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 | |
58 | #ifndef EIO_FINISH |
63 | #ifndef EIO_FINISH |
… | |
… | |
70 | #ifdef _WIN32 |
75 | #ifdef _WIN32 |
71 | |
76 | |
72 | /*doh*/ |
77 | /*doh*/ |
73 | #else |
78 | #else |
74 | |
79 | |
75 | # include "config.h" |
|
|
76 | # include <sys/time.h> |
80 | # include <sys/time.h> |
77 | # include <sys/select.h> |
81 | # include <sys/select.h> |
78 | # include <sys/mman.h> |
|
|
79 | # include <unistd.h> |
82 | # include <unistd.h> |
80 | # include <utime.h> |
83 | # include <utime.h> |
81 | # include <signal.h> |
84 | # include <signal.h> |
82 | # include <dirent.h> |
85 | # include <dirent.h> |
83 | |
86 | |
|
|
87 | #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES |
|
|
88 | # include <sys/mman.h> |
|
|
89 | #endif |
|
|
90 | |
84 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
91 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
85 | # if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) |
92 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
86 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
93 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
87 | # define D_INO(de) (de)->d_fileno |
94 | # define D_INO(de) (de)->d_fileno |
88 | # define D_NAMLEN(de) (de)->d_namlen |
95 | # define D_NAMLEN(de) (de)->d_namlen |
89 | # elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 |
96 | # elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
90 | # define D_INO(de) (de)->d_ino |
97 | # define D_INO(de) (de)->d_ino |
91 | # endif |
98 | # endif |
92 | |
99 | |
93 | #ifdef _D_EXACT_NAMLEN |
100 | #ifdef _D_EXACT_NAMLEN |
94 | # undef D_NAMLEN |
101 | # undef D_NAMLEN |
… | |
… | |
106 | #endif |
113 | #endif |
107 | |
114 | |
108 | #if HAVE_SENDFILE |
115 | #if HAVE_SENDFILE |
109 | # if __linux |
116 | # if __linux |
110 | # include <sys/sendfile.h> |
117 | # include <sys/sendfile.h> |
111 | # elif __freebsd |
118 | # elif __FreeBSD__ || defined __APPLE__ |
112 | # include <sys/socket.h> |
119 | # include <sys/socket.h> |
113 | # include <sys/uio.h> |
120 | # include <sys/uio.h> |
114 | # elif __hpux |
121 | # elif __hpux |
115 | # include <sys/socket.h> |
122 | # include <sys/socket.h> |
116 | # elif __solaris /* not yet */ |
123 | # elif __solaris |
117 | # include <sys/sendfile.h> |
124 | # include <sys/sendfile.h> |
118 | # else |
125 | # else |
119 | # error sendfile support requested but not available |
126 | # error sendfile support requested but not available |
120 | # endif |
127 | # endif |
121 | #endif |
128 | #endif |
… | |
… | |
128 | #endif |
135 | #endif |
129 | #ifndef D_NAMLEN |
136 | #ifndef D_NAMLEN |
130 | # define D_NAMLEN(de) strlen ((de)->d_name) |
137 | # define D_NAMLEN(de) strlen ((de)->d_name) |
131 | #endif |
138 | #endif |
132 | |
139 | |
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 */ |
140 | /* used for struct dirent, AIX doesn't provide it */ |
137 | #ifndef NAME_MAX |
141 | #ifndef NAME_MAX |
138 | # define NAME_MAX 4096 |
142 | # define NAME_MAX 4096 |
|
|
143 | #endif |
|
|
144 | |
|
|
145 | /* used for readlink etc. */ |
|
|
146 | #ifndef PATH_MAX |
|
|
147 | # define PATH_MAX 4096 |
139 | #endif |
148 | #endif |
140 | |
149 | |
141 | /* buffer size for various temporary buffers */ |
150 | /* buffer size for various temporary buffers */ |
142 | #define EIO_BUFSIZE 65536 |
151 | #define EIO_BUFSIZE 65536 |
143 | |
152 | |
… | |
… | |
196 | |
205 | |
197 | /*****************************************************************************/ |
206 | /*****************************************************************************/ |
198 | |
207 | |
199 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
208 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
200 | |
209 | |
201 | /* calculcate time difference in ~1/EIO_TICKS of a second */ |
210 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
202 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
211 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
203 | { |
212 | { |
204 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
213 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
205 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
214 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
206 | } |
215 | } |
… | |
… | |
214 | static unsigned int max_poll_reqs; /* reslock */ |
223 | static unsigned int max_poll_reqs; /* reslock */ |
215 | |
224 | |
216 | static volatile unsigned int nreqs; /* reqlock */ |
225 | static volatile unsigned int nreqs; /* reqlock */ |
217 | static volatile unsigned int nready; /* reqlock */ |
226 | static volatile unsigned int nready; /* reqlock */ |
218 | static volatile unsigned int npending; /* reqlock */ |
227 | static volatile unsigned int npending; /* reqlock */ |
219 | static volatile unsigned int max_idle = 4; |
228 | static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ |
|
|
229 | static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ |
220 | |
230 | |
221 | static mutex_t wrklock = X_MUTEX_INIT; |
231 | static xmutex_t wrklock; |
222 | static mutex_t reslock = X_MUTEX_INIT; |
232 | static xmutex_t reslock; |
223 | static mutex_t reqlock = X_MUTEX_INIT; |
233 | static xmutex_t reqlock; |
224 | static cond_t reqwait = X_COND_INIT; |
234 | static xcond_t reqwait; |
225 | |
235 | |
226 | #if !HAVE_PREADWRITE |
236 | #if !HAVE_PREADWRITE |
227 | /* |
237 | /* |
228 | * make our pread/pwrite emulation safe against themselves, but not against |
238 | * make our pread/pwrite emulation safe against themselves, but not against |
229 | * normal read/write by using a mutex. slows down execution a lot, |
