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 |
… | |
… | |
120 | |
152 | |
121 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
153 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
122 | |
154 | |
123 | /*****************************************************************************/ |
155 | /*****************************************************************************/ |
124 | |
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) |
|
|
165 | |
|
|
166 | /*****************************************************************************/ |
|
|
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 |
|
|
170 | |
|
|
171 | struct etp_worker; |
127 | |
172 | |
128 | #define ETP_REQ eio_req |
173 | #define ETP_REQ eio_req |
129 | #define ETP_DESTROY(req) eio_destroy (req) |
174 | #define ETP_DESTROY(req) eio_destroy (req) |
130 | static int eio_finish (eio_req *req); |
175 | static int eio_finish (eio_req *req); |
131 | #define ETP_FINISH(req) eio_finish (req) |
176 | #define ETP_FINISH(req) eio_finish (req) |
|
|
177 | static void eio_execute (struct etp_worker *self, eio_req *req); |
|
|
178 | #define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) |
132 | |
179 | |
133 | #define ETP_WORKER_CLEAR(req) \ |
180 | #define ETP_WORKER_CLEAR(req) \ |
134 | if (wrk->dbuf) \ |
181 | if (wrk->dbuf) \ |
135 | { \ |
182 | { \ |
136 | free (wrk->dbuf); \ |
183 | free (wrk->dbuf); \ |
… | |
… | |
140 | if (wrk->dirp) \ |
187 | if (wrk->dirp) \ |
141 | { \ |
188 | { \ |
142 | closedir (wrk->dirp); \ |
189 | closedir (wrk->dirp); \ |
143 | wrk->dirp = 0; \ |
190 | wrk->dirp = 0; \ |
144 | } |
191 | } |
|
|
192 | |
145 | #define ETP_WORKER_COMMON \ |
193 | #define ETP_WORKER_COMMON \ |
146 | void *dbuf; \ |
194 | void *dbuf; \ |
147 | DIR *dirp; |
195 | DIR *dirp; |
148 | |
196 | |
149 | /*****************************************************************************/ |
197 | /*****************************************************************************/ |
… | |
… | |
172 | |
220 | |
173 | static mutex_t wrklock = X_MUTEX_INIT; |
221 | static mutex_t wrklock = X_MUTEX_INIT; |
174 | static mutex_t reslock = X_MUTEX_INIT; |
222 | static mutex_t reslock = X_MUTEX_INIT; |
175 | static mutex_t reqlock = X_MUTEX_INIT; |
223 | static mutex_t reqlock = X_MUTEX_INIT; |
176 | static cond_t reqwait = X_COND_INIT; |
224 | static cond_t reqwait = X_COND_INIT; |
|
|
225 | |
|
|
226 | #if !HAVE_PREADWRITE |
|
|
227 | /* |
|
|
228 | * make our pread/pwrite emulation safe against themselves, but not against |
|
|
229 | * normal read/write by using a mutex. slows down execution a lot, |
|
|
230 | * but that's your problem, not mine. |
|
|
231 | */ |
|
|
232 | static mutex_t preadwritelock = X_MUTEX_INIT; |
|
|
233 | #endif |
177 | |
234 | |
178 | typedef struct etp_worker |
235 | typedef struct etp_worker |
179 | { |
236 | { |
180 | /* locked by wrklock */ |
237 | /* locked by wrklock */ |
181 | struct etp_worker *prev, *next; |
238 | struct etp_worker *prev, *next; |
… | |
… | |
310 | X_LOCK (reqlock); |
367 | X_LOCK (reqlock); |
311 | X_LOCK (reslock); |
368 | X_LOCK (reslock); |
312 | #if !HAVE_PREADWRITE |
369 | #if !HAVE_PREADWRITE |
313 | X_LOCK (preadwritelock); |
370 | X_LOCK (preadwritelock); |
314 | #endif |
371 | #endif |
315 | #if !HAVE_READDIR_R |
|
|
316 | X_LOCK (readdirlock); |
|
|
317 | #endif |
|
|
318 | } |
372 | } |
319 | |
373 | |
320 | static void etp_atfork_parent (void) |
374 | static void etp_atfork_parent (void) |
321 | { |
375 | { |
322 | #if !HAVE_READDIR_R |
|
|
323 | X_UNLOCK (readdirlock); |
|
|
324 | #endif |
|
|
325 | #if !HAVE_PREADWRITE |
376 | #if !HAVE_PREADWRITE |
326 | X_UNLOCK (preadwritelock); |
377 | X_UNLOCK (preadwritelock); |
327 | #endif |
378 | #endif |
328 | X_UNLOCK (reslock); |
379 | X_UNLOCK (reslock); |
329 | X_UNLOCK (reqlock); |
380 | X_UNLOCK (reqlock); |
… | |
… | |
332 | |
383 | |
333 | static void etp_atfork_child (void) |
384 | static void etp_atfork_child (void) |
334 | { |
385 | { |
335 | ETP_REQ *prv; |
386 | ETP_REQ *prv; |
336 | |
387 | |
337 | while (prv = reqq_shift (&req_queue)) |
388 | while ((prv = reqq_shift (&req_queue))) |
338 | ETP_DESTROY (prv); |
389 | ETP_DESTROY (prv); |
339 | |
390 | |
340 | while (prv = reqq_shift (&res_queue)) |
391 | while ((prv = reqq_shift (&res_queue))) |
341 | ETP_DESTROY (prv); |
392 | ETP_DESTROY (prv); |
342 | |
393 | |
343 | while (wrk_first.next != &wrk_first) |
394 | while (wrk_first.next != &wrk_first) |
344 | { |
395 | { |
345 | etp_worker *wrk = wrk_first.next; |
396 | etp_worker *wrk = wrk_first.next; |
… | |
… | |
373 | |
424 | |
374 | pthread_once (&doinit, etp_once_init); |
425 | pthread_once (&doinit, etp_once_init); |
375 | |
426 | |
376 | want_poll_cb = want_poll; |
427 | want_poll_cb = want_poll; |
377 | done_poll_cb = done_poll; |
428 | done_poll_cb = done_poll; |
|
|
429 | |
|
|
430 | return 0; |
378 | } |
431 | } |
379 | |
432 | |
380 | X_THREAD_PROC (etp_proc); |
433 | X_THREAD_PROC (etp_proc); |
381 | |
434 | |
382 | static void etp_start_thread (void) |
435 | static void etp_start_thread (void) |
… | |
… | |
402 | X_UNLOCK (wrklock); |
455 | X_UNLOCK (wrklock); |
403 | } |
456 | } |
404 | |
457 | |
405 | static void etp_maybe_start_thread (void) |
458 | static void etp_maybe_start_thread (void) |
406 | { |
459 | { |
407 | if (etp_nthreads () >= wanted) |
460 | if (expect_true (etp_nthreads () >= wanted)) |
408 | return; |
461 | return; |
409 | |
462 | |
410 | /* todo: maybe use idle here, but might be less exact */ |
463 | /* todo: maybe use idle here, but might be less exact */ |
411 | 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 ())) |
412 | return; |
465 | return; |
413 | |
466 | |
414 | etp_start_thread (); |
467 | etp_start_thread (); |
415 | } |
468 | } |
416 | |
469 | |
… | |
… | |
469 | |
522 | |
470 | X_LOCK (reqlock); |
523 | X_LOCK (reqlock); |
471 | --nreqs; |
524 | --nreqs; |
472 | X_UNLOCK (reqlock); |
525 | X_UNLOCK (reqlock); |
473 | |
526 | |
474 | if (req->type == EIO_GROUP && req->size) |
527 | if (expect_false (req->type == EIO_GROUP && req->size)) |
475 | { |
528 | { |
476 | req->int1 = 1; /* mark request as delayed */ |
529 | req->int1 = 1; /* mark request as delayed */ |
477 | continue; |
530 | continue; |
478 | } |
531 | } |
479 | else |
532 | else |
480 | { |
533 | { |
481 | int res = ETP_FINISH (req); |
534 | int res = ETP_FINISH (req); |
482 | if (res) |
535 | if (expect_false (res)) |
483 | return res; |
536 | return res; |
484 | } |
537 | } |
485 | |
538 | |
486 | if (maxreqs && !--maxreqs) |
539 | if (expect_false (maxreqs && !--maxreqs)) |
487 | break; |
540 | break; |
488 | |
541 | |
489 | if (maxtime) |
542 | if (maxtime) |
490 | { |
543 | { |
491 | gettimeofday (&tv_now, 0); |
544 | gettimeofday (&tv_now, 0); |
… | |
… | |
510 | |
563 | |
511 | static void etp_submit (ETP_REQ *req) |
564 | static void etp_submit (ETP_REQ *req) |
512 | { |
565 | { |
513 | req->pri -= ETP_PRI_MIN; |
566 | req->pri -= ETP_PRI_MIN; |
514 | |
567 | |
515 | 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; |
516 | 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; |
517 | |
570 | |
|
|
571 | if (expect_false (req->type == EIO_GROUP)) |
|
|
572 | { |
|
|
573 | /* I hope this is worth it :/ */ |
518 | X_LOCK (reqlock); |
574 | X_LOCK (reqlock); |
519 | ++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; |
520 | ++nready; |
591 | ++nready; |
521 | reqq_push (&req_queue, req); |
592 | reqq_push (&req_queue, req); |
522 | X_COND_SIGNAL (reqwait); |
593 | X_COND_SIGNAL (reqwait); |
523 | X_UNLOCK (reqlock); |
594 | X_UNLOCK (reqlock); |
524 | |
595 | |
525 | etp_maybe_start_thread (); |
596 | etp_maybe_start_thread (); |
|
|
597 | } |
526 | } |
598 | } |
527 | |
599 | |
528 | static void etp_set_max_poll_time (double nseconds) |
600 | static void etp_set_max_poll_time (double nseconds) |
529 | { |
601 | { |
530 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
602 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
… | |
… | |
565 | |
637 | |
566 | static void grp_try_feed (eio_req *grp) |
638 | static void grp_try_feed (eio_req *grp) |
567 | { |
639 | { |
568 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
640 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
569 | { |
641 | { |
570 | int old_len = grp->size; |
642 | grp->flags &= ~EIO_FLAG_GROUPADD; |
571 | |
643 | |
572 | EIO_FEED (grp); |
644 | EIO_FEED (grp); |
573 | |
645 | |
574 | /* stop if no progress has been made */ |
646 | /* stop if no progress has been made */ |
575 | if (old_len == grp->size) |
647 | if (!(grp->flags & EIO_FLAG_GROUPADD)) |
576 | { |
648 | { |
577 | grp->feed = 0; |
649 | grp->feed = 0; |
578 | break; |
650 | break; |
579 | } |
651 | } |
580 | } |
652 | } |
… | |
… | |
697 | |
769 | |
698 | /*****************************************************************************/ |
770 | /*****************************************************************************/ |
699 | /* work around various missing functions */ |
771 | /* work around various missing functions */ |
700 | |
772 | |
701 | #if !HAVE_PREADWRITE |
773 | #if !HAVE_PREADWRITE |
|
|
774 | # undef pread |
|
|
775 | # undef pwrite |
702 | # define pread eio__pread |
776 | # define pread eio__pread |
703 | # define pwrite eio__pwrite |
777 | # 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 | |
778 | |
712 | static ssize_t |
779 | static ssize_t |
713 | 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) |
714 | { |
781 | { |
715 | ssize_t res; |
782 | ssize_t res; |
… | |
… | |
733 | |
800 | |
734 | X_LOCK (preadwritelock); |
801 | X_LOCK (preadwritelock); |
735 | ooffset = lseek (fd, 0, SEEK_CUR); |
802 | ooffset = lseek (fd, 0, SEEK_CUR); |
736 | lseek (fd, offset, SEEK_SET); |
803 | lseek (fd, offset, SEEK_SET); |
737 | res = write (fd, buf, count); |
804 | res = write (fd, buf, count); |
738 | lseek (fd, offset, SEEK_SET); |
805 | lseek (fd, ooffset, SEEK_SET); |
739 | X_UNLOCK (preadwritelock); |
806 | X_UNLOCK (preadwritelock); |
740 | |
807 | |
741 | return res; |
808 | return res; |
742 | } |
809 | } |
743 | #endif |
810 | #endif |
744 | |
811 | |
745 | #ifndef HAVE_FUTIMES |
812 | #ifndef HAVE_FUTIMES |
746 | |
813 | |
|
|
814 | # undef utimes |
|
|
815 | # undef futimes |
747 | # define utimes(path,times) eio__utimes (path, times) |
816 | # define utimes(path,times) eio__utimes (path, times) |
748 | # define futimes(fd,times) eio__futimes (fd, times) |
817 | # define futimes(fd,times) eio__futimes (fd, times) |
749 | |
818 | |
750 | static int |
819 | static int |
751 | eio__utimes (const char *filename, const struct timeval times[2]) |
820 | eio__utimes (const char *filename, const struct timeval times[2]) |
… | |
… | |
770 | } |
839 | } |
771 | |
840 | |
772 | #endif |
841 | #endif |
773 | |
842 | |
774 | #if !HAVE_FDATASYNC |
843 | #if !HAVE_FDATASYNC |
|
|
844 | # undef fdatasync |
775 | # define fdatasync fsync |
845 | # define fdatasync(fd) fsync (fd) |
776 | #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 | } |
777 | |
875 | |
778 | #if !HAVE_READAHEAD |
876 | #if !HAVE_READAHEAD |
|
|
877 | # undef readahead |
779 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
878 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
780 | |
879 | |
781 | static ssize_t |
880 | static ssize_t |
782 | eio__readahead (int fd, off_t offset, size_t count, worker *self) |
881 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
783 | { |
882 | { |
784 | size_t todo = count; |
883 | size_t todo = count; |
785 | dBUF; |
884 | dBUF; |
786 | |
885 | |
787 | while (todo > 0) |
886 | while (todo > 0) |
… | |
… | |
795 | |
894 | |
796 | errno = 0; |
895 | errno = 0; |
797 | return count; |
896 | return count; |
798 | } |
897 | } |
799 | |
898 | |
800 | #endif |
|
|
801 | |
|
|
802 | #if !HAVE_READDIR_R |
|
|
803 | # define readdir_r eio__readdir_r |
|
|
804 | |
|
|
805 | static mutex_t readdirlock = X_MUTEX_INIT; |
|
|
806 | |
|
|
807 | static int |
|
|
808 | eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res) |
|
|
809 | { |
|
|
810 | EIO_STRUCT_DIRENT *e; |
|
|
811 | int errorno; |
|
|
812 | |
|
|
813 | X_LOCK (readdirlock); |
|
|
814 | |
|
|
815 | e = readdir (dirp); |
|
|
816 | errorno = errno; |
|
|
817 | |
|
|
818 | if (e) |
|
|
819 | { |
|
|
820 | *res = ent; |
|
|
821 | strcpy (ent->d_name, e->d_name); |
|
|
822 | } |
|
|
823 | else |
|
|
824 | *res = 0; |
|
|
825 | |
|
|
826 | X_UNLOCK (readdirlock); |
|
|
827 | |
|
|
828 | errno = errorno; |
|
|
829 | return e ? 0 : -1; |
|
|
830 | } |
|
|
831 | #endif |
899 | #endif |
832 | |
900 | |
833 | /* sendfile always needs emulation */ |
901 | /* sendfile always needs emulation */ |
834 | static ssize_t |
902 | static ssize_t |
835 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
903 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
… | |
… | |
922 | } |
990 | } |
923 | |
991 | |
924 | return res; |
992 | return res; |
925 | } |
993 | } |
926 | |
994 | |
|
|
995 | static signed char |
|
|
996 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
|
|
997 | { |
|
|
998 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
|
|
999 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; |
|
|
1000 | } |
|
|
1001 | |
|
|
1002 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
|
|
1003 | |
|
|
1004 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
|
|
1005 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
|
|
1006 | |
|
|
1007 | static void |
|
|
1008 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1009 | { |
|
|
1010 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
|
|
1011 | unsigned char *bit = bits; |
|
|
1012 | |
|
|
1013 | assert (CHAR_BIT == 8); |
|
|
1014 | assert (sizeof (eio_dirent) * 8 < 256); |
|
|
1015 | assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ |
|
|
1016 | assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ |
|
|
1017 | |
|
|
1018 | if (size <= EIO_SORT_FAST) |
|
|
1019 | return; |
|
|
1020 | |
|
|
1021 | /* first prepare an array of bits to test in our radix sort */ |
|
|
1022 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
|
|
1023 | /* inode_bits must contain all inodes ORed together */ |
|
|
1024 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
|
|
1025 | { |
|
|
1026 | ino_t endianness; |
|
|
1027 | int i, j; |
|
|
1028 | |
|
|
1029 | /* we store the byte offset of byte n into byte n of "endianness" */ |
|
|
1030 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1031 | ((unsigned char *)&endianness)[i] = i; |
|
|
1032 | |
|
|
1033 | *bit++ = 0; |
|
|
1034 | |
|
|
1035 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1036 | { |
|
|
1037 | /* shifting off the byte offsets out of "endianness" */ |
|
|
1038 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
|
|
1039 | endianness >>= 8; |
|
|
1040 | |
|
|
1041 | for (j = 0; j < 8; ++j) |
|
|
1042 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
|
|
1043 | *bit++ = offs + j; |
|
|
1044 | } |
|
|
1045 | |
|
|
1046 | for (j = 0; j < 8; ++j) |
|
|
1047 | if (score_bits & (1 << j)) |
|
|
1048 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
|
|
1049 | } |
|
|
1050 | |
|
|
1051 | /* now actually do the sorting (a variant of MSD radix sort) */ |
|
|
1052 | { |
|
|
1053 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
|
|
1054 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
|
|
1055 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
|
|
1056 | int stk_idx = 0; |
|
|
1057 | |
|
|
1058 | base_stk [stk_idx] = dents; |
|
|
1059 | end_stk [stk_idx] = dents + size; |
|
|
1060 | bit_stk [stk_idx] = bit - 1; |
|
|
1061 | |
|
|
1062 | do |
|
|
1063 | { |
|
|
1064 | base = base_stk [stk_idx]; |
|
|
1065 | end = end_stk [stk_idx]; |
|
|
1066 | bit = bit_stk [stk_idx]; |
|
|
1067 | |
|
|
1068 | for (;;) |
|
|
1069 | { |
|
|
1070 | unsigned char O = *bit >> 3; |
|
|
1071 | unsigned char M = 1 << (*bit & 7); |
|
|
1072 | |
|
|
1073 | eio_dirent *a = base; |
|
|
1074 | eio_dirent *b = end; |
|
|
1075 | |
|
|
1076 | if (b - a < EIO_SORT_CUTOFF) |
|
|
1077 | break; |
|
|
1078 | |
|
|
1079 | /* now bit-partition the array on the bit */ |
|
|
1080 | /* this ugly asymmetric loop seems to perform much better than typical */ |
|
|
1081 | /* partition algos found in the literature */ |
|
|
1082 | do |
|
|
1083 | if (!(((unsigned char *)a)[O] & M)) |
|
|
1084 | ++a; |
|
|
1085 | else if (!(((unsigned char *)--b)[O] & M)) |
|
|
1086 | { |
|
|
1087 | eio_dirent tmp = *a; *a = *b; *b = tmp; |
|
|
1088 | ++a; |
|
|
1089 | } |
|
|
1090 | while (b > a); |
|
|
1091 | |
|
|
1092 | /* next bit, or stop, if no bits left in this path */ |
|
|
1093 | if (!*--bit) |
|
|
1094 | break; |
|
|
1095 | |
|
|
1096 | base_stk [stk_idx] = a; |
|
|
1097 | end_stk [stk_idx] = end; |
|
|
1098 | bit_stk [stk_idx] = bit; |
|
|
1099 | ++stk_idx; |
|
|
1100 | |
|
|
1101 | end = a; |
|
|
1102 | } |
|
|
1103 | } |
|
|
1104 | while (stk_idx--); |
|
|
1105 | } |
|
|
1106 | } |
|
|
1107 | |
|
|
1108 | static void |
|
|
1109 | eio_dent_insertion_sort (eio_dirent *dents, int size) |
|
|
1110 | { |
|
|
1111 | /* first move the smallest element to the front, to act as a sentinel */ |
|
|
1112 | { |
|
|
1113 | int i; |
|
|
1114 | eio_dirent *min = dents; |
|
|
1115 | |
|
|
1116 | /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
|
|
1117 | for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
|
|
1118 | if (EIO_DENT_CMP (dents [i], <, *min)) |
|
|
1119 | min = &dents [i]; |
|
|
1120 | |
|
|
1121 | /* swap elements 0 and j (minimum) */ |
|
|
1122 | { |
|
|
1123 | eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
|
|
1124 | } |
|
|
1125 | } |
|
|
1126 | |
|
|
1127 | /* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
|
|
1128 | { |
|
|
1129 | eio_dirent *i, *j; |
|
|
1130 | |
|
|
1131 | for (i = dents + 1; i < dents + size; ++i) |
|
|
1132 | { |
|
|
1133 | eio_dirent value = *i; |
|
|
1134 | |
|
|
1135 | for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
|
|
1136 | j [1] = j [0]; |
|
|
1137 | |
|
|
1138 | j [1] = value; |
|
|
1139 | } |
|
|
1140 | } |
|
|
1141 | } |
|
|
1142 | |
|
|
1143 | static void |
|
|
1144 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1145 | { |
|
|
1146 | if (size <= 1) |
|
|
1147 | return; /* our insertion sort relies on size > 0 */ |
|
|
1148 | |
|
|
1149 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
|
|
1150 | /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
|
|
1151 | eio_dent_radix_sort (dents, size, score_bits, inode_bits); |
|
|
1152 | |
|
|
1153 | /* use an insertion sort at the end, or for small arrays, */ |
|
|
1154 | /* as insertion sort is more efficient for small partitions */ |
|
|
1155 | eio_dent_insertion_sort (dents, size); |
|
|
1156 | } |
|
|
1157 | |
927 | /* read a full directory */ |
1158 | /* read a full directory */ |
928 | static void |
1159 | static void |
929 | eio__scandir (eio_req *req, etp_worker *self) |
1160 | eio__scandir (eio_req *req, etp_worker *self) |
930 | { |
1161 | { |
931 | DIR *dirp; |
1162 | DIR *dirp; |
932 | union |
|
|
933 | { |
|
|
934 | EIO_STRUCT_DIRENT d; |
|
|
935 | char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1]; |
|
|
936 | } *u; |
|
|
937 | EIO_STRUCT_DIRENT *entp; |
1163 | EIO_STRUCT_DIRENT *entp; |
938 | char *name, *names; |
1164 | char *name, *names; |
939 | int memlen = 4096; |
1165 | int namesalloc = 4096; |
940 | int memofs = 0; |
1166 | int namesoffs = 0; |
|
|
1167 | int flags = req->int1; |
|
|
1168 | eio_dirent *dents = 0; |
|
|
1169 | int dentalloc = 128; |
941 | int res = 0; |
1170 | int dentoffs = 0; |
|
|
1171 | ino_t inode_bits = 0; |
|
|
1172 | |
|
|
1173 | req->result = -1; |
|
|
1174 | |
|
|
1175 | if (!(flags & EIO_READDIR_DENTS)) |
|
|
1176 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
942 | |
1177 | |
943 | X_LOCK (wrklock); |
1178 | X_LOCK (wrklock); |
|
|
1179 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
944 | self->dirp = dirp = opendir (req->ptr1); |
1180 | self->dirp = dirp = opendir (req->ptr1); |
945 | self->dbuf = u = malloc (sizeof (*u)); |
|
|
946 | req->flags |= EIO_FLAG_PTR2_FREE; |
1181 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
|
|
1182 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
947 | req->ptr2 = names = malloc (memlen); |
1183 | req->ptr2 = names = malloc (namesalloc); |
948 | X_UNLOCK (wrklock); |
1184 | X_UNLOCK (wrklock); |
949 | |
1185 | |
950 | if (dirp && u && names) |
1186 | if (dirp && names && (!flags || dents)) |
951 | for (;;) |
1187 | for (;;) |
952 | { |
1188 | { |
953 | errno = 0; |
1189 | errno = 0; |
954 | readdir_r (dirp, &u->d, &entp); |
1190 | entp = readdir (dirp); |
955 | |
1191 | |
956 | if (!entp) |
1192 | if (!entp) |
|
|
1193 | { |
|
|
1194 | if (errno) |
|
|
1195 | break; |
|
|
1196 | |
|
|
1197 | /* sort etc. */ |
|
|
1198 | req->int1 = flags; |
|
|
1199 | req->result = dentoffs; |
|
|
1200 | |
|
|
1201 | if (flags & EIO_READDIR_STAT_ORDER) |
|
|
1202 | eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ |
|
|
1203 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1204 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
|
|
1205 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
|
|
1206 | else |
|
|
1207 | { |
|
|
1208 | /* in this case, all is known, and we just put dirs first and sort them */ |
|
|
1209 | eio_dirent *oth = dents + dentoffs; |
|
|
1210 | eio_dirent *dir = dents; |
|
|
1211 | |
|
|
1212 | /* now partition dirs to the front, and non-dirs to the back */ |
|
|
1213 | /* by walking from both sides and swapping if necessary */ |
|
|
1214 | /* also clear score, so it doesn't influence sorting */ |
|
|
1215 | while (oth > dir) |
|
|
1216 | { |
|
|
1217 | if (dir->type == EIO_DT_DIR) |
|
|
1218 | ++dir; |
|
|
1219 | else if ((--oth)->type == EIO_DT_DIR) |
|
|
1220 | { |
|
|
1221 | eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
|
|
1222 | |
|
|
1223 | ++dir; |
|
|
1224 | } |
|
|
1225 | } |
|
|
1226 | |
|
|
1227 | /* now sort the dirs only */ |
|
|
1228 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
|
|
1229 | } |
|
|
1230 | |
957 | break; |
1231 | break; |
|
|
1232 | } |
958 | |
1233 | |
|
|
1234 | /* now add the entry to our list(s) */ |
959 | name = entp->d_name; |
1235 | name = entp->d_name; |
960 | |
1236 | |
|
|
1237 | /* skip . and .. entries */ |
961 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1238 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
962 | { |
1239 | { |
963 | int len = strlen (name) + 1; |
1240 | int len = D_NAMLEN (entp) + 1; |
964 | |
1241 | |
965 | res++; |
1242 | while (expect_false (namesoffs + len > namesalloc)) |
966 | |
|
|
967 | while (memofs + len > memlen) |
|
|
968 | { |
1243 | { |
969 | memlen *= 2; |
1244 | namesalloc *= 2; |
970 | X_LOCK (wrklock); |
1245 | X_LOCK (wrklock); |
971 | req->ptr2 = names = realloc (names, memlen); |
1246 | req->ptr2 = names = realloc (names, namesalloc); |
972 | X_UNLOCK (wrklock); |
1247 | X_UNLOCK (wrklock); |
973 | |
1248 | |
974 | if (!names) |
1249 | if (!names) |
975 | break; |
1250 | break; |
976 | } |
1251 | } |
977 | |
1252 | |
978 | memcpy (names + memofs, name, len); |
1253 | memcpy (names + namesoffs, name, len); |
|
|
1254 | |
|
|
1255 | if (dents) |
|
|
1256 | { |
|
|
1257 | struct eio_dirent *ent; |
|
|
1258 | |
|
|
1259 | if (expect_false (dentoffs == dentalloc)) |
|
|
1260 | { |
|
|
1261 | dentalloc *= 2; |
|
|
1262 | X_LOCK (wrklock); |
|
|
1263 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
|
|
1264 | X_UNLOCK (wrklock); |
|
|
1265 | |
|
|
1266 | if (!dents) |
|
|
1267 | break; |
|
|
1268 | } |
|
|
1269 | |
|
|
1270 | ent = dents + dentoffs; |
|
|
1271 | |
|
|
1272 | ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
|
|
1273 | ent->namelen = len - 1; |
|
|
1274 | ent->inode = D_INO (entp); |
|
|
1275 | |
|
|
1276 | inode_bits |= ent->inode; |
|
|
1277 | |
|
|
1278 | switch (D_TYPE (entp)) |
|
|
1279 | { |
|
|
1280 | default: |
|
|
1281 | ent->type = EIO_DT_UNKNOWN; |
|
|
1282 | flags |= EIO_READDIR_FOUND_UNKNOWN; |
|
|
1283 | break; |
|
|
1284 | |
|
|
1285 | #ifdef DT_FIFO |
|
|
1286 | case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
|
|
1287 | #endif |
|
|
1288 | #ifdef DT_CHR |
|
|
1289 | case DT_CHR: ent->type = EIO_DT_CHR; break; |
|
|
1290 | #endif |
|
|
1291 | #ifdef DT_MPC |
|
|
1292 | case DT_MPC: ent->type = EIO_DT_MPC; break; |
|
|
1293 | #endif |
|
|
1294 | #ifdef DT_DIR |
|
|
1295 | case DT_DIR: ent->type = EIO_DT_DIR; break; |
|
|
1296 | #endif |
|
|
1297 | #ifdef DT_NAM |
|
|
1298 | case DT_NAM: ent->type = EIO_DT_NAM; break; |
|
|
1299 | #endif |
|
|
1300 | #ifdef DT_BLK |
|
|
1301 | case DT_BLK: ent->type = EIO_DT_BLK; break; |
|
|
1302 | #endif |
|
|
1303 | #ifdef DT_MPB |
|
|
1304 | case DT_MPB: ent->type = EIO_DT_MPB; break; |
|
|
1305 | #endif |
|
|
1306 | #ifdef DT_REG |
|
|
1307 | case DT_REG: ent->type = EIO_DT_REG; break; |
|
|
1308 | #endif |
|
|
1309 | #ifdef DT_NWK |
|
|
1310 | case DT_NWK: ent->type = EIO_DT_NWK; break; |
|
|
1311 | #endif |
|
|
1312 | #ifdef DT_CMP |
|
|
1313 | case DT_CMP: ent->type = EIO_DT_CMP; break; |
|
|
1314 | #endif |
|
|
1315 | #ifdef DT_LNK |
|
|
1316 | case DT_LNK: ent->type = EIO_DT_LNK; break; |
|
|
1317 | #endif |
|
|
1318 | #ifdef DT_SOCK |
|
|
1319 | case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
|
|
1320 | #endif |
|
|
1321 | #ifdef DT_DOOR |
|
|
1322 | case DT_DOOR: ent->type = EIO_DT_DOOR; break; |
|
|
1323 | #endif |
|
|
1324 | #ifdef DT_WHT |
|
|
1325 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
|
|
1326 | #endif |
|
|
1327 | } |
|
|
1328 | |
|
|
1329 | ent->score = 7; |
|
|
1330 | |
|
|
1331 | if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1332 | { |
|
|
1333 | if (ent->type == EIO_DT_UNKNOWN) |
|
|
1334 | { |
|
|
1335 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
|
|
1336 | ent->score = 1; |
|
|
1337 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
|
|
1338 | ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
|
|
1339 | } |
|
|
1340 | else if (ent->type == EIO_DT_DIR) |
|
|
1341 | ent->score = 0; |
|
|
1342 | } |
|
|
1343 | } |
|
|
1344 | |
979 | memofs += len; |
1345 | namesoffs += len; |
|
|
1346 | ++dentoffs; |
|
|
1347 | } |
|
|
1348 | |
|
|
1349 | if (EIO_CANCELLED (req)) |
|
|
1350 | { |
|
|
1351 | errno = ECANCELED; |
|
|
1352 | break; |
980 | } |
1353 | } |
981 | } |
1354 | } |
|
|
1355 | } |
982 | |
1356 | |
983 | if (errno) |
1357 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
984 | res = -1; |
1358 | # undef msync |
985 | |
1359 | # define msync(a,b,c) ((errno = ENOSYS), -1) |
986 | req->result = res; |
1360 | #endif |
|
|
1361 | |
|
|
1362 | int |
|
|
1363 | eio__mtouch (void *mem, size_t len, int flags) |
|
|
1364 | { |
|
|
1365 | intptr_t addr = (intptr_t)mem; |
|
|
1366 | intptr_t end = addr + len; |
|
|
1367 | #ifdef PAGESIZE |
|
|
1368 | const intptr_t page = PAGESIZE; |
|
|
1369 | #else |
|
|
1370 | static intptr_t page; |
|
|
1371 | |
|
|
1372 | if (!page) |
|
|
1373 | page = sysconf (_SC_PAGESIZE); |
|
|
1374 | #endif |
|
|
1375 | |
|
|
1376 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
|
|
1377 | |
|
|
1378 | if (addr < end) |
|
|
1379 | if (flags) /* modify */ |
|
|
1380 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
|
|
1381 | else |
|
|
1382 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
|
|
1383 | |
|
|
1384 | return 0; |
987 | } |
1385 | } |
988 | |
1386 | |
989 | /*****************************************************************************/ |
1387 | /*****************************************************************************/ |
990 | |
1388 | |
991 | #define ALLOC(len) \ |
1389 | #define ALLOC(len) \ |
… | |
… | |
1047 | |
1445 | |
1048 | --nready; |
1446 | --nready; |
1049 | |
1447 | |
1050 | X_UNLOCK (reqlock); |
1448 | X_UNLOCK (reqlock); |
1051 | |
1449 | |
1052 | errno = 0; /* strictly unnecessary */ |
1450 | if (req->type < 0) |
|
|
1451 | goto quit; |
1053 | |
1452 | |
1054 | if (!EIO_CANCELLED (req)) |
1453 | if (!EIO_CANCELLED (req)) |
1055 | switch (req->type) |
1454 | ETP_EXECUTE (self, req); |
1056 | { |
|
|
1057 | case EIO_READ: ALLOC (req->size); |
|
|
1058 | req->result = req->offs >= 0 |
|
|
1059 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
|
|
1060 | : read (req->int1, req->ptr2, req->size); break; |
|
|
1061 | case EIO_WRITE: req->result = req->offs >= 0 |
|
|
1062 | ? pwrite (req->int1, req->ptr2, req->size, req->offs) |
|
|
1063 | : write (req->int1, req->ptr2, req->size); break; |
|
|
1064 | |
|
|
1065 | case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; |
|
|
1066 | case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; |
|
|
1067 | |
|
|
1068 | case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
|
|
1069 | req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
|
|
1070 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
|
|
1071 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
|
|
1072 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
|
|
1073 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
|
|
1074 | |
|
|
1075 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
|
|
1076 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
|
|
1077 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
|
|
1078 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
|
|
1079 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
|
|
1080 | case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; |
|
|
1081 | |
|
|
1082 | case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; |
|
|
1083 | case EIO_CLOSE: req->result = close (req->int1); break; |
|
|
1084 | case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; |
|
|
1085 | case EIO_UNLINK: req->result = unlink (req->ptr1); break; |
|
|
1086 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
|
|
1087 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
|
|
1088 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
|
|
1089 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
|
|
1090 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
|
|
1091 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; |
|
|
1092 | |
|
|
1093 | case EIO_READLINK: ALLOC (NAME_MAX); |
|
|
1094 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
|
|
1095 | |
|
|
1096 | case EIO_SYNC: req->result = 0; sync (); break; |
|
|
1097 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
|
|
1098 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
|
|
1099 | |
|
|
1100 | case EIO_READDIR: eio__scandir (req, self); break; |
|
|
1101 | |
|
|
1102 | case EIO_BUSY: |
|
|
1103 | #ifdef _WIN32 |
|
|
1104 | Sleep (req->nv1 * 1000.); |
|
|
1105 | #else |
|
|
1106 | { |
|
|
1107 | struct timeval tv; |
|
|
1108 | |
|
|
1109 | tv.tv_sec = req->nv1; |
|
|
1110 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
|
|
1111 | |
|
|
1112 | req->result = select (0, 0, 0, 0, &tv); |
|
|
1113 | } |
|
|
1114 | #endif |
|
|
1115 | break; |
|
|
1116 | |
|
|
1117 | case EIO_UTIME: |
|
|
1118 | case EIO_FUTIME: |
|
|
1119 | { |
|
|
1120 | struct timeval tv[2]; |
|
|
1121 | struct timeval *times; |
|
|
1122 | |
|
|
1123 | if (req->nv1 != -1. || req->nv2 != -1.) |
|
|
1124 | { |
|
|
1125 | tv[0].tv_sec = req->nv1; |
|
|
1126 | tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; |
|
|
1127 | tv[1].tv_sec = req->nv2; |
|
|
1128 | tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; |
|
|
1129 | |
|
|
1130 | times = tv; |
|
|
1131 | } |
|
|
1132 | else |
|
|
1133 | times = 0; |
|
|
1134 | |
|
|
1135 | |
|
|
1136 | req->result = req->type == EIO_FUTIME |
|
|
1137 | ? futimes (req->int1, times) |
|
|
1138 | : utimes (req->ptr1, times); |
|
|
1139 | } |
|
|
1140 | |
|
|
1141 | case EIO_GROUP: |
|
|
1142 | case EIO_NOP: |
|
|
1143 | req->result = 0; |
|
|
1144 | break; |
|
|
1145 | |
|
|
1146 | case -1: |
|
|
1147 | goto quit; |
|
|
1148 | |
|
|
1149 | default: |
|
|
1150 | req->result = -1; |
|
|
1151 | break; |
|
|
1152 | } |
|
|
1153 | |
|
|
1154 | req->errorno = errno; |
|
|
1155 | |
1455 | |
1156 | X_LOCK (reslock); |
1456 | X_LOCK (reslock); |
1157 | |
1457 | |
1158 | ++npending; |
1458 | ++npending; |
1159 | |
1459 | |
… | |
… | |
1176 | |
1476 | |
1177 | /*****************************************************************************/ |
1477 | /*****************************************************************************/ |
1178 | |
1478 | |
1179 | int eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1479 | int eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1180 | { |
1480 | { |
1181 | etp_init (want_poll, done_poll); |
1481 | return etp_init (want_poll, done_poll); |
1182 | } |
1482 | } |
1183 | |
1483 | |
1184 | static void eio_api_destroy (eio_req *req) |
1484 | static void eio_api_destroy (eio_req *req) |
1185 | { |
1485 | { |
1186 | free (req); |
1486 | free (req); |
… | |
… | |
1208 | { \ |
1508 | { \ |
1209 | eio_api_destroy (req); \ |
1509 | eio_api_destroy (req); \ |
1210 | return 0; \ |
1510 | return 0; \ |
1211 | } |
1511 | } |
1212 | |
1512 | |
|
|
1513 | static void eio_execute (etp_worker *self, eio_req *req) |
|
|
1514 | { |
|
|
1515 | errno = 0; |
|
|
1516 | |
|
|
1517 | switch (req->type) |
|
|
1518 | { |
|
|
1519 | case EIO_READ: ALLOC (req->size); |
|
|
1520 | req->result = req->offs >= 0 |
|
|
1521 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
|
|
1522 | : read (req->int1, req->ptr2, req->size); break; |
|
|
1523 | case EIO_WRITE: req->result = req->offs >= 0 |
|
|
1524 | ? pwrite (req->int1, req->ptr2, req->size, req->offs) |
|
|
1525 | : write (req->int1, req->ptr2, req->size); break; |
|
|
1526 | |
|
|
1527 | case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; |
|
|
1528 | case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; |
|
|
1529 | |
|
|
1530 | case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
|
|
1531 | req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
|
|
1532 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
|
|
1533 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
|
|
1534 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
|
|
1535 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
|
|
1536 | |
|
|
1537 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
|
|
1538 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
|
|
1539 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
|
|
1540 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
|
|
1541 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
|
|
1542 | case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; |
|
|
1543 | |
|
|
1544 | case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; |
|
|
1545 | case EIO_CLOSE: req->result = close (req->int1); break; |
|
|
1546 | case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; |
|
|
1547 | case EIO_UNLINK: req->result = unlink (req->ptr1); break; |
|
|
1548 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
|
|
1549 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
|
|
1550 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
|
|
1551 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
|
|
1552 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
|
|
1553 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
|
|
1554 | |
|
|
1555 | case EIO_READLINK: ALLOC (NAME_MAX); |
|
|
1556 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
|
|
1557 | |
|
|
1558 | case EIO_SYNC: req->result = 0; sync (); break; |
|
|
1559 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
|
|
1560 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
|
|
1561 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
|
|
1562 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
|
|
1563 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
|
|
1564 | |
|
|
1565 | case EIO_READDIR: eio__scandir (req, self); break; |
|
|
1566 | |
|
|
1567 | case EIO_BUSY: |
|
|
1568 | #ifdef _WIN32 |
|
|
1569 | Sleep (req->nv1 * 1000.); |
|
|
1570 | #else |
|
|
1571 | { |
|
|
1572 | struct timeval tv; |
|
|
1573 | |
|
|
1574 | tv.tv_sec = req->nv1; |
|
|
1575 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
|
|
1576 | |
|
|
1577 | req->result = select (0, 0, 0, 0, &tv); |
|
|
1578 | } |
|
|
1579 | #endif |
|
|
1580 | break; |
|
|
1581 | |
|
|
1582 | case EIO_UTIME: |
|
|
1583 | case EIO_FUTIME: |
|
|
1584 | { |
|
|
1585 | struct timeval tv[2]; |
|
|
1586 | struct timeval *times; |
|
|
1587 | |
|
|
1588 | if (req->nv1 != -1. || req->nv2 != -1.) |
|
|
1589 | { |
|
|
1590 | tv[0].tv_sec = req->nv1; |
|
|
1591 | tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; |
|
|
1592 | tv[1].tv_sec = req->nv2; |
|
|
1593 | tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; |
|
|
1594 | |
|
|
1595 | times = tv; |
|
|
1596 | } |
|
|
1597 | else |
|
