1 | /* |
1 | /* |
2 | * libeio implementation |
2 | * libeio implementation |
3 | * |
3 | * |
4 | * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> |
4 | * Copyright (c) 2007,2008,2009 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 | * |
… | |
… | |
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 | #include "eio.h" |
40 | #include "eio.h" |
|
|
41 | |
|
|
42 | #ifdef EIO_STACKSIZE |
|
|
43 | # define XTHREAD_STACKSIZE EIO_STACKSIZE |
|
|
44 | #endif |
41 | #include "xthread.h" |
45 | #include "xthread.h" |
42 | |
46 | |
43 | #include <errno.h> |
47 | #include <errno.h> |
44 | #include <stddef.h> |
48 | #include <stddef.h> |
45 | #include <stdlib.h> |
49 | #include <stdlib.h> |
… | |
… | |
64 | #endif |
68 | #endif |
65 | |
69 | |
66 | #ifdef _WIN32 |
70 | #ifdef _WIN32 |
67 | |
71 | |
68 | /*doh*/ |
72 | /*doh*/ |
69 | |
|
|
70 | #else |
73 | #else |
71 | |
74 | |
72 | # include "config.h" |
75 | # include "config.h" |
73 | # include <sys/time.h> |
76 | # include <sys/time.h> |
74 | # include <sys/select.h> |
77 | # include <sys/select.h> |
|
|
78 | # include <sys/mman.h> |
75 | # include <unistd.h> |
79 | # include <unistd.h> |
76 | # include <utime.h> |
80 | # include <utime.h> |
77 | # include <signal.h> |
81 | # include <signal.h> |
78 | # include <dirent.h> |
82 | # include <dirent.h> |
|
|
83 | |
|
|
84 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
|
|
85 | # if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) |
|
|
86 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
|
|
87 | # define D_INO(de) (de)->d_fileno |
|
|
88 | # define D_NAMLEN(de) (de)->d_namlen |
|
|
89 | # elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 |
|
|
90 | # define D_INO(de) (de)->d_ino |
|
|
91 | # endif |
|
|
92 | |
|
|
93 | #ifdef _D_EXACT_NAMLEN |
|
|
94 | # undef D_NAMLEN |
|
|
95 | # define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
|
|
96 | #endif |
|
|
97 | |
|
|
98 | # ifdef _DIRENT_HAVE_D_TYPE |
|
|
99 | # define D_TYPE(de) (de)->d_type |
|
|
100 | # endif |
79 | |
101 | |
80 | # ifndef EIO_STRUCT_DIRENT |
102 | # ifndef EIO_STRUCT_DIRENT |
81 | # define EIO_STRUCT_DIRENT struct dirent |
103 | # define EIO_STRUCT_DIRENT struct dirent |
82 | # endif |
104 | # endif |
83 | |
105 | |
… | |
… | |
96 | # else |
118 | # else |
97 | # error sendfile support requested but not available |
119 | # error sendfile support requested but not available |
98 | # endif |
120 | # endif |
99 | #endif |
121 | #endif |
100 | |
122 | |
|
|
123 | #ifndef D_TYPE |
|
|
124 | # define D_TYPE(de) 0 |
|
|
125 | #endif |
|
|
126 | #ifndef D_INO |
|
|
127 | # define D_INO(de) 0 |
|
|
128 | #endif |
|
|
129 | #ifndef D_NAMLEN |
|
|
130 | # define D_NAMLEN(de) strlen ((de)->d_name) |
|
|
131 | #endif |
|
|
132 | |
101 | /* number of seconds after which an idle threads exit */ |
133 | /* number of seconds after which an idle threads exit */ |
102 | #define IDLE_TIMEOUT 10 |
134 | #define IDLE_TIMEOUT 10 |
103 | |
135 | |
104 | /* used for struct dirent, AIX doesn't provide it */ |
136 | /* used for struct dirent, AIX doesn't provide it */ |
105 | #ifndef NAME_MAX |
137 | #ifndef NAME_MAX |
… | |
… | |
117 | errno = ENOMEM; \ |
149 | errno = ENOMEM; \ |
118 | if (!eio_buf) \ |
150 | if (!eio_buf) \ |
119 | return -1; |
151 | return -1; |
120 | |
152 | |
121 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
153 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
|
|
154 | |
|
|
155 | /*****************************************************************************/ |
|
|
156 | |
|
|
157 | #if __GNUC__ >= 3 |
|
|
158 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
159 | #else |
|
|
160 | # define expect(expr,value) (expr) |
|
|
161 | #endif |
|
|
162 | |
|
|
163 | #define expect_false(expr) expect ((expr) != 0, 0) |
|
|
164 | #define expect_true(expr) expect ((expr) != 0, 1) |
122 | |
165 | |
123 | /*****************************************************************************/ |
166 | /*****************************************************************************/ |
124 | |
167 | |
125 | #define ETP_PRI_MIN EIO_PRI_MIN |
168 | #define ETP_PRI_MIN EIO_PRI_MIN |
126 | #define ETP_PRI_MAX EIO_PRI_MAX |
169 | #define ETP_PRI_MAX EIO_PRI_MAX |
… | |
… | |
144 | if (wrk->dirp) \ |
187 | if (wrk->dirp) \ |
145 | { \ |
188 | { \ |
146 | closedir (wrk->dirp); \ |
189 | closedir (wrk->dirp); \ |
147 | wrk->dirp = 0; \ |
190 | wrk->dirp = 0; \ |
148 | } |
191 | } |
|
|
192 | |
149 | #define ETP_WORKER_COMMON \ |
193 | #define ETP_WORKER_COMMON \ |
150 | void *dbuf; \ |
194 | void *dbuf; \ |
151 | DIR *dirp; |
195 | DIR *dirp; |
152 | |
196 | |
153 | /*****************************************************************************/ |
197 | /*****************************************************************************/ |
… | |
… | |
411 | X_UNLOCK (wrklock); |
455 | X_UNLOCK (wrklock); |
412 | } |
456 | } |
413 | |
457 | |
414 | static void etp_maybe_start_thread (void) |
458 | static void etp_maybe_start_thread (void) |
415 | { |
459 | { |
416 | if (etp_nthreads () >= wanted) |
460 | if (expect_true (etp_nthreads () >= wanted)) |
417 | return; |
461 | return; |
418 | |
462 | |
419 | /* todo: maybe use idle here, but might be less exact */ |
463 | /* todo: maybe use idle here, but might be less exact */ |
420 | if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) |
464 | if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
421 | return; |
465 | return; |
422 | |
466 | |
423 | etp_start_thread (); |
467 | etp_start_thread (); |
424 | } |
468 | } |
425 | |
469 | |
… | |
… | |
478 | |
522 | |
479 | X_LOCK (reqlock); |
523 | X_LOCK (reqlock); |
480 | --nreqs; |
524 | --nreqs; |
481 | X_UNLOCK (reqlock); |
525 | X_UNLOCK (reqlock); |
482 | |
526 | |
483 | if (req->type == EIO_GROUP && req->size) |
527 | if (expect_false (req->type == EIO_GROUP && req->size)) |
484 | { |
528 | { |
485 | req->int1 = 1; /* mark request as delayed */ |
529 | req->int1 = 1; /* mark request as delayed */ |
486 | continue; |
530 | continue; |
487 | } |
531 | } |
488 | else |
532 | else |
489 | { |
533 | { |
490 | int res = ETP_FINISH (req); |
534 | int res = ETP_FINISH (req); |
491 | if (res) |
535 | if (expect_false (res)) |
492 | return res; |
536 | return res; |
493 | } |
537 | } |
494 | |
538 | |
495 | if (maxreqs && !--maxreqs) |
539 | if (expect_false (maxreqs && !--maxreqs)) |
496 | break; |
540 | break; |
497 | |
541 | |
498 | if (maxtime) |
542 | if (maxtime) |
499 | { |
543 | { |
500 | gettimeofday (&tv_now, 0); |
544 | gettimeofday (&tv_now, 0); |
… | |
… | |
519 | |
563 | |
520 | static void etp_submit (ETP_REQ *req) |
564 | static void etp_submit (ETP_REQ *req) |
521 | { |
565 | { |
522 | req->pri -= ETP_PRI_MIN; |
566 | req->pri -= ETP_PRI_MIN; |
523 | |
567 | |
524 | if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
568 | if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
525 | if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
569 | if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
526 | |
570 | |
|
|
571 | if (expect_false (req->type == EIO_GROUP)) |
|
|
572 | { |
|
|
573 | /* I hope this is worth it :/ */ |
527 | X_LOCK (reqlock); |
574 | X_LOCK (reqlock); |
528 | ++nreqs; |
575 | ++nreqs; |
|
|
576 | X_UNLOCK (reqlock); |
|
|
577 | |
|
|
578 | X_LOCK (reslock); |
|
|
579 | |
|
|
580 | ++npending; |
|
|
581 | |
|
|
582 | if (!reqq_push (&res_queue, req) && want_poll_cb) |
|
|
583 | want_poll_cb (); |
|
|
584 | |
|
|
585 | X_UNLOCK (reslock); |
|
|
586 | } |
|
|
587 | else |
|
|
588 | { |
|
|
589 | X_LOCK (reqlock); |
|
|
590 | ++nreqs; |
529 | ++nready; |
591 | ++nready; |
530 | reqq_push (&req_queue, req); |
592 | reqq_push (&req_queue, req); |
