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> |
75 | # include <unistd.h> |
78 | # include <unistd.h> |
76 | # include <utime.h> |
79 | # include <utime.h> |
77 | # include <signal.h> |
80 | # include <signal.h> |
78 | # include <dirent.h> |
81 | # include <dirent.h> |
|
|
82 | |
|
|
83 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
|
|
84 | # if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) |
|
|
85 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
|
|
86 | # define D_INO(de) (de)->d_fileno |
|
|
87 | # define D_NAMLEN(de) (de)->d_namlen |
|
|
88 | # elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 |
|
|
89 | # define D_INO(de) (de)->d_ino |
|
|
90 | # endif |
|
|
91 | |
|
|
92 | #ifdef _D_EXACT_NAMLEN |
|
|
93 | # undef D_NAMLEN |
|
|
94 | # define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
|
|
95 | #endif |
|
|
96 | |
|
|
97 | # ifdef _DIRENT_HAVE_D_TYPE |
|
|
98 | # define D_TYPE(de) (de)->d_type |
|
|
99 | # endif |
79 | |
100 | |
80 | # ifndef EIO_STRUCT_DIRENT |
101 | # ifndef EIO_STRUCT_DIRENT |
81 | # define EIO_STRUCT_DIRENT struct dirent |
102 | # define EIO_STRUCT_DIRENT struct dirent |
82 | # endif |
103 | # endif |
83 | |
104 | |
… | |
… | |
96 | # else |
117 | # else |
97 | # error sendfile support requested but not available |
118 | # error sendfile support requested but not available |
98 | # endif |
119 | # endif |
99 | #endif |
120 | #endif |
100 | |
121 | |
|
|
122 | #ifndef D_TYPE |
|
|
123 | # define D_TYPE(de) 0 |
|
|
124 | #endif |
|
|
125 | #ifndef D_INO |
|
|
126 | # define D_INO(de) 0 |
|
|
127 | #endif |
|
|
128 | #ifndef D_NAMLEN |
|
|
129 | # define D_NAMLEN(de) strlen ((de)->d_name) |
|
|
130 | #endif |
|
|
131 | |
101 | /* number of seconds after which an idle threads exit */ |
132 | /* number of seconds after which an idle threads exit */ |
102 | #define IDLE_TIMEOUT 10 |
133 | #define IDLE_TIMEOUT 10 |
103 | |
134 | |
104 | /* used for struct dirent, AIX doesn't provide it */ |
135 | /* used for struct dirent, AIX doesn't provide it */ |
105 | #ifndef NAME_MAX |
136 | #ifndef NAME_MAX |
… | |
… | |
117 | errno = ENOMEM; \ |
148 | errno = ENOMEM; \ |
118 | if (!eio_buf) \ |
149 | if (!eio_buf) \ |
119 | return -1; |
150 | return -1; |
120 | |
151 | |
121 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
152 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
|
|
153 | |
|
|
154 | /*****************************************************************************/ |
|
|
155 | |
|
|
156 | #if __GNUC__ >= 3 |
|
|
157 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
158 | #else |
|
|
159 | # define expect(expr,value) (expr) |
|
|
160 | #endif |
|
|
161 | |
|
|
162 | #define expect_false(expr) expect ((expr) != 0, 0) |
|
|
163 | #define expect_true(expr) expect ((expr) != 0, 1) |
122 | |
164 | |
123 | /*****************************************************************************/ |
165 | /*****************************************************************************/ |
124 | |
166 | |
125 | #define ETP_PRI_MIN EIO_PRI_MIN |
167 | #define ETP_PRI_MIN EIO_PRI_MIN |
126 | #define ETP_PRI_MAX EIO_PRI_MAX |
168 | #define ETP_PRI_MAX EIO_PRI_MAX |
… | |
… | |
144 | if (wrk->dirp) \ |
186 | if (wrk->dirp) \ |
145 | { \ |
187 | { \ |
146 | closedir (wrk->dirp); \ |
188 | closedir (wrk->dirp); \ |
147 | wrk->dirp = 0; \ |
189 | wrk->dirp = 0; \ |
148 | } |
190 | } |
|
|
191 | |
149 | #define ETP_WORKER_COMMON \ |
192 | #define ETP_WORKER_COMMON \ |
150 | void *dbuf; \ |
193 | void *dbuf; \ |
151 | DIR *dirp; |
194 | DIR *dirp; |
