_SENDFILE
902# if __linux 953# if __linux
903 res = sendfile (ofd, ifd, &offset, count); 954 res = sendfile (ofd, ifd, &offset, count);
904 955
905# elif __freebsd 956# elif __FreeBSD__
906 /* 957 /*
907 * Of course, the freebsd sendfile is a dire hack with no thoughts 958 * Of course, the freebsd sendfile is a dire hack with no thoughts
908 * wasted on making it similar to other I/O functions. 959 * wasted on making it similar to other I/O functions.
909 */ 960 */
910 { 961 {
911 off_t sbytes; 962 off_t sbytes;
912 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 963 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
913 964
914 if (res < 0 && sbytes) 965 #if 0 /* according to the manpage, this is correct, but broken behaviour */
915 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 966 /* freebsd' sendfile will return 0 on success */
967 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
968 /* not on e.g. EIO or EPIPE - sounds broken */
969 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
970 res = sbytes;
971 #endif
972
973 /* according to source inspection, this is correct, and useful behaviour */
974 if (sbytes)
975 res = sbytes;
976 }
977
978# elif defined (__APPLE__)
979
980 {
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)
916 res = sbytes; 986 res = sbytes;
917 } 987 }
918 988
919# elif __hpux 989# elif __hpux
920 res = sendfile (ofd, ifd, offset, count, 0, 0); 990 res = sendfile (ofd, ifd, offset, count, 0, 0);
934 if (res < 0 && sbytes) 1004 if (res < 0 && sbytes)
935 res = sbytes; 1005 res = sbytes;
936 } 1006 }
937 1007
938# endif 1008# endif
1009
1010#elif defined (_WIN32)
1011
1012 /* does not work, just for documentation of what would need to be done */
1013 {
1014 HANDLE h = TO_SOCKET (ifd);
1015 SetFilePointer (h, offset, 0, FILE_BEGIN);
1016 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1017 }
1018
939#else 1019#else
940 res = -1; 1020 res = -1;
941 errno = ENOSYS; 1021 errno = ENOSYS;
942#endif 1022#endif
943 1023
944 if (res < 0 1024 if (res < 0
945 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1025 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1026 /* BSDs */
1027#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1028 || errno == ENOTSUP
1029#endif
1030 || errno == EOPNOTSUPP /* BSDs */
946#if __solaris 1031#if __solaris
947 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1032 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
948#endif 1033#endif
949 ) 1034 )
950 ) 1035 )
981 } 1066 }
982 1067
983 return res; 1068 return res;
984} 1069}
985 1070
986static int 1071static signed char
987eio_dent_cmp (const void *a_, const void *b_) 1072eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
988{ 1073{
989 const eio_dirent *a = (const eio_dirent *)a_; 1074 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_; 1075 : a->inode < b->inode ? -1
1076 : a->inode > b->inode ? 1
1077 : 0;
1078}
991 1079
992 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score 1080#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 */ 1081
1082#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1083#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1084
1085static void
1086eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1087{
1088 unsigned char bits [9 + sizeof (ino_t) * 8];
1089 unsigned char *bit = bits;
1090
1091 assert (CHAR_BIT == 8);
1092 assert (sizeof (eio_dirent) * 8 < 256);
1093 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1094 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1095
1096 if (size <= EIO_SORT_FAST)
1097 return;
1098
1099 /* first prepare an array of bits to test in our radix sort */
1100 /* try to take endianness into account, as well as differences in ino_t sizes */
1101 /* inode_bits must contain all inodes ORed together */
1102 /* which is used to skip bits that are 0 everywhere, which is very common */
1103 {
1104 ino_t endianness;
1105 int i, j;
1106
1107 /* we store the byte offset of byte n into byte n of "endianness" */
1108 for (i = 0; i < sizeof (ino_t); ++i)
1109 ((unsigned char *)&endianness)[i] = i;
1110
1111 *bit++ = 0;
1112
1113 for (i = 0; i < sizeof (ino_t); ++i)
1114 {
1115 /* shifting off the byte offsets out of "endianness" */
1116 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1117 endianness >>= 8;
1118
1119 for (j = 0; j < 8; ++j)
1120 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1121 *bit++ = offs + j;
1122 }
1123
1124 for (j = 0; j < 8; ++j)
1125 if (score_bits & (1 << j))
