1 | /* |
1 | /* |
2 | * libeio implementation |
2 | * libeio implementation |
3 | * |
3 | * |
4 | * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> |
4 | * Copyright (c) 2007,2008,2009,2010 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> |
46 | #include <string.h> |
50 | #include <string.h> |
47 | #include <errno.h> |
51 | #include <errno.h> |
48 | #include <sys/types.h> |
52 | #include <sys/types.h> |
49 | #include <sys/stat.h> |
53 | #include <sys/stat.h> |
|
|
54 | #include <sys/statvfs.h> |
50 | #include <limits.h> |
55 | #include <limits.h> |
51 | #include <fcntl.h> |
56 | #include <fcntl.h> |
52 | #include <assert.h> |
57 | #include <assert.h> |
53 | |
58 | |
54 | #ifndef EIO_FINISH |
59 | #ifndef EIO_FINISH |
… | |
… | |
64 | #endif |
69 | #endif |
65 | |
70 | |
66 | #ifdef _WIN32 |
71 | #ifdef _WIN32 |
67 | |
72 | |
68 | /*doh*/ |
73 | /*doh*/ |
69 | |
|
|
70 | #else |
74 | #else |
71 | |
75 | |
72 | # include "config.h" |
76 | # include "config.h" |
73 | # include <sys/time.h> |
77 | # include <sys/time.h> |
74 | # include <sys/select.h> |
78 | # include <sys/select.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> |
79 | |
83 | |
|
|
84 | #if _POSIX_MEMLOCK || _POSIX_MAPPED_FILES |
|
|
85 | # include <sys/mman.h> |
|
|
86 | #endif |
|
|
87 | |
|
|
88 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
|
|
89 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
|
|
90 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
|
|
91 | # define D_INO(de) (de)->d_fileno |
|
|
92 | # define D_NAMLEN(de) (de)->d_namlen |
|
|
93 | # elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
|
|
94 | # define D_INO(de) (de)->d_ino |
|
|
95 | # endif |
|
|
96 | |
|
|
97 | #ifdef _D_EXACT_NAMLEN |
|
|
98 | # undef D_NAMLEN |
|
|
99 | # define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
|
|
100 | #endif |
|
|
101 | |
|
|
102 | # ifdef _DIRENT_HAVE_D_TYPE |
|
|
103 | # define D_TYPE(de) (de)->d_type |
|
|
104 | # endif |
|
|
105 | |
80 | # ifndef EIO_STRUCT_DIRENT |
106 | # ifndef EIO_STRUCT_DIRENT |
81 | # define EIO_STRUCT_DIRENT struct dirent |
107 | # define EIO_STRUCT_DIRENT struct dirent |
82 | # endif |
108 | # endif |
83 | |
109 | |
84 | #endif |
110 | #endif |
85 | |
111 | |
86 | #if HAVE_SENDFILE |
112 | #if HAVE_SENDFILE |
87 | # if __linux |
113 | # if __linux |
88 | # include <sys/sendfile.h> |
114 | # include <sys/sendfile.h> |
89 | # elif __freebsd |
115 | # elif __FreeBSD__ || defined __APPLE__ |
90 | # include <sys/socket.h> |
116 | # include <sys/socket.h> |
91 | # include <sys/uio.h> |
117 | # include <sys/uio.h> |
92 | # elif __hpux |
118 | # elif __hpux |
93 | # include <sys/socket.h> |
119 | # include <sys/socket.h> |
94 | # elif __solaris /* not yet */ |
120 | # elif __solaris |
95 | # include <sys/sendfile.h> |
121 | # include <sys/sendfile.h> |
96 | # else |
122 | # else |
97 | # error sendfile support requested but not available |
123 | # error sendfile support requested but not available |
98 | # endif |
124 | # endif |
99 | #endif |
125 | #endif |
100 | |
126 | |
|
|
127 | #ifndef D_TYPE |
|
|
128 | # define D_TYPE(de) 0 |
|
|
129 | #endif |
|
|
130 | #ifndef D_INO |
|
|
131 | # define D_INO(de) 0 |
|
|
132 | #endif |
|
|
133 | #ifndef D_NAMLEN |
|
|
134 | # define D_NAMLEN(de) strlen ((de)->d_name) |
|
|
135 | #endif |
|
|
136 | |
101 | /* number of seconds after which an idle threads exit */ |
137 | /* number of seconds after which an idle threads exit */ |
102 | #define IDLE_TIMEOUT 10 |
138 | #define IDLE_TIMEOUT 10 |
103 | |
139 | |
104 | /* used for struct dirent, AIX doesn't provide it */ |
140 | /* used for struct dirent, AIX doesn't provide it */ |
105 | #ifndef NAME_MAX |
141 | #ifndef NAME_MAX |
106 | # define NAME_MAX 4096 |
142 | # define NAME_MAX 4096 |
|
|
143 | #endif |
|
|
144 | |
|
|
145 | /* used for readlink etc. */ |
|
|
146 | #ifndef PATH_MAX |
|
|
147 | # define PATH_MAX 4096 |
107 | #endif |
148 | #endif |
108 | |
149 | |
109 | /* buffer size for various temporary buffers */ |
150 | /* buffer size for various temporary buffers */ |
110 | #define EIO_BUFSIZE 65536 |
151 | #define EIO_BUFSIZE 65536 |
111 | |
152 | |
… | |
… | |
117 | errno = ENOMEM; \ |
158 | errno = ENOMEM; \ |
118 | if (!eio_buf) \ |
159 | if (!eio_buf) \ |
119 | return -1; |
160 | return -1; |
120 | |
161 | |
121 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
162 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
|
|
163 | |
|
|
164 | /*****************************************************************************/ |
|
|
165 | |
|
|
166 | #if __GNUC__ >= 3 |
|
|
167 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
168 | #else |
|
|
169 | # define expect(expr,value) (expr) |
|
|
170 | #endif |
|
|
171 | |
|
|
172 | #define expect_false(expr) expect ((expr) != 0, 0) |
|
|
173 | #define expect_true(expr) expect ((expr) != 0, 1) |
122 | |
174 | |
123 | /*****************************************************************************/ |
175 | /*****************************************************************************/ |
124 | |
176 | |
125 | #define ETP_PRI_MIN EIO_PRI_MIN |
177 | #define ETP_PRI_MIN EIO_PRI_MIN |
126 | #define ETP_PRI_MAX EIO_PRI_MAX |
178 | #define ETP_PRI_MAX EIO_PRI_MAX |
… | |
… | |
144 | if (wrk->dirp) \ |
196 | if (wrk->dirp) \ |
145 | { \ |
197 | { \ |
146 | closedir (wrk->dirp); \ |
198 | closedir (wrk->dirp); \ |
147 | wrk->dirp = 0; \ |
199 | wrk->dirp = 0; \ |
148 | } |
200 | } |
|
|
201 | |
149 | #define ETP_WORKER_COMMON \ |
202 | #define ETP_WORKER_COMMON \ |
150 | void *dbuf; \ |
203 | void *dbuf; \ |
151 | DIR *dirp; |
204 | DIR *dirp; |
152 | |
205 | |
153 | /*****************************************************************************/ |
206 | /*****************************************************************************/ |
154 | |
207 | |
155 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
208 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
156 | |
209 | |
157 | /* calculcate time difference in ~1/EIO_TICKS of a second */ |
210 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