239 | * normal read/write by using a mutex. slows down execution a lot, |
230 | * but that's your problem, not mine. |
240 | * but that's your problem, not mine. |
231 | */ |
241 | */ |
232 | static mutex_t preadwritelock = X_MUTEX_INIT; |
242 | static xmutex_t preadwritelock = X_MUTEX_INIT; |
233 | #endif |
243 | #endif |
234 | |
244 | |
235 | typedef struct etp_worker |
245 | typedef struct etp_worker |
236 | { |
246 | { |
237 | /* locked by wrklock */ |
247 | /* locked by wrklock */ |
238 | struct etp_worker *prev, *next; |
248 | struct etp_worker *prev, *next; |
239 | |
249 | |
240 | thread_t tid; |
250 | xthread_t tid; |
241 | |
251 | |
242 | /* locked by reslock, reqlock or wrklock */ |
252 | /* locked by reslock, reqlock or wrklock */ |
243 | ETP_REQ *req; /* currently processed request */ |
253 | ETP_REQ *req; /* currently processed request */ |
244 | |
254 | |
245 | ETP_WORKER_COMMON |
255 | ETP_WORKER_COMMON |
… | |
… | |
359 | } |
369 | } |
360 | |
370 | |
361 | abort (); |
371 | abort (); |
362 | } |
372 | } |
363 | |
373 | |
|
|
374 | static void etp_thread_init (void) |
|
|
375 | { |
|
|
376 | X_MUTEX_CREATE (wrklock); |
|
|
377 | X_MUTEX_CREATE (reslock); |
|
|
378 | X_MUTEX_CREATE (reqlock); |
|
|
379 | X_COND_CREATE (reqwait); |
|
|
380 | } |
|
|
381 | |
364 | static void etp_atfork_prepare (void) |
382 | static void etp_atfork_prepare (void) |
365 | { |
383 | { |
366 | X_LOCK (wrklock); |
384 | X_LOCK (wrklock); |
367 | X_LOCK (reqlock); |
385 | X_LOCK (reqlock); |
368 | X_LOCK (reslock); |
386 | X_LOCK (reslock); |
… | |
… | |
406 | idle = 0; |
424 | idle = 0; |
407 | nreqs = 0; |
425 | nreqs = 0; |
408 | nready = 0; |
426 | nready = 0; |
409 | npending = 0; |
427 | npending = 0; |
410 | |
428 | |
411 | etp_atfork_parent (); |
429 | etp_thread_init (); |
412 | } |
430 | } |
413 | |
431 | |
414 | static void |
432 | static void |
415 | etp_once_init (void) |
433 | etp_once_init (void) |
416 | { |
434 | { |
|
|
435 | etp_thread_init (); |
417 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
436 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
418 | } |
437 | } |
419 | |
438 | |
420 | static int |
439 | static int |
421 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
440 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
… | |
… | |
598 | } |
617 | } |
599 | |
618 | |
600 | static void etp_set_max_poll_time (double nseconds) |
619 | static void etp_set_max_poll_time (double nseconds) |
601 | { |
620 | { |
602 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
621 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
603 | max_poll_time = nseconds; |
622 | max_poll_time = nseconds * EIO_TICKS; |
604 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
623 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
605 | } |
624 | } |
606 | |
625 | |
607 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
626 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
608 | { |
627 | { |
… | |
… | |
612 | } |
631 | } |
613 | |
632 | |
614 | static void etp_set_max_idle (unsigned int nthreads) |
633 | static void etp_set_max_idle (unsigned int nthreads) |
615 | { |
634 | { |
616 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
635 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
617 | max_idle = nthreads <= 0 ? 1 : nthreads; |
636 | max_idle = nthreads; |
|
|
637 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
|
|
638 | } |
|
|
639 | |
|
|
640 | static void etp_set_idle_timeout (unsigned int seconds) |
|
|
641 | { |
|
|
642 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
|
|
643 | idle_timeout = seconds; |
618 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
644 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
619 | } |
645 | } |
620 | |
646 | |
621 | static void etp_set_min_parallel (unsigned int nthreads) |
647 | static void etp_set_min_parallel (unsigned int nthreads) |
622 | { |
648 | { |
… | |
… | |
750 | void eio_set_max_idle (unsigned int nthreads) |
776 | void eio_set_max_idle (unsigned int nthreads) |
751 | { |
777 | { |
752 | etp_set_max_idle (nthreads); |
778 | etp_set_max_idle (nthreads); |
753 | } |
779 | } |
754 | |
780 | |
|
|
781 | void eio_set_idle_timeout (unsigned int seconds) |
|
|
782 | { |
|
|
783 | etp_set_idle_timeout (seconds); |
|
|
784 | } |
|
|
785 | |
755 | void eio_set_min_parallel (unsigned int nthreads) |
786 | void eio_set_min_parallel (unsigned int nthreads) |
756 | { |
787 | { |
757 | etp_set_min_parallel (nthreads); |
788 | etp_set_min_parallel (nthreads); |
758 | } |
789 | } |
759 | |
790 | |
… | |
… | |
807 | |
838 | |
808 | return res; |
839 | return res; |
809 | } |
840 | } |
810 | #endif |
841 | #endif |
811 | |
842 | |
812 | #ifndef HAVE_FUTIMES |
843 | #ifndef HAVE_UTIMES |
813 | |
844 | |
814 | # undef utimes |
845 | # undef utimes |
815 | # undef futimes |
|
|
816 | # define utimes(path,times) eio__utimes (path, times) |
846 | # define utimes(path,times) eio__utimes (path, times) |
817 | # define futimes(fd,times) eio__futimes (fd, times) |
|
|
818 | |
847 | |
819 | static int |
848 | static int |
820 | eio__utimes (const char *filename, const struct timeval times[2]) |
849 | eio__utimes (const char *filename, const struct timeval times[2]) |
821 | { |
850 | { |
822 | if (times) |
851 | if (times) |
… | |
… | |
829 | return utime (filename, &buf); |
858 | return utime (filename, &buf); |
830 | } |
859 | } |
831 | else |
860 | else |
832 | return utime (filename, 0); |
861 | return utime (filename, 0); |
833 | } |
862 | } |
|
|
863 | |
|
|
864 | #endif |
|
|
865 | |
|
|
866 | #ifndef HAVE_FUTIMES |
|
|
867 | |
|
|
868 | # undef futimes |
|
|
869 | # define futimes(fd,times) eio__futimes (fd, times) |
834 | |
870 | |
835 | static int eio__futimes (int fd, const struct timeval tv[2]) |
871 | static int eio__futimes (int fd, const struct timeval tv[2]) |
836 | { |
872 | { |
837 | errno = ENOSYS; |
873 | errno = ENOSYS; |
838 | return -1; |
874 | return -1; |
… | |
… | |
867 | if (!res || errno != ENOSYS) |
903 | if (!res || errno != ENOSYS) |
868 | return res; |
904 | return res; |
869 | #endif |
905 | #endif |
870 | |
906 | |
871 | /* even though we could play tricks with the flags, it's better to always |
907 | /* even though we could play tricks with the flags, it's better to always |
872 | * call fdatasync, as thta matches the expectation of it's users best */ |
908 | * call fdatasync, as that matches the expectation of its users best */ |
873 | return fdatasync (fd); |
909 | return fdatasync (fd); |
874 | } |
910 | } |
875 | |
911 | |
876 | #if !HAVE_READAHEAD |
912 | #if !HAVE_READAHEAD |
877 | # undef readahead |
913 | # undef readahead |
… | |
… | |
909 | |
945 | |
910 | #if HAVE_SENDFILE |
946 | #if HAVE_SENDFILE |
911 | # if __linux |
947 | # if __linux |
912 | res = sendfile (ofd, ifd, &offset, count); |
948 | res = sendfile (ofd, ifd, &offset, count); |
913 | |
949 | |
914 | # elif __freebsd |
950 | # elif __FreeBSD__ |
915 | /* |
951 | /* |
916 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
952 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
917 | * wasted on making it similar to other I/O functions. |
953 | * wasted on making it similar to other I/O functions. |
918 | */ |
954 | */ |
919 | { |
955 | { |
920 | off_t sbytes; |
956 | off_t sbytes; |
921 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
957 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
922 | |
958 | |
923 | if (res < 0 && sbytes) |
959 | #if 0 /* according to the manpage, this is correct, but broken behaviour */ |
924 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
960 | /* freebsd' sendfile will return 0 on success */ |
|
|
961 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
|
|
962 | /* not on e.g. EIO or EPIPE - sounds broken */ |
|
|
963 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
|
|
964 | res = sbytes; |
|
|
965 | #endif |
|
|
966 | |
|
|
967 | /* according to source inspection, this is correct, and useful behaviour */ |
|
|
968 | if (sbytes) |
|
|
969 | res = sbytes; |
|
|
970 | } |
|
|
971 | |
|
|
972 | # elif defined (__APPLE__) |
|
|
973 | |
|
|
974 | { |
|
|
975 | off_t sbytes = count; |
|
|
976 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
|
|
977 | |
|
|
978 | /* according to the manpage, sbytes is always valid */ |
|
|
979 | if (sbytes) |
925 | res = sbytes; |
980 | res = sbytes; |
926 | } |
981 | } |
927 | |
982 | |
928 | # elif __hpux |
983 | # elif __hpux |
929 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
984 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
… | |
… | |
943 | if (res < 0 && sbytes) |
998 | if (res < 0 && sbytes) |
944 | res = sbytes; |
999 | res = sbytes; |
945 | } |
1000 | } |
946 | |
1001 | |
947 | # endif |
1002 | # endif |
|
|
1003 | |
|
|
1004 | #elif defined (_WIN32) |
|
|
1005 | |
|
|
1006 | /* does not work, just for documentation of what would need to be done */ |
|
|
1007 | { |
|
|
1008 | HANDLE h = TO_SOCKET (ifd); |
|
|
1009 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
|
|
1010 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
|
|
1011 | } |
|
|
1012 | |
948 | #else |
1013 | #else |
949 | res = -1; |
1014 | res = -1; |
950 | errno = ENOSYS; |
1015 | errno = ENOSYS; |
951 | #endif |
1016 | #endif |
952 | |
1017 | |
953 | if (res < 0 |
1018 | if (res < 0 |
954 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
1019 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
|
|
1020 | /* BSDs */ |
|
|
1021 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
|
|
1022 | || errno == ENOTSUP |
|
|
1023 | #endif |
|
|
1024 | || errno == EOPNOTSUPP /* BSDs */ |
955 | #if __solaris |
1025 | #if __solaris |
956 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
1026 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
957 | #endif |
1027 | #endif |
958 | ) |
1028 | ) |
959 | ) |
1029 | ) |
… | |
… | |
993 | } |
1063 | } |
994 | |
1064 | |
995 | static signed char |
1065 | static signed char |
996 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
1066 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
997 | { |
1067 | { |
998 | return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */ |
1068 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
999 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; |
1069 | : a->inode < b->inode ? -1 |
|
|
1070 | : a->inode > b->inode ? 1 |
|
|
1071 | : 0; |
1000 | } |
1072 | } |
1001 | |
1073 | |
|
|
1074 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
|
|
1075 | |
1002 | #define EIO_QSORT_CUTOFF 20 /* quite high, but performs well on many filesystems */ |
1076 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
|
|
1077 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
1003 | |
1078 | |
1004 | static void |
1079 | static void |
1005 | eio_dent_sort (eio_dirent *dents, int size) |
1080 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
1006 | { |
1081 | { |
|
|
1082 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
|
|
1083 | unsigned char *bit = bits; |
|
|
1084 | |
|
|
1085 | assert (CHAR_BIT == 8); |
|
|
1086 | assert (sizeof (eio_dirent) * 8 < 256); |
|
|
1087 | assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ |
|
|
1088 | assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */ |
|
|
1089 | |
|
|
1090 | if (size <= EIO_SORT_FAST) |
|
|
1091 | return; |
|
|
1092 | |
|
|
1093 | /* first prepare an array of bits to test in our radix sort */ |
|
|
1094 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
|
|
1095 | /* inode_bits must contain all inodes ORed together */ |
|
|
1096 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
|
|
1097 | { |
|
|
1098 | ino_t endianness; |
1007 | int i, j; |
1099 | int i, j; |
1008 | |
1100 | |
1009 | if (size > EIO_QSORT_CUTOFF * 3) /* skip quicksort for small directories */ |
1101 | /* we store the byte offset of byte n into byte n of "endianness" */ |
|
|
1102 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1103 | ((unsigned char *)&endianness)[i] = i; |
|
|
1104 | |
|
|
1105 | *bit++ = 0; |
|
|
1106 | |
|
|
1107 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1108 | { |
|
|
1109 | /* shifting off the byte offsets out of "endianness" */ |
|
|
1110 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
|
|
1111 | endianness >>= 8; |
|
|
1112 | |
|
|
1113 | for (j = 0; j < 8; ++j) |
|
|
1114 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
|
|
1115 | *bit++ = offs + j; |
|
|
1116 | } |
|
|
1117 | |
|
|
1118 | for (j = 0; j < 8; ++j) |
|
|
1119 | if (score_bits & (1 << j)) |
|
|
1120 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
|
|
1121 | } |
|
|
1122 | |
|
|
1123 | /* now actually do the sorting (a variant of MSD radix sort) */ |
|
|
1124 | { |
|
|
1125 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
|
|
1126 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
|
|
1127 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
|
|
1128 | int stk_idx = 0; |
|
|
1129 | |
|
|
1130 | base_stk [stk_idx] = dents; |
|
|
1131 | end_stk [stk_idx] = dents + size; |
|
|
1132 | bit_stk [stk_idx] = bit - 1; |
|
|
1133 | |
|
|
1134 | do |
|
|
1135 | { |
|
|
1136 | base = base_stk [stk_idx]; |
|
|
1137 | end = end_stk [stk_idx]; |
|
|
1138 | bit = bit_stk [stk_idx]; |
|
|
1139 | |
|
|
1140 | for (;;) |
|
|
1141 | { |
|
|
1142 | unsigned char O = *bit >> 3; |
|
|
1143 | unsigned char M = 1 << (*bit & 7); |
|
|
1144 | |
|
|
1145 | eio_dirent *a = base; |
|
|
1146 | eio_dirent *b = end; |
|
|
1147 | |
|
|
1148 | if (b - a < EIO_SORT_CUTOFF) |
|
|
1149 | break; |
|
|
1150 | |
|
|
1151 | /* now bit-partition the array on the bit */ |
|
|
1152 | /* this ugly asymmetric loop seems to perform much better than typical */ |
|
|
1153 | /* partition algos found in the literature */ |
|
|
1154 | do |
|
|
1155 | if (!(((unsigned char *)a)[O] & M)) |
|
|
1156 | ++a; |
|
|
1157 | else if (!(((unsigned char *)--b)[O] & M)) |
|
|
1158 | { |
|
|
1159 | eio_dirent tmp = *a; *a = *b; *b = tmp; |
|
|
1160 | ++a; |
|
|
1161 | } |
|
|
1162 | while (b > a); |
|
|
1163 | |
|
|
1164 | /* next bit, or stop, if no bits left in this path */ |
|
|
1165 | if (!*--bit) |
|
|
1166 | break; |
|
|
1167 | |
|
|
1168 | base_stk [stk_idx] = a; |
|
|
1169 | end_stk [stk_idx] = end; |
|
|
1170 | bit_stk [stk_idx] = bit; |
|
|
1171 | ++stk_idx; |
|
|
1172 | |
|
|
1173 | end = a; |
|
|
1174 | } |
|
|
1175 | } |
|
|
1176 | while (stk_idx--); |
|
|
1177 | } |
|
|
1178 | } |
|
|
1179 | |
|
|
1180 | static void |
|
|
1181 | eio_dent_insertion_sort (eio_dirent *dents, int size) |
|
|
1182 | { |
|
|
1183 | /* first move the smallest element to the front, to act as a sentinel */ |
|
|
1184 | { |
|
|
1185 | int i; |
|
|
1186 | eio_dirent *min = dents; |
|
|
1187 | |
|
|
1188 | /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
|
|
1189 | for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
|
|
1190 | if (EIO_DENT_CMP (dents [i], <, *min)) |
|
|
1191 | min = &dents [i]; |
|
|
1192 | |
|
|
1193 | /* swap elements 0 and j (minimum) */ |
1010 | { |
1194 | { |
1011 | /* first, use quicksort */ |
1195 | eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
1012 | /* should be good for 2**31 entries */ |
1196 | } |
1013 | struct rng { int l, r; } rng [32]; |
1197 | } |
1014 | |
1198 | |
1015 | i = 0; |
1199 | /* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
1016 | rng[0].l = 0; |
1200 | { |
1017 | rng[0].r = size; |
1201 | eio_dirent *i, *j; |
1018 | |
1202 | |
1019 | while (expect_true (i >= 0)) |
1203 | for (i = dents + 1; i < dents + size; ++i) |
1020 | { |
1204 | { |
1021 | int L = rng [i].l; |
1205 | eio_dirent value = *i; |
1022 | int R = rng [i].r - 1; |
|
|
1023 | |
1206 | |
1024 | if (expect_false (L + EIO_QSORT_CUTOFF < R)) |
1207 | for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
1025 | { |
1208 | j [1] = j [0]; |
1026 | eio_dirent piv = dents [L]; |
|
|
1027 | |
1209 | |
1028 | while (L < R) |
1210 | j [1] = value; |
1029 | { |
|
|
1030 | while (eio_dent_cmp (&dents [R], &piv) >= 0 && L < R) |
|
|
1031 | --R; |
|
|
1032 | |
|
|
1033 | if (L < R) |
|
|
1034 | dents [L++] = dents [R]; |
|
|
1035 | |
|
|
1036 | while (eio_dent_cmp (&dents [L], &piv) <= 0 && L < R) |
|
|
1037 | ++L; |
|
|
1038 | |
|
|
1039 | if (L < R) |
|
|
1040 | dents [R--] = dents [L]; |
|
|
1041 | } |
|
|
1042 | |
|
|
1043 | dents [L] = piv; |
|
|
1044 | |
|
|
1045 | ++i; |
|
|
1046 | rng [i].l = L + 1; |
|
|
1047 | rng [i].r = rng [i - 1].r; |
|
|
1048 | rng [i - 1].r = L; |
|
|
1049 | |
|
|
1050 | if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l) |
|
|
1051 | { |
|
|
1052 | struct rng t; |
|
|
1053 | |
|
|
1054 | t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t; |
|
|
1055 | } |
|
|
1056 | } |
|
|
1057 | else |
|
|
1058 | --i; |
|
|
1059 | } |
1211 | } |
1060 | } |
1212 | } |
|
|
1213 | } |
1061 | |
1214 | |
1062 | /* use a simple insertion sort at the end */ |
1215 | static void |
1063 | for (i = 1; i < size; ++i) |
1216 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
1064 | { |
1217 | { |
1065 | eio_dirent value = dents [i]; |
1218 | if (size <= 1) |
|
|
1219 | return; /* our insertion sort relies on size > 0 */ |
1066 | |
1220 | |
1067 | for (j = i - 1; j >= 0 && eio_dent_cmp (&dents [j], &value) > 0; --j) |
1221 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
1068 | dents [j + 1] = dents [j]; |
1222 | /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
|
|
1223 | eio_dent_radix_sort (dents, size, score_bits, inode_bits); |
1069 | |
1224 | |
1070 | dents [j + 1] = value; |
1225 | /* use an insertion sort at the end, or for small arrays, */ |
1071 | } |
1226 | /* as insertion sort is more efficient for small partitions */ |
|
|
1227 | eio_dent_insertion_sort (dents, size); |
1072 | } |
1228 | } |
1073 | |
1229 | |
1074 | /* read a full directory */ |
1230 | /* read a full directory */ |
1075 | static void |
1231 | static void |
1076 | eio__scandir (eio_req *req, etp_worker *self) |
1232 | eio__scandir (eio_req *req, etp_worker *self) |
… | |
… | |
1082 | int namesoffs = 0; |
1238 | int namesoffs = 0; |
1083 | int flags = req->int1; |
1239 | int flags = req->int1; |
1084 | eio_dirent *dents = 0; |
1240 | eio_dirent *dents = 0; |
1085 | int dentalloc = 128; |
1241 | int dentalloc = 128; |
1086 | int dentoffs = 0; |
1242 | int dentoffs = 0; |
|
|
1243 | ino_t inode_bits = 0; |
1087 | |
1244 | |
1088 | req->result = -1; |
1245 | req->result = -1; |
1089 | |
1246 | |
1090 | if (!(flags & EIO_READDIR_DENTS)) |
1247 | if (!(flags & EIO_READDIR_DENTS)) |
1091 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1248 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1092 | |
1249 | |
1093 | X_LOCK (wrklock); |
1250 | X_LOCK (wrklock); |
1094 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1251 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1095 | self->dirp = dirp = opendir (req->ptr1); |
1252 | self->dirp = dirp = opendir (req->ptr1); |
|
|
1253 | |
1096 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1254 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1097 | req->ptr1 = names = malloc (namesalloc); |
|
|
1098 | req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1255 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
|
|
1256 | req->ptr2 = names = malloc (namesalloc); |
1099 | X_UNLOCK (wrklock); |
1257 | X_UNLOCK (wrklock); |
1100 | |
1258 | |
1101 | if (dirp && names && (!flags || dents)) |
1259 | if (dirp && names && (!flags || dents)) |
1102 | for (;;) |
1260 | for (;;) |
1103 | { |
1261 | { |
… | |
… | |
1111 | |
1269 | |
1112 | /* sort etc. */ |
1270 | /* sort etc. */ |
1113 | req->int1 = flags; |
1271 | req->int1 = flags; |
1114 | req->result = dentoffs; |
1272 | req->result = dentoffs; |
1115 | |
1273 | |
1116 | if (dents) |
|
|
1117 | { |
|
|
1118 | eio_dirent *ent = dents + dentoffs; |
|
|
1119 | |
|
|
1120 | while (ent > dents) |
|
|
1121 | (--ent)->name = names + (size_t)ent->name; |
|
|
1122 | } |
|
|
1123 | |
|
|
1124 | if (flags & EIO_READDIR_STAT_ORDER |
1274 | if (flags & EIO_READDIR_STAT_ORDER) |
1125 | || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) |
1275 | eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); |
1126 | eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */ |
|
|
1127 | else if (flags & EIO_READDIR_DIRS_FIRST) |
1276 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1277 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
|
|
1278 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
|
|
1279 | else |
1128 | { |
1280 | { |
1129 | /* in this case, all is known, and we just put dirs first and sort them */ |
1281 | /* in this case, all is known, and we just put dirs first and sort them */ |
1130 | eio_dirent *ent = dents + dentoffs; |
1282 | eio_dirent *oth = dents + dentoffs; |
1131 | eio_dirent *dir = dents; |
1283 | eio_dirent *dir = dents; |
1132 | |
1284 | |
1133 | /* now move dirs to the front, and non-dirs to the back */ |
1285 | /* now partition dirs to the front, and non-dirs to the back */ |
1134 | /* by walking from both sides and swapping if necessary */ |
1286 | /* by walking from both sides and swapping if necessary */ |
1135 | while (ent > dir) |
1287 | while (oth > dir) |
1136 | { |
1288 | { |
1137 | if (dir->type == DT_DIR) |
1289 | if (dir->type == EIO_DT_DIR) |
1138 | ++dir; |
1290 | ++dir; |
1139 | else |
|
|
1140 | { |
|
|
1141 | --ent; |
|
|
1142 | |
|
|
1143 | if (ent->type == DT_DIR) |
1291 | else if ((--oth)->type == EIO_DT_DIR) |
1144 | { |
1292 | { |
1145 | eio_dirent tmp = *dir; |
1293 | eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
1146 | *dir = *ent; |
|
|
1147 | *ent = tmp; |
|
|
1148 | |
1294 | |
1149 | ++dir; |
1295 | ++dir; |
1150 | } |
1296 | } |
1151 | } |
1297 | } |
|
|
1298 | |
|
|
1299 | /* now sort the dirs only (dirs all have the same score) */ |
|
|
1300 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1152 | } |
1301 | } |
1153 | |
|
|
1154 | /* now sort the dirs only */ |
|
|
1155 | eio_dent_sort (dents, dir - dents); |
|
|
1156 | } |
|
|
1157 | |
|
|
1158 | /* only provide the names array unless DENTS is specified */ |
|
|
1159 | if (!(flags & EIO_READDIR_DENTS)) |
|
|
1160 | { |
|
|
1161 | X_LOCK (wrklock); |
|
|
1162 | assert (!dents); |
|
|
1163 | req->ptr1 = 0; |
|
|
1164 | req->ptr2 = names; |
|
|
1165 | X_UNLOCK (wrklock); |
|
|
1166 | } |
|
|
1167 | |
1302 | |
1168 | break; |
1303 | break; |
1169 | } |
1304 | } |
1170 | |
1305 | |
1171 | /* now add the entry to our list(s) */ |
1306 | /* now add the entry to our list(s) */ |
… | |
… | |
1178 | |
1313 | |
1179 | while (expect_false (namesoffs + len > namesalloc)) |
1314 | while (expect_false (namesoffs + len > namesalloc)) |
1180 | { |
1315 | { |
1181 | namesalloc *= 2; |
1316 | namesalloc *= 2; |
1182 | X_LOCK (wrklock); |
1317 | X_LOCK (wrklock); |
1183 | req->ptr1 = names = realloc (names, namesalloc); |
1318 | req->ptr2 = names = realloc (names, namesalloc); |
1184 | X_UNLOCK (wrklock); |
1319 | X_UNLOCK (wrklock); |
1185 | |
1320 | |
1186 | if (!names) |
1321 | if (!names) |
1187 | break; |
1322 | break; |
1188 | } |
1323 | } |
… | |
… | |
1195 | |
1330 | |
1196 | if (expect_false (dentoffs == dentalloc)) |
1331 | if (expect_false (dentoffs == dentalloc)) |
1197 | { |
1332 | { |
1198 | dentalloc *= 2; |
1333 | dentalloc *= 2; |
1199 | X_LOCK (wrklock); |
1334 | X_LOCK (wrklock); |
1200 | req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1335 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1201 | X_UNLOCK (wrklock); |
1336 | X_UNLOCK (wrklock); |
1202 | |
1337 | |
1203 | if (!dents) |
1338 | if (!dents) |
1204 | break; |
1339 | break; |
1205 | } |
1340 | } |
1206 | |
1341 | |
1207 | ent = dents + dentoffs; |
1342 | ent = dents + dentoffs; |
1208 | |
1343 | |
1209 | ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ |
1344 | ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
1210 | ent->namelen = len - 1; |
1345 | ent->namelen = len - 1; |
1211 | ent->inode = D_INO (entp); |
1346 | ent->inode = D_INO (entp); |
|
|
1347 | |
|
|
1348 | inode_bits |= ent->inode; |
1212 | |
1349 | |
1213 | switch (D_TYPE (entp)) |
1350 | switch (D_TYPE (entp)) |
1214 | { |
1351 | { |
1215 | default: |
1352 | default: |
1216 | ent->type = EIO_DT_UNKNOWN; |
1353 | ent->type = EIO_DT_UNKNOWN; |
… | |
… | |
1259 | #ifdef DT_WHT |
1396 | #ifdef DT_WHT |
1260 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1397 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1261 | #endif |
1398 | #endif |
1262 | } |
1399 | } |
1263 | |
1400 | |
1264 | ent->score = 0; |
1401 | ent->score = 7; |
1265 | |
1402 | |
1266 | if (flags & EIO_READDIR_DIRS_FIRST) |
1403 | if (flags & EIO_READDIR_DIRS_FIRST) |
1267 | { |
1404 | { |
1268 | if (ent->type == EIO_DT_UNKNOWN) |
1405 | if (ent->type == EIO_DT_UNKNOWN) |
1269 | { |
1406 | { |
1270 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1407 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1271 | ent->score = 98; |
1408 | ent->score = 1; |
1272 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1409 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1273 | ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ |
1410 | ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
1274 | } |
1411 | } |
1275 | else if (ent->type == EIO_DT_DIR) |
1412 | else if (ent->type == EIO_DT_DIR) |
1276 | ent->score = 100; |
1413 | ent->score = 0; |
1277 | } |
1414 | } |
1278 | } |
1415 | } |
1279 | |
1416 | |
1280 | namesoffs += len; |
1417 | namesoffs += len; |
1281 | ++dentoffs; |
1418 | ++dentoffs; |
… | |
… | |
1287 | break; |
1424 | break; |
1288 | } |
1425 | } |
1289 | } |
1426 | } |
1290 | } |
1427 | } |
1291 | |
1428 | |
|
|
1429 | #ifdef PAGESIZE |
|
|
1430 | # define eio_pagesize() PAGESIZE |
|
|
1431 | #else |
|
|
1432 | static intptr_t |
|
|
1433 | eio_pagesize (void) |
|
|
1434 | { |
|
|
1435 | static intptr_t page; |
|
|
1436 | |
|
|
1437 | if (!page) |
|
|
1438 | page = sysconf (_SC_PAGESIZE); |
|
|
1439 | |
|
|
1440 | return page; |
|
|
1441 | } |
|
|
1442 | #endif |
|
|
1443 | |
|
|
1444 | static void |
|
|
1445 | eio_page_align (void **addr, size_t *length) |
|
|
1446 | { |
|
|
1447 | intptr_t mask = eio_pagesize () - 1; |
|
|
1448 | |
|
|
1449 | /* round down addr */ |
|
|
1450 | intptr_t adj = mask & (intptr_t)*addr; |
|
|
1451 | |
|
|
1452 | *addr = (void *)((intptr_t)*addr - adj); |
|
|
1453 | *length += adj; |
|
|
1454 | |
|
|
1455 | /* round up length */ |
|
|
1456 | *length = (*length + mask) & ~mask; |
|
|
1457 | } |
|
|
1458 | |
|
|
1459 | #if !_POSIX_MEMLOCK |
|
|
1460 | # define eio__mlockall(a) ((errno = ENOSYS), -1) |
|
|
1461 | #else |
|
|
1462 | |
|
|
1463 | static int |
|
|
1464 | eio__mlockall (int flags) |
|
|
1465 | { |
|
|
1466 | #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 |
|
|
1467 | extern int mallopt (int, int); |
|
|
1468 | mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ |
|
|
1469 | #endif |
|
|
1470 | |
|
|
1471 | if (EIO_MCL_CURRENT != MCL_CURRENT |
|
|
1472 | || EIO_MCL_FUTURE != MCL_FUTURE) |
|
|
1473 | { |
|
|
1474 | flags = 0 |
|
|
1475 | | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0) |
|
|
1476 | | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0); |
|
|
1477 | } |
|
|
1478 | |
|
|
1479 | return mlockall (flags); |
|
|
1480 | } |
|
|
1481 | #endif |
|
|
1482 | |
|
|
1483 | #if !_POSIX_MEMLOCK_RANGE |
|
|
1484 | # define eio__mlock(a,b) ((errno = ENOSYS), -1) |
|
|
1485 | #else |
|
|
1486 | |
|
|
1487 | static int |
|
|
1488 | eio__mlock (void *addr, size_t length) |
|
|
1489 | { |
|
|
1490 | eio_page_align (&addr, &length); |
|
|
1491 | |
|
|
1492 | return mlock (addr, length); |
|
|
1493 | } |
|
|
1494 | |
|
|
1495 | #endif |
|
|
1496 | |
1292 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1497 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1293 | # undef msync |
|
|
1294 | # define msync(a,b,c) ((errno = ENOSYS), -1) |
1498 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
|
|
1499 | #else |
|
|
1500 | |
|
|
1501 | int |
|
|
1502 | eio__msync (void *mem, size_t len, int flags) |
|
|
1503 | { |
|
|
1504 | eio_page_align (&mem, &len); |
|
|
1505 | |
|
|
1506 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1507 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1508 | || EIO_MS_SYNC != MS_SYNC) |
|
|
1509 | { |
|
|
1510 | flags = 0 |
|
|
1511 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
|
|
1512 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
|
|
1513 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
|
|
1514 | } |
|
|
1515 | |
|
|
1516 | return msync (mem, len, flags); |
|
|
1517 | } |
|
|
1518 | |
1295 | #endif |
1519 | #endif |
1296 | |
1520 | |
1297 | int |
1521 | int |
1298 | eio__mtouch (void *mem, size_t len, int flags) |
1522 | eio__mtouch (void *mem, size_t len, int flags) |
1299 | { |
1523 | { |
|
|
1524 | eio_page_align (&mem, &len); |
|
|
1525 | |
|
|
1526 | { |
1300 | intptr_t addr = (intptr_t)mem; |
1527 | intptr_t addr = (intptr_t)mem; |
1301 | intptr_t end = addr + len; |
1528 | intptr_t end = addr + len; |
1302 | #ifdef PAGESIZE |
1529 | intptr_t page = eio_pagesize (); |
1303 | const intptr_t page = PAGESIZE; |
|
|
1304 | #else |
|
|
1305 | static intptr_t page; |
|
|
1306 | |
1530 | |
1307 | if (!page) |
|
|
1308 | page = sysconf (_SC_PAGESIZE); |
|
|
1309 | #endif |
|
|
1310 | |
|
|
1311 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
|
|
1312 | |
|
|
1313 | if (addr < end) |
1531 | if (addr < end) |
1314 | if (flags) /* modify */ |
1532 | if (flags & EIO_MT_MODIFY) /* modify */ |
1315 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1533 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1316 | else |
1534 | else |
1317 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1535 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
|
|
1536 | } |
1318 | |
1537 | |
1319 | return 0; |
1538 | return 0; |
1320 | } |
1539 | } |
1321 | |
1540 | |
1322 | /*****************************************************************************/ |
1541 | /*****************************************************************************/ |
… | |
… | |
1356 | if (req) |
1575 | if (req) |
1357 | break; |
1576 | break; |
1358 | |
1577 | |
1359 | ++idle; |
1578 | ++idle; |
1360 | |
1579 | |
1361 | ts.tv_sec = time (0) + IDLE_TIMEOUT; |
1580 | ts.tv_sec = time (0) + idle_timeout; |
1362 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1581 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1363 | { |
1582 | { |
1364 | if (idle > max_idle) |
1583 | if (idle > max_idle) |
1365 | { |
1584 | { |
1366 | --idle; |
1585 | --idle; |
… | |
… | |
1445 | return 0; \ |
1664 | return 0; \ |
1446 | } |
1665 | } |
1447 | |
1666 | |
1448 | static void eio_execute (etp_worker *self, eio_req *req) |
1667 | static void eio_execute (etp_worker *self, eio_req *req) |
1449 | { |
1668 | { |
1450 | errno = 0; |
|
|
1451 | |
|
|
1452 | switch (req->type) |
1669 | switch (req->type) |
1453 | { |
1670 | { |
1454 | case EIO_READ: ALLOC (req->size); |
1671 | case EIO_READ: ALLOC (req->size); |
1455 | req->result = req->offs >= 0 |
1672 | req->result = req->offs >= 0 |
1456 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
1673 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
… | |
… | |
1467 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1684 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1468 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1685 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1469 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1686 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1470 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1687 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1471 | |
1688 | |
|
|
1689 | case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1690 | req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1691 | case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1692 | req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1693 | |
1472 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1694 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1473 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1695 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1474 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1696 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1475 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1697 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1476 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
1698 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
… | |
… | |
1483 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1705 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1484 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1706 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1485 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1707 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1486 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1708 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1487 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1709 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1488 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
1710 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; |
1489 | |
1711 | |
1490 | case EIO_READLINK: ALLOC (NAME_MAX); |
1712 | case EIO_READLINK: ALLOC (PATH_MAX); |
1491 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1713 | req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; |
1492 | |
1714 | |
1493 | case EIO_SYNC: req->result = 0; sync (); break; |
1715 | case EIO_SYNC: req->result = 0; sync (); break; |
1494 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1716 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1495 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1717 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1496 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
1718 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
1497 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1719 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
|
|
1720 | case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; |
|
|
1721 | case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; |
1498 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1722 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1499 | |
1723 | |
1500 | case EIO_READDIR: eio__scandir (req, self); break; |
1724 | case EIO_READDIR: eio__scandir (req, self); break; |
1501 | |
1725 | |
1502 | case EIO_BUSY: |
1726 | case EIO_BUSY: |
1503 | #ifdef _WIN32 |
1727 | #ifdef _WIN32 |
1504 | Sleep (req->nv1 * 1000.); |
1728 | Sleep (req->nv1 * 1e3); |
1505 | #else |
1729 | #else |
1506 | { |
1730 | { |
1507 | struct timeval tv; |
1731 | struct timeval tv; |
1508 | |
1732 | |
1509 | tv.tv_sec = req->nv1; |
1733 | tv.tv_sec = req->nv1; |
1510 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
1734 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6; |
1511 | |
1735 | |
1512 | req->result = select (0, 0, 0, 0, &tv); |
1736 | req->result = select (0, 0, 0, 0, &tv); |
1513 | } |
1737 | } |
1514 | #endif |
1738 | #endif |
1515 | break; |
1739 | break; |
… | |
… | |
1530 | times = tv; |
1754 | times = tv; |
1531 | } |
1755 | } |
1532 | else |
1756 | else |
1533 | times = 0; |
1757 | times = 0; |
1534 | |
1758 | |
1535 | |
|
|
1536 | req->result = req->type == EIO_FUTIME |
1759 | req->result = req->type == EIO_FUTIME |
1537 | ? futimes (req->int1, times) |
1760 | ? futimes (req->int1, times) |
1538 | : utimes (req->ptr1, times); |
1761 | : utimes (req->ptr1, times); |
1539 | } |
1762 | } |
1540 | break; |
1763 | break; |
… | |
… | |
1549 | case EIO_CUSTOM: |
1772 | case EIO_CUSTOM: |
1550 | ((void (*)(eio_req *))req->feed) (req); |
1773 | ((void (*)(eio_req *))req->feed) (req); |
1551 | break; |
1774 | break; |
1552 | |
1775 | |
1553 | default: |
1776 | default: |
|
|
1777 | errno = ENOSYS; |
1554 | req->result = -1; |
1778 | req->result = -1; |
1555 | break; |
1779 | break; |
1556 | } |
1780 | } |
1557 | |
1781 | |
1558 | req->errorno = errno; |
1782 | req->errorno = errno; |
… | |
… | |
1588 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1812 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1589 | { |
1813 | { |
1590 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1814 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1591 | } |
1815 | } |
1592 | |
1816 | |
|
|
1817 | eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data) |
|
|
1818 | { |
|
|
1819 | REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND; |
|
|
1820 | } |
|
|
1821 | |
|
|
1822 | eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) |
|
|
1823 | { |
|
|
1824 | REQ (EIO_MLOCKALL); req->int1 = flags; SEND; |
|
|
1825 | } |
|
|
1826 | |
1593 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
1827 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
1594 | { |
1828 | { |
1595 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
1829 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
1596 | } |
1830 | } |
1597 | |
1831 | |
… | |
… | |
1621 | } |
1855 | } |
1622 | |
1856 | |
1623 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1857 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1624 | { |
1858 | { |
1625 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
1859 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
|
|
1860 | } |
|
|
1861 | |
|
|
1862 | eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data) |
|
|
1863 | { |
|
|
1864 | REQ (EIO_FSTATVFS); req->int1 = fd; SEND; |
1626 | } |
1865 | } |
1627 | |
1866 | |
1628 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1867 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1629 | { |
1868 | { |
1630 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
1869 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
… | |
… | |
1704 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1943 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1705 | { |
1944 | { |
1706 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1945 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1707 | } |
1946 | } |
1708 | |
1947 | |
|
|
1948 | eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data) |
|
|
1949 | { |
|
|
1950 | return eio__1path (EIO_STATVFS, path, pri, cb, data); |
|
|
1951 | } |
|
|
1952 | |
1709 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1953 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1710 | { |
1954 | { |
1711 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1955 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1712 | } |
1956 | } |
1713 | |
1957 | |
… | |
… | |
1721 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1965 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1722 | } |
1966 | } |
1723 | |
1967 | |
1724 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1968 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1725 | { |
1969 | { |
1726 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
1970 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND; |
1727 | } |
1971 | } |
1728 | |
1972 | |
1729 | static eio_req * |
1973 | static eio_req * |
1730 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1974 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1731 | { |
1975 | { |