1126 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1127 }
1128
1129 /* now actually do the sorting (a variant of MSD radix sort) */
1130 {
1131 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1132 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1133 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1134 int stk_idx = 0;
1135
1136 base_stk [stk_idx] = dents;
1137 end_stk [stk_idx] = dents + size;
1138 bit_stk [stk_idx] = bit - 1;
1139
1140 do
1141 {
1142 base = base_stk [stk_idx];
1143 end = end_stk [stk_idx];
1144 bit = bit_stk [stk_idx];
1145
1146 for (;;)
1147 {
1148 unsigned char O = *bit >> 3;
1149 unsigned char M = 1 << (*bit & 7);
1150
1151 eio_dirent *a = base;
1152 eio_dirent *b = end;
1153
1154 if (b - a < EIO_SORT_CUTOFF)
1155 break;
1156
1157 /* now bit-partition the array on the bit */
1158 /* this ugly asymmetric loop seems to perform much better than typical */
1159 /* partition algos found in the literature */
1160 do
1161 if (!(((unsigned char *)a)[O] & M))
1162 ++a;
1163 else if (!(((unsigned char *)--b)[O] & M))
1164 {
1165 eio_dirent tmp = *a; *a = *b; *b = tmp;
1166 ++a;
1167 }
1168 while (b > a);
1169
1170 /* next bit, or stop, if no bits left in this path */
1171 if (!*--bit)
1172 break;
1173
1174 base_stk [stk_idx] = a;
1175 end_stk [stk_idx] = end;
1176 bit_stk [stk_idx] = bit;
1177 ++stk_idx;
1178
1179 end = a;
1180 }
1181 }
1182 while (stk_idx--);
1183 }
1184}
1185
1186static void
1187eio_dent_insertion_sort (eio_dirent *dents, int size)
1188{
1189 /* first move the smallest element to the front, to act as a sentinel */
1190 {
1191 int i;
1192 eio_dirent *min = dents;
1193
1194 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1195 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1196 if (EIO_DENT_CMP (dents [i], <, *min))
1197 min = &dents [i];
1198
1199 /* swap elements 0 and j (minimum) */
1200 {
1201 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1202 }
1203 }
1204
1205 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1206 {
1207 eio_dirent *i, *j;
1208
1209 for (i = dents + 1; i < dents + size; ++i)
1210 {
1211 eio_dirent value = *i;
1212
1213 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1214 j [1] = j [0];
1215
1216 j [1] = value;
1217 }
1218 }
1219}
1220
1221static void
1222eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1223{
1224 if (size <= 1)
1225 return; /* our insertion sort relies on size > 0 */
1226
1227 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1228 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1229 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1230
1231 /* use an insertion sort at the end, or for small arrays, */
1232 /* as insertion sort is more efficient for small partitions */
1233 eio_dent_insertion_sort (dents, size);
994} 1234}
995 1235
996/* read a full directory */ 1236/* read a full directory */
997static void 1237static void
998eio__scandir (eio_req *req, etp_worker *self) 1238eio__scandir (eio_req *req, etp_worker *self)
999{ 1239{
1000 DIR *dirp; 1240 DIR *dirp;
1001 EIO_STRUCT_DIRENT *entp; 1241 EIO_STRUCT_DIRENT *entp;
1002 unsigned char *name, *names; 1242 char *name, *names;
1003 int namesalloc = 4096; 1243 int namesalloc = 4096;
1004 int namesoffs = 0; 1244 int namesoffs = 0;
1005 int flags = req->int1; 1245 int flags = req->int1;
1006 eio_dirent *dents = 0; 1246 eio_dirent *dents = 0;
1007 int dentalloc = 128; 1247 int dentalloc = 128;
1008 int dentoffs = 0; 1248 int dentoffs = 0;
1249 ino_t inode_bits = 0;
1009 1250
1010 req->result = -1; 1251 req->result = -1;
1011 1252
1012 if (!(flags & EIO_READDIR_DENTS)) 1253 if (!(flags & EIO_READDIR_DENTS))
1013 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1254 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1014 1255
1015 X_LOCK (wrklock); 1256 X_LOCK (wrklock);
1016 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1257 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1017 self->dirp = dirp = opendir (req->ptr1); 1258 self->dirp = dirp = opendir (req->ptr1);
1259
1018 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1260 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; 1261 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1262 req->ptr2 = names = malloc (namesalloc);
1021 X_UNLOCK (wrklock); 1263 X_UNLOCK (wrklock);
1022 1264
1023 if (dirp && names && (!flags || dents)) 1265 if (dirp && names && (!flags || dents))
1024 for (;;) 1266 for (;;)
1025 { 1267 {
1033 1275
1034 /* sort etc. */ 1276 /* sort etc. */
