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 |
… | |
… | |
54 | #include <sys/statvfs.h> |
58 | #include <sys/statvfs.h> |
55 | #include <limits.h> |
59 | #include <limits.h> |
56 | #include <fcntl.h> |
60 | #include <fcntl.h> |
57 | #include <assert.h> |
61 | #include <assert.h> |
58 | |
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 | |
59 | #ifndef EIO_FINISH |
69 | #ifndef EIO_FINISH |
60 | # 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 |
61 | #endif |
71 | #endif |
62 | |
72 | |
63 | #ifndef EIO_DESTROY |
73 | #ifndef EIO_DESTROY |
… | |
… | |
71 | #ifdef _WIN32 |
81 | #ifdef _WIN32 |
72 | |
82 | |
73 | /*doh*/ |
83 | /*doh*/ |
74 | #else |
84 | #else |
75 | |
85 | |
76 | # include "config.h" |
|
|
77 | # include <sys/time.h> |
86 | # include <sys/time.h> |
78 | # include <sys/select.h> |
87 | # include <sys/select.h> |
79 | # include <sys/mman.h> |
|
|
80 | # include <unistd.h> |
88 | # include <unistd.h> |
81 | # include <utime.h> |
89 | # include <utime.h> |
82 | # include <signal.h> |
90 | # include <signal.h> |
83 | # include <dirent.h> |
91 | # include <dirent.h> |
|
|
92 | |
|
|
93 | #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES |
|
|
94 | # include <sys/mman.h> |
|
|
95 | #endif |
84 | |
96 | |
85 | /* 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 */ |
86 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
98 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
87 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
99 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
88 | # define D_INO(de) (de)->d_fileno |
100 | # define D_INO(de) (de)->d_fileno |
… | |
… | |
129 | #endif |
141 | #endif |
130 | #ifndef D_NAMLEN |
142 | #ifndef D_NAMLEN |
131 | # define D_NAMLEN(de) strlen ((de)->d_name) |
143 | # define D_NAMLEN(de) strlen ((de)->d_name) |
132 | #endif |
144 | #endif |
133 | |
145 | |
134 | /* number of seconds after which an idle threads exit */ |
|
|
135 | #define IDLE_TIMEOUT 10 |
|
|
136 | |
|
|
137 | /* used for struct dirent, AIX doesn't provide it */ |
146 | /* used for struct dirent, AIX doesn't provide it */ |
138 | #ifndef NAME_MAX |
147 | #ifndef NAME_MAX |
139 | # 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 |
140 | #endif |
154 | #endif |
141 | |
155 | |
142 | /* buffer size for various temporary buffers */ |
156 | /* buffer size for various temporary buffers */ |
143 | #define EIO_BUFSIZE 65536 |
157 | #define EIO_BUFSIZE 65536 |
144 | |
158 | |
… | |
… | |
215 | static unsigned int max_poll_reqs; /* reslock */ |
229 | static unsigned int max_poll_reqs; /* reslock */ |
216 | |
230 | |
217 | static volatile unsigned int nreqs; /* reqlock */ |
231 | static volatile unsigned int nreqs; /* reqlock */ |
218 | static volatile unsigned int nready; /* reqlock */ |
232 | static volatile unsigned int nready; /* reqlock */ |
219 | static volatile unsigned int npending; /* reqlock */ |
233 | static volatile unsigned int npending; /* reqlock */ |
220 | 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 */ |
221 | |
236 | |
222 | static mutex_t wrklock = X_MUTEX_INIT; |
237 | static xmutex_t wrklock; |
223 | static mutex_t reslock = X_MUTEX_INIT; |
238 | static xmutex_t reslock; |
224 | static mutex_t reqlock = X_MUTEX_INIT; |
239 | static xmutex_t reqlock; |
225 | static cond_t reqwait = X_COND_INIT; |
240 | static xcond_t reqwait; |
226 | |
241 | |
227 | #if !HAVE_PREADWRITE |
242 | #if !HAVE_PREADWRITE |
228 | /* |
243 | /* |
229 | * make our pread/pwrite emulation safe against themselves, but not against |
244 | * make our pread/pwrite emulation safe against themselves, but not against |
230 | * 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, |
231 | * but that's your problem, not mine. |
246 | * but that's your problem, not mine. |
232 | */ |
247 | */ |
233 | static mutex_t preadwritelock = X_MUTEX_INIT; |
248 | static xmutex_t preadwritelock = X_MUTEX_INIT; |
234 | #endif |
249 | #endif |
235 | |
250 | |
236 | typedef struct etp_worker |
251 | typedef struct etp_worker |
237 | { |
252 | { |
238 | /* locked by wrklock */ |
253 | /* locked by wrklock */ |
239 | struct etp_worker *prev, *next; |
254 | struct etp_worker *prev, *next; |
240 | |
255 | |
241 | thread_t tid; |
256 | xthread_t tid; |
242 | |
257 | |
243 | /* locked by reslock, reqlock or wrklock */ |
258 | /* locked by reslock, reqlock or wrklock */ |
244 | ETP_REQ *req; /* currently processed request */ |
259 | ETP_REQ *req; /* currently processed request */ |
245 | |
260 | |
246 | ETP_WORKER_COMMON |
261 | ETP_WORKER_COMMON |
… | |
… | |
360 | } |
375 | } |
361 | |
376 | |
362 | abort (); |
377 | abort (); |
363 | } |
378 | } |
364 | |
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 | |
365 | static void etp_atfork_prepare (void) |
388 | static void etp_atfork_prepare (void) |
366 | { |
389 | { |
367 | X_LOCK (wrklock); |
390 | X_LOCK (wrklock); |
368 | X_LOCK (reqlock); |
391 | X_LOCK (reqlock); |
369 | X_LOCK (reslock); |
392 | X_LOCK (reslock); |
… | |
… | |
407 | idle = 0; |
430 | idle = 0; |
408 | nreqs = 0; |
431 | nreqs = 0; |
409 | nready = 0; |
432 | nready = 0; |
410 | npending = 0; |
433 | npending = 0; |
411 | |
434 | |
412 | etp_atfork_parent (); |
435 | etp_thread_init (); |
413 | } |
436 | } |
414 | |
437 | |
415 | static void |
438 | static void |
416 | etp_once_init (void) |
439 | etp_once_init (void) |
417 | { |
440 | { |
|
|
441 | etp_thread_init (); |
418 | 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); |
419 | } |
443 | } |
420 | |
444 | |
421 | static int |
445 | static int |
422 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
446 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
… | |
… | |
613 | } |
637 | } |
614 | |
638 | |
615 | static void etp_set_max_idle (unsigned int nthreads) |
639 | static void etp_set_max_idle (unsigned int nthreads) |
616 | { |
640 | { |
617 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
641 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
618 | 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; |
619 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
650 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
620 | } |
651 | } |
621 | |
652 | |
622 | static void etp_set_min_parallel (unsigned int nthreads) |
653 | static void etp_set_min_parallel (unsigned int nthreads) |
623 | { |
654 | { |
… | |
… | |
751 | void eio_set_max_idle (unsigned int nthreads) |
782 | void eio_set_max_idle (unsigned int nthreads) |
752 | { |
783 | { |
753 | etp_set_max_idle (nthreads); |
784 | etp_set_max_idle (nthreads); |
754 | } |
785 | } |
755 | |
786 | |
|
|
787 | void eio_set_idle_timeout (unsigned int seconds) |
|
|
788 | { |
|
|
789 | etp_set_idle_timeout (seconds); |
|
|
790 | } |
|
|
791 | |
756 | void eio_set_min_parallel (unsigned int nthreads) |
792 | void eio_set_min_parallel (unsigned int nthreads) |
757 | { |
793 | { |
758 | etp_set_min_parallel (nthreads); |
794 | etp_set_min_parallel (nthreads); |
759 | } |
795 | } |
760 | |
796 | |
… | |
… | |
808 | |
844 | |
809 | return res; |
845 | return res; |
810 | } |
846 | } |
811 | #endif |
847 | #endif |
812 | |
848 | |
813 | #ifndef HAVE_FUTIMES |
849 | #ifndef HAVE_UTIMES |
814 | |
850 | |
815 | # undef utimes |
851 | # undef utimes |
816 | # undef futimes |
|
|
817 | # define utimes(path,times) eio__utimes (path, times) |
852 | # define utimes(path,times) eio__utimes (path, times) |
818 | # define futimes(fd,times) eio__futimes (fd, times) |
|
|
819 | |
853 | |
820 | static int |
854 | static int |
821 | eio__utimes (const char *filename, const struct timeval times[2]) |
855 | eio__utimes (const char *filename, const struct timeval times[2]) |
822 | { |
856 | { |
823 | if (times) |
857 | if (times) |
… | |
… | |
830 | return utime (filename, &buf); |
864 | return utime (filename, &buf); |
831 | } |
865 | } |
832 | else |
866 | else |
833 | return utime (filename, 0); |
867 | return utime (filename, 0); |
834 | } |
868 | } |
|
|
869 | |
|
|
870 | #endif |
|
|
871 | |
|
|
872 | #ifndef HAVE_FUTIMES |
|
|
873 | |
|
|
874 | # undef futimes |
|
|
875 | # define futimes(fd,times) eio__futimes (fd, times) |
835 | |
876 | |
836 | static int eio__futimes (int fd, const struct timeval tv[2]) |
877 | static int eio__futimes (int fd, const struct timeval tv[2]) |
837 | { |
878 | { |
838 | errno = ENOSYS; |
879 | errno = ENOSYS; |
839 | return -1; |
880 | return -1; |
… | |
… | |
901 | |
942 | |
902 | /* sendfile always needs emulation */ |
943 | /* sendfile always needs emulation */ |
903 | static ssize_t |
944 | static ssize_t |
904 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
945 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
905 | { |
946 | { |
|
|
947 | ssize_t written = 0; |
906 | ssize_t res; |
948 | ssize_t res; |
907 | |
949 | |
908 | if (!count) |
950 | if (!count) |
909 | return 0; |
951 | return 0; |
910 | |
952 | |
|
|
953 | for (;;) |
|
|
954 | { |
911 | #if HAVE_SENDFILE |
955 | #if HAVE_SENDFILE |
912 | # if __linux |
956 | # if __linux |
|
|
957 | off_t soffset = offset; |
913 | res = sendfile (ofd, ifd, &offset, count); |
958 | res = sendfile (ofd, ifd, &soffset, count); |
914 | |
959 | |
915 | # elif __FreeBSD__ |
960 | # elif __FreeBSD__ |
916 | /* |
961 | /* |
917 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
962 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
918 | * wasted on making it similar to other I/O functions. |
963 | * wasted on making it similar to other I/O functions. |
919 | */ |
964 | */ |
920 | { |
|
|
921 | off_t sbytes; |
965 | off_t sbytes; |
922 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
966 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
923 | |
967 | |
|
|
968 | #if 0 /* according to the manpage, this is correct, but broken behaviour */ |
924 | /* freebsd' sendfile will return 0 on success */ |
969 | /* freebsd' sendfile will return 0 on success */ |
925 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
970 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
926 | /* not on e.g. EIO or EPIPE - sounds broken */ |
971 | /* not on e.g. EIO or EPIPE - sounds broken */ |
927 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
972 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
928 | res = sbytes; |
973 | res = sbytes; |
929 | } |
974 | #endif |
|
|
975 | |
|
|
976 | /* according to source inspection, this is correct, and useful behaviour */ |
|
|
977 | if (sbytes) |
|
|
978 | res = sbytes; |
930 | |
979 | |
931 | # elif defined (__APPLE__) |
980 | # elif defined (__APPLE__) |
932 | |
|
|
933 | { |
|
|
934 | off_t sbytes = count; |
981 | off_t sbytes = count; |
935 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
982 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
936 | |
983 | |
937 | /* according to the manpage, sbytes is always valid */ |
984 | /* according to the manpage, sbytes is always valid */ |
938 | if (sbytes) |
985 | if (sbytes) |
939 | res = sbytes; |
986 | res = sbytes; |
940 | } |
|
|
941 | |
987 | |
942 | # elif __hpux |
988 | # elif __hpux |
943 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
989 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
944 | |
990 | |
945 | # elif __solaris |
991 | # elif __solaris |
946 | { |
|
|
947 | struct sendfilevec vec; |
992 | struct sendfilevec vec; |
948 | size_t sbytes; |
993 | size_t sbytes; |
949 | |
994 | |
950 | vec.sfv_fd = ifd; |
995 | vec.sfv_fd = ifd; |
951 | vec.sfv_flag = 0; |
996 | vec.sfv_flag = 0; |
952 | vec.sfv_off = offset; |
997 | vec.sfv_off = offset; |
953 | vec.sfv_len = count; |
998 | vec.sfv_len = count; |
954 | |
999 | |
955 | res = sendfilev (ofd, &vec, 1, &sbytes); |
1000 | res = sendfilev (ofd, &vec, 1, &sbytes); |
956 | |
1001 | |
957 | if (res < 0 && sbytes) |
1002 | if (res < 0 && sbytes) |
958 | res = sbytes; |
1003 | res = sbytes; |
959 | } |
|
|
960 | |
1004 | |
961 | # endif |
1005 | # endif |
962 | |
1006 | |
963 | #elif defined (_WIN32) |
1007 | #elif defined (_WIN32) |
964 | |
|
|
965 | /* does not work, just for documentation of what would need to be done */ |
1008 | /* does not work, just for documentation of what would need to be done */ |
966 | { |
1009 | /* actually, cannot be done like this, as TransmitFile changes the file offset, */ |
|
|
1010 | /* libeio guarantees that the file offset does not change, and windows */ |
|
|
1011 | /* has no way to get an independent handle to the same file description */ |
967 | HANDLE h = TO_SOCKET (ifd); |
1012 | HANDLE h = TO_SOCKET (ifd); |
968 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
1013 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
969 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
1014 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
970 | } |
|
|
971 | |
1015 | |
972 | #else |
1016 | #else |
973 | res = -1; |
1017 | res = -1; |
974 | errno = ENOSYS; |
1018 | errno = ENOSYS; |
975 | #endif |
1019 | #endif |
976 | |
1020 | |
|
|
1021 | /* we assume sendfile can copy at least 128mb in one go */ |
|
|
1022 | if (res <= 128 * 1024 * 1024) |
|
|
1023 | { |
|
|
1024 | if (res > 0) |
|
|
1025 | written += res; |
|
|
1026 | |
|
|
1027 | if (written) |
|
|
1028 | return written; |
|
|
1029 | |
|
|
1030 | break; |
|
|
1031 | } |
|
|
1032 | else |
|
|
1033 | { |
|
|
1034 | /* if we requested more, then probably the kernel was lazy */ |
|
|
1035 | written += res; |
|
|
1036 | offset += res; |
|
|
1037 | count -= res; |
|
|
1038 | |
|
|
1039 | if (!count) |
|
|
1040 | return written; |
|
|
1041 | } |
|
|
1042 | } |
|
|
1043 | |
977 | if (res < 0 |
1044 | if (res < 0 |
978 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
1045 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
979 | /* BSDs */ |
1046 | /* BSDs */ |
980 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
1047 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
981 | || errno == ENOTSUP |
1048 | || errno == ENOTSUP |
982 | #endif |
1049 | #endif |
… | |
… | |
1022 | } |
1089 | } |
1023 | |
1090 | |
1024 | static signed char |
1091 | static signed char |
1025 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
1092 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
1026 | { |
1093 | { |
1027 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
1094 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
1028 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; |
1095 | : a->inode < b->inode ? -1 |
|
|
1096 | : a->inode > b->inode ? 1 |
|
|
1097 | : 0; |
1029 | } |
1098 | } |
1030 | |
1099 | |
1031 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
1100 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
1032 | |
1101 | |
1033 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
1102 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
… | |
… | |
1039 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
1108 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
1040 | unsigned char *bit = bits; |
1109 | unsigned char *bit = bits; |
1041 | |
1110 | |
1042 | assert (CHAR_BIT == 8); |
1111 | assert (CHAR_BIT == 8); |
1043 | assert (sizeof (eio_dirent) * 8 < 256); |
1112 | assert (sizeof (eio_dirent) * 8 < 256); |
1044 | assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ |
1113 | assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ |
1045 | assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ |
1114 | assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */ |
1046 | |
1115 | |
1047 | if (size <= EIO_SORT_FAST) |
1116 | if (size <= EIO_SORT_FAST) |
1048 | return; |
1117 | return; |
1049 | |
1118 | |
1050 | /* first prepare an array of bits to test in our radix sort */ |
1119 | /* first prepare an array of bits to test in our radix sort */ |
… | |
… | |
1205 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1274 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1206 | |
1275 | |
1207 | X_LOCK (wrklock); |
1276 | X_LOCK (wrklock); |
1208 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1277 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1209 | self->dirp = dirp = opendir (req->ptr1); |
1278 | self->dirp = dirp = opendir (req->ptr1); |
|
|
1279 | |
1210 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1280 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1211 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1281 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1212 | req->ptr2 = names = malloc (namesalloc); |
1282 | req->ptr2 = names = malloc (namesalloc); |
1213 | X_UNLOCK (wrklock); |
1283 | X_UNLOCK (wrklock); |
1214 | |
1284 | |
… | |
… | |
1226 | /* sort etc. */ |
1296 | /* sort etc. */ |
1227 | req->int1 = flags; |
1297 | req->int1 = flags; |
1228 | req->result = dentoffs; |
1298 | req->result = dentoffs; |
1229 | |
1299 | |
1230 | if (flags & EIO_READDIR_STAT_ORDER) |
1300 | if (flags & EIO_READDIR_STAT_ORDER) |
1231 | eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ |
1301 | eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); |
1232 | else if (flags & EIO_READDIR_DIRS_FIRST) |
1302 | else if (flags & EIO_READDIR_DIRS_FIRST) |
1233 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
1303 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
1234 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
1304 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
1235 | else |
1305 | else |
1236 | { |
1306 | { |
… | |
… | |
1238 | eio_dirent *oth = dents + dentoffs; |
1308 | eio_dirent *oth = dents + dentoffs; |
1239 | eio_dirent *dir = dents; |
1309 | eio_dirent *dir = dents; |
1240 | |
1310 | |
1241 | /* now partition dirs to the front, and non-dirs to the back */ |
1311 | /* now partition dirs to the front, and non-dirs to the back */ |
1242 | /* by walking from both sides and swapping if necessary */ |
1312 | /* by walking from both sides and swapping if necessary */ |
1243 | /* also clear score, so it doesn't influence sorting */ |
|
|
1244 | while (oth > dir) |
1313 | while (oth > dir) |
1245 | { |
1314 | { |
1246 | if (dir->type == EIO_DT_DIR) |
1315 | if (dir->type == EIO_DT_DIR) |
1247 | ++dir; |
1316 | ++dir; |
1248 | else if ((--oth)->type == EIO_DT_DIR) |
1317 | else if ((--oth)->type == EIO_DT_DIR) |
… | |
… | |
1251 | |
1320 | |
1252 | ++dir; |
1321 | ++dir; |
1253 | } |
1322 | } |
1254 | } |
1323 | } |
1255 | |
1324 | |
1256 | /* now sort the dirs only */ |
1325 | /* now sort the dirs only (dirs all have the same score) */ |
1257 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1326 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1258 | } |
1327 | } |
1259 | |
1328 | |
1260 | break; |
1329 | break; |
1261 | } |
1330 | } |
… | |
… | |
1381 | break; |
1450 | break; |
1382 | } |
1451 | } |
1383 | } |
1452 | } |
1384 | } |
1453 | } |
1385 | |
1454 | |
|
|
1455 | #ifdef PAGESIZE |
|
|
1456 | # define eio_pagesize() PAGESIZE |
|
|
1457 | #else |
|
|
1458 | static intptr_t |
|
|
1459 | eio_pagesize (void) |
|
|
1460 | { |
|
|
1461 | static intptr_t page; |
|
|
1462 | |
|
|
1463 | if (!page) |
|
|
1464 | page = sysconf (_SC_PAGESIZE); |
|
|
1465 | |
|
|
1466 | return page; |
|
|
1467 | } |
|
|
1468 | #endif |
|
|
1469 | |
|
|
1470 | static void |
|
|
1471 | eio_page_align (void **addr, size_t *length) |
|
|
1472 | { |
|
|
1473 | intptr_t mask = eio_pagesize () - 1; |
|
|
1474 | |
|
|
1475 | /* round down addr */ |
|
|
1476 | intptr_t adj = mask & (intptr_t)*addr; |
|
|
1477 | |
|
|
1478 | *addr = (void *)((intptr_t)*addr - adj); |
|
|
1479 | *length += adj; |
|
|
1480 | |
|
|
1481 | /* round up length */ |
|
|
1482 | *length = (*length + mask) & ~mask; |
|
|
1483 | } |
|
|
1484 | |
|
|
1485 | #if !_POSIX_MEMLOCK |
|
|
1486 | # define eio__mlockall(a) ((errno = ENOSYS), -1) |
|
|
1487 | #else |
|
|
1488 | |
|
|
1489 | static int |
|
|
1490 | eio__mlockall (int flags) |
|
|
1491 | { |
|
|
1492 | #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 |
|
|
1493 | extern int mallopt (int, int); |
|
|
1494 | mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ |
|
|
1495 | #endif |
|
|
1496 | |
|
|
1497 | if (EIO_MCL_CURRENT != MCL_CURRENT |
|
|
1498 | || EIO_MCL_FUTURE != MCL_FUTURE) |
|
|
1499 | { |
|
|
1500 | flags = 0 |
|
|
1501 | | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0) |
|
|
1502 | | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0); |
|
|
1503 | } |
|
|
1504 | |
|
|
1505 | return mlockall (flags); |
|
|
1506 | } |
|
|
1507 | #endif |
|
|
1508 | |
|
|
1509 | #if !_POSIX_MEMLOCK_RANGE |
|
|
1510 | # define eio__mlock(a,b) ((errno = ENOSYS), -1) |
|
|
1511 | #else |
|
|
1512 | |
|
|
1513 | static int |
|
|
1514 | eio__mlock (void *addr, size_t length) |
|
|
1515 | { |
|
|
1516 | eio_page_align (&addr, &length); |
|
|
1517 | |
|
|
1518 | return mlock (addr, length); |
|
|
1519 | } |
|
|
1520 | |
|
|
1521 | #endif |
|
|
1522 | |
1386 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1523 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1387 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
1524 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
1388 | #else |
1525 | #else |
1389 | |
1526 | |
1390 | int |
1527 | int |
1391 | eio__msync (void *mem, size_t len, int flags) |
1528 | eio__msync (void *mem, size_t len, int flags) |
1392 | { |
1529 | { |
|
|
1530 | eio_page_align (&mem, &len); |
|
|
1531 | |
1393 | if (EIO_MS_ASYNC != MS_SYNC |
1532 | if (EIO_MS_ASYNC != MS_SYNC |
1394 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
1533 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
1395 | || EIO_MS_SYNC != MS_SYNC) |
1534 | || EIO_MS_SYNC != MS_SYNC) |
1396 | { |
1535 | { |
1397 | flags = 0 |
1536 | flags = 0 |
… | |
… | |
1404 | } |
1543 | } |
1405 | |
1544 | |
1406 | #endif |
1545 | #endif |
1407 | |
1546 | |
1408 | int |
1547 | int |
1409 | eio__mtouch (void *mem, size_t len, int flags) |
1548 | eio__mtouch (eio_req *req) |
1410 | { |
1549 | { |
|
|
1550 | void *mem = req->ptr2; |
|
|
1551 | size_t len = req->size; |
|
|
1552 | int flags = req->int1; |
|
|
1553 | |
|
|
1554 | eio_page_align (&mem, &len); |
|
|
1555 | |
|
|
1556 | { |
1411 | intptr_t addr = (intptr_t)mem; |
1557 | intptr_t addr = (intptr_t)mem; |
1412 | intptr_t end = addr + len; |
1558 | intptr_t end = addr + len; |
1413 | #ifdef PAGESIZE |
1559 | intptr_t page = eio_pagesize (); |
1414 | const intptr_t page = PAGESIZE; |
|
|
1415 | #else |
|
|
1416 | static intptr_t page; |
|
|
1417 | |
1560 | |
1418 | if (!page) |
|
|
1419 | page = sysconf (_SC_PAGESIZE); |
|
|
1420 | #endif |
|
|
1421 | |
|
|
1422 | /* round down to start of page, although this is probably useless */ |
|
|
1423 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
|
|
1424 | |
|
|
1425 | if (addr < end) |
1561 | if (addr < end) |
1426 | if (flags & EIO_MT_MODIFY) /* modify */ |
1562 | if (flags & EIO_MT_MODIFY) /* modify */ |
1427 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1563 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
1428 | else |
1564 | else |
1429 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1565 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
|
|
1566 | } |
1430 | |
1567 | |
1431 | return 0; |
1568 | return 0; |
1432 | } |
1569 | } |
1433 | |
1570 | |
1434 | /*****************************************************************************/ |
1571 | /*****************************************************************************/ |
… | |
… | |
1468 | if (req) |
1605 | if (req) |
1469 | break; |
1606 | break; |
1470 | |
1607 | |
1471 | ++idle; |
1608 | ++idle; |
1472 | |
1609 | |
1473 | ts.tv_sec = time (0) + IDLE_TIMEOUT; |
1610 | ts.tv_sec = time (0) + idle_timeout; |
1474 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1611 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1475 | { |
1612 | { |
1476 | if (idle > max_idle) |
1613 | if (idle > max_idle) |
1477 | { |
1614 | { |
1478 | --idle; |
1615 | --idle; |
… | |
… | |
1557 | return 0; \ |
1694 | return 0; \ |
1558 | } |
1695 | } |
1559 | |
1696 | |
1560 | static void eio_execute (etp_worker *self, eio_req *req) |
1697 | static void eio_execute (etp_worker *self, eio_req *req) |
1561 | { |
1698 | { |
1562 | errno = 0; |
|
|
1563 | |
|
|
1564 | switch (req->type) |
1699 | switch (req->type) |
1565 | { |
1700 | { |
1566 | case EIO_READ: ALLOC (req->size); |
1701 | case EIO_READ: ALLOC (req->size); |
1567 | req->result = req->offs >= 0 |
1702 | req->result = req->offs >= 0 |
1568 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
1703 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
… | |
… | |
1600 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1735 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1601 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1736 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1602 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1737 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1603 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1738 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1604 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1739 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1605 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
1740 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; |
1606 | |
1741 | |
1607 | case EIO_READLINK: ALLOC (NAME_MAX); |
1742 | case EIO_READLINK: ALLOC (PATH_MAX); |
1608 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1743 | req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; |
1609 | |
1744 | |
1610 | case EIO_SYNC: req->result = 0; sync (); break; |
1745 | case EIO_SYNC: req->result = 0; sync (); break; |
1611 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1746 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1612 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1747 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1613 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
1748 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
|
|
1749 | case EIO_MTOUCH: req->result = eio__mtouch (req); break; |
1614 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1750 | case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; |
|
|
1751 | case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; |
1615 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1752 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1616 | |
1753 | |
1617 | case EIO_READDIR: eio__scandir (req, self); break; |
1754 | case EIO_READDIR: eio__scandir (req, self); break; |
1618 | |
1755 | |
1619 | case EIO_BUSY: |
1756 | case EIO_BUSY: |
1620 | #ifdef _WIN32 |
1757 | #ifdef _WIN32 |
1621 | Sleep (req->nv1 * 1000.); |
1758 | Sleep (req->nv1 * 1e3); |
1622 | #else |
1759 | #else |
1623 | { |
1760 | { |
1624 | struct timeval tv; |
1761 | struct timeval tv; |
1625 | |
1762 | |
1626 | tv.tv_sec = req->nv1; |
1763 | tv.tv_sec = req->nv1; |
1627 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
1764 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6; |
1628 | |
1765 | |
1629 | req->result = select (0, 0, 0, 0, &tv); |
1766 | req->result = select (0, 0, 0, 0, &tv); |
1630 | } |
1767 | } |
1631 | #endif |
1768 | #endif |
1632 | break; |
1769 | break; |
… | |
… | |
1647 | times = tv; |
1784 | times = tv; |
1648 | } |
1785 | } |
1649 | else |
1786 | else |
1650 | times = 0; |
1787 | times = 0; |
1651 | |
1788 | |
1652 | |
|
|
1653 | req->result = req->type == EIO_FUTIME |
1789 | req->result = req->type == EIO_FUTIME |
1654 | ? futimes (req->int1, times) |
1790 | ? futimes (req->int1, times) |
1655 | : utimes (req->ptr1, times); |
1791 | : utimes (req->ptr1, times); |
1656 | } |
1792 | } |
1657 | break; |
1793 | break; |
… | |
… | |
1662 | case EIO_NOP: |
1798 | case EIO_NOP: |
1663 | req->result = 0; |
1799 | req->result = 0; |
1664 | break; |
1800 | break; |
1665 | |
1801 | |
1666 | case EIO_CUSTOM: |
1802 | case EIO_CUSTOM: |
1667 | ((void (*)(eio_req *))req->feed) (req); |
1803 | req->feed (req); |
1668 | break; |
1804 | break; |
1669 | |
1805 | |
1670 | default: |
1806 | default: |
|
|
1807 | errno = ENOSYS; |
1671 | req->result = -1; |
1808 | req->result = -1; |
1672 | break; |
1809 | break; |
1673 | } |
1810 | } |
1674 | |
1811 | |
1675 | req->errorno = errno; |
1812 | req->errorno = errno; |
… | |
… | |
1705 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1842 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1706 | { |
1843 | { |
1707 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1844 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1708 | } |
1845 | } |
1709 | |
1846 | |
|
|
1847 | eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data) |
|
|
1848 | { |
|
|
1849 | REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND; |
|
|
1850 | } |
|
|
1851 | |
|
|
1852 | eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) |
|
|
1853 | { |
|
|
1854 | REQ (EIO_MLOCKALL); req->int1 = flags; SEND; |
|
|
1855 | } |
|
|
1856 | |
1710 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
1857 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
1711 | { |
1858 | { |
1712 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
1859 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
1713 | } |
1860 | } |
1714 | |
1861 | |
… | |
… | |
1848 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1995 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1849 | } |
1996 | } |
1850 | |
1997 | |
1851 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1998 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1852 | { |
1999 | { |
1853 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
2000 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND; |
1854 | } |
2001 | } |
1855 | |
2002 | |
1856 | static eio_req * |
2003 | static eio_req * |
1857 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2004 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1858 | { |
2005 | { |
… | |
… | |
1882 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2029 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1883 | { |
2030 | { |
1884 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
2031 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1885 | } |
2032 | } |
1886 | |
2033 | |
1887 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
2034 | eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data) |
1888 | { |
2035 | { |
1889 | REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; |
2036 | REQ (EIO_CUSTOM); req->feed = execute; SEND; |
1890 | } |
2037 | } |
1891 | |
2038 | |
1892 | #endif |
2039 | #endif |
1893 | |
2040 | |
1894 | eio_req *eio_grp (eio_cb cb, void *data) |
2041 | eio_req *eio_grp (eio_cb cb, void *data) |