1 | /* |
1 | /* |
2 | * libeio implementation |
2 | * libeio implementation |
3 | * |
3 | * |
4 | * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> |
4 | * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de> |
5 | * All rights reserved. |
5 | * All rights reserved. |
6 | * |
6 | * |
7 | * Redistribution and use in source and binary forms, with or without modifica- |
7 | * Redistribution and use in source and binary forms, with or without modifica- |
8 | * tion, are permitted provided that the following conditions are met: |
8 | * tion, are permitted provided that the following conditions are met: |
9 | * |
9 | * |
… | |
… | |
35 | * and other provisions required by the GPL. If you do not delete the |
35 | * and other provisions required by the GPL. If you do not delete the |
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 | #ifndef _WIN32 |
|
|
41 | # include "config.h" |
|
|
42 | #endif |
|
|
43 | |
40 | #include "eio.h" |
44 | #include "eio.h" |
|
|
45 | |
|
|
46 | #ifdef EIO_STACKSIZE |
|
|
47 | # define XTHREAD_STACKSIZE EIO_STACKSIZE |
|
|
48 | #endif |
41 | #include "xthread.h" |
49 | #include "xthread.h" |
42 | |
50 | |
43 | #include <errno.h> |
51 | #include <errno.h> |
44 | #include <stddef.h> |
52 | #include <stddef.h> |
45 | #include <stdlib.h> |
53 | #include <stdlib.h> |
46 | #include <string.h> |
54 | #include <string.h> |
47 | #include <errno.h> |
55 | #include <errno.h> |
48 | #include <sys/types.h> |
56 | #include <sys/types.h> |
49 | #include <sys/stat.h> |
57 | #include <sys/stat.h> |
|
|
58 | #include <sys/statvfs.h> |
50 | #include <limits.h> |
59 | #include <limits.h> |
51 | #include <fcntl.h> |
60 | #include <fcntl.h> |
52 | #include <assert.h> |
61 | #include <assert.h> |
53 | |
62 | |
|
|
63 | /* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ |
|
|
64 | /* intptr_t only comes form stdint.h, says idiot openbsd coder */ |
|
|
65 | #if HAVE_STDINT_H |
|
|
66 | # include <stdint.h> |
|
|
67 | #endif |
|
|
68 | |
54 | #ifndef EIO_FINISH |
69 | #ifndef EIO_FINISH |
55 | # define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
70 | # define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
56 | #endif |
71 | #endif |
57 | |
72 | |
58 | #ifndef EIO_DESTROY |
73 | #ifndef EIO_DESTROY |
… | |
… | |
64 | #endif |
79 | #endif |
65 | |
80 | |
66 | #ifdef _WIN32 |
81 | #ifdef _WIN32 |
67 | |
82 | |
68 | /*doh*/ |
83 | /*doh*/ |
69 | |
|
|
70 | #else |
84 | #else |
71 | |
85 | |
72 | # include "config.h" |
|
|
73 | # include <sys/time.h> |
86 | # include <sys/time.h> |
74 | # include <sys/select.h> |
87 | # include <sys/select.h> |
75 | # include <unistd.h> |
88 | # include <unistd.h> |
76 | # include <utime.h> |
89 | # include <utime.h> |
77 | # include <signal.h> |
90 | # include <signal.h> |
78 | # include <dirent.h> |
91 | # include <dirent.h> |
79 | |
92 | |
|
|
93 | #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES |
|
|
94 | # include <sys/mman.h> |
|
|
95 | #endif |
|
|
96 | |
|
|
97 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
|
|
98 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
|
|
99 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
|
|
100 | # define D_INO(de) (de)->d_fileno |
|
|
101 | # define D_NAMLEN(de) (de)->d_namlen |
|
|
102 | # elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
|
|
103 | # define D_INO(de) (de)->d_ino |
|
|
104 | # endif |
|
|
105 | |
|
|
106 | #ifdef _D_EXACT_NAMLEN |
|
|
107 | # undef D_NAMLEN |
|
|
108 | # define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
|
|
109 | #endif |
|
|
110 | |
|
|
111 | # ifdef _DIRENT_HAVE_D_TYPE |
|
|
112 | # define D_TYPE(de) (de)->d_type |
|
|
113 | # endif |
|
|
114 | |
80 | # ifndef EIO_STRUCT_DIRENT |
115 | # ifndef EIO_STRUCT_DIRENT |
81 | # define EIO_STRUCT_DIRENT struct dirent |
116 | # define EIO_STRUCT_DIRENT struct dirent |
82 | # endif |
117 | # endif |
83 | |
118 | |
84 | #endif |
119 | #endif |
85 | |
120 | |
86 | #if HAVE_SENDFILE |
121 | #if HAVE_SENDFILE |
87 | # if __linux |
122 | # if __linux |
88 | # include <sys/sendfile.h> |
123 | # include <sys/sendfile.h> |
89 | # elif __freebsd |
124 | # elif __FreeBSD__ || defined __APPLE__ |
90 | # include <sys/socket.h> |
125 | # include <sys/socket.h> |
91 | # include <sys/uio.h> |
126 | # include <sys/uio.h> |
92 | # elif __hpux |
127 | # elif __hpux |
93 | # include <sys/socket.h> |
128 | # include <sys/socket.h> |
94 | # elif __solaris /* not yet */ |
129 | # elif __solaris |
95 | # include <sys/sendfile.h> |
130 | # include <sys/sendfile.h> |
96 | # else |
131 | # else |
97 | # error sendfile support requested but not available |
132 | # error sendfile support requested but not available |
98 | # endif |
133 | # endif |
99 | #endif |
134 | #endif |
100 | |
135 | |
101 | /* number of seconds after which an idle threads exit */ |
136 | #ifndef D_TYPE |
102 | #define IDLE_TIMEOUT 10 |
137 | # define D_TYPE(de) 0 |
|
|
138 | #endif |
|
|
139 | #ifndef D_INO |
|
|
140 | # define D_INO(de) 0 |
|
|
141 | #endif |
|
|
142 | #ifndef D_NAMLEN |
|
|
143 | # define D_NAMLEN(de) strlen ((de)->d_name) |
|
|
144 | #endif |
103 | |
145 | |
104 | /* used for struct dirent, AIX doesn't provide it */ |
146 | /* used for struct dirent, AIX doesn't provide it */ |
105 | #ifndef NAME_MAX |
147 | #ifndef NAME_MAX |
106 | # define NAME_MAX 4096 |
148 | # define NAME_MAX 4096 |
|
|
149 | #endif |
|
|
150 | |
|
|
151 | /* used for readlink etc. */ |
|
|
152 | #ifndef PATH_MAX |
|
|
153 | # define PATH_MAX 4096 |
107 | #endif |
154 | #endif |
108 | |
155 | |
109 | /* buffer size for various temporary buffers */ |
156 | /* buffer size for various temporary buffers */ |
110 | #define EIO_BUFSIZE 65536 |
157 | #define EIO_BUFSIZE 65536 |
111 | |
158 | |
… | |
… | |
117 | errno = ENOMEM; \ |
164 | errno = ENOMEM; \ |
118 | if (!eio_buf) \ |
165 | if (!eio_buf) \ |
119 | return -1; |
166 | return -1; |
120 | |
167 | |
121 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
168 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
|
|
169 | |
|
|
170 | /*****************************************************************************/ |
|
|
171 | |
|
|
172 | #if __GNUC__ >= 3 |
|
|
173 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
174 | #else |
|
|
175 | # define expect(expr,value) (expr) |
|
|
176 | #endif |
|
|
177 | |
|
|
178 | #define expect_false(expr) expect ((expr) != 0, 0) |
|
|
179 | #define expect_true(expr) expect ((expr) != 0, 1) |
122 | |
180 | |
123 | /*****************************************************************************/ |
181 | /*****************************************************************************/ |
124 | |
182 | |
125 | #define ETP_PRI_MIN EIO_PRI_MIN |
183 | #define ETP_PRI_MIN EIO_PRI_MIN |
126 | #define ETP_PRI_MAX EIO_PRI_MAX |
184 | #define ETP_PRI_MAX EIO_PRI_MAX |
… | |
… | |
144 | if (wrk->dirp) \ |
202 | if (wrk->dirp) \ |
145 | { \ |
203 | { \ |
146 | closedir (wrk->dirp); \ |
204 | closedir (wrk->dirp); \ |
147 | wrk->dirp = 0; \ |
205 | wrk->dirp = 0; \ |
148 | } |
206 | } |
|
|
207 | |
149 | #define ETP_WORKER_COMMON \ |
208 | #define ETP_WORKER_COMMON \ |
150 | void *dbuf; \ |
209 | void *dbuf; \ |
151 | DIR *dirp; |
210 | DIR *dirp; |
152 | |
211 | |
153 | /*****************************************************************************/ |
212 | /*****************************************************************************/ |
154 | |
213 | |
155 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
214 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
156 | |
215 | |
157 | /* calculcate time difference in ~1/EIO_TICKS of a second */ |
216 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
158 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
217 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
159 | { |
218 | { |
160 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
219 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
161 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
220 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
162 | } |
221 | } |
… | |
… | |
170 | static unsigned int max_poll_reqs; /* reslock */ |
229 | static unsigned int max_poll_reqs; /* reslock */ |
171 | |
230 | |
172 | static volatile unsigned int nreqs; /* reqlock */ |
231 | static volatile unsigned int nreqs; /* reqlock */ |
173 | static volatile unsigned int nready; /* reqlock */ |
232 | static volatile unsigned int nready; /* reqlock */ |
174 | static volatile unsigned int npending; /* reqlock */ |
233 | static volatile unsigned int npending; /* reqlock */ |
175 | static volatile unsigned int max_idle = 4; |
234 | static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ |
|
|
235 | static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ |
176 | |
236 | |
177 | static mutex_t wrklock = X_MUTEX_INIT; |
237 | static xmutex_t wrklock; |
178 | static mutex_t reslock = X_MUTEX_INIT; |
238 | static xmutex_t reslock; |
179 | static mutex_t reqlock = X_MUTEX_INIT; |
239 | static xmutex_t reqlock; |
180 | static cond_t reqwait = X_COND_INIT; |
240 | static xcond_t reqwait; |
181 | |
241 | |
182 | #if !HAVE_PREADWRITE |
242 | #if !HAVE_PREADWRITE |
183 | /* |
243 | /* |
184 | * make our pread/pwrite emulation safe against themselves, but not against |
244 | * make our pread/pwrite emulation safe against themselves, but not against |
185 | * normal read/write by using a mutex. slows down execution a lot, |
245 | * normal read/write by using a mutex. slows down execution a lot, |
186 | * but that's your problem, not mine. |
246 | * but that's your problem, not mine. |
187 | */ |
247 | */ |
188 | static mutex_t preadwritelock = X_MUTEX_INIT; |
248 | static xmutex_t preadwritelock = X_MUTEX_INIT; |
189 | #endif |
249 | #endif |
190 | |
250 | |
191 | typedef struct etp_worker |
251 | typedef struct etp_worker |
192 | { |
252 | { |
193 | /* locked by wrklock */ |
253 | /* locked by wrklock */ |
194 | struct etp_worker *prev, *next; |
254 | struct etp_worker *prev, *next; |
195 | |
255 | |
196 | thread_t tid; |
256 | xthread_t tid; |
197 | |
257 | |
198 | /* locked by reslock, reqlock or wrklock */ |
258 | /* locked by reslock, reqlock or wrklock */ |
199 | ETP_REQ *req; /* currently processed request */ |
259 | ETP_REQ *req; /* currently processed request */ |
200 | |
260 | |
201 | ETP_WORKER_COMMON |
261 | ETP_WORKER_COMMON |
… | |
… | |
315 | } |
375 | } |
316 | |
376 | |
317 | abort (); |
377 | abort (); |
318 | } |
378 | } |
319 | |
379 | |
|
|
380 | static void etp_thread_init (void) |
|
|
381 | { |
|
|
382 | X_MUTEX_CREATE (wrklock); |
|
|
383 | X_MUTEX_CREATE (reslock); |
|
|
384 | X_MUTEX_CREATE (reqlock); |
|
|
385 | X_COND_CREATE (reqwait); |
|
|
386 | } |
|
|
387 | |
320 | static void etp_atfork_prepare (void) |
388 | static void etp_atfork_prepare (void) |
321 | { |
389 | { |
322 | X_LOCK (wrklock); |
390 | X_LOCK (wrklock); |
323 | X_LOCK (reqlock); |
391 | X_LOCK (reqlock); |
324 | X_LOCK (reslock); |
392 | X_LOCK (reslock); |
… | |
… | |
362 | idle = 0; |
430 | idle = 0; |
363 | nreqs = 0; |
431 | nreqs = 0; |
364 | nready = 0; |
432 | nready = 0; |
365 | npending = 0; |
433 | npending = 0; |
366 | |
434 | |
367 | etp_atfork_parent (); |
435 | etp_thread_init (); |
368 | } |
436 | } |
369 | |
437 | |
370 | static void |
438 | static void |
371 | etp_once_init (void) |
439 | etp_once_init (void) |
372 | { |
440 | { |
|
|
441 | etp_thread_init (); |
373 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
442 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
374 | } |
443 | } |
375 | |
444 | |
376 | static int |
445 | static int |
377 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
446 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
… | |
… | |
411 | X_UNLOCK (wrklock); |
480 | X_UNLOCK (wrklock); |
412 | } |
481 | } |
413 | |
482 | |
414 | static void etp_maybe_start_thread (void) |
483 | static void etp_maybe_start_thread (void) |
415 | { |
484 | { |
416 | if (etp_nthreads () >= wanted) |
485 | if (expect_true (etp_nthreads () >= wanted)) |
417 | return; |
486 | return; |
418 | |
487 | |
419 | /* todo: maybe use idle here, but might be less exact */ |
488 | /* todo: maybe use idle here, but might be less exact */ |
420 | if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) |
489 | if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
421 | return; |
490 | return; |
422 | |
491 | |
423 | etp_start_thread (); |
492 | etp_start_thread (); |
424 | } |
493 | } |
425 | |
494 | |
… | |
… | |
478 | |
547 | |
479 | X_LOCK (reqlock); |
548 | X_LOCK (reqlock); |
480 | --nreqs; |
549 | --nreqs; |
481 | X_UNLOCK (reqlock); |
550 | X_UNLOCK (reqlock); |
482 | |
551 | |
483 | if (req->type == EIO_GROUP && req->size) |
552 | if (expect_false (req->type == EIO_GROUP && req->size)) |
484 | { |
553 | { |
485 | req->int1 = 1; /* mark request as delayed */ |
554 | req->int1 = 1; /* mark request as delayed */ |
486 | continue; |
555 | continue; |
487 | } |
556 | } |
488 | else |
557 | else |
489 | { |
558 | { |
490 | int res = ETP_FINISH (req); |
559 | int res = ETP_FINISH (req); |
491 | if (res) |
560 | if (expect_false (res)) |
492 | return res; |
561 | return res; |
493 | } |
562 | } |
494 | |
563 | |
495 | if (maxreqs && !--maxreqs) |
564 | if (expect_false (maxreqs && !--maxreqs)) |
496 | break; |
565 | break; |
497 | |
566 | |
498 | if (maxtime) |
567 | if (maxtime) |
499 | { |
568 | { |
500 | gettimeofday (&tv_now, 0); |
569 | gettimeofday (&tv_now, 0); |
… | |
… | |
519 | |
588 | |
520 | static void etp_submit (ETP_REQ *req) |
589 | static void etp_submit (ETP_REQ *req) |
521 | { |
590 | { |
522 | req->pri -= ETP_PRI_MIN; |
591 | req->pri -= ETP_PRI_MIN; |
523 | |
592 | |
524 | if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
593 | if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
525 | if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
594 | if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
526 | |
595 | |
|
|
596 | if (expect_false (req->type == EIO_GROUP)) |
|
|
597 | { |
|
|
598 | /* I hope this is worth it :/ */ |
527 | X_LOCK (reqlock); |
599 | X_LOCK (reqlock); |
528 | ++nreqs; |
600 | ++nreqs; |
|
|
601 | X_UNLOCK (reqlock); |
|
|
602 | |
|
|
603 | X_LOCK (reslock); |
|
|
604 | |
|
|
605 | ++npending; |
|
|
606 | |
|
|
607 | if (!reqq_push (&res_queue, req) && want_poll_cb) |
|
|
608 | want_poll_cb (); |
|
|
609 | |
|
|
610 | X_UNLOCK (reslock); |
|
|
611 | } |
|
|
612 | else |
|
|
613 | { |
|
|
614 | X_LOCK (reqlock); |
|
|
615 | ++nreqs; |
529 | ++nready; |
616 | ++nready; |
530 | reqq_push (&req_queue, req); |
617 | reqq_push (&req_queue, req); |
531 | X_COND_SIGNAL (reqwait); |
618 | X_COND_SIGNAL (reqwait); |
532 | X_UNLOCK (reqlock); |
619 | X_UNLOCK (reqlock); |
533 | |
620 | |
534 | etp_maybe_start_thread (); |
621 | etp_maybe_start_thread (); |
|
|
622 | } |
535 | } |
623 | } |
536 | |
624 | |
537 | static void etp_set_max_poll_time (double nseconds) |
625 | static void etp_set_max_poll_time (double nseconds) |
538 | { |
626 | { |
539 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
627 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
540 | max_poll_time = nseconds; |
628 | max_poll_time = nseconds * EIO_TICKS; |
541 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
629 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
542 | } |
630 | } |
543 | |
631 | |
544 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
632 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
545 | { |
633 | { |
… | |
… | |
549 | } |
637 | } |
550 | |
638 | |
551 | static void etp_set_max_idle (unsigned int nthreads) |
639 | static void etp_set_max_idle (unsigned int nthreads) |
552 | { |
640 | { |
553 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
641 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
554 | max_idle = nthreads <= 0 ? 1 : nthreads; |
642 | max_idle = nthreads; |
|
|
643 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
|
|
644 | } |
|
|
645 | |
|
|
646 | static void etp_set_idle_timeout (unsigned int seconds) |
|
|
647 | { |
|
|
648 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
|
|
649 | idle_timeout = seconds; |
555 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
650 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
556 | } |
651 | } |
557 | |
652 | |
558 | static void etp_set_min_parallel (unsigned int nthreads) |
653 | static void etp_set_min_parallel (unsigned int nthreads) |
559 | { |
654 | { |
… | |
… | |
574 | |
669 | |
575 | static void grp_try_feed (eio_req *grp) |
670 | static void grp_try_feed (eio_req *grp) |
576 | { |
671 | { |
577 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
672 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
578 | { |
673 | { |
579 | int old_len = grp->size; |
674 | grp->flags &= ~EIO_FLAG_GROUPADD; |
580 | |
675 | |
581 | EIO_FEED (grp); |
676 | EIO_FEED (grp); |
582 | |
677 | |
583 | /* stop if no progress has been made */ |
678 | /* stop if no progress has been made */ |
584 | if (old_len == grp->size) |
679 | if (!(grp->flags & EIO_FLAG_GROUPADD)) |
585 | { |
680 | { |
586 | grp->feed = 0; |
681 | grp->feed = 0; |
587 | break; |
682 | break; |
588 | } |
683 | } |
589 | } |
684 | } |
… | |
… | |
687 | void eio_set_max_idle (unsigned int nthreads) |
782 | void eio_set_max_idle (unsigned int nthreads) |
688 | { |
783 | { |
689 | etp_set_max_idle (nthreads); |
784 | etp_set_max_idle (nthreads); |
690 | } |
785 | } |
691 | |
786 | |
|
|
787 | void eio_set_idle_timeout (unsigned int seconds) |
|
|
788 | { |
|
|
789 | etp_set_idle_timeout (seconds); |
|
|
790 | } |
|
|
791 | |
692 | void eio_set_min_parallel (unsigned int nthreads) |
792 | void eio_set_min_parallel (unsigned int nthreads) |
693 | { |
793 | { |
694 | etp_set_min_parallel (nthreads); |
794 | etp_set_min_parallel (nthreads); |
695 | } |
795 | } |
696 | |
796 | |
… | |
… | |
706 | |
806 | |
707 | /*****************************************************************************/ |
807 | /*****************************************************************************/ |
708 | /* work around various missing functions */ |
808 | /* work around various missing functions */ |
709 | |
809 | |
710 | #if !HAVE_PREADWRITE |
810 | #if !HAVE_PREADWRITE |
|
|
811 | # undef pread |
|
|
812 | # undef pwrite |
711 | # define pread eio__pread |
813 | # define pread eio__pread |
712 | # define pwrite eio__pwrite |
814 | # define pwrite eio__pwrite |
713 | |
815 | |
714 | static ssize_t |
816 | static ssize_t |
715 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
817 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
… | |
… | |
735 | |
837 | |
736 | X_LOCK (preadwritelock); |
838 | X_LOCK (preadwritelock); |
737 | ooffset = lseek (fd, 0, SEEK_CUR); |
839 | ooffset = lseek (fd, 0, SEEK_CUR); |
738 | lseek (fd, offset, SEEK_SET); |
840 | lseek (fd, offset, SEEK_SET); |
739 | res = write (fd, buf, count); |
841 | res = write (fd, buf, count); |
740 | lseek (fd, offset, SEEK_SET); |
842 | lseek (fd, ooffset, SEEK_SET); |
741 | X_UNLOCK (preadwritelock); |
843 | X_UNLOCK (preadwritelock); |
742 | |
844 | |
743 | return res; |
845 | return res; |
744 | } |
846 | } |
745 | #endif |
847 | #endif |
746 | |
848 | |
747 | #ifndef HAVE_FUTIMES |
849 | #ifndef HAVE_UTIMES |
748 | |
850 | |
|
|
851 | # undef utimes |
749 | # define utimes(path,times) eio__utimes (path, times) |
852 | # define utimes(path,times) eio__utimes (path, times) |
750 | # define futimes(fd,times) eio__futimes (fd, times) |
|
|
751 | |
853 | |
752 | static int |
854 | static int |
753 | eio__utimes (const char *filename, const struct timeval times[2]) |
855 | eio__utimes (const char *filename, const struct timeval times[2]) |
754 | { |
856 | { |
755 | if (times) |
857 | if (times) |
… | |
… | |
763 | } |
865 | } |
764 | else |
866 | else |
765 | return utime (filename, 0); |
867 | return utime (filename, 0); |
766 | } |
868 | } |
767 | |
869 | |
|
|
870 | #endif |
|
|
871 | |
|
|
872 | #ifndef HAVE_FUTIMES |
|
|
873 | |
|
|
874 | # undef futimes |
|
|
875 | # define futimes(fd,times) eio__futimes (fd, times) |
|
|
876 | |
768 | static int eio__futimes (int fd, const struct timeval tv[2]) |
877 | static int eio__futimes (int fd, const struct timeval tv[2]) |
769 | { |
878 | { |
770 | errno = ENOSYS; |
879 | errno = ENOSYS; |
771 | return -1; |
880 | return -1; |
772 | } |
881 | } |
773 | |
882 | |
774 | #endif |
883 | #endif |
775 | |
884 | |
776 | #if !HAVE_FDATASYNC |
885 | #if !HAVE_FDATASYNC |
|
|
886 | # undef fdatasync |
777 | # define fdatasync fsync |
887 | # define fdatasync(fd) fsync (fd) |
778 | #endif |
888 | #endif |
|
|
889 | |
|
|
890 | /* sync_file_range always needs emulation */ |
|
|
891 | int |
|
|
892 | eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
|
|
893 | { |
|
|
894 | #if HAVE_SYNC_FILE_RANGE |
|
|
895 | int res; |
|
|
896 | |
|
|
897 | if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE |
|
|
898 | || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE |
|
|
899 | || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER) |
|
|
900 | { |
|
|
901 | flags = 0 |
|
|
902 | | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0) |
|
|
903 | | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0) |
|
|
