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 1.65 by root, Mon May 30 12:56:50 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
58#ifndef EIO_FINISH 63#ifndef EIO_FINISH
70#ifdef _WIN32 75#ifdef _WIN32
71 76
72 /*doh*/ 77 /*doh*/
73#else 78#else
74 79
75# include "config.h"
76# include <sys/time.h> 80# include <sys/time.h>
77# include <sys/select.h> 81# include <sys/select.h>
78# include <unistd.h> 82# include <unistd.h>
79# include <utime.h> 83# include <utime.h>
80# include <signal.h> 84# include <signal.h>
81# include <dirent.h> 85# include <dirent.h>
82 86
87#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
88# include <sys/mman.h>
89#endif
90
83/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 91/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) 92# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
93# define _DIRENT_HAVE_D_TYPE /* sigh */
85# define D_INO(de) (de)->d_fileno 94# define D_INO(de) (de)->d_fileno
86# define _DIRENT_HAVE_D_TYPE /* sigh */ 95# define D_NAMLEN(de) (de)->d_namlen
87# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 96# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
88# define D_INO(de) (de)->d_ino 97# define D_INO(de) (de)->d_ino
89# endif 98# endif
99
100#ifdef _D_EXACT_NAMLEN
101# undef D_NAMLEN
102# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
103#endif
90 104
91# ifdef _DIRENT_HAVE_D_TYPE 105# ifdef _DIRENT_HAVE_D_TYPE
92# define D_TYPE(de) (de)->d_type 106# define D_TYPE(de) (de)->d_type
93# endif 107# endif
94 108
99#endif 113#endif
100 114
101#if HAVE_SENDFILE 115#if HAVE_SENDFILE
102# if __linux 116# if __linux
103# include <sys/sendfile.h> 117# include <sys/sendfile.h>
104# elif __freebsd 118# elif __FreeBSD__ || defined __APPLE__
105# include <sys/socket.h> 119# include <sys/socket.h>
106# include <sys/uio.h> 120# include <sys/uio.h>
107# elif __hpux 121# elif __hpux
108# include <sys/socket.h> 122# include <sys/socket.h>
109# elif __solaris /* not yet */ 123# elif __solaris
110# include <sys/sendfile.h> 124# include <sys/sendfile.h>
111# else 125# else
112# error sendfile support requested but not available 126# error sendfile support requested but not available
113# endif 127# endif
114#endif 128#endif
115 129
116#ifndef D_TYPE 130#ifndef D_TYPE
117# define D_TYPE(de) 0 131# define D_TYPE(de) 0
118#endif 132#endif
119
120#ifndef D_INO 133#ifndef D_INO
121# define D_INO(de) 0 134# define D_INO(de) 0
122#endif 135#endif
123 136#ifndef D_NAMLEN
124/* number of seconds after which an idle threads exit */ 137# define D_NAMLEN(de) strlen ((de)->d_name)
125#define IDLE_TIMEOUT 10 138#endif
126 139
127/* used for struct dirent, AIX doesn't provide it */ 140/* used for struct dirent, AIX doesn't provide it */
128#ifndef NAME_MAX 141#ifndef NAME_MAX
129# define NAME_MAX 4096 142# define NAME_MAX 4096
143#endif
144
145/* used for readlink etc. */
146#ifndef PATH_MAX
147# define PATH_MAX 4096
130#endif 148#endif
131 149
132/* buffer size for various temporary buffers */ 150/* buffer size for various temporary buffers */
133#define EIO_BUFSIZE 65536 151#define EIO_BUFSIZE 65536
134 152
187 205
188/*****************************************************************************/ 206/*****************************************************************************/
189 207
190#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
191 209
192/* calculcate time difference in ~1/EIO_TICKS of a second */ 210/* calculate time difference in ~1/EIO_TICKS of a second */
193static int tvdiff (struct timeval *tv1, struct timeval *tv2) 211static int tvdiff (struct timeval *tv1, struct timeval *tv2)
194{ 212{
195 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 213 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
196 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 214 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
197} 215}
205static unsigned int max_poll_reqs; /* reslock */ 223static unsigned int max_poll_reqs; /* reslock */
206 224
207static volatile unsigned int nreqs; /* reqlock */ 225static volatile unsigned int nreqs; /* reqlock */
208static volatile unsigned int nready; /* reqlock */ 226static volatile unsigned int nready; /* reqlock */
209static volatile unsigned int npending; /* reqlock */ 227static volatile unsigned int npending; /* reqlock */
210static volatile unsigned int max_idle = 4; 228static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
229static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
211 230
212static mutex_t wrklock = X_MUTEX_INIT; 231static xmutex_t wrklock;
213static mutex_t reslock = X_MUTEX_INIT; 232static xmutex_t reslock;
214static mutex_t reqlock = X_MUTEX_INIT; 233static xmutex_t reqlock;
215static cond_t reqwait = X_COND_INIT; 234static xcond_t reqwait;
216 235
217#if !HAVE_PREADWRITE 236#if !HAVE_PREADWRITE
218/* 237/*
219 * make our pread/pwrite emulation safe against themselves, but not against 238 * make our pread/pwrite emulation safe against themselves, but not against
220 * normal read/write by using a mutex. slows down execution a lot, 239 * normal read/write by using a mutex. slows down execution a lot,
221 * but that's your problem, not mine. 240 * but that's your problem, not mine.
222 */ 241 */
223static mutex_t preadwritelock = X_MUTEX_INIT; 242static xmutex_t preadwritelock = X_MUTEX_INIT;
224#endif 243#endif
225 244
226typedef struct etp_worker 245typedef struct etp_worker
227{ 246{
228 /* locked by wrklock */ 247 /* locked by wrklock */
229 struct etp_worker *prev, *next; 248 struct etp_worker *prev, *next;
230 249
231 thread_t tid; 250 xthread_t tid;
232 251
233 /* locked by reslock, reqlock or wrklock */ 252 /* locked by reslock, reqlock or wrklock */
234 ETP_REQ *req; /* currently processed request */ 253 ETP_REQ *req; /* currently processed request */
235 254
236 ETP_WORKER_COMMON 255 ETP_WORKER_COMMON
350 } 369 }
351 370
352 abort (); 371 abort ();
353} 372}
354 373
374static void etp_thread_init (void)
375{
376 X_MUTEX_CREATE (wrklock);
377 X_MUTEX_CREATE (reslock);
378 X_MUTEX_CREATE (reqlock);
379 X_COND_CREATE (reqwait);
380}
381
355static void etp_atfork_prepare (void) 382static void etp_atfork_prepare (void)
356{ 383{
357 X_LOCK (wrklock); 384 X_LOCK (wrklock);
358 X_LOCK (reqlock); 385 X_LOCK (reqlock);
359 X_LOCK (reslock); 386 X_LOCK (reslock);
397 idle = 0; 424 idle = 0;
398 nreqs = 0; 425 nreqs = 0;
399 nready = 0; 426 nready = 0;
400 npending = 0; 427 npending = 0;
401 428
402 etp_atfork_parent (); 429 etp_thread_init ();
403} 430}
404 431
405static void 432static void
406etp_once_init (void) 433etp_once_init (void)
407{ 434{
435 etp_thread_init ();
408 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 436 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
409} 437}
410 438
411static int 439static int
412etp_init (void (*want_poll)(void), void (*done_poll)(void)) 440etp_init (void (*want_poll)(void), void (*done_poll)(void))
589} 617}
590 618
591static void etp_set_max_poll_time (double nseconds) 619static void etp_set_max_poll_time (double nseconds)
592{ 620{
593 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 621 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
594 max_poll_time = nseconds; 622 max_poll_time = nseconds * EIO_TICKS;
595 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 623 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
596} 624}
597 625
598static void etp_set_max_poll_reqs (unsigned int maxreqs) 626static void etp_set_max_poll_reqs (unsigned int maxreqs)
599{ 627{
603} 631}
604 632
605static void etp_set_max_idle (unsigned int nthreads) 633static void etp_set_max_idle (unsigned int nthreads)
606{ 634{
607 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 635 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
608 max_idle = nthreads <= 0 ? 1 : nthreads; 636 max_idle = nthreads;
637 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
638}
639
640static void etp_set_idle_timeout (unsigned int seconds)
641{
642 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
643 idle_timeout = seconds;
609 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 644 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
610} 645}
611 646
612static void etp_set_min_parallel (unsigned int nthreads) 647static void etp_set_min_parallel (unsigned int nthreads)
613{ 648{
741void eio_set_max_idle (unsigned int nthreads) 776void eio_set_max_idle (unsigned int nthreads)
742{ 777{
743 etp_set_max_idle (nthreads); 778 etp_set_max_idle (nthreads);
744} 779}
745 780
781void eio_set_idle_timeout (unsigned int seconds)
782{
783 etp_set_idle_timeout (seconds);
784}
785
746void eio_set_min_parallel (unsigned int nthreads) 786void eio_set_min_parallel (unsigned int nthreads)
747{ 787{
748 etp_set_min_parallel (nthreads); 788 etp_set_min_parallel (nthreads);
749} 789}
750 790
798 838
799 return res; 839 return res;
800} 840}
801#endif 841#endif
802 842
803#ifndef HAVE_FUTIMES 843#ifndef HAVE_UTIMES
804 844
805# undef utimes 845# undef utimes
806# undef futimes
807# define utimes(path,times) eio__utimes (path, times) 846# define utimes(path,times) eio__utimes (path, times)
808# define futimes(fd,times) eio__futimes (fd, times)
809 847
810static int 848static int
811eio__utimes (const char *filename, const struct timeval times[2]) 849eio__utimes (const char *filename, const struct timeval times[2])
812{ 850{
813 if (times) 851 if (times)
820 return utime (filename, &buf); 858 return utime (filename, &buf);
821 } 859 }
822 else 860 else
823 return utime (filename, 0); 861 return utime (filename, 0);
824} 862}
863
864#endif
865
866#ifndef HAVE_FUTIMES
867
868# undef futimes
869# define futimes(fd,times) eio__futimes (fd, times)
825 870
826static int eio__futimes (int fd, const struct timeval tv[2]) 871static int eio__futimes (int fd, const struct timeval tv[2])
827{ 872{
828 errno = ENOSYS; 873 errno = ENOSYS;
829 return -1; 874 return -1;
858 if (!res || errno != ENOSYS) 903 if (!res || errno != ENOSYS)
859 return res; 904 return res;
860#endif 905#endif
861 906
862 /* even though we could play tricks with the flags, it's better to always 907 /* 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 */ 908 * call fdatasync, as that matches the expectation of its users best */
864 return fdatasync (fd); 909 return fdatasync (fd);
865} 910}
866 911
867#if !HAVE_READAHEAD 912#if !HAVE_READAHEAD
868# undef readahead 913# undef readahead
900 945
901#if HAVE_SENDFILE 946#if HAVE_SENDFILE
902# if __linux 947# if __linux
903 res = sendfile (ofd, ifd, &offset, count); 948 res = sendfile (ofd, ifd, &offset, count);
904 949
905# elif __freebsd 950# elif __FreeBSD__
906 /* 951 /*
907 * Of course, the freebsd sendfile is a dire hack with no thoughts 952 * Of course, the freebsd sendfile is a dire hack with no thoughts
908 * wasted on making it similar to other I/O functions. 953 * wasted on making it similar to other I/O functions.
909 */ 954 */
910 { 955 {
911 off_t sbytes; 956 off_t sbytes;
912 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 957 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
913 958
914 if (res < 0 && sbytes) 959 #if 0 /* according to the manpage, this is correct, but broken behaviour */
915 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 960 /* freebsd' sendfile will return 0 on success */
961 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
962 /* not on e.g. EIO or EPIPE - sounds broken */
963 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
964 res = sbytes;
965 #endif
966
967 /* according to source inspection, this is correct, and useful behaviour */
968 if (sbytes)
969 res = sbytes;
970 }
971
972# elif defined (__APPLE__)
973
974 {
975 off_t sbytes = count;
976 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
977
978 /* according to the manpage, sbytes is always valid */
979 if (sbytes)
916 res = sbytes; 980 res = sbytes;
917 } 981 }
918 982
919# elif __hpux 983# elif __hpux
920 res = sendfile (ofd, ifd, offset, count, 0, 0); 984 res = sendfile (ofd, ifd, offset, count, 0, 0);
934 if (res < 0 && sbytes) 998 if (res < 0 && sbytes)
935 res = sbytes; 999 res = sbytes;
936 } 1000 }
937 1001
938# endif 1002# endif
1003
1004#elif defined (_WIN32)
1005
1006 /* does not work, just for documentation of what would need to be done */
1007 {
1008 HANDLE h = TO_SOCKET (ifd);
1009 SetFilePointer (h, offset, 0, FILE_BEGIN);
1010 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1011 }
1012
939#else 1013#else
940 res = -1; 1014 res = -1;
941 errno = ENOSYS; 1015 errno = ENOSYS;
942#endif 1016#endif
943 1017
944 if (res < 0 1018 if (res < 0
945 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1019 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1020 /* BSDs */
1021#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1022 || errno == ENOTSUP
1023#endif
1024 || errno == EOPNOTSUPP /* BSDs */
946#if __solaris 1025#if __solaris
947 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1026 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
948#endif 1027#endif
949 ) 1028 )
950 ) 1029 )
981 } 1060 }
982 1061
983 return res; 1062 return res;
984} 1063}
985 1064
986static int 1065static signed char
987eio_dent_cmp (const void *a_, const void *b_) 1066eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
988{ 1067{
989 const eio_dirent *a = (const eio_dirent *)a_; 1068 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_; 1069 : a->inode < b->inode ? -1
1070 : a->inode > b->inode ? 1
1071 : 0;
1072}
991 1073
992 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score 1074#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 */ 1075
1076#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1077#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1078
1079static void
1080eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1081{
1082 unsigned char bits [9 + sizeof (ino_t) * 8];
1083 unsigned char *bit = bits;
1084
1085 assert (CHAR_BIT == 8);
1086 assert (sizeof (eio_dirent) * 8 < 256);
1087 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1088 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1089
1090 if (size <= EIO_SORT_FAST)
1091 return;
1092
1093 /* first prepare an array of bits to test in our radix sort */
1094 /* try to take endianness into account, as well as differences in ino_t sizes */
1095 /* inode_bits must contain all inodes ORed together */
1096 /* which is used to skip bits that are 0 everywhere, which is very common */
1097 {
1098 ino_t endianness;
1099 int i, j;
1100
1101 /* we store the byte offset of byte n into byte n of "endianness" */
1102 for (i = 0; i < sizeof (ino_t); ++i)
1103 ((unsigned char *)&endianness)[i] = i;
1104
1105 *bit++ = 0;
1106
1107 for (i = 0; i < sizeof (ino_t); ++i)
1108 {
1109 /* shifting off the byte offsets out of "endianness" */
1110 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1111 endianness >>= 8;
1112
1113 for (j = 0; j < 8; ++j)
1114 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1115 *bit++ = offs + j;
1116 }
1117
1118 for (j = 0; j < 8; ++j)
1119 if (score_bits & (1 << j))
1120 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1121 }
1122
1123 /* now actually do the sorting (a variant of MSD radix sort) */
1124 {
1125 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1126 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1127 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1128 int stk_idx = 0;
1129
1130 base_stk [stk_idx] = dents;
1131 end_stk [stk_idx] = dents + size;
1132 bit_stk [stk_idx] = bit - 1;
1133
1134 do
1135 {
1136 base = base_stk [stk_idx];
1137 end = end_stk [stk_idx];
1138 bit = bit_stk [stk_idx];
1139
1140 for (;;)
1141 {
1142 unsigned char O = *bit >> 3;
1143 unsigned char M = 1 << (*bit & 7);
1144
1145 eio_dirent *a = base;
1146 eio_dirent *b = end;
1147
1148 if (b - a < EIO_SORT_CUTOFF)
1149 break;
1150
1151 /* now bit-partition the array on the bit */
1152 /* this ugly asymmetric loop seems to perform much better than typical */
1153 /* partition algos found in the literature */
1154 do
1155 if (!(((unsigned char *)a)[O] & M))
1156 ++a;
1157 else if (!(((unsigned char *)--b)[O] & M))
1158 {
1159 eio_dirent tmp = *a; *a = *b; *b = tmp;
1160 ++a;
1161 }
1162 while (b > a);
1163
1164 /* next bit, or stop, if no bits left in this path */
1165 if (!*--bit)
1166 break;
1167
1168 base_stk [stk_idx] = a;
1169 end_stk [stk_idx] = end;
1170 bit_stk [stk_idx] = bit;
1171 ++stk_idx;
1172
1173 end = a;
1174 }
1175 }
1176 while (stk_idx--);
1177 }
1178}
1179
1180static void
1181eio_dent_insertion_sort (eio_dirent *dents, int size)
1182{
1183 /* first move the smallest element to the front, to act as a sentinel */
1184 {
1185 int i;
1186 eio_dirent *min = dents;
1187
1188 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1189 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1190 if (EIO_DENT_CMP (dents [i], <, *min))
1191 min = &dents [i];
1192
1193 /* swap elements 0 and j (minimum) */
1194 {
1195 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1196 }
1197 }
1198
1199 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1200 {
1201 eio_dirent *i, *j;
1202
1203 for (i = dents + 1; i < dents + size; ++i)
1204 {
1205 eio_dirent value = *i;
1206
1207 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1208 j [1] = j [0];
1209
1210 j [1] = value;
1211 }
1212 }
1213}
1214
1215static void
1216eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1217{
1218 if (size <= 1)
1219 return; /* our insertion sort relies on size > 0 */
1220
1221 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1222 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1223 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1224
1225 /* use an insertion sort at the end, or for small arrays, */
1226 /* as insertion sort is more efficient for small partitions */
1227 eio_dent_insertion_sort (dents, size);
994} 1228}
995 1229
996/* read a full directory */ 1230/* read a full directory */
997static void 1231static void
998eio__scandir (eio_req *req, etp_worker *self) 1232eio__scandir (eio_req *req, etp_worker *self)
999{ 1233{
1000 DIR *dirp; 1234 DIR *dirp;
1001 EIO_STRUCT_DIRENT *entp; 1235 EIO_STRUCT_DIRENT *entp;
1002 unsigned char *name, *names; 1236 char *name, *names;
1003 int namesalloc = 4096; 1237 int namesalloc = 4096;
1004 int namesoffs = 0; 1238 int namesoffs = 0;
1005 int flags = req->int1; 1239 int flags = req->int1;
1006 eio_dirent *dents = 0; 1240 eio_dirent *dents = 0;
1007 int dentalloc = 128; 1241 int dentalloc = 128;
1008 int dentoffs = 0; 1242 int dentoffs = 0;
1243 ino_t inode_bits = 0;
1009 1244
1010 req->result = -1; 1245 req->result = -1;
1011 1246
1012 if (!(flags & EIO_READDIR_DENTS)) 1247 if (!(flags & EIO_READDIR_DENTS))
1013 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1248 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1014 1249
1015 X_LOCK (wrklock); 1250 X_LOCK (wrklock);
1016 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1251 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1017 self->dirp = dirp = opendir (req->ptr1); 1252 self->dirp = dirp = opendir (req->ptr1);
1253
1018 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1254 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; 1255 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1256 req->ptr2 = names = malloc (namesalloc);
1021 X_UNLOCK (wrklock); 1257 X_UNLOCK (wrklock);
1022 1258
1023 if (dirp && names && (!flags || dents)) 1259 if (dirp && names && (!flags || dents))
1024 for (;;) 1260 for (;;)
1025 { 1261 {
1033 1269
1034 /* sort etc. */ 1270 /* sort etc. */
1035 req->int1 = flags; 1271 req->int1 = flags;
1036 req->result = dentoffs; 1272 req->result = dentoffs;
1037 1273
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 1274 if (flags & EIO_READDIR_STAT_ORDER)
1047 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1275 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) 1276 else if (flags & EIO_READDIR_DIRS_FIRST)
1277 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1278 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1279 else
1053 { 1280 {
1054 /* in this case, all is known, and we just put dirs first and sort them */ 1281 /* in this case, all is known, and we just put dirs first and sort them */
1055 eio_dirent *ent = dents + dentoffs; 1282 eio_dirent *oth = dents + dentoffs;
1056 eio_dirent *dir = dents; 1283 eio_dirent *dir = dents;
1057 1284
1285 /* now partition dirs to the front, and non-dirs to the back */
1286 /* by walking from both sides and swapping if necessary */
1058 while (ent > dir) 1287 while (oth > dir)
1059 { 1288 {
1060 if (dir->type == DT_DIR) 1289 if (dir->type == EIO_DT_DIR)
1061 ++dir; 1290 ++dir;
1062 else
1063 {
1064 --ent;
1065
1066 if (ent->type == DT_DIR) 1291 else if ((--oth)->type == EIO_DT_DIR)
1067 { 1292 {
1068 eio_dirent tmp = *dir; 1293 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1069 *dir = *ent;
1070 *ent = tmp;
1071 1294
1072 ++dir; 1295 ++dir;
1073 } 1296 }
1074 } 1297 }
1298
1299 /* now sort the dirs only (dirs all have the same score) */
1300 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1075 } 1301 }
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 1302
1091 break; 1303 break;
1092 } 1304 }
1093 1305
1094 /* now add the entry to our list(s) */ 1306 /* now add the entry to our list(s) */
1095 name = entp->d_name; 1307 name = entp->d_name;
1096 1308
1097 /* skip . and .. entries */ 1309 /* skip . and .. entries */
1098 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1310 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1099 { 1311 {
1100 int len = strlen (name) + 1; 1312 int len = D_NAMLEN (entp) + 1;
1101 1313
1102 while (expect_false (namesoffs + len > namesalloc)) 1314 while (expect_false (namesoffs + len > namesalloc))
1103 { 1315 {
1104 namesalloc *= 2; 1316 namesalloc *= 2;
1105 X_LOCK (wrklock); 1317 X_LOCK (wrklock);
1106 req->ptr1 = names = realloc (names, namesalloc); 1318 req->ptr2 = names = realloc (names, namesalloc);
1107 X_UNLOCK (wrklock); 1319 X_UNLOCK (wrklock);
1108 1320
1109 if (!names) 1321 if (!names)
1110 break; 1322 break;
1111 } 1323 }
1118 1330
1119 if (expect_false (dentoffs == dentalloc)) 1331 if (expect_false (dentoffs == dentalloc))
1120 { 1332 {
1121 dentalloc *= 2; 1333 dentalloc *= 2;
1122 X_LOCK (wrklock); 1334 X_LOCK (wrklock);
1123 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1335 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1124 X_UNLOCK (wrklock); 1336 X_UNLOCK (wrklock);
1125 1337
1126 if (!dents) 1338 if (!dents)
1127 break; 1339 break;
1128 } 1340 }
1129 1341
1130 ent = dents + dentoffs; 1342 ent = dents + dentoffs;
1131 1343
1132 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1344 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1133 ent->namelen = len - 1; 1345 ent->namelen = len - 1;
1134 ent->inode = D_INO (entp); 1346 ent->inode = D_INO (entp);
1347
1348 inode_bits |= ent->inode;
1135 1349
1136 switch (D_TYPE (entp)) 1350 switch (D_TYPE (entp))
1137 { 1351 {
1138 default: 1352 default:
1139 ent->type = EIO_DT_UNKNOWN; 1353 ent->type = EIO_DT_UNKNOWN;
1144 case DT_FIFO: ent->type = EIO_DT_FIFO; break; 1358 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1145 #endif 1359 #endif
1146 #ifdef DT_CHR 1360 #ifdef DT_CHR
1147 case DT_CHR: ent->type = EIO_DT_CHR; break; 1361 case DT_CHR: ent->type = EIO_DT_CHR; break;
1148 #endif 1362 #endif
1363 #ifdef DT_MPC
1364 case DT_MPC: ent->type = EIO_DT_MPC; break;
1365 #endif
1149 #ifdef DT_DIR 1366 #ifdef DT_DIR
1150 case DT_DIR: ent->type = EIO_DT_DIR; break; 1367 case DT_DIR: ent->type = EIO_DT_DIR; break;
1368 #endif
1369 #ifdef DT_NAM
1370 case DT_NAM: ent->type = EIO_DT_NAM; break;
1151 #endif 1371 #endif
1152 #ifdef DT_BLK 1372 #ifdef DT_BLK
1153 case DT_BLK: ent->type = EIO_DT_BLK; break; 1373 case DT_BLK: ent->type = EIO_DT_BLK; break;
1154 #endif 1374 #endif
1375 #ifdef DT_MPB
1376 case DT_MPB: ent->type = EIO_DT_MPB; break;
1377 #endif
1155 #ifdef DT_REG 1378 #ifdef DT_REG
1156 case DT_REG: ent->type = EIO_DT_REG; break; 1379 case DT_REG: ent->type = EIO_DT_REG; break;
1380 #endif
1381 #ifdef DT_NWK
1382 case DT_NWK: ent->type = EIO_DT_NWK; break;
1383 #endif
1384 #ifdef DT_CMP
1385 case DT_CMP: ent->type = EIO_DT_CMP; break;
1157 #endif 1386 #endif
1158 #ifdef DT_LNK 1387 #ifdef DT_LNK
1159 case DT_LNK: ent->type = EIO_DT_LNK; break; 1388 case DT_LNK: ent->type = EIO_DT_LNK; break;
1160 #endif 1389 #endif
1161 #ifdef DT_SOCK 1390 #ifdef DT_SOCK
1162 case DT_SOCK: ent->type = EIO_DT_SOCK; break; 1391 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1163 #endif 1392 #endif
1393 #ifdef DT_DOOR
1394 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1395 #endif
1164 #ifdef DT_WHT 1396 #ifdef DT_WHT
1165 case DT_WHT: ent->type = EIO_DT_WHT; break; 1397 case DT_WHT: ent->type = EIO_DT_WHT; break;
1166 #endif 1398 #endif
1167 } 1399 }
1168 1400
1169 ent->score = 0; 1401 ent->score = 7;
1170 1402
1171 if (flags & EIO_READDIR_DIRS_FIRST) 1403 if (flags & EIO_READDIR_DIRS_FIRST)
1172 { 1404 {
1173 if (ent->type == EIO_DT_UNKNOWN) 1405 if (ent->type == EIO_DT_UNKNOWN)
1174 { 1406 {
1175 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1407 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1176 ent->score = 98; 1408 ent->score = 1;
1177 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1409 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 */ 1410 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1179 } 1411 }
1180 else if (ent->type == DT_DIR) 1412 else if (ent->type == EIO_DT_DIR)
1181 ent->score = 100; 1413 ent->score = 0;
1182 } 1414 }
1183 } 1415 }
1184 1416
1185 namesoffs += len; 1417 namesoffs += len;
1186 ++dentoffs; 1418 ++dentoffs;
1187 } 1419 }
1420
1421 if (EIO_CANCELLED (req))
1422 {
1423 errno = ECANCELED;
1424 break;
1425 }
1188 } 1426 }
1189 else
1190 req->result = -1;
1191} 1427}
1428
1429#ifdef PAGESIZE
1430# define eio_pagesize() PAGESIZE
1431#else
1432static intptr_t
1433eio_pagesize (void)
1434{
1435 static intptr_t page;
1436
1437 if (!page)
1438 page = sysconf (_SC_PAGESIZE);
1439
1440 return page;
1441}
1442#endif
1443
1444static void
1445eio_page_align (void **addr, size_t *length)
1446{
1447 intptr_t mask = eio_pagesize () - 1;
1448
1449 /* round down addr */
1450 intptr_t adj = mask & (intptr_t)*addr;
1451
1452 *addr = (void *)((intptr_t)*addr - adj);
1453 *length += adj;
1454
1455 /* round up length */
1456 *length = (*length + mask) & ~mask;
1457}
1458
1459#if !_POSIX_MEMLOCK
1460# define eio__mlockall(a) ((errno = ENOSYS), -1)
1461#else
1462
1463static int
1464eio__mlockall (int flags)
1465{
1466 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1467 extern int mallopt (int, int);
1468 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1469 #endif
1470
1471 if (EIO_MCL_CURRENT != MCL_CURRENT
1472 || EIO_MCL_FUTURE != MCL_FUTURE)
1473 {
1474 flags = 0
1475 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1476 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1477 }
1478
1479 return mlockall (flags);
1480}
1481#endif
1482
1483#if !_POSIX_MEMLOCK_RANGE
1484# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1485#else
1486
1487static int
1488eio__mlock (void *addr, size_t length)
1489{
1490 eio_page_align (&addr, &length);
1491
1492 return mlock (addr, length);
1493}
1494
1495#endif
1192 1496
1193#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1497#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1194# undef msync
1195# define msync(a,b,c) ((errno = ENOSYS), -1) 1498# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1499#else
1500
1501int
1502eio__msync (void *mem, size_t len, int flags)
1503{
1504 eio_page_align (&mem, &len);
1505
1506 if (EIO_MS_ASYNC != MS_SYNC
1507 || EIO_MS_INVALIDATE != MS_INVALIDATE
1508 || EIO_MS_SYNC != MS_SYNC)
1509 {
1510 flags = 0
1511 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1512 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1513 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1514 }
1515
1516 return msync (mem, len, flags);
1517}
1518
1196#endif 1519#endif
1197 1520
1198int 1521int
1199eio__mtouch (void *mem, size_t len, int flags) 1522eio__mtouch (void *mem, size_t len, int flags)
1200{ 1523{
1524 eio_page_align (&mem, &len);
1525
1526 {
1201 intptr_t addr = (intptr_t)mem; 1527 intptr_t addr = (intptr_t)mem;
1202 intptr_t end = addr + len; 1528 intptr_t end = addr + len;
1203#ifdef PAGESIZE 1529 intptr_t page = eio_pagesize ();
1204 const intptr_t page = PAGESIZE;
1205#else
1206 static intptr_t page;
1207 1530
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) 1531 if (addr < end)
1215 if (flags) /* modify */ 1532 if (flags & EIO_MT_MODIFY) /* modify */
1216 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1533 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1217 else 1534 else
1218 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1535 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1536 }
1219 1537
1220 return 0; 1538 return 0;
1221} 1539}
1222 1540
1223/*****************************************************************************/ 1541/*****************************************************************************/
1257 if (req) 1575 if (req)
1258 break; 1576 break;
1259 1577
1260 ++idle; 1578 ++idle;
1261 1579
1262 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1580 ts.tv_sec = time (0) + idle_timeout;
1263 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1581 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1264 { 1582 {
1265 if (idle > max_idle) 1583 if (idle > max_idle)
1266 { 1584 {
1267 --idle; 1585 --idle;
1346 return 0; \ 1664 return 0; \
1347 } 1665 }
1348 1666
1349static void eio_execute (etp_worker *self, eio_req *req) 1667static void eio_execute (etp_worker *self, eio_req *req)
1350{ 1668{
1351 errno = 0;
1352
1353 switch (req->type) 1669 switch (req->type)
1354 { 1670 {
1355 case EIO_READ: ALLOC (req->size); 1671 case EIO_READ: ALLOC (req->size);
1356 req->result = req->offs >= 0 1672 req->result = req->offs >= 0
1357 ? pread (req->int1, req->ptr2, req->size, req->offs) 1673 ? pread (req->int1, req->ptr2, req->size, req->offs)
1368 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1684 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1369 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1685 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1370 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1686 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1371 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1687 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1372 1688
1689 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1690 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1691 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1692 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1693
1373 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1694 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; 1695 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; 1696 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; 1697 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; 1698 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1384 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1705 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1385 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1706 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; 1707 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1387 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1708 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1388 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1709 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; 1710 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1390 1711
1391 case EIO_READLINK: ALLOC (NAME_MAX); 1712 case EIO_READLINK: ALLOC (PATH_MAX);
1392 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1713 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1393 1714
1394 case EIO_SYNC: req->result = 0; sync (); break; 1715 case EIO_SYNC: req->result = 0; sync (); break;
1395 case EIO_FSYNC: req->result = fsync (req->int1); break; 1716 case EIO_FSYNC: req->result = fsync (req->int1); break;
1396 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1717 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1397 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1718 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; 1719 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1720 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1721 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; 1722 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1400 1723
1401 case EIO_READDIR: eio__scandir (req, self); break; 1724 case EIO_READDIR: eio__scandir (req, self); break;
1402 1725
1403 case EIO_BUSY: 1726 case EIO_BUSY:
1404#ifdef _WIN32 1727#ifdef _WIN32
1405 Sleep (req->nv1 * 1000.); 1728 Sleep (req->nv1 * 1e3);
1406#else 1729#else
1407 { 1730 {
1408 struct timeval tv; 1731 struct timeval tv;
1409 1732
1410 tv.tv_sec = req->nv1; 1733 tv.tv_sec = req->nv1;
1411 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1734 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1412 1735
1413 req->result = select (0, 0, 0, 0, &tv); 1736 req->result = select (0, 0, 0, 0, &tv);
1414 } 1737 }
1415#endif 1738#endif
1416 break; 1739 break;
1431 times = tv; 1754 times = tv;
1432 } 1755 }
1433 else 1756 else
1434 times = 0; 1757 times = 0;
1435 1758
1436
1437 req->result = req->type == EIO_FUTIME 1759 req->result = req->type == EIO_FUTIME
1438 ? futimes (req->int1, times) 1760 ? futimes (req->int1, times)
1439 : utimes (req->ptr1, times); 1761 : utimes (req->ptr1, times);
1440 } 1762 }
1441 break; 1763 break;
1450 case EIO_CUSTOM: 1772 case EIO_CUSTOM:
1451 ((void (*)(eio_req *))req->feed) (req); 1773 ((void (*)(eio_req *))req->feed) (req);
1452 break; 1774 break;
1453 1775
1454 default: 1776 default:
1777 errno = ENOSYS;
1455 req->result = -1; 1778 req->result = -1;
1456 break; 1779 break;
1457 } 1780 }
1458 1781
1459 req->errorno = errno; 1782 req->errorno = errno;
1489eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 1812eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1490{ 1813{
1491 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 1814 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1492} 1815}
1493 1816
1817eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1818{
1819 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1820}
1821
1822eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1823{
1824 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1825}
1826
1494eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 1827eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1495{ 1828{
1496 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 1829 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1497} 1830}
1498 1831
1522} 1855}
1523 1856
1524eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1857eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1525{ 1858{
1526 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1859 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1860}
1861
1862eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1863{
1864 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1527} 1865}
1528 1866
1529eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1867eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1530{ 1868{
1531 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1869 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) 1943eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1606{ 1944{
1607 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1945 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1608} 1946}
1609 1947
1948eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1949{
1950 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1951}
1952
1610eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1953eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1611{ 1954{
1612 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1955 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1613} 1956}
1614 1957
1622 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 1965 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1623} 1966}
1624 1967
1625eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1968eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1626{ 1969{
1627 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1970 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1628} 1971}
1629 1972
1630static eio_req * 1973static eio_req *
1631eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1974eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1632{ 1975{
1716/* misc garbage */ 2059/* misc garbage */
1717 2060
1718ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2061ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1719{ 2062{
1720 etp_worker wrk; 2063 etp_worker wrk;
2064 ssize_t ret;
1721 2065
1722 wrk.dbuf = 0; 2066 wrk.dbuf = 0;
1723 2067
1724 eio__sendfile (ofd, ifd, offset, count, &wrk); 2068 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1725 2069
1726 if (wrk.dbuf) 2070 if (wrk.dbuf)
1727 free (wrk.dbuf); 2071 free (wrk.dbuf);
1728}
1729 2072
2073 return ret;
2074}
2075

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines