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 1.59 by root, Wed Dec 1 07:30:39 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
778 820
779#ifndef HAVE_FUTIMES 821#ifndef HAVE_UTIMES
780 822
781# undef utimes 823# undef utimes
782# undef futimes
783# define utimes(path,times) eio__utimes (path, times) 824# define utimes(path,times) eio__utimes (path, times)
784# define futimes(fd,times) eio__futimes (fd, times)
785 825
786static int 826static int
787eio__utimes (const char *filename, const struct timeval times[2]) 827eio__utimes (const char *filename, const struct timeval times[2])
788{ 828{
789 if (times) 829 if (times)
… …
796 return utime (filename, &buf); 836 return utime (filename, &buf);
797 } 837 }
798 else 838 else
799 return utime (filename, 0); 839 return utime (filename, 0);
800} 840}
841
842#endif
843
844#ifndef HAVE_FUTIMES
845
846# undef futimes
847# define futimes(fd,times) eio__futimes (fd, times)
801 848
802static int eio__futimes (int fd, const struct timeval tv[2]) 849static int eio__futimes (int fd, const struct timeval tv[2])
803{ 850{
804 errno = ENOSYS; 851 errno = ENOSYS;
805 return -1; 852 return -1;
… …
829 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); 876 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
830 } 877 }
831 878
832 res = sync_file_range (fd, offset, nbytes, flags); 879 res = sync_file_range (fd, offset, nbytes, flags);
833 880
834 if (res != ENOSYS) 881 if (!res || errno != ENOSYS)
835 return res; 882 return res;
836#endif 883#endif
837 884
838 /* even though we could play tricks with the flags, it's better to always 885 /* 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 */ 886 * call fdatasync, as that matches the expectation of its users best */
840 return fdatasync (fd); 887 return fdatasync (fd);
841} 888}
842 889
843#if !HAVE_READAHEAD 890#if !HAVE_READAHEAD
844# undef readahead 891# undef readahead
… …
876 923
877#if HAVE_SENDFILE 924#if HAVE_SENDFILE
878# if __linux 925# if __linux
879 res = sendfile (ofd, ifd, &offset, count); 926 res = sendfile (ofd, ifd, &offset, count);
880 927
881# elif __freebsd 928# elif __FreeBSD__
882 /* 929 /*
883 * Of course, the freebsd sendfile is a dire hack with no thoughts 930 * Of course, the freebsd sendfile is a dire hack with no thoughts
884 * wasted on making it similar to other I/O functions. 931 * wasted on making it similar to other I/O functions.
885 */ 932 */
886 { 933 {
887 off_t sbytes; 934 off_t sbytes;
888 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 935 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
889 936
890 if (res < 0 && sbytes) 937 #if 0 /* according to the manpage, this is correct, but broken behaviour */
891 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 938 /* freebsd' sendfile will return 0 on success */
939 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
940 /* not on e.g. EIO or EPIPE - sounds broken */
941 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
942 res = sbytes;
943 #endif
944
945 /* according to source inspection, this is correct, and useful behaviour */
946 if (sbytes)
947 res = sbytes;
948 }
949
950# elif defined (__APPLE__)
951
952 {
953 off_t sbytes = count;
954 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
955
956 /* according to the manpage, sbytes is always valid */
957 if (sbytes)
892 res = sbytes; 958 res = sbytes;
893 } 959 }
894 960
895# elif __hpux 961# elif __hpux
896 res = sendfile (ofd, ifd, offset, count, 0, 0); 962 res = sendfile (ofd, ifd, offset, count, 0, 0);
… …
910 if (res < 0 && sbytes) 976 if (res < 0 && sbytes)
911 res = sbytes; 977 res = sbytes;
912 } 978 }
913 979
914# endif 980# endif
981
982#elif defined (_WIN32)
983
984 /* does not work, just for documentation of what would need to be done */
985 {
986 HANDLE h = TO_SOCKET (ifd);
987 SetFilePointer (h, offset, 0, FILE_BEGIN);
988 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
989 }
990
915#else 991#else
916 res = -1; 992 res = -1;
917 errno = ENOSYS; 993 errno = ENOSYS;
918#endif 994#endif
919 995
920 if (res < 0 996 if (res < 0
921 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 997 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
998 /* BSDs */
999#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1000 || errno == ENOTSUP
1001#endif
1002 || errno == EOPNOTSUPP /* BSDs */
922#if __solaris 1003#if __solaris
923 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1004 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
924#endif 1005#endif
925 ) 1006 )
926 ) 1007 )
… …
957 } 1038 }
958 1039
959 return res; 1040 return res;
960} 1041}
961 1042
1043static signed char
1044eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1045{
1046 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1047 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1048}
1049
1050#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1051
1052#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1053#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1054
1055static void
1056eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1057{
1058 unsigned char bits [9 + sizeof (ino_t) * 8];
1059 unsigned char *bit = bits;
1060
1061 assert (CHAR_BIT == 8);
1062 assert (sizeof (eio_dirent) * 8 < 256);
1063 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1064 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1065
1066 if (size <= EIO_SORT_FAST)
1067 return;
1068
1069 /* first prepare an array of bits to test in our radix sort */
1070 /* try to take endianness into account, as well as differences in ino_t sizes */
1071 /* inode_bits must contain all inodes ORed together */
1072 /* which is used to skip bits that are 0 everywhere, which is very common */
1073 {
1074 ino_t endianness;
1075 int i, j;
1076
1077 /* we store the byte offset of byte n into byte n of "endianness" */
1078 for (i = 0; i < sizeof (ino_t); ++i)
1079 ((unsigned char *)&endianness)[i] = i;
1080
1081 *bit++ = 0;
1082
1083 for (i = 0; i < sizeof (ino_t); ++i)
1084 {
1085 /* shifting off the byte offsets out of "endianness" */
1086 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1087 endianness >>= 8;
1088
1089 for (j = 0; j < 8; ++j)
1090 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1091 *bit++ = offs + j;
1092 }
1093
1094 for (j = 0; j < 8; ++j)
1095 if (score_bits & (1 << j))
1096 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1097 }
1098
1099 /* now actually do the sorting (a variant of MSD radix sort) */
1100 {
1101 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1102 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1103 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1104 int stk_idx = 0;
1105
1106 base_stk [stk_idx] = dents;
1107 end_stk [stk_idx] = dents + size;
1108 bit_stk [stk_idx] = bit - 1;
1109
1110 do
1111 {
1112 base = base_stk [stk_idx];
1113 end = end_stk [stk_idx];
1114 bit = bit_stk [stk_idx];
1115
1116 for (;;)
1117 {
1118 unsigned char O = *bit >> 3;
1119 unsigned char M = 1 << (*bit & 7);
1120
1121 eio_dirent *a = base;
1122 eio_dirent *b = end;
1123
1124 if (b - a < EIO_SORT_CUTOFF)
1125 break;
1126
1127 /* now bit-partition the array on the bit */
1128 /* this ugly asymmetric loop seems to perform much better than typical */
1129 /* partition algos found in the literature */
1130 do
1131 if (!(((unsigned char *)a)[O] & M))
1132 ++a;
1133 else if (!(((unsigned char *)--b)[O] & M))
1134 {
1135 eio_dirent tmp = *a; *a = *b; *b = tmp;
1136 ++a;
1137 }
1138 while (b > a);
1139
1140 /* next bit, or stop, if no bits left in this path */
1141 if (!*--bit)
1142 break;
1143
1144 base_stk [stk_idx] = a;
1145 end_stk [stk_idx] = end;
1146 bit_stk [stk_idx] = bit;
1147 ++stk_idx;
1148
1149 end = a;
1150 }
1151 }
1152 while (stk_idx--);
1153 }
1154}
1155
1156static void
1157eio_dent_insertion_sort (eio_dirent *dents, int size)
1158{
1159 /* first move the smallest element to the front, to act as a sentinel */
1160 {
1161 int i;
1162 eio_dirent *min = dents;
1163
1164 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1165 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1166 if (EIO_DENT_CMP (dents [i], <, *min))
1167 min = &dents [i];
1168
1169 /* swap elements 0 and j (minimum) */
1170 {
1171 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1172 }
1173 }
1174
1175 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1176 {
1177 eio_dirent *i, *j;
1178
1179 for (i = dents + 1; i < dents + size; ++i)
1180 {
1181 eio_dirent value = *i;
1182
1183 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1184 j [1] = j [0];
1185
1186 j [1] = value;
1187 }
1188 }
1189}
1190
1191static void
1192eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1193{
1194 if (size <= 1)
