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

Comparing libeio/eio.c (file contents):
Revision 1.27 by root, Wed Oct 22 18:15:36 2008 UTC vs.
Revision 1.56 by root, Sun Sep 12 03:36:28 2010 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010 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 *
… …
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#include "eio.h" 40#include "eio.h"
41
42#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif
41#include "xthread.h" 45#include "xthread.h"
42 46
43#include <errno.h> 47#include <errno.h>
44#include <stddef.h> 48#include <stddef.h>
45#include <stdlib.h> 49#include <stdlib.h>
46#include <string.h> 50#include <string.h>
47#include <errno.h> 51#include <errno.h>
48#include <sys/types.h> 52#include <sys/types.h>
49#include <sys/stat.h> 53#include <sys/stat.h>
54#include <sys/statvfs.h>
50#include <limits.h> 55#include <limits.h>
51#include <fcntl.h> 56#include <fcntl.h>
52#include <assert.h> 57#include <assert.h>
53 58
54#ifndef EIO_FINISH 59#ifndef EIO_FINISH
… …
64#endif 69#endif
65 70
66#ifdef _WIN32 71#ifdef _WIN32
67 72
68 /*doh*/ 73 /*doh*/
69
70#else 74#else
71 75
72# include "config.h" 76# include "config.h"
73# include <sys/time.h> 77# include <sys/time.h>
74# include <sys/select.h> 78# include <sys/select.h>
75# include <unistd.h> 79# include <unistd.h>
76# include <utime.h> 80# include <utime.h>
77# include <signal.h> 81# include <signal.h>
78# include <dirent.h> 82# include <dirent.h>
79 83
84#if _POSIX_MEMLOCK || _POSIX_MAPPED_FILES
85# include <sys/mman.h>
86#endif
87
88/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
89# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
90# define _DIRENT_HAVE_D_TYPE /* sigh */
91# define D_INO(de) (de)->d_fileno
92# define D_NAMLEN(de) (de)->d_namlen
93# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
94# define D_INO(de) (de)->d_ino
95# endif
96
97#ifdef _D_EXACT_NAMLEN
98# undef D_NAMLEN
99# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
100#endif
101
102# ifdef _DIRENT_HAVE_D_TYPE
103# define D_TYPE(de) (de)->d_type
104# endif
105
80# ifndef EIO_STRUCT_DIRENT 106# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 107# define EIO_STRUCT_DIRENT struct dirent
82# endif 108# endif
83 109
84#endif 110#endif
85 111
86#if HAVE_SENDFILE 112#if HAVE_SENDFILE
87# if __linux 113# if __linux
88# include <sys/sendfile.h> 114# include <sys/sendfile.h>
89# elif __freebsd 115# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 116# include <sys/socket.h>
91# include <sys/uio.h> 117# include <sys/uio.h>
92# elif __hpux 118# elif __hpux
93# include <sys/socket.h> 119# include <sys/socket.h>
94# elif __solaris /* not yet */ 120# elif __solaris
95# include <sys/sendfile.h> 121# include <sys/sendfile.h>
96# else 122# else
97# error sendfile support requested but not available 123# error sendfile support requested but not available
98# endif 124# endif
99#endif 125#endif
100 126
127#ifndef D_TYPE
128# define D_TYPE(de) 0
129#endif
130#ifndef D_INO
131# define D_INO(de) 0
132#endif
133#ifndef D_NAMLEN
134# define D_NAMLEN(de) strlen ((de)->d_name)
135#endif
136
101/* number of seconds after which an idle threads exit */ 137/* number of seconds after which an idle threads exit */
102#define IDLE_TIMEOUT 10 138#define IDLE_TIMEOUT 10
103 139
104/* used for struct dirent, AIX doesn't provide it */ 140/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 141#ifndef NAME_MAX
106# 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
107#endif 148#endif
108 149
109/* buffer size for various temporary buffers */ 150/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 151#define EIO_BUFSIZE 65536
111 152
… …
155 if (wrk->dirp) \ 196 if (wrk->dirp) \
156 { \ 197 { \
157 closedir (wrk->dirp); \ 198 closedir (wrk->dirp); \
158 wrk->dirp = 0; \ 199 wrk->dirp = 0; \
159 } 200 }
201
160#define ETP_WORKER_COMMON \ 202#define ETP_WORKER_COMMON \
161 void *dbuf; \ 203 void *dbuf; \
162 DIR *dirp; 204 DIR *dirp;
163 205
164/*****************************************************************************/ 206/*****************************************************************************/
165 207
166#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
167 209
168/* calculcate time difference in ~1/EIO_TICKS of a second */ 210/* calculate time difference in ~1/EIO_TICKS of a second */
169static int tvdiff (struct timeval *tv1, struct timeval *tv2) 211static int tvdiff (struct timeval *tv1, struct timeval *tv2)
170{ 212{
