ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libeio/eio.c
Revision: 1.148
Committed: Tue Feb 20 06:05:19 2018 UTC (6 years, 3 months ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.147: +2 -0 lines
Log Message:
*** empty log message ***

File Contents

# Content
1 /*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013,2016,2017 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
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
37 * either the BSD or the GPL.
38 */
39
40 #ifndef _WIN32
41 # include "config.h"
42 #endif
43
44 #include "eio.h"
45 #include "ecb.h"
46
47 #include <errno.h>
48 #include <stddef.h>
49 #include <stdlib.h>
50 #include <string.h>
51 #include <errno.h>
52 #include <sys/types.h>
53 #include <sys/stat.h>
54 #include <limits.h>
55 #include <fcntl.h>
56 #include <assert.h>
57
58 /* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
59 /* intptr_t only comes from stdint.h, says idiot openbsd coder */
60 #if HAVE_STDINT_H
61 # include <stdint.h>
62 #endif
63
64 #ifndef ECANCELED
65 # define ECANCELED EDOM
66 #endif
67 #ifndef ELOOP
68 # define ELOOP EDOM
69 #endif
70
71 #if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
72 # define ENOTSOCK WSAENOTSOCK
73 #endif
74
75 static void eio_destroy (eio_req *req);
76
77 #ifndef EIO_FINISH
78 # define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
79 #endif
80
81 #ifndef EIO_DESTROY
82 # define EIO_DESTROY(req) do { if ((req)->destroy) (req)->destroy (req); } while (0)
83 #endif
84
85 #ifndef EIO_FEED
86 # define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
87 #endif
88
89 #ifndef EIO_FD_TO_WIN32_HANDLE
90 # define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
91 #endif
92 #ifndef EIO_WIN32_HANDLE_TO_FD
93 # define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
94 #endif
95
96 #define EIO_ERRNO(errval,retval) ((errno = errval), retval)
97
98 #define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
99
100 #ifdef _WIN32
101
102 #undef PAGESIZE
103 #define PAGESIZE 4096 /* GetSystemInfo? */
104
105 #define O_CLOEXEC 0
106
107 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
108
109 #ifdef EIO_STRUCT_STATI64
110 /* look at perl's non-sloppy stat */
111 #define stat(path,buf) _stati64 (path,buf)
112 #define fstat(fd,buf) _fstati64 (fd,buf)
113 #endif
114 #define lstat(path,buf) stat (path,buf)
115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
116 #define mkdir(path,mode) _mkdir (path)
117 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
118
119 #define chmod(path,mode) _chmod (path, mode)
120 #define dup(fd) _dup (fd)
121 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
122 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
123
124 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
125 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
202
203 /* we could even stat and see if it exists */
204 static int
205 symlink (const char *old, const char *neu)
206 {
207 #if WINVER >= 0x0600
208 int flags;
209
210 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
211 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
212 * with directory first.
213 */
214 for (flags = 3; flags >= 0; --flags)
215 if (CreateSymbolicLink (neu, old, flags))
216 return 0;
217 #endif
218
219 return EIO_ERRNO (ENOENT, -1);
220 }
221
222 /* POSIX API only, causing trouble for win32 apps */
223 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
224 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
225
226 struct statvfs
227 {
228 int dummy;
229 };
230
231 #define DT_DIR EIO_DT_DIR
232 #define DT_REG EIO_DT_REG
233 #define D_NAME(entp) entp.cFileName
234 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
235
236 #else
237
238 #include <sys/ioctl.h>
239 #include <sys/time.h>
240 #include <sys/select.h>
241 #include <unistd.h>
242 #include <signal.h>
243 #include <dirent.h>
244
245 #ifdef ANDROID
246 #include <sys/vfs.h>
247 #define statvfs statfs
248 #define fstatvfs fstatfs
249 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
250 #else
251 #include <sys/statvfs.h>
252 #endif
253
254 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
255 #include <sys/mman.h>
256 #endif
257
258 #define D_NAME(entp) entp->d_name
259
260 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
261 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
262 #define _DIRENT_HAVE_D_TYPE /* sigh */
263 #define D_INO(de) (de)->d_fileno
264 #define D_NAMLEN(de) (de)->d_namlen
265 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
266 #define D_INO(de) (de)->d_ino
267 #endif
268
269 #ifdef _D_EXACT_NAMLEN
270 #undef D_NAMLEN
271 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
272 #endif
273
274 #ifdef _DIRENT_HAVE_D_TYPE
275 #define D_TYPE(de) (de)->d_type
276 #endif
277
278 #ifndef EIO_STRUCT_DIRENT
279 #define EIO_STRUCT_DIRENT struct dirent
280 #endif
281
282 #endif
283
284 #if HAVE_UTIMES
285 # include <utime.h>
286 #endif
287
288 #if HAVE_SYS_SYSCALL_H
289 # include <sys/syscall.h>
290 #endif
291
292 #if HAVE_SENDFILE
293 # if __linux
294 # include <sys/sendfile.h>
295 # elif __FreeBSD__ || defined __APPLE__
296 # include <sys/socket.h>
297 # include <sys/uio.h>
298 # elif __hpux
299 # include <sys/socket.h>
300 # elif __solaris
301 # include <sys/sendfile.h>
302 # else
303 # error sendfile support requested but not available
304 # endif
305 #endif
306
307 #if HAVE_RENAMEAT2
308 # include <sys/syscall.h>
309 # include <linux/fs.h>
310 #endif
311
312 #ifndef D_TYPE
313 # define D_TYPE(de) 0
314 #endif
315 #ifndef D_INO
316 # define D_INO(de) 0
317 #endif
318 #ifndef D_NAMLEN
319 # define D_NAMLEN(entp) strlen (D_NAME (entp))
320 #endif
321