1035 req->int1 = flags; 1277 req->int1 = flags;
1036 req->result = dentoffs; 1278 req->result = dentoffs;
1037 1279
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 1280 if (flags & EIO_READDIR_STAT_ORDER)
1047 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1281 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) 1282 else if (flags & EIO_READDIR_DIRS_FIRST)
1283 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1284 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1285 else
1053 { 1286 {
1054 /* in this case, all is known, and we just put dirs first and sort them */ 1287 /* in this case, all is known, and we just put dirs first and sort them */
1055 eio_dirent *ent = dents + dentoffs; 1288 eio_dirent *oth = dents + dentoffs;
1056 eio_dirent *dir = dents; 1289 eio_dirent *dir = dents;
1057 1290
1291 /* now partition dirs to the front, and non-dirs to the back */
1292 /* by walking from both sides and swapping if necessary */
1058 while (ent > dir) 1293 while (oth > dir)
1059 { 1294 {
1060 if (dir->type == DT_DIR) 1295 if (dir->type == EIO_DT_DIR)
1061 ++dir; 1296 ++dir;
1062 else
1063 {
1064 --ent;
1065
1066 if (ent->type == DT_DIR) 1297 else if ((--oth)->type == EIO_DT_DIR)
1067 { 1298 {
1068 eio_dirent tmp = *dir; 1299 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1069 *dir = *ent;
1070 *ent = tmp;
1071 1300
1072 ++dir; 1301 ++dir;
1073 } 1302 }
1074 } 1303 }
1304
1305 /* now sort the dirs only (dirs all have the same score) */
1306 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1075 } 1307 }
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 1308
1091 break; 1309 break;
1092 } 1310 }
1093 1311
1094 /* now add the entry to our list(s) */ 1312 /* now add the entry to our list(s) */
1095 name = entp->d_name; 1313 name = entp->d_name;
1096 1314
1097 /* skip . and .. entries */ 1315 /* skip . and .. entries */
1098 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1316 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1099 { 1317 {
1100 int len = strlen (name) + 1; 1318 int len = D_NAMLEN (entp) + 1;
1101 1319
1102 while (expect_false (namesoffs + len > namesalloc)) 1320 while (expect_false (namesoffs + len > namesalloc))
1103 { 1321 {
1104 namesalloc *= 2; 1322 namesalloc *= 2;
1105 X_LOCK (wrklock); 1323 X_LOCK (wrklock);
1106 req->ptr1 = names = realloc (names, namesalloc); 1324 req->ptr2 = names = realloc (names, namesalloc);
1107 X_UNLOCK (wrklock); 1325 X_UNLOCK (wrklock);
1108 1326
1109 if (!names) 1327 if (!names)
1110 break; 1328 break;
1111 } 1329 }
1118 1336
1119 if (expect_false (dentoffs == dentalloc)) 1337 if (expect_false (dentoffs == dentalloc))
1120 { 1338 {
1121 dentalloc *= 2; 1339 dentalloc *= 2;
1122 X_LOCK (wrklock); 1340 X_LOCK (wrklock);
1123 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1341 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1124 X_UNLOCK (wrklock); 1342 X_UNLOCK (wrklock);
1125 1343
1126 if (!dents) 1344 if (!dents)
1127 break; 1345 break;
1128 } 1346 }
1129 1347
1130 ent = dents + dentoffs; 1348 ent = dents + dentoffs;
1131 1349
1132 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1350 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1133 ent->namelen = len - 1; 1351 ent->namelen = len - 1;
1134 ent->inode = D_INO (entp); 1352 ent->inode = D_INO (entp);
1353
1354 inode_bits |= ent->inode;
1135 1355
1136 switch (D_TYPE (entp)) 1356 switch (D_TYPE (entp))
1137 { 1357 {
1138 default: 1358 default:
1139 ent->type = EIO_DT_UNKNOWN; 1359 ent->type = EIO_DT_UNKNOWN;
1144 case DT_FIFO: ent->type = EIO_DT_FIFO; break; 1364 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1145 #endif 1365 #endif
1146 #ifdef DT_CHR 1366 #ifdef DT_CHR
1147 case DT_CHR: ent->type = EIO_DT_CHR; break; 1367 case DT_CHR: ent->type = EIO_DT_CHR; break;
1148 #endif 1368 #endif
1369 #ifdef DT_MPC
1370 case DT_MPC: ent->type = EIO_DT_MPC; break;
1371 #endif
1149 #ifdef DT_DIR 1372 #ifdef DT_DIR
1150 case DT_DIR: ent->type = EIO_DT_DIR; break; 1373 case DT_DIR: ent->type = EIO_DT_DIR; break;
1374 #endif
1375 #ifdef DT_NAM
1376 case DT_NAM: ent->type = EIO_DT_NAM; break;
1151 #endif 1377 #endif
1152 #ifdef DT_BLK 1378 #ifdef DT_BLK
1153 case DT_BLK: ent->type = EIO_DT_BLK; break; 1379 case DT_BLK: ent->type = EIO_DT_BLK; break;
1154 #endif 1380 #endif
1381 #ifdef DT_MPB
1382 case DT_MPB: ent->type = EIO_DT_MPB; break;
1383 #endif
1155 #ifdef DT_REG 1384 #ifdef DT_REG
1156 case DT_REG: ent->type = EIO_DT_REG; break; 1385 case DT_REG: ent->type = EIO_DT_REG; break;