152 | |
195 | |
153 | /*****************************************************************************/ |
196 | /*****************************************************************************/ |
… | |
… | |
176 | |
219 | |
177 | static mutex_t wrklock = X_MUTEX_INIT; |
220 | static mutex_t wrklock = X_MUTEX_INIT; |
178 | static mutex_t reslock = X_MUTEX_INIT; |
221 | static mutex_t reslock = X_MUTEX_INIT; |
179 | static mutex_t reqlock = X_MUTEX_INIT; |
222 | static mutex_t reqlock = X_MUTEX_INIT; |
180 | static cond_t reqwait = X_COND_INIT; |
223 | static cond_t reqwait = X_COND_INIT; |
|
|
224 | |
|
|
225 | #if !HAVE_PREADWRITE |
|
|
226 | /* |
|
|
227 | * make our pread/pwrite emulation safe against themselves, but not against |
|
|
228 | * normal read/write by using a mutex. slows down execution a lot, |
|
|
229 | * but that's your problem, not mine. |
|
|
230 | */ |
|
|
231 | static mutex_t preadwritelock = X_MUTEX_INIT; |
|
|
232 | #endif |
181 | |
233 | |
182 | typedef struct etp_worker |
234 | typedef struct etp_worker |
183 | { |
235 | { |
184 | /* locked by wrklock */ |
236 | /* locked by wrklock */ |
185 | struct etp_worker *prev, *next; |
237 | struct etp_worker *prev, *next; |
… | |
… | |
330 | |
382 | |
331 | static void etp_atfork_child (void) |
383 | static void etp_atfork_child (void) |
332 | { |
384 | { |
333 | ETP_REQ *prv; |
385 | ETP_REQ *prv; |
334 | |
386 | |
335 | while (prv = reqq_shift (&req_queue)) |
387 | while ((prv = reqq_shift (&req_queue))) |
336 | ETP_DESTROY (prv); |
388 | ETP_DESTROY (prv); |
337 | |
389 | |
338 | while (prv = reqq_shift (&res_queue)) |
390 | while ((prv = reqq_shift (&res_queue))) |
339 | ETP_DESTROY (prv); |
391 | ETP_DESTROY (prv); |
340 | |
392 | |
341 | while (wrk_first.next != &wrk_first) |
393 | while (wrk_first.next != &wrk_first) |
342 | { |
394 | { |
343 | etp_worker *wrk = wrk_first.next; |
395 | etp_worker *wrk = wrk_first.next; |
… | |
… | |
402 | X_UNLOCK (wrklock); |
454 | X_UNLOCK (wrklock); |
403 | } |
455 | } |
404 | |
456 | |
405 | static void etp_maybe_start_thread (void) |
457 | static void etp_maybe_start_thread (void) |
406 | { |
458 | { |
407 | if (etp_nthreads () >= wanted) |
459 | if (expect_true (etp_nthreads () >= wanted)) |
408 | return; |
460 | return; |
409 | |
461 | |
410 | /* todo: maybe use idle here, but might be less exact */ |
462 | /* todo: maybe use idle here, but might be less exact */ |
411 | if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) |
463 | if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
412 | return; |
464 | return; |
413 | |
465 | |
414 | etp_start_thread (); |
466 | etp_start_thread (); |
415 | } |
467 | } |
416 | |
468 | |
… | |
… | |
469 | |
521 | |
470 | X_LOCK (reqlock); |
522 | X_LOCK (reqlock); |
471 | --nreqs; |
523 | --nreqs; |
472 | X_UNLOCK (reqlock); |
524 | X_UNLOCK (reqlock); |
473 | |
525 | |
474 | if (req->type == EIO_GROUP && req->size) |
526 | if (expect_false (req->type == EIO_GROUP && req->size)) |
475 | { |
527 | { |
476 | req->int1 = 1; /* mark request as delayed */ |
528 | req->int1 = 1; /* mark request as delayed */ |
477 | continue; |
529 | continue; |
478 | } |
530 | } |
479 | else |
531 | else |
480 | { |
532 | { |
481 | int res = ETP_FINISH (req); |
533 | int res = ETP_FINISH (req); |
482 | if (res) |
534 | if (expect_false (res)) |
483 | return res; |
535 | return res; |
484 | } |
536 | } |
485 | |
537 | |
486 | if (maxreqs && !--maxreqs) |
538 | if (expect_false (maxreqs && !--maxreqs)) |
487 | break; |
539 | break; |
488 | |
540 | |
489 | if (maxtime) |
541 | if (maxtime) |
490 | { |
542 | { |
491 | gettimeofday (&tv_now, 0); |
543 | gettimeofday (&tv_now, 0); |
… | |
… | |
510 | |
562 | |
511 | static void etp_submit (ETP_REQ *req) |
563 | static void etp_submit (ETP_REQ *req) |
512 | { |
564 | { |
513 | req->pri -= ETP_PRI_MIN; |
565 | req->pri -= ETP_PRI_MIN; |
514 | |
566 | |
515 | if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
567 | if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
516 | if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
568 | if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
517 | |
569 | |
|
|
570 | if (expect_false (req->type == EIO_GROUP)) |
|
|
571 | { |
|
|
572 | /* I hope this is worth it :/ */ |
518 | X_LOCK (reqlock); |
573 | X_LOCK (reqlock); |
519 | ++nreqs; |
574 | ++nreqs; |
|
|
575 | X_UNLOCK (reqlock); |
|
|
576 | |
|
|
577 | X_LOCK (reslock); |
|
|
578 | |
|
|
579 | ++npending; |
|
|
580 | |
|
|
581 | if (!reqq_push (&res_queue, req) && want_poll_cb) |
|
|
582 | want_poll_cb (); |
|
|
583 | |
|
|
584 | X_UNLOCK (reslock); |
|
|
585 | } |
|
|
586 | else |
|
|
587 | { |
|
|
588 | X_LOCK (reqlock); |
|
|
589 | ++nreqs; |
520 | ++nready; |
590 | ++nready; |
521 | reqq_push (&req_queue, req); |
591 | reqq_push (&req_queue, req); |
522 | X_COND_SIGNAL (reqwait); |
592 | X_COND_SIGNAL (reqwait); |
523 | X_UNLOCK (reqlock); |
593 | X_UNLOCK (reqlock); |
524 | |
594 | |
525 | etp_maybe_start_thread (); |
595 | etp_maybe_start_thread (); |
|
|
596 | } |
526 | } |
597 | } |
527 | |
598 | |
528 | static void etp_set_max_poll_time (double nseconds) |
599 | static void etp_set_max_poll_time (double nseconds) |
529 | { |
600 | { |
530 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
601 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
… | |
… | |
565 | |
636 | |
566 | static void grp_try_feed (eio_req *grp) |
637 | static void grp_try_feed (eio_req *grp) |
567 | { |
638 | { |
568 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
639 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
569 | { |
640 | { |
570 | int old_len = grp->size; |
641 | grp->flags &= ~EIO_FLAG_GROUPADD; |
571 | |
642 | |
572 | EIO_FEED (grp); |
643 | EIO_FEED (grp); |
573 | |
644 | |
574 | /* stop if no progress has been made */ |
645 | /* stop if no progress has been made */ |
575 | if (old_len == grp->size) |
646 | if (!(grp->flags & EIO_FLAG_GROUPADD)) |
576 | { |
647 | { |
577 | grp->feed = 0; |
648 | grp->feed = 0; |
578 | break; |
649 | break; |
579 | } |
650 | } |
580 | } |
651 | } |
… | |
… | |
697 | |
768 | |
698 | /*****************************************************************************/ |
769 | /*****************************************************************************/ |
699 | /* work around various missing functions */ |
770 | /* work around various missing functions */ |
700 | |
771 | |
701 | #if !HAVE_PREADWRITE |
772 | #if !HAVE_PREADWRITE |
|
|
773 | # undef pread |
|
|
774 | # undef pwrite |
702 | # define pread eio__pread |
775 | # define pread eio__pread |
703 | # define pwrite eio__pwrite |
776 | # define pwrite eio__pwrite |
704 | |
|
|
705 | /* |
|
|
706 | * make our pread/pwrite safe against themselves, but not against |
|
|
707 | * normal read/write by using a mutex. slows down execution a lot, |
|
|
708 | * but that's your problem, not mine. |
|
|
709 | */ |
|
|
710 | static mutex_t preadwritelock = X_MUTEX_INIT; |
|
|
711 | |
777 | |
712 | static ssize_t |
778 | static ssize_t |
713 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
779 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