322 /* used for struct dirent, AIX doesn't provide it */
323 #ifndef NAME_MAX
324 # define NAME_MAX 4096
325 #endif
326
327 /* used for readlink etc. */
328 #ifndef PATH_MAX
329 # define PATH_MAX 0
330 #endif
331
332 #ifndef EIO_PATH_MIN
333 # define EIO_PATH_MIN 8160
334 #endif
335
336 #define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
337
338 /* buffer size for various temporary buffers */
339 #define EIO_BUFSIZE 65536
340
341 #define dBUF \
342 char *eio_buf = malloc (EIO_BUFSIZE); \
343 errno = ENOMEM; \
344 if (!eio_buf) \
345 return -1
346
347 #define FUBd \
348 free (eio_buf)
349
350 /*****************************************************************************/
351
352 struct etp_tmpbuf;
353
354 #if _POSIX_VERSION >= 200809L
355 #define HAVE_AT 1
356 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
357 #ifndef O_SEARCH
358 #define O_SEARCH O_RDONLY
359 #endif
360 #else
361 #define HAVE_AT 0
362 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
363 #endif
364
365 struct eio_pwd
366 {
367 #if HAVE_AT
368 int fd;
369 #endif
370 int len;
371 char str[1]; /* actually, a 0-terminated canonical path */
372 };
373
374 /*****************************************************************************/
375
376 #define ETP_PRI_MIN EIO_PRI_MIN
377 #define ETP_PRI_MAX EIO_PRI_MAX
378
379 #define ETP_TYPE_QUIT -1
380 #define ETP_TYPE_GROUP EIO_GROUP
381
382 static void eio_nop_callback (void) { }
383 static void (*eio_want_poll_cb)(void) = eio_nop_callback;
384 static void (*eio_done_poll_cb)(void) = eio_nop_callback;
385
386 #define ETP_WANT_POLL(pool) eio_want_poll_cb ()
387 #define ETP_DONE_POLL(pool) eio_done_poll_cb ()
388
389 struct etp_worker;
390 #define ETP_REQ eio_req
391 #define ETP_DESTROY(req) eio_destroy (req)
392 static int eio_finish (eio_req *req);
393 #define ETP_FINISH(req) eio_finish (req)
394 static void eio_execute (struct etp_worker *self, eio_req *req);
395 #define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
396
397 #include "etp.c"
398
399 static struct etp_pool eio_pool;
400 #define EIO_POOL (&eio_pool)
401
402 /*****************************************************************************/
403
404 static void
405 grp_try_feed (eio_req *grp)
406 {
407 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
408 {
409 grp->flags &= ~ETP_FLAG_GROUPADD;
410
411 EIO_FEED (grp);
412
413 /* stop if no progress has been made */
414 if (!(grp->flags & ETP_FLAG_GROUPADD))
415 {
416 grp->feed = 0;
417 break;
418 }
419 }
420 }
421
422 static int
423 grp_dec (eio_req *grp)
424 {
425 --grp->size;
426
427 /* call feeder, if applicable */
428 grp_try_feed (grp);
429
430 /* finish, if done */
431 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
432 return eio_finish (grp);
433 else
434 return 0;
435 }
436
437 static void
438 eio_destroy (eio_req *req)
439 {
440 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
441 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
442
443 EIO_DESTROY (req);
444 }
445
446 static int
447 eio_finish (eio_req *req)
448 {
449 int res = EIO_FINISH (req);
450
451 if (req->grp)
452 {
453 int res2;
454 eio_req *grp = req->grp;
455
456 /* unlink request */
457 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
458 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
459
460 if (grp->grp_first == req)
461 grp->grp_first = req->grp_next;
462
463 res2 = grp_dec (grp);
464
465 if (!res)
466 res = res2;
467 }
468
469 eio_destroy (req);
470
471 return res;
472 }
473
474 void
475 eio_grp_cancel (eio_req *grp)
476 {
477 etp_grp_cancel (EIO_POOL, grp);
478 }
479
480 void
481 eio_cancel (eio_req *req)
482 {
483 etp_cancel (EIO_POOL, req);
484 }
485
486 void
487 eio_submit (eio_req *req)
488 {
489 etp_submit (EIO_POOL, req);
490 }
491
492 unsigned int
493 eio_nreqs (void)
494 {
495 return etp_nreqs (EIO_POOL);
496 }
497
498 unsigned int
499 eio_nready (void)
500 {
501 return etp_nready (EIO_POOL);
502 }
503
504 unsigned int
505 eio_npending (void)
506 {
507 return etp_npending (EIO_POOL);
508 }
509
510 unsigned int ecb_cold
511 eio_nthreads (void)
512 {
513 return etp_nthreads (EIO_POOL);
514 }
515
516 void ecb_cold
517 eio_set_max_poll_time (double nseconds)
518 {
519 etp_set_max_poll_time (EIO_POOL, nseconds);
520 }
521
522 void ecb_cold
523 eio_set_max_poll_reqs (unsigned int maxreqs)
524 {
525 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
526 }
527
528 void ecb_cold
529 eio_set_max_idle (unsigned int nthreads)
530 {
531 etp_set_max_idle (EIO_POOL, nthreads);
532 }
533
534 void ecb_cold
535 eio_set_idle_timeout (unsigned int seconds)
536 {
537 etp_set_idle_timeout (EIO_POOL, seconds);
538 }
539
540 void ecb_cold
541 eio_set_min_parallel (unsigned int nthreads)
542 {
543 etp_set_min_parallel (EIO_POOL, nthreads);
544 }
545
546 void ecb_cold
547 eio_set_max_parallel (unsigned int nthreads)
548 {
549 etp_set_max_parallel (EIO_POOL, nthreads);
550 }
551
552 int eio_poll (void)
553 {
554 return etp_poll (EIO_POOL);
555 }
556
557 /*****************************************************************************/
558 /* work around various missing functions */
559
560 #if HAVE_POSIX_CLOSE && !__linux
561 # define eio__close(fd) posix_close (fd, 0)
562 #else
563 # define eio__close(fd) close (fd)
564 #endif
565
566 /* close() without disturbing errno */
567 static void
568 silent_close (int fd)
569 {
570 int saved_errno = errno;
571 eio__close (fd);
572 errno = saved_errno;
573 }
574
575 #ifndef HAVE_UTIMES
576
577 # undef utimes
578 # define utimes(path,times) eio__utimes (path, times)
579
580 static int
581 eio__utimes (const char *filename, const struct timeval times[2])
582 {
583 if (times)
584 {
585 struct utimbuf buf;
586
587 buf.actime = times[0].tv_sec;
588 buf.modtime = times[1].tv_sec;
589
590 return utime (filename, &buf);
591 }
592 else
593 return utime (filename, 0);
594 }
595
596 #endif
597
598 #ifndef HAVE_FUTIMES
599
600 # undef futimes
601 # define futimes(fd,times) eio__futimes (fd, times)
602
603 static int
604 eio__futimes (int fd, const struct timeval tv[2])
605 {
606 errno = ENOSYS;
607 return -1;
608 }
609
610 #endif
611
612 #if !HAVE_FDATASYNC
613 # undef fdatasync
614 # define fdatasync(fd) fsync (fd)
615 #endif
616
617 static int
618 eio__syncfs (int fd)
619 {
620 int res;
621
622 #if HAVE_SYS_SYNCFS
623 res = (int)syscall (__NR_syncfs, (int)(fd));
624 #else
625 res = EIO_ENOSYS ();
626 #endif
627
628 if (res < 0 && errno == ENOSYS && fd >= 0)
629 sync ();
630
631 return res;
632 }
633
634 /* sync_file_range always needs emulation */
635 static int
636 eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
637 {
638 #if HAVE_SYNC_FILE_RANGE
639 int res;
640
641 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
642 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
643 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
644 {
645 flags = 0
646 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
647 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
648 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
649 }
650
651 res = sync_file_range (fd, offset, nbytes, flags);
652
653 if (!res || errno != ENOSYS)
654 return res;
655 #endif
656
657 /* even though we could play tricks with the flags, it's better to always
658 * call fdatasync, as that matches the expectation of its users best */
659 return fdatasync (fd);
660 }
661
662 static int
663 eio__fallocate (int fd, int mode, off_t offset, size_t len)
664 {
665 #if HAVE_LINUX_FALLOCATE
666 return fallocate (fd, mode, offset, len);
667 #else
668 return EIO_ENOSYS ();
669 #endif
670 }
671
672 #if !HAVE_READAHEAD
673 # undef readahead
674 # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
675
676 static eio_ssize_t
677 eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
678 {
679 size_t todo = count;
680 dBUF;
681
682 while (todo > 0)
683 {
684 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
685
686 pread (fd, eio_buf, len, offset);
687 offset += len;
688 todo -= len;
689 }
690
691 FUBd;
692
693 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
694 /* but not for e.g. EIO or eof, so we also never fail */
695 return 0;
696 }
697
698 #endif
699
700 /* sendfile always needs emulation */
701 static eio_ssize_t
702 eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
703 {
704 eio_ssize_t written = 0;
705 eio_ssize_t res;
706
707 if (!count)
708 return 0;
709
710 for (;;)
711 {
712 #ifdef __APPLE__
713 # undef HAVE_SENDFILE /* broken, as everything on os x */
714 #endif
715 #if HAVE_SENDFILE
716 # if __linux
717 off_t soffset = offset;
718 res = sendfile (ofd, ifd, &soffset, count);
719
720 # elif __FreeBSD__
721 /*
722 * Of course, the freebsd sendfile is a dire hack with no thoughts
723 * wasted on making it similar to other I/O functions.
724 */
725 off_t sbytes;
726 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
727
728 #if 0 /* according to the manpage, this is correct, but broken behaviour */
729 /* freebsd' sendfile will return 0 on success */
730 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
731 /* not on e.g. EIO or EPIPE - sounds broken */
732 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
733 res = sbytes;
734 #endif
735
736 /* according to source inspection, this is correct, and useful behaviour */
737 if (sbytes)
738 res = sbytes;
739
740 # elif defined __APPLE__
741 off_t sbytes = count;
742 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
743
744 /* according to the manpage, sbytes is always valid */
745 if (sbytes)
746 res = sbytes;
747
748 # elif __hpux
749 res = sendfile (ofd, ifd, offset, count, 0, 0);
750
751 # elif __solaris
752 struct sendfilevec vec;
753 size_t sbytes;
754
755 vec.sfv_fd = ifd;
756 vec.sfv_flag = 0;
757 vec.sfv_off = offset;
758 vec.sfv_len = count;
759
760 res = sendfilev (ofd, &vec, 1, &sbytes);
761
762 if (res < 0 && sbytes)
763 res = sbytes;
764
765 # endif
766
767 #elif defined (_WIN32) && 0
768 /* does not work, just for documentation of what would need to be done */
769 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
770 /* libeio guarantees that the file offset does not change, and windows */
771 /* has no way to get an independent handle to the same file description */
772 HANDLE h = TO_SOCKET (ifd);
773 SetFilePointer (h, offset, 0, FILE_BEGIN);
774 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
775
776 #else
777 res = EIO_ENOSYS ();
778 #endif
779
780 /* we assume sendfile can copy at least 128mb in one go */
781 if (res <= 128 * 1024 * 1024)
782 {
783 if (res > 0)
784 written += res;
785
786 if (written)
787 return written;
788
789 break;
790 }
791 else
792 {
793 /* if we requested more, then probably the kernel was lazy */
794 written += res;
795 offset += res;
796 count -= res;
797
798 if (!count)
799 return written;
800 }
801 }
802
803 if (res < 0
804 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
805 /* BSDs */
806 #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
807 || errno == ENOTSUP
808 #endif
809 #ifdef EOPNOTSUPP /* windows */
810 || errno == EOPNOTSUPP /* BSDs */
811 #endif
812 #if __solaris
813 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
814 #endif
815 )
816 )
817 {
818 /* emulate sendfile. this is a major pain in the ass */
819 dBUF;
820
821 res = 0;
822
823 while (count)
824 {
825 eio_ssize_t cnt;
826
827 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
828