|
1598 | times = 0; |
|
|
1599 | |
|
|
1600 | |
|
|
1601 | req->result = req->type == EIO_FUTIME |
|
|
1602 | ? futimes (req->int1, times) |
|
|
1603 | : utimes (req->ptr1, times); |
|
|
1604 | } |
|
|
1605 | break; |
|
|
1606 | |
|
|
1607 | case EIO_GROUP: |
|
|
1608 | abort (); /* handled in eio_request */ |
|
|
1609 | |
|
|
1610 | case EIO_NOP: |
|
|
1611 | req->result = 0; |
|
|
1612 | break; |
|
|
1613 | |
|
|
1614 | case EIO_CUSTOM: |
|
|
1615 | ((void (*)(eio_req *))req->feed) (req); |
|
|
1616 | break; |
|
|
1617 | |
|
|
1618 | default: |
|
|
1619 | req->result = -1; |
|
|
1620 | break; |
|
|
1621 | } |
|
|
1622 | |
|
|
1623 | req->errorno = errno; |
|
|
1624 | } |
|
|
1625 | |
1213 | #ifndef EIO_NO_WRAPPERS |
1626 | #ifndef EIO_NO_WRAPPERS |
1214 | |
1627 | |
1215 | eio_req *eio_nop (int pri, eio_cb cb, void *data) |
1628 | eio_req *eio_nop (int pri, eio_cb cb, void *data) |
1216 | { |
1629 | { |
1217 | REQ (EIO_NOP); SEND; |
1630 | REQ (EIO_NOP); SEND; |
… | |
… | |
1228 | } |
1641 | } |
1229 | |
1642 | |
1230 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1643 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1231 | { |
1644 | { |
1232 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
1645 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
|
|
1646 | } |
|
|
1647 | |
|
|
1648 | eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
1649 | { |
|
|
1650 | REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
|
|
1651 | } |
|
|
1652 | |
|
|
1653 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
1654 | { |
|
|
1655 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
|
|
1656 | } |
|
|
1657 | |
|
|
1658 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
|
|
1659 | { |
|
|
1660 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
1233 | } |
1661 | } |
1234 | |
1662 | |
1235 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1663 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1236 | { |
1664 | { |
1237 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1665 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
… | |
… | |
1351 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
1779 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
1352 | { |
1780 | { |
1353 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
1781 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
1354 | } |
1782 | } |
1355 | |
1783 | |
1356 | eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) |
1784 | eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data) |
1357 | { |
1785 | { |
1358 | return eio__1path (EIO_READDIR, path, pri, cb, data); |
1786 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1359 | } |
1787 | } |
1360 | |
1788 | |
1361 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1789 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1362 | { |
1790 | { |
1363 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; |
1791 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
1364 | } |
1792 | } |
1365 | |
1793 | |
1366 | static eio_req * |
1794 | static eio_req * |
1367 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1795 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1368 | { |
1796 | { |
… | |
… | |
1392 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1820 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1393 | { |
1821 | { |
1394 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1822 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1395 | } |
1823 | } |
1396 | |
1824 | |
|
|
1825 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
|
|
1826 | { |
|
|
1827 | REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; |
|
|
1828 | } |
|
|
1829 | |
1397 | #endif |
1830 | #endif |
1398 | |
1831 | |
1399 | eio_req *eio_grp (eio_cb cb, void *data) |
1832 | eio_req *eio_grp (eio_cb cb, void *data) |
1400 | { |
1833 | { |
1401 | const int pri = EIO_PRI_MAX; |
1834 | const int pri = EIO_PRI_MAX; |
… | |
… | |
1427 | |
1860 | |
1428 | void eio_grp_add (eio_req *grp, eio_req *req) |
1861 | void eio_grp_add (eio_req *grp, eio_req *req) |
1429 | { |
1862 | { |
1430 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
1863 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
1431 | |
1864 | |
|
|
1865 | grp->flags |= EIO_FLAG_GROUPADD; |
|
|
1866 | |
1432 | ++grp->size; |
1867 | ++grp->size; |
1433 | req->grp = grp; |
1868 | req->grp = grp; |
1434 | |
1869 | |
1435 | req->grp_prev = 0; |
1870 | req->grp_prev = 0; |
1436 | req->grp_next = grp->grp_first; |
1871 | req->grp_next = grp->grp_first; |
… | |
… | |
1445 | /* misc garbage */ |
1880 | /* misc garbage */ |
1446 | |
1881 | |
1447 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1882 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1448 | { |
1883 | { |
1449 | etp_worker wrk; |
1884 | etp_worker wrk; |
|
|
1885 | ssize_t ret; |
1450 | |
1886 | |
1451 | wrk.dbuf = 0; |
1887 | wrk.dbuf = 0; |
1452 | |
1888 | |
1453 | eio__sendfile (ofd, ifd, offset, count, &wrk); |
1889 | ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
1454 | |
1890 | |
1455 | if (wrk.dbuf) |
1891 | if (wrk.dbuf) |
1456 | free (wrk.dbuf); |
1892 | free (wrk.dbuf); |
1457 | } |
|
|
1458 | |
1893 | |
|
|
1894 | return ret; |
|
|
1895 | } |
|
|
1896 | |