1 | /* |
1 | /* |
2 | * libeio implementation |
2 | * libeio implementation |
3 | * |
3 | * |
4 | * Copyright (c) 2007,2008,2009,2010 Marc Alexander Lehmann <libeio@schmorp.de> |
4 | * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de> |
5 | * All rights reserved. |
5 | * All rights reserved. |
6 | * |
6 | * |
7 | * Redistribution and use in source and binary forms, with or without modifica- |
7 | * Redistribution and use in source and binary forms, with or without modifica- |
8 | * tion, are permitted provided that the following conditions are met: |
8 | * tion, are permitted provided that the following conditions are met: |
9 | * |
9 | * |
… | |
… | |
35 | * and other provisions required by the GPL. If you do not delete the |
35 | * and other provisions required by the GPL. If you do not delete the |
36 | * provisions above, a recipient may use your version of this file under |
36 | * provisions above, a recipient may use your version of this file under |
37 | * either the BSD or the GPL. |
37 | * either the BSD or the GPL. |
38 | */ |
38 | */ |
39 | |
39 | |
|
|
40 | #ifndef _WIN32 |
|
|
41 | # include "config.h" |
|
|
42 | #endif |
|
|
43 | |
40 | #include "eio.h" |
44 | #include "eio.h" |
41 | |
45 | |
42 | #ifdef EIO_STACKSIZE |
46 | #ifdef EIO_STACKSIZE |
43 | # define XTHREAD_STACKSIZE EIO_STACKSIZE |
47 | # define XTHREAD_STACKSIZE EIO_STACKSIZE |
44 | #endif |
48 | #endif |
… | |
… | |
49 | #include <stdlib.h> |
53 | #include <stdlib.h> |
50 | #include <string.h> |
54 | #include <string.h> |
51 | #include <errno.h> |
55 | #include <errno.h> |
52 | #include <sys/types.h> |
56 | #include <sys/types.h> |
53 | #include <sys/stat.h> |
57 | #include <sys/stat.h> |
|
|
58 | #include <sys/statvfs.h> |
54 | #include <limits.h> |
59 | #include <limits.h> |
55 | #include <fcntl.h> |
60 | #include <fcntl.h> |
56 | #include <assert.h> |
61 | #include <assert.h> |
57 | |
62 | |
|
|
63 | /* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ |
|
|
64 | /* intptr_t only comes form stdint.h, says idiot openbsd coder */ |
|
|
65 | #if HAVE_STDINT_H |
|
|
66 | # include <stdint.h> |
|
|
67 | #endif |
|
|
68 | |
58 | #ifndef EIO_FINISH |
69 | #ifndef EIO_FINISH |
59 | # define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
70 | # define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
60 | #endif |
71 | #endif |
61 | |
72 | |
62 | #ifndef EIO_DESTROY |
73 | #ifndef EIO_DESTROY |
… | |
… | |
70 | #ifdef _WIN32 |
81 | #ifdef _WIN32 |
71 | |
82 | |
72 | /*doh*/ |
83 | /*doh*/ |
73 | #else |
84 | #else |
74 | |
85 | |
75 | # include "config.h" |
|
|
76 | # include <sys/time.h> |
86 | # include <sys/time.h> |
77 | # include <sys/select.h> |
87 | # include <sys/select.h> |
78 | # include <sys/mman.h> |
|
|
79 | # include <unistd.h> |
88 | # include <unistd.h> |
80 | # include <utime.h> |
89 | # include <utime.h> |
81 | # include <signal.h> |
90 | # include <signal.h> |
82 | # include <dirent.h> |
91 | # include <dirent.h> |
83 | |
92 | |
|
|
93 | #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES |
|
|
94 | # include <sys/mman.h> |
|
|
95 | #endif |
|
|
96 | |
84 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
97 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
85 | # if __freebsd || defined __NetBSD__ || defined __OpenBSD__ |
98 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
86 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
99 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
87 | # define D_INO(de) (de)->d_fileno |
100 | # define D_INO(de) (de)->d_fileno |
88 | # define D_NAMLEN(de) (de)->d_namlen |
101 | # define D_NAMLEN(de) (de)->d_namlen |
89 | # elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
102 | # elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
90 | # define D_INO(de) (de)->d_ino |
103 | # define D_INO(de) (de)->d_ino |
… | |
… | |
106 | #endif |
119 | #endif |
107 | |
120 | |
108 | #if HAVE_SENDFILE |
121 | #if HAVE_SENDFILE |
109 | # if __linux |
122 | # if __linux |
110 | # include <sys/sendfile.h> |
123 | # include <sys/sendfile.h> |
111 | # elif __freebsd || defined __APPLE__ |
124 | # elif __FreeBSD__ || defined __APPLE__ |
112 | # include <sys/socket.h> |
125 | # include <sys/socket.h> |
113 | # include <sys/uio.h> |
126 | # include <sys/uio.h> |
114 | # elif __hpux |
127 | # elif __hpux |
115 | # include <sys/socket.h> |
128 | # include <sys/socket.h> |
116 | # elif __solaris |
129 | # elif __solaris |
… | |
… | |
128 | #endif |
141 | #endif |
129 | #ifndef D_NAMLEN |
142 | #ifndef D_NAMLEN |
130 | # define D_NAMLEN(de) strlen ((de)->d_name) |
143 | # define D_NAMLEN(de) strlen ((de)->d_name) |
131 | #endif |
144 | #endif |
132 | |
145 | |
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 */ |
146 | /* used for struct dirent, AIX doesn't provide it */ |
137 | #ifndef NAME_MAX |
147 | #ifndef NAME_MAX |
138 | # define NAME_MAX 4096 |
148 | # define NAME_MAX 4096 |
|
|
149 | #endif |
|
|
150 | |
|
|
151 | /* used for readlink etc. */ |
|
|
152 | #ifndef PATH_MAX |
|
|
153 | # define PATH_MAX 4096 |
139 | #endif |
154 | #endif |
140 | |
155 | |
141 | /* buffer size for various temporary buffers */ |
156 | /* buffer size for various temporary buffers */ |
142 | #define EIO_BUFSIZE 65536 |
157 | #define EIO_BUFSIZE 65536 |
143 | |
158 | |
… | |
… | |
214 | static unsigned int max_poll_reqs; /* reslock */ |
229 | static unsigned int max_poll_reqs; /* reslock */ |
215 | |
230 | |
216 | static volatile unsigned int nreqs; /* reqlock */ |
231 | static volatile unsigned int nreqs; /* reqlock */ |
217 | static volatile unsigned int nready; /* reqlock */ |
232 | static volatile unsigned int nready; /* reqlock */ |
218 | static volatile unsigned int npending; /* reqlock */ |
233 | static volatile unsigned int npending; /* reqlock */ |
219 | static volatile unsigned int max_idle = 4; |
234 | static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ |
|
|
235 | static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ |
220 | |
236 | |
221 | static mutex_t wrklock = X_MUTEX_INIT; |
237 | static xmutex_t wrklock; |
222 | static mutex_t reslock = X_MUTEX_INIT; |
238 | static xmutex_t reslock; |
223 | static mutex_t reqlock = X_MUTEX_INIT; |
239 | static xmutex_t reqlock; |
224 | static cond_t reqwait = X_COND_INIT; |
240 | static xcond_t reqwait; |
225 | |
241 | |
226 | #if !HAVE_PREADWRITE |
242 | #if !HAVE_PREADWRITE |
227 | /* |
243 | /* |
228 | * make our pread/pwrite emulation safe against themselves, but not against |
244 | * make our pread/pwrite emulation safe against themselves, but not against |
229 | * normal read/write by using a mutex. slows down execution a lot, |
245 | * normal read/write by using a mutex. slows down execution a lot, |
230 | * but that's your problem, not mine. |
246 | * but that's your problem, not mine. |
231 | */ |
247 | */ |
232 | static mutex_t preadwritelock = X_MUTEX_INIT; |
248 | static xmutex_t preadwritelock = X_MUTEX_INIT; |
233 | #endif |
249 | #endif |
234 | |
250 | |
235 | typedef struct etp_worker |
251 | typedef struct etp_worker |
236 | { |
252 | { |
237 | /* locked by wrklock */ |
253 | /* locked by wrklock */ |
238 | struct etp_worker *prev, *next; |
254 | struct etp_worker *prev, *next; |
239 | |
255 | |
240 | thread_t tid; |
256 | xthread_t tid; |
241 | |
257 | |
242 | /* locked by reslock, reqlock or wrklock */ |
258 | /* locked by reslock, reqlock or wrklock */ |
243 | ETP_REQ *req; /* currently processed request */ |
259 | ETP_REQ *req; /* currently processed request */ |
244 | |
260 | |
245 | ETP_WORKER_COMMON |
261 | ETP_WORKER_COMMON |
… | |
… | |
359 | } |
375 | } |
360 | |
376 | |
361 | abort (); |
377 | abort (); |
362 | } |
378 | } |
363 | |
379 | |
|
|
380 | static void etp_thread_init (void) |
|
|
381 | { |
|
|
382 | X_MUTEX_CREATE (wrklock); |
|
|
383 | X_MUTEX_CREATE (reslock); |
|
|
384 | X_MUTEX_CREATE (reqlock); |
|
|
385 | X_COND_CREATE (reqwait); |
|
|
386 | } |
|
|
387 | |
364 | static void etp_atfork_prepare (void) |
388 | static void etp_atfork_prepare (void) |
365 | { |
389 | { |
366 | X_LOCK (wrklock); |
390 | X_LOCK (wrklock); |
367 | X_LOCK (reqlock); |
391 | X_LOCK (reqlock); |
368 | X_LOCK (reslock); |
392 | X_LOCK (reslock); |
… | |
… | |
406 | idle = 0; |
430 | idle = 0; |
407 | nreqs = 0; |
431 | nreqs = 0; |
408 | nready = 0; |
432 | nready = 0; |
409 | npending = 0; |
433 | npending = 0; |
410 | |
434 | |
411 | etp_atfork_parent (); |
435 | etp_thread_init (); |
412 | } |
436 | } |
413 | |
437 | |
414 | static void |
438 | static void |
415 | etp_once_init (void) |
439 | etp_once_init (void) |
416 | { |
440 | { |
|
|
441 | etp_thread_init (); |
417 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
442 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
418 | } |
443 | } |
419 | |
444 | |
420 | static int |
445 | static int |
421 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
446 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
… | |
… | |
612 | } |
637 | } |
613 | |
638 | |
614 | static void etp_set_max_idle (unsigned int nthreads) |
639 | static void etp_set_max_idle (unsigned int nthreads) |
615 | { |
640 | { |
616 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
641 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
617 | max_idle = nthreads <= 0 ? 1 : nthreads; |
642 | max_idle = nthreads; |
|
|
643 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
|
|
644 | } |
|
|
645 | |
|
|
646 | static void etp_set_idle_timeout (unsigned int seconds) |
|
|
647 | { |
|
|
648 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
|
|
649 | idle_timeout = seconds; |
618 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
650 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
619 | } |
651 | } |
620 | |
652 | |
621 | static void etp_set_min_parallel (unsigned int nthreads) |
653 | static void etp_set_min_parallel (unsigned int nthreads) |
622 | { |
654 | { |
… | |
… | |
750 | void eio_set_max_idle (unsigned int nthreads) |
782 | void eio_set_max_idle (unsigned int nthreads) |
751 | { |
783 | { |
752 | etp_set_max_idle (nthreads); |
784 | etp_set_max_idle (nthreads); |
753 | } |
785 | } |
754 | |
786 | |
|
|
787 | void eio_set_idle_timeout (unsigned int seconds) |
|
|
788 | { |
|
|
789 | etp_set_idle_timeout (seconds); |
|
|
790 | } |
|
|
791 | |
755 | void eio_set_min_parallel (unsigned int nthreads) |
792 | void eio_set_min_parallel (unsigned int nthreads) |
756 | { |
793 | { |
757 | etp_set_min_parallel (nthreads); |
794 | etp_set_min_parallel (nthreads); |
758 | } |
795 | } |
759 | |
796 | |
… | |
… | |
807 | |
844 | |
808 | return res; |
845 | return res; |
809 | } |
846 | } |
810 | #endif |
847 | #endif |
811 | |
848 | |
812 | #ifndef HAVE_FUTIMES |
849 | #ifndef HAVE_UTIMES |
813 | |
850 | |
814 | # undef utimes |
851 | # undef utimes |
815 | # undef futimes |
|
|
816 | # define utimes(path,times) eio__utimes (path, times) |
852 | # define utimes(path,times) eio__utimes (path, times) |
817 | # define futimes(fd,times) eio__futimes (fd, times) |
|
|
818 | |
853 | |
819 | static int |
854 | static int |
820 | eio__utimes (const char *filename, const struct timeval times[2]) |
855 | eio__utimes (const char *filename, const struct timeval times[2]) |
821 | { |
856 | { |
822 | if (times) |
857 | if (times) |
… | |
… | |
829 | return utime (filename, &buf); |
864 | return utime (filename, &buf); |
830 | } |
865 | } |
831 | else |
866 | else |
832 | return utime (filename, 0); |
867 | return utime (filename, 0); |
833 | } |
868 | } |
|
|
869 | |
|
|
870 | #endif |
|
|
871 | |
|
|
872 | #ifndef HAVE_FUTIMES |
|
|
873 | |
|
|
874 | # undef futimes |
|
|
875 | # define futimes(fd,times) eio__futimes (fd, times) |
834 | |
876 | |
835 | static int eio__futimes (int fd, const struct timeval tv[2]) |
877 | static int eio__futimes (int fd, const struct timeval tv[2]) |
836 | { |
878 | { |
837 | errno = ENOSYS; |
879 | errno = ENOSYS; |
838 | return -1; |
880 | return -1; |
… | |
… | |
909 | |
951 | |
910 | #if HAVE_SENDFILE |
952 | #if HAVE_SENDFILE |
911 | # if __linux |
953 | # if __linux |
912 | res = sendfile (ofd, ifd, &offset, count); |
954 | res = sendfile (ofd, ifd, &offset, count); |
913 | |
955 | |
914 | # elif __freebsd |
956 | # elif __FreeBSD__ |
915 | /* |
957 | /* |
916 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
958 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
917 | * wasted on making it similar to other I/O functions. |
959 | * wasted on making it similar to other I/O functions. |
918 | */ |
960 | */ |
919 | { |
961 | { |
920 | off_t sbytes; |
962 | off_t sbytes; |
921 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
963 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
922 | |
964 | |
923 | if (res < 0 && sbytes) |
965 | #if 0 /* according to the manpage, this is correct, but broken behaviour */ |
924 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
966 | /* freebsd' sendfile will return 0 on success */ |
|
|
967 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
|
|
968 | /* not on e.g. EIO or EPIPE - sounds broken */ |
|
|
969 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
|
|
970 | res = sbytes; |
|
|
971 | #endif |
|
|
972 | |
|
|
973 | /* according to source inspection, this is correct, and useful behaviour */ |
|
|
974 | if (sbytes) |
925 | res = sbytes; |
975 | res = sbytes; |
926 | } |
976 | } |
927 | |
977 | |
928 | # elif defined (__APPLE__) |
978 | # elif defined (__APPLE__) |
929 | |
979 | |
930 | { |
980 | { |
931 | off_t sbytes = count; |
981 | off_t sbytes = count; |
932 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
982 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
933 | |
983 | |
934 | if (res < 0 && errno == EAGAIN && sbytes) |
984 | /* according to the manpage, sbytes is always valid */ |
|
|
985 | if (sbytes) |
935 | res = sbytes; |
986 | res = sbytes; |
936 | } |
987 | } |
937 | |
988 | |
938 | # elif __hpux |
989 | # elif __hpux |
939 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
990 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
… | |
… | |
970 | errno = ENOSYS; |
1021 | errno = ENOSYS; |
971 | #endif |
1022 | #endif |
972 | |
1023 | |
973 | if (res < 0 |
1024 | if (res < 0 |
974 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
1025 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
|
|
1026 | /* BSDs */ |
|
|
1027 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
|
|
1028 | || errno == ENOTSUP |
|
|
1029 | #endif |
975 | || errno == ENOTSUP || errno == EOPNOTSUPP /* BSDs */ |
1030 | || errno == EOPNOTSUPP /* BSDs */ |
976 | #if __solaris |
1031 | #if __solaris |
977 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
1032 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
978 | #endif |
1033 | #endif |
979 | ) |
1034 | ) |
980 | ) |
1035 | ) |
… | |
… | |
1014 | } |
1069 | } |
1015 | |
1070 | |
1016 | static signed char |
1071 | static signed char |
1017 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
1072 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
1018 | { |
1073 | { |
1019 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
1074 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
1020 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; |
1075 | : a->inode < b->inode ? -1 |
|
|
1076 | : a->inode > b->inode ? 1 |
|
|
1077 | : 0; |
1021 | } |
1078 | } |
1022 | |
1079 | |
1023 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
1080 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
1024 | |
1081 | |
1025 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
1082 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
… | |
… | |
1031 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
1088 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
1032 | unsigned char *bit = bits; |
1089 | unsigned char *bit = bits; |
1033 | |
1090 | |
1034 | assert (CHAR_BIT == 8); |
1091 | assert (CHAR_BIT == 8); |
1035 | assert (sizeof (eio_dirent) * 8 < 256); |
1092 | assert (sizeof (eio_dirent) * 8 < 256); |
1036 | assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ |
1093 | assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ |
1037 | assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ |
1094 | assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */ |
1038 | |
1095 | |
1039 | if (size <= EIO_SORT_FAST) |
1096 | if (size <= EIO_SORT_FAST) |
1040 | return; |
1097 | return; |
1041 | |
1098 | |
1042 | /* first prepare an array of bits to test in our radix sort */ |
1099 | /* first prepare an array of bits to test in our radix sort */ |
… | |
… | |
1197 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1254 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1198 | |
1255 | |
1199 | X_LOCK (wrklock); |
1256 | X_LOCK (wrklock); |
1200 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1257 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1201 | self->dirp = dirp = opendir (req->ptr1); |
1258 | self->dirp = dirp = opendir (req->ptr1); |
|
|
1259 | |
1202 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1260 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1203 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1261 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1204 | req->ptr2 = names = malloc (namesalloc); |
1262 | req->ptr2 = names = malloc (namesalloc); |
1205 | X_UNLOCK (wrklock); |
1263 | X_UNLOCK (wrklock); |
1206 | |
1264 | |
… | |
… | |
1218 | /* sort etc. */ |
1276 | /* sort etc. */ |
1219 | req->int1 = flags; |
1277 | req->int1 = flags; |
1220 | req->result = dentoffs; |
1278 | req->result = dentoffs; |
1221 | |
1279 | |
1222 | if (flags & EIO_READDIR_STAT_ORDER) |
1280 | if (flags & EIO_READDIR_STAT_ORDER) |
1223 | eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ |
1281 | eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); |
1224 | else if (flags & EIO_READDIR_DIRS_FIRST) |
1282 | else if (flags & EIO_READDIR_DIRS_FIRST) |
1225 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
1283 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
1226 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
1284 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
1227 | else |
1285 | else |
1228 | { |
1286 | { |
… | |
… | |
1230 | eio_dirent *oth = dents + dentoffs; |
1288 | eio_dirent *oth = dents + dentoffs; |
1231 | eio_dirent *dir = dents; |
1289 | eio_dirent *dir = dents; |
1232 | |
1290 | |
1233 | /* now partition dirs to the front, and non-dirs to the back */ |
1291 | /* now partition dirs to the front, and non-dirs to the back */ |
1234 | /* by walking from both sides and swapping if necessary */ |
1292 | /* by walking from both sides and swapping if necessary */ |
1235 | /* also clear score, so it doesn't influence sorting */ |
|
|
1236 | while (oth > dir) |
1293 | while (oth > dir) |
1237 | { |
1294 | { |
1238 | if (dir->type == EIO_DT_DIR) |
1295 | if (dir->type == EIO_DT_DIR) |
1239 | ++dir; |
1296 | ++dir; |
1240 | else if ((--oth)->type == EIO_DT_DIR) |
1297 | else if ((--oth)->type == EIO_DT_DIR) |
… | |
… | |
1243 | |
1300 | |
1244 | ++dir; |
1301 | ++dir; |
1245 | } |
1302 | } |
1246 | } |
1303 | } |
1247 | |
1304 | |
1248 | /* now sort the dirs only */ |
1305 | /* now sort the dirs only (dirs all have the same score) */ |
1249 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1306 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1250 | } |
1307 | } |
1251 | |
1308 | |
1252 | break; |
1309 | break; |
1253 | } |
1310 | } |
… | |
… | |
1373 | break; |
1430 | break; |
1374 | } |
1431 | } |
1375 | } |
1432 | } |
1376 | } |
1433 | } |
1377 | |
1434 | |
|
|
1435 | #ifdef PAGESIZE |
|
|
1436 | # define eio_pagesize() PAGESIZE |
|
|
1437 | #else |
|
|
1438 | static intptr_t |
|
|
1439 | eio_pagesize (void) |
|
|
1440 | { |
|
|
1441 | static intptr_t page; |
|
|
1442 | |
|
|
1443 | if (!page) |
|
|
1444 | page = sysconf (_SC_PAGESIZE); |
|
|
1445 | |
|
|
1446 | return page; |
|
|
1447 | } |
|
|
1448 | #endif |
|
|
1449 | |
|
|
1450 | static void |
|
|
1451 | eio_page_align (void **addr, size_t *length) |
|
|
1452 | { |
|
|
1453 | intptr_t mask = eio_pagesize () - 1; |
|
|
1454 | |
|
|
1455 | /* round down addr */ |
|
|
1456 | intptr_t adj = mask & (intptr_t)*addr; |
|
|
1457 | |
|
|
1458 | *addr = (void *)((intptr_t)*addr - adj); |
|
|
1459 | *length += adj; |
|
|
1460 | |
|
|
1461 | /* round up length */ |
|
|
1462 | *length = (*length + mask) & ~mask; |
|
|
1463 | } |
|
|
1464 | |
|
|
1465 | #if !_POSIX_MEMLOCK |
|
|
1466 | # define eio__mlockall(a) ((errno = ENOSYS), -1) |
|
|
1467 | #else |
|
|
1468 | |
|
|
1469 | static int |
|
|
1470 | eio__mlockall (int flags) |
|
|
1471 | { |
|
|
1472 | #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 |
|
|
1473 | extern int mallopt (int, int); |
|
|
1474 | mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ |
|
|
1475 | #endif |
|
|
1476 | |
|
|
1477 | if (EIO_MCL_CURRENT != MCL_CURRENT |
|
|
1478 | || EIO_MCL_FUTURE != MCL_FUTURE) |
|
|
1479 | { |
|
|
1480 | flags = 0 |
|
|
1481 | | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0) |
|
|
1482 | | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0); |
|
|
1483 | } |
|
|
1484 | |
|
|
1485 | return mlockall (flags); |
|
|
1486 | } |
|
|
1487 | #endif |
|
|
1488 | |
|
|
1489 | #if !_POSIX_MEMLOCK_RANGE |
|
|
1490 | # define eio__mlock(a,b) ((errno = ENOSYS), -1) |
|
|
1491 | #else |
|
|
1492 | |
|
|
1493 | static int |
|
|
1494 | eio__mlock (void *addr, size_t length) |
|
|
1495 | { |
|
|
1496 | eio_page_align (&addr, &length); |
|
|
1497 | |
|
|
1498 | return mlock (addr, length); |
|
|
1499 | } |
|
|
1500 | |
|
|
1501 | #endif |
|
|
1502 | |
1378 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1503 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1379 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
1504 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
1380 | #else |
1505 | #else |
1381 | |
1506 | |
1382 | int |
1507 | int |
1383 | eio__msync (void *mem, size_t len, int flags) |
1508 | eio__msync (void *mem, size_t len, int flags) |
1384 | { |
1509 | { |
|
|
1510 | eio_page_align (&mem, &len); |
|
|
1511 | |
1385 | if (EIO_MS_ASYNC != MS_SYNC |
1512 | if (EIO_MS_ASYNC != MS_SYNC |
1386 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
1513 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
1387 | || EIO_MS_SYNC != MS_SYNC) |
1514 | || EIO_MS_SYNC != MS_SYNC) |
1388 | { |
1515 | { |
1389 | flags = 0 |
1516 | flags = 0 |
… | |
… | |
1398 | #endif |
1525 | #endif |
1399 | |
1526 | |
1400 | int |
1527 | int |
1401 | eio__mtouch (void *mem, size_t len, int flags) |
1528 | eio__mtouch (void *mem, size_t len, int flags) |
1402 | { |
1529 | { |
|
|
1530 | eio_page_align (&mem, &len); |
|
|
1531 | |
|
|
1532 | { |
1403 | intptr_t addr = (intptr_t)mem; |
1533 | intptr_t addr = (intptr_t)mem; |
1404 | intptr_t end = addr + len; |
1534 | intptr_t end = addr + len; |
1405 | #ifdef PAGESIZE |
1535 | intptr_t page = eio_pagesize (); |
1406 | const intptr_t page = PAGESIZE; |
|
|
1407 | #else |
|
|
1408 | static intptr_t page; |
|
|
1409 | |
1536 | |
1410 | if (!page) |
|
|
1411 | page = sysconf (_SC_PAGESIZE); |
|
|
1412 | #endif |
|
|
1413 | |
|
|
1414 | /* round down to start of page, although this is probably useless */ |
|
|
1415 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
|
|
1416 | |
|
|
1417 | if (addr < end) |
1537 | if (addr < end) |
1418 | if (flags & EIO_MT_MODIFY) /* modify */ |
1538 | if (flags & EIO_MT_MODIFY) /* modify */ |
1419 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1539 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1420 | else |
1540 | else |
1421 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1541 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
|
|
1542 | } |
1422 | |
1543 | |
1423 | return 0; |
1544 | return 0; |
1424 | } |
1545 | } |
1425 | |
1546 | |
1426 | /*****************************************************************************/ |
1547 | /*****************************************************************************/ |
… | |
… | |
1460 | if (req) |
1581 | if (req) |
1461 | break; |
1582 | break; |
1462 | |
1583 | |
1463 | ++idle; |
1584 | ++idle; |
1464 | |
1585 | |
1465 | ts.tv_sec = time (0) + IDLE_TIMEOUT; |
1586 | ts.tv_sec = time (0) + idle_timeout; |
1466 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1587 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1467 | { |
1588 | { |
1468 | if (idle > max_idle) |
1589 | if (idle > max_idle) |
1469 | { |
1590 | { |
1470 | --idle; |
1591 | --idle; |
… | |
… | |
1549 | return 0; \ |
1670 | return 0; \ |
1550 | } |
1671 | } |
1551 | |
1672 | |
1552 | static void eio_execute (etp_worker *self, eio_req *req) |
1673 | static void eio_execute (etp_worker *self, eio_req *req) |
1553 | { |
1674 | { |
1554 | errno = 0; |
|
|
1555 | |
|
|
1556 | switch (req->type) |
1675 | switch (req->type) |
1557 | { |
1676 | { |
1558 | case EIO_READ: ALLOC (req->size); |
1677 | case EIO_READ: ALLOC (req->size); |
1559 | req->result = req->offs >= 0 |
1678 | req->result = req->offs >= 0 |
1560 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
1679 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
… | |
… | |
1571 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1690 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1572 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1691 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1573 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1692 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1574 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1693 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1575 | |
1694 | |
|
|
1695 | case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1696 | req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1697 | case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1698 | req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1699 | |
1576 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1700 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1577 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1701 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1578 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1702 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1579 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1703 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1580 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
1704 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
… | |
… | |
1587 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1711 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1588 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1712 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1589 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1713 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1590 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1714 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1591 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1715 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1592 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
1716 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; |
1593 | |
1717 | |
1594 | case EIO_READLINK: ALLOC (NAME_MAX); |
1718 | case EIO_READLINK: ALLOC (PATH_MAX); |
1595 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1719 | req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; |
1596 | |
1720 | |
1597 | case EIO_SYNC: req->result = 0; sync (); break; |
1721 | case EIO_SYNC: req->result = 0; sync (); break; |
1598 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1722 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1599 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1723 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1600 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
1724 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
1601 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1725 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
|
|
1726 | case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; |
|
|
1727 | case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; |
1602 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1728 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1603 | |
1729 | |
1604 | case EIO_READDIR: eio__scandir (req, self); break; |
1730 | case EIO_READDIR: eio__scandir (req, self); break; |
1605 | |
1731 | |
1606 | case EIO_BUSY: |
1732 | case EIO_BUSY: |
1607 | #ifdef _WIN32 |
1733 | #ifdef _WIN32 |
1608 | Sleep (req->nv1 * 1000.); |
1734 | Sleep (req->nv1 * 1e3); |
1609 | #else |
1735 | #else |
1610 | { |
1736 | { |
1611 | struct timeval tv; |
1737 | struct timeval tv; |
1612 | |
1738 | |
1613 | tv.tv_sec = req->nv1; |
1739 | tv.tv_sec = req->nv1; |
1614 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
1740 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6; |
1615 | |
1741 | |
1616 | req->result = select (0, 0, 0, 0, &tv); |
1742 | req->result = select (0, 0, 0, 0, &tv); |
1617 | } |
1743 | } |
1618 | #endif |
1744 | #endif |
1619 | break; |
1745 | break; |
… | |
… | |
1634 | times = tv; |
1760 | times = tv; |
1635 | } |
1761 | } |
1636 | else |
1762 | else |
1637 | times = 0; |
1763 | times = 0; |
1638 | |
1764 | |
1639 | |
|
|
1640 | req->result = req->type == EIO_FUTIME |
1765 | req->result = req->type == EIO_FUTIME |
1641 | ? futimes (req->int1, times) |
1766 | ? futimes (req->int1, times) |
1642 | : utimes (req->ptr1, times); |
1767 | : utimes (req->ptr1, times); |
1643 | } |
1768 | } |
1644 | break; |
1769 | break; |
… | |
… | |
1653 | case EIO_CUSTOM: |
1778 | case EIO_CUSTOM: |
1654 | ((void (*)(eio_req *))req->feed) (req); |
1779 | ((void (*)(eio_req *))req->feed) (req); |
1655 | break; |
1780 | break; |
1656 | |
1781 | |
1657 | default: |
1782 | default: |
|
|
1783 | errno = ENOSYS; |
1658 | req->result = -1; |
1784 | req->result = -1; |
1659 | break; |
1785 | break; |
1660 | } |
1786 | } |
1661 | |
1787 | |
1662 | req->errorno = errno; |
1788 | req->errorno = errno; |
… | |
… | |
1692 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1818 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1693 | { |
1819 | { |
1694 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1820 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1695 | } |
1821 | } |
1696 | |
1822 | |
|
|
1823 | eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data) |
|
|
1824 | { |
|
|
1825 | REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND; |
|
|
1826 | } |
|
|
1827 | |
|
|
1828 | eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) |
|
|
1829 | { |
|
|
1830 | REQ (EIO_MLOCKALL); req->int1 = flags; SEND; |
|
|
1831 | } |
|
|
1832 | |
1697 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
1833 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
1698 | { |
1834 | { |
1699 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
1835 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
1700 | } |
1836 | } |
1701 | |
1837 | |
… | |
… | |
1725 | } |
1861 | } |
1726 | |
1862 | |
1727 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1863 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1728 | { |
1864 | { |
1729 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
1865 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
|
|
1866 | } |
|
|
1867 | |
|
|
1868 | eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data) |
|
|
1869 | { |
|
|
1870 | REQ (EIO_FSTATVFS); req->int1 = fd; SEND; |
1730 | } |
1871 | } |
1731 | |
1872 | |
1732 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1873 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1733 | { |
1874 | { |
1734 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
1875 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
… | |
… | |
1808 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1949 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1809 | { |
1950 | { |
1810 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1951 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1811 | } |
1952 | } |
1812 | |
1953 | |
|
|
1954 | eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data) |
|
|
1955 | { |
|
|
1956 | return eio__1path (EIO_STATVFS, path, pri, cb, data); |
|
|
1957 | } |
|
|
1958 | |
1813 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1959 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1814 | { |
1960 | { |
1815 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1961 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1816 | } |
1962 | } |
1817 | |
1963 | |
… | |
… | |
1825 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1971 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1826 | } |
1972 | } |
1827 | |
1973 | |
1828 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1974 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1829 | { |
1975 | { |
1830 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
1976 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND; |
1831 | } |
1977 | } |
1832 | |
1978 | |
1833 | static eio_req * |
1979 | static eio_req * |
1834 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1980 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1835 | { |
1981 | { |