714 | { |
780 | { |
715 | ssize_t res; |
781 | ssize_t res; |
… | |
… | |
733 | |
799 | |
734 | X_LOCK (preadwritelock); |
800 | X_LOCK (preadwritelock); |
735 | ooffset = lseek (fd, 0, SEEK_CUR); |
801 | ooffset = lseek (fd, 0, SEEK_CUR); |
736 | lseek (fd, offset, SEEK_SET); |
802 | lseek (fd, offset, SEEK_SET); |
737 | res = write (fd, buf, count); |
803 | res = write (fd, buf, count); |
738 | lseek (fd, offset, SEEK_SET); |
804 | lseek (fd, ooffset, SEEK_SET); |
739 | X_UNLOCK (preadwritelock); |
805 | X_UNLOCK (preadwritelock); |
740 | |
806 | |
741 | return res; |
807 | return res; |
742 | } |
808 | } |
743 | #endif |
809 | #endif |
744 | |
810 | |
745 | #ifndef HAVE_FUTIMES |
811 | #ifndef HAVE_FUTIMES |
746 | |
812 | |
|
|
813 | # undef utimes |
|
|
814 | # undef futimes |
747 | # define utimes(path,times) eio__utimes (path, times) |
815 | # define utimes(path,times) eio__utimes (path, times) |
748 | # define futimes(fd,times) eio__futimes (fd, times) |
816 | # define futimes(fd,times) eio__futimes (fd, times) |
749 | |
817 | |
750 | static int |
818 | static int |
751 | eio__utimes (const char *filename, const struct timeval times[2]) |
819 | eio__utimes (const char *filename, const struct timeval times[2]) |
… | |
… | |
770 | } |
838 | } |
771 | |
839 | |
772 | #endif |
840 | #endif |
773 | |
841 | |
774 | #if !HAVE_FDATASYNC |
842 | #if !HAVE_FDATASYNC |
|
|
843 | # undef fdatasync |
775 | # define fdatasync fsync |
844 | # define fdatasync(fd) fsync (fd) |
776 | #endif |
845 | #endif |
|
|
846 | |
|
|
847 | /* sync_file_range always needs emulation */ |
|
|
848 | int |
|
|
849 | eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
|
|
850 | { |
|
|
851 | #if HAVE_SYNC_FILE_RANGE |
|
|
852 | int res; |
|
|
853 | |
|
|
854 | if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE |
|
|
855 | || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE |
|
|
856 | || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER) |
|
|
857 | { |
|
|
858 | flags = 0 |
|
|
859 | | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0) |
|
|
860 | | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0) |
|
|
861 | | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); |
|
|
862 | } |
|
|
863 | |
|
|
864 | res = sync_file_range (fd, offset, nbytes, flags); |
|
|
865 | |
|
|
866 | if (!res || errno != ENOSYS) |
|
|
867 | return res; |
|
|
868 | #endif |
|
|
869 | |
|
|
870 | /* even though we could play tricks with the flags, it's better to always |
|
|
871 | * call fdatasync, as thta matches the expectation of it's users best */ |
|
|
872 | return fdatasync (fd); |
|
|
873 | } |
777 | |
874 | |
778 | #if !HAVE_READAHEAD |
875 | #if !HAVE_READAHEAD |
|
|
876 | # undef readahead |
779 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
877 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
780 | |
878 | |
781 | static ssize_t |
879 | static ssize_t |
782 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
880 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
783 | { |
881 | { |
… | |
… | |
891 | } |
989 | } |
892 | |
990 | |
893 | return res; |
991 | return res; |
894 | } |
992 | } |
895 | |
993 | |
|
|
994 | static int |
|
|
995 | eio_dent_cmp (const void *a_, const void *b_) |
|
|
996 | { |
|
|
997 | const eio_dirent *a = (const eio_dirent *)a_; |
|
|
998 | const eio_dirent *b = (const eio_dirent *)b_; |
|
|
999 | |
|
|
1000 | return (int)b->score - (int)a->score ? (int)b->score - (int)a->score |
|
|
1001 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */ |
|
|
1002 | } |
|
|
1003 | |
896 | /* read a full directory */ |
1004 | /* read a full directory */ |
897 | static void |
1005 | static void |
898 | eio__scandir (eio_req *req, etp_worker *self) |
1006 | eio__scandir (eio_req *req, etp_worker *self) |
899 | { |
1007 | { |
900 | DIR *dirp; |
1008 | DIR *dirp; |
901 | EIO_STRUCT_DIRENT *entp; |
1009 | EIO_STRUCT_DIRENT *entp; |
902 | char *name, *names; |
1010 | unsigned char *name, *names; |
903 | int memlen = 4096; |
1011 | int namesalloc = 4096; |
904 | int memofs = 0; |
1012 | int namesoffs = 0; |
|
|
1013 | int flags = req->int1; |
|
|
1014 | eio_dirent *dents = 0; |
|
|
1015 | int dentalloc = 128; |
905 | int res = 0; |
1016 | int dentoffs = 0; |
|
|
1017 | |
|
|
1018 | req->result = -1; |
|
|
1019 | |
|
|
1020 | if (!(flags & EIO_READDIR_DENTS)) |
|
|
1021 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
906 | |
1022 | |
907 | X_LOCK (wrklock); |
1023 | X_LOCK (wrklock); |
|
|
1024 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
908 | self->dirp = dirp = opendir (req->ptr1); |
1025 | self->dirp = dirp = opendir (req->ptr1); |
909 | req->flags |= EIO_FLAG_PTR2_FREE; |
1026 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
910 | req->ptr2 = names = malloc (memlen); |
1027 | req->ptr1 = names = malloc (namesalloc); |
|
|
1028 | req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
911 | X_UNLOCK (wrklock); |
1029 | X_UNLOCK (wrklock); |
912 | |
1030 | |
913 | if (dirp && names) |
1031 | if (dirp && names && (!flags || dents)) |
914 | for (;;) |
1032 | for (;;) |
915 | { |
1033 | { |
916 | errno = 0; |
1034 | errno = 0; |
917 | entp = readdir (dirp); |
1035 | entp = readdir (dirp); |
918 | |
1036 | |
919 | if (!entp) |
1037 | if (!entp) |
|
|
1038 | { |
|
|
1039 | if (errno) |
|
|
1040 | break; |
|
|
1041 | |
|
|
1042 | /* sort etc. */ |
|
|
1043 | req->int1 = flags; |
|
|
1044 | req->result = dentoffs; |
|
|
1045 | |
|
|
1046 | if (dents) |
|
|
1047 | { |
|
|
1048 | eio_dirent *ent = dents + dentoffs; |
|
|
1049 | |
|
|
1050 | while (ent > dents) |
|
|
1051 | (--ent)->name = names + (size_t)ent->name; |
|
|
1052 | } |
|
|
1053 | |
|
|
1054 | if (flags & EIO_READDIR_STAT_ORDER |
|
|
1055 | || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) |
|
|
1056 | { |
|
|
1057 | /* pray your qsort doesn't use quicksort */ |
|
|
1058 | qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */ |
|
|
1059 | } |
|
|
1060 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1061 | { |
|
|
1062 | /* in this case, all is known, and we just put dirs first and sort them */ |
|
|
1063 | eio_dirent *ent = dents + dentoffs; |
|
|
1064 | eio_dirent *dir = dents; |
|
|
1065 | |
|
|
1066 | /* now move dirs to the front, and non-dirs to the back */ |
|
|
1067 | /* by walkign from both sides and swapping if necessary */ |
|
|
1068 | while (ent > dir) |
|
|
1069 | { |
|
|
1070 | if (dir->type == DT_DIR) |
|
|
1071 | ++dir; |
|
|
1072 | else |
|
|
1073 | { |
|
|
1074 | --ent; |
|
|
1075 | |
|
|
1076 | if (ent->type == DT_DIR) |
|
|
1077 | { |
|
|
1078 | eio_dirent tmp = *dir; |
|
|
1079 | *dir = *ent; |
|
|
1080 | *ent = tmp; |
|
|
1081 | |
|
|
1082 | ++dir; |
|
|
1083 | } |
|
|
1084 | } |
|
|
1085 | } |
|
|
1086 | |
|
|
1087 | /* now sort the dirs only */ |
|
|
1088 | qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp); |
|
|
1089 | } |
|
|
1090 | |
|
|
1091 | /* only provide the names array unless DENTS is specified */ |
|
|
1092 | if (!(flags & EIO_READDIR_DENTS)) |
|
|
1093 | { |
|
|
1094 | X_LOCK (wrklock); |
|
|
1095 | assert (!dents); |
|
|
1096 | req->ptr1 = 0; |
|
|
1097 | req->ptr2 = names; |
|
|
1098 | X_UNLOCK (wrklock); |
|
|
1099 | } |
|
|
1100 | |
920 | break; |
1101 | break; |
|
|
1102 | } |
921 | |
1103 | |
|
|
1104 | /* now add the entry to our list(s) */ |
922 | name = entp->d_name; |
1105 | name = entp->d_name; |
923 | |
1106 | |
|
|
1107 | /* skip . and .. entries */ |
924 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1108 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
925 | { |
1109 | { |
926 | int len = strlen (name) + 1; |
1110 | int len = D_NAMLEN (entp) + 1; |
927 | |
1111 | |
928 | res++; |
1112 | while (expect_false (namesoffs + len > namesalloc)) |
929 | |
|
|
930 | while (memofs + len > memlen) |
|
|
931 | { |
1113 | { |
932 | memlen *= 2; |
1114 | namesalloc *= 2; |
933 | X_LOCK (wrklock); |
1115 | X_LOCK (wrklock); |
934 | req->ptr2 = names = realloc (names, memlen); |
1116 | req->ptr1 = names = realloc (names, namesalloc); |
935 | X_UNLOCK (wrklock); |
1117 | X_UNLOCK (wrklock); |
936 | |
1118 | |
937 | if (!names) |
1119 | if (!names) |
938 | break; |
1120 | break; |
939 | } |
1121 | } |
940 | |
1122 | |
941 | memcpy (names + memofs, name, len); |
1123 | memcpy (names + namesoffs, name, len); |
|
|
1124 | |
|
|
1125 | if (dents) |
|
|
1126 | { |
|
|
1127 | struct eio_dirent *ent; |
|
|
1128 | |
|
|
1129 | if (expect_false (dentoffs == dentalloc)) |
|
|
1130 | { |
|
|
1131 | dentalloc *= 2; |
|
|
1132 | X_LOCK (wrklock); |
|
|
1133 | req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
|
|
1134 | X_UNLOCK (wrklock); |
|
|
1135 | |
|
|
1136 | if (!dents) |
|
|
1137 | break; |
|
|
1138 | } |
|
|
1139 | |
|
|
1140 | ent = dents + dentoffs; |
|
|
1141 | |
|
|
1142 | ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ |
|
|
1143 | ent->namelen = len - 1; |
|
|
1144 | ent->inode = D_INO (entp); |
|
|
1145 | |
|
|
1146 | switch (D_TYPE (entp)) |
|
|
1147 | { |
|
|
1148 | default: |
|
|
1149 | ent->type = EIO_DT_UNKNOWN; |
|
|
1150 | flags |= EIO_READDIR_FOUND_UNKNOWN; |
|
|
1151 | break; |
|
|
1152 | |
|
|
1153 | #ifdef DT_FIFO |
|
|
1154 | case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
|
|
1155 | #endif |
|
|
1156 | #ifdef DT_CHR |
|
|
1157 | case DT_CHR: ent->type = EIO_DT_CHR; break; |
|
|
1158 | #endif |
|
|
1159 | #ifdef DT_MPC |
|
|
1160 | case DT_MPC: ent->type = EIO_DT_MPC; break; |
|
|
1161 | #endif |
|
|
1162 | #ifdef DT_DIR |
|
|
1163 | case DT_DIR: ent->type = EIO_DT_DIR; break; |
|
|
1164 | #endif |
|
|
1165 | #ifdef DT_NAM |
|
|
1166 | case DT_NAM: ent->type = EIO_DT_NAM; break; |
|
|
1167 | #endif |
|
|
1168 | #ifdef DT_BLK |
|
|
1169 | case DT_BLK: ent->type = EIO_DT_BLK; break; |
|
|
1170 | #endif |
|
|
1171 | #ifdef DT_MPB |
|
|
1172 | case DT_MPB: ent->type = EIO_DT_MPB; break; |
|
|
1173 | #endif |
|
|
1174 | #ifdef DT_REG |
|
|
1175 | case DT_REG: ent->type = EIO_DT_REG; break; |
|
|
1176 | #endif |
|
|
1177 | #ifdef DT_NWK |
|
|
1178 | case DT_NWK: ent->type = EIO_DT_NWK; break; |
|
|
1179 | #endif |
|
|
1180 | #ifdef DT_CMP |
|
|
1181 | case DT_CMP: ent->type = EIO_DT_CMP; break; |
|
|
1182 | #endif |
|
|
1183 | #ifdef DT_LNK |
|
|
1184 | case DT_LNK: ent->type = EIO_DT_LNK; break; |
|
|
1185 | #endif |
|
|
1186 | #ifdef DT_SOCK |
|
|
1187 | case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
|
|
1188 | #endif |
|
|
1189 | #ifdef DT_DOOR |
|
|
1190 | case DT_DOOR: ent->type = EIO_DT_DOOR; break; |
|
|
1191 | #endif |
|
|
1192 | #ifdef DT_WHT |
|
|
1193 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
|
|
1194 | #endif |
|
|
1195 | } |
|
|
1196 | |
|
|