829 if (cnt <= 0)
830 {
831 if (cnt && !res) res = -1;
832 break;
833 }
834
835 cnt = write (ofd, eio_buf, cnt);
836
837 if (cnt <= 0)
838 {
839 if (cnt && !res) res = -1;
840 break;
841 }
842
843 offset += cnt;
844 res += cnt;
845 count -= cnt;
846 }
847
848 FUBd;
849 }
850
851 return res;
852 }
853
854 #ifdef PAGESIZE
855 # define eio_pagesize() PAGESIZE
856 #else
857 static intptr_t
858 eio_pagesize (void)
859 {
860 static intptr_t page;
861
862 if (!page)
863 page = sysconf (_SC_PAGESIZE);
864
865 return page;
866 }
867 #endif
868
869 static void
870 eio_page_align (void **addr, size_t *length)
871 {
872 intptr_t mask = eio_pagesize () - 1;
873
874 /* round down addr */
875 intptr_t adj = mask & (intptr_t)*addr;
876
877 *addr = (void *)((intptr_t)*addr - adj);
878 *length += adj;
879
880 /* round up length */
881 *length = (*length + mask) & ~mask;
882 }
883
884 #if !_POSIX_MEMLOCK
885 # define eio__mlockall(a) EIO_ENOSYS ()
886 #else
887
888 static int
889 eio__mlockall (int flags)
890 {
891 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
892 extern int mallopt (int, int);
893 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
894 #endif
895
896 if (EIO_MCL_CURRENT != MCL_CURRENT
897 || EIO_MCL_FUTURE != MCL_FUTURE)
898 {
899 flags = 0
900 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
901 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
902 }
903
904 return mlockall (flags);
905 }
906 #endif
907
908 #if !_POSIX_MEMLOCK_RANGE
909 # define eio__mlock(a,b) EIO_ENOSYS ()
910 #else
911
912 static int
913 eio__mlock (void *addr, size_t length)
914 {
915 eio_page_align (&addr, &length);
916
917 return mlock (addr, length);
918 }
919
920 #endif
921
922 #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
923 # define eio__msync(a,b,c) EIO_ENOSYS ()
924 #else
925
926 static int
927 eio__msync (void *mem, size_t len, int flags)
928 {
929 eio_page_align (&mem, &len);
930
931 if (EIO_MS_ASYNC != MS_SYNC
932 || EIO_MS_INVALIDATE != MS_INVALIDATE
933 || EIO_MS_SYNC != MS_SYNC)
934 {
935 flags = 0
936 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
937 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
938 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
939 }
940
941 return msync (mem, len, flags);
942 }
943
944 #endif
945
946 static int
947 eio__mtouch (eio_req *req)
948 {
949 void *mem = req->ptr2;
950 size_t len = req->size;
951 int flags = req->int1;
952
953 eio_page_align (&mem, &len);
954
955 {
956 intptr_t addr = (intptr_t)mem;
957 intptr_t end = addr + len;
958 intptr_t page = eio_pagesize ();
959
960 if (addr < end)
961 if (flags & EIO_MT_MODIFY) /* modify */
962 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
963 else
964 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
965 }
966
967 return 0;
968 }
969
970 /*****************************************************************************/
971 /* requests implemented outside eio_execute, because they are so large */
972
973 static void
974 eio__lseek (eio_req *req)
975 {
976 /* this usually gets optimised away completely, or your compiler sucks, */
977 /* or the whence constants really are not 0, 1, 2 */
978 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
979 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
980 : req->int2 == EIO_SEEK_END ? SEEK_END
981 : req->int2;
982
983 req->offs = lseek (req->int1, req->offs, whence);
984 req->result = req->offs == (off_t)-1 ? -1 : 0;
985 }
986
987 /* result will always end up in tmpbuf, there is always space for adding a 0-byte */
988 static int
989 eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
990 {
991 char *res;
992 const char *rel = path;
993 char *tmp1, *tmp2;
994 #if SYMLOOP_MAX > 32
995 int symlinks = SYMLOOP_MAX;
996 #else
997 int symlinks = 32;
998 #endif
999
1000 errno = EINVAL;
1001 if (!rel)
1002 return -1;
1003
1004 errno = ENOENT;
1005 if (!*rel)
1006 return -1;
1007
1008 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1009 #ifdef _WIN32
1010 if (_access (rel, 4) != 0)
1011 return -1;
1012
1013 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1014
1015 errno = ENAMETOOLONG;
1016 if (symlinks >= EIO_PATH_MAX * 3)
1017 return -1;
1018
1019 errno = EIO;
1020 if (symlinks <= 0)
1021 return -1;
1022
1023 return symlinks;
1024
1025 #else
1026 tmp1 = res + EIO_PATH_MAX;
1027 tmp2 = tmp1 + EIO_PATH_MAX;
1028
1029 #if 0 /* disabled, the musl way to do things is just too racy */
1030 #if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1031 /* on linux we may be able to ask the kernel */
1032 {
1033 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1034
1035 if (fd >= 0)
1036 {
1037 sprintf (tmp1, "/proc/self/fd/%d", fd);
1038 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1039 /* here we should probably stat the open file and the disk file, to make sure they still match */
1040 eio__close (fd);
1041
1042 if (req->result > 0)
1043 goto done;
1044 }
1045 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1046 return -1;
1047 }
1048 #endif
1049 #endif
1050
1051 if (*rel != '/')
1052 {
1053 int len;
1054
1055 errno = ENOENT;
1056 if (wd == EIO_INVALID_WD)
1057 return -1;
1058
1059 if (wd == EIO_CWD)
1060 {
1061 if (!getcwd (res, EIO_PATH_MAX))
1062 return -1;
1063
1064 len = strlen (res);
1065 }
1066 else
1067 memcpy (res, wd->str, len = wd->len);
1068
1069 if (res [1]) /* only use if not / */
1070 res += len;
1071 }
1072
1073 while (*rel)
1074 {
1075 eio_ssize_t len, linklen;
1076 const char *beg = rel;
1077
1078 while (*rel && *rel != '/')
1079 ++rel;
1080
1081 len = rel - beg;
1082
1083 if (!len) /* skip slashes */
1084 {
1085 ++rel;
1086 continue;
1087 }
1088
1089 if (beg [0] == '.')
1090 {
1091 if (len == 1)
1092 continue; /* . - nop */
1093
1094 if (beg [1] == '.' && len == 2)
1095 {
1096 /* .. - back up one component, if possible */
1097
1098 while (res != tmpbuf->ptr)
1099 if (*--res == '/')
1100 break;
1101
1102 continue;
1103 }
1104 }
1105
1106 errno = ENAMETOOLONG;
1107 if (res + 1 + len + 1 >= tmp1)
1108 return -1;
1109
1110 /* copy one component */
1111 *res = '/';
1112 memcpy (res + 1, beg, len);
1113
1114 /* zero-terminate, for readlink */
1115 res [len + 1] = 0;
1116
1117 /* now check if it's a symlink */
1118 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1119
1120 if (linklen < 0)
1121 {
1122 if (errno != EINVAL)
1123 return -1;
1124
1125 /* it's a normal directory. hopefully */
1126 res += len + 1;
1127 }
1128 else
1129 {
1130 /* yay, it was a symlink - build new path in tmp2 */
1131 int rellen = strlen (rel);
1132
1133 errno = ENAMETOOLONG;
1134 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1135 return -1;
1136
1137 errno = ELOOP;
1138 if (!--symlinks)
1139 return -1;
1140
1141 if (*tmp1 == '/')
1142 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1143
1144 /* we need to be careful, as rel might point into tmp2 already */
1145 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1146 tmp2 [linklen] = '/';
1147 memcpy (tmp2, tmp1, linklen);
1148
1149 rel = tmp2;
1150 }
1151 }
1152
1153 /* special case for the lone root path */
1154 if (res == tmpbuf->ptr)
1155 *res++ = '/';
1156
1157 return res - (char *)tmpbuf->ptr;
1158 #endif
1159 }
1160
1161 static signed char
1162 eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1163 {
1164 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1165 : a->inode < b->inode ? -1
1166 : a->inode > b->inode ? 1
1167 : 0;
1168 }
1169
1170 #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1171
1172 #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1173 #define EIO_SORT_FAST 60 /* when to only use insertion sort */
1174
1175 static void
1176 eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1177 {
1178 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1179 unsigned char *bit = bits;
1180
1181 assert (CHAR_BIT == 8);
1182 assert (sizeof (eio_dirent) * 8 < 256);
1183 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1184 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1185
1186 if (size <= EIO_SORT_FAST)
1187 return;
1188
1189 /* first prepare an array of bits to test in our radix sort */
1190 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1191 /* inode_bits must contain all inodes ORed together */
1192 /* which is used to skip bits that are 0 everywhere, which is very common */
1193 {
1194 eio_ino_t endianness;
1195 int i, j;
1196
1197 /* we store the byte offset of byte n into byte n of "endianness" */
1198 for (i = 0; i < sizeof (eio_ino_t); ++i)
1199 ((unsigned char *)&endianness)[i] = i;
1200
1201 *bit++ = 0;
1202
1203 for (i = 0; i < sizeof (eio_ino_t); ++i)
1204 {
1205 /* shifting off the byte offsets out of "endianness" */
1206 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1207 endianness >>= 8;
1208
1209 for (j = 0; j < 8; ++j)
1210 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1211 *bit++ = offs + j;
1212 }
1213
1214 for (j = 0; j < 8; ++j)
1215 if (score_bits & (1 << j))
1216 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1217 }
1218
1219 /* now actually do the sorting (a variant of MSD radix sort) */
1220 {
1221 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1222 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1223 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1224 int stk_idx = 0;
1225
1226 base_stk [stk_idx] = dents;
1227 end_stk [stk_idx] = dents + size;
1228 bit_stk [stk_idx] = bit - 1;
1229
1230 do
1231 {
1232 base = base_stk [stk_idx];
1233 end = end_stk [stk_idx];
1234 bit = bit_stk [stk_idx];
1235
1236 for (;;)
1237 {
1238 unsigned char O = *bit >> 3;
1239 unsigned char M = 1 << (*bit & 7);
1240
1241 eio_dirent *a = base;
1242 eio_dirent *b = end;
1243
1244 if (b - a < EIO_SORT_CUTOFF)
1245 break;
1246
1247 /* now bit-partition the array on the bit */
1248 /* this ugly asymmetric loop seems to perform much better than typical */
1249 /* partition algos found in the literature */
1250 do
1251 if (!(((unsigned char *)a)[O] & M))
1252 ++a;
1253 else if (!(((unsigned char *)--b)[O] & M))
1254 {
1255 eio_dirent tmp = *a; *a = *b; *b = tmp;
1256 ++a;
1257 }
1258 while (b > a);
1259
1260 /* next bit, or stop, if no bits left in this path */
1261 if (!*--bit)
1262 break;
1263
1264 base_stk [stk_idx] = a;
1265 end_stk [stk_idx] = end;
1266 bit_stk [stk_idx] = bit;
1267 ++stk_idx;
1268
1269 end = a;
1270 }
1271 }
1272 while (stk_idx--);
1273 }
1274 }
1275
1276 static void
1277 eio_dent_insertion_sort (eio_dirent *dents, int size)
1278 {
1279 /* first move the smallest element to the front, to act as a sentinel */
1280 {
1281 int i;
1282 eio_dirent *min = dents;
1283
1284 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1285 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1286 if (EIO_DENT_CMP (dents [i], <, *min))
1287 min = &dents [i];
1288
1289 /* swap elements 0 and j (minimum) */
1290 {
1291 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1292 }
1293 }
1294
1295 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1296 {
1297 eio_dirent *i, *j;
1298
1299 for (i = dents + 1; i < dents + size; ++i)
1300 {
1301 eio_dirent value = *i;
1302
1303 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1304 j [1] = j [0];
1305
1306 j [1] = value;
1307 }
1308 }
1309 }
1310
1311 static void
1312 eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1313 {
1314 if (size <= 1)
1315 return; /* our insertion sort relies on size > 0 */
1316
1317 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1318 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1319 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1320
1321 /* use an insertion sort at the end, or for small arrays, */
1322 /* as insertion sort is more efficient for small partitions */
1323 eio_dent_insertion_sort (dents, size);
1324 }
1325
1326 /* read a full directory */
1327 static void
1328 eio__scandir (eio_req *req, etp_worker *self)
1329 {
1330 char *name, *names;
1331 int namesalloc = 4096 - sizeof (void *) * 4;
1332 int namesoffs = 0;
1333 int flags = req->int1;
1334 eio_dirent *dents = 0;
1335 int dentalloc = 128;
1336 int dentoffs = 0;
1337 eio_ino_t inode_bits = 0;
1338 #ifdef _WIN32
1339 HANDLE dirp;
1340 WIN32_FIND_DATA entp;
1341 #else
1342 DIR *dirp;
1343 EIO_STRUCT_DIRENT *entp;
1344 #endif
1345
1346 req->result = -1;
1347
1348 if (!(flags & EIO_READDIR_DENTS))
1349 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1350
1351 #ifdef _WIN32
1352 {
1353 int len = strlen ((const char *)req->ptr1);
1354 char *path = malloc (MAX_PATH);
1355 const char *fmt;
1356 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1357
1358 if (!len)
1359 fmt = "./*";
1360 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1361 fmt = "%s*";
1362 else
1363 fmt = "%s/*";
1364
1365 _snprintf (path, MAX_PATH, fmt, reqpath);
1366 dirp = FindFirstFile (path, &entp);
1367 free (path);
1368
1369 if (dirp == INVALID_HANDLE_VALUE)
1370 {
1371 /* should steal _dosmaperr */
1372 switch (GetLastError ())
1373 {
1374 case ERROR_FILE_NOT_FOUND:
1375 req->result = 0;
1376 break;
1377
1378 case ERROR_INVALID_NAME:
1379 case ERROR_PATH_NOT_FOUND:
1380 case ERROR_NO_MORE_FILES:
1381 errno = ENOENT;
1382 break;
1383
1384 case ERROR_NOT_ENOUGH_MEMORY:
1385 errno = ENOMEM;
1386 break;
1387
1388 default:
1389 errno = EINVAL;
1390 break;
1391 }
1392
1393 return;
1394 }
1395 }
1396 #else
1397 #if HAVE_AT
1398 if (req->wd)
1399 {
1400 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1401
1402 if (fd < 0)
1403 return;
1404
1405 dirp = fdopendir (fd);
1406
1407 if (!dirp)
1408 silent_close (fd);
1409 }
1410 else
1411 dirp = opendir (req->ptr1);
1412 #else
1413 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1414 #endif
1415
1416 if (!dirp)
1417 return;
1418 #endif
1419
1420 if (req->flags & EIO_FLAG_PTR1_FREE)
1421 free (req->ptr1);
1422
1423 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1424 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1425 req->ptr2 = names = malloc (namesalloc);
1426
1427 if (!names || (flags && !dents))
1428 return;
1429
1430 for (;;)
1431 {
1432 int done;
1433
1434 #ifdef _WIN32
1435 done = !dirp;
1436 #else
1437 errno = 0;
1438 entp = readdir (dirp);
1439 done = !entp;
1440 #endif
1441
1442 if (done)
1443 {
1444 #ifndef _WIN32
1445 int old_errno = errno;
1446 closedir (dirp);
1447 errno = old_errno;
1448
1449 if (errno)
1450 break;
1451 #endif
1452
1453 /* sort etc. */
1454 req->int1 = flags;
1455 req->result = dentoffs;
1456
1457 if (flags & EIO_READDIR_STAT_ORDER)
1458 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1459 else if (flags & EIO_READDIR_DIRS_FIRST)
1460 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1461 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1462 else
1463 {
1464 /* in this case, all is known, and we just put dirs first and sort them */
1465 eio_dirent *oth = dents + dentoffs;
1466 eio_dirent *dir = dents;
1467
1468 /* now partition dirs to the front, and non-dirs to the back */
1469 /* by walking from both sides and swapping if necessary */
1470 while (oth > dir)
1471 {
1472 if (dir->type == EIO_DT_DIR)
1473 ++dir;
1474 else if ((--oth)->type == EIO_DT_DIR)
1475 {
1476 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1477
1478 ++dir;
1479 }
1480 }
1481
1482 /* now sort the dirs only (dirs all have the same score) */
1483 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1484 }
1485
1486 break;
1487 }
1488
1489 /* now add the entry to our list(s) */
1490 name = D_NAME (entp);
1491
1492 /* skip . and .. entries */
1493 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1494 {
1495 int len = D_NAMLEN (entp) + 1;
1496