531 | X_COND_SIGNAL (reqwait); |
593 | X_COND_SIGNAL (reqwait); |
532 | X_UNLOCK (reqlock); |
594 | X_UNLOCK (reqlock); |
533 | |
595 | |
534 | etp_maybe_start_thread (); |
596 | etp_maybe_start_thread (); |
|
|
597 | } |
535 | } |
598 | } |
536 | |
599 | |
537 | static void etp_set_max_poll_time (double nseconds) |
600 | static void etp_set_max_poll_time (double nseconds) |
538 | { |
601 | { |
539 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
602 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
… | |
… | |
574 | |
637 | |
575 | static void grp_try_feed (eio_req *grp) |
638 | static void grp_try_feed (eio_req *grp) |
576 | { |
639 | { |
577 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
640 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
578 | { |
641 | { |
579 | int old_len = grp->size; |
642 | grp->flags &= ~EIO_FLAG_GROUPADD; |
580 | |
643 | |
581 | EIO_FEED (grp); |
644 | EIO_FEED (grp); |
582 | |
645 | |
583 | /* stop if no progress has been made */ |
646 | /* stop if no progress has been made */ |
584 | if (old_len == grp->size) |
647 | if (!(grp->flags & EIO_FLAG_GROUPADD)) |
585 | { |
648 | { |
586 | grp->feed = 0; |
649 | grp->feed = 0; |
587 | break; |
650 | break; |
588 | } |
651 | } |
589 | } |
652 | } |
… | |
… | |
706 | |
769 | |
707 | /*****************************************************************************/ |
770 | /*****************************************************************************/ |
708 | /* work around various missing functions */ |
771 | /* work around various missing functions */ |
709 | |
772 | |
710 | #if !HAVE_PREADWRITE |
773 | #if !HAVE_PREADWRITE |
|
|
774 | # undef pread |
|
|
775 | # undef pwrite |
711 | # define pread eio__pread |
776 | # define pread eio__pread |
712 | # define pwrite eio__pwrite |
777 | # define pwrite eio__pwrite |
713 | |
778 | |
714 | static ssize_t |
779 | static ssize_t |
715 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
780 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
… | |
… | |
735 | |
800 | |
736 | X_LOCK (preadwritelock); |
801 | X_LOCK (preadwritelock); |
737 | ooffset = lseek (fd, 0, SEEK_CUR); |
802 | ooffset = lseek (fd, 0, SEEK_CUR); |
738 | lseek (fd, offset, SEEK_SET); |
803 | lseek (fd, offset, SEEK_SET); |
739 | res = write (fd, buf, count); |
804 | res = write (fd, buf, count); |
740 | lseek (fd, offset, SEEK_SET); |
805 | lseek (fd, ooffset, SEEK_SET); |
741 | X_UNLOCK (preadwritelock); |
806 | X_UNLOCK (preadwritelock); |
742 | |
807 | |
743 | return res; |
808 | return res; |
744 | } |
809 | } |
745 | #endif |
810 | #endif |
746 | |
811 | |
747 | #ifndef HAVE_FUTIMES |
812 | #ifndef HAVE_FUTIMES |
748 | |
813 | |
|
|
814 | # undef utimes |
|
|
815 | # undef futimes |
749 | # define utimes(path,times) eio__utimes (path, times) |
816 | # define utimes(path,times) eio__utimes (path, times) |
750 | # define futimes(fd,times) eio__futimes (fd, times) |
817 | # define futimes(fd,times) eio__futimes (fd, times) |
751 | |
818 | |
752 | static int |
819 | static int |
753 | eio__utimes (const char *filename, const struct timeval times[2]) |
820 | eio__utimes (const char *filename, const struct timeval times[2]) |
… | |
… | |
772 | } |
839 | } |
773 | |
840 | |
774 | #endif |
841 | #endif |
775 | |
842 | |
776 | #if !HAVE_FDATASYNC |
843 | #if !HAVE_FDATASYNC |
|
|
844 | # undef fdatasync |
777 | # define fdatasync fsync |
845 | # define fdatasync(fd) fsync (fd) |
778 | #endif |
846 | #endif |
|
|
847 | |
|
|
848 | /* sync_file_range always needs emulation */ |
|
|
849 | int |
|
|
850 | eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
|
|
851 | { |
|
|
852 | #if HAVE_SYNC_FILE_RANGE |
|
|
853 | int res; |
|
|
854 | |
|
|
855 | if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE |
|
|
856 | || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE |
|
|
857 | || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER) |
|
|
858 | { |
|
|
859 | flags = 0 |
|
|
860 | | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0) |
|
|
861 | | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0) |
|
|
862 | | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); |
|
|
863 | } |
|
|
864 | |
|
|
865 | res = sync_file_range (fd, offset, nbytes, flags); |
|
|
866 | |
|
|
867 | if (!res || errno != ENOSYS) |
|
|
868 | return res; |
|
|
869 | #endif |
|
|
870 | |
|
|
871 | /* 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 */ |
|
|
873 | return fdatasync (fd); |
|
|
874 | } |
779 | |
875 | |
780 | #if !HAVE_READAHEAD |
876 | #if !HAVE_READAHEAD |
|
|
877 | # undef readahead |
781 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
878 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
782 | |
879 | |
783 | static ssize_t |
880 | static ssize_t |
784 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
881 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
785 | { |
882 | { |
… | |
… | |
893 | } |
990 | } |
894 | |
991 | |
895 | return res; |
992 | return res; |
896 | } |
993 | } |
897 | |
994 | |
|
|
995 | static signed char |
|
|
996 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
|
|
997 | { |
|
|
998 | return b->score - a->score ? b->score - a->score /* works because our signed char is always 0..100 */ |
|
|
999 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; |
|
|
1000 | } |
|
|
1001 | |
|
|
1002 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
|
|
1003 | |
|
|
1004 | #define EIO_QSORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
|
|
1005 | #define EIO_QSORT_SKIP 60 /* when to skip qsort completely */ |
|
|
1006 | |
|
|
1007 | static void |
|
|
1008 | eio_dent_sort (eio_dirent *dents, int size) |
|
|
1009 | { |
|
|
1010 | int i, j; |
|
|
1011 | |
|
|
1012 | if (size <= 1) |
|
|
1013 | return; /* our insertion sort relies on size > 0 */ |
|
|
1014 | |
|
|
1015 | if (size > EIO_QSORT_SKIP) /* skip quicksort for small directories */ |
|
|
1016 | { |
|
|
1017 | /* first, use quicksort */ |
|
|
1018 | /* should be good for 2**31 entries */ |
|
|
1019 | struct rng { int l, r; } rng [32]; |
|
|
1020 | |
|
|
1021 | i = 0; |
|
|
1022 | rng[0].l = 0; |
|
|
1023 | rng[0].r = size; |
|
|
1024 | |
|
|
1025 | while (expect_true (i >= 0)) |
|
|
1026 | { |
|
|
1027 | int L = rng [i].l; |
|
|
1028 | int R = rng [i].r - 1; |
|
|
1029 | |
|
|
1030 | if (expect_false (L + EIO_QSORT_CUTOFF < R)) |
|
|
1031 | { |
|
|
1032 | eio_dirent piv = dents [L]; |
|
|
1033 | |
|
|
1034 | while (L < R) |
|
|
1035 | { |
|
|
1036 | while (EIO_DENT_CMP (dents [R], >=, piv) && L < R) |
|
|
1037 | --R; |
|
|
1038 | |
|
|
1039 | if (L < R) |
|
|
1040 | dents [L++] = dents [R]; |
|
|
1041 | |
|
|
1042 | while (EIO_DENT_CMP (dents [L], <=, piv) && L < R) |
|
|
1043 | ++L; |
|
|
1044 | |
|
|
1045 | if (L < R) |
|
|
1046 | dents [R--] = dents [L]; |
|
|
1047 | } |
|
|
1048 | |
|
|
1049 | dents [L] = piv; |
|
|
1050 | |
|
|
1051 | ++i; |
|
|
1052 | rng [i].l = L + 1; |
|
|
1053 | rng [i].r = rng [i - 1].r; |
|
|
1054 | rng [i - 1].r = L; |
|
|
1055 | |
|
|
1056 | if (rng [i].r - rng [i].l > rng [i - 1].r - rng [i - 1].l) |
|
|
1057 | { |
|
|
1058 | struct rng t; |
|
|
1059 | |
|
|
1060 | t = rng [i]; rng [i] = rng [i - 1]; rng [i - 1] = t; |
|
|
1061 | } |
|
|
1062 | } |
|
|
1063 | else |
|
|
1064 | --i; |
|
|
1065 | } |
|
|
1066 | } |
|
|
1067 | |
|
|
1068 | /* use an insertion sort after qsort, or for small arrays */ |
|
|
1069 | /* first move the smallest element to the front, to act as a sentinel */ |
|
|
1070 | { |
|
|
1071 | int min = 0; |
|
|
1072 | |
|
|
1073 | for (i = size > EIO_QSORT_SKIP ? EIO_QSORT_CUTOFF : size; --i; ) |
|
|
1074 | if (EIO_DENT_CMP (dents [i], <, dents [min])) |
|
|
1075 | min = i; |
|
|
1076 | |
|
|
1077 | /* swap elements 0 and j (minimum) */ |
|
|
1078 | { |
|
|
1079 | eio_dirent tmp = dents [0]; dents [0] = dents [min]; dents [min] = tmp; |
|
|
1080 | } |
|
|
1081 | } |
|
|
1082 | |
|
|
1083 | /* then do standard insertion sort */ |
|
|
1084 | for (i = 1; i < size; ++i) |
|
|
1085 | { |
|
|
1086 | eio_dirent value = dents [i]; |
|
|
1087 | |
|
|
1088 | for (j = i - 1; EIO_DENT_CMP (dents [j], >, value); --j) |
|
|
1089 | dents [j + 1] = dents [j]; |
|
|
1090 | |
|
|
1091 | dents [j + 1] = value; |
|
|
1092 | } |
|
|
1093 | } |
|
|
1094 | |
898 | /* read a full directory */ |
1095 | /* read a full directory */ |
899 | static void |
1096 | static void |
900 | eio__scandir (eio_req *req, etp_worker *self) |
1097 | eio__scandir (eio_req *req, etp_worker *self) |
901 | { |
1098 | { |
902 | DIR *dirp; |
1099 | DIR *dirp; |
903 | EIO_STRUCT_DIRENT *entp; |
1100 | EIO_STRUCT_DIRENT *entp; |
904 | char *name, *names; |
1101 | char *name, *names; |
905 | int memlen = 4096; |
1102 | int namesalloc = 4096; |
906 | int memofs = 0; |
1103 | int namesoffs = 0; |
|
|
1104 | int flags = req->int1; |
|
|
1105 | eio_dirent *dents = 0; |
|
|
1106 | int dentalloc = 128; |
907 | int res = 0; |
1107 | int dentoffs = 0; |
|
|
1108 | |
|
|
1109 | req->result = -1; |
|
|
1110 | |
|
|
1111 | if (!(flags & EIO_READDIR_DENTS)) |
|
|
1112 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
908 | |
1113 | |
909 | X_LOCK (wrklock); |
1114 | X_LOCK (wrklock); |
910 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1115 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
911 | self->dirp = dirp = opendir (req->ptr1); |
1116 | self->dirp = dirp = opendir (req->ptr1); |
912 | req->flags |= EIO_FLAG_PTR2_FREE; |
1117 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
913 | req->ptr2 = names = malloc (memlen); |
1118 | req->ptr1 = names = malloc (namesalloc); |
|
|
1119 | req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
914 | X_UNLOCK (wrklock); |
1120 | X_UNLOCK (wrklock); |
915 | |
1121 | |
916 | if (dirp && names) |
1122 | if (dirp && names && (!flags || dents)) |
917 | for (;;) |
1123 | for (;;) |
918 | { |
1124 | { |
919 | errno = 0; |
1125 | errno = 0; |
920 | entp = readdir (dirp); |
1126 | entp = readdir (dirp); |
921 | |
1127 | |
922 | if (!entp) |
1128 | if (!entp) |
|
|
1129 | { |
|
|
1130 | if (errno) |
|
|
1131 | break; |
|
|
1132 | |
|
|
1133 | /* sort etc. */ |
|
|
1134 | req->int1 = flags; |
|
|
1135 | req->result = dentoffs; |
|
|
1136 | |
|
|
1137 | if (dents) |
|
|
1138 | { |
|
|
1139 | eio_dirent *ent = dents + dentoffs; |
|
|
1140 | |
|
|
1141 | while (ent > dents) |
|
|
1142 | (--ent)->name = names + (size_t)ent->name; |
|
|
1143 | } |
|
|
1144 | |
|
|
1145 | if (flags & EIO_READDIR_STAT_ORDER |
|
|
1146 | || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) |
|
|
1147 | eio_dent_sort (dents, dentoffs); /* score depends of DIRS_FIRST */ |
|
|
1148 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1149 | { |
|
|
1150 | /* in this case, all is known, and we just put dirs first and sort them */ |
|
|
1151 | eio_dirent *ent = dents + dentoffs; |
|
|
1152 | eio_dirent *dir = dents; |
|
|
1153 | |
|
|
1154 | /* now move dirs to the front, and non-dirs to the back */ |
|
|
1155 | /* by walking from both sides and swapping if necessary */ |
|
|
1156 | while (ent > dir) |
|
|
1157 | { |
|
|
1158 | if (dir->type == DT_DIR) |
|
|
1159 | ++dir; |
|
|
1160 | else |
|
|
1161 | { |
|
|
1162 | --ent; |
|
|
1163 | |
|
|
1164 | if (ent->type == DT_DIR) |
|
|
1165 | { |
|
|
1166 | eio_dirent tmp = *dir; |
|
|
1167 | *dir = *ent; |
|
|
1168 | *ent = tmp; |
|
|
1169 | |
|
|
1170 | ++dir; |
|
|
1171 | } |
|
|
1172 | } |
|
|
1173 | } |
|
|
1174 | |
|
|
1175 | /* now sort the dirs only */ |
|
|
1176 | eio_dent_sort (dents, dir - dents); |
|
|
1177 | } |
|
|
1178 | |
|
|
1179 | /* only provide the names array unless DENTS is specified */ |
|
|
1180 | if (!(flags & EIO_READDIR_DENTS)) |
|
|
1181 | { |
|
|
1182 | X_LOCK (wrklock); |
|
|
1183 | assert (!dents); |
|
|
1184 | req->ptr1 = 0; |
|
|
1185 | req->ptr2 = names; |
|
|
1186 | X_UNLOCK (wrklock); |
|
|
1187 | } |
|
|
1188 | |
923 | break; |
1189 | break; |
|
|
1190 | } |
924 | |
1191 | |
|
|
1192 | /* now add the entry to our list(s) */ |
925 | name = entp->d_name; |
1193 | name = entp->d_name; |
926 | |
1194 | |
|
|
1195 | /* skip . and .. entries */ |
927 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1196 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
928 | { |
1197 | { |
929 | int len = strlen (name) + 1; |
1198 | int len = D_NAMLEN (entp) + 1; |
930 | |
1199 | |
931 | res++; |
1200 | while (expect_false (namesoffs + len > namesalloc)) |
932 | |
|
|
933 | while (memofs + len > memlen) |
|
|
934 | { |
1201 | { |
935 | memlen *= 2; |
1202 | namesalloc *= 2; |
936 | X_LOCK (wrklock); |
1203 | X_LOCK (wrklock); |
937 | req->ptr2 = names = realloc (names, memlen); |
1204 | req->ptr1 = names = realloc (names, namesalloc); |
938 | X_UNLOCK (wrklock); |
1205 | X_UNLOCK (wrklock); |
939 | |
1206 | |
940 | if (!names) |
1207 | if (!names) |
941 | break; |
1208 | break; |
942 | } |
1209 | } |
943 | |
1210 | |
944 | memcpy (names + memofs, name, len); |
1211 | memcpy (names + namesoffs, name, len); |
|
|
1212 | |
|
|
1213 | if (dents) |
|
|
1214 | { |
|
|
1215 | struct eio_dirent *ent; |
|
|
1216 | |
|
|
1217 | if (expect_false (dentoffs == dentalloc)) |
|
|
1218 | { |
|
|
1219 | dentalloc *= 2; |
|
|
1220 | X_LOCK (wrklock); |
|
|
1221 | req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
|
|
1222 | X_UNLOCK (wrklock); |
|
|
1223 | |
|
|
1224 | if (!dents) |
|
|
1225 | break; |
|
|
1226 | } |
|
|
1227 | |
|
|
1228 | ent = dents + dentoffs; |
|
|
1229 | |
|
|
1230 | ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ |
|
|
1231 | ent->namelen = len - 1; |
|
|
1232 | ent->inode = D_INO (entp); |
|
|
1233 | |
|
|
1234 | switch (D_TYPE (entp)) |
|
|
1235 | { |
|
|
1236 | default: |
|
|
1237 | ent->type = EIO_DT_UNKNOWN; |
|
|
1238 | flags |= EIO_READDIR_FOUND_UNKNOWN; |
|
|
1239 | break; |
|
|
1240 | |
|
|
1241 | #ifdef DT_FIFO |
|
|
1242 | case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
|
|
1243 | #endif |
|
|
1244 | #ifdef DT_CHR |
|
|
1245 | case DT_CHR: ent->type = EIO_DT_CHR; break; |
|
|
1246 | #endif |
|
|
1247 | #ifdef DT_MPC |
|
|
1248 | case DT_MPC: ent->type = EIO_DT_MPC; break; |
|
|
1249 | #endif |
|
|
1250 | #ifdef DT_DIR |
|
|
1251 | case DT_DIR: ent->type = EIO_DT_DIR; break; |
|
|
1252 | #endif |
|
|
1253 | #ifdef DT_NAM |
|
|
1254 | case DT_NAM: ent->type = EIO_DT_NAM; break; |
|
|
1255 | #endif |
|
|
1256 | #ifdef DT_BLK |
|
|
1257 | case DT_BLK: ent->type = EIO_DT_BLK; break; |
|
|
1258 | #endif |
|
|
1259 | #ifdef DT_MPB |
|
|
1260 | case DT_MPB: ent->type = EIO_DT_MPB; break; |
|
|
1261 | #endif |
|
|
1262 | #ifdef DT_REG |
|
|
1263 | case DT_REG: ent->type = EIO_DT_REG; break; |
|
|
1264 | #endif |
|
|
1265 | #ifdef DT_NWK |
|
|
1266 | case DT_NWK: ent->type = EIO_DT_NWK; break; |
|
|
1267 | #endif |
|
|
1268 | #ifdef DT_CMP |
|
|
1269 | case DT_CMP: ent->type = EIO_DT_CMP; break; |
|
|
1270 | #endif |
|
|
1271 | #ifdef DT_LNK |
|
|
1272 | case DT_LNK: ent->type = EIO_DT_LNK; break; |
|
|
1273 | #endif |
|
|
1274 | #ifdef DT_SOCK |
|
|
1275 | case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
|
|
1276 | #endif |
|
|
1277 | #ifdef DT_DOOR |
|
|
1278 | case DT_DOOR: ent->type = EIO_DT_DOOR; break; |
|
|
1279 | #endif |
|
|
1280 | #ifdef DT_WHT |
|
|
1281 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
|
|
1282 | #endif |
|
|
1283 | } |
|
|
1284 | |
|
|
1285 | ent->score = 0; |
|
|
1286 | |
|
|
1287 | if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1288 | { |
|
|
1289 | if (ent->type == EIO_DT_UNKNOWN) |
|
|
1290 | { |
|
|
1291 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
|
|
1292 | ent->score = 98; |
|
|
1293 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
|
|
1294 | ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ |
|
|
1295 | } |
|
|
1296 | else if (ent->type == EIO_DT_DIR) |
|
|
1297 | ent->score = 100; |
|
|
1298 | } |
|
|
1299 | } |
|
|
1300 | |
945 | memofs += len; |
1301 | namesoffs += len; |
|
|
1302 | ++dentoffs; |
|
|
1303 | } |
|
|
1304 | |
|
|
1305 | if (EIO_CANCELLED (req)) |
|
|
1306 | { |
|
|
1307 | errno = ECANCELED; |
|
|
1308 | break; |
946 | } |
1309 | } |
947 | } |
1310 | } |
|
|
1311 | } |
948 | |
1312 | |
949 | if (errno) |
1313 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
950 | res = -1; |
1314 | # undef msync |
951 | |
1315 | # define msync(a,b,c) ((errno = ENOSYS), -1) |
952 | req->result = res; |
1316 | #endif |
|
|
1317 | |
|
|
1318 | int |
|
|
1319 | eio__mtouch (void *mem, size_t len, int flags) |
|
|
1320 | { |
|
|
1321 | intptr_t addr = (intptr_t)mem; |
|
|
1322 | intptr_t end = addr + len; |
|
|
1323 | #ifdef PAGESIZE |
|
|
1324 | const intptr_t page = PAGESIZE; |
|
|
1325 | #else |
|
|
1326 | static intptr_t page; |
|
|
1327 | |
|
|
1328 | if (!page) |
|
|
1329 | page = sysconf (_SC_PAGESIZE); |
|
|
1330 | #endif |
|
|
1331 | |
|
|
1332 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
|
|
1333 | |
|
|
1334 | if (addr < end) |
|
|
1335 | if (flags) /* modify */ |
|
|
1336 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
|
|
1337 | else |
|
|
1338 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
|
|
1339 | |
|
|
1340 | return 0; |
953 | } |
1341 | } |
954 | |
1342 | |
955 | /*****************************************************************************/ |
1343 | /*****************************************************************************/ |
956 | |
1344 | |
957 | #define ALLOC(len) \ |
1345 | #define ALLOC(len) \ |
… | |
… | |
1124 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1512 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1125 | |
1513 | |
1126 | case EIO_SYNC: req->result = 0; sync (); break; |
1514 | case EIO_SYNC: req->result = 0; sync (); break; |
1127 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1515 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1128 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1516 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
|
|
1517 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
|
|
1518 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
|
|
1519 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1129 | |
1520 | |
1130 | case EIO_READDIR: eio__scandir (req, self); break; |
1521 | case EIO_READDIR: eio__scandir (req, self); break; |
1131 | |
1522 | |
1132 | case EIO_BUSY: |
1523 | case EIO_BUSY: |
1133 | #ifdef _WIN32 |
1524 | #ifdef _WIN32 |
… | |
… | |
1165 | |
1556 | |
1166 | req->result = req->type == EIO_FUTIME |
1557 | req->result = req->type == EIO_FUTIME |
1167 | ? futimes (req->int1, times) |
1558 | ? futimes (req->int1, times) |
1168 | : utimes (req->ptr1, times); |
1559 | : utimes (req->ptr1, times); |
1169 | } |
1560 | } |
|
|
1561 | break; |
1170 | |
1562 | |
1171 | case EIO_GROUP: |
1563 | case EIO_GROUP: |
|
|
1564 | abort (); /* handled in eio_request */ |
|
|
1565 | |
1172 | case EIO_NOP: |
1566 | case EIO_NOP: |
1173 | req->result = 0; |
1567 | req->result = 0; |
1174 | break; |
1568 | break; |
1175 | |
1569 | |
1176 | case EIO_CUSTOM: |
1570 | case EIO_CUSTOM: |
1177 | req->feed (req); |
1571 | ((void (*)(eio_req *))req->feed) (req); |
1178 | break; |
1572 | break; |
1179 | |
1573 | |
1180 | default: |
1574 | default: |
1181 | req->result = -1; |
1575 | req->result = -1; |
1182 | break; |
1576 | break; |
… | |
… | |