1197 | ent->score = 0; |
|
|
1198 | |
|
|
1199 | if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1200 | { |
|
|
1201 | if (ent->type == EIO_DT_UNKNOWN) |
|
|
1202 | { |
|
|
1203 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
|
|
1204 | ent->score = 98; |
|
|
1205 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
|
|
1206 | ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ |
|
|
1207 | } |
|
|
1208 | else if (ent->type == EIO_DT_DIR) |
|
|
1209 | ent->score = 100; |
|
|
1210 | } |
|
|
1211 | } |
|
|
1212 | |
942 | memofs += len; |
1213 | namesoffs += len; |
|
|
1214 | ++dentoffs; |
|
|
1215 | } |
|
|
1216 | |
|
|
1217 | if (EIO_CANCELLED (req)) |
|
|
1218 | { |
|
|
1219 | errno = ECANCELED; |
|
|
1220 | break; |
943 | } |
1221 | } |
944 | } |
1222 | } |
|
|
1223 | } |
945 | |
1224 | |
946 | if (errno) |
1225 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
947 | res = -1; |
1226 | # undef msync |
948 | |
1227 | # define msync(a,b,c) ((errno = ENOSYS), -1) |
949 | req->result = res; |
1228 | #endif |
|
|
1229 | |
|
|
1230 | int |
|
|
1231 | eio__mtouch (void *mem, size_t len, int flags) |
|
|
1232 | { |
|
|
1233 | intptr_t addr = (intptr_t)mem; |
|
|
1234 | intptr_t end = addr + len; |
|
|
1235 | #ifdef PAGESIZE |
|
|
1236 | const intptr_t page = PAGESIZE; |
|
|
1237 | #else |
|
|
1238 | static intptr_t page; |
|
|
1239 | |
|
|
1240 | if (!page) |
|
|
1241 | page = sysconf (_SC_PAGESIZE); |
|
|
1242 | #endif |
|
|
1243 | |
|
|
1244 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
|
|
1245 | |
|
|
1246 | if (addr < end) |
|
|
1247 | if (flags) /* modify */ |
|
|
1248 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
|
|
1249 | else |
|
|
1250 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
|
|
1251 | |
|
|
1252 | return 0; |
950 | } |
1253 | } |
951 | |
1254 | |
952 | /*****************************************************************************/ |
1255 | /*****************************************************************************/ |
953 | |
1256 | |
954 | #define ALLOC(len) \ |
1257 | #define ALLOC(len) \ |
… | |
… | |
1121 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1424 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1122 | |
1425 | |
1123 | case EIO_SYNC: req->result = 0; sync (); break; |
1426 | case EIO_SYNC: req->result = 0; sync (); break; |
1124 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1427 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1125 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1428 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
|
|
1429 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
|
|
1430 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
|
|
1431 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1126 | |
1432 | |
1127 | case EIO_READDIR: eio__scandir (req, self); break; |
1433 | case EIO_READDIR: eio__scandir (req, self); break; |
1128 | |
1434 | |
1129 | case EIO_BUSY: |
1435 | case EIO_BUSY: |
1130 | #ifdef _WIN32 |
1436 | #ifdef _WIN32 |
… | |
… | |
1162 | |
1468 | |
1163 | req->result = req->type == EIO_FUTIME |
1469 | req->result = req->type == EIO_FUTIME |
1164 | ? futimes (req->int1, times) |
1470 | ? futimes (req->int1, times) |
1165 | : utimes (req->ptr1, times); |
1471 | : utimes (req->ptr1, times); |
1166 | } |
1472 | } |
|
|
1473 | break; |
1167 | |
1474 | |
1168 | case EIO_GROUP: |
1475 | case EIO_GROUP: |
|
|
1476 | abort (); /* handled in eio_request */ |
|
|
1477 | |
1169 | case EIO_NOP: |
1478 | case EIO_NOP: |
1170 | req->result = 0; |
1479 | req->result = 0; |
