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 1.40 by root, Sat Jun 13 14:58:33 2009 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 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>
64#endif 68#endif
65 69
66#ifdef _WIN32 70#ifdef _WIN32
67 71
68 /*doh*/ 72 /*doh*/
69
70#else 73#else
71 74
72# include "config.h" 75# include "config.h"
73# include <sys/time.h> 76# include <sys/time.h>
74# include <sys/select.h> 77# include <sys/select.h>
78# include <sys/mman.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>
83
84/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
85# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
86# define _DIRENT_HAVE_D_TYPE /* sigh */
87# define D_INO(de) (de)->d_fileno
88# define D_NAMLEN(de) (de)->d_namlen
89# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
90# define D_INO(de) (de)->d_ino
91# endif
92
93#ifdef _D_EXACT_NAMLEN
94# undef D_NAMLEN
95# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
96#endif
97
98# ifdef _DIRENT_HAVE_D_TYPE
99# define D_TYPE(de) (de)->d_type
100# endif
79 101
80# ifndef EIO_STRUCT_DIRENT 102# ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 103# define EIO_STRUCT_DIRENT struct dirent
82# endif 104# endif
83 105
96# else 118# else
97# error sendfile support requested but not available 119# error sendfile support requested but not available
98# endif 120# endif
99#endif 121#endif
100 122
123#ifndef D_TYPE
124# define D_TYPE(de) 0
125#endif
126#ifndef D_INO
127# define D_INO(de) 0
128#endif
129#ifndef D_NAMLEN
130# define D_NAMLEN(de) strlen ((de)->d_name)
131#endif
132
101/* number of seconds after which an idle threads exit */ 133/* number of seconds after which an idle threads exit */
102#define IDLE_TIMEOUT 10 134#define IDLE_TIMEOUT 10
103 135
104/* used for struct dirent, AIX doesn't provide it */ 136/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 137#ifndef NAME_MAX
155 if (wrk->dirp) \ 187 if (wrk->dirp) \
156 { \ 188 { \
157 closedir (wrk->dirp); \ 189 closedir (wrk->dirp); \
158 wrk->dirp = 0; \ 190 wrk->dirp = 0; \
159 } 191 }
192
160#define ETP_WORKER_COMMON \ 193#define ETP_WORKER_COMMON \
161 void *dbuf; \ 194 void *dbuf; \
162 DIR *dirp; 195 DIR *dirp;
163 196
164/*****************************************************************************/ 197/*****************************************************************************/
767 800
768 X_LOCK (preadwritelock); 801 X_LOCK (preadwritelock);
769 ooffset = lseek (fd, 0, SEEK_CUR); 802 ooffset = lseek (fd, 0, SEEK_CUR);
770 lseek (fd, offset, SEEK_SET); 803 lseek (fd, offset, SEEK_SET);
771 res = write (fd, buf, count); 804 res = write (fd, buf, count);
772 lseek (fd, offset, SEEK_SET); 805 lseek (fd, ooffset, SEEK_SET);
773 X_UNLOCK (preadwritelock); 806 X_UNLOCK (preadwritelock);
774 807
775 return res; 808 return res;
776} 809}
777#endif 810#endif
829 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); 862 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
830 } 863 }
831 864
832 res = sync_file_range (fd, offset, nbytes, flags); 865 res = sync_file_range (fd, offset, nbytes, flags);
833 866
834 if (res != ENOSYS) 867 if (!res || errno != ENOSYS)
835 return res; 868 return res;
836#endif 869#endif
837 870
838 /* even though we could play tricks with the flags, it's better to always 871 /* 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 */ 872 * call fdatasync, as thta matches the expectation of it's users best */
957 } 990 }
958 991
959 return res; 992 return res;
960} 993}
961 994
995static signed char
996eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
997{
998 return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */
999 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1000}
1001
1002#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1003
1004#define EIO_QSORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1005#define EIO_QSORT_SKIP 60 /* when to skip qsort completely */
1006
1007static void
1008eio_dent_sort (eio_dirent *dents, int size)
1009{
1010 int i, j;
1011
1012 if (size <= 1)
1013 return; /* our insertion sort relies on size > 0 */
1014
1015 if (size > EIO_QSORT_SKIP) /* skip quicksort for small directories */
1016 {
1017 /* first, use quicksort */
1018 /* should be good for 2**31 entries */
1019 struct rng { int l, r; } rng [32];
1020
1021 i = 0;
1022 rng[0].l = 0;
1023 rng[0].r = size;
1024
1025 while (expect_true (i >= 0))
1026 {
1027 int L = rng [i].l;
1028 int R = rng [i].r - 1;
1029
1030 if (expect_false (L + EIO_QSORT_CUTOFF < R))
1031 {
1032 eio_dirent piv = dents [L];
1033
1034 while (L < R)
1035 {
1036 while (EIO_DENT_CMP (dents [R], >=, piv) && L < R)
1037 --R;
1038
1039 if (L < R)
1040 dents [L++] = dents [R];
1041
1042 while (EIO_DENT_CMP (dents [L], <=, piv) && L < R)
1043 ++L;
1044
1045 if (L < R)
1046 dents [R--] = dents [L];
1047 }
1048
1049 dents [L] = piv;
1050
1051 ++i;
1052 rng [i].l = L + 1;
1053 rng [i].r = rng [i - 1].r;
1054 rng [i - 1].r = L;
1055
1056 if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l)
1057 {
1058 struct rng t;
1059
1060 t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t;
1061 }
1062 }
1063 else
1064 --i;
1065 }
1066 }
1067
1068 /* use an insertion sort after qsort, or for small arrays */
1069 /* first move the smallest element to the front, to act as a sentinel */
1070 {
1071 int min = 0;
1072
1073 for (i = size > EIO_QSORT_SKIP ? EIO_QSORT_CUTOFF + 1 : size; --i; )
1074 if (EIO_DENT_CMP (dents [i], <, dents [min]))
1075 min = i;
1076
1077 /* swap elements 0 and j (minimum) */
1078 {
1079 eio_dirent tmp = dents [0]; dents [0] = dents [min]; dents [min] = tmp;
1080 }
1081 }
1082
1083 /* then do standard insertion sort */
1084 for (i = 1; i < size; ++i)
1085 {
1086 eio_dirent value = dents [i];
1087
1088 for (j = i - 1; EIO_DENT_CMP (dents [j], >, value); --j)
1089 {
1090 assert (j >= 0);
1091 dents [j + 1] = dents [j];
1092 }
1093
1094 dents [j + 1] = value;
1095 }
1096}
1097
962/* read a full directory */ 1098/* read a full directory */
963static void 1099static void
964eio__scandir (eio_req *req, etp_worker *self) 1100eio__scandir (eio_req *req, etp_worker *self)
965{ 1101{
966 DIR *dirp; 1102 DIR *dirp;
967 EIO_STRUCT_DIRENT *entp; 1103 EIO_STRUCT_DIRENT *entp;
968 char *name, *names; 1104 char *name, *names;
969 int memlen = 4096; 1105 int namesalloc = 4096;
970 int memofs = 0; 1106 int namesoffs = 0;
1107 int flags = req->int1;
1108 eio_dirent *dents = 0;
1109 int dentalloc = 128;
971 int res = 0; 1110 int dentoffs = 0;
1111
1112 req->result = -1;
1113
1114 if (!(flags & EIO_READDIR_DENTS))
1115 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
972 1116
973 X_LOCK (wrklock); 1117 X_LOCK (wrklock);
974 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1118 /* the corresponding closedir is in ETP_WORKER_CLEAR */
975 self->dirp = dirp = opendir (req->ptr1); 1119 self->dirp = dirp = opendir (req->ptr1);