1497 while (ecb_expect_false (namesoffs + len > namesalloc))
1498 {
1499 namesalloc *= 2;
1500 req->ptr2 = names = realloc (names, namesalloc);
1501
1502 if (!names)
1503 break;
1504 }
1505
1506 memcpy (names + namesoffs, name, len);
1507
1508 if (dents)
1509 {
1510 struct eio_dirent *ent;
1511
1512 if (ecb_expect_false (dentoffs == dentalloc))
1513 {
1514 dentalloc *= 2;
1515 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1516
1517 if (!dents)
1518 break;
1519 }
1520
1521 ent = dents + dentoffs;
1522
1523 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1524 ent->namelen = len - 1;
1525 ent->inode = D_INO (entp);
1526
1527 inode_bits |= ent->inode;
1528
1529 switch (D_TYPE (entp))
1530 {
1531 default:
1532 ent->type = EIO_DT_UNKNOWN;
1533 flags |= EIO_READDIR_FOUND_UNKNOWN;
1534 break;
1535
1536 #ifdef DT_FIFO
1537 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1538 #endif
1539 #ifdef DT_CHR
1540 case DT_CHR: ent->type = EIO_DT_CHR; break;
1541 #endif
1542 #ifdef DT_MPC
1543 case DT_MPC: ent->type = EIO_DT_MPC; break;
1544 #endif
1545 #ifdef DT_DIR
1546 case DT_DIR: ent->type = EIO_DT_DIR; break;
1547 #endif
1548 #ifdef DT_NAM
1549 case DT_NAM: ent->type = EIO_DT_NAM; break;
1550 #endif
1551 #ifdef DT_BLK
1552 case DT_BLK: ent->type = EIO_DT_BLK; break;
1553 #endif
1554 #ifdef DT_MPB
1555 case DT_MPB: ent->type = EIO_DT_MPB; break;
1556 #endif
1557 #ifdef DT_REG
1558 case DT_REG: ent->type = EIO_DT_REG; break;
1559 #endif
1560 #ifdef DT_NWK
1561 case DT_NWK: ent->type = EIO_DT_NWK; break;
1562 #endif
1563 #ifdef DT_CMP
1564 case DT_CMP: ent->type = EIO_DT_CMP; break;
1565 #endif
1566 #ifdef DT_LNK
1567 case DT_LNK: ent->type = EIO_DT_LNK; break;
1568 #endif
1569 #ifdef DT_SOCK
1570 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1571 #endif
1572 #ifdef DT_DOOR
1573 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1574 #endif
1575 #ifdef DT_WHT
1576 case DT_WHT: ent->type = EIO_DT_WHT; break;
1577 #endif
1578 }
1579
1580 ent->score = 7;
1581
1582 if (flags & EIO_READDIR_DIRS_FIRST)
1583 {
1584 if (ent->type == EIO_DT_UNKNOWN)
1585 {
1586 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1587 ent->score = 1;
1588 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1589 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1590 }
1591 else if (ent->type == EIO_DT_DIR)
1592 ent->score = 0;
1593 }
1594 }
1595
1596 namesoffs += len;
1597 ++dentoffs;
1598 }
1599
1600 if (EIO_CANCELLED (req))
1601 {
1602 errno = ECANCELED;
1603 break;
1604 }
1605
1606 #ifdef _WIN32
1607 if (!FindNextFile (dirp, &entp))
1608 {
1609 FindClose (dirp);
1610 dirp = 0;
1611 }
1612 #endif
1613 }
1614 }
1615
1616 /*****************************************************************************/
1617 /* working directory stuff */
1618 /* various deficiencies in the posix 2008 api force us to */
1619 /* keep the absolute path in string form at all times */
1620 /* fuck yeah. */
1621
1622 #if !HAVE_AT
1623
1624 /* a bit like realpath, but usually faster because it doesn'T have to return */
1625 /* an absolute or canonical path */
1626 static const char *
1627 wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1628 {
1629 if (!wd || *path == '/')
1630 return path;
1631
1632 if (path [0] == '.' && !path [1])
1633 return wd->str;
1634
1635 {
1636 int l1 = wd->len;
1637 int l2 = strlen (path);
1638
1639 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1640
1641 memcpy (res, wd->str, l1);
1642 res [l1] = '/';
1643 memcpy (res + l1 + 1, path, l2 + 1);
1644
1645 return res;
1646 }
1647 }
1648
1649 #endif
1650
1651 static eio_wd
1652 eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1653 {
1654 int fd;
1655 eio_wd res;
1656 int len = eio__realpath (tmpbuf, wd, path);
1657
1658 if (len < 0)
1659 return EIO_INVALID_WD;
1660
1661 #if HAVE_AT
1662 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1663
1664 if (fd < 0)
1665 return EIO_INVALID_WD;
1666 #endif
1667
1668 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1669
1670 #if HAVE_AT
1671 res->fd = fd;
1672 #endif
1673
1674 res->len = len;
1675 memcpy (res->str, tmpbuf->ptr, len);
1676 res->str [len] = 0;
1677
1678 return res;
1679 }
1680
1681 eio_wd
1682 eio_wd_open_sync (eio_wd wd, const char *path)
1683 {
1684 struct etp_tmpbuf tmpbuf = { };
1685 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1686 free (tmpbuf.ptr);
1687
1688 return wd;
1689 }
1690
1691 void
1692 eio_wd_close_sync (eio_wd wd)
1693 {
1694 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1695 {
1696 #if HAVE_AT
1697 eio__close (wd->fd);
1698 #endif
1699 free (wd);
1700 }
1701 }
1702
1703 #if HAVE_AT
1704
1705 static int
1706 eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1707 {
1708 #if HAVE_RENAMEAT2
1709 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1710 #else
1711 if (flags)
1712 return EIO_ENOSYS ();
1713
1714 return renameat (olddirfd, oldpath, newdirfd, newpath);
1715 #endif
1716 }
1717
1718 /* they forgot these */
1719
1720 static int