1386 #endif
1387 #ifdef DT_NWK
1388 case DT_NWK: ent->type = EIO_DT_NWK; break;
1389 #endif
1390 #ifdef DT_CMP
1391 case DT_CMP: ent->type = EIO_DT_CMP; break;
1157 #endif 1392 #endif
1158 #ifdef DT_LNK 1393 #ifdef DT_LNK
1159 case DT_LNK: ent->type = EIO_DT_LNK; break; 1394 case DT_LNK: ent->type = EIO_DT_LNK; break;
1160 #endif 1395 #endif
1161 #ifdef DT_SOCK 1396 #ifdef DT_SOCK
1162 case DT_SOCK: ent->type = EIO_DT_SOCK; break; 1397 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1163 #endif 1398 #endif
1399 #ifdef DT_DOOR
1400 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1401 #endif
1164 #ifdef DT_WHT 1402 #ifdef DT_WHT
1165 case DT_WHT: ent->type = EIO_DT_WHT; break; 1403 case DT_WHT: ent->type = EIO_DT_WHT; break;
1166 #endif 1404 #endif
1167 } 1405 }
1168 1406
1169 ent->score = 0; 1407 ent->score = 7;
1170 1408
1171 if (flags & EIO_READDIR_DIRS_FIRST) 1409 if (flags & EIO_READDIR_DIRS_FIRST)
1172 { 1410 {
1173 if (ent->type == EIO_DT_UNKNOWN) 1411 if (ent->type == EIO_DT_UNKNOWN)
1174 { 1412 {
1175 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1413 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1176 ent->score = 98; 1414 ent->score = 1;
1177 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1415 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 */ 1416 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1179 } 1417 }
1180 else if (ent->type == DT_DIR) 1418 else if (ent->type == EIO_DT_DIR)
1181 ent->score = 100; 1419 ent->score = 0;
1182 } 1420 }
1183 } 1421 }
1184 1422
1185 namesoffs += len; 1423 namesoffs += len;
1186 ++dentoffs; 1424 ++dentoffs;
1187 } 1425 }
1426
1427 if (EIO_CANCELLED (req))
1428 {
1429 errno = ECANCELED;
1430 break;
1431 }
1188 } 1432 }
1189 else
1190 req->result = -1;
1191} 1433}
1434
1435#ifdef PAGESIZE
1436# define eio_pagesize() PAGESIZE
1437#else
1438static intptr_t
1439eio_pagesize (void)
1440{
1441 static intptr_t page;
1442
1443 if (!page)
1444 page = sysconf (_SC_PAGESIZE);
1445
1446 return page;
1447}
1448#endif
1449
1450static void
1451eio_page_align (void **addr, size_t *length)
1452{
1453 intptr_t mask = eio_pagesize () - 1;
1454
1455 /* round down addr */
1456 intptr_t adj = mask & (intptr_t)*addr;
1457
1458 *addr = (void *)((intptr_t)*addr - adj);
1459 *length += adj;
1460
1461 /* round up length */
1462 *length = (*length + mask) & ~mask;
1463}
1464
1465#if !_POSIX_MEMLOCK
1466# define eio__mlockall(a) ((errno = ENOSYS), -1)
1467#else
1468
1469static int
1470eio__mlockall (int flags)
1471{
1472 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1473 extern int mallopt (int, int);
1474 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1475 #endif
1476
1477 if (EIO_MCL_CURRENT != MCL_CURRENT
1478 || EIO_MCL_FUTURE != MCL_FUTURE)
1479 {
1480 flags = 0
1481 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1482 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1483 }
1484
1485 return mlockall (flags);
1486}
1487#endif
1488
1489#if !_POSIX_MEMLOCK_RANGE
1490# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1491#else
1492
1493static int
1494eio__mlock (void *addr, size_t length)
1495{
1496 eio_page_align (&addr, &length);
1497
1498 return mlock (addr, length);
1499}
1500
1501#endif
1192 1502
1193#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1503#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1194# undef msync
1195# define msync(a,b,c) ((errno = ENOSYS), -1) 1504# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1505#else
1506
1507int
1508eio__msync (void *mem, size_t len, int flags)
1509{
1510 eio_page_align (&mem, &len);
1511
1512 if (EIO_MS_ASYNC != MS_SYNC
1513 || EIO_MS_INVALIDATE != MS_INVALIDATE
1514 || EIO_MS_SYNC != MS_SYNC)
1515 {
1516 flags = 0
1517 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1518 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1519 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1520 }
1521
1522 return msync (mem, len, flags);
1523}
1524
1196#endif 1525#endif
1197 1526
1198int 1527int
1199eio__mtouch (void *mem, size_t len, int flags) 1528eio__mtouch (void *mem, size_t len, int flags)
1200{ 1529{
1530 eio_page_align (&mem, &len);
1531
1532 {
1201 intptr_t addr = (intptr_t)mem; 1533 intptr_t addr = (intptr_t)mem;
1202 intptr_t end = addr + len; 1534 intptr_t end = addr + len;
1203#ifdef PAGESIZE 1535 intptr_t page = eio_pagesize ();
1204 const intptr_t page = PAGESIZE;
1205#else
1206 static intptr_t page;
1207 1536
1208 if (!page)