158 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
211 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
159 | { |
212 | { |
160 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
213 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
161 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
214 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
162 | } |
215 | } |
… | |
… | |
172 | static volatile unsigned int nreqs; /* reqlock */ |
225 | static volatile unsigned int nreqs; /* reqlock */ |
173 | static volatile unsigned int nready; /* reqlock */ |
226 | static volatile unsigned int nready; /* reqlock */ |
174 | static volatile unsigned int npending; /* reqlock */ |
227 | static volatile unsigned int npending; /* reqlock */ |
175 | static volatile unsigned int max_idle = 4; |
228 | static volatile unsigned int max_idle = 4; |
176 | |
229 | |
177 | static mutex_t wrklock = X_MUTEX_INIT; |
230 | static xmutex_t wrklock = X_MUTEX_INIT; |
178 | static mutex_t reslock = X_MUTEX_INIT; |
231 | static xmutex_t reslock = X_MUTEX_INIT; |
179 | static mutex_t reqlock = X_MUTEX_INIT; |
232 | static xmutex_t reqlock = X_MUTEX_INIT; |
180 | static cond_t reqwait = X_COND_INIT; |
233 | static xcond_t reqwait = X_COND_INIT; |
|
|
234 | |
|
|
235 | #if !HAVE_PREADWRITE |
|
|
236 | /* |
|
|
237 | * make our pread/pwrite emulation safe against themselves, but not against |
|
|
238 | * normal read/write by using a mutex. slows down execution a lot, |
|
|
239 | * but that's your problem, not mine. |
|
|
240 | */ |
|
|
241 | static xmutex_t preadwritelock = X_MUTEX_INIT; |
|
|
242 | #endif |
181 | |
243 | |
182 | typedef struct etp_worker |
244 | typedef struct etp_worker |
183 | { |
245 | { |
184 | /* locked by wrklock */ |
246 | /* locked by wrklock */ |
185 | struct etp_worker *prev, *next; |
247 | struct etp_worker *prev, *next; |
186 | |
248 | |
187 | thread_t tid; |
249 | xthread_t tid; |
188 | |
250 | |
189 | /* locked by reslock, reqlock or wrklock */ |
251 | /* locked by reslock, reqlock or wrklock */ |
190 | ETP_REQ *req; /* currently processed request */ |
252 | ETP_REQ *req; /* currently processed request */ |
191 | |
253 | |
192 | ETP_WORKER_COMMON |
254 | ETP_WORKER_COMMON |
… | |
… | |
330 | |
392 | |
331 | static void etp_atfork_child (void) |
393 | static void etp_atfork_child (void) |
332 | { |
394 | { |
333 | ETP_REQ *prv; |
395 | ETP_REQ *prv; |
334 | |
396 | |
335 | while (prv = reqq_shift (&req_queue)) |
397 | while ((prv = reqq_shift (&req_queue))) |
336 | ETP_DESTROY (prv); |
398 | ETP_DESTROY (prv); |
337 | |
399 | |
338 | while (prv = reqq_shift (&res_queue)) |
400 | while ((prv = reqq_shift (&res_queue))) |
339 | ETP_DESTROY (prv); |
401 | ETP_DESTROY (prv); |
340 | |
402 | |
341 | while (wrk_first.next != &wrk_first) |
403 | while (wrk_first.next != &wrk_first) |
342 | { |
404 | { |
343 | etp_worker *wrk = wrk_first.next; |
405 | etp_worker *wrk = wrk_first.next; |
… | |
… | |
402 | X_UNLOCK (wrklock); |
464 | X_UNLOCK (wrklock); |
403 | } |
465 | } |
404 | |
466 | |
405 | static void etp_maybe_start_thread (void) |
467 | static void etp_maybe_start_thread (void) |
406 | { |
468 | { |
407 | if (etp_nthreads () >= wanted) |
469 | if (expect_true (etp_nthreads () >= wanted)) |
408 | return; |
470 | return; |
409 | |
471 | |
410 | /* todo: maybe use idle here, but might be less exact */ |
472 | /* todo: maybe use idle here, but might be less exact */ |
411 | if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) |
473 | if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
412 | return; |
474 | return; |
413 | |
475 | |
414 | etp_start_thread (); |
476 | etp_start_thread (); |
415 | } |
477 | } |
416 | |
478 | |
… | |
… | |
469 | |
531 | |
470 | X_LOCK (reqlock); |
532 | X_LOCK (reqlock); |
471 | --nreqs; |
533 | --nreqs; |
472 | X_UNLOCK (reqlock); |
534 | X_UNLOCK (reqlock); |
473 | |
535 | |
474 | if (req->type == EIO_GROUP && req->size) |
536 | if (expect_false (req->type == EIO_GROUP && req->size)) |
475 | { |
537 | { |
476 | req->int1 = 1; /* mark request as delayed */ |
538 | req->int1 = 1; /* mark request as delayed */ |
477 | continue; |
539 | continue; |
478 | } |
540 | } |
479 | else |
541 | else |
480 | { |
542 | { |
481 | int res = ETP_FINISH (req); |
543 | int res = ETP_FINISH (req); |
482 | if (res) |
544 | if (expect_false (res)) |
483 | return res; |
545 | return res; |
484 | } |
546 | } |
485 | |
547 | |
486 | if (maxreqs && !--maxreqs) |
548 | if (expect_false (maxreqs && !--maxreqs)) |
487 | break; |
549 | break; |
488 | |
550 | |
489 | if (maxtime) |
551 | if (maxtime) |
490 | { |
552 | { |
491 | gettimeofday (&tv_now, 0); |
553 | gettimeofday (&tv_now, 0); |
… | |
… | |
510 | |
572 | |
511 | static void etp_submit (ETP_REQ *req) |
573 | static void etp_submit (ETP_REQ *req) |
512 | { |
574 | { |
513 | req->pri -= ETP_PRI_MIN; |
575 | req->pri -= ETP_PRI_MIN; |
514 | |
576 | |
515 | if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
577 | 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; |
578 | if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
517 | |
579 | |
|
|
580 | if (expect_false (req->type == EIO_GROUP)) |
|
|
581 | { |
|
|
582 | /* I hope this is worth it :/ */ |
518 | X_LOCK (reqlock); |
583 | X_LOCK (reqlock); |
519 | ++nreqs; |
584 | ++nreqs; |
|
|
585 | X_UNLOCK (reqlock); |
|
|
586 | |
|
|
587 | X_LOCK (reslock); |
|
|
588 | |
|
|
589 | ++npending; |
|
|
590 | |
|
|
591 | if (!reqq_push (&res_queue, req) && want_poll_cb) |
|
|
592 | want_poll_cb (); |
|
|
593 | |
|
|
594 | X_UNLOCK (reslock); |
|
|
595 | } |
|
|
596 | else |
|
|
597 | { |
|
|
598 | X_LOCK (reqlock); |
|
|
599 | ++nreqs; |
520 | ++nready; |
600 | ++nready; |
521 | reqq_push (&req_queue, req); |
601 | reqq_push (&req_queue, req); |
522 | X_COND_SIGNAL (reqwait); |
602 | X_COND_SIGNAL (reqwait); |
523 | X_UNLOCK (reqlock); |
603 | X_UNLOCK (reqlock); |
524 | |
604 | |
525 | etp_maybe_start_thread (); |
605 | etp_maybe_start_thread (); |
|
|
606 | } |
526 | } |
607 | } |
527 | |
608 | |
528 | static void etp_set_max_poll_time (double nseconds) |