1171 | break; |
1480 | break; |
1172 | |
1481 | |
1173 | case EIO_CUSTOM: |
1482 | case EIO_CUSTOM: |
1174 | req->feed (req); |
1483 | ((void (*)(eio_req *))req->feed) (req); |
1175 | break; |
1484 | break; |
1176 | |
1485 | |
1177 | default: |
1486 | default: |
1178 | req->result = -1; |
1487 | req->result = -1; |
1179 | break; |
1488 | break; |
… | |
… | |
1202 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1511 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1203 | { |
1512 | { |
1204 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
1513 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
1205 | } |
1514 | } |
1206 | |
1515 | |
|
|
1516 | eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
1517 | { |
|
|
1518 | REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
|
|
1519 | } |
|
|
1520 | |
|
|
1521 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
1522 | { |
|
|
1523 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
|
|
1524 | } |
|
|
1525 | |
|
|
1526 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
|
|
1527 | { |
|
|
1528 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
|
|
1529 | } |
|
|
1530 | |
1207 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1531 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1208 | { |
1532 | { |
1209 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1533 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1210 | } |
1534 | } |
1211 | |
1535 | |
… | |
… | |
1323 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
1647 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
1324 | { |
1648 | { |
1325 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
1649 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
1326 | } |
1650 | } |
1327 | |
1651 | |
1328 | eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) |
1652 | eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data) |
1329 | { |
1653 | { |
1330 | return eio__1path (EIO_READDIR, path, pri, cb, data); |
1654 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1331 | } |
1655 | } |
1332 | |
1656 | |
1333 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1657 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1334 | { |
1658 | { |
1335 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
1659 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
… | |
… | |
1366 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1690 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1367 | } |
1691 | } |
1368 | |
1692 | |
1369 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
1693 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
1370 | { |
1694 | { |
1371 | REQ (EIO_CUSTOM); req->feed = execute; SEND; |
1695 | REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; |
1372 | } |
1696 | } |
1373 | |
1697 | |
1374 | #endif |
1698 | #endif |
1375 | |
1699 | |
1376 | eio_req *eio_grp (eio_cb cb, void *data) |
1700 | eio_req *eio_grp (eio_cb cb, void *data) |
… | |
… | |
1404 | |
1728 | |
1405 | void eio_grp_add (eio_req *grp, eio_req *req) |
1729 | void eio_grp_add (eio_req *grp, eio_req *req) |
1406 | { |
1730 | { |
1407 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
1731 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
1408 | |
1732 | |
|
|
1733 | grp->flags |= EIO_FLAG_GROUPADD; |
|
|
1734 | |
1409 | ++grp->size; |
1735 | ++grp->size; |
1410 | req->grp = grp; |
1736 | req->grp = grp; |
1411 | |
1737 | |
1412 | req->grp_prev = 0; |
1738 | req->grp_prev = 0; |
1413 | req->grp_next = grp->grp_first; |
1739 | req->grp_next = grp->grp_first; |