1195 return; /* our insertion sort relies on size > 0 */
1196
1197 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1198 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1199 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1200
1201 /* use an insertion sort at the end, or for small arrays, */
1202 /* as insertion sort is more efficient for small partitions */
1203 eio_dent_insertion_sort (dents, size);
1204}
1205
962/* read a full directory */ 1206/* read a full directory */
963static void 1207static void
964eio__scandir (eio_req *req, etp_worker *self) 1208eio__scandir (eio_req *req, etp_worker *self)
965{ 1209{
966 DIR *dirp; 1210 DIR *dirp;
967 EIO_STRUCT_DIRENT *entp; 1211 EIO_STRUCT_DIRENT *entp;
968 char *name, *names; 1212 char *name, *names;
969 int memlen = 4096; 1213 int namesalloc = 4096;
970 int memofs = 0; 1214 int namesoffs = 0;
1215 int flags = req->int1;
1216 eio_dirent *dents = 0;
1217 int dentalloc = 128;
971 int res = 0; 1218 int dentoffs = 0;
1219 ino_t inode_bits = 0;
1220
1221 req->result = -1;
1222
1223 if (!(flags & EIO_READDIR_DENTS))
1224 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
972 1225
973 X_LOCK (wrklock); 1226 X_LOCK (wrklock);
974 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1227 /* the corresponding closedir is in ETP_WORKER_CLEAR */
975 self->dirp = dirp = opendir (req->ptr1); 1228 self->dirp = dirp = opendir (req->ptr1);
976 req->flags |= EIO_FLAG_PTR2_FREE; 1229 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1230 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
977 req->ptr2 = names = malloc (memlen); 1231 req->ptr2 = names = malloc (namesalloc);
978 X_UNLOCK (wrklock); 1232 X_UNLOCK (wrklock);
979 1233
980 if (dirp && names) 1234 if (dirp && names && (!flags || dents))
981 for (;;) 1235 for (;;)
982 { 1236 {
983 errno = 0; 1237 errno = 0;
984 entp = readdir (dirp); 1238 entp = readdir (dirp);
985 1239
986 if (!entp) 1240 if (!entp)
1241 {
1242 if (errno)
1243 break;
1244
1245 /* sort etc. */
1246 req->int1 = flags;
1247 req->result = dentoffs;
1248
1249 if (flags & EIO_READDIR_STAT_ORDER)
1250 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
1251 else if (flags & EIO_READDIR_DIRS_FIRST)
1252 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1253 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1254 else
1255 {
1256 /* in this case, all is known, and we just put dirs first and sort them */
1257 eio_dirent *oth = dents + dentoffs;
1258 eio_dirent *dir = dents;
1259
1260 /* now partition dirs to the front, and non-dirs to the back */
1261 /* by walking from both sides and swapping if necessary */
1262 /* also clear score, so it doesn't influence sorting */
1263 while (oth > dir)
1264 {
1265 if (dir->type == EIO_DT_DIR)
1266 ++dir;
1267 else if ((--oth)->type == EIO_DT_DIR)
1268 {
1269 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1270
1271 ++dir;
1272 }
1273 }
1274
1275 /* now sort the dirs only */
1276 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1277 }
1278
987 break; 1279 break;
1280 }
988 1281
1282 /* now add the entry to our list(s) */
989 name = entp->d_name; 1283 name = entp->d_name;
990 1284
1285 /* skip . and .. entries */
991 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1286 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
992 { 1287 {
993 int len = strlen (name) + 1; 1288 int len = D_NAMLEN (entp) + 1;
994 1289
995 res++; 1290 while (expect_false (namesoffs + len > namesalloc))
996
997 while (memofs + len > memlen)
998 { 1291 {
999 memlen *= 2; 1292 namesalloc *= 2;
1000 X_LOCK (wrklock); 1293 X_LOCK (wrklock);
1001 req->ptr2 = names = realloc (names, memlen); 1294 req->ptr2 = names = realloc (names, namesalloc);
1002 X_UNLOCK (wrklock); 1295 X_UNLOCK (wrklock);
1003 1296
1004 if (!names) 1297 if (!names)
1005 break; 1298 break;
1006 } 1299 }
1007 1300
1008 memcpy (names + memofs, name, len); 1301 memcpy (names + namesoffs, name, len);
1302
1303 if (dents)
1304 {
1305 struct eio_dirent *ent;
1306
1307 if (expect_false (dentoffs == dentalloc))
1308 {
1309 dentalloc *= 2;
1310 X_LOCK (wrklock);
1311 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1312 X_UNLOCK (wrklock);
1313
1314 if (!dents)
1315 break;
1316 }
1317
1318 ent = dents + dentoffs;
1319
1320 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1321 ent->namelen = len - 1;
1322 ent->inode = D_INO (entp);
1323
1324 inode_bits |= ent->inode;
1325
1326 switch (D_TYPE (entp))
1327 {
1328 default:
1329 ent->type = EIO_DT_UNKNOWN;
1330 flags |= EIO_READDIR_FOUND_UNKNOWN;
1331 break;
1332
1333 #ifdef DT_FIFO
1334 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1335 #endif
1336 #ifdef DT_CHR
1337 case DT_CHR: ent->type = EIO_DT_CHR; break;
1338 #endif
1339 #ifdef DT_MPC
1340 case DT_MPC: ent->type = EIO_DT_MPC; break;
1341 #endif
1342 #ifdef DT_DIR
1343 case DT_DIR: ent->type = EIO_DT_DIR; break;
1344 #endif
1345 #ifdef DT_NAM
1346 case DT_NAM: ent->type = EIO_DT_NAM; break;
1347 #endif
1348 #ifdef DT_BLK
1349 case DT_BLK: ent->type = EIO_DT_BLK; break;
1350 #endif
1351 #ifdef DT_MPB
1352 case DT_MPB: ent->type = EIO_DT_MPB; break;
1353 #endif
1354 #ifdef DT_REG
1355 case DT_REG: ent->type = EIO_DT_REG; break;
1356 #endif
1357 #ifdef DT_NWK
1358 case DT_NWK: ent->type = EIO_DT_NWK; break;
1359 #endif
1360 #ifdef DT_CMP
1361 case DT_CMP: ent->type = EIO_DT_CMP; break;
1362 #endif
1363 #ifdef DT_LNK
1364 case DT_LNK: ent->type = EIO_DT_LNK; break;
1365 #endif
1366 #ifdef DT_SOCK
1367 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1368 #endif
1369 #ifdef DT_DOOR
1370 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1371 #endif
1372 #ifdef DT_WHT
1373 case DT_WHT: ent->type = EIO_DT_WHT; break;
1374 #endif
1375 }
1376
1377 ent->score = 7;
1378
1379 if (flags & EIO_READDIR_DIRS_FIRST)
1380 {
1381 if (ent->type == EIO_DT_UNKNOWN)
1382 {
1383 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1384 ent->score = 1;
1385 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1386 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1387 }
1388 else if (ent->type == EIO_DT_DIR)
1389 ent->score = 0;
1390 }
1391 }
1392
1009 memofs += len; 1393 namesoffs += len;
1394 ++dentoffs;
1395 }
1396
1397 if (EIO_CANCELLED (req))
1398 {
1399 errno = ECANCELED;
1400 break;
1010 } 1401 }
1011 } 1402 }
1012
1013 if (errno)
1014 res = -1;
1015
1016 req->result = res;
1017} 1403}
1404
1405#ifdef PAGESIZE
1406# define eio_pagesize() PAGESIZE
1407#else
1408static intptr_t
1409eio_pagesize (void)
1410{
1411 static intptr_t page;
1412
1413 if (!page)
1414 page = sysconf (_SC_PAGESIZE);
1415
1416 return page;
1417}
1418#endif
1419
1420static void
1421eio_page_align (void **addr, size_t *length)
1422{
1423 intptr_t mask = eio_pagesize () - 1;
1424
1425 /* round down addr */
1426 intptr_t adj = mask & (intptr_t)*addr;
1427
1428 *addr = (void *)((intptr_t)*addr - adj);
1429 *length += adj;
1430
1431 /* round up length */
1432 *length = (*length + mask) & ~mask;
1433}
1434
1435#if !_POSIX_MEMLOCK
1436# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1437# define eio__mlockall(a) ((errno = ENOSYS), -1)
1438#else
1439
1440static int
1441eio__mlock (void *addr, size_t length)
1442{
1443 eio_page_align (&addr, &length);
1444
1445 return mlock (addr, length);
1446}
1447
1448static int
1449eio__mlockall (int flags)
1450{
1451 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1452 extern int mallopt (int, int);
1453 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1454 #endif
1455
1456 if (EIO_MCL_CURRENT != MCL_CURRENT
1457 || EIO_MCL_FUTURE != MCL_FUTURE)
1458 {
1459 flags = 0
1460 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1461 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1462 }
1463
1464 return mlockall (flags);
1465}
1466#endif
1018 1467
1019#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1468#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1020# undef msync 1469# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1021# define msync(a,b,c) ENOSYS 1470#else
1471
1472int
1473eio__msync (void *mem, size_t len, int flags)
1474{
1475 eio_page_align (&mem, &len);
1476
1477 if (EIO_MS_ASYNC != MS_SYNC
1478 || EIO_MS_INVALIDATE != MS_INVALIDATE
1479 || EIO_MS_SYNC != MS_SYNC)
1480 {
1481 flags = 0
1482 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1483 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1484 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1485 }
1486
1487 return msync (mem, len, flags);
1488}
1489
1022#endif 1490#endif
1023 1491
1024int 1492int
1025eio__mtouch (void *mem, size_t len, int flags) 1493eio__mtouch (void *mem, size_t len, int flags)
1026{ 1494{
1495 eio_page_align (&mem, &len);
1496
1497 {
1027 intptr_t addr = (intptr_t)mem; 1498 intptr_t addr = (intptr_t)mem;
1028 intptr_t end = addr + len; 1499 intptr_t end = addr + len;
1029#ifdef PAGESIZE 1500 intptr_t page = eio_pagesize ();
1030 const intptr_t page = PAGESIZE;
1031#else
1032 static intptr_t page;
1033 1501
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) 1502 if (addr < end)
1041 if (flags) /* modify */ 1503 if (flags & EIO_MT_MODIFY) /* modify */
1042 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); 1504 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1043 else 1505 else
1044 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); 1506 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1507 }
1045 1508
1046 return 0; 1509 return 0;
1047} 1510}
1048 1511
1049/*****************************************************************************/ 1512/*****************************************************************************/
… …
1172 return 0; \ 1635 return 0; \
1173 } 1636 }
1174 1637
1175static void eio_execute (etp_worker *self, eio_req *req) 1638static void eio_execute (etp_worker *self, eio_req *req)
1176{ 1639{
1177 errno = 0;
1178
1179 switch (req->type) 1640 switch (req->type)
1180 { 1641 {
1181 case EIO_READ: ALLOC (req->size); 1642 case EIO_READ: ALLOC (req->size);
1182 req->result = req->offs >= 0 1643 req->result = req->offs >= 0
1183 ? pread (req->int1, req->ptr2, req->size, req->offs) 1644 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1193 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1654 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1194 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1655 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1195 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1656 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1196 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1657 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1658 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1659
1660 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1661 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1662 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1663 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1198 1664
1199 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1665 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; 1666 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; 1667 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; 1668 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
… …
1212 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 1678 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1213 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 1679 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1214 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 1680 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; 1681 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1216 1682
1217 case EIO_READLINK: ALLOC (NAME_MAX); 1683 case EIO_READLINK: ALLOC (PATH_MAX);
1218 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 1684 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1219 1685
1220 case EIO_SYNC: req->result = 0; sync (); break; 1686 case EIO_SYNC: req->result = 0; sync (); break;
1221 case EIO_FSYNC: req->result = fsync (req->int1); break; 1687 case EIO_FSYNC: req->result = fsync (req->int1); break;
1222 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 1688 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1223 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 1689 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; 1690 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1691 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1692 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; 1693 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1226 1694
1227 case EIO_READDIR: eio__scandir (req, self); break; 1695 case EIO_READDIR: eio__scandir (req, self); break;
1228 1696
1229 case EIO_BUSY: 1697 case EIO_BUSY:
1230#ifdef _WIN32 1698#ifdef _WIN32
1231 Sleep (req->nv1 * 1000.); 1699 Sleep (req->nv1 * 1e3);
1232#else 1700#else
1233 { 1701 {
1234 struct timeval tv; 1702 struct timeval tv;
1235 1703
1236 tv.tv_sec = req->nv1; 1704 tv.tv_sec = req->nv1;
1237 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 1705 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1238 1706
1239 req->result = select (0, 0, 0, 0, &tv); 1707 req->result = select (0, 0, 0, 0, &tv);
1240 } 1708 }
1241#endif 1709#endif
1242 break; 1710 break;
… …
1257 times = tv; 1725 times = tv;
1258 } 1726 }
1259 else 1727 else
1260 times = 0; 1728 times = 0;
1261 1729
1262
1263 req->result = req->type == EIO_FUTIME 1730 req->result = req->type == EIO_FUTIME
1264 ? futimes (req->int1, times) 1731 ? futimes (req->int1, times)
1265 : utimes (req->ptr1, times); 1732 : utimes (req->ptr1, times);
1266 } 1733 }
1267 break; 1734 break;
… …
1276 case EIO_CUSTOM: 1743 case EIO_CUSTOM:
1277 ((void (*)(eio_req *))req->feed) (req); 1744 ((void (*)(eio_req *))req->feed) (req);
1278 break; 1745 break;
1279 1746
1280 default: 1747 default:
1748 errno = ENOSYS;
1281 req->result = -1; 1749 req->result = -1;
1282 break; 1750 break;
1283 } 1751 }
1284 1752
1285 req->errorno = errno; 1753 req->errorno = errno;
… …
1315eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 1783eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1316{ 1784{
1317 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 1785 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1318} 1786}
1319 1787
1788eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1789{
1790 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1791}
1792
1793eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1794{
1795 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1796}
1797
1320eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 1798eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1321{ 1799{
1322 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 1800 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1323} 1801}
1324 1802
… …
1348} 1826}
1349 1827
1350eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 1828eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1351{ 1829{
1352 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1830 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1831}
1832
1833eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1834{
1835 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1353} 1836}
1354 1837
1355eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 1838eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1356{ 1839{
1357 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1840 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) 1914eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1432{ 1915{
1433 return eio__1path (EIO_LSTAT, path, pri, cb, data); 1916 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1434} 1917}
1435 1918
1919eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1920{
1921 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1922}
1923
1436eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 1924eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1437{ 1925{
1438 return eio__1path (EIO_UNLINK, path, pri, cb, data); 1926 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1439} 1927}
1440 1928
1441eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1929eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1442{ 1930{
1443 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1931 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1444} 1932}
1445 1933
1446eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1934eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1447{ 1935{
1448 return eio__1path (EIO_READDIR, path, pri, cb, data); 1936 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1449} 1937}
1450 1938
1451eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1939eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1452{ 1940{
1453 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1941 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
… …
1542/* misc garbage */ 2030/* misc garbage */
1543 2031
1544ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2032ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1545{ 2033{
1546 etp_worker wrk; 2034 etp_worker wrk;
2035 ssize_t ret;
1547 2036
1548 wrk.dbuf = 0; 2037 wrk.dbuf = 0;
1549 2038
1550 eio__sendfile (ofd, ifd, offset, count, &wrk); 2039 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1551 2040
1552 if (wrk.dbuf) 2041 if (wrk.dbuf)
1553 free (wrk.dbuf); 2042 free (wrk.dbuf);
1554}
1555 2043
2044 return ret;
2045}
2046

Diff Legend

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