1209 page = sysconf (_SC_PAGESIZE);
1210#endif
1211
1212 addr &= ~(page - 1); /* assume page size is always a power of two */
1213
1214 if (addr < end) 1537 if (addr < end)
1215 if (flags) /* modify */ 1538 if (flags & EIO_MT_MODIFY) /* modify */
1216 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1539 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1217 else 1540 else
1218 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1541 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1542 }
1219 1543
1220 return 0; 1544 return 0;
1221} 1545}
1222 1546
1223/*****************************************************************************/ 1547/*****************************************************************************/
1257 if (req) 1581 if (req)
1258 break; 1582 break;
1259 1583
1260 ++idle; 1584 ++idle;
1261 1585
1262 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1586 ts.tv_sec = time (0) + idle_timeout;
1263 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1587 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1264 { 1588 {
1265 if (idle > max_idle) 1589 if (idle > max_idle)
1266 { 1590 {
1267 --idle; 1591 --idle;
1346 return 0; \ 1670 return 0; \
1347 } 1671 }
1348 1672
1349static void eio_execute (etp_worker *self, eio_req *req) 1673static void eio_execute (etp_worker *self, eio_req *req)
1350{ 1674{
1351 errno = 0;
1352
1353 switch (req->type) 1675 switch (req->type)
1354 { 1676 {
1355 case EIO_READ: ALLOC (req->size); 1677 case EIO_READ: ALLOC (req->size);
1356 req->result = req->offs >= 0 1678 req->result = req->offs >= 0
1357 ? pread (req->int1, req->ptr2, req->size, req->offs) 1679 ? pread (req->int1, req->ptr2, req->size, req->offs)
1368 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1690 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1369 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1691 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1370 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1692 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1371 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1693 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1372 1694
1695 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1696 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1697 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1698 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1699
1373 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1700 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; 1701 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; 1702 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; 1703 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; 1704 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1384 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1711 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1385 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1712 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; 1713 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1387 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1714 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1388 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1715 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; 1716 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1390 1717
1391 case EIO_READLINK: ALLOC (NAME_MAX); 1718 case EIO_READLINK: ALLOC (PATH_MAX);
1392 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1719 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1393 1720
1394 case EIO_SYNC: req->result = 0; sync (); break; 1721 case EIO_SYNC: req->result = 0; sync (); break;
1395 case EIO_FSYNC: req->result = fsync (req->int1); break; 1722 case EIO_FSYNC: req->result = fsync (req->int1); break;
1396 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1723 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1397 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1724 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1398 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1725 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1726 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1727 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; 1728 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1400 1729
1401 case EIO_READDIR: eio__scandir (req, self); break; 1730 case EIO_READDIR: eio__scandir (req, self); break;
1402 1731
1403 case EIO_BUSY: 1732 case EIO_BUSY:
1404#ifdef _WIN32 1733#ifdef _WIN32
1405 Sleep (req->nv1 * 1000.); 1734 Sleep (req->nv1 * 1e3);
1406#else 1735#else
1407 { 1736 {
1408 struct timeval tv; 1737 struct timeval tv;
1409 1738
1410 tv.tv_sec = req->nv1; 1739 tv.tv_sec = req->nv1;
1411 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1740 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1412 1741
1413 req->result = select (0, 0, 0, 0, &tv); 1742 req->result = select (0, 0, 0, 0, &tv);
1414 } 1743 }
1415#endif 1744#endif
1416 break; 1745 break;
1431 times = tv; 1760 times = tv;
1432 } 1761 }
1433 else 1762 else
1434 times = 0; 1763 times = 0;
1435 1764
1436
1437 req->result = req->type == EIO_FUTIME 1765 req->result = req->type == EIO_FUTIME
1438 ? futimes (req->int1, times) 1766 ? futimes (req->int1, times)
1439 : utimes (req->ptr1, times); 1767 : utimes (req->ptr1, times);
1440 } 1768 }
1441 break; 1769 break;
1450 case EIO_CUSTOM: 1778 case EIO_CUSTOM:
1451 ((void (*)(eio_req *))req->feed) (req); 1779 ((void (*)(eio_req *))req->feed) (req);
1452 break; 1780 break;
1453 1781
1454 default: 1782 default:
1783 errno = ENOSYS;
1455 req->result = -1; 1784 req->result = -1;
1456 break; 1785 break;
1457 } 1786 }
1458 1787
1459 req->errorno = errno; 1788 req->errorno = errno;
1489eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 1818eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1490{ 1819{
1491 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 1820 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1492} 1821}
1493 1822
1823eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1824{
1825 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1826}
1827
1828eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1829{
1830 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1831}
1832
1494eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 1833eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1495{ 1834{
1496 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 1835 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1497} 1836}
1498 1837
1522} 1861}
1523 1862
1524eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1863eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1525{ 1864{
1526 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1865 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1866}
1867
1868eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1869{
1870 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1527} 1871}
1528 1872
1529eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1873eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1530{ 1874{
1531 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1875 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) 1949eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1606{ 1950{
1607 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1951 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1608} 1952}
1609 1953
1954eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1955{
1956 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1957}
1958
1610eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1959eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1611{ 1960{
1612 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1961 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1613} 1962}
1614 1963
1622 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 1971 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1623} 1972}
1624 1973
1625eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1974eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1626{ 1975{
1627 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1976 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1628} 1977}
1629 1978
1630static eio_req * 1979static eio_req *
1631eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1980eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1632{ 1981{
1716/* misc garbage */ 2065/* misc garbage */
1717 2066
1718ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2067ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1719{ 2068{
1720 etp_worker wrk; 2069 etp_worker wrk;
2070 ssize_t ret;
1721 2071
1722 wrk.dbuf = 0; 2072 wrk.dbuf = 0;
1723 2073
1724 eio__sendfile (ofd, ifd, offset, count, &wrk); 2074 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1725 2075
1726 if (wrk.dbuf) 2076 if (wrk.dbuf)
1727 free (wrk.dbuf); 2077 free (wrk.dbuf);
1728}
1729 2078
2079 return ret;
2080}
2081

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines