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

Comparing libeio/eio.c (file contents):
Revision 1.41 by root, Sun Jun 14 20:36:59 2009 UTC vs.
Revision 1.62 by root, Tue Feb 15 03:15:16 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 *
… …
49#include <stdlib.h> 49#include <stdlib.h>
50#include <string.h> 50#include <string.h>
51#include <errno.h> 51#include <errno.h>
52#include <sys/types.h> 52#include <sys/types.h>
53#include <sys/stat.h> 53#include <sys/stat.h>
54#include <sys/statvfs.h>
54#include <limits.h> 55#include <limits.h>
55#include <fcntl.h> 56#include <fcntl.h>
56#include <assert.h> 57#include <assert.h>
57 58
58#ifndef EIO_FINISH 59#ifndef EIO_FINISH
… …
73#else 74#else
74 75
75# include "config.h" 76# include "config.h"
76# include <sys/time.h> 77# include <sys/time.h>
77# include <sys/select.h> 78# include <sys/select.h>
78# include <sys/mman.h>
79# include <unistd.h> 79# include <unistd.h>
80# include <utime.h> 80# include <utime.h>
81# include <signal.h> 81# include <signal.h>
82# include <dirent.h> 82# include <dirent.h>
83 83
84#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
85# include <sys/mman.h>
86#endif
87
84/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 88/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) 89# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
86# define _DIRENT_HAVE_D_TYPE /* sigh */ 90# define _DIRENT_HAVE_D_TYPE /* sigh */
87# define D_INO(de) (de)->d_fileno 91# define D_INO(de) (de)->d_fileno
88# define D_NAMLEN(de) (de)->d_namlen 92# define D_NAMLEN(de) (de)->d_namlen
89# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 93# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
90# define D_INO(de) (de)->d_ino 94# define D_INO(de) (de)->d_ino
91# endif 95# endif
92 96
93#ifdef _D_EXACT_NAMLEN 97#ifdef _D_EXACT_NAMLEN
94# undef D_NAMLEN 98# undef D_NAMLEN
… …
106#endif 110#endif
107 111
108#if HAVE_SENDFILE 112#if HAVE_SENDFILE
109# if __linux 113# if __linux
110# include <sys/sendfile.h> 114# include <sys/sendfile.h>
111# elif __freebsd 115# elif __FreeBSD__ || defined __APPLE__
112# include <sys/socket.h> 116# include <sys/socket.h>
113# include <sys/uio.h> 117# include <sys/uio.h>
114# elif __hpux 118# elif __hpux
115# include <sys/socket.h> 119# include <sys/socket.h>
116# elif __solaris /* not yet */ 120# elif __solaris
117# include <sys/sendfile.h> 121# include <sys/sendfile.h>
118# else 122# else
119# error sendfile support requested but not available 123# error sendfile support requested but not available
120# endif 124# endif
121#endif 125#endif
… …
128#endif 132#endif
129#ifndef D_NAMLEN 133#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name) 134# define D_NAMLEN(de) strlen ((de)->d_name)
131#endif 135#endif
132 136
133/* number of seconds after which an idle threads exit */
134#define IDLE_TIMEOUT 10
135
136/* used for struct dirent, AIX doesn't provide it */ 137/* used for struct dirent, AIX doesn't provide it */
137#ifndef NAME_MAX 138#ifndef NAME_MAX
138# define NAME_MAX 4096 139# define NAME_MAX 4096
140#endif
141
142/* used for readlink etc. */
143#ifndef PATH_MAX
144# define PATH_MAX 4096
139#endif 145#endif
140 146
141/* buffer size for various temporary buffers */ 147/* buffer size for various temporary buffers */
142#define EIO_BUFSIZE 65536 148#define EIO_BUFSIZE 65536
143 149
… …
196 202
197/*****************************************************************************/ 203/*****************************************************************************/
198 204
199#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 205#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
200 206
201/* calculcate time difference in ~1/EIO_TICKS of a second */ 207/* calculate time difference in ~1/EIO_TICKS of a second */
202static int tvdiff (struct timeval *tv1, struct timeval *tv2) 208static int tvdiff (struct timeval *tv1, struct timeval *tv2)
203{ 209{
204 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 210 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
205 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 211 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
206} 212}
… …
214static unsigned int max_poll_reqs; /* reslock */ 220static unsigned int max_poll_reqs; /* reslock */
215 221
216static volatile unsigned int nreqs; /* reqlock */ 222static volatile unsigned int nreqs; /* reqlock */
217static volatile unsigned int nready; /* reqlock */ 223static volatile unsigned int nready; /* reqlock */
218static volatile unsigned int npending; /* reqlock */ 224static volatile unsigned int npending; /* reqlock */
219static volatile unsigned int max_idle = 4; 225static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
226static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
220 227
221static mutex_t wrklock = X_MUTEX_INIT; 228static xmutex_t wrklock;
222static mutex_t reslock = X_MUTEX_INIT; 229static xmutex_t reslock;
223static mutex_t reqlock = X_MUTEX_INIT; 230static xmutex_t reqlock;
224static cond_t reqwait = X_COND_INIT; 231static xcond_t reqwait;
225 232
226#if !HAVE_PREADWRITE 233#if !HAVE_PREADWRITE
227/* 234/*
228 * make our pread/pwrite emulation safe against themselves, but not against 235 * make our pread/pwrite emulation safe against themselves, but not against
229 * normal read/write by using a mutex. slows down execution a lot, 236 * normal read/write by using a mutex. slows down execution a lot,
230 * but that's your problem, not mine. 237 * but that's your problem, not mine.
231 */ 238 */
232static mutex_t preadwritelock = X_MUTEX_INIT; 239static xmutex_t preadwritelock = X_MUTEX_INIT;
233#endif 240#endif
234 241
235typedef struct etp_worker 242typedef struct etp_worker
236{ 243{
237 /* locked by wrklock */ 244 /* locked by wrklock */
238 struct etp_worker *prev, *next; 245 struct etp_worker *prev, *next;
239 246
240 thread_t tid; 247 xthread_t tid;
241 248
242 /* locked by reslock, reqlock or wrklock */ 249 /* locked by reslock, reqlock or wrklock */
243 ETP_REQ *req; /* currently processed request */ 250 ETP_REQ *req; /* currently processed request */
244 251
245 ETP_WORKER_COMMON 252 ETP_WORKER_COMMON
… …
359 } 366 }
360 367
361 abort (); 368 abort ();
362} 369}
363 370
371static void etp_thread_init (void)
372{
373 X_MUTEX_CREATE (wrklock);
374 X_MUTEX_CREATE (reslock);
375 X_MUTEX_CREATE (reqlock);
376 X_COND_CREATE (reqwait);
377}
378
364static void etp_atfork_prepare (void) 379static void etp_atfork_prepare (void)
365{ 380{
366 X_LOCK (wrklock); 381 X_LOCK (wrklock);
367 X_LOCK (reqlock); 382 X_LOCK (reqlock);
368 X_LOCK (reslock); 383 X_LOCK (reslock);
… …
406 idle = 0; 421 idle = 0;
407 nreqs = 0; 422 nreqs = 0;
408 nready = 0; 423 nready = 0;
409 npending = 0; 424 npending = 0;
410 425
411 etp_atfork_parent (); 426 etp_thread_init ();
412} 427}
413 428
414static void 429static void
415etp_once_init (void) 430etp_once_init (void)
416{ 431{
432 etp_thread_init ();
417 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 433 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
418} 434}
419 435
420static int 436static int
421etp_init (void (*want_poll)(void), void (*done_poll)(void)) 437etp_init (void (*want_poll)(void), void (*done_poll)(void))
… …
598} 614}
599 615
600static void etp_set_max_poll_time (double nseconds) 616static void etp_set_max_poll_time (double nseconds)
601{ 617{
602 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 618 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
603 max_poll_time = nseconds; 619 max_poll_time = nseconds * EIO_TICKS;
604 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 620 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
605} 621}
606 622
607static void etp_set_max_poll_reqs (unsigned int maxreqs) 623static void etp_set_max_poll_reqs (unsigned int maxreqs)
608{ 624{
… …
612} 628}
613 629
614static void etp_set_max_idle (unsigned int nthreads) 630static void etp_set_max_idle (unsigned int nthreads)
615{ 631{
616 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 632 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
617 max_idle = nthreads <= 0 ? 1 : nthreads; 633 max_idle = nthreads;
634 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
635}
636
637static void etp_set_idle_timeout (unsigned int seconds)
638{
639 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
640 idle_timeout = seconds;
618 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 641 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
619} 642}
620 643
621static void etp_set_min_parallel (unsigned int nthreads) 644static void etp_set_min_parallel (unsigned int nthreads)
622{ 645{
… …
750void eio_set_max_idle (unsigned int nthreads) 773void eio_set_max_idle (unsigned int nthreads)
751{ 774{
752 etp_set_max_idle (nthreads); 775 etp_set_max_idle (nthreads);
753} 776}
754 777
778void eio_set_idle_timeout (unsigned int seconds)
779{
780 etp_set_idle_timeout (seconds);
781}
782
755void eio_set_min_parallel (unsigned int nthreads) 783void eio_set_min_parallel (unsigned int nthreads)
756{ 784{
757 etp_set_min_parallel (nthreads); 785 etp_set_min_parallel (nthreads);
758} 786}
759 787
… …
807 835
808 return res; 836 return res;
809} 837}
810#endif 838#endif
811 839
812#ifndef HAVE_FUTIMES 840#ifndef HAVE_UTIMES
813 841
814# undef utimes 842# undef utimes
815# undef futimes
816# define utimes(path,times) eio__utimes (path, times) 843# define utimes(path,times) eio__utimes (path, times)
817# define futimes(fd,times) eio__futimes (fd, times)
818 844
819static int 845static int
820eio__utimes (const char *filename, const struct timeval times[2]) 846eio__utimes (const char *filename, const struct timeval times[2])
821{ 847{
822 if (times) 848 if (times)
… …
829 return utime (filename, &buf); 855 return utime (filename, &buf);
830 } 856 }
831 else 857 else
832 return utime (filename, 0); 858 return utime (filename, 0);
833} 859}
860
861#endif
862
863#ifndef HAVE_FUTIMES
864
865# undef futimes
866# define futimes(fd,times) eio__futimes (fd, times)
834 867
835static int eio__futimes (int fd, const struct timeval tv[2]) 868static int eio__futimes (int fd, const struct timeval tv[2])
836{ 869{
837 errno = ENOSYS; 870 errno = ENOSYS;
838 return -1; 871 return -1;
… …
867 if (!res || errno != ENOSYS) 900 if (!res || errno != ENOSYS)
868 return res; 901 return res;
869#endif 902#endif
870 903
871 /* even though we could play tricks with the flags, it's better to always 904 /* even though we could play tricks with the flags, it's better to always
872 * call fdatasync, as thta matches the expectation of it's users best */ 905 * call fdatasync, as that matches the expectation of its users best */
873 return fdatasync (fd); 906 return fdatasync (fd);
874} 907}
875 908
876#if !HAVE_READAHEAD 909#if !HAVE_READAHEAD
877# undef readahead 910# undef readahead
… …
909 942
910#if HAVE_SENDFILE 943#if HAVE_SENDFILE
911# if __linux 944# if __linux
912 res = sendfile (ofd, ifd, &offset, count); 945 res = sendfile (ofd, ifd, &offset, count);
913 946
914# elif __freebsd 947# elif __FreeBSD__
915 /* 948 /*
916 * Of course, the freebsd sendfile is a dire hack with no thoughts 949 * Of course, the freebsd sendfile is a dire hack with no thoughts
917 * wasted on making it similar to other I/O functions. 950 * wasted on making it similar to other I/O functions.
918 */ 951 */
919 { 952 {
920 off_t sbytes; 953 off_t sbytes;
921 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 954 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
922 955
923 if (res < 0 && sbytes) 956 #if 0 /* according to the manpage, this is correct, but broken behaviour */
924 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 957 /* freebsd' sendfile will return 0 on success */
958 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
959 /* not on e.g. EIO or EPIPE - sounds broken */
960 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
961 res = sbytes;
962 #endif
963
964 /* according to source inspection, this is correct, and useful behaviour */
965 if (sbytes)
966 res = sbytes;
967 }
968
969# elif defined (__APPLE__)
970
971 {
972 off_t sbytes = count;
973 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
974
975 /* according to the manpage, sbytes is always valid */
976 if (sbytes)
925 res = sbytes; 977 res = sbytes;
926 } 978 }
927 979
928# elif __hpux 980# elif __hpux
929 res = sendfile (ofd, ifd, offset, count, 0, 0); 981 res = sendfile (ofd, ifd, offset, count, 0, 0);
… …
943 if (res < 0 && sbytes) 995 if (res < 0 && sbytes)
944 res = sbytes; 996 res = sbytes;
945 } 997 }
946 998
947# endif 999# endif
1000
1001#elif defined (_WIN32)
1002
1003 /* does not work, just for documentation of what would need to be done */
1004 {
1005 HANDLE h = TO_SOCKET (ifd);
1006 SetFilePointer (h, offset, 0, FILE_BEGIN);
1007 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1008 }
1009
948#else 1010#else
949 res = -1; 1011 res = -1;
950 errno = ENOSYS; 1012 errno = ENOSYS;
951#endif 1013#endif
952 1014
953 if (res < 0 1015 if (res < 0
954 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1016 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1017 /* BSDs */
1018#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1019 || errno == ENOTSUP
1020#endif
1021 || errno == EOPNOTSUPP /* BSDs */
955#if __solaris 1022#if __solaris
956 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1023 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
957#endif 1024#endif
958 ) 1025 )
959 ) 1026 )
… …
993} 1060}
994 1061
995static signed char 1062static signed char
996eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1063eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997{ 1064{
998 return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */ 1065 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
999 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1066 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1000} 1067}
1001 1068
1002#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1069#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1003#define EIO_DENT_NUM_SCORES 9
1004 1070
1005#define EIO_QSORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1071#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1006#define EIO_QSORT_SKIP 60 /* when to skip qsort completely */ 1072#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1007 1073
1008static void 1074static void
1009eio_dent_sort (eio_dirent *dents, int size) 1075eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1010{ 1076{
1011 if (size <= 1) 1077 unsigned char bits [9 + sizeof (ino_t) * 8];
1012 return; /* our insertion sort relies on size > 0 */ 1078 unsigned char *bit = bits;
1013 1079
1014 if (size > EIO_QSORT_SKIP) /* skip quicksort for small directories */ 1080 assert (CHAR_BIT == 8);
1081 assert (sizeof (eio_dirent) * 8 < 256);
1082 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1083 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1084
1085 if (size <= EIO_SORT_FAST)
1086 return;
1087
1088 /* first prepare an array of bits to test in our radix sort */
1089 /* try to take endianness into account, as well as differences in ino_t sizes */
1090 /* inode_bits must contain all inodes ORed together */
1091 /* which is used to skip bits that are 0 everywhere, which is very common */
1015 { 1092 {
1016 /* first, use quicksort */ 1093 ino_t endianness;
1017 /* should be good for 2**31 entries */
1018 int i; 1094 int i, j;
1019 struct rng { eio_dirent *l, *r; } rng [32];
1020 1095
1021 i = 0; 1096 /* we store the byte offset of byte n into byte n of "endianness" */
1022 rng[0].l = dents; 1097 for (i = 0; i < sizeof (ino_t); ++i)
1023 rng[0].r = dents + size; 1098 ((unsigned char *)&endianness)[i] = i;
1024 1099
1025 while (expect_true (i >= 0)) 1100 *bit++ = 0;
1101
1102 for (i = 0; i < sizeof (ino_t); ++i)
1026 { 1103 {
1027 eio_dirent *L = rng [i].l; 1104 /* shifting off the byte offsets out of "endianness" */
1028 eio_dirent *R = rng [i].r - 1; 1105 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1106 endianness >>= 8;
1029 1107
1030 if (expect_false (L + EIO_QSORT_CUTOFF < R)) 1108 for (j = 0; j < 8; ++j)
1109 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1110 *bit++ = offs + j;
1111 }
1112
1113 for (j = 0; j < 8; ++j)
1114 if (score_bits & (1 << j))
1115 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1116 }
1117
1118 /* now actually do the sorting (a variant of MSD radix sort) */
1119 {
1120 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1121 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1122 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1123 int stk_idx = 0;
1124
1125 base_stk [stk_idx] = dents;
1126 end_stk [stk_idx] = dents + size;
1127 bit_stk [stk_idx] = bit - 1;
1128
1129 do
1130 {
1131 base = base_stk [stk_idx];
1132 end = end_stk [stk_idx];
1133 bit = bit_stk [stk_idx];
1134
1135 for (;;)
1031 { 1136 {
1032 eio_dirent *mid = &dents [((L - dents) + (R - dents)) >> 1]; 1137 unsigned char O = *bit >> 3;
1033 eio_dirent piv = *mid; *mid = *L; *L = piv; 1138 unsigned char M = 1 << (*bit & 7);
1034 1139
1035 while (L < R) 1140 eio_dirent *a = base;
1141 eio_dirent *b = end;
1142
1143 if (b - a < EIO_SORT_CUTOFF)
1144 break;
1145
1146 /* now bit-partition the array on the bit */
1147 /* this ugly asymmetric loop seems to perform much better than typical */
1148 /* partition algos found in the literature */
1149 do
1150 if (!(((unsigned char *)a)[O] & M))
1151 ++a;
1152 else if (!(((unsigned char *)--b)[O] & M))
1036 { 1153 {
1037 while (EIO_DENT_CMP (*R, >=, piv) && L < R) 1154 eio_dirent tmp = *a; *a = *b; *b = tmp;
1038 --R;
1039
1040 if (L < R)
1041 *L++ = *R;
1042
1043 while (EIO_DENT_CMP (*L, <=, piv) && L < R)
1044 ++L; 1155 ++a;
1045
1046 if (L < R)
1047 *R-- = *L;
1048 } 1156 }
1157 while (b > a);
1049 1158
1050 *L = piv; 1159 /* next bit, or stop, if no bits left in this path */
1160 if (!*--bit)
1161 break;
1051 1162
1163 base_stk [stk_idx] = a;
1164 end_stk [stk_idx] = end;
1165 bit_stk [stk_idx] = bit;
1052 ++i; 1166 ++stk_idx;
1053 rng [i].l = L + 1;
1054 rng [i].r = rng [i - 1].r;
1055 rng [i - 1].r = L;
1056 1167
1057 if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l) 1168 end = a;
1058 {
1059 struct rng t;
1060
1061 t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t;
1062 }
1063 } 1169 }
1064 else
1065 --i;
1066 } 1170 }
1171 while (stk_idx--);
1067 } 1172 }
1173}
1068 1174
1069 /* use an insertion sort after qsort, or for small arrays */ 1175static void
1176eio_dent_insertion_sort (eio_dirent *dents, int size)
1177{
1070 /* first move the smallest element to the front, to act as a sentinel */ 1178 /* first move the smallest element to the front, to act as a sentinel */
1071 { 1179 {
1072 int i; 1180 int i;
1073 eio_dirent *min = dents; 1181 eio_dirent *min = dents;
1074 1182
1183 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1075 for (i = size > EIO_QSORT_SKIP ? EIO_QSORT_CUTOFF + 1 : size; --i; ) 1184 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1076 if (EIO_DENT_CMP (dents [i], <, *min)) 1185 if (EIO_DENT_CMP (dents [i], <, *min))
1077 min = &dents [i]; 1186 min = &dents [i];
1078 1187
1079 /* swap elements 0 and j (minimum) */ 1188 /* swap elements 0 and j (minimum) */
1080 { 1189 {
1081 eio_dirent tmp = *dents; *dents = *min; *min = tmp; 1190 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1082 } 1191 }
1083 } 1192 }
1084 1193
1194 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1085 { 1195 {
1086 /* then do standard insertion sort */
1087 eio_dirent *i, *j; 1196 eio_dirent *i, *j;
1088 1197
1089 for (i = dents + 1; i < dents + size; ++i) 1198 for (i = dents + 1; i < dents + size; ++i)
1090 { 1199 {
1091 eio_dirent value = *i; 1200 eio_dirent value = *i;
… …
1094 j [1] = j [0]; 1203 j [1] = j [0];
1095 1204
1096 j [1] = value; 1205 j [1] = value;
1097 } 1206 }
1098 } 1207 }
1208}
1209
1210static void
1211eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1212{
1213 if (size <= 1)
1214 return; /* our insertion sort relies on size > 0 */
1215
1216 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1217 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1218 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1219
1220 /* use an insertion sort at the end, or for small arrays, */
1221 /* as insertion sort is more efficient for small partitions */
1222 eio_dent_insertion_sort (dents, size);
1099} 1223}
1100 1224
1101/* read a full directory */ 1225/* read a full directory */
1102static void 1226static void
1103eio__scandir (eio_req *req, etp_worker *self) 1227eio__scandir (eio_req *req, etp_worker *self)
… …
1109 int namesoffs = 0; 1233 int namesoffs = 0;
1110 int flags = req->int1; 1234 int flags = req->int1;
1111 eio_dirent *dents = 0; 1235 eio_dirent *dents = 0;
1112 int dentalloc = 128; 1236 int dentalloc = 128;
1113 int dentoffs = 0; 1237 int dentoffs = 0;
1238 ino_t inode_bits = 0;
1114 1239
1115 req->result = -1; 1240 req->result = -1;
1116 1241
1117 if (!(flags & EIO_READDIR_DENTS)) 1242 if (!(flags & EIO_READDIR_DENTS))
1118 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1243 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1119 1244
1120 X_LOCK (wrklock); 1245 X_LOCK (wrklock);
1121 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1246 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1122 self->dirp = dirp = opendir (req->ptr1); 1247 self->dirp = dirp = opendir (req->ptr1);
1123 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1248 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1124 req->ptr1 = names = malloc (namesalloc);
1125 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1249 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1250 req->ptr2 = names = malloc (namesalloc);
1126 X_UNLOCK (wrklock); 1251 X_UNLOCK (wrklock);
1127 1252
1128 if (dirp && names && (!flags || dents)) 1253 if (dirp && names && (!flags || dents))
1129 for (;;) 1254 for (;;)
1130 { 1255 {
… …
1138 1263
1139 /* sort etc. */ 1264 /* sort etc. */
1140 req->int1 = flags; 1265 req->int1 = flags;
1141 req->result = dentoffs; 1266 req->result = dentoffs;
1142 1267
1143 if (dents)
1144 {
1145 eio_dirent *ent = dents + dentoffs;
1146
1147 while (ent > dents)
1148 (--ent)->name = names + (size_t)ent->name;
1149 }
1150
1151 if (flags & EIO_READDIR_STAT_ORDER 1268 if (flags & EIO_READDIR_STAT_ORDER)
1152 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1269 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
1153 eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */
1154 else if (flags & EIO_READDIR_DIRS_FIRST) 1270 else if (flags & EIO_READDIR_DIRS_FIRST)
1271 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1272 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1273 else
1155 { 1274 {
1156 /* in this case, all is known, and we just put dirs first and sort them */ 1275 /* in this case, all is known, and we just put dirs first and sort them */
1157 eio_dirent *ent = dents + dentoffs; 1276 eio_dirent *oth = dents + dentoffs;
1158 eio_dirent *dir = dents; 1277 eio_dirent *dir = dents;
1159 1278
1160 /* now move dirs to the front, and non-dirs to the back */ 1279 /* now partition dirs to the front, and non-dirs to the back */
1161 /* by walking from both sides and swapping if necessary */ 1280 /* by walking from both sides and swapping if necessary */
1281 /* also clear score, so it doesn't influence sorting */
1162 while (ent > dir) 1282 while (oth > dir)
1163 {
1164 if (dir->type == DT_DIR)
1165 { 1283 {
1166 dir->score = 0; 1284 if (dir->type == EIO_DT_DIR)
1167 ++dir; 1285 ++dir;
1168 }
1169 else
1170 {
1171 --ent;
1172
1173 if (ent->type == DT_DIR) 1286 else if ((--oth)->type == EIO_DT_DIR)
1174 { 1287 {
1175 eio_dirent tmp = *dir; 1288 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1176 *dir = *ent;
1177 *ent = tmp;
1178 1289
1179 dir->score = 0;
1180 ++dir; 1290 ++dir;
1181 } 1291 }
1182 } 1292 }
1183 }
1184 1293
1185 /* now sort the dirs only */ 1294 /* now sort the dirs only */
1186 eio_dent_sort (dents, dir - dents); 1295 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1187 } 1296 }
1188
1189 /* only provide the names array unless DENTS is specified */
1190 if (!(flags & EIO_READDIR_DENTS))
1191 {
1192 X_LOCK (wrklock);
1193 assert (!dents);
1194 req->ptr1 = 0;
1195 req->ptr2 = names;
1196 X_UNLOCK (wrklock);
1197 }
1198 1297
1199 break; 1298 break;
1200 } 1299 }
1201 1300
1202 /* now add the entry to our list(s) */ 1301 /* now add the entry to our list(s) */
… …
1209 1308
1210 while (expect_false (namesoffs + len > namesalloc)) 1309 while (expect_false (namesoffs + len > namesalloc))
1211 { 1310 {
1212 namesalloc *= 2; 1311 namesalloc *= 2;
1213 X_LOCK (wrklock); 1312 X_LOCK (wrklock);
1214 req->ptr1 = names = realloc (names, namesalloc); 1313 req->ptr2 = names = realloc (names, namesalloc);
1215 X_UNLOCK (wrklock); 1314 X_UNLOCK (wrklock);
1216 1315
1217 if (!names) 1316 if (!names)
1218 break; 1317 break;
1219 } 1318 }
… …
1226 1325
1227 if (expect_false (dentoffs == dentalloc)) 1326 if (expect_false (dentoffs == dentalloc))
1228 { 1327 {
1229 dentalloc *= 2; 1328 dentalloc *= 2;
1230 X_LOCK (wrklock); 1329 X_LOCK (wrklock);
1231 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1330 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1232 X_UNLOCK (wrklock); 1331 X_UNLOCK (wrklock);
1233 1332
1234 if (!dents) 1333 if (!dents)
1235 break; 1334 break;
1236 } 1335 }
1237 1336
1238 ent = dents + dentoffs; 1337 ent = dents + dentoffs;
1239 1338
1240 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1339 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1241 ent->namelen = len - 1; 1340 ent->namelen = len - 1;
1242 ent->inode = D_INO (entp) ? D_INO (entp) : dentoffs; 1341 ent->inode = D_INO (entp);
1342
1343 inode_bits |= ent->inode;
1243 1344
1244 switch (D_TYPE (entp)) 1345 switch (D_TYPE (entp))
1245 { 1346 {
1246 default: 1347 default:
1247 ent->type = EIO_DT_UNKNOWN; 1348 ent->type = EIO_DT_UNKNOWN;
… …
1290 #ifdef DT_WHT 1391 #ifdef DT_WHT
1291 case DT_WHT: ent->type = EIO_DT_WHT; break; 1392 case DT_WHT: ent->type = EIO_DT_WHT; break;
1292 #endif 1393 #endif
1293 } 1394 }
1294 1395
1295 ent->score = 0; 1396 ent->score = 7;
1296 1397
1297 if (flags & EIO_READDIR_DIRS_FIRST) 1398 if (flags & EIO_READDIR_DIRS_FIRST)
1298 { 1399 {
1299 if (ent->type == EIO_DT_UNKNOWN) 1400 if (ent->type == EIO_DT_UNKNOWN)
1300 { 1401 {
1301 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1402 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1302 ent->score = 7; 1403 ent->score = 1;
1303 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1404 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1304 ent->score = len <= 2 ? len + 4 : len <= 4 ? 3 : len <= 7 ? 2 : 1; /* shorter == more likely dir, but avoid too many classes */ 1405 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1305 } 1406 }
1306 else if (ent->type == EIO_DT_DIR) 1407 else if (ent->type == EIO_DT_DIR)
1307 ent->score = 8; 1408 ent->score = 0;
1308 } 1409 }
1309 } 1410 }
1310 1411
1311 namesoffs += len; 1412 namesoffs += len;
1312 ++dentoffs; 1413 ++dentoffs;
… …
1318 break; 1419 break;
1319 } 1420 }
1320 } 1421 }
1321} 1422}
1322 1423
1424#ifdef PAGESIZE
1425# define eio_pagesize() PAGESIZE
1426#else
1427static intptr_t
1428eio_pagesize (void)
1429{
1430 static intptr_t page;
1431
1432 if (!page)
1433 page = sysconf (_SC_PAGESIZE);
1434
1435 return page;
1436}
1437#endif
1438
1439static void
1440eio_page_align (void **addr, size_t *length)
1441{
1442 intptr_t mask = eio_pagesize () - 1;
1443
1444 /* round down addr */
1445 intptr_t adj = mask & (intptr_t)*addr;
1446
1447 *addr = (void *)((intptr_t)*addr - adj);
1448 *length += adj;
1449
1450 /* round up length */
1451 *length = (*length + mask) & ~mask;
1452}
1453
1454#if !_POSIX_MEMLOCK
1455# define eio__mlockall(a) ((errno = ENOSYS), -1)
1456#else
1457
1458static int
1459eio__mlockall (int flags)
1460{
1461 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1462 extern int mallopt (int, int);
1463 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1464 #endif
1465
1466 if (EIO_MCL_CURRENT != MCL_CURRENT
1467 || EIO_MCL_FUTURE != MCL_FUTURE)
1468 {
1469 flags = 0
1470 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1471 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1472 }
1473
1474 return mlockall (flags);
1475}
1476#endif
1477
1478#if !_POSIX_MEMLOCK_RANGE
1479# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1480#else
1481
1482static int
1483eio__mlock (void *addr, size_t length)
1484{
1485 eio_page_align (&addr, &length);
1486
1487 return mlock (addr, length);
1488}
1489
1490#endif
1491
1323#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1492#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1324# undef msync
1325# define msync(a,b,c) ((errno = ENOSYS), -1) 1493# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1494#else
1495
1496int
1497eio__msync (void *mem, size_t len, int flags)
1498{
1499 eio_page_align (&mem, &len);
1500
1501 if (EIO_MS_ASYNC != MS_SYNC
1502 || EIO_MS_INVALIDATE != MS_INVALIDATE
1503 || EIO_MS_SYNC != MS_SYNC)
1504 {
1505 flags = 0
1506 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1507 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1508 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1509 }
1510
1511 return msync (mem, len, flags);
1512}
1513
1326#endif 1514#endif
1327 1515
1328int 1516int
1329eio__mtouch (void *mem, size_t len, int flags) 1517eio__mtouch (void *mem, size_t len, int flags)
1330{ 1518{
1519 eio_page_align (&mem, &len);
1520
1521 {
1331 intptr_t addr = (intptr_t)mem; 1522 intptr_t addr = (intptr_t)mem;
1332 intptr_t end = addr + len; 1523 intptr_t end = addr + len;
1333#ifdef PAGESIZE 1524 intptr_t page = eio_pagesize ();
1334 const intptr_t page = PAGESIZE;
1335#else
1336 static intptr_t page;
1337 1525
1338 if (!page)
1339 page = sysconf (_SC_PAGESIZE);
1340#endif
1341
1342 addr &= ~(page - 1); /* assume page size is always a power of two */
1343
1344 if (addr < end) 1526 if (addr < end)
1345 if (flags) /* modify */ 1527 if (flags & EIO_MT_MODIFY) /* modify */
1346 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1528 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1347 else 1529 else
1348 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1530 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1531 }
1349 1532
1350 return 0; 1533 return 0;
1351} 1534}
1352 1535
1353/*****************************************************************************/ 1536/*****************************************************************************/
… …
1387 if (req) 1570 if (req)
1388 break; 1571 break;
1389 1572
1390 ++idle; 1573 ++idle;
1391 1574
1392 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1575 ts.tv_sec = time (0) + idle_timeout;
1393 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1576 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1394 { 1577 {
1395 if (idle > max_idle) 1578 if (idle > max_idle)
1396 { 1579 {
1397 --idle; 1580 --idle;
… …
1476 return 0; \ 1659 return 0; \
1477 } 1660 }
1478 1661
1479static void eio_execute (etp_worker *self, eio_req *req) 1662static void eio_execute (etp_worker *self, eio_req *req)
1480{ 1663{
1481 errno = 0;
1482
1483 switch (req->type) 1664 switch (req->type)
1484 { 1665 {
1485 case EIO_READ: ALLOC (req->size); 1666 case EIO_READ: ALLOC (req->size);
1486 req->result = req->offs >= 0 1667 req->result = req->offs >= 0
1487 ? pread (req->int1, req->ptr2, req->size, req->offs) 1668 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1498 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1679 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1499 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1680 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1500 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1681 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1501 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1682 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1502 1683
1684 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1685 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1686 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1687 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1688
1503 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1689 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1504 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1690 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1505 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1691 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1506 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1692 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1507 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1693 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
… …
1514 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1700 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1515 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 1701 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1516 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1702 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1517 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1703 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1518 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1704 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1519 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1705 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1520 1706
1521 case EIO_READLINK: ALLOC (NAME_MAX); 1707 case EIO_READLINK: ALLOC (PATH_MAX);
1522 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1708 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1523 1709
1524 case EIO_SYNC: req->result = 0; sync (); break; 1710 case EIO_SYNC: req->result = 0; sync (); break;
1525 case EIO_FSYNC: req->result = fsync (req->int1); break; 1711 case EIO_FSYNC: req->result = fsync (req->int1); break;
1526 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1712 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1527 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1713 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1528 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1714 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1715 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1716 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1529 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 1717 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1530 1718
1531 case EIO_READDIR: eio__scandir (req, self); break; 1719 case EIO_READDIR: eio__scandir (req, self); break;
1532 1720
1533 case EIO_BUSY: 1721 case EIO_BUSY:
1534#ifdef _WIN32 1722#ifdef _WIN32
1535 Sleep (req->nv1 * 1000.); 1723 Sleep (req->nv1 * 1e3);
1536#else 1724#else
1537 { 1725 {
1538 struct timeval tv; 1726 struct timeval tv;
1539 1727
1540 tv.tv_sec = req->nv1; 1728 tv.tv_sec = req->nv1;
1541 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1729 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1542 1730
1543 req->result = select (0, 0, 0, 0, &tv); 1731 req->result = select (0, 0, 0, 0, &tv);
1544 } 1732 }
1545#endif 1733#endif
1546 break; 1734 break;
… …
1561 times = tv; 1749 times = tv;
1562 } 1750 }
1563 else 1751 else
1564 times = 0; 1752 times = 0;
1565 1753
1566
1567 req->result = req->type == EIO_FUTIME 1754 req->result = req->type == EIO_FUTIME
1568 ? futimes (req->int1, times) 1755 ? futimes (req->int1, times)
1569 : utimes (req->ptr1, times); 1756 : utimes (req->ptr1, times);
1570 } 1757 }
1571 break; 1758 break;
… …
1580 case EIO_CUSTOM: 1767 case EIO_CUSTOM:
1581 ((void (*)(eio_req *))req->feed) (req); 1768 ((void (*)(eio_req *))req->feed) (req);
1582 break; 1769 break;
1583 1770
1584 default: 1771 default:
1772 errno = ENOSYS;
1585 req->result = -1; 1773 req->result = -1;
1586 break; 1774 break;
1587 } 1775 }
1588 1776
1589 req->errorno = errno; 1777 req->errorno = errno;
… …
1619eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 1807eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1620{ 1808{
1621 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 1809 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1622} 1810}
1623 1811
1812eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1813{
1814 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1815}
1816
1817eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1818{
1819 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1820}
1821
1624eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 1822eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1625{ 1823{
1626 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 1824 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1627} 1825}
1628 1826
… …
1652} 1850}
1653 1851
1654eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1852eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1655{ 1853{
1656 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1854 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1855}
1856
1857eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1858{
1859 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1657} 1860}
1658 1861
1659eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1862eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1660{ 1863{
1661 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1864 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
… …
1735eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 1938eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1736{ 1939{
1737 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1940 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1738} 1941}
1739 1942
1943eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1944{
1945 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1946}
1947
1740eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1948eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1741{ 1949{
1742 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1950 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1743} 1951}
1744 1952
… …
1752 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 1960 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1753} 1961}
1754 1962
1755eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1963eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1756{ 1964{
1757 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1965 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1758} 1966}
1759 1967
1760static eio_req * 1968static eio_req *
1761eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1969eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1762{ 1970{

Diff Legend

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