1 | /* |
1 | /* |
2 | * libeio implementation |
2 | * libeio implementation |
3 | * |
3 | * |
4 | * Copyright (c) 2007,2008,2009 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 <unistd.h> |
88 | # include <unistd.h> |
79 | # include <utime.h> |
89 | # include <utime.h> |
80 | # include <signal.h> |
90 | # include <signal.h> |
81 | # include <dirent.h> |
91 | # include <dirent.h> |
82 | |
92 | |
|
|
93 | #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES |
|
|
94 | # include <sys/mman.h> |
|
|
95 | #endif |
|
|
96 | |
83 | /* 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 */ |
84 | # if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) |
98 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
85 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
99 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
86 | # define D_INO(de) (de)->d_fileno |
100 | # define D_INO(de) (de)->d_fileno |
87 | # define D_NAMLEN(de) (de)->d_namlen |
101 | # define D_NAMLEN(de) (de)->d_namlen |
88 | # elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 |
102 | # elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
89 | # define D_INO(de) (de)->d_ino |
103 | # define D_INO(de) (de)->d_ino |
90 | # endif |
104 | # endif |
91 | |
105 | |
92 | #ifdef _D_EXACT_NAMLEN |
106 | #ifdef _D_EXACT_NAMLEN |
93 | # undef D_NAMLEN |
107 | # undef D_NAMLEN |
… | |
… | |
105 | #endif |
119 | #endif |
106 | |
120 | |
107 | #if HAVE_SENDFILE |
121 | #if HAVE_SENDFILE |
108 | # if __linux |
122 | # if __linux |
109 | # include <sys/sendfile.h> |
123 | # include <sys/sendfile.h> |
110 | # elif __freebsd |
124 | # elif __FreeBSD__ || defined __APPLE__ |
111 | # include <sys/socket.h> |
125 | # include <sys/socket.h> |
112 | # include <sys/uio.h> |
126 | # include <sys/uio.h> |
113 | # elif __hpux |
127 | # elif __hpux |
114 | # include <sys/socket.h> |
128 | # include <sys/socket.h> |
115 | # elif __solaris /* not yet */ |
129 | # elif __solaris |
116 | # include <sys/sendfile.h> |
130 | # include <sys/sendfile.h> |
117 | # else |
131 | # else |
118 | # error sendfile support requested but not available |
132 | # error sendfile support requested but not available |
119 | # endif |
133 | # endif |
120 | #endif |
134 | #endif |
… | |
… | |
127 | #endif |
141 | #endif |
128 | #ifndef D_NAMLEN |
142 | #ifndef D_NAMLEN |
129 | # define D_NAMLEN(de) strlen ((de)->d_name) |
143 | # define D_NAMLEN(de) strlen ((de)->d_name) |
130 | #endif |
144 | #endif |
131 | |
145 | |
132 | /* number of seconds after which an idle threads exit */ |
|
|
133 | #define IDLE_TIMEOUT 10 |
|
|
134 | |
|
|
135 | /* used for struct dirent, AIX doesn't provide it */ |
146 | /* used for struct dirent, AIX doesn't provide it */ |
136 | #ifndef NAME_MAX |
147 | #ifndef NAME_MAX |
137 | # 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 |
138 | #endif |
154 | #endif |
139 | |
155 | |
140 | /* buffer size for various temporary buffers */ |
156 | /* buffer size for various temporary buffers */ |
141 | #define EIO_BUFSIZE 65536 |
157 | #define EIO_BUFSIZE 65536 |
142 | |
158 | |
… | |
… | |
195 | |
211 | |
196 | /*****************************************************************************/ |
212 | /*****************************************************************************/ |
197 | |
213 | |
198 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
214 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
199 | |
215 | |
200 | /* calculcate time difference in ~1/EIO_TICKS of a second */ |
216 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
201 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
217 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
202 | { |
218 | { |
203 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
219 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
204 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
220 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
205 | } |
221 | } |
… | |
… | |
213 | static unsigned int max_poll_reqs; /* reslock */ |
229 | static unsigned int max_poll_reqs; /* reslock */ |
214 | |
230 | |
215 | static volatile unsigned int nreqs; /* reqlock */ |
231 | static volatile unsigned int nreqs; /* reqlock */ |
216 | static volatile unsigned int nready; /* reqlock */ |
232 | static volatile unsigned int nready; /* reqlock */ |
217 | static volatile unsigned int npending; /* reqlock */ |
233 | static volatile unsigned int npending; /* reqlock */ |
218 | 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 */ |
219 | |
236 | |
220 | static mutex_t wrklock = X_MUTEX_INIT; |
237 | static xmutex_t wrklock; |
221 | static mutex_t reslock = X_MUTEX_INIT; |
238 | static xmutex_t reslock; |
222 | static mutex_t reqlock = X_MUTEX_INIT; |
239 | static xmutex_t reqlock; |
223 | static cond_t reqwait = X_COND_INIT; |
240 | static xcond_t reqwait; |
224 | |
241 | |
225 | #if !HAVE_PREADWRITE |
242 | #if !HAVE_PREADWRITE |
226 | /* |
243 | /* |
227 | * make our pread/pwrite emulation safe against themselves, but not against |
244 | * make our pread/pwrite emulation safe against themselves, but not against |
228 | * 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, |
229 | * but that's your problem, not mine. |
246 | * but that's your problem, not mine. |
230 | */ |
247 | */ |
231 | static mutex_t preadwritelock = X_MUTEX_INIT; |
248 | static xmutex_t preadwritelock = X_MUTEX_INIT; |
232 | #endif |
249 | #endif |
233 | |
250 | |
234 | typedef struct etp_worker |
251 | typedef struct etp_worker |
235 | { |
252 | { |
236 | /* locked by wrklock */ |
253 | /* locked by wrklock */ |
237 | struct etp_worker *prev, *next; |
254 | struct etp_worker *prev, *next; |
238 | |
255 | |
239 | thread_t tid; |
256 | xthread_t tid; |
240 | |
257 | |
241 | /* locked by reslock, reqlock or wrklock */ |
258 | /* locked by reslock, reqlock or wrklock */ |
242 | ETP_REQ *req; /* currently processed request */ |
259 | ETP_REQ *req; /* currently processed request */ |
243 | |
260 | |
244 | ETP_WORKER_COMMON |
261 | ETP_WORKER_COMMON |
… | |
… | |
358 | } |
375 | } |
359 | |
376 | |
360 | abort (); |
377 | abort (); |
361 | } |
378 | } |
362 | |
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 | |
363 | static void etp_atfork_prepare (void) |
388 | static void etp_atfork_prepare (void) |
364 | { |
389 | { |
365 | X_LOCK (wrklock); |
390 | X_LOCK (wrklock); |
366 | X_LOCK (reqlock); |
391 | X_LOCK (reqlock); |
367 | X_LOCK (reslock); |
392 | X_LOCK (reslock); |
… | |
… | |
405 | idle = 0; |
430 | idle = 0; |
406 | nreqs = 0; |
431 | nreqs = 0; |
407 | nready = 0; |
432 | nready = 0; |
408 | npending = 0; |
433 | npending = 0; |
409 | |
434 | |
410 | etp_atfork_parent (); |
435 | etp_thread_init (); |
411 | } |
436 | } |
412 | |
437 | |
413 | static void |
438 | static void |
414 | etp_once_init (void) |
439 | etp_once_init (void) |
415 | { |
440 | { |
|
|
441 | etp_thread_init (); |
416 | 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); |
417 | } |
443 | } |
418 | |
444 | |
419 | static int |
445 | static int |
420 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
446 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
… | |
… | |
597 | } |
623 | } |
598 | |
624 | |
599 | static void etp_set_max_poll_time (double nseconds) |
625 | static void etp_set_max_poll_time (double nseconds) |
600 | { |
626 | { |
601 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
627 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
602 | max_poll_time = nseconds; |
628 | max_poll_time = nseconds * EIO_TICKS; |
603 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
629 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
604 | } |
630 | } |
605 | |
631 | |
606 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
632 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
607 | { |
633 | { |
… | |
… | |
611 | } |
637 | } |
612 | |
638 | |
613 | static void etp_set_max_idle (unsigned int nthreads) |
639 | static void etp_set_max_idle (unsigned int nthreads) |
614 | { |
640 | { |
615 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
641 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
616 | 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; |
617 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
650 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
618 | } |
651 | } |
619 | |
652 | |
620 | static void etp_set_min_parallel (unsigned int nthreads) |
653 | static void etp_set_min_parallel (unsigned int nthreads) |
621 | { |
654 | { |
… | |
… | |
749 | void eio_set_max_idle (unsigned int nthreads) |
782 | void eio_set_max_idle (unsigned int nthreads) |
750 | { |
783 | { |
751 | etp_set_max_idle (nthreads); |
784 | etp_set_max_idle (nthreads); |
752 | } |
785 | } |
753 | |
786 | |
|
|
787 | void eio_set_idle_timeout (unsigned int seconds) |
|
|
788 | { |
|
|
789 | etp_set_idle_timeout (seconds); |
|
|
790 | } |
|
|
791 | |
754 | void eio_set_min_parallel (unsigned int nthreads) |
792 | void eio_set_min_parallel (unsigned int nthreads) |
755 | { |
793 | { |
756 | etp_set_min_parallel (nthreads); |
794 | etp_set_min_parallel (nthreads); |
757 | } |
795 | } |
758 | |
796 | |
… | |
… | |
806 | |
844 | |
807 | return res; |
845 | return res; |
808 | } |
846 | } |
809 | #endif |
847 | #endif |
810 | |
848 | |
811 | #ifndef HAVE_FUTIMES |
849 | #ifndef HAVE_UTIMES |
812 | |
850 | |
813 | # undef utimes |
851 | # undef utimes |
814 | # undef futimes |
|
|
815 | # define utimes(path,times) eio__utimes (path, times) |
852 | # define utimes(path,times) eio__utimes (path, times) |
816 | # define futimes(fd,times) eio__futimes (fd, times) |
|
|
817 | |
853 | |
818 | static int |
854 | static int |
819 | eio__utimes (const char *filename, const struct timeval times[2]) |
855 | eio__utimes (const char *filename, const struct timeval times[2]) |
820 | { |
856 | { |
821 | if (times) |
857 | if (times) |
… | |
… | |
828 | return utime (filename, &buf); |
864 | return utime (filename, &buf); |
829 | } |
865 | } |
830 | else |
866 | else |
831 | return utime (filename, 0); |
867 | return utime (filename, 0); |
832 | } |
868 | } |
|
|
869 | |
|
|
870 | #endif |
|
|
871 | |
|
|
872 | #ifndef HAVE_FUTIMES |
|
|
873 | |
|
|
874 | # undef futimes |
|
|
875 | # define futimes(fd,times) eio__futimes (fd, times) |
833 | |
876 | |
834 | static int eio__futimes (int fd, const struct timeval tv[2]) |
877 | static int eio__futimes (int fd, const struct timeval tv[2]) |
835 | { |
878 | { |
836 | errno = ENOSYS; |
879 | errno = ENOSYS; |
837 | return -1; |
880 | return -1; |
… | |
… | |
866 | if (!res || errno != ENOSYS) |
909 | if (!res || errno != ENOSYS) |
867 | return res; |
910 | return res; |
868 | #endif |
911 | #endif |
869 | |
912 | |
870 | /* even though we could play tricks with the flags, it's better to always |
913 | /* even though we could play tricks with the flags, it's better to always |
871 | * call fdatasync, as thta matches the expectation of it's users best */ |
914 | * call fdatasync, as that matches the expectation of its users best */ |
872 | return fdatasync (fd); |
915 | return fdatasync (fd); |
873 | } |
916 | } |
874 | |
917 | |
875 | #if !HAVE_READAHEAD |
918 | #if !HAVE_READAHEAD |
876 | # undef readahead |
919 | # undef readahead |
… | |
… | |
908 | |
951 | |
909 | #if HAVE_SENDFILE |
952 | #if HAVE_SENDFILE |
910 | # if __linux |
953 | # if __linux |
911 | res = sendfile (ofd, ifd, &offset, count); |
954 | res = sendfile (ofd, ifd, &offset, count); |
912 | |
955 | |
913 | # elif __freebsd |
956 | # elif __FreeBSD__ |
914 | /* |
957 | /* |
915 | * 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 |
916 | * wasted on making it similar to other I/O functions. |
959 | * wasted on making it similar to other I/O functions. |
917 | */ |
960 | */ |
918 | { |
961 | { |
919 | off_t sbytes; |
962 | off_t sbytes; |
920 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
963 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
921 | |
964 | |
922 | if (res < 0 && sbytes) |
965 | #if 0 /* according to the manpage, this is correct, but broken behaviour */ |
923 | /* 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) |
|
|
975 | res = sbytes; |
|
|
976 | } |
|
|
977 | |
|
|
978 | # elif defined (__APPLE__) |
|
|
979 | |
|
|
980 | { |
|
|
981 | off_t sbytes = count; |
|
|
982 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
|
|
983 | |
|
|
984 | /* according to the manpage, sbytes is always valid */ |
|
|
985 | if (sbytes) |
924 | res = sbytes; |
986 | res = sbytes; |
925 | } |
987 | } |
926 | |
988 | |
927 | # elif __hpux |
989 | # elif __hpux |
928 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
990 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
… | |
… | |
942 | if (res < 0 && sbytes) |
1004 | if (res < 0 && sbytes) |
943 | res = sbytes; |
1005 | res = sbytes; |
944 | } |
1006 | } |
945 | |
1007 | |
946 | # endif |
1008 | # endif |
|
|
1009 | |
|
|
1010 | #elif defined (_WIN32) |
|
|
1011 | |
|
|
1012 | /* does not work, just for documentation of what would need to be done */ |
|
|
1013 | { |
|
|
1014 | HANDLE h = TO_SOCKET (ifd); |
|
|
1015 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
|
|
1016 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
|
|
1017 | } |
|
|
1018 | |
947 | #else |
1019 | #else |
948 | res = -1; |
1020 | res = -1; |
949 | errno = ENOSYS; |
1021 | errno = ENOSYS; |
950 | #endif |
1022 | #endif |
951 | |
1023 | |
952 | if (res < 0 |
1024 | if (res < 0 |
953 | && (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 |
|
|
1030 | || errno == EOPNOTSUPP /* BSDs */ |
954 | #if __solaris |
1031 | #if __solaris |
955 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
1032 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
956 | #endif |
1033 | #endif |
957 | ) |
1034 | ) |
958 | ) |
1035 | ) |
… | |
… | |
989 | } |
1066 | } |
990 | |
1067 | |
991 | return res; |
1068 | return res; |
992 | } |
1069 | } |
993 | |
1070 | |
994 | static int |
1071 | static signed char |
995 | eio_dent_cmp (const void *a_, const void *b_) |
1072 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
996 | { |
1073 | { |
997 | const eio_dirent *a = (const eio_dirent *)a_; |
1074 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
998 | const eio_dirent *b = (const eio_dirent *)b_; |
1075 | : a->inode < b->inode ? -1 |
|
|
1076 | : a->inode > b->inode ? 1 |
|
|
1077 | : 0; |
|
|
1078 | } |
999 | |
1079 | |
1000 | return (int)b->score - (int)a->score ? (int)b->score - (int)a->score |
1080 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
1001 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */ |
1081 | |
|
|
1082 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
|
|
1083 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
|
|
1084 | |
|
|
1085 | static void |
|
|
1086 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1087 | { |
|
|
1088 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
|
|
1089 | unsigned char *bit = bits; |
|
|
1090 | |
|
|
1091 | assert (CHAR_BIT == 8); |
|
|
1092 | assert (sizeof (eio_dirent) * 8 < 256); |
|
|
1093 | assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ |
|
|
1094 | assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */ |
|
|
1095 | |
|
|
1096 | if (size <= EIO_SORT_FAST) |
|
|
1097 | return; |
|
|
1098 | |
|
|
1099 | /* first prepare an array of bits to test in our radix sort */ |
|
|
1100 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
|
|
1101 | /* inode_bits must contain all inodes ORed together */ |
|
|
1102 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
|
|
1103 | { |
|
|
1104 | ino_t endianness; |
|
|
1105 | int i, j; |
|
|
1106 | |
|
|
1107 | /* we store the byte offset of byte n into byte n of "endianness" */ |
|
|
1108 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1109 | ((unsigned char *)&endianness)[i] = i; |
|
|
1110 | |
|
|
1111 | *bit++ = 0; |
|
|
1112 | |
|
|
1113 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1114 | { |
|
|
1115 | /* shifting off the byte offsets out of "endianness" */ |
|
|
1116 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
|
|
1117 | endianness >>= 8; |
|
|
1118 | |
|
|
1119 | for (j = 0; j < 8; ++j) |
|
|
1120 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
|
|
1121 | *bit++ = offs + j; |
|
|
1122 | } |
|
|
1123 | |
|
|
1124 | for (j = 0; j < 8; ++j) |
|
|
1125 | if (score_bits & (1 << j)) |
|
|
1126 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
|
|
1127 | } |
|
|
1128 | |
|
|
1129 | /* now actually do the sorting (a variant of MSD radix sort) */ |
|
|
1130 | { |
|
|
1131 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
|
|
1132 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
|
|
1133 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
|
|
1134 | int stk_idx = 0; |
|
|
1135 | |
|
|
1136 | base_stk [stk_idx] = dents; |
|
|
1137 | end_stk [stk_idx] = dents + size; |
|
|
1138 | bit_stk [stk_idx] = bit - 1; |
|
|
1139 | |
|
|
1140 | do |
|
|
1141 | { |
|
|
1142 | base = base_stk [stk_idx]; |
|
|
1143 | end = end_stk [stk_idx]; |
|
|
1144 | bit = bit_stk [stk_idx]; |
|
|
1145 | |
|
|
1146 | for (;;) |
|
|
1147 | { |
|
|
1148 | unsigned char O = *bit >> 3; |
|
|
1149 | unsigned char M = 1 << (*bit & 7); |
|
|
1150 | |
|
|
1151 | eio_dirent *a = base; |
|
|
1152 | eio_dirent *b = end; |
|
|
1153 | |
|
|
1154 | if (b - a < EIO_SORT_CUTOFF) |
|
|
1155 | break; |
|
|
1156 | |
|
|
1157 | /* now bit-partition the array on the bit */ |
|
|
1158 | /* this ugly asymmetric loop seems to perform much better than typical */ |
|
|
1159 | /* partition algos found in the literature */ |
|
|
1160 | do |
|
|
1161 | if (!(((unsigned char *)a)[O] & M)) |
|
|
1162 | ++a; |
|
|
1163 | else if (!(((unsigned char *)--b)[O] & M)) |
|
|
1164 | { |
|
|
1165 | eio_dirent tmp = *a; *a = *b; *b = tmp; |
|
|
1166 | ++a; |
|
|
1167 | } |
|
|
1168 | while (b > a); |
|
|
1169 | |
|
|
1170 | /* next bit, or stop, if no bits left in this path */ |
|
|
1171 | if (!*--bit) |
|
|
1172 | break; |
|
|
1173 | |
|
|
1174 | base_stk [stk_idx] = a; |
|
|
1175 | end_stk [stk_idx] = end; |
|
|
1176 | bit_stk [stk_idx] = bit; |
|
|
1177 | ++stk_idx; |
|
|
1178 | |
|
|
1179 | end = a; |
|
|
1180 | } |
|
|
1181 | } |
|
|
1182 | while (stk_idx--); |
|
|
1183 | } |
|
|
1184 | } |
|
|
1185 | |
|
|
1186 | static void |
|
|
1187 | eio_dent_insertion_sort (eio_dirent *dents, int size) |
|
|
1188 | { |
|
|
1189 | /* first move the smallest element to the front, to act as a sentinel */ |
|
|
1190 | { |
|
|
1191 | int i; |
|
|
1192 | eio_dirent *min = dents; |
|
|
1193 | |
|
|
1194 | /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
|
|
1195 | for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
|
|
1196 | if (EIO_DENT_CMP (dents [i], <, *min)) |
|
|
1197 | min = &dents [i]; |
|
|
1198 | |
|
|
1199 | /* swap elements 0 and j (minimum) */ |
|
|
1200 | { |
|
|
1201 | eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
|
|
1202 | } |
|
|
1203 | } |
|
|
1204 | |
|
|
1205 | /* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
|
|
1206 | { |
|
|
1207 | eio_dirent *i, *j; |
|
|
1208 | |
|
|
1209 | for (i = dents + 1; i < dents + size; ++i) |
|
|
1210 | { |
|
|
1211 | eio_dirent value = *i; |
|
|
1212 | |
|
|
1213 | for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
|
|
1214 | j [1] = j [0]; |
|
|
1215 | |
|
|
1216 | j [1] = value; |
|
|
1217 | } |
|
|
1218 | } |
|
|
1219 | } |
|
|
1220 | |
|
|
1221 | static void |
|
|
1222 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1223 | { |
|
|
1224 | if (size <= 1) |
|
|
1225 | return; /* our insertion sort relies on size > 0 */ |
|
|
1226 | |
|
|
1227 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
|
|
1228 | /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
|
|
1229 | eio_dent_radix_sort (dents, size, score_bits, inode_bits); |
|
|
1230 | |
|
|
1231 | /* use an insertion sort at the end, or for small arrays, */ |
|
|
1232 | /* as insertion sort is more efficient for small partitions */ |
|
|
1233 | eio_dent_insertion_sort (dents, size); |
1002 | } |
1234 | } |
1003 | |
1235 | |
1004 | /* read a full directory */ |
1236 | /* read a full directory */ |
1005 | static void |
1237 | static void |
1006 | eio__scandir (eio_req *req, etp_worker *self) |
1238 | eio__scandir (eio_req *req, etp_worker *self) |
1007 | { |
1239 | { |
1008 | DIR *dirp; |
1240 | DIR *dirp; |
1009 | EIO_STRUCT_DIRENT *entp; |
1241 | EIO_STRUCT_DIRENT *entp; |
1010 | unsigned char *name, *names; |
1242 | char *name, *names; |
1011 | int namesalloc = 4096; |
1243 | int namesalloc = 4096; |
1012 | int namesoffs = 0; |
1244 | int namesoffs = 0; |
1013 | int flags = req->int1; |
1245 | int flags = req->int1; |
1014 | eio_dirent *dents = 0; |
1246 | eio_dirent *dents = 0; |
1015 | int dentalloc = 128; |
1247 | int dentalloc = 128; |
1016 | int dentoffs = 0; |
1248 | int dentoffs = 0; |
|
|
1249 | ino_t inode_bits = 0; |
1017 | |
1250 | |
1018 | req->result = -1; |
1251 | req->result = -1; |
1019 | |
1252 | |
1020 | if (!(flags & EIO_READDIR_DENTS)) |
1253 | if (!(flags & EIO_READDIR_DENTS)) |
1021 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1254 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1022 | |
1255 | |
1023 | X_LOCK (wrklock); |
1256 | X_LOCK (wrklock); |
1024 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1257 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1025 | self->dirp = dirp = opendir (req->ptr1); |
1258 | self->dirp = dirp = opendir (req->ptr1); |
|
|
1259 | |
1026 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1260 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1027 | req->ptr1 = names = malloc (namesalloc); |
|
|
1028 | req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1261 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
|
|
1262 | req->ptr2 = names = malloc (namesalloc); |
1029 | X_UNLOCK (wrklock); |
1263 | X_UNLOCK (wrklock); |
1030 | |
1264 | |
1031 | if (dirp && names && (!flags || dents)) |
1265 | if (dirp && names && (!flags || dents)) |
1032 | for (;;) |
1266 | for (;;) |
1033 | { |
1267 | { |
… | |
… | |
1041 | |
1275 | |
1042 | /* sort etc. */ |
1276 | /* sort etc. */ |
1043 | req->int1 = flags; |
1277 | req->int1 = flags; |
1044 | req->result = dentoffs; |
1278 | req->result = dentoffs; |
1045 | |
1279 | |
1046 | if (dents) |
|
|
1047 | { |
|
|
1048 | eio_dirent *ent = dents + dentoffs; |
|
|
1049 | |
|
|
1050 | while (ent > dents) |
|
|
1051 | (--ent)->name = names + (size_t)ent->name; |
|
|
1052 | } |
|
|
1053 | |
|
|
1054 | if (flags & EIO_READDIR_STAT_ORDER |
1280 | if (flags & EIO_READDIR_STAT_ORDER) |
1055 | || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) |
1281 | eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); |
1056 | { |
|
|
1057 | /* pray your qsort doesn't use quicksort */ |
|
|
1058 | qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */ |
|
|
1059 | } |
|
|
1060 | else if (flags & EIO_READDIR_DIRS_FIRST) |
1282 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1283 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
|
|
1284 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
|
|
1285 | else |
1061 | { |
1286 | { |
1062 | /* in this case, all is known, and we just put dirs first and sort them */ |
1287 | /* in this case, all is known, and we just put dirs first and sort them */ |
1063 | eio_dirent *ent = dents + dentoffs; |
1288 | eio_dirent *oth = dents + dentoffs; |
1064 | eio_dirent *dir = dents; |
1289 | eio_dirent *dir = dents; |
1065 | |
1290 | |
|
|
1291 | /* now partition dirs to the front, and non-dirs to the back */ |
|
|
1292 | /* by walking from both sides and swapping if necessary */ |
1066 | while (ent > dir) |
1293 | while (oth > dir) |
1067 | { |
1294 | { |
1068 | if (dir->type == DT_DIR) |
1295 | if (dir->type == EIO_DT_DIR) |
1069 | ++dir; |
1296 | ++dir; |
1070 | else |
|
|
1071 | { |
|
|
1072 | --ent; |
|
|
1073 | |
|
|
1074 | if (ent->type == DT_DIR) |
1297 | else if ((--oth)->type == EIO_DT_DIR) |
1075 | { |
1298 | { |
1076 | eio_dirent tmp = *dir; |
1299 | eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
1077 | *dir = *ent; |
|
|
1078 | *ent = tmp; |
|
|
1079 | |
1300 | |
1080 | ++dir; |
1301 | ++dir; |
1081 | } |
1302 | } |
1082 | } |
1303 | } |
|
|
1304 | |
|
|
1305 | /* now sort the dirs only (dirs all have the same score) */ |
|
|
1306 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1083 | } |
1307 | } |
1084 | |
|
|
1085 | /* now sort the dirs only */ |
|
|
1086 | qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp); |
|
|
1087 | } |
|
|
1088 | |
|
|
1089 | /* only provide the names array unless DENTS is specified */ |
|
|
1090 | if (!(flags & EIO_READDIR_DENTS)) |
|
|
1091 | { |
|
|
1092 | X_LOCK (wrklock); |
|
|
1093 | assert (!dents); |
|
|
1094 | req->ptr1 = 0; |
|
|
1095 | req->ptr2 = names; |
|
|
1096 | X_UNLOCK (wrklock); |
|
|
1097 | } |
|
|
1098 | |
1308 | |
1099 | break; |
1309 | break; |
1100 | } |
1310 | } |
1101 | |
1311 | |
1102 | /* now add the entry to our list(s) */ |
1312 | /* now add the entry to our list(s) */ |
… | |
… | |
1109 | |
1319 | |
1110 | while (expect_false (namesoffs + len > namesalloc)) |
1320 | while (expect_false (namesoffs + len > namesalloc)) |
1111 | { |
1321 | { |
1112 | namesalloc *= 2; |
1322 | namesalloc *= 2; |
1113 | X_LOCK (wrklock); |
1323 | X_LOCK (wrklock); |
1114 | req->ptr1 = names = realloc (names, namesalloc); |
1324 | req->ptr2 = names = realloc (names, namesalloc); |
1115 | X_UNLOCK (wrklock); |
1325 | X_UNLOCK (wrklock); |
1116 | |
1326 | |
1117 | if (!names) |
1327 | if (!names) |
1118 | break; |
1328 | break; |
1119 | } |
1329 | } |
… | |
… | |
1126 | |
1336 | |
1127 | if (expect_false (dentoffs == dentalloc)) |
1337 | if (expect_false (dentoffs == dentalloc)) |
1128 | { |
1338 | { |
1129 | dentalloc *= 2; |
1339 | dentalloc *= 2; |
1130 | X_LOCK (wrklock); |
1340 | X_LOCK (wrklock); |
1131 | req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1341 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1132 | X_UNLOCK (wrklock); |
1342 | X_UNLOCK (wrklock); |
1133 | |
1343 | |
1134 | if (!dents) |
1344 | if (!dents) |
1135 | break; |
1345 | break; |
1136 | } |
1346 | } |
1137 | |
1347 | |
1138 | ent = dents + dentoffs; |
1348 | ent = dents + dentoffs; |
1139 | |
1349 | |
1140 | ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ |
1350 | ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
1141 | ent->namelen = len - 1; |
1351 | ent->namelen = len - 1; |
1142 | ent->inode = D_INO (entp); |
1352 | ent->inode = D_INO (entp); |
|
|
1353 | |
|
|
1354 | inode_bits |= ent->inode; |
1143 | |
1355 | |
1144 | switch (D_TYPE (entp)) |
1356 | switch (D_TYPE (entp)) |
1145 | { |
1357 | { |
1146 | default: |
1358 | default: |
1147 | ent->type = EIO_DT_UNKNOWN; |
1359 | ent->type = EIO_DT_UNKNOWN; |
… | |
… | |
1190 | #ifdef DT_WHT |
1402 | #ifdef DT_WHT |
1191 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1403 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1192 | #endif |
1404 | #endif |
1193 | } |
1405 | } |
1194 | |
1406 | |
1195 | ent->score = 0; |
1407 | ent->score = 7; |
1196 | |
1408 | |
1197 | if (flags & EIO_READDIR_DIRS_FIRST) |
1409 | if (flags & EIO_READDIR_DIRS_FIRST) |
1198 | { |
1410 | { |
1199 | if (ent->type == EIO_DT_UNKNOWN) |
1411 | if (ent->type == EIO_DT_UNKNOWN) |
1200 | { |
1412 | { |
1201 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1413 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1202 | ent->score = 98; |
1414 | ent->score = 1; |
1203 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1415 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1204 | ent->score = len <= 2 ? len + 60 : len <= 4 ? 50 : len <= 7 ? 40 : 10; /* shorter == more likely dir, but avoid too many classes */ |
1416 | ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
1205 | } |
1417 | } |
1206 | else if (ent->type == EIO_DT_DIR) |
1418 | else if (ent->type == EIO_DT_DIR) |
1207 | ent->score = 100; |
1419 | ent->score = 0; |
1208 | } |
1420 | } |
1209 | } |
1421 | } |
1210 | |
1422 | |
1211 | namesoffs += len; |
1423 | namesoffs += len; |
1212 | ++dentoffs; |
1424 | ++dentoffs; |
1213 | } |
1425 | } |
|
|
1426 | |
|
|
1427 | if (EIO_CANCELLED (req)) |
|
|
1428 | { |
|
|
1429 | errno = ECANCELED; |
|
|
1430 | break; |
|
|
1431 | } |
1214 | } |
1432 | } |
1215 | else |
|
|
1216 | req->result = -1; |
|
|
1217 | } |
1433 | } |
|
|
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 |
1218 | |
1502 | |
1219 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1503 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1220 | # undef msync |
|
|
1221 | # define msync(a,b,c) ((errno = ENOSYS), -1) |
1504 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
|
|
1505 | #else |
|
|
1506 | |
|
|
1507 | int |
|
|
1508 | eio__msync (void *mem, size_t len, int flags) |
|
|
1509 | { |
|
|
1510 | eio_page_align (&mem, &len); |
|
|
1511 | |
|
|
1512 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1513 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1514 | || EIO_MS_SYNC != MS_SYNC) |
|
|
1515 | { |
|
|
1516 | flags = 0 |
|
|
1517 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
|
|
1518 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
|
|
1519 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
|
|
1520 | } |
|
|
1521 | |
|
|
1522 | return msync (mem, len, flags); |
|
|
1523 | } |
|
|
1524 | |
1222 | #endif |
1525 | #endif |
1223 | |
1526 | |
1224 | int |
1527 | int |
1225 | eio__mtouch (void *mem, size_t len, int flags) |
1528 | eio__mtouch (void *mem, size_t len, int flags) |
1226 | { |
1529 | { |
|
|
1530 | eio_page_align (&mem, &len); |
|
|
1531 | |
|
|
1532 | { |
1227 | intptr_t addr = (intptr_t)mem; |
1533 | intptr_t addr = (intptr_t)mem; |
1228 | intptr_t end = addr + len; |
1534 | intptr_t end = addr + len; |
1229 | #ifdef PAGESIZE |
1535 | intptr_t page = eio_pagesize (); |
1230 | const intptr_t page = PAGESIZE; |
|
|
1231 | #else |
|
|
1232 | static intptr_t page; |
|
|
1233 | |
1536 | |
1234 | if (!page) |
|
|
1235 | page = sysconf (_SC_PAGESIZE); |
|
|
1236 | #endif |
|
|
1237 | |
|
|
1238 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
|
|
1239 | |
|
|
1240 | if (addr < end) |
1537 | if (addr < end) |
1241 | if (flags) /* modify */ |
1538 | if (flags & EIO_MT_MODIFY) /* modify */ |
1242 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1539 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1243 | else |
1540 | else |
1244 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1541 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
|
|
1542 | } |
1245 | |
1543 | |
1246 | return 0; |
1544 | return 0; |
1247 | } |
1545 | } |
1248 | |
1546 | |
1249 | /*****************************************************************************/ |
1547 | /*****************************************************************************/ |
… | |
… | |
1283 | if (req) |
1581 | if (req) |
1284 | break; |
1582 | break; |
1285 | |
1583 | |
1286 | ++idle; |
1584 | ++idle; |
1287 | |
1585 | |
1288 | ts.tv_sec = time (0) + IDLE_TIMEOUT; |
1586 | ts.tv_sec = time (0) + idle_timeout; |
1289 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1587 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1290 | { |
1588 | { |
1291 | if (idle > max_idle) |
1589 | if (idle > max_idle) |
1292 | { |
1590 | { |
1293 | --idle; |
1591 | --idle; |
… | |
… | |
1372 | return 0; \ |
1670 | return 0; \ |
1373 | } |
1671 | } |
1374 | |
1672 | |
1375 | static void eio_execute (etp_worker *self, eio_req *req) |
1673 | static void eio_execute (etp_worker *self, eio_req *req) |
1376 | { |
1674 | { |
1377 | errno = 0; |
|
|
1378 | |
|
|
1379 | switch (req->type) |
1675 | switch (req->type) |
1380 | { |
1676 | { |
1381 | case EIO_READ: ALLOC (req->size); |
1677 | case EIO_READ: ALLOC (req->size); |
1382 | req->result = req->offs >= 0 |
1678 | req->result = req->offs >= 0 |
1383 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
1679 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
… | |
… | |
1394 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1690 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1395 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1691 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1396 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1692 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1397 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1693 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1398 | |
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 | |
1399 | 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; |
1400 | 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; |
1401 | 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; |
1402 | 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; |
1403 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
1704 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
… | |
… | |
1410 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1711 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1411 | 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; |
1412 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1713 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1413 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1714 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1414 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1715 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1415 | 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; |
1416 | |
1717 | |
1417 | case EIO_READLINK: ALLOC (NAME_MAX); |
1718 | case EIO_READLINK: ALLOC (PATH_MAX); |
1418 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1719 | req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; |
1419 | |
1720 | |
1420 | case EIO_SYNC: req->result = 0; sync (); break; |
1721 | case EIO_SYNC: req->result = 0; sync (); break; |
1421 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1722 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1422 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1723 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1423 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
1724 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
1424 | 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; |
1425 | 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; |
1426 | |
1729 | |
1427 | case EIO_READDIR: eio__scandir (req, self); break; |
1730 | case EIO_READDIR: eio__scandir (req, self); break; |
1428 | |
1731 | |
1429 | case EIO_BUSY: |
1732 | case EIO_BUSY: |
1430 | #ifdef _WIN32 |
1733 | #ifdef _WIN32 |
1431 | Sleep (req->nv1 * 1000.); |
1734 | Sleep (req->nv1 * 1e3); |
1432 | #else |
1735 | #else |
1433 | { |
1736 | { |
1434 | struct timeval tv; |
1737 | struct timeval tv; |
1435 | |
1738 | |
1436 | tv.tv_sec = req->nv1; |
1739 | tv.tv_sec = req->nv1; |
1437 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
1740 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6; |
1438 | |
1741 | |
1439 | req->result = select (0, 0, 0, 0, &tv); |
1742 | req->result = select (0, 0, 0, 0, &tv); |
1440 | } |
1743 | } |
1441 | #endif |
1744 | #endif |
1442 | break; |
1745 | break; |
… | |
… | |
1457 | times = tv; |
1760 | times = tv; |
1458 | } |
1761 | } |
1459 | else |
1762 | else |
1460 | times = 0; |
1763 | times = 0; |
1461 | |
1764 | |
1462 | |
|
|
1463 | req->result = req->type == EIO_FUTIME |
1765 | req->result = req->type == EIO_FUTIME |
1464 | ? futimes (req->int1, times) |
1766 | ? futimes (req->int1, times) |
1465 | : utimes (req->ptr1, times); |
1767 | : utimes (req->ptr1, times); |
1466 | } |
1768 | } |
1467 | break; |
1769 | break; |
… | |
… | |
1476 | case EIO_CUSTOM: |
1778 | case EIO_CUSTOM: |
1477 | ((void (*)(eio_req *))req->feed) (req); |
1779 | ((void (*)(eio_req *))req->feed) (req); |
1478 | break; |
1780 | break; |
1479 | |
1781 | |
1480 | default: |
1782 | default: |
|
|
1783 | errno = ENOSYS; |
1481 | req->result = -1; |
1784 | req->result = -1; |
1482 | break; |
1785 | break; |
1483 | } |
1786 | } |
1484 | |
1787 | |
1485 | req->errorno = errno; |
1788 | req->errorno = errno; |
… | |
… | |
1515 | 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) |
1516 | { |
1819 | { |
1517 | 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; |
1518 | } |
1821 | } |
1519 | |
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 | |
1520 | 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) |
1521 | { |
1834 | { |
1522 | 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; |
1523 | } |
1836 | } |
1524 | |
1837 | |
… | |
… | |
1548 | } |
1861 | } |
1549 | |
1862 | |
1550 | 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) |
1551 | { |
1864 | { |
1552 | 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; |
1553 | } |
1871 | } |
1554 | |
1872 | |
1555 | 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) |
1556 | { |
1874 | { |
1557 | 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; |
… | |
… | |
1631 | 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) |
1632 | { |
1950 | { |
1633 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1951 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1634 | } |
1952 | } |
1635 | |
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 | |
1636 | 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) |
1637 | { |
1960 | { |
1638 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1961 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1639 | } |
1962 | } |
1640 | |
1963 | |
… | |
… | |
1648 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1971 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1649 | } |
1972 | } |
1650 | |
1973 | |
1651 | 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) |
1652 | { |
1975 | { |
1653 | 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; |
1654 | } |
1977 | } |
1655 | |
1978 | |
1656 | static eio_req * |
1979 | static eio_req * |
1657 | 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) |
1658 | { |
1981 | { |
… | |
… | |
1742 | /* misc garbage */ |
2065 | /* misc garbage */ |
1743 | |
2066 | |
1744 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
2067 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1745 | { |
2068 | { |
1746 | etp_worker wrk; |
2069 | etp_worker wrk; |
|
|
2070 | ssize_t ret; |
1747 | |
2071 | |
1748 | wrk.dbuf = 0; |
2072 | wrk.dbuf = 0; |
1749 | |
2073 | |
1750 | eio__sendfile (ofd, ifd, offset, count, &wrk); |
2074 | ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
1751 | |
2075 | |
1752 | if (wrk.dbuf) |
2076 | if (wrk.dbuf) |
1753 | free (wrk.dbuf); |
2077 | free (wrk.dbuf); |
1754 | } |
|
|
1755 | |
2078 | |
|
|
2079 | return ret; |
|
|
2080 | } |
|
|
2081 | |