976 req->flags |= EIO_FLAG_PTR2_FREE; 1120 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
977 req->ptr2 = names = malloc (memlen); 1121 req->ptr1 = names = malloc (namesalloc);
1122 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
978 X_UNLOCK (wrklock); 1123 X_UNLOCK (wrklock);
979 1124
980 if (dirp && names) 1125 if (dirp && names && (!flags || dents))
981 for (;;) 1126 for (;;)
982 { 1127 {
983 errno = 0; 1128 errno = 0;
984 entp = readdir (dirp); 1129 entp = readdir (dirp);
985 1130
986 if (!entp) 1131 if (!entp)
1132 {
1133 if (errno)
1134 break;
1135
1136 /* sort etc. */
1137 req->int1 = flags;
1138 req->result = dentoffs;
1139
1140 if (dents)
1141 {
1142 eio_dirent *ent = dents + dentoffs;
1143
1144 while (ent > dents)
1145 (--ent)->name = names + (size_t)ent->name;
1146 }
1147
1148 if (flags & EIO_READDIR_STAT_ORDER
1149 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))
1150 eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */
1151 else if (flags & EIO_READDIR_DIRS_FIRST)
1152 {
1153 /* in this case, all is known, and we just put dirs first and sort them */
1154 eio_dirent *ent = dents + dentoffs;
1155 eio_dirent *dir = dents;
1156
1157 /* now move dirs to the front, and non-dirs to the back */
1158 /* by walking from both sides and swapping if necessary */
1159 while (ent > dir)
1160 {
1161 if (dir->type == DT_DIR)
1162 ++dir;
1163 else
1164 {
1165 --ent;
1166
1167 if (ent->type == DT_DIR)
1168 {
1169 eio_dirent tmp = *dir;
1170 *dir = *ent;
1171 *ent = tmp;
1172
1173 ++dir;
1174 }
1175 }
1176 }
1177
1178 /* now sort the dirs only */
1179 eio_dent_sort (dents, dir - dents);
1180 }
1181
1182 /* only provide the names array unless DENTS is specified */
1183 if (!(flags & EIO_READDIR_DENTS))
1184 {
1185 X_LOCK (wrklock);
1186 assert (!dents);
1187 req->ptr1 = 0;
1188 req->ptr2 = names;
1189 X_UNLOCK (wrklock);
1190 }
1191
987 break; 1192 break;
1193 }
988 1194
1195 /* now add the entry to our list(s) */
989 name = entp->d_name; 1196 name = entp->d_name;
990 1197
1198 /* skip . and .. entries */
991 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1199 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
992 { 1200 {
993 int len = strlen (name) + 1; 1201 int len = D_NAMLEN (entp) + 1;
994 1202
995 res++; 1203 while (expect_false (namesoffs + len > namesalloc))
996
997 while (memofs + len > memlen)
998 { 1204 {
999 memlen *= 2; 1205 namesalloc *= 2;
1000 X_LOCK (wrklock); 1206 X_LOCK (wrklock);
1001 req->ptr2 = names = realloc (names, memlen); 1207 req->ptr1 = names = realloc (names, namesalloc);
1002 X_UNLOCK (wrklock); 1208 X_UNLOCK (wrklock);
1003 1209
1004 if (!names) 1210 if (!names)
1005 break; 1211 break;
1006 } 1212 }
1007 1213
1008 memcpy (names + memofs, name, len); 1214 memcpy (names + namesoffs, name, len);
1215
1216 if (dents)
1217 {
1218 struct eio_dirent *ent;
1219
1220 if (expect_false (dentoffs == dentalloc))
1221 {
1222 dentalloc *= 2;
1223 X_LOCK (wrklock);
1224 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1225 X_UNLOCK (wrklock);
1226
1227 if (!dents)
1228 break;
1229 }
1230
1231 ent = dents + dentoffs;
1232
1233 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */
1234 ent->namelen = len - 1;
1235 ent->inode = D_INO (entp);
1236
1237 switch (D_TYPE (entp))
1238 {
1239 default:
1240 ent->type = EIO_DT_UNKNOWN;
1241 flags |= EIO_READDIR_FOUND_UNKNOWN;
1242 break;
1243
1244 #ifdef DT_FIFO
1245 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1246 #endif
1247 #ifdef DT_CHR
1248 case DT_CHR: ent->type = EIO_DT_CHR; break;
1249 #endif
1250 #ifdef DT_MPC
1251 case DT_MPC: ent->type = EIO_DT_MPC; break;
1252 #endif
1253 #ifdef DT_DIR
1254 case DT_DIR: ent->type = EIO_DT_DIR; break;
1255 #endif
1256 #ifdef DT_NAM
1257 case DT_NAM: ent->type = EIO_DT_NAM; break;
1258 #endif
1259 #ifdef DT_BLK
1260 case DT_BLK: ent->type = EIO_DT_BLK; break;
1261 #endif
1262 #ifdef DT_MPB
1263 case DT_MPB: ent->type = EIO_DT_MPB; break;
1264 #endif
1265 #ifdef DT_REG
1266 case DT_REG: ent->type = EIO_DT_REG; break;
1267 #endif
1268 #ifdef DT_NWK
1269 case DT_NWK: ent->type = EIO_DT_NWK; break;
1270 #endif
1271 #ifdef DT_CMP
1272 case DT_CMP: ent->type = EIO_DT_CMP; break;
1273 #endif
1274 #ifdef DT_LNK
1275 case DT_LNK: ent->type = EIO_DT_LNK; break;
1276 #endif
1277 #ifdef DT_SOCK
1278 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1279 #endif
1280 #ifdef DT_DOOR
1281 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1282 #endif
1283 #ifdef DT_WHT
1284 case DT_WHT: ent->type = EIO_DT_WHT; break;
1285 #endif
1286 }
1287
1288 ent->score = 0;
1289
1290 if (flags & EIO_READDIR_DIRS_FIRST)
1291 {
1292 if (ent->type == EIO_DT_UNKNOWN)
1293 {
1294 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1295 ent->score = 98;
1296 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1297 ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */
1298 }
1299 else if (ent->type == EIO_DT_DIR)
1300 ent->score = 100;
1301 }
1302 }
1303
1009 memofs += len; 1304 namesoffs += len;
1305 ++dentoffs;
1306 }
1307
1308 if (EIO_CANCELLED (req))
1309 {
1310 errno = ECANCELED;
1311 break;
1010 } 1312 }
1011 } 1313 }
1012
1013 if (errno)
1014 res = -1;
1015
1016 req->result = res;
1017} 1314}
1018 1315
1019#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1316#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1020# undef msync 1317# undef msync
1021# define msync(a,b,c) ENOSYS 1318# define msync(a,b,c) ((errno = ENOSYS), -1)
1022#endif 1319#endif
1023 1320
1024int 1321int
1025eio__mtouch (void *mem, size_t len, int flags) 1322eio__mtouch (void *mem, size_t len, int flags)
1026{ 1323{
1441eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1738eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1442{ 1739{
1443 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1740 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1444} 1741}
1445 1742
1446eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1743eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1447{ 1744{
1448 return eio__1path (EIO_READDIR, path, pri, cb, data); 1745 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1449} 1746}
1450 1747
1451eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1748eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1452{ 1749{
1453 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 1750 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1542/* misc garbage */ 1839/* misc garbage */
1543 1840
1544ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 1841ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1545{ 1842{
1546 etp_worker wrk; 1843 etp_worker wrk;
1844 ssize_t ret;
1547 1845
1548 wrk.dbuf = 0; 1846 wrk.dbuf = 0;
1549 1847
1550 eio__sendfile (ofd, ifd, offset, count, &wrk); 1848 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1551 1849
1552 if (wrk.dbuf) 1850 if (wrk.dbuf)
1553 free (wrk.dbuf); 1851 free (wrk.dbuf);
1554}
1555 1852
1853 return ret;
1854}
1855

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines