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 | * |
… | |
… | |
49 | #include <stdlib.h> |
49 | #include <stdlib.h> |
50 | #include <string.h> |
50 | #include <string.h> |
51 | #include <errno.h> |
51 | #include <errno.h> |
52 | #include <sys/types.h> |
52 | #include <sys/types.h> |
53 | #include <sys/stat.h> |
53 | #include <sys/stat.h> |
|
|
54 | #include <sys/statvfs.h> |
54 | #include <limits.h> |
55 | #include <limits.h> |
55 | #include <fcntl.h> |
56 | #include <fcntl.h> |
56 | #include <assert.h> |
57 | #include <assert.h> |
57 | |
58 | |
58 | #ifndef EIO_FINISH |
59 | #ifndef EIO_FINISH |
… | |
… | |
78 | # include <unistd.h> |
79 | # include <unistd.h> |
79 | # include <utime.h> |
80 | # include <utime.h> |
80 | # include <signal.h> |
81 | # include <signal.h> |
81 | # include <dirent.h> |
82 | # include <dirent.h> |
82 | |
83 | |
|
|
84 | #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES |
|
|
85 | # include <sys/mman.h> |
|
|
86 | #endif |
|
|
87 | |
83 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
88 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
84 | # if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) |
89 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
|
|
90 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
85 | # define D_INO(de) (de)->d_fileno |
91 | # define D_INO(de) (de)->d_fileno |
86 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
92 | # define D_NAMLEN(de) (de)->d_namlen |
87 | # elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 |
93 | # elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
88 | # define D_INO(de) (de)->d_ino |
94 | # define D_INO(de) (de)->d_ino |
89 | # endif |
95 | # endif |
|
|
96 | |
|
|
97 | #ifdef _D_EXACT_NAMLEN |
|
|
98 | # undef D_NAMLEN |
|
|
99 | # define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
|
|
100 | #endif |
90 | |
101 | |
91 | # ifdef _DIRENT_HAVE_D_TYPE |
102 | # ifdef _DIRENT_HAVE_D_TYPE |
92 | # define D_TYPE(de) (de)->d_type |
103 | # define D_TYPE(de) (de)->d_type |
93 | # endif |
104 | # endif |
94 | |
105 | |
… | |
… | |
99 | #endif |
110 | #endif |
100 | |
111 | |
101 | #if HAVE_SENDFILE |
112 | #if HAVE_SENDFILE |
102 | # if __linux |
113 | # if __linux |
103 | # include <sys/sendfile.h> |
114 | # include <sys/sendfile.h> |
104 | # elif __freebsd |
115 | # elif __FreeBSD__ || defined __APPLE__ |
105 | # include <sys/socket.h> |
116 | # include <sys/socket.h> |
106 | # include <sys/uio.h> |
117 | # include <sys/uio.h> |
107 | # elif __hpux |
118 | # elif __hpux |
108 | # include <sys/socket.h> |
119 | # include <sys/socket.h> |
109 | # elif __solaris /* not yet */ |
120 | # elif __solaris |
110 | # include <sys/sendfile.h> |
121 | # include <sys/sendfile.h> |
111 | # else |
122 | # else |
112 | # error sendfile support requested but not available |
123 | # error sendfile support requested but not available |
113 | # endif |
124 | # endif |
114 | #endif |
125 | #endif |
115 | |
126 | |
116 | #ifndef D_TYPE |
127 | #ifndef D_TYPE |
117 | # define D_TYPE(de) 0 |
128 | # define D_TYPE(de) 0 |
118 | #endif |
129 | #endif |
119 | |
|
|
120 | #ifndef D_INO |
130 | #ifndef D_INO |
121 | # define D_INO(de) 0 |
131 | # define D_INO(de) 0 |
122 | #endif |
132 | #endif |
123 | |
133 | #ifndef D_NAMLEN |
124 | /* number of seconds after which an idle threads exit */ |
134 | # define D_NAMLEN(de) strlen ((de)->d_name) |
125 | #define IDLE_TIMEOUT 10 |
135 | #endif |
126 | |
136 | |
127 | /* used for struct dirent, AIX doesn't provide it */ |
137 | /* used for struct dirent, AIX doesn't provide it */ |
128 | #ifndef NAME_MAX |
138 | #ifndef NAME_MAX |
129 | # define NAME_MAX 4096 |
139 | # define NAME_MAX 4096 |
|
|
140 | #endif |
|
|
141 | |
|
|
142 | /* used for readlink etc. */ |
|
|
143 | #ifndef PATH_MAX |
|
|
144 | # define PATH_MAX 4096 |
130 | #endif |
145 | #endif |
131 | |
146 | |
132 | /* buffer size for various temporary buffers */ |
147 | /* buffer size for various temporary buffers */ |
133 | #define EIO_BUFSIZE 65536 |
148 | #define EIO_BUFSIZE 65536 |
134 | |
149 | |
… | |
… | |
178 | if (wrk->dirp) \ |
193 | if (wrk->dirp) \ |
179 | { \ |
194 | { \ |
180 | closedir (wrk->dirp); \ |
195 | closedir (wrk->dirp); \ |
181 | wrk->dirp = 0; \ |
196 | wrk->dirp = 0; \ |
182 | } |
197 | } |
|
|
198 | |
183 | #define ETP_WORKER_COMMON \ |
199 | #define ETP_WORKER_COMMON \ |
184 | void *dbuf; \ |
200 | void *dbuf; \ |
185 | DIR *dirp; |
201 | DIR *dirp; |
186 | |
202 | |
187 | /*****************************************************************************/ |
203 | /*****************************************************************************/ |
188 | |
204 | |
189 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
205 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
190 | |
206 | |
191 | /* calculcate time difference in ~1/EIO_TICKS of a second */ |
207 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
192 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
208 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
193 | { |
209 | { |
194 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
210 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
195 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
211 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
196 | } |
212 | } |
… | |
… | |
204 | static unsigned int max_poll_reqs; /* reslock */ |
220 | static unsigned int max_poll_reqs; /* reslock */ |
205 | |
221 | |
206 | static volatile unsigned int nreqs; /* reqlock */ |
222 | static volatile unsigned int nreqs; /* reqlock */ |
207 | static volatile unsigned int nready; /* reqlock */ |
223 | static volatile unsigned int nready; /* reqlock */ |
208 | static volatile unsigned int npending; /* reqlock */ |
224 | static volatile unsigned int npending; /* reqlock */ |
209 | static volatile unsigned int max_idle = 4; |
225 | static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ |
|
|
226 | static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ |
210 | |
227 | |
211 | static mutex_t wrklock = X_MUTEX_INIT; |
228 | static xmutex_t wrklock; |
212 | static mutex_t reslock = X_MUTEX_INIT; |
229 | static xmutex_t reslock; |
213 | static mutex_t reqlock = X_MUTEX_INIT; |
230 | static xmutex_t reqlock; |
214 | static cond_t reqwait = X_COND_INIT; |
231 | static xcond_t reqwait; |
215 | |
232 | |
216 | #if !HAVE_PREADWRITE |
233 | #if !HAVE_PREADWRITE |
217 | /* |
234 | /* |
218 | * make our pread/pwrite emulation safe against themselves, but not against |
235 | * make our pread/pwrite emulation safe against themselves, but not against |
219 | * normal read/write by using a mutex. slows down execution a lot, |
236 | * normal read/write by using a mutex. slows down execution a lot, |
220 | * but that's your problem, not mine. |
237 | * but that's your problem, not mine. |
221 | */ |
238 | */ |
222 | static mutex_t preadwritelock = X_MUTEX_INIT; |
239 | static xmutex_t preadwritelock = X_MUTEX_INIT; |
223 | #endif |
240 | #endif |
224 | |
241 | |
225 | typedef struct etp_worker |
242 | typedef struct etp_worker |
226 | { |
243 | { |
227 | /* locked by wrklock */ |
244 | /* locked by wrklock */ |
228 | struct etp_worker *prev, *next; |
245 | struct etp_worker *prev, *next; |
229 | |
246 | |
230 | thread_t tid; |
247 | xthread_t tid; |
231 | |
248 | |
232 | /* locked by reslock, reqlock or wrklock */ |
249 | /* locked by reslock, reqlock or wrklock */ |
233 | ETP_REQ *req; /* currently processed request */ |
250 | ETP_REQ *req; /* currently processed request */ |
234 | |
251 | |
235 | ETP_WORKER_COMMON |
252 | ETP_WORKER_COMMON |
… | |
… | |
349 | } |
366 | } |
350 | |
367 | |
351 | abort (); |
368 | abort (); |
352 | } |
369 | } |
353 | |
370 | |
|
|
371 | static void etp_thread_init (void) |
|
|
372 | { |
|
|
373 | X_MUTEX_CREATE (wrklock); |
|
|
374 | X_MUTEX_CREATE (reslock); |
|
|
375 | X_MUTEX_CREATE (reqlock); |
|
|
376 | X_COND_CREATE (reqwait); |
|
|
377 | } |
|
|
378 | |
354 | static void etp_atfork_prepare (void) |
379 | static void etp_atfork_prepare (void) |
355 | { |
380 | { |
356 | X_LOCK (wrklock); |
381 | X_LOCK (wrklock); |
357 | X_LOCK (reqlock); |
382 | X_LOCK (reqlock); |
358 | X_LOCK (reslock); |
383 | X_LOCK (reslock); |
… | |
… | |
396 | idle = 0; |
421 | idle = 0; |
397 | nreqs = 0; |
422 | nreqs = 0; |
398 | nready = 0; |
423 | nready = 0; |
399 | npending = 0; |
424 | npending = 0; |
400 | |
425 | |
401 | etp_atfork_parent (); |
426 | etp_thread_init (); |
402 | } |
427 | } |
403 | |
428 | |
404 | static void |
429 | static void |
405 | etp_once_init (void) |
430 | etp_once_init (void) |
406 | { |
431 | { |
|
|
432 | etp_thread_init (); |
407 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
433 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
408 | } |
434 | } |
409 | |
435 | |
410 | static int |
436 | static int |
411 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
437 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
… | |
… | |
588 | } |
614 | } |
589 | |
615 | |
590 | static void etp_set_max_poll_time (double nseconds) |
616 | static void etp_set_max_poll_time (double nseconds) |
591 | { |
617 | { |
592 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
618 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
593 | max_poll_time = nseconds; |
619 | max_poll_time = nseconds * EIO_TICKS; |
594 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
620 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
595 | } |
621 | } |
596 | |
622 | |
597 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
623 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
598 | { |
624 | { |
… | |
… | |
602 | } |
628 | } |
603 | |
629 | |
604 | static void etp_set_max_idle (unsigned int nthreads) |
630 | static void etp_set_max_idle (unsigned int nthreads) |
605 | { |
631 | { |
606 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
632 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
607 | max_idle = nthreads <= 0 ? 1 : nthreads; |
633 | max_idle = nthreads; |
|
|
634 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
|
|
635 | } |
|
|
636 | |
|
|
637 | static void etp_set_idle_timeout (unsigned int seconds) |
|
|
638 | { |
|
|
639 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
|
|
640 | idle_timeout = seconds; |
608 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
641 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
609 | } |
642 | } |
610 | |
643 | |
611 | static void etp_set_min_parallel (unsigned int nthreads) |
644 | static void etp_set_min_parallel (unsigned int nthreads) |
612 | { |
645 | { |
… | |
… | |
740 | void eio_set_max_idle (unsigned int nthreads) |
773 | void eio_set_max_idle (unsigned int nthreads) |
741 | { |
774 | { |
742 | etp_set_max_idle (nthreads); |
775 | etp_set_max_idle (nthreads); |
743 | } |
776 | } |
744 | |
777 | |
|
|
778 | void eio_set_idle_timeout (unsigned int seconds) |
|
|
779 | { |
|
|
780 | etp_set_idle_timeout (seconds); |
|
|
781 | } |
|
|
782 | |
745 | void eio_set_min_parallel (unsigned int nthreads) |
783 | void eio_set_min_parallel (unsigned int nthreads) |
746 | { |
784 | { |
747 | etp_set_min_parallel (nthreads); |
785 | etp_set_min_parallel (nthreads); |
748 | } |
786 | } |
749 | |
787 | |
… | |
… | |
797 | |
835 | |
798 | return res; |
836 | return res; |
799 | } |
837 | } |
800 | #endif |
838 | #endif |
801 | |
839 | |
802 | #ifndef HAVE_FUTIMES |
840 | #ifndef HAVE_UTIMES |
803 | |
841 | |
804 | # undef utimes |
842 | # undef utimes |
805 | # undef futimes |
|
|
806 | # define utimes(path,times) eio__utimes (path, times) |
843 | # define utimes(path,times) eio__utimes (path, times) |
807 | # define futimes(fd,times) eio__futimes (fd, times) |
|
|
808 | |
844 | |
809 | static int |
845 | static int |
810 | eio__utimes (const char *filename, const struct timeval times[2]) |
846 | eio__utimes (const char *filename, const struct timeval times[2]) |
811 | { |
847 | { |
812 | if (times) |
848 | if (times) |
… | |
… | |
819 | return utime (filename, &buf); |
855 | return utime (filename, &buf); |
820 | } |
856 | } |
821 | else |
857 | else |
822 | return utime (filename, 0); |
858 | return utime (filename, 0); |
823 | } |
859 | } |
|
|
860 | |
|
|
861 | #endif |
|
|
862 | |
|
|
863 | #ifndef HAVE_FUTIMES |
|
|
864 | |
|
|
865 | # undef futimes |
|
|
866 | # define futimes(fd,times) eio__futimes (fd, times) |
824 | |
867 | |
825 | static int eio__futimes (int fd, const struct timeval tv[2]) |
868 | static int eio__futimes (int fd, const struct timeval tv[2]) |
826 | { |
869 | { |
827 | errno = ENOSYS; |
870 | errno = ENOSYS; |
828 | return -1; |
871 | return -1; |
… | |
… | |
857 | if (!res || errno != ENOSYS) |
900 | if (!res || errno != ENOSYS) |
858 | return res; |
901 | return res; |
859 | #endif |
902 | #endif |
860 | |
903 | |
861 | /* even though we could play tricks with the flags, it's better to always |
904 | /* even though we could play tricks with the flags, it's better to always |
862 | * call fdatasync, as thta matches the expectation of it's users best */ |
905 | * call fdatasync, as that matches the expectation of its users best */ |
863 | return fdatasync (fd); |
906 | return fdatasync (fd); |
864 | } |
907 | } |
865 | |
908 | |
866 | #if !HAVE_READAHEAD |
909 | #if !HAVE_READAHEAD |
867 | # undef readahead |
910 | # undef readahead |
… | |
… | |
899 | |
942 | |
900 | #if HAVE_SENDFILE |
943 | #if HAVE_SENDFILE |
901 | # if __linux |
944 | # if __linux |
902 | res = sendfile (ofd, ifd, &offset, count); |
945 | res = sendfile (ofd, ifd, &offset, count); |
903 | |
946 | |
904 | # elif __freebsd |
947 | # elif __FreeBSD__ |
905 | /* |
948 | /* |
906 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
949 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
907 | * wasted on making it similar to other I/O functions. |
950 | * wasted on making it similar to other I/O functions. |
908 | */ |
951 | */ |
909 | { |
952 | { |
910 | off_t sbytes; |
953 | off_t sbytes; |
911 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
954 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
912 | |
955 | |
913 | if (res < 0 && sbytes) |
956 | #if 0 /* according to the manpage, this is correct, but broken behaviour */ |
914 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
957 | /* freebsd' sendfile will return 0 on success */ |
|
|
958 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
|
|
959 | /* not on e.g. EIO or EPIPE - sounds broken */ |
|
|
960 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
|
|
961 | res = sbytes; |
|
|
962 | #endif |
|
|
963 | |
|
|
964 | /* according to source inspection, this is correct, and useful behaviour */ |
|
|
965 | if (sbytes) |
|
|
966 | res = sbytes; |
|
|
967 | } |
|
|
968 | |
|
|
969 | # elif defined (__APPLE__) |
|
|
970 | |
|
|
971 | { |
|
|
972 | off_t sbytes = count; |
|
|
973 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
|
|
974 | |
|
|
975 | /* according to the manpage, sbytes is always valid */ |
|
|
976 | if (sbytes) |
915 | res = sbytes; |
977 | res = sbytes; |
916 | } |
978 | } |
917 | |
979 | |
918 | # elif __hpux |
980 | # elif __hpux |
919 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
981 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
… | |
… | |
933 | if (res < 0 && sbytes) |
995 | if (res < 0 && sbytes) |
934 | res = sbytes; |
996 | res = sbytes; |
935 | } |
997 | } |
936 | |
998 | |
937 | # endif |
999 | # endif |
|
|
1000 | |
|
|
1001 | #elif defined (_WIN32) |
|
|
1002 | |
|
|
1003 | /* does not work, just for documentation of what would need to be done */ |
|
|
1004 | { |
|
|
1005 | HANDLE h = TO_SOCKET (ifd); |
|
|
1006 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
|
|
1007 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
|
|
1008 | } |
|
|
1009 | |
938 | #else |
1010 | #else |
939 | res = -1; |
1011 | res = -1; |
940 | errno = ENOSYS; |
1012 | errno = ENOSYS; |
941 | #endif |
1013 | #endif |
942 | |
1014 | |
943 | if (res < 0 |
1015 | if (res < 0 |
944 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
1016 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
|
|
1017 | /* BSDs */ |
|
|
1018 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
|
|
1019 | || errno == ENOTSUP |
|
|
1020 | #endif |
|
|
1021 | || errno == EOPNOTSUPP /* BSDs */ |
945 | #if __solaris |
1022 | #if __solaris |
946 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
1023 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
947 | #endif |
1024 | #endif |
948 | ) |
1025 | ) |
949 | ) |
1026 | ) |
… | |
… | |
980 | } |
1057 | } |
981 | |
1058 | |
982 | return res; |
1059 | return res; |
983 | } |
1060 | } |
984 | |
1061 | |
985 | static int |
1062 | static signed char |
986 | eio_dent_cmp (const void *a_, const void *b_) |
1063 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
987 | { |
1064 | { |
988 | const eio_dirent *a = (const eio_dirent *)a_; |
1065 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
989 | const eio_dirent *b = (const eio_dirent *)b_; |
|
|
990 | |
|
|
991 | return (int)b->score - (int)a->score ? (int)b->score - (int)a->score |
|
|
992 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */ |
1066 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; |
|
|
1067 | } |
|
|
1068 | |
|
|
1069 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
|
|
1070 | |
|
|
1071 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
|
|
1072 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
|
|
1073 | |
|
|
1074 | static void |
|
|
1075 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1076 | { |
|
|
1077 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
|
|
1078 | unsigned char *bit = bits; |
|
|
1079 | |
|
|
1080 | assert (CHAR_BIT == 8); |
|
|
1081 | assert (sizeof (eio_dirent) * 8 < 256); |
|
|
1082 | assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ |
|
|
1083 | assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ |
|
|
1084 | |
|
|
1085 | if (size <= EIO_SORT_FAST) |
|
|
1086 | return; |
|
|
1087 | |
|
|
1088 | /* first prepare an array of bits to test in our radix sort */ |
|
|
1089 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
|
|
1090 | /* inode_bits must contain all inodes ORed together */ |
|
|
1091 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
|
|
1092 | { |
|
|
1093 | ino_t endianness; |
|
|
1094 | int i, j; |
|
|
1095 | |
|
|
1096 | /* we store the byte offset of byte n into byte n of "endianness" */ |
|
|
1097 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1098 | ((unsigned char *)&endianness)[i] = i; |
|
|
1099 | |
|
|
1100 | *bit++ = 0; |
|
|
1101 | |
|
|
1102 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1103 | { |
|
|
1104 | /* shifting off the byte offsets out of "endianness" */ |
|
|
1105 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
|
|
1106 | endianness >>= 8; |
|
|
1107 | |
|
|
1108 | for (j = 0; j < 8; ++j) |
|
|
1109 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
|
|
1110 | *bit++ = offs + j; |
|
|
1111 | } |
|
|
1112 | |
|
|
1113 | for (j = 0; j < 8; ++j) |
|
|
1114 | if (score_bits & (1 << j)) |
|
|
1115 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
|
|
1116 | } |
|
|
1117 | |
|
|
1118 | /* now actually do the sorting (a variant of MSD radix sort) */ |
|
|
1119 | { |
|
|
1120 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
|
|
1121 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
|
|
1122 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
|
|
1123 | int stk_idx = 0; |
|
|
1124 | |
|
|
1125 | base_stk [stk_idx] = dents; |
|
|
1126 | end_stk [stk_idx] = dents + size; |
|
|
1127 | bit_stk [stk_idx] = bit - 1; |
|
|
1128 | |
|
|
1129 | do |
|
|
1130 | { |
|
|
1131 | base = base_stk [stk_idx]; |
|
|
1132 | end = end_stk [stk_idx]; |
|
|
1133 | bit = bit_stk [stk_idx]; |
|
|
1134 | |
|
|
1135 | for (;;) |
|
|
1136 | { |
|
|
1137 | unsigned char O = *bit >> 3; |
|
|
1138 | unsigned char M = 1 << (*bit & 7); |
|
|
1139 | |
|
|
1140 | eio_dirent *a = base; |
|
|
1141 | eio_dirent *b = end; |
|
|
1142 | |
|
|
1143 | if (b - a < EIO_SORT_CUTOFF) |
|
|
1144 | break; |
|
|
1145 | |
|
|
1146 | /* now bit-partition the array on the bit */ |
|
|
1147 | /* this ugly asymmetric loop seems to perform much better than typical */ |
|
|
1148 | /* partition algos found in the literature */ |
|
|
1149 | do |
|
|
1150 | if (!(((unsigned char *)a)[O] & M)) |
|
|
1151 | ++a; |
|
|
1152 | else if (!(((unsigned char *)--b)[O] & M)) |
|
|
1153 | { |
|
|
1154 | eio_dirent tmp = *a; *a = *b; *b = tmp; |
|
|
1155 | ++a; |
|
|
1156 | } |
|
|
1157 | while (b > a); |
|
|
1158 | |
|
|
1159 | /* next bit, or stop, if no bits left in this path */ |
|
|
1160 | if (!*--bit) |
|
|
1161 | break; |
|
|
1162 | |
|
|
1163 | base_stk [stk_idx] = a; |
|
|
1164 | end_stk [stk_idx] = end; |
|
|
1165 | bit_stk [stk_idx] = bit; |
|
|
1166 | ++stk_idx; |
|
|
1167 | |
|
|
1168 | end = a; |
|
|
1169 | } |
|
|
1170 | } |
|
|
1171 | while (stk_idx--); |
|
|
1172 | } |
|
|
1173 | } |
|
|
1174 | |
|
|
1175 | static void |
|
|
1176 | eio_dent_insertion_sort (eio_dirent *dents, int size) |
|
|
1177 | { |
|
|
1178 | /* first move the smallest element to the front, to act as a sentinel */ |
|
|
1179 | { |
|
|
1180 | int i; |
|
|
1181 | eio_dirent *min = dents; |
|
|
1182 | |
|
|
1183 | /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
|
|
1184 | for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
|
|
1185 | if (EIO_DENT_CMP (dents [i], <, *min)) |
|
|
1186 | min = &dents [i]; |
|
|
1187 | |
|
|
1188 | /* swap elements 0 and j (minimum) */ |
|
|
1189 | { |
|
|
1190 | eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
|
|
1191 | } |
|
|
1192 | } |
|
|
1193 | |
|
|
1194 | /* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
|
|
1195 | { |
|
|
1196 | eio_dirent *i, *j; |
|
|
1197 | |
|
|
1198 | for (i = dents + 1; i < dents + size; ++i) |
|
|
1199 | { |
|
|
1200 | eio_dirent value = *i; |
|
|
1201 | |
|
|
1202 | for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
|
|
1203 | j [1] = j [0]; |
|
|
1204 | |
|
|
1205 | j [1] = value; |
|
|
1206 | } |
|
|
1207 | } |
|
|
1208 | } |
|
|
1209 | |
|
|
1210 | static void |
|
|
1211 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1212 | { |
|
|
1213 | if (size <= 1) |
|
|
1214 | return; /* our insertion sort relies on size > 0 */ |
|
|
1215 | |
|
|
1216 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
|
|
1217 | /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
|
|
1218 | eio_dent_radix_sort (dents, size, score_bits, inode_bits); |
|
|
1219 | |
|
|
1220 | /* use an insertion sort at the end, or for small arrays, */ |
|
|
1221 | /* as insertion sort is more efficient for small partitions */ |
|
|
1222 | eio_dent_insertion_sort (dents, size); |
993 | } |
1223 | } |
994 | |
1224 | |
995 | /* read a full directory */ |
1225 | /* read a full directory */ |
996 | static void |
1226 | static void |
997 | eio__scandir (eio_req *req, etp_worker *self) |
1227 | eio__scandir (eio_req *req, etp_worker *self) |
998 | { |
1228 | { |
999 | DIR *dirp; |
1229 | DIR *dirp; |
1000 | EIO_STRUCT_DIRENT *entp; |
1230 | EIO_STRUCT_DIRENT *entp; |
1001 | unsigned char *name, *names; |
1231 | char *name, *names; |
1002 | int namesalloc = 4096; |
1232 | int namesalloc = 4096; |
1003 | int namesoffs = 0; |
1233 | int namesoffs = 0; |
1004 | int flags = req->int1; |
1234 | int flags = req->int1; |
1005 | eio_dirent *dents = 0; |
1235 | eio_dirent *dents = 0; |
1006 | int dentalloc = 128; |
1236 | int dentalloc = 128; |
1007 | int dentoffs = 0; |
1237 | int dentoffs = 0; |
|
|
1238 | ino_t inode_bits = 0; |
1008 | |
1239 | |
1009 | req->result = -1; |
1240 | req->result = -1; |
1010 | |
1241 | |
1011 | if (!(flags & EIO_READDIR_DENTS)) |
1242 | if (!(flags & EIO_READDIR_DENTS)) |
1012 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1243 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1013 | |
1244 | |
1014 | X_LOCK (wrklock); |
1245 | X_LOCK (wrklock); |
1015 | |
|
|
1016 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1246 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1017 | self->dirp = dirp = opendir (req->ptr1); |
1247 | self->dirp = dirp = opendir (req->ptr1); |
1018 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1248 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1019 | req->ptr1 = names = malloc (namesalloc); |
|
|
1020 | req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1249 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1021 | |
1250 | req->ptr2 = names = malloc (namesalloc); |
1022 | X_UNLOCK (wrklock); |
1251 | X_UNLOCK (wrklock); |
1023 | |
1252 | |
1024 | if (dirp && names && (!flags || dents)) |
1253 | if (dirp && names && (!flags || dents)) |
1025 | for (;;) |
1254 | for (;;) |
1026 | { |
1255 | { |
… | |
… | |
1034 | |
1263 | |
1035 | /* sort etc. */ |
1264 | /* sort etc. */ |
1036 | req->int1 = flags; |
1265 | req->int1 = flags; |
1037 | req->result = dentoffs; |
1266 | req->result = dentoffs; |
1038 | |
1267 | |
1039 | if (flags & EIO_READDIR_STAT_ORDER || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)) |
1268 | if (flags & EIO_READDIR_STAT_ORDER) |
|
|
1269 | eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ |
|
|
1270 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1271 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
|
|
1272 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
|
|
1273 | else |
1040 | { |
1274 | { |
1041 | /* pray your qsort doesn't use quicksort */ |
|
|
1042 | qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */ |
|
|
1043 | } |
|
|
1044 | else if (flags & EIO_READDIR_DIRS_FIRST && !(flags & EIO_READDIR_FOUND_UNKNOWN)) |
|
|
1045 | { |
|
|
1046 | /* in this case, all is known, and we just put dirs first and sort them */ |
1275 | /* in this case, all is known, and we just put dirs first and sort them */ |
1047 | eio_dirent *ent = dents + dentoffs; |
1276 | eio_dirent *oth = dents + dentoffs; |
1048 | eio_dirent *dir = dents; |
1277 | eio_dirent *dir = dents; |
1049 | |
1278 | |
|
|
1279 | /* now partition dirs to the front, and non-dirs to the back */ |
|
|
1280 | /* by walking from both sides and swapping if necessary */ |
|
|
1281 | /* also clear score, so it doesn't influence sorting */ |
1050 | while (ent > dir) |
1282 | while (oth > dir) |
1051 | { |
1283 | { |
1052 | if (dir->type == DT_DIR) |
1284 | if (dir->type == EIO_DT_DIR) |
1053 | ++dir; |
1285 | ++dir; |
1054 | else |
|
|
1055 | { |
|
|
1056 | --ent; |
|
|
1057 | |
|
|
1058 | if (ent->type == DT_DIR) |
1286 | else if ((--oth)->type == EIO_DT_DIR) |
1059 | { |
1287 | { |
1060 | eio_dirent tmp = *dir; |
1288 | eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
1061 | *dir = *ent; |
|
|
1062 | *ent = tmp; |
|
|
1063 | |
1289 | |
1064 | ++dir; |
1290 | ++dir; |
1065 | } |
1291 | } |
1066 | } |
1292 | } |
1067 | } |
|
|
1068 | |
1293 | |
1069 | /* now sort the dirs only */ |
1294 | /* now sort the dirs only */ |
1070 | qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp); |
1295 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1071 | } |
1296 | } |
1072 | |
|
|
1073 | {int i; for(i=0;i<dentoffs;++i){eio_dirent *e=dents+i; printf ("%9ld %3d %s\n", e->inode,e->score,e->name);}}//D |
|
|
1074 | |
1297 | |
1075 | break; |
1298 | break; |
1076 | } |
1299 | } |
1077 | |
1300 | |
|
|
1301 | /* now add the entry to our list(s) */ |
1078 | name = entp->d_name; |
1302 | name = entp->d_name; |
1079 | |
1303 | |
|
|
1304 | /* skip . and .. entries */ |
1080 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1305 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1081 | { |
1306 | { |
1082 | int len = strlen (name) + 1; |
1307 | int len = D_NAMLEN (entp) + 1; |
1083 | |
1308 | |
1084 | while (expect_false (namesoffs + len > namesalloc)) |
1309 | while (expect_false (namesoffs + len > namesalloc)) |
1085 | { |
1310 | { |
1086 | namesalloc *= 2; |
1311 | namesalloc *= 2; |
1087 | X_LOCK (wrklock); |
1312 | X_LOCK (wrklock); |
1088 | req->ptr1 = names = realloc (names, namesalloc); |
1313 | req->ptr2 = names = realloc (names, namesalloc); |
1089 | X_UNLOCK (wrklock); |
1314 | X_UNLOCK (wrklock); |
1090 | |
1315 | |
1091 | if (!names) |
1316 | if (!names) |
1092 | break; |
1317 | break; |
1093 | } |
1318 | } |
… | |
… | |
1100 | |
1325 | |
1101 | if (expect_false (dentoffs == dentalloc)) |
1326 | if (expect_false (dentoffs == dentalloc)) |
1102 | { |
1327 | { |
1103 | dentalloc *= 2; |
1328 | dentalloc *= 2; |
1104 | X_LOCK (wrklock); |
1329 | X_LOCK (wrklock); |
1105 | req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1330 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1106 | X_UNLOCK (wrklock); |
1331 | X_UNLOCK (wrklock); |
1107 | |
1332 | |
1108 | if (!dents) |
1333 | if (!dents) |
1109 | break; |
1334 | break; |
1110 | } |
1335 | } |
1111 | |
1336 | |
1112 | ent = dents + dentoffs; |
1337 | ent = dents + dentoffs; |
1113 | |
1338 | |
1114 | ent->name = names + namesoffs; |
1339 | ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
1115 | ent->namelen = len - 1; |
1340 | ent->namelen = len - 1; |
1116 | ent->inode = D_INO (entp); |
1341 | ent->inode = D_INO (entp); |
|
|
1342 | |
|
|
1343 | inode_bits |= ent->inode; |
1117 | |
1344 | |
1118 | switch (D_TYPE (entp)) |
1345 | switch (D_TYPE (entp)) |
1119 | { |
1346 | { |
1120 | default: |
1347 | default: |
1121 | ent->type = EIO_DT_UNKNOWN; |
1348 | ent->type = EIO_DT_UNKNOWN; |
… | |
… | |
1126 | case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
1353 | case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
1127 | #endif |
1354 | #endif |
1128 | #ifdef DT_CHR |
1355 | #ifdef DT_CHR |
1129 | case DT_CHR: ent->type = EIO_DT_CHR; break; |
1356 | case DT_CHR: ent->type = EIO_DT_CHR; break; |
1130 | #endif |
1357 | #endif |
|
|
1358 | #ifdef DT_MPC |
|
|
1359 | case DT_MPC: ent->type = EIO_DT_MPC; break; |
|
|
1360 | #endif |
1131 | #ifdef DT_DIR |
1361 | #ifdef DT_DIR |
1132 | case DT_DIR: ent->type = EIO_DT_DIR; break; |
1362 | case DT_DIR: ent->type = EIO_DT_DIR; break; |
1133 | #endif |
1363 | #endif |
|
|
1364 | #ifdef DT_NAM |
|
|
1365 | case DT_NAM: ent->type = EIO_DT_NAM; break; |
|
|
1366 | #endif |
1134 | #ifdef DT_BLK |
1367 | #ifdef DT_BLK |
1135 | case DT_BLK: ent->type = EIO_DT_BLK; break; |
1368 | case DT_BLK: ent->type = EIO_DT_BLK; break; |
1136 | #endif |
1369 | #endif |
|
|
1370 | #ifdef DT_MPB |
|
|
1371 | case DT_MPB: ent->type = EIO_DT_MPB; break; |
|
|
1372 | #endif |
1137 | #ifdef DT_REG |
1373 | #ifdef DT_REG |
1138 | case DT_REG: ent->type = EIO_DT_REG; break; |
1374 | case DT_REG: ent->type = EIO_DT_REG; break; |
1139 | #endif |
1375 | #endif |
|
|
1376 | #ifdef DT_NWK |
|
|
1377 | case DT_NWK: ent->type = EIO_DT_NWK; break; |
|
|
1378 | #endif |
|
|
1379 | #ifdef DT_CMP |
|
|
1380 | case DT_CMP: ent->type = EIO_DT_CMP; break; |
|
|
1381 | #endif |
1140 | #ifdef DT_LNK |
1382 | #ifdef DT_LNK |
1141 | case DT_LNK: ent->type = EIO_DT_LNK; break; |
1383 | case DT_LNK: ent->type = EIO_DT_LNK; break; |
1142 | #endif |
1384 | #endif |
1143 | #ifdef DT_SOCK |
1385 | #ifdef DT_SOCK |
1144 | case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
1386 | case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
1145 | #endif |
1387 | #endif |
|
|
1388 | #ifdef DT_DOOR |
|
|
1389 | case DT_DOOR: ent->type = EIO_DT_DOOR; break; |
|
|
1390 | #endif |
1146 | #ifdef DT_WHT |
1391 | #ifdef DT_WHT |
1147 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1392 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
1148 | #endif |
1393 | #endif |
1149 | } |
1394 | } |
1150 | |
1395 | |
1151 | ent->score = 0; |
1396 | ent->score = 7; |
1152 | |
1397 | |
1153 | if (flags & EIO_READDIR_DIRS_FIRST) |
1398 | if (flags & EIO_READDIR_DIRS_FIRST) |
1154 | { |
1399 | { |
1155 | if (ent->type == EIO_DT_UNKNOWN) |
1400 | if (ent->type == EIO_DT_UNKNOWN) |
1156 | { |
1401 | { |
1157 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1402 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1158 | ent->score = 98; |
1403 | ent->score = 1; |
1159 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1404 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1160 | ent->score = len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ |
1405 | ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
1161 | } |
1406 | } |
1162 | else if (ent->type == DT_DIR) |
1407 | else if (ent->type == EIO_DT_DIR) |
1163 | ent->score = 100; |
1408 | ent->score = 0; |
1164 | } |
1409 | } |
1165 | } |
1410 | } |
1166 | |
1411 | |
1167 | namesoffs += len; |
1412 | namesoffs += len; |
1168 | ++dentoffs; |
1413 | ++dentoffs; |
1169 | } |
1414 | } |
|
|
1415 | |
|
|
1416 | if (EIO_CANCELLED (req)) |
|
|
1417 | { |
|
|
1418 | errno = ECANCELED; |
|
|
1419 | break; |
|
|
1420 | } |
1170 | } |
1421 | } |
1171 | else |
1422 | } |
1172 | req->result = -1; |
|
|
1173 | |
1423 | |
1174 | /* if user doesn't want the dents, do not provide it */ |
1424 | #ifdef PAGESIZE |
1175 | if (!(flags & EIO_READDIR_DENTS)) |
1425 | # define eio_pagesize() PAGESIZE |
|
|
1426 | #else |
|
|
1427 | static intptr_t |
|
|
1428 | eio_pagesize (void) |
|
|
1429 | { |
|
|
1430 | static intptr_t page; |
|
|
1431 | |
|
|
1432 | if (!page) |
|
|
1433 | page = sysconf (_SC_PAGESIZE); |
|
|
1434 | |
|
|
1435 | return page; |
|
|
1436 | } |
|
|
1437 | #endif |
|
|
1438 | |
|
|
1439 | static void |
|
|
1440 | eio_page_align (void **addr, size_t *length) |
|
|
1441 | { |
|
|
1442 | intptr_t mask = eio_pagesize () - 1; |
|
|
1443 | |
|
|
1444 | /* round down addr */ |
|
|
1445 | intptr_t adj = mask & (intptr_t)*addr; |
|
|
1446 | |
|
|
1447 | *addr = (void *)((intptr_t)*addr - adj); |
|
|
1448 | *length += adj; |
|
|
1449 | |
|
|
1450 | /* round up length */ |
|
|
1451 | *length = (*length + mask) & ~mask; |
|
|
1452 | } |
|
|
1453 | |
|
|
1454 | #if !_POSIX_MEMLOCK |
|
|
1455 | # define eio__mlockall(a) ((errno = ENOSYS), -1) |
|
|
1456 | #else |
|
|
1457 | |
|
|
1458 | static int |
|
|
1459 | eio__mlockall (int flags) |
|
|
1460 | { |
|
|
1461 | #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 |
|
|
1462 | extern int mallopt (int, int); |
|
|
1463 | mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ |
|
|
1464 | #endif |
|
|
1465 | |
|
|
1466 | if (EIO_MCL_CURRENT != MCL_CURRENT |
|
|
1467 | || EIO_MCL_FUTURE != MCL_FUTURE) |
1176 | { |
1468 | { |
1177 | X_LOCK (wrklock); |
1469 | flags = 0 |
1178 | free (dents); |
1470 | | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0) |
1179 | req->ptr2 = req->ptr1; |
1471 | | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0); |
1180 | req->ptr1 = 0; |
|
|
1181 | X_UNLOCK (wrklock); |
|
|
1182 | } |
1472 | } |
|
|
1473 | |
|
|
1474 | return mlockall (flags); |
1183 | } |
1475 | } |
|
|
1476 | #endif |
|
|
1477 | |
|
|
1478 | #if !_POSIX_MEMLOCK_RANGE |
|
|
1479 | # define eio__mlock(a,b) ((errno = ENOSYS), -1) |
|
|
1480 | #else |
|
|
1481 | |
|
|
1482 | static int |
|
|
1483 | eio__mlock (void *addr, size_t length) |
|
|
1484 | { |
|
|
1485 | eio_page_align (&addr, &length); |
|
|
1486 | |
|
|
1487 | return mlock (addr, length); |
|
|
1488 | } |
|
|
1489 | |
|
|
1490 | #endif |
1184 | |
1491 | |
1185 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1492 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1186 | # undef msync |
|
|
1187 | # define msync(a,b,c) ((errno = ENOSYS), -1) |
1493 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
|
|
1494 | #else |
|
|
1495 | |
|
|
1496 | int |
|
|
1497 | eio__msync (void *mem, size_t len, int flags) |
|
|
1498 | { |
|
|
1499 | eio_page_align (&mem, &len); |
|
|
1500 | |
|
|
1501 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1502 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1503 | || EIO_MS_SYNC != MS_SYNC) |
|
|
1504 | { |
|
|
1505 | flags = 0 |
|
|
1506 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
|
|
1507 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
|
|
1508 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
|
|
1509 | } |
|
|
1510 | |
|
|
1511 | return msync (mem, len, flags); |
|
|
1512 | } |
|
|
1513 | |
1188 | #endif |
1514 | #endif |
1189 | |
1515 | |
1190 | int |
1516 | int |
1191 | eio__mtouch (void *mem, size_t len, int flags) |
1517 | eio__mtouch (void *mem, size_t len, int flags) |
1192 | { |
1518 | { |
|
|
1519 | eio_page_align (&mem, &len); |
|
|
1520 | |
|
|
1521 | { |
1193 | intptr_t addr = (intptr_t)mem; |
1522 | intptr_t addr = (intptr_t)mem; |
1194 | intptr_t end = addr + len; |
1523 | intptr_t end = addr + len; |
1195 | #ifdef PAGESIZE |
1524 | intptr_t page = eio_pagesize (); |
1196 | const intptr_t page = PAGESIZE; |
|
|
1197 | #else |
|
|
1198 | static intptr_t page; |
|
|
1199 | |
1525 | |
1200 | if (!page) |
|
|
1201 | page = sysconf (_SC_PAGESIZE); |
|
|
1202 | #endif |
|
|
1203 | |
|
|
1204 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
|
|
1205 | |
|
|
1206 | if (addr < end) |
1526 | if (addr < end) |
1207 | if (flags) /* modify */ |
1527 | if (flags & EIO_MT_MODIFY) /* modify */ |
1208 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1528 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1209 | else |
1529 | else |
1210 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1530 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
|
|
1531 | } |
1211 | |
1532 | |
1212 | return 0; |
1533 | return 0; |
1213 | } |
1534 | } |
1214 | |
1535 | |
1215 | /*****************************************************************************/ |
1536 | /*****************************************************************************/ |
… | |
… | |
1249 | if (req) |
1570 | if (req) |
1250 | break; |
1571 | break; |
1251 | |
1572 | |
1252 | ++idle; |
1573 | ++idle; |
1253 | |
1574 | |
1254 | ts.tv_sec = time (0) + IDLE_TIMEOUT; |
1575 | ts.tv_sec = time (0) + idle_timeout; |
1255 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1576 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1256 | { |
1577 | { |
1257 | if (idle > max_idle) |
1578 | if (idle > max_idle) |
1258 | { |
1579 | { |
1259 | --idle; |
1580 | --idle; |
… | |
… | |
1338 | return 0; \ |
1659 | return 0; \ |
1339 | } |
1660 | } |
1340 | |
1661 | |
1341 | static void eio_execute (etp_worker *self, eio_req *req) |
1662 | static void eio_execute (etp_worker *self, eio_req *req) |
1342 | { |
1663 | { |
1343 | errno = 0; |
|
|
1344 | |
|
|
1345 | switch (req->type) |
1664 | switch (req->type) |
1346 | { |
1665 | { |
1347 | case EIO_READ: ALLOC (req->size); |
1666 | case EIO_READ: ALLOC (req->size); |
1348 | req->result = req->offs >= 0 |
1667 | req->result = req->offs >= 0 |
1349 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
1668 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
… | |
… | |
1360 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1679 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1361 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1680 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1362 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1681 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1363 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1682 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1364 | |
1683 | |
|
|
1684 | case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1685 | req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1686 | case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1687 | req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1688 | |
1365 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1689 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1366 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1690 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1367 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1691 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1368 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1692 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1369 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
1693 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
… | |
… | |
1376 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1700 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1377 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1701 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1378 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1702 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1379 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1703 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1380 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1704 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1381 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
1705 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; |
1382 | |
1706 | |
1383 | case EIO_READLINK: ALLOC (NAME_MAX); |
1707 | case EIO_READLINK: ALLOC (PATH_MAX); |
1384 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1708 | req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; |
1385 | |
1709 | |
1386 | case EIO_SYNC: req->result = 0; sync (); break; |
1710 | case EIO_SYNC: req->result = 0; sync (); break; |
1387 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1711 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1388 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1712 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1389 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
1713 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
1390 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1714 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
|
|
1715 | case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; |
|
|
1716 | case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; |
1391 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1717 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1392 | |
1718 | |
1393 | case EIO_READDIR: eio__scandir (req, self); break; |
1719 | case EIO_READDIR: eio__scandir (req, self); break; |
1394 | |
1720 | |
1395 | case EIO_BUSY: |
1721 | case EIO_BUSY: |
1396 | #ifdef _WIN32 |
1722 | #ifdef _WIN32 |
1397 | Sleep (req->nv1 * 1000.); |
1723 | Sleep (req->nv1 * 1e3); |
1398 | #else |
1724 | #else |
1399 | { |
1725 | { |
1400 | struct timeval tv; |
1726 | struct timeval tv; |
1401 | |
1727 | |
1402 | tv.tv_sec = req->nv1; |
1728 | tv.tv_sec = req->nv1; |
1403 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
1729 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6; |
1404 | |
1730 | |
1405 | req->result = select (0, 0, 0, 0, &tv); |
1731 | req->result = select (0, 0, 0, 0, &tv); |
1406 | } |
1732 | } |
1407 | #endif |
1733 | #endif |
1408 | break; |
1734 | break; |
… | |
… | |
1423 | times = tv; |
1749 | times = tv; |
1424 | } |
1750 | } |
1425 | else |
1751 | else |
1426 | times = 0; |
1752 | times = 0; |
1427 | |
1753 | |
1428 | |
|
|
1429 | req->result = req->type == EIO_FUTIME |
1754 | req->result = req->type == EIO_FUTIME |
1430 | ? futimes (req->int1, times) |
1755 | ? futimes (req->int1, times) |
1431 | : utimes (req->ptr1, times); |
1756 | : utimes (req->ptr1, times); |
1432 | } |
1757 | } |
1433 | break; |
1758 | break; |
… | |
… | |
1442 | case EIO_CUSTOM: |
1767 | case EIO_CUSTOM: |
1443 | ((void (*)(eio_req *))req->feed) (req); |
1768 | ((void (*)(eio_req *))req->feed) (req); |
1444 | break; |
1769 | break; |
1445 | |
1770 | |
1446 | default: |
1771 | default: |
|
|
1772 | errno = ENOSYS; |
1447 | req->result = -1; |
1773 | req->result = -1; |
1448 | break; |
1774 | break; |
1449 | } |
1775 | } |
1450 | |
1776 | |
1451 | req->errorno = errno; |
1777 | req->errorno = errno; |
… | |
… | |
1481 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1807 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1482 | { |
1808 | { |
1483 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1809 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1484 | } |
1810 | } |
1485 | |
1811 | |
|
|
1812 | eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data) |
|
|
1813 | { |
|
|
1814 | REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND; |
|
|
1815 | } |
|
|
1816 | |
|
|
1817 | eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) |
|
|
1818 | { |
|
|
1819 | REQ (EIO_MLOCKALL); req->int1 = flags; SEND; |
|
|
1820 | } |
|
|
1821 | |
1486 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
1822 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
1487 | { |
1823 | { |
1488 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
1824 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
1489 | } |
1825 | } |
1490 | |
1826 | |
… | |
… | |
1514 | } |
1850 | } |
1515 | |
1851 | |
1516 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1852 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1517 | { |
1853 | { |
1518 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
1854 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
|
|
1855 | } |
|
|
1856 | |
|
|
1857 | eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data) |
|
|
1858 | { |
|
|
1859 | REQ (EIO_FSTATVFS); req->int1 = fd; SEND; |
1519 | } |
1860 | } |
1520 | |
1861 | |
1521 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1862 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1522 | { |
1863 | { |
1523 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
1864 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
… | |
… | |
1597 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1938 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1598 | { |
1939 | { |
1599 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1940 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1600 | } |
1941 | } |
1601 | |
1942 | |
|
|
1943 | eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data) |
|
|
1944 | { |
|
|
1945 | return eio__1path (EIO_STATVFS, path, pri, cb, data); |
|
|
1946 | } |
|
|
1947 | |
1602 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1948 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1603 | { |
1949 | { |
1604 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1950 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1605 | } |
1951 | } |
1606 | |
1952 | |
… | |
… | |
1614 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1960 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1615 | } |
1961 | } |
1616 | |
1962 | |
1617 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1963 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1618 | { |
1964 | { |
1619 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
1965 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND; |
1620 | } |
1966 | } |
1621 | |
1967 | |
1622 | static eio_req * |
1968 | static eio_req * |
1623 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1969 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1624 | { |
1970 | { |
… | |
… | |
1708 | /* misc garbage */ |
2054 | /* misc garbage */ |
1709 | |
2055 | |
1710 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
2056 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1711 | { |
2057 | { |
1712 | etp_worker wrk; |
2058 | etp_worker wrk; |
|
|
2059 | ssize_t ret; |
1713 | |
2060 | |
1714 | wrk.dbuf = 0; |
2061 | wrk.dbuf = 0; |
1715 | |
2062 | |
1716 | eio__sendfile (ofd, ifd, offset, count, &wrk); |
2063 | ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
1717 | |
2064 | |
1718 | if (wrk.dbuf) |
2065 | if (wrk.dbuf) |
1719 | free (wrk.dbuf); |
2066 | free (wrk.dbuf); |
1720 | } |
|
|
1721 | |
2067 | |
|
|
2068 | return ret; |
|
|
2069 | } |
|
|
2070 | |