904 | | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); |
|
|
905 | } |
|
|
906 | |
|
|
907 | res = sync_file_range (fd, offset, nbytes, flags); |
|
|
908 | |
|
|
909 | if (!res || errno != ENOSYS) |
|
|
910 | return res; |
|
|
911 | #endif |
|
|
912 | |
|
|
913 | /* even though we could play tricks with the flags, it's better to always |
|
|
914 | * call fdatasync, as that matches the expectation of its users best */ |
|
|
915 | return fdatasync (fd); |
|
|
916 | } |
779 | |
917 | |
780 | #if !HAVE_READAHEAD |
918 | #if !HAVE_READAHEAD |
|
|
919 | # undef readahead |
781 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
920 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
782 | |
921 | |
783 | static ssize_t |
922 | static ssize_t |
784 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
923 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
785 | { |
924 | { |
… | |
… | |
803 | |
942 | |
804 | /* sendfile always needs emulation */ |
943 | /* sendfile always needs emulation */ |
805 | static ssize_t |
944 | static ssize_t |
806 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
945 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
807 | { |
946 | { |
|
|
947 | ssize_t written = 0; |
808 | ssize_t res; |
948 | ssize_t res; |
809 | |
949 | |
810 | if (!count) |
950 | if (!count) |
811 | return 0; |
951 | return 0; |
812 | |
952 | |
|
|
953 | for (;;) |
|
|
954 | { |
813 | #if HAVE_SENDFILE |
955 | #if HAVE_SENDFILE |
814 | # if __linux |
956 | # if __linux |
|
|
957 | off_t soffset = offset; |
815 | res = sendfile (ofd, ifd, &offset, count); |
958 | res = sendfile (ofd, ifd, &soffset, count); |
816 | |
959 | |
817 | # elif __freebsd |
960 | # elif __FreeBSD__ |
818 | /* |
961 | /* |
819 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
962 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
820 | * wasted on making it similar to other I/O functions. |
963 | * wasted on making it similar to other I/O functions. |
821 | */ |
964 | */ |
822 | { |
|
|
823 | off_t sbytes; |
965 | off_t sbytes; |
824 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
966 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
825 | |
967 | |
826 | if (res < 0 && sbytes) |
968 | #if 0 /* according to the manpage, this is correct, but broken behaviour */ |
827 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
969 | /* freebsd' sendfile will return 0 on success */ |
|
|
970 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
|
|
971 | /* not on e.g. EIO or EPIPE - sounds broken */ |
|
|
972 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
828 | res = sbytes; |
973 | res = sbytes; |
829 | } |
974 | #endif |
|
|
975 | |
|
|
976 | /* according to source inspection, this is correct, and useful behaviour */ |
|
|
977 | if (sbytes) |
|
|
978 | res = sbytes; |
|
|
979 | |
|
|
980 | # elif defined (__APPLE__) |
|
|
981 | off_t sbytes = count; |
|
|
982 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
|
|
983 | |
|
|
984 | /* according to the manpage, sbytes is always valid */ |
|
|
985 | if (sbytes) |
|
|
986 | res = sbytes; |
830 | |
987 | |
831 | # elif __hpux |
988 | # elif __hpux |
832 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
989 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
833 | |
990 | |
834 | # elif __solaris |
991 | # elif __solaris |
835 | { |
|
|
836 | struct sendfilevec vec; |
992 | struct sendfilevec vec; |
837 | size_t sbytes; |
993 | size_t sbytes; |
838 | |
994 | |
839 | vec.sfv_fd = ifd; |
995 | vec.sfv_fd = ifd; |
840 | vec.sfv_flag = 0; |
996 | vec.sfv_flag = 0; |
841 | vec.sfv_off = offset; |
997 | vec.sfv_off = offset; |
842 | vec.sfv_len = count; |
998 | vec.sfv_len = count; |
843 | |
999 | |
844 | res = sendfilev (ofd, &vec, 1, &sbytes); |
1000 | res = sendfilev (ofd, &vec, 1, &sbytes); |
845 | |
1001 | |
846 | if (res < 0 && sbytes) |
1002 | if (res < 0 && sbytes) |
847 | res = sbytes; |
1003 | res = sbytes; |
848 | } |
|
|
849 | |
1004 | |
850 | # endif |
1005 | # endif |
|
|
1006 | |
|
|
1007 | #elif defined (_WIN32) |
|
|
1008 | /* does not work, just for documentation of what would need to be done */ |
|
|
1009 | /* actually, cannot be done like this, as TransmitFile changes the file offset, */ |
|
|
1010 | /* libeio guarantees that the file offset does not change, and windows */ |
|
|
1011 | /* has no way to get an independent handle to the same file description */ |
|
|
1012 | HANDLE h = TO_SOCKET (ifd); |
|
|
1013 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
|
|
1014 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
|
|
1015 | |
851 | #else |
1016 | #else |
852 | res = -1; |
1017 | res = -1; |
853 | errno = ENOSYS; |
1018 | errno = ENOSYS; |
854 | #endif |
1019 | #endif |
855 | |
1020 | |
|
|
1021 | /* we assume sendfile can copy at least 128mb in one go */ |
|
|
1022 | if (res <= 128 * 1024 * 1024) |
|
|
1023 | { |
|
|
1024 | if (res > 0) |
|
|
1025 | written += res; |
|
|
1026 | |
|
|
1027 | if (written) |
|
|
1028 | return written; |
|
|
1029 | |
|
|
1030 | break; |
|
|
1031 | } |
|
|
1032 | else |
|
|
1033 | { |
|
|
1034 | /* if we requested more, then probably the kernel was lazy */ |
|
|
1035 | written += res; |
|
|
1036 | offset += res; |
|
|
1037 | count -= res; |
|
|
1038 | |
|
|
1039 | if (!count) |
|
|
1040 | return written; |
|
|
1041 | } |
|
|
1042 | } |
|
|
1043 | |
856 | if (res < 0 |
1044 | if (res < 0 |
857 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
1045 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
|
|
1046 | /* BSDs */ |
|
|
1047 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
|
|
1048 | || errno == ENOTSUP |
|
|
1049 | #endif |
|
|
1050 | || errno == EOPNOTSUPP /* BSDs */ |
858 | #if __solaris |
1051 | #if __solaris |
859 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
1052 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
860 | #endif |
1053 | #endif |
861 | ) |
1054 | ) |
862 | ) |
1055 | ) |
… | |
… | |
893 | } |
1086 | } |
894 | |
1087 | |
895 | return res; |
1088 | return res; |
896 | } |
1089 | } |
897 | |
1090 | |
|
|
1091 | static signed char |
|
|
1092 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
|
|
1093 | { |
|
|
1094 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
|
|
1095 | : a->inode < b->inode ? -1 |
|
|
1096 | : a->inode > b->inode ? 1 |
|
|
1097 | : 0; |
|
|
1098 | } |
|
|
1099 | |
|
|
1100 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
|
|
1101 | |
|
|
1102 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
|
|
1103 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
|
|
1104 | |
|
|
1105 | static void |
|
|
1106 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1107 | { |
|
|
1108 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
|
|
1109 | unsigned char *bit = bits; |
|
|
1110 | |
|
|
1111 | assert (CHAR_BIT == 8); |
|
|
1112 | assert (sizeof (eio_dirent) * 8 < 256); |
|
|
1113 | assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ |
|
|
1114 | assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */ |
|
|
1115 | |
|
|
1116 | if (size <= EIO_SORT_FAST) |