609 | static void etp_set_max_poll_time (double nseconds) |
529 | { |
610 | { |
530 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
611 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
531 | max_poll_time = nseconds; |
612 | max_poll_time = nseconds * EIO_TICKS; |
532 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
613 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
533 | } |
614 | } |
534 | |
615 | |
535 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
616 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
536 | { |
617 | { |
… | |
… | |
565 | |
646 | |
566 | static void grp_try_feed (eio_req *grp) |
647 | static void grp_try_feed (eio_req *grp) |
567 | { |
648 | { |
568 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
649 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
569 | { |
650 | { |
570 | int old_len = grp->size; |
651 | grp->flags &= ~EIO_FLAG_GROUPADD; |
571 | |
652 | |
572 | EIO_FEED (grp); |
653 | EIO_FEED (grp); |
573 | |
654 | |
574 | /* stop if no progress has been made */ |
655 | /* stop if no progress has been made */ |
575 | if (old_len == grp->size) |
656 | if (!(grp->flags & EIO_FLAG_GROUPADD)) |
576 | { |
657 | { |
577 | grp->feed = 0; |
658 | grp->feed = 0; |
578 | break; |
659 | break; |
579 | } |
660 | } |
580 | } |
661 | } |
… | |
… | |
697 | |
778 | |
698 | /*****************************************************************************/ |
779 | /*****************************************************************************/ |
699 | /* work around various missing functions */ |
780 | /* work around various missing functions */ |
700 | |
781 | |
701 | #if !HAVE_PREADWRITE |
782 | #if !HAVE_PREADWRITE |
|
|
783 | # undef pread |
|
|
784 | # undef pwrite |
702 | # define pread eio__pread |
785 | # define pread eio__pread |
703 | # define pwrite eio__pwrite |
786 | # 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 | |
787 | |
712 | static ssize_t |
788 | static ssize_t |
713 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
789 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
714 | { |
790 | { |
715 | ssize_t res; |
791 | ssize_t res; |
… | |
… | |
733 | |
809 | |
734 | X_LOCK (preadwritelock); |
810 | X_LOCK (preadwritelock); |
735 | ooffset = lseek (fd, 0, SEEK_CUR); |
811 | ooffset = lseek (fd, 0, SEEK_CUR); |
736 | lseek (fd, offset, SEEK_SET); |
812 | lseek (fd, offset, SEEK_SET); |
737 | res = write (fd, buf, count); |
813 | res = write (fd, buf, count); |
738 | lseek (fd, offset, SEEK_SET); |
814 | lseek (fd, ooffset, SEEK_SET); |
739 | X_UNLOCK (preadwritelock); |
815 | X_UNLOCK (preadwritelock); |
740 | |
816 | |
741 | return res; |
817 | return res; |
742 | } |
818 | } |
743 | #endif |
819 | #endif |
744 | |
820 | |
745 | #ifndef HAVE_FUTIMES |
821 | #ifndef HAVE_FUTIMES |
746 | |
822 | |
|
|
823 | # undef utimes |
|
|
824 | # undef futimes |
747 | # define utimes(path,times) eio__utimes (path, times) |
825 | # define utimes(path,times) eio__utimes (path, times) |
748 | # define futimes(fd,times) eio__futimes (fd, times) |
826 | # define futimes(fd,times) eio__futimes (fd, times) |
749 | |
827 | |
750 | static int |
828 | static int |
751 | eio__utimes (const char *filename, const struct timeval times[2]) |
829 | eio__utimes (const char *filename, const struct timeval times[2]) |
… | |
… | |
770 | } |
848 | } |
771 | |
849 | |
772 | #endif |
850 | #endif |
773 | |
851 | |
774 | #if !HAVE_FDATASYNC |
852 | #if !HAVE_FDATASYNC |
|
|
853 | # undef fdatasync |
775 | # define fdatasync fsync |
854 | # define fdatasync(fd) fsync (fd) |
776 | #endif |
855 | #endif |
|
|
856 | |
|
|
857 | /* sync_file_range always needs emulation */ |
|
|
858 | int |
|
|
859 | eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
|
|
860 | { |
|
|
861 | #if HAVE_SYNC_FILE_RANGE |
|
|
862 | int res; |
|
|
863 | |
|
|
864 | if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE |
|
|
865 | || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE |
|
|
866 | || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER) |
|
|
867 | { |
|
|
868 | flags = 0 |
|
|
869 | | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0) |
|
|
870 | | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0) |
|
|
871 | | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); |
|
|
872 | } |
|
|
873 | |
|
|
874 | res = sync_file_range (fd, offset, nbytes, flags); |
|
|
875 | |
|
|
876 | if (!res || errno != ENOSYS) |
|
|
877 | return res; |
|
|
878 | #endif |
|
|
879 | |
|
|
880 | /* even though we could play tricks with the flags, it's better to always |
|
|
881 | * call fdatasync, as that matches the expectation of its users best */ |
|
|
882 | return fdatasync (fd); |
|
|
883 | } |
777 | |
884 | |
778 | #if !HAVE_READAHEAD |
885 | #if !HAVE_READAHEAD |
|
|
886 | # undef readahead |
779 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
887 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
780 | |
888 | |
781 | static ssize_t |
889 | static ssize_t |
782 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
890 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
783 | { |
891 | { |
… | |
… | |
810 | |
918 | |
811 | #if HAVE_SENDFILE |
919 | #if HAVE_SENDFILE |
812 | # if __linux |
920 | # if __linux |
813 | res = sendfile (ofd, ifd, &offset, count); |
921 | res = sendfile (ofd, ifd, &offset, count); |
814 | |
922 | |
815 | # elif __freebsd |
923 | # elif __FreeBSD__ |
816 | /* |
924 | /* |
817 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
925 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
818 | * wasted on making it similar to other I/O functions. |
926 | * wasted on making it similar to other I/O functions. |
819 | */ |
927 | */ |
820 | { |
928 | { |
821 | off_t sbytes; |
929 | off_t sbytes; |
822 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
930 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
823 | |
931 | |
824 | if (res < 0 && sbytes) |
932 | #if 0 /* according to the manpage, this is correct, but broken behaviour */ |
825 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
933 | /* freebsd' sendfile will return 0 on success */ |
|
|
934 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
|
|
935 | /* not on e.g. EIO or EPIPE - sounds broken */ |
|
|
936 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
|
|
937 | res = sbytes; |
|
|
938 | #endif |
|
|
939 | |
|
|
940 | /* according to source inspection, this is correct, and useful behaviour */ |
|
|
941 | if (sbytes) |
|
|
942 | res = sbytes; |
|
|
943 | } |
|
|
944 | |
|
|
945 | # elif defined (__APPLE__) |
|
|
946 | |
|
|
947 | { |
|
|
948 | off_t sbytes = count; |
|
|
949 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
|
|
950 | |
|
|
951 | /* according to the manpage, sbytes is always valid */ |
|
|
952 | if (sbytes) |
826 | res = sbytes; |
953 | res = sbytes; |
827 | } |
954 | } |
828 | |
955 | |
829 | # elif __hpux |
956 | # elif __hpux |
830 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
957 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
… | |
… | |
844 | if (res < 0 && sbytes) |
971 | if (res < 0 && sbytes) |
845 | res = sbytes; |
972 | res = sbytes; |
846 | } |
973 | } |
847 | |
974 | |
848 | # endif |
975 | # endif |
|
|
976 | |
|
|
977 | #elif defined (_WIN32) |
|
|
978 | |
|
|
979 | /* does not work, just for documentation of what would need to be done */ |
|
|
980 | { |
|
|
981 | HANDLE h = TO_SOCKET (ifd); |
|
|
982 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
|
|
983 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
|
|
984 | } |
|
|
985 | |
849 | #else |
986 | #else |
850 | res = -1; |
987 | res = -1; |
851 | errno = ENOSYS; |
988 | errno = ENOSYS; |
852 | #endif |
989 | #endif |
853 | |
990 | |
854 | if (res < 0 |
991 | if (res < 0 |
855 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
992 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
|
|
993 | /* BSDs */ |
|
|
994 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
|
|
995 | || errno == ENOTSUP |
|
|
996 | #endif |
|
|
997 | || errno == EOPNOTSUPP /* BSDs */ |
856 | #if __solaris |
998 | #if __solaris |
857 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
999 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
858 | #endif |
1000 | #endif |
859 | ) |
1001 | ) |
860 | ) |
1002 | ) |
… | |
… | |
891 | } |
1033 | } |
892 | |
1034 | |
893 | return res; |
1035 | return res; |
894 | } |
1036 | } |
895 | |
1037 | |
|
|
1038 | static signed char |
|
|
1039 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
|
|
1040 | { |
|
|
1041 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
|
|
1042 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; |
|
|
1043 | } |
|
|
1044 | |
|
|
1045 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
|
|
1046 | |
|
|
1047 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
|
|
1048 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
|
|
1049 | |
|
|
1050 | static void |
|
|
1051 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1052 | { |
|
|
1053 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
|
|
1054 | unsigned char *bit = bits; |
|
|
1055 | |
|
|
1056 | assert (CHAR_BIT == 8); |
|
|
1057 | assert (sizeof (eio_dirent) * 8 < 256); |
|
|
1058 | assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ |
|
|
1059 | assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ |
|
|
1060 | |
|
|
1061 | if (size <= EIO_SORT_FAST) |
|
|
1062 | return; |
|
|
1063 | |
|
|
1064 | /* first prepare an array of bits to test in our radix sort */ |
|
|
1065 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
|
|
1066 | /* inode_bits must contain all inodes ORed together */ |
|
|
1067 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
|
|
1068 | { |
|
|
1069 | ino_t endianness; |
|
|
1070 | int i, j; |
|
|
1071 | |
|
|
1072 | /* we store the byte offset of byte n into byte n of "endianness" */ |
|
|
1073 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1074 | ((unsigned char *)&endianness)[i] = i; |
|
|
1075 | |
|
|
1076 | *bit++ = 0; |
|
|
1077 | |
|
|
1078 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1079 | { |
|
|
1080 | /* shifting off the byte offsets out of "endianness" */ |
|
|
1081 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
|
|
1082 | endianness >>= 8; |
|
|
1083 | |
|
|
1084 | for (j = 0; j < 8; ++j) |
|
|
1085 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
|
|
1086 | *bit++ = offs + j; |
|
|
1087 | } |
|
|
1088 | |
|
|
1089 | for (j = 0; j < 8; ++j) |
|
|
1090 | if (score_bits & (1 << j)) |
|
|
1091 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
|
|
1092 | } |
|
|
1093 | |
|
|
1094 | /* now actually do the sorting (a variant of MSD radix sort) */ |
|
|
1095 | { |
|
|
1096 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
|
|
1097 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
|
|
1098 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
|
|
1099 | int stk_idx = 0; |
|
|
1100 | |
|
|
1101 | base_stk [stk_idx] = dents; |
|
|
1102 | end_stk [stk_idx] = dents + size; |
|
|
1103 | bit_stk [stk_idx] = bit - 1; |
|
|
1104 | |
|
|
1105 | do |
|
|
1106 | { |
|
|
1107 | base = base_stk [stk_idx]; |
|
|
1108 | end = end_stk [stk_idx]; |
|
|
1109 | bit = bit_stk [stk_idx]; |
|
|
1110 | |
|
|
1111 | for (;;) |
|
|
1112 | { |
|
|
1113 | unsigned char O = *bit >> 3; |
|
|
1114 | unsigned char M = 1 << (*bit & 7); |
|
|
1115 | |
|
|
1116 | eio_dirent *a = base; |
|
|
1117 | eio_dirent *b = end; |
|
|
1118 | |
|
|
1119 | if (b - a < EIO_SORT_CUTOFF) |
|
|
1120 | break; |
|
|
1121 | |
|
|
1122 | /* now bit-partition the array on the bit */ |
|
|
1123 | /* this ugly asymmetric loop seems to perform much better than typical */ |
|
|
1124 | /* partition algos found in the literature */ |
|
|
1125 | do |
|
|
1126 | if (!(((unsigned char *)a)[O] & M)) |
|
|
1127 | ++a; |
|
|
1128 | else if (!(((unsigned char *)--b)[O] & M)) |
|
|
1129 | { |
|
|
1130 | eio_dirent tmp = *a; *a = *b; *b = tmp; |
|
|
1131 | ++a; |
|
|
1132 | } |
|
|
1133 | while (b > a); |
|
|
1134 | |
|
|
1135 | /* next bit, or stop, if no bits left in this path */ |
|
|
1136 | if (!*--bit) |
|
|
1137 | break; |
|
|
1138 | |
|
|
1139 | base_stk [stk_idx] = a; |
|
|
1140 | end_stk [stk_idx] = end; |
|
|
1141 | bit_stk [stk_idx] = bit; |
|
|
1142 | ++stk_idx; |
|
|
1143 | |
|
|
1144 | end = a; |
|
|
1145 | } |
|
|
1146 | } |
|
|
1147 | while (stk_idx--); |
|
|
1148 | } |
|
|
1149 | } |
|
|
1150 | |
|
|
1151 | static void |
|
|
1152 | eio_dent_insertion_sort (eio_dirent *dents, int size) |
|
|
1153 | { |
|
|
1154 | /* first move the smallest element to the front, to act as a sentinel */ |
|
|
1155 | { |
|
|
1156 | int i; |
|
|
1157 | eio_dirent *min = dents; |
|
|
1158 | |
|
|
1159 | /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
|
|
1160 | for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
|
|
1161 | if (EIO_DENT_CMP (dents [i], <, *min)) |
|
|
1162 | min = &dents [i]; |
|
|
1163 | |
|
|
1164 | /* swap elements 0 and j (minimum) */ |
|
|
1165 | { |
|
|
1166 | eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
|
|
1167 | } |
|
|
1168 | } |
|
|
1169 | |
|
|
1170 | /* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
|
|
1171 | { |
|
|
1172 | eio_dirent *i, *j; |
|
|
1173 | |
|
|
1174 | for (i = dents + 1; i < dents + size; ++i) |
|
|
1175 | { |
|
|
1176 | eio_dirent value = *i; |
|
|
1177 | |
|
|
1178 | for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
|
|
1179 | j [1] = j [0]; |
|
|
1180 | |
|
|
1181 | j [1] = value; |
|
|
1182 | } |
|
|
1183 | } |
|
|
1184 | } |
|
|
1185 | |
|
|
1186 | static void |
|
|
1187 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1188 | { |
|
|
1189 | if (size <= 1) |
|
|
1190 | return; /* our insertion sort relies on size > 0 */ |
|
|
1191 | |
|
|
1192 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
|
|
1193 | /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
|
|
1194 | eio_dent_radix_sort (dents, size, score_bits, inode_bits); |
|
|
1195 | |
|
|
1196 | /* use an insertion sort at the end, or for small arrays, */ |
|
|
1197 | /* as insertion sort is more efficient for small partitions */ |
|
|
1198 | eio_dent_insertion_sort (dents, size); |
|
|
1199 | } |
|
|
1200 | |
896 | /* read a full directory */ |
1201 | /* read a full directory */ |
897 | static void |
1202 | static void |
898 | eio__scandir (eio_req *req, etp_worker *self) |
1203 | eio__scandir (eio_req *req, etp_worker *self) |
899 | { |
1204 | { |
900 | DIR *dirp; |
1205 | DIR *dirp; |
901 | EIO_STRUCT_DIRENT *entp; |
1206 | EIO_STRUCT_DIRENT *entp; |
902 | char *name, *names; |
1207 | char *name, *names; |
903 | int memlen = 4096; |
1208 | int namesalloc = 4096; |
904 | int memofs = 0; |
1209 | int namesoffs = 0; |
|
|
1210 | int flags = req->int1; |
|
|
1211 | eio_dirent *dents = 0; |
|
|
1212 | int dentalloc = 128; |
905 | int res = 0; |
1213 | int dentoffs = 0; |
|
|
1214 | ino_t inode_bits = 0; |
|
|
1215 | |
|
|
1216 | req->result = -1; |
|
|
1217 | |
|
|
1218 | if (!(flags & EIO_READDIR_DENTS)) |
|
|
1219 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
906 | |
1220 | |
907 | X_LOCK (wrklock); |
1221 | X_LOCK (wrklock); |
|
|
1222 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
908 | self->dirp = dirp = opendir (req->ptr1); |
1223 | self->dirp = dirp = opendir (req->ptr1); |
909 | req->flags |= EIO_FLAG_PTR2_FREE; |
1224 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
|
|
1225 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
910 | req->ptr2 = names = malloc (memlen); |
1226 | req->ptr2 = names = malloc (namesalloc); |
911 | X_UNLOCK (wrklock); |
1227 | X_UNLOCK (wrklock); |
912 | |
1228 | |
913 | if (dirp && names) |
1229 | if (dirp && names && (!flags || dents)) |
914 | for (;;) |
1230 | for (;;) |
915 | { |
1231 | { |
916 | errno = 0; |
1232 | errno = 0; |
917 | entp = readdir (dirp); |
1233 | entp = readdir (dirp); |
918 | |
1234 | |
919 | if (!entp) |
1235 | if (!entp) |
|
|
1236 | { |
|
|
1237 | if (errno) |
|
|
1238 | break; |
|
|
1239 | |
|
|
1240 | /* sort etc. */ |
|
|
1241 | req->int1 = flags; |
|
|
1242 | req->result = dentoffs; |
|
|
1243 | |
|
|
1244 | if (flags & EIO_READDIR_STAT_ORDER) |
|
|
1245 | eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ |
|
|
1246 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1247 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
|
|
1248 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
|
|
1249 | else |
|
|
1250 | { |
|
|
1251 | /* in this case, all is known, and we just put dirs first and sort them */ |
|
|
1252 | eio_dirent *oth = dents + dentoffs; |
|
|
1253 | eio_dirent *dir = dents; |
|
|
1254 | |
|
|
1255 | /* now partition dirs to the front, and non-dirs to the back */ |
|
|
1256 | /* by walking from both sides and swapping if necessary */ |
|
|
1257 | /* also clear score, so it doesn't influence sorting */ |
|
|
1258 | while (oth > dir) |
|
|
1259 | { |
|
|
1260 | if (dir->type == EIO_DT_DIR) |
|
|
1261 | ++dir; |
|
|
1262 | else if ((--oth)->type == EIO_DT_DIR) |
|
|
1263 | { |
|
|
1264 | eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
|
|
1265 | |
|
|
1266 | ++dir; |
|
|
1267 | } |
|
|
1268 | } |
|
|
1269 | |
|
|
1270 | /* now sort the dirs only */ |
|
|
1271 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
|
|
1272 | } |
|
|
1273 | |
920 | break; |
1274 | break; |
|
|
1275 | } |
921 | |
1276 | |
|
|
1277 | /* now add the entry to our list(s) */ |
922 | name = entp->d_name; |
1278 | name = entp->d_name; |
923 | |
1279 | |
|
|
1280 | /* skip . and .. entries */ |
924 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1281 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
925 | { |
1282 | { |
926 | int len = strlen (name) + 1; |
1283 | int len = D_NAMLEN (entp) + 1; |
927 | |
1284 | |
928 | res++; |
1285 | while (expect_false (namesoffs + len > namesalloc)) |
929 | |
|
|
930 | while (memofs + len > memlen) |
|
|
931 | { |
1286 | { |
932 | memlen *= 2; |
1287 | namesalloc *= 2; |
933 | X_LOCK (wrklock); |
1288 | X_LOCK (wrklock); |
934 | req->ptr2 = names = realloc (names, memlen); |
1289 | req->ptr2 = names = realloc (names, namesalloc); |
935 | X_UNLOCK (wrklock); |
1290 | X_UNLOCK (wrklock); |
936 | |
1291 | |
937 | if (!names) |