171 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 213 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
172 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 214 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
173} 215}
… …
183static volatile unsigned int nreqs; /* reqlock */ 225static volatile unsigned int nreqs; /* reqlock */
184static volatile unsigned int nready; /* reqlock */ 226static volatile unsigned int nready; /* reqlock */
185static volatile unsigned int npending; /* reqlock */ 227static volatile unsigned int npending; /* reqlock */
186static volatile unsigned int max_idle = 4; 228static volatile unsigned int max_idle = 4;
187 229
188static mutex_t wrklock = X_MUTEX_INIT; 230static xmutex_t wrklock = X_MUTEX_INIT;
189static mutex_t reslock = X_MUTEX_INIT; 231static xmutex_t reslock = X_MUTEX_INIT;
190static mutex_t reqlock = X_MUTEX_INIT; 232static xmutex_t reqlock = X_MUTEX_INIT;
191static cond_t reqwait = X_COND_INIT; 233static xcond_t reqwait = X_COND_INIT;
192 234
193#if !HAVE_PREADWRITE 235#if !HAVE_PREADWRITE
194/* 236/*
195 * make our pread/pwrite emulation safe against themselves, but not against 237 * make our pread/pwrite emulation safe against themselves, but not against
196 * normal read/write by using a mutex. slows down execution a lot, 238 * normal read/write by using a mutex. slows down execution a lot,
197 * but that's your problem, not mine. 239 * but that's your problem, not mine.
198 */ 240 */
199static mutex_t preadwritelock = X_MUTEX_INIT; 241static xmutex_t preadwritelock = X_MUTEX_INIT;
200#endif 242#endif
201 243
202typedef struct etp_worker 244typedef struct etp_worker
203{ 245{
204 /* locked by wrklock */ 246 /* locked by wrklock */
205 struct etp_worker *prev, *next; 247 struct etp_worker *prev, *next;
206 248
207 thread_t tid; 249 xthread_t tid;
208 250
209 /* locked by reslock, reqlock or wrklock */ 251 /* locked by reslock, reqlock or wrklock */
210 ETP_REQ *req; /* currently processed request */ 252 ETP_REQ *req; /* currently processed request */
211 253
212 ETP_WORKER_COMMON 254 ETP_WORKER_COMMON
… …
565} 607}
566 608
567static void etp_set_max_poll_time (double nseconds) 609static void etp_set_max_poll_time (double nseconds)
568{ 610{
569 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 611 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
570 max_poll_time = nseconds; 612 max_poll_time = nseconds * EIO_TICKS;
571 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 613 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
572} 614}
573 615
574static void etp_set_max_poll_reqs (unsigned int maxreqs) 616static void etp_set_max_poll_reqs (unsigned int maxreqs)
575{ 617{
… …
767 809
768 X_LOCK (preadwritelock); 810 X_LOCK (preadwritelock);
769 ooffset = lseek (fd, 0, SEEK_CUR); 811 ooffset = lseek (fd, 0, SEEK_CUR);
770 lseek (fd, offset, SEEK_SET); 812 lseek (fd, offset, SEEK_SET);
771 res = write (fd, buf, count); 813 res = write (fd, buf, count);
772 lseek (fd, offset, SEEK_SET); 814 lseek (fd, ooffset, SEEK_SET);
773 X_UNLOCK (preadwritelock); 815 X_UNLOCK (preadwritelock);
774 816
775 return res; 817 return res;
776} 818}
777#endif 819#endif
… …
829 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); 871 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
830 } 872 }
831 873
832 res = sync_file_range (fd, offset, nbytes, flags); 874 res = sync_file_range (fd, offset, nbytes, flags);
833 875
834 if (res != ENOSYS) 876 if (!res || errno != ENOSYS)
835 return res; 877 return res;
836#endif 878#endif
837 879
838 /* even though we could play tricks with the flags, it's better to always 880 /* even though we could play tricks with the flags, it's better to always
839 * call fdatasync, as thta matches the expectation of it's users best */ 881 * call fdatasync, as that matches the expectation of its users best */
840 return fdatasync (fd); 882 return fdatasync (fd);
841} 883}
842 884
843#if !HAVE_READAHEAD 885#if !HAVE_READAHEAD
844# undef readahead 886# undef readahead
… …
876 918
877#if HAVE_SENDFILE 919#if HAVE_SENDFILE
878# if __linux 920# if __linux
879 res = sendfile (ofd, ifd, &offset, count); 921 res = sendfile (ofd, ifd, &offset, count);
880 922
881# elif __freebsd 923# elif __FreeBSD__
882 /* 924 /*
883 * Of course, the freebsd sendfile is a dire hack with no thoughts 925 * Of course, the freebsd sendfile is a dire hack with no thoughts
884 * wasted on making it similar to other I/O functions. 926 * wasted on making it similar to other I/O functions.
885 */ 927 */
886 { 928 {
887 off_t sbytes; 929 off_t sbytes;
888 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 930 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
889 931
890 if (res < 0 && sbytes) 932 #if 0 /* according to the manpage, this is correct, but broken behaviour */
891 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 933 /* freebsd' sendfile will return 0 on success */
934 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
935 /* not on e.g. EIO or EPIPE - sounds broken */
936 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
937 res = sbytes;
938 #endif
939
940 /* according to source inspection, this is correct, and useful behaviour */
941 if (sbytes)
942 res = sbytes;
943 }
944
945# elif defined (__APPLE__)
946
947 {
948 off_t sbytes = count;
949 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
950
951 /* according to the manpage, sbytes is always valid */
952 if (sbytes)
892 res = sbytes; 953 res = sbytes;
893 } 954 }
894 955
895# elif __hpux 956# elif __hpux
896 res = sendfile (ofd, ifd, offset, count, 0, 0); 957 res = sendfile (ofd, ifd, offset, count, 0, 0);
… …
910 if (res < 0 && sbytes) 971 if (res < 0 && sbytes)
911 res = sbytes; 972 res = sbytes;
912 } 973 }
913 974
914# endif 975# endif
976
977#elif defined (_WIN32)
978
979 /* does not work, just for documentation of what would need to be done */
980 {
981 HANDLE h = TO_SOCKET (ifd);
982 SetFilePointer (h, offset, 0, FILE_BEGIN);
983 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
984 }
985
915#else 986#else
916 res = -1; 987 res = -1;
917 errno = ENOSYS; 988 errno = ENOSYS;
918#endif 989#endif
919 990
920 if (res < 0 991 if (res < 0
921 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 992 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
993 /* BSDs */
994#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
995 || errno == ENOTSUP
996#endif
997 || errno == EOPNOTSUPP /* BSDs */
922#if __solaris 998#if __solaris
923 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 999 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
924#endif 1000#endif
925 ) 1001 )
926 ) 1002 )
… …
957 } 1033 }
958 1034
959 return res; 1035 return res;
960} 1036}
961 1037
1038static signed char
1039eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1040{
1041 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1042 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1043}
1044
1045#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1046
1047#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1048#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1049
1050static void
1051eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1052{
1053 unsigned char bits [9 + sizeof (ino_t) * 8];
1054 unsigned char *bit = bits;
1055
1056 assert (CHAR_BIT == 8);
1057 assert (sizeof (eio_dirent) * 8 < 256);
1058 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1059 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1060
1061 if (size <= EIO_SORT_FAST)
1062 return;
1063
1064 /* first prepare an array of bits to test in our radix sort */
1065 /* try to take endianness into account, as well as differences in ino_t sizes */
1066 /* inode_bits must contain all inodes ORed together */
1067 /* which is used to skip bits that are 0 everywhere, which is very common */
1068 {
1069 ino_t endianness;
1070 int i, j;
1071
1072 /* we store the byte offset of byte n into byte n of "endianness" */
1073 for (i = 0; i < sizeof (ino_t); ++i)
1074 ((unsigned char *)&endianness)[i] = i;
1075
1076 *bit++ = 0;
1077
1078 for (i = 0; i < sizeof (ino_t); ++i)
1079 {
1080 /* shifting off the byte offsets out of "endianness" */
1081 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1082 endianness >>= 8;
1083
1084 for (j = 0; j < 8; ++j)
1085 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1086 *bit++ = offs + j;
1087 }
1088
1089 for (j = 0; j < 8; ++j)
1090 if (score_bits & (1 << j))
1091 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1092 }
1093
1094 /* now actually do the sorting (a variant of MSD radix sort) */
1095 {
1096 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1097 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1098 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1099 int stk_idx = 0;
1100
1101 base_stk [stk_idx] = dents;
1102 end_stk [stk_idx] = dents + size;
1103 bit_stk [stk_idx] = bit - 1;
1104
1105 do
1106 {
1107 base = base_stk [stk_idx];
1108 end = end_stk [stk_idx];
1109 bit = bit_stk [stk_idx];
1110
1111 for (;;)
1112 {
1113 unsigned char O = *bit >> 3;
1114 unsigned char M = 1 << (*bit & 7);
1115
1116 eio_dirent *a = base;
1117 eio_dirent *b = end;
1118
1119 if (b - a < EIO_SORT_CUTOFF)
1120 break;
1121
1122 /* now bit-partition the array on the bit */
1123 /* this ugly asymmetric loop seems to perform much better than typical */
1124 /* partition algos found in the literature */
1125 do
1126 if (!(((unsigned char *)a)[O] & M))
1127 ++a;
1128 else if (!(((unsigned char *)--b)[O] & M))
1129 {
1130 eio_dirent tmp = *a; *a = *b; *b = tmp;
1131 ++a;
1132 }
1133 while (b > a);
1134
1135 /* next bit, or stop, if no bits left in this path */
1136 if (!*--bit)
1137 break;
1138
1139 base_stk [stk_idx] = a;
1140 end_stk [stk_idx] = end;
1141 bit_stk [stk_idx] = bit;
1142 ++stk_idx;
1143
1144 end = a;
1145 }
1146 }
1147 while (stk_idx--);
1148 }
1149}
1150
1151static void
1152eio_dent_insertion_sort (eio_dirent *dents, int size)
1153{
1154 /* first move the smallest element to the front, to act as a sentinel */
1155 {
1156 int i;
1157 eio_dirent *min = dents;
1158
1159 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1160 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1161 if (EIO_DENT_CMP (dents [i], <, *min))
1162 min = &dents [i];
1163
1164 /* swap elements 0 and j (minimum) */
1165 {
1166 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1167 }
1168 }
1169
1170 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1171 {
1172 eio_dirent *i, *j;
1173
1174 for (i = dents + 1; i < dents + size; ++i)
1175 {
1176 eio_dirent value = *i;
1177
1178 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1179 j [1] = j [0];
1180
1181 j [1] = value;
1182 }
1183 }
1184}
1185
1186static void
1187eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1188{
1189 if (size <= 1)
1190 return; /* our insertion sort relies on size > 0 */
1191
1192 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1193 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1194 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1195
1196 /* use an insertion sort at the end, or for small arrays, */
1197 /* as insertion sort is more efficient for small partitions */
1198 eio_dent_insertion_sort (dents, size);
1199}
1200
962/* read a full directory */ 1201/* read a full directory */
963static void 1202static void
964eio__scandir (eio_req *req, etp_worker *self) 1203eio__scandir (eio_req *req, etp_worker *self)
965{ 1204{
966 DIR *dirp; 1205 DIR *dirp;
967 EIO_STRUCT_DIRENT *entp; 1206 EIO_STRUCT_DIRENT *entp;
968 char *name, *names; 1207 char *name, *names;
969 int memlen = 4096; 1208 int namesalloc = 4096;
970 int memofs = 0; 1209 int namesoffs = 0;
1210 int flags = req->int1;
1211 eio_dirent *dents = 0;
1212 int dentalloc = 128;
971 int res = 0; 1213 int dentoffs = 0;
1214 ino_t inode_bits = 0;
1215
1216 req->result = -1;
1217
1218 if (!(flags & EIO_READDIR_DENTS))
1219 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
972 1220
973 X_LOCK (wrklock); 1221 X_LOCK (wrklock);
974 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1222 /* the corresponding closedir is in ETP_WORKER_CLEAR */
975 self->dirp = dirp = opendir (req->ptr1); 1223 self->dirp = dirp = opendir (req->ptr1);
976 req->flags |= EIO_FLAG_PTR2_FREE; 1224 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1225 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
977 req->ptr2 = names = malloc (memlen); 1226 req->ptr2 = names = malloc (namesalloc);
978 X_UNLOCK (wrklock); 1227 X_UNLOCK (wrklock);
979 1228
980 if (dirp && names) 1229 if (dirp && names && (!flags || dents))
981 for (;;) 1230 for (;;)
982 { 1231 {
983 errno = 0; 1232 errno = 0;
984 entp = readdir (dirp); 1233 entp = readdir (dirp);
985 1234
986 if (!entp) 1235 if (!entp)
1236 {
1237 if (errno)
1238 break;
1239
1240 /* sort etc. */
1241 req->int1 = flags;
1242 req->result = dentoffs;
1243
1244 if (flags & EIO_READDIR_STAT_ORDER)
1245 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
1246 else if (flags & EIO_READDIR_DIRS_FIRST)
1247 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1248 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1249 else
1250 {
1251 /* in this case, all is known, and we just put dirs first and sort them */
1252 eio_dirent *oth = dents + dentoffs;
1253 eio_dirent *dir = dents;
1254
1255 /* now partition dirs to the front, and non-dirs to the back */
1256 /* by walking from both sides and swapping if necessary */
1257 /* also clear score, so it doesn't influence sorting */
1258 while (oth > dir)
1259 {
1260 if (dir->type == EIO_DT_DIR)
1261 ++dir;
1262 else if ((--oth)->type == EIO_DT_DIR)
1263 {
1264 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1265
1266 ++dir;
1267 }
1268 }
1269
1270 /* now sort the dirs only */
1271 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1272 }
1273
987 break; 1274 break;
1275 }
988 1276
1277 /* now add the entry to our list(s) */
989 name = entp->d_name; 1278 name = entp->d_name;
990 1279
1280 /* skip . and .. entries */
991 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1281 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
992 { 1282 {
993 int len = strlen (name) + 1; 1283 int len = D_NAMLEN (entp) + 1;
994 1284
995 res++; 1285 while (expect_false (namesoffs + len > namesalloc))
996
997 while (memofs + len > memlen)
998 { 1286 {
999 memlen *= 2; 1287 namesalloc *= 2;
1000 X_LOCK (wrklock); 1288 X_LOCK (wrklock);
1001 req->ptr2 = names = realloc (names, memlen); 1289 req->ptr2 = names = realloc (names, namesalloc);
1002 X_UNLOCK (wrklock); 1290 X_UNLOCK (wrklock);
1003 1291
1004 if (!names) 1292 if (!names)
1005 break; 1293 break;
1006 } 1294 }
1007 1295
1008 memcpy (names + memofs, name, len); 1296 memcpy (names + namesoffs, name, len);
1297
1298 if (dents)
1299 {
1300 struct eio_dirent *ent;
1301
1302 if (expect_false (dentoffs == dentalloc))
1303 {
1304 dentalloc *= 2;
1305 X_LOCK (wrklock);
1306 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1307 X_UNLOCK (wrklock);
1308
1309 if (!dents)
1310 break;
1311 }
1312
1313 ent = dents + dentoffs;
1314
1315 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1316 ent->namelen = len - 1;
1317 ent->inode = D_INO (entp);
1318
1319 inode_bits |= ent->inode;
1320
1321 switch (D_TYPE (entp))
1322 {
1323 default:
1324 ent->type = EIO_DT_UNKNOWN;
1325 flags |= EIO_READDIR_FOUND_UNKNOWN;
1326 break;
1327
1328 #ifdef DT_FIFO
1329 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1330 #endif
1331 #ifdef DT_CHR
1332 case DT_CHR: ent->type = EIO_DT_CHR; break;
1333 #endif
1334 #ifdef DT_MPC
1335 case DT_MPC: ent->type = EIO_DT_MPC; break;
1336 #endif
1337 #ifdef DT_DIR
1338 case DT_DIR: ent->type = EIO_DT_DIR; break;
1339 #endif
1340 #ifdef DT_NAM
1341 case DT_NAM: ent->type = EIO_DT_NAM; break;
1342 #endif
1343 #ifdef DT_BLK
1344 case DT_BLK: ent->type = EIO_DT_BLK; break;
1345 #endif
1346 #ifdef DT_MPB
1347 case DT_MPB: ent->type = EIO_DT_MPB; break;
1348 #endif
1349 #ifdef DT_REG
1350 case DT_REG: ent->type = EIO_DT_REG; break;
1351 #endif
1352 #ifdef DT_NWK
1353 case DT_NWK: ent->type = EIO_DT_NWK; break;
1354 #endif
1355 #ifdef DT_CMP
1356 case DT_CMP: ent->type = EIO_DT_CMP; break;
1357 #endif
1358 #ifdef DT_LNK
1359 case DT_LNK: ent->type = EIO_DT_LNK; break;
1360 #endif
1361 #ifdef DT_SOCK
1362 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1363 #endif
1364 #ifdef DT_DOOR
1365 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1366 #endif
1367 #ifdef DT_WHT
1368 case DT_WHT: ent->type = EIO_DT_WHT; break;
1369 #endif
1370 }
1371
1372 ent->score = 7;
1373
1374 if (flags & EIO_READDIR_DIRS_FIRST)
1375 {
1376 if (ent->type == EIO_DT_UNKNOWN)
1377 {
1378 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1379 ent->score = 1;
1380 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1381 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1382 }
1383 else if (ent->type == EIO_DT_DIR)
1384 ent->score = 0;
1385 }
1386 }
1387
1009 memofs += len; 1388 namesoffs += len;
1389 ++dentoffs;
1390 }
1391
1392 if (EIO_CANCELLED (req))
1393 {
1394 errno = ECANCELED;
1395 break;
1010 } 1396 }
1011 } 1397 }
1012
1013 if (errno)
1014 res = -1;
1015
1016 req->result = res;
1017} 1398}
1399
1400#ifdef PAGESIZE
1401# define eio_pagesize() PAGESIZE
1402#else
1403static intptr_t
1404eio_pagesize (void)
1405{
1406 static intptr_t page;
1407
1408 if (!page)
1409 page = sysconf (_SC_PAGESIZE);
1410
1411 return page;
1412}
1413#endif
1414
1415static void
1416eio_page_align (void **addr, size_t *length)
1417{
1418 intptr_t mask = eio_pagesize () - 1;
1419
1420 /* round down addr */
1421 intptr_t adj = mask & (intptr_t)*addr;
1422
1423 *addr = (void *)((intptr_t)*addr - adj);
1424 *length += adj;
1425
1426 /* round up length */
1427 *length = (*length + mask) & ~mask;
1428}
1429
1430#if !_POSIX_MEMLOCK
1431# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1432# define eio__mlockall(a) ((errno = ENOSYS), -1)
1433#else
1434
1435static int
1436eio__mlock (void *addr, size_t length)
1437{
1438 eio_page_align (&addr, &length);
1439
1440 mlock (addr, length);
1441}
1442
1443static int
1444eio__mlockall (int flags)
1445{
1446 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1447 extern int mallopt (int, int);
1448 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1449 #endif
1450
1451 if (EIO_MCL_CURRENT != MCL_CURRENT
1452 || EIO_MCL_FUTURE != MCL_FUTURE)
1453 {
1454 flags = 0
1455 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1456 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1457 }
1458
1459 mlockall (flags);
1460}
1461#endif
1018 1462
1019#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1463#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1020# undef msync 1464# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1021# define msync(a,b,c) ENOSYS 1465#else
1466
1467int
1468eio__msync (void *mem, size_t len, int flags)
1469{
1470 eio_page_align (&mem, &len);
1471
1472 if (EIO_MS_ASYNC != MS_SYNC
1473 || EIO_MS_INVALIDATE != MS_INVALIDATE
1474 || EIO_MS_SYNC != MS_SYNC)
1475 {
1476 flags = 0
1477 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1478 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1479 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1480 }
1481
1482 return msync (mem, len, flags);
1483}
1484
1022#endif 1485#endif
1023 1486
1024int 1487int
1025eio__mtouch (void *mem, size_t len, int flags) 1488eio__mtouch (void *mem, size_t len, int flags)
1026{ 1489{
1490 eio_page_align (&mem, &len);
1491
1492 {
1027 intptr_t addr = (intptr_t)mem; 1493 intptr_t addr = (intptr_t)mem;
1028 intptr_t end = addr + len; 1494 intptr_t end = addr + len;
1029#ifdef PAGESIZE 1495 intptr_t page = eio_pagesize ();
1030 const intptr_t page = PAGESIZE;
1031#else
1032 static intptr_t page;
1033 1496
1034 if (!page)
1035 page = sysconf (_SC_PAGESIZE);
1036#endif
1037
1038 addr &= ~(page - 1); /* assume page size is always a power of two */
1039
1040 if (addr < end) 1497 if (addr < end)
1041 if (flags) /* modify */ 1498 if (flags & EIO_MT_MODIFY) /* modify */
1042 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1499 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1043 else 1500 else
1044 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1501 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1502 }
1045 1503
1046 return 0; 1504 return 0;
1047} 1505}
1048 1506
1049/*****************************************************************************/ 1507/*****************************************************************************/
… …
1194 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1652 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1195 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1653 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1196 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1654 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1655 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1198 1656
1657 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1658 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1659 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1660 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1661
1199 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1662 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1200 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1663 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1201 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1664 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1202 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1665 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1203 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1666 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
… …
1212 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1675 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1213 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1676 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1214 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1677 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1215 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 1678 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1216 1679
1217 case EIO_READLINK: ALLOC (NAME_MAX); 1680 case EIO_READLINK: ALLOC (PATH_MAX);
1218 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1681 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1219 1682
1220 case EIO_SYNC: req->result = 0; sync (); break; 1683 case EIO_SYNC: req->result = 0; sync (); break;
1221 case EIO_FSYNC: req->result = fsync (req->int1); break; 1684 case EIO_FSYNC: req->result = fsync (req->int1); break;
1222 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1685 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1223 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1686 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1224 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 1687 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1688 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1689 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1225 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 1690 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1226 1691
1227 case EIO_READDIR: eio__scandir (req, self); break; 1692 case EIO_READDIR: eio__scandir (req, self); break;
1228 1693
1229 case EIO_BUSY: 1694 case EIO_BUSY:
… …
1315eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 1780eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1316{ 1781{
1317 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 1782 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1318} 1783}
1319 1784
1785eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1786{
1787 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1788}
1789
1790eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1791{
1792 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1793}
1794
1320eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 1795eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1321{ 1796{
1322 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 1797 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1323} 1798}
1324 1799
… …
1348} 1823}
1349 1824
1350eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1825eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1351{ 1826{
1352 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1827 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1828}
1829
1830eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1831{
1832 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1353} 1833}
1354 1834
1355eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1835eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1356{ 1836{
1357 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1837 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
… …
1431eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 1911eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1432{ 1912{
1433 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1913 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1434} 1914}
1435 1915
1916eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1917{
1918 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1919}
1920
1436eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1921eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1437{ 1922{
1438 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1923 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1439} 1924}
1440 1925
1441eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1926eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1442{ 1927{
1443 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1928 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1444} 1929}
1445 1930
1446eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1931eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1447{ 1932{
1448 return eio__1path (EIO_READDIR, path, pri, cb, data); 1933 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1449} 1934}
1450 1935
1451eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1936eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1452{ 1937{
1453 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1938 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
… …
1542/* misc garbage */ 2027/* misc garbage */
1543 2028
1544ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2029ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1545{ 2030{
1546 etp_worker wrk; 2031 etp_worker wrk;
2032 ssize_t ret;
1547 2033
1548 wrk.dbuf = 0; 2034 wrk.dbuf = 0;
1549 2035
1550 eio__sendfile (ofd, ifd, offset, count, &wrk); 2036 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1551 2037
1552 if (wrk.dbuf) 2038 if (wrk.dbuf)
1553 free (wrk.dbuf); 2039 free (wrk.dbuf);
1554}
1555 2040
2041 return ret;
2042}
2043

Diff Legend

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