|
|
1117 | return; |
|
|
1118 | |
|
|
1119 | /* first prepare an array of bits to test in our radix sort */ |
|
|
1120 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
|
|
1121 | /* inode_bits must contain all inodes ORed together */ |
|
|
1122 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
|
|
1123 | { |
|
|
1124 | ino_t endianness; |
|
|
1125 | int i, j; |
|
|
1126 | |
|
|
1127 | /* we store the byte offset of byte n into byte n of "endianness" */ |
|
|
1128 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1129 | ((unsigned char *)&endianness)[i] = i; |
|
|
1130 | |
|
|
1131 | *bit++ = 0; |
|
|
1132 | |
|
|
1133 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1134 | { |
|
|
1135 | /* shifting off the byte offsets out of "endianness" */ |
|
|
1136 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
|
|
1137 | endianness >>= 8; |
|
|
1138 | |
|
|
1139 | for (j = 0; j < 8; ++j) |
|
|
1140 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
|
|
1141 | *bit++ = offs + j; |
|
|
1142 | } |
|
|
1143 | |
|
|
1144 | for (j = 0; j < 8; ++j) |
|
|
1145 | if (score_bits & (1 << j)) |
|
|
1146 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
|
|
1147 | } |
|
|
1148 | |
|
|
1149 | /* now actually do the sorting (a variant of MSD radix sort) */ |
|
|
1150 | { |
|
|
1151 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
|
|
1152 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
|
|
1153 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
|
|
1154 | int stk_idx = 0; |
|
|
1155 | |
|
|
1156 | base_stk [stk_idx] = dents; |
|
|
1157 | end_stk [stk_idx] = dents + size; |
|
|
1158 | bit_stk [stk_idx] = bit - 1; |
|
|
1159 | |
|
|
1160 | do |
|
|
1161 | { |
|
|
1162 | base = base_stk [stk_idx]; |
|
|
1163 | end = end_stk [stk_idx]; |
|
|
1164 | bit = bit_stk [stk_idx]; |
|
|
1165 | |
|
|
1166 | for (;;) |
|
|
1167 | { |
|
|
1168 | unsigned char O = *bit >> 3; |
|
|
1169 | unsigned char M = 1 << (*bit & 7); |
|
|
1170 | |
|
|
1171 | eio_dirent *a = base; |
|
|
1172 | eio_dirent *b = end; |
|
|
1173 | |
|
|
1174 | if (b - a < EIO_SORT_CUTOFF) |
|
|
1175 | break; |
|
|
1176 | |
|
|
1177 | /* now bit-partition the array on the bit */ |
|
|
1178 | /* this ugly asymmetric loop seems to perform much better than typical */ |
|
|
1179 | /* partition algos found in the literature */ |
|
|
1180 | do |
|
|
1181 | if (!(((unsigned char *)a)[O] & M)) |
|
|
1182 | ++a; |
|
|
1183 | else if (!(((unsigned char *)--b)[O] & M)) |
|
|
1184 | { |
|
|
1185 | eio_dirent tmp = *a; *a = *b; *b = tmp; |
|
|
1186 | ++a; |
|
|
1187 | } |
|
|
1188 | while (b > a); |
|
|
1189 | |
|
|
1190 | /* next bit, or stop, if no bits left in this path */ |
|
|
1191 | if (!*--bit) |
|
|
1192 | break; |
|
|
1193 | |
|
|
1194 | base_stk [stk_idx] = a; |
|
|
1195 | end_stk [stk_idx] = end; |
|
|
1196 | bit_stk [stk_idx] = bit; |
|
|
1197 | ++stk_idx; |
|
|
1198 | |
|
|
1199 | end = a; |
|
|
1200 | } |
|
|
1201 | } |
|
|
1202 | while (stk_idx--); |
|
|
1203 | } |
|
|
1204 | } |
|
|
1205 | |
|
|
1206 | static void |
|
|
1207 | eio_dent_insertion_sort (eio_dirent *dents, int size) |
|
|
1208 | { |
|
|
1209 | /* first move the smallest element to the front, to act as a sentinel */ |
|
|
1210 | { |
|
|
1211 | int i; |
|
|
1212 | eio_dirent *min = dents; |
|
|
1213 | |
|
|
1214 | /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
|
|
1215 | for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
|
|
1216 | if (EIO_DENT_CMP (dents [i], <, *min)) |
|
|
1217 | min = &dents [i]; |
|
|
1218 | |
|
|
1219 | /* swap elements 0 and j (minimum) */ |
|
|
1220 | { |
|
|
1221 | eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
|
|
1222 | } |
|
|
1223 | } |
|
|
1224 | |
|
|
1225 | /* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
|
|
1226 | { |
|
|
1227 | eio_dirent *i, *j; |
|
|
1228 | |
|
|
1229 | for (i = dents + 1; i < dents + size; ++i) |
|
|
1230 | { |
|
|
1231 | eio_dirent value = *i; |
|
|
1232 | |
|
|
1233 | for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
|
|
1234 | j [1] = j [0]; |
|
|
1235 | |
|
|
1236 | j [1] = value; |
|
|
1237 | } |
|
|
1238 | } |
|
|
1239 | } |
|
|
1240 | |
|
|
1241 | static void |
|
|
1242 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1243 | { |
|
|
1244 | if (size <= 1) |
|
|
1245 | return; /* our insertion sort relies on size > 0 */ |
|
|
1246 | |
|
|
1247 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
|
|
1248 | /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
|
|
1249 | eio_dent_radix_sort (dents, size, score_bits, inode_bits); |
|
|
1250 | |
|
|
1251 | /* use an insertion sort at the end, or for small arrays, */ |
|
|
1252 | /* as insertion sort is more efficient for small partitions */ |
|
|
1253 | eio_dent_insertion_sort (dents, size); |
|
|
1254 | } |
|
|
1255 | |
898 | /* read a full directory */ |
1256 | /* read a full directory */ |
899 | static void |
1257 | static void |
900 | eio__scandir (eio_req *req, etp_worker *self) |
1258 | eio__scandir (eio_req *req, etp_worker *self) |
901 | { |
1259 | { |
902 | DIR *dirp; |
1260 | DIR *dirp; |
903 | EIO_STRUCT_DIRENT *entp; |
1261 | EIO_STRUCT_DIRENT *entp; |
904 | char *name, *names; |
1262 | char *name, *names; |
905 | int memlen = 4096; |
1263 | int namesalloc = 4096; |
906 | int memofs = 0; |
1264 | int namesoffs = 0; |
|
|
1265 | int flags = req->int1; |
|
|
1266 | eio_dirent *dents = 0; |
|
|
1267 | int dentalloc = 128; |
907 | int res = 0; |
1268 | int dentoffs = 0; |
|
|
1269 | ino_t inode_bits = 0; |
|
|
1270 | |
|
|
1271 | req->result = -1; |
|
|
1272 | |
|
|
1273 | if (!(flags & EIO_READDIR_DENTS)) |
|
|
1274 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
908 | |
1275 | |
909 | X_LOCK (wrklock); |
1276 | X_LOCK (wrklock); |
910 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1277 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
911 | self->dirp = dirp = opendir (req->ptr1); |
1278 | self->dirp = dirp = opendir (req->ptr1); |
|
|
1279 | |
912 | req->flags |= EIO_FLAG_PTR2_FREE; |
1280 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
|
|
1281 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
913 | req->ptr2 = names = malloc (memlen); |
1282 | req->ptr2 = names = malloc (namesalloc); |
914 | X_UNLOCK (wrklock); |
1283 | X_UNLOCK (wrklock); |
915 | |
1284 | |
916 | if (dirp && names) |
1285 | if (dirp && names && (!flags || dents)) |
917 | for (;;) |
1286 | for (;;) |
918 | { |
1287 | { |
919 | errno = 0; |
1288 | errno = 0; |
920 | entp = readdir (dirp); |
1289 | entp = readdir (dirp); |
921 | |
1290 | |
922 | if (!entp) |
1291 | if (!entp) |
|
|
1292 | { |
|
|
1293 | if (errno) |
|
|
1294 | break; |
|
|
1295 | |
|
|
1296 | /* sort etc. */ |
|
|
1297 | req->int1 = flags; |
|
|
1298 | req->result = dentoffs; |
|
|
1299 | |
|
|
1300 | if (flags & EIO_READDIR_STAT_ORDER) |
|
|
1301 | eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); |
|
|
1302 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1303 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
|
|
1304 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
|
|
1305 | else |
|
|
1306 | { |
|
|
1307 | /* in this case, all is known, and we just put dirs first and sort them */ |
|
|
1308 | eio_dirent *oth = dents + dentoffs; |
|
|
1309 | eio_dirent *dir = dents; |
|
|
1310 | |
|
|
1311 | /* now partition dirs to the front, and non-dirs to the back */ |
|
|
1312 | /* by walking from both sides and swapping if necessary */ |
|
|
1313 | while (oth > dir) |
|
|
1314 | { |
|
|
1315 | if (dir->type == EIO_DT_DIR) |
|
|
1316 | ++dir; |
|
|
1317 | else if ((--oth)->type == EIO_DT_DIR) |
|
|
1318 | { |
|
|
1319 | eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
|
|
1320 | |
|
|
1321 | ++dir; |
|
|
1322 | } |
|
|
1323 | } |
|
|
1324 | |
|
|
1325 | /* now sort the dirs only (dirs all have the same score) */ |
|
|
1326 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
|
|
1327 | } |
|
|
1328 | |
923 | break; |
1329 | break; |
|
|
1330 | } |
924 | |
1331 | |
|
|
1332 | /* now add the entry to our list(s) */ |
925 | name = entp->d_name; |
1333 | name = entp->d_name; |
926 | |
1334 | |
|
|
1335 | /* skip . and .. entries */ |
927 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1336 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
928 | { |
1337 | { |
929 | int len = strlen (name) + 1; |
1338 | int len = D_NAMLEN (entp) + 1; |
930 | |
1339 | |
931 | res++; |
1340 | while (expect_false (namesoffs + len > namesalloc)) |
932 | |
|
|
933 | while (memofs + len > memlen) |
|
|
934 | { |
1341 | { |
935 | memlen *= 2; |
1342 | namesalloc *= 2; |
936 | X_LOCK (wrklock); |
1343 | X_LOCK (wrklock); |
937 | req->ptr2 = names = realloc (names, memlen); |
1344 | req->ptr2 = names = realloc (names, namesalloc); |
938 | X_UNLOCK (wrklock); |
1345 | X_UNLOCK (wrklock); |
939 | |
1346 | |
940 | if (!names) |
1347 | if (!names) |
941 | break; |
1348 | break; |
942 | } |
1349 | } |
943 | |
1350 | |
944 | memcpy (names + memofs, name, len); |
1351 | memcpy (names + namesoffs, name, len); |
|
|
1352 | |
|
|
1353 | if (dents) |
|
|
1354 | { |
|
|
1355 | struct eio_dirent *ent; |
|
|
1356 | |
|
|
1357 | if (expect_false (dentoffs == dentalloc)) |
|
|
1358 | { |
|
|
1359 | dentalloc *= 2; |
|
|
1360 | X_LOCK (wrklock); |
|
|
1361 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
|
|
1362 | X_UNLOCK (wrklock); |
|
|
1363 | |
|
|
1364 | if (!dents) |
|
|
1365 | break; |
|
|
1366 | } |
|
|
1367 | |
|
|
1368 | ent = dents + dentoffs; |
|
|
1369 | |
|
|
1370 | ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
|
|
1371 | ent->namelen = len - 1; |
|
|
1372 | ent->inode = D_INO (entp); |
|
|
1373 | |
|
|
1374 | inode_bits |= ent->inode; |
|
|
1375 | |
|
|
1376 | switch (D_TYPE (entp)) |
|
|
1377 | { |
|
|
1378 | default: |
|
|
1379 | ent->type = EIO_DT_UNKNOWN; |
|
|
1380 | flags |= EIO_READDIR_FOUND_UNKNOWN; |
|
|
1381 | break; |
|
|
1382 | |
|
|
1383 | #ifdef DT_FIFO |
|
|
1384 | case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
|
|
1385 | #endif |
|
|
1386 | #ifdef DT_CHR |
|
|
1387 | case DT_CHR: ent->type = EIO_DT_CHR; break; |
|
|
1388 | #endif |
|
|
1389 | #ifdef DT_MPC |
|
|
1390 | case DT_MPC: ent->type = EIO_DT_MPC; break; |
|
|
1391 | #endif |
|
|
1392 | #ifdef DT_DIR |
|
|
1393 | case DT_DIR: ent->type = EIO_DT_DIR; break; |
|
|
1394 | #endif |
|
|
1395 | #ifdef DT_NAM |
|
|
1396 | case DT_NAM: ent->type = EIO_DT_NAM; break; |
|
|
1397 | #endif |
|
|
1398 | #ifdef DT_BLK |
|
|
1399 | case DT_BLK: ent->type = EIO_DT_BLK; break; |
|
|
1400 | #endif |
|
|
1401 | #ifdef DT_MPB |
|
|
1402 | case DT_MPB: ent->type = EIO_DT_MPB; break; |
|
|
1403 | #endif |
|
|
1404 | #ifdef DT_REG |
|
|
1405 | case DT_REG: ent->type = EIO_DT_REG; break; |
|
|
1406 | #endif |
|
|
1407 | #ifdef DT_NWK |
|
|
1408 | case DT_NWK: ent->type = EIO_DT_NWK; break; |
|
|
1409 | #endif |
|
|
1410 | #ifdef DT_CMP |
|
|
1411 | case DT_CMP: ent->type = EIO_DT_CMP; break; |
|
|
1412 | #endif |
|
|
1413 | #ifdef DT_LNK |
|
|
1414 | case DT_LNK: ent->type = EIO_DT_LNK; break; |
|
|
1415 | #endif |
|
|
1416 | #ifdef DT_SOCK |
|
|
1417 | case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
|
|
1418 | #endif |
|
|
1419 | #ifdef DT_DOOR |
|
|
1420 | case DT_DOOR: ent->type = EIO_DT_DOOR; break; |
|
|
1421 | #endif |
|
|
1422 | #ifdef DT_WHT |
|
|
1423 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
|
|
1424 | #endif |
|
|
1425 | } |
|
|
1426 | |
|
|
1427 | ent->score = 7; |
|
|
1428 | |
|
|
1429 | if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1430 | { |
|
|
1431 | if (ent->type == EIO_DT_UNKNOWN) |
|
|
1432 | { |
|
|
1433 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
|
|
1434 | ent->score = 1; |
|
|
1435 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
|
|
1436 | ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
|
|
1437 | } |
|
|
1438 | else if (ent->type == EIO_DT_DIR) |
|
|
1439 | ent->score = 0; |
|
|
1440 | } |
|
|
1441 | } |
|
|
1442 | |
945 | memofs += len; |
1443 | namesoffs += len; |
|
|
1444 | ++dentoffs; |
|
|
1445 | } |
|
|
1446 | |
|
|
1447 | if (EIO_CANCELLED (req)) |
|
|
1448 | { |
|
|
1449 | errno = ECANCELED; |
|
|
1450 | break; |
946 | } |
1451 | } |
947 | } |
1452 | } |
|
|
1453 | } |
948 | |
1454 | |
949 | if (errno) |
1455 | #ifdef PAGESIZE |
950 | res = -1; |
1456 | # define eio_pagesize() PAGESIZE |
|
|
1457 | #else |
|
|
1458 | static intptr_t |
|
|
1459 | eio_pagesize (void) |
|
|
1460 | { |
|
|
1461 | static intptr_t page; |
|
|
1462 | |
|
|
1463 | if (!page) |
|
|
1464 | page = sysconf (_SC_PAGESIZE); |
|
|
1465 | |
|
|
1466 | return page; |
|
|
1467 | } |
|
|
1468 | #endif |
|
|
1469 | |
|
|
1470 | static void |
|
|
1471 | eio_page_align (void **addr, size_t *length) |
|
|
1472 | { |
|
|
1473 | intptr_t mask = eio_pagesize () - 1; |
|
|
1474 | |
|
|
1475 | /* round down addr */ |
|
|
1476 | intptr_t adj = mask & (intptr_t)*addr; |
|
|
1477 | |
|
|
1478 | *addr = (void *)((intptr_t)*addr - adj); |
|
|
1479 | *length += adj; |
|
|
1480 | |
|
|
1481 | /* round up length */ |
|
|
1482 | *length = (*length + mask) & ~mask; |
|
|
1483 | } |
|
|
1484 | |
|
|
1485 | #if !_POSIX_MEMLOCK |
|
|
1486 | # define eio__mlockall(a) ((errno = ENOSYS), -1) |
|
|
1487 | #else |
|
|
1488 | |
|
|
1489 | static int |
|
|
1490 | eio__mlockall (int flags) |
|
|
1491 | { |
|
|
1492 | #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 |
|
|
1493 | extern int mallopt (int, int); |
|
|
1494 | mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ |
|
|
1495 | #endif |
|
|
1496 | |
|
|
1497 | if (EIO_MCL_CURRENT != MCL_CURRENT |
|
|
1498 | || EIO_MCL_FUTURE != MCL_FUTURE) |
|
|
1499 | { |
|
|
1500 | flags = 0 |
|
|
1501 | | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0) |
|
|
1502 | | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0); |
|
|
1503 | } |
|
|
1504 | |
|
|
1505 | return mlockall (flags); |
|
|
1506 | } |
|
|
1507 | #endif |
|
|
1508 | |
|
|
1509 | #if !_POSIX_MEMLOCK_RANGE |
|
|
1510 | # define eio__mlock(a,b) ((errno = ENOSYS), -1) |
|
|
1511 | #else |
|
|
1512 | |
|
|
1513 | static int |
|
|
1514 | eio__mlock (void *addr, size_t length) |
|
|
1515 | { |
|
|
1516 | eio_page_align (&addr, &length); |
|
|
1517 | |
|
|
1518 | return mlock (addr, length); |
|
|
1519 | } |
|
|
1520 | |
|
|
1521 | #endif |
|
|
1522 | |
|
|
1523 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
|
|
1524 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
|
|
1525 | #else |
|
|
1526 | |
|
|
1527 | int |
|
|
1528 | eio__msync (void *mem, size_t len, int flags) |
|
|
1529 | { |
|
|
1530 | eio_page_align (&mem, &len); |
|
|
1531 | |
|
|
1532 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1533 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1534 | || EIO_MS_SYNC != MS_SYNC) |
|
|
1535 | { |
|
|
1536 | flags = 0 |
|
|
1537 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
|
|
1538 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
|
|
1539 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
|
|
1540 | } |
|
|
1541 | |
|
|
1542 | return msync (mem, len, flags); |
|
|
1543 | } |
|
|
1544 | |
|
|
1545 | #endif |
|
|
1546 | |
|
|
1547 | int |
|
|
1548 | eio__mtouch (eio_req *req) |
|
|
1549 | { |
|
|
1550 | void *mem = req->ptr2; |
|
|
1551 | size_t len = req->size; |
|
|
1552 | int flags = req->int1; |
|
|
1553 | |
|
|
1554 | eio_page_align (&mem, &len); |
|
|
1555 | |
951 | |
1556 | { |
952 | req->result = res; |
1557 | intptr_t addr = (intptr_t)mem; |
|
|
1558 | intptr_t end = addr + len; |
|
|
1559 | intptr_t page = eio_pagesize (); |
|
|
1560 | |
|
|
1561 | if (addr < end) |
|
|
1562 | if (flags & EIO_MT_MODIFY) /* modify */ |
|
|
1563 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
|
|
1564 | else |
|
|
1565 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
|
|
1566 | } |
|
|
1567 | |
|
|
1568 | return 0; |
953 | } |
1569 | } |
954 | |
1570 | |
955 | /*****************************************************************************/ |
1571 | /*****************************************************************************/ |
956 | |
1572 | |
957 | #define ALLOC(len) \ |
1573 | #define ALLOC(len) \ |
… | |
… | |
989 | if (req) |
1605 | if (req) |
990 | break; |
1606 | break; |
991 | |
1607 | |
992 | ++idle; |
1608 | ++idle; |
993 | |
1609 | |
994 | ts.tv_sec = time (0) + IDLE_TIMEOUT; |
1610 | ts.tv_sec = time (0) + idle_timeout; |
995 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1611 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
996 | { |
1612 | { |
997 | if (idle > max_idle) |
1613 | if (idle > max_idle) |
998 | { |
1614 | { |
999 | --idle; |
1615 | --idle; |
… | |
… | |
1078 | return 0; \ |
1694 | return 0; \ |
1079 | } |
1695 | } |
1080 | |
1696 | |
1081 | static void eio_execute (etp_worker *self, eio_req *req) |
1697 | static void eio_execute (etp_worker *self, eio_req *req) |
1082 | { |
1698 | { |
1083 | errno = 0; |
|
|
1084 | |
|
|
1085 | switch (req->type) |
1699 | switch (req->type) |
1086 | { |
1700 | { |
1087 | case EIO_READ: ALLOC (req->size); |
1701 | case EIO_READ: ALLOC (req->size); |
1088 | req->result = req->offs >= 0 |
1702 | req->result = req->offs >= 0 |
1089 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
1703 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
… | |
… | |
1100 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1714 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1101 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1715 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1102 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1716 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1103 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1717 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1104 | |
1718 | |
|
|
1719 | case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1720 | req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1721 | case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1722 | req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1723 | |
1105 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1724 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1106 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1725 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1107 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1726 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1108 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1727 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1109 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
1728 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
… | |
… | |
1116 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1735 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1117 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1736 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1118 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1737 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1119 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1738 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1120 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1739 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1121 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
1740 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; |
1122 | |
1741 | |
1123 | case EIO_READLINK: ALLOC (NAME_MAX); |
1742 | case EIO_READLINK: ALLOC (PATH_MAX); |
1124 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1743 | req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; |
1125 | |
1744 | |
1126 | case EIO_SYNC: req->result = 0; sync (); break; |
1745 | case EIO_SYNC: req->result = 0; sync (); break; |
1127 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1746 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1128 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1747 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
|
|
1748 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
|
|
1749 | case EIO_MTOUCH: req->result = eio__mtouch (req); break; |
|
|
1750 | case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; |
|
|
1751 | case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; |
|
|
1752 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1129 | |
1753 | |
1130 | case EIO_READDIR: eio__scandir (req, self); break; |
1754 | case EIO_READDIR: eio__scandir (req, self); break; |
1131 | |
1755 | |
1132 | case EIO_BUSY: |
1756 | case EIO_BUSY: |
1133 | #ifdef _WIN32 |
1757 | #ifdef _WIN32 |
1134 | Sleep (req->nv1 * 1000.); |
1758 | Sleep (req->nv1 * 1e3); |
1135 | #else |
1759 | #else |
1136 | { |
1760 | { |
1137 | struct timeval tv; |
1761 | struct timeval tv; |
1138 | |
1762 | |
1139 | tv.tv_sec = req->nv1; |
1763 | tv.tv_sec = req->nv1; |
1140 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
1764 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6; |
1141 | |
1765 | |
1142 | req->result = select (0, 0, 0, 0, &tv); |
1766 | req->result = select (0, 0, 0, 0, &tv); |
1143 | } |
1767 | } |
1144 | #endif |
1768 | #endif |
1145 | break; |
1769 | break; |
… | |
… | |
1160 | times = tv; |
1784 | times = tv; |
1161 | } |
1785 | } |
1162 | else |
1786 | else |
1163 | times = 0; |
1787 | times = 0; |
1164 | |
1788 | |
1165 | |
|
|
1166 | req->result = req->type == EIO_FUTIME |
1789 | req->result = req->type == EIO_FUTIME |
1167 | ? futimes (req->int1, times) |
1790 | ? futimes (req->int1, times) |
1168 | : utimes (req->ptr1, times); |
1791 | : utimes (req->ptr1, times); |
1169 | } |
1792 | } |
|
|
1793 | break; |
1170 | |
1794 | |
1171 | case EIO_GROUP: |
1795 | case EIO_GROUP: |
|
|
1796 | abort (); /* handled in eio_request */ |
|
|
1797 | |
1172 | case EIO_NOP: |
1798 | case EIO_NOP: |
1173 | req->result = 0; |
1799 | req->result = 0; |
1174 | break; |
1800 | break; |
1175 | |
1801 | |
1176 | case EIO_CUSTOM: |
1802 | case EIO_CUSTOM: |
1177 | req->feed (req); |
1803 | req->feed (req); |
1178 | break; |
1804 | break; |
1179 | |
1805 | |
1180 | default: |
1806 | default: |
|
|
1807 | errno = ENOSYS; |
1181 | req->result = -1; |
1808 | req->result = -1; |
1182 | break; |
1809 | break; |
1183 | } |
1810 | } |
1184 | |
1811 | |
1185 | req->errorno = errno; |
1812 | req->errorno = errno; |
… | |
… | |
1205 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1832 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1206 | { |
1833 | { |
1207 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
1834 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
1208 | } |
1835 | } |
1209 | |
1836 | |
|
|
1837 | eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
1838 | { |
|
|
1839 | REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
|
|
1840 | } |
|
|
1841 | |
|
|
1842 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
1843 | { |
|
|
1844 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
|
|
1845 | } |
|
|
1846 | |
|
|
1847 | eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data) |
|
|
1848 | { |
|
|
1849 | REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND; |
|
|
1850 | } |
|
|
1851 | |
|
|
1852 | eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) |
|
|
1853 | { |
|
|
1854 | REQ (EIO_MLOCKALL); req->int1 = flags; SEND; |
|
|
1855 | } |
|
|
1856 | |
|
|
1857 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
|
|
1858 | { |
|
|
1859 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
|
|
1860 | } |
|
|
1861 | |
1210 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1862 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1211 | { |
1863 | { |
1212 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1864 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1213 | } |
1865 | } |
1214 | |
1866 | |
… | |
… | |
1233 | } |
1885 | } |
1234 | |
1886 | |
1235 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1887 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1236 | { |
1888 | { |
1237 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
1889 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
|
|
1890 | } |
|
|
1891 | |
|
|
1892 | eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data) |
|
|
1893 | { |
|
|
1894 | REQ (EIO_FSTATVFS); req->int1 = fd; SEND; |
1238 | } |
1895 | } |
1239 | |
1896 | |
1240 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1897 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1241 | { |
1898 | { |
1242 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
1899 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
… | |
… | |
1316 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1973 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1317 | { |
1974 | { |
1318 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1975 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1319 | } |
1976 | } |
1320 | |
1977 | |
|
|
1978 | eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data) |
|
|
1979 | { |
|
|
1980 | return eio__1path (EIO_STATVFS, path, pri, cb, data); |
|
|
1981 | } |
|
|
1982 | |
1321 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1983 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1322 | { |
1984 | { |
1323 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1985 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1324 | } |
1986 | } |
1325 | |
1987 | |
1326 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
1988 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
1327 | { |
1989 | { |
1328 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
1990 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
1329 | } |
1991 | } |
1330 | |
1992 | |
1331 | eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) |
1993 | eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data) |
1332 | { |
1994 | { |
1333 | return eio__1path (EIO_READDIR, path, pri, cb, data); |
1995 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1334 | } |
1996 | } |
1335 | |
1997 | |
1336 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1998 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1337 | { |
1999 | { |
1338 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
2000 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND; |
1339 | } |
2001 | } |
1340 | |
2002 | |
1341 | static eio_req * |
2003 | static eio_req * |
1342 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2004 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1343 | { |
2005 | { |
… | |
… | |
1367 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2029 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1368 | { |
2030 | { |
1369 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
2031 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1370 | } |
2032 | } |
1371 | |
2033 | |
1372 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
2034 | eio_req *eio_custom (void (*)(eio_req *) execute, int pri, eio_cb cb, void *data); |
1373 | { |
2035 | { |
1374 | REQ (EIO_CUSTOM); req->feed = execute; SEND; |
2036 | REQ (EIO_CUSTOM); req->feed = execute; SEND; |
1375 | } |
2037 | } |
1376 | |
2038 | |
1377 | #endif |
2039 | #endif |
… | |
… | |
1407 | |
2069 | |
1408 | void eio_grp_add (eio_req *grp, eio_req *req) |
2070 | void eio_grp_add (eio_req *grp, eio_req *req) |
1409 | { |
2071 | { |
1410 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
2072 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
1411 | |
2073 | |
|
|
2074 | grp->flags |= EIO_FLAG_GROUPADD; |
|
|
2075 | |
1412 | ++grp->size; |
2076 | ++grp->size; |
1413 | req->grp = grp; |
2077 | req->grp = grp; |
1414 | |
2078 | |
1415 | req->grp_prev = 0; |
2079 | req->grp_prev = 0; |
1416 | req->grp_next = grp->grp_first; |
2080 | req->grp_next = grp->grp_first; |
… | |
… | |
1425 | /* misc garbage */ |
2089 | /* misc garbage */ |
1426 | |
2090 | |
1427 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
2091 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1428 | { |
2092 | { |
1429 | etp_worker wrk; |
2093 | etp_worker wrk; |
|
|
2094 | ssize_t ret; |
1430 | |
2095 | |
1431 | wrk.dbuf = 0; |
2096 | wrk.dbuf = 0; |
1432 | |
2097 | |
1433 | eio__sendfile (ofd, ifd, offset, count, &wrk); |
2098 | ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
1434 | |
2099 | |
1435 | if (wrk.dbuf) |
2100 | if (wrk.dbuf) |
1436 | free (wrk.dbuf); |
2101 | free (wrk.dbuf); |
1437 | } |
|
|
1438 | |
2102 | |
|
|
2103 | return ret; |
|
|
2104 | } |
|
|
2105 | |