1292 | if (!names) |
938 | break; |
1293 | break; |
939 | } |
1294 | } |
940 | |
1295 | |
941 | memcpy (names + memofs, name, len); |
1296 | memcpy (names + namesoffs, name, len); |
|
|
1297 | |
|
|
1298 | if (dents) |
|
|
1299 | { |
|
|
1300 | struct eio_dirent *ent; |
|
|
1301 | |
|
|
1302 | if (expect_false (dentoffs == dentalloc)) |
|
|
1303 | { |
|
|
1304 | dentalloc *= 2; |
|
|
1305 | X_LOCK (wrklock); |
|
|
1306 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
|
|
1307 | X_UNLOCK (wrklock); |
|
|
1308 | |
|
|
1309 | if (!dents) |
|
|
1310 | break; |
|
|
1311 | } |
|
|
1312 | |
|
|
1313 | ent = dents + dentoffs; |
|
|
1314 | |
|
|
1315 | ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
|
|
1316 | ent->namelen = len - 1; |
|
|
1317 | ent->inode = D_INO (entp); |
|
|
1318 | |
|
|
1319 | inode_bits |= ent->inode; |
|
|
1320 | |
|
|
1321 | switch (D_TYPE (entp)) |
|
|
1322 | { |
|
|
1323 | default: |
|
|
1324 | ent->type = EIO_DT_UNKNOWN; |
|
|
1325 | flags |= EIO_READDIR_FOUND_UNKNOWN; |
|
|
1326 | break; |
|
|
1327 | |
|
|
1328 | #ifdef DT_FIFO |
|
|
1329 | case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
|
|
1330 | #endif |
|
|
1331 | #ifdef DT_CHR |
|
|
1332 | case DT_CHR: ent->type = EIO_DT_CHR; break; |
|
|
1333 | #endif |
|
|
1334 | #ifdef DT_MPC |
|
|
1335 | case DT_MPC: ent->type = EIO_DT_MPC; break; |
|
|
1336 | #endif |
|
|
1337 | #ifdef DT_DIR |
|
|
1338 | case DT_DIR: ent->type = EIO_DT_DIR; break; |
|
|
1339 | #endif |
|
|
1340 | #ifdef DT_NAM |
|
|
1341 | case DT_NAM: ent->type = EIO_DT_NAM; break; |
|
|
1342 | #endif |
|
|
1343 | #ifdef DT_BLK |
|
|
1344 | case DT_BLK: ent->type = EIO_DT_BLK; break; |
|
|
1345 | #endif |
|
|
1346 | #ifdef DT_MPB |
|
|
1347 | case DT_MPB: ent->type = EIO_DT_MPB; break; |
|
|
1348 | #endif |
|
|
1349 | #ifdef DT_REG |
|
|
1350 | case DT_REG: ent->type = EIO_DT_REG; break; |
|
|
1351 | #endif |
|
|
1352 | #ifdef DT_NWK |
|
|
1353 | case DT_NWK: ent->type = EIO_DT_NWK; break; |
|
|
1354 | #endif |
|
|
1355 | #ifdef DT_CMP |
|
|
1356 | case DT_CMP: ent->type = EIO_DT_CMP; break; |
|
|
1357 | #endif |
|
|
1358 | #ifdef DT_LNK |
|
|
1359 | case DT_LNK: ent->type = EIO_DT_LNK; break; |
|
|
1360 | #endif |
|
|
1361 | #ifdef DT_SOCK |
|
|
1362 | case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
|
|
1363 | #endif |
|
|
1364 | #ifdef DT_DOOR |
|
|
1365 | case DT_DOOR: ent->type = EIO_DT_DOOR; break; |
|
|
1366 | #endif |
|
|
1367 | #ifdef DT_WHT |
|
|
1368 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
|
|
1369 | #endif |
|
|
1370 | } |
|
|
1371 | |
|
|
1372 | ent->score = 7; |
|
|
1373 | |
|
|
1374 | if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1375 | { |
|
|
1376 | if (ent->type == EIO_DT_UNKNOWN) |
|
|
1377 | { |
|
|
1378 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
|
|
1379 | ent->score = 1; |
|
|
1380 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
|
|
1381 | ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
|
|
1382 | } |
|
|
1383 | else if (ent->type == EIO_DT_DIR) |
|
|
1384 | ent->score = 0; |
|
|
1385 | } |
|
|
1386 | } |
|
|
1387 | |
942 | memofs += len; |
1388 | namesoffs += len; |
|
|
1389 | ++dentoffs; |
|
|
1390 | } |
|
|
1391 | |
|
|
1392 | if (EIO_CANCELLED (req)) |
|
|
1393 | { |
|
|
1394 | errno = ECANCELED; |
|
|
1395 | break; |
943 | } |
1396 | } |
944 | } |
1397 | } |
|
|
1398 | } |
945 | |
1399 | |
946 | if (errno) |
1400 | #ifdef PAGESIZE |
947 | res = -1; |
1401 | # define eio_pagesize() PAGESIZE |
|
|
1402 | #else |
|
|
1403 | static intptr_t |
|
|
1404 | eio_pagesize (void) |
|
|
1405 | { |
|
|
1406 | static intptr_t page; |
|
|
1407 | |
|
|
1408 | if (!page) |
|
|
1409 | page = sysconf (_SC_PAGESIZE); |
|
|
1410 | |
|
|
1411 | return page; |
|
|
1412 | } |
|
|
1413 | #endif |
|
|
1414 | |
|
|
1415 | static void |
|
|
1416 | eio_page_align (void **addr, size_t *length) |
|
|
1417 | { |
|
|
1418 | intptr_t mask = eio_pagesize () - 1; |
|
|
1419 | |
|
|
1420 | /* round down addr */ |
|
|
1421 | intptr_t adj = mask & (intptr_t)*addr; |
|
|
1422 | |
|
|
1423 | *addr = (void *)((intptr_t)*addr - adj); |
|
|
1424 | *length += adj; |
|
|
1425 | |
|
|
1426 | /* round up length */ |
|
|
1427 | *length = (*length + mask) & ~mask; |
|
|
1428 | } |
|
|
1429 | |
|
|
1430 | #if !_POSIX_MEMLOCK |
|
|
1431 | # define eio__mlock(a,b) ((errno = ENOSYS), -1) |
|
|
1432 | # define eio__mlockall(a) ((errno = ENOSYS), -1) |
|
|
1433 | #else |
|
|
1434 | |
|
|
1435 | static int |
|
|
1436 | eio__mlock (void *addr, size_t length) |
|
|
1437 | { |
|
|
1438 | eio_page_align (&addr, &length); |
|
|
1439 | |
|
|
1440 | return mlock (addr, length); |
|
|
1441 | } |
|
|
1442 | |
|
|
1443 | static int |
|
|
1444 | eio__mlockall (int flags) |
|
|
1445 | { |
|
|
1446 | #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 |
|
|
1447 | extern int mallopt (int, int); |
|
|
1448 | mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ |
|
|
1449 | #endif |
|
|
1450 | |
|
|
1451 | if (EIO_MCL_CURRENT != MCL_CURRENT |
|
|
1452 | || EIO_MCL_FUTURE != MCL_FUTURE) |
|
|
1453 | { |
|
|
1454 | flags = 0 |
|
|
1455 | | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0) |
|
|
1456 | | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0); |
|
|
1457 | } |
|
|
1458 | |
|
|
1459 | return mlockall (flags); |
|
|
1460 | } |
|
|
1461 | #endif |
|
|
1462 | |
|
|
1463 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
|
|
1464 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
|
|
1465 | #else |
|
|
1466 | |
|
|
1467 | int |
|
|
1468 | eio__msync (void *mem, size_t len, int flags) |
|
|
1469 | { |
|
|
1470 | eio_page_align (&mem, &len); |
|
|
1471 | |
|
|
1472 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1473 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1474 | || EIO_MS_SYNC != MS_SYNC) |
|
|
1475 | { |
|
|
1476 | flags = 0 |
|
|
1477 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
|
|
1478 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
|
|
1479 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
|
|
1480 | } |
|
|
1481 | |
|
|
1482 | return msync (mem, len, flags); |
|
|
1483 | } |
|
|
1484 | |
|
|
1485 | #endif |
|
|
1486 | |
|
|
1487 | int |
|
|
1488 | eio__mtouch (void *mem, size_t len, int flags) |
|
|
1489 | { |
|
|
1490 | eio_page_align (&mem, &len); |
|
|
1491 | |
948 | |
1492 | { |
949 | req->result = res; |
1493 | intptr_t addr = (intptr_t)mem; |
|
|
1494 | intptr_t end = addr + len; |
|
|
1495 | intptr_t page = eio_pagesize (); |
|
|
1496 | |
|
|
1497 | if (addr < end) |
|
|
1498 | if (flags & EIO_MT_MODIFY) /* modify */ |
|
|
1499 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
|
|
1500 | else |
|
|
1501 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
|
|
1502 | } |
|
|
1503 | |
|
|
1504 | return 0; |
950 | } |
1505 | } |
951 | |
1506 | |
952 | /*****************************************************************************/ |
1507 | /*****************************************************************************/ |
953 | |
1508 | |
954 | #define ALLOC(len) \ |
1509 | #define ALLOC(len) \ |
… | |
… | |
1075 | return 0; \ |
1630 | return 0; \ |
1076 | } |
1631 | } |
1077 | |
1632 | |
1078 | static void eio_execute (etp_worker *self, eio_req *req) |
1633 | static void eio_execute (etp_worker *self, eio_req *req) |
1079 | { |
1634 | { |
1080 | errno = 0; |
|
|
1081 | |
|
|
1082 | switch (req->type) |
1635 | switch (req->type) |
1083 | { |
1636 | { |
1084 | case EIO_READ: ALLOC (req->size); |
1637 | case EIO_READ: ALLOC (req->size); |
1085 | req->result = req->offs >= 0 |
1638 | req->result = req->offs >= 0 |
1086 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
1639 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
… | |
… | |
1096 | req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1649 | req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1097 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1650 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1098 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1651 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1099 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1652 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1100 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1653 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
|
|
1654 | |
|
|
1655 | case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1656 | req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1657 | case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1658 | req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
1101 | |
1659 | |
1102 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1660 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1103 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1661 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1104 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1662 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1105 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1663 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
… | |
… | |
1115 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1673 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1116 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1674 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1117 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1675 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1118 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
1676 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
1119 | |
1677 | |
1120 | case EIO_READLINK: ALLOC (NAME_MAX); |
1678 | case EIO_READLINK: ALLOC (PATH_MAX); |
1121 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1679 | req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; |
1122 | |
1680 | |
1123 | case EIO_SYNC: req->result = 0; sync (); break; |
1681 | case EIO_SYNC: req->result = 0; sync (); break; |
1124 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1682 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1125 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1683 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
|
|
1684 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
|
|
1685 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
|
|
1686 | case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; |
|
|
1687 | case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; |
|
|
1688 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1126 | |
1689 | |
1127 | case EIO_READDIR: eio__scandir (req, self); break; |
1690 | case EIO_READDIR: eio__scandir (req, self); break; |
1128 | |
1691 | |
1129 | case EIO_BUSY: |
1692 | case EIO_BUSY: |
1130 | #ifdef _WIN32 |
1693 | #ifdef _WIN32 |
1131 | Sleep (req->nv1 * 1000.); |
1694 | Sleep (req->nv1 * 1e3); |
1132 | #else |
1695 | #else |
1133 | { |
1696 | { |
1134 | struct timeval tv; |
1697 | struct timeval tv; |
1135 | |
1698 | |
1136 | tv.tv_sec = req->nv1; |
1699 | tv.tv_sec = req->nv1; |
1137 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
1700 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6; |
1138 | |
1701 | |
1139 | req->result = select (0, 0, 0, 0, &tv); |
1702 | req->result = select (0, 0, 0, 0, &tv); |
1140 | } |
1703 | } |
1141 | #endif |
1704 | #endif |
1142 | break; |
1705 | break; |
… | |
… | |
1157 | times = tv; |
1720 | times = tv; |
1158 | } |
1721 | } |
1159 | else |
1722 | else |
1160 | times = 0; |
1723 | times = 0; |
1161 | |
1724 | |
1162 | |
|
|
1163 | req->result = req->type == EIO_FUTIME |
1725 | req->result = req->type == EIO_FUTIME |
1164 | ? futimes (req->int1, times) |
1726 | ? futimes (req->int1, times) |
1165 | : utimes (req->ptr1, times); |
1727 | : utimes (req->ptr1, times); |
1166 | } |
1728 | } |
|
|
1729 | break; |
1167 | |
1730 | |
1168 | case EIO_GROUP: |
1731 | case EIO_GROUP: |
|
|
1732 | abort (); /* handled in eio_request */ |
|
|
1733 | |
1169 | case EIO_NOP: |
1734 | case EIO_NOP: |
1170 | req->result = 0; |
1735 | req->result = 0; |
1171 | break; |
1736 | break; |
1172 | |
1737 | |
1173 | case EIO_CUSTOM: |
1738 | case EIO_CUSTOM: |
1174 | req->feed (req); |
1739 | ((void (*)(eio_req *))req->feed) (req); |
1175 | break; |
1740 | break; |
1176 | |
1741 | |
1177 | default: |
1742 | default: |
|
|
1743 | errno = ENOSYS; |
1178 | req->result = -1; |
1744 | req->result = -1; |
1179 | break; |
1745 | break; |
1180 | } |
1746 | } |
1181 | |
1747 | |
1182 | req->errorno = errno; |
1748 | req->errorno = errno; |
… | |
… | |
1202 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1768 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1203 | { |
1769 | { |
1204 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
1770 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
1205 | } |
1771 | } |
1206 | |
1772 | |
|
|
1773 | eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
1774 | { |
|
|
1775 | REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
|
|
1776 | } |
|
|
1777 | |
|
|
1778 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
1779 | { |
|
|
1780 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
|
|
1781 | } |
|
|
1782 | |
|
|
1783 | eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data) |
|
|
1784 | { |
|
|
1785 | REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND; |
|
|
1786 | } |
|
|
1787 | |
|
|
1788 | eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) |
|
|
1789 | { |
|
|
1790 | REQ (EIO_MLOCKALL); req->int1 = flags; SEND; |
|
|
1791 | } |
|
|
1792 | |
|
|
1793 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
|
|
1794 | { |
|
|
1795 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
|
|
1796 | } |
|
|
1797 | |
1207 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1798 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1208 | { |
1799 | { |
1209 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1800 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1210 | } |
1801 | } |
1211 | |
1802 | |
… | |
… | |
1230 | } |
1821 | } |
1231 | |
1822 | |
1232 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1823 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1233 | { |
1824 | { |
1234 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
1825 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
|
|
1826 | } |
|
|
1827 | |
|
|
1828 | eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data) |
|
|
1829 | { |
|
|
1830 | REQ (EIO_FSTATVFS); req->int1 = fd; SEND; |
1235 | } |
1831 | } |
1236 | |
1832 | |
1237 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1833 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1238 | { |
1834 | { |
1239 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
1835 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
… | |
… | |
1313 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1909 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1314 | { |
1910 | { |
1315 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1911 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1316 | } |
1912 | } |
1317 | |
1913 | |
|
|
1914 | eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data) |
|
|
1915 | { |
|
|
1916 | return eio__1path (EIO_STATVFS, path, pri, cb, data); |
|
|
1917 | } |
|
|
1918 | |
1318 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1919 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1319 | { |
1920 | { |
1320 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1921 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1321 | } |
1922 | } |
1322 | |
1923 | |
1323 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
1924 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
1324 | { |
1925 | { |
1325 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
1926 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
1326 | } |
1927 | } |
1327 | |
1928 | |
1328 | eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) |
1929 | eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data) |
1329 | { |
1930 | { |
1330 | return eio__1path (EIO_READDIR, path, pri, cb, data); |
1931 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1331 | } |
1932 | } |
1332 | |
1933 | |
1333 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1934 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1334 | { |
1935 | { |
1335 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
1936 | 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); |
1967 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1367 | } |
1968 | } |
1368 | |
1969 | |
1369 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
1970 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
1370 | { |
1971 | { |
1371 | REQ (EIO_CUSTOM); req->feed = execute; SEND; |
1972 | REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; |
1372 | } |
1973 | } |
1373 | |
1974 | |
1374 | #endif |
1975 | #endif |
1375 | |
1976 | |
1376 | eio_req *eio_grp (eio_cb cb, void *data) |
1977 | eio_req *eio_grp (eio_cb cb, void *data) |
… | |
… | |
1404 | |
2005 | |
1405 | void eio_grp_add (eio_req *grp, eio_req *req) |
2006 | void eio_grp_add (eio_req *grp, eio_req *req) |
1406 | { |
2007 | { |
1407 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
2008 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
1408 | |
2009 | |
|
|
2010 | grp->flags |= EIO_FLAG_GROUPADD; |
|
|
2011 | |
1409 | ++grp->size; |
2012 | ++grp->size; |
1410 | req->grp = grp; |
2013 | req->grp = grp; |
1411 | |
2014 | |
1412 | req->grp_prev = 0; |
2015 | req->grp_prev = 0; |
1413 | req->grp_next = grp->grp_first; |
2016 | req->grp_next = grp->grp_first; |
… | |
… | |
1422 | /* misc garbage */ |
2025 | /* misc garbage */ |
1423 | |
2026 | |
1424 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
2027 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1425 | { |
2028 | { |
1426 | etp_worker wrk; |
2029 | etp_worker wrk; |
|
|
2030 | ssize_t ret; |
1427 | |
2031 | |
1428 | wrk.dbuf = 0; |
2032 | wrk.dbuf = 0; |
1429 | |
2033 | |
1430 | eio__sendfile (ofd, ifd, offset, count, &wrk); |
2034 | ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
1431 | |
2035 | |
1432 | if (wrk.dbuf) |
2036 | if (wrk.dbuf) |
1433 | free (wrk.dbuf); |
2037 | free (wrk.dbuf); |
1434 | } |
|
|
1435 | |
2038 | |
|
|
2039 | return ret; |
|
|
2040 | } |
|
|
2041 | |