1205 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1599 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1206 | { |
1600 | { |
1207 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
1601 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
1208 | } |
1602 | } |
1209 | |
1603 | |
|
|
1604 | eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
1605 | { |
|
|
1606 | REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
|
|
1607 | } |
|
|
1608 | |
|
|
1609 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
1610 | { |
|
|
1611 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
|
|
1612 | } |
|
|
1613 | |
|
|
1614 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
|
|
1615 | { |
|
|
1616 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
|
|
1617 | } |
|
|
1618 | |
1210 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1619 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1211 | { |
1620 | { |
1212 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1621 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1213 | } |
1622 | } |
1214 | |
1623 | |
… | |
… | |
1326 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
1735 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
1327 | { |
1736 | { |
1328 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
1737 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
1329 | } |
1738 | } |
1330 | |
1739 | |
1331 | eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) |
1740 | eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data) |
1332 | { |
1741 | { |
1333 | return eio__1path (EIO_READDIR, path, pri, cb, data); |
1742 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1334 | } |
1743 | } |
1335 | |
1744 | |
1336 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1745 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1337 | { |
1746 | { |
1338 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
1747 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
… | |
… | |
1369 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1778 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1370 | } |
1779 | } |
1371 | |
1780 | |
1372 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
1781 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
1373 | { |
1782 | { |
1374 | REQ (EIO_CUSTOM); req->feed = execute; SEND; |
1783 | REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; |
1375 | } |
1784 | } |
1376 | |
1785 | |
1377 | #endif |
1786 | #endif |
1378 | |
1787 | |
1379 | eio_req *eio_grp (eio_cb cb, void *data) |
1788 | eio_req *eio_grp (eio_cb cb, void *data) |
… | |
… | |
1407 | |
1816 | |
1408 | void eio_grp_add (eio_req *grp, eio_req *req) |
1817 | void eio_grp_add (eio_req *grp, eio_req *req) |
1409 | { |
1818 | { |
1410 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
1819 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
1411 | |
1820 | |
|
|
1821 | grp->flags |= EIO_FLAG_GROUPADD; |
|
|
1822 | |
1412 | ++grp->size; |
1823 | ++grp->size; |
1413 | req->grp = grp; |
1824 | req->grp = grp; |
1414 | |
1825 | |
1415 | req->grp_prev = 0; |
1826 | req->grp_prev = 0; |
1416 | req->grp_next = grp->grp_first; |
1827 | req->grp_next = grp->grp_first; |
… | |
… | |
1425 | /* misc garbage */ |
1836 | /* misc garbage */ |
1426 | |
1837 | |
1427 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1838 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1428 | { |
1839 | { |
1429 | etp_worker wrk; |
1840 | etp_worker wrk; |
|
|
1841 | ssize_t ret; |
1430 | |
1842 | |
1431 | wrk.dbuf = 0; |
1843 | wrk.dbuf = 0; |
1432 | |
1844 | |
1433 | eio__sendfile (ofd, ifd, offset, count, &wrk); |
1845 | ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
1434 | |
1846 | |
1435 | if (wrk.dbuf) |
1847 | if (wrk.dbuf) |
1436 | free (wrk.dbuf); |
1848 | free (wrk.dbuf); |
1437 | } |
|
|
1438 | |
1849 | |
|
|
1850 | return ret; |
|
|
1851 | } |
|
|
1852 | |