1721 eio__truncateat (int dirfd, const char *path, off_t length)
1722 {
1723 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1724 int res;
1725
1726 if (fd < 0)
1727 return fd;
1728
1729 res = ftruncate (fd, length);
1730 silent_close (fd);
1731 return res;
1732 }
1733
1734 static int
1735 eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1736 {
1737 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1738 int res;
1739
1740 if (fd < 0)
1741 return fd;
1742
1743 res = fstatvfs (fd, buf);
1744 silent_close (fd);
1745 return res;
1746 }
1747
1748 #endif
1749
1750 /*****************************************************************************/
1751
1752 #define ALLOC(len) \
1753 if (!req->ptr2) \
1754 { \
1755 X_LOCK (EIO_POOL->wrklock); \
1756 req->flags |= EIO_FLAG_PTR2_FREE; \
1757 X_UNLOCK (EIO_POOL->wrklock); \
1758 req->ptr2 = malloc (len); \
1759 if (!req->ptr2) \
1760 { \
1761 errno = ENOMEM; \
1762 req->result = -1; \
1763 goto alloc_fail; \
1764 } \
1765 }
1766
1767 /*****************************************************************************/
1768
1769 static void
1770 eio__slurp (int fd, eio_req *req)
1771 {
1772 req->result = fd;
1773
1774 if (fd < 0)
1775 return;
1776
1777 if (req->offs < 0 || !req->size) /* do we need the size? */
1778 {
1779 off_t size = lseek (fd, 0, SEEK_END);
1780
1781 if (req->offs < 0)
1782 req->offs += size;
1783
1784 if (!req->size)
1785 req->size = size - req->offs;
1786 }
1787
1788 ALLOC (req->size);
1789 req->result = pread (fd, req->ptr2, req->size, req->offs);
1790
1791 silent_close (fd);
1792
1793 alloc_fail:
1794 ;
1795 }
1796
1797 int ecb_cold
1798 eio_init (void (*want_poll)(void), void (*done_poll)(void))
1799 {
1800 eio_want_poll_cb = want_poll;
1801 eio_done_poll_cb = done_poll;
1802
1803 return etp_init (EIO_POOL, 0, 0, 0);
1804 }
1805
1806 ecb_inline void
1807 eio_api_destroy (eio_req *req)
1808 {
1809 free (req);
1810 }
1811
1812 #define REQ(rtype) \
1813 eio_req *req; \
1814 \
1815 req = (eio_req *)calloc (1, sizeof *req); \
1816 if (!req) \
1817 return 0; \
1818 \
1819 req->type = rtype; \
1820 req->pri = pri; \
1821 req->finish = cb; \
1822 req->data = data; \
1823 req->destroy = eio_api_destroy;
1824
1825 #define SEND eio_submit (req); return req
1826
1827 #define PATH \
1828 req->flags |= EIO_FLAG_PTR1_FREE; \
1829 req->ptr1 = strdup (path); \
1830 if (!req->ptr1) \
1831 { \
1832 eio_api_destroy (req); \
1833 return 0; \
1834 }
1835
1836 #define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1837
1838 static void
1839 eio_execute (etp_worker *self, eio_req *req)
1840 {
1841 #if HAVE_AT
1842 int dirfd;
1843 #else
1844 const char *path;
1845 #endif
1846
1847 if (ecb_expect_false (EIO_CANCELLED (req)))
1848 {
1849 req->result = -1;
1850 req->errorno = ECANCELED;
1851 return;
1852 }
1853
1854 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1855 {
1856 req->result = -1;
1857 req->errorno = ENOENT;
1858 return;
1859 }
1860
1861 if (req->type >= EIO_OPEN)
1862 {
1863 #if HAVE_AT
1864 dirfd = WD2FD (req->wd);
1865 #else
1866 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1867 #endif
1868 }
1869
1870 switch (req->type)
1871 {
1872 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1873 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1874 break;
1875 case EIO_WD_CLOSE: req->result = 0;
1876 eio_wd_close_sync (req->wd); break;
1877
1878 case EIO_SEEK: eio__lseek (req); break;
1879 case EIO_READ: ALLOC (req->size);
1880 req->result = req->offs >= 0
1881 ? pread (req->int1, req->ptr2, req->size, req->offs)
1882 : read (req->int1, req->ptr2, req->size); break;
1883 case EIO_WRITE: req->result = req->offs >= 0
1884 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1885 : write (req->int1, req->ptr2, req->size); break;
1886
1887 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1888 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1889
1890 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1891 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1892
1893 #if HAVE_AT
1894
1895 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1896 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1897 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1898 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1899 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1900 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1901 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1902 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1903 case EIO_SLURP: eio__slurp ( openat (dirfd, req->ptr1, O_RDONLY | O_CLOEXEC), req); break;
1904
1905 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1906 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1907 req->result = req->wd && SINGLEDOT (req->ptr1)
1908 ? rmdir (req->wd->str)
1909 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1910 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1911 case EIO_RENAME: req->result = eio__renameat2 (
1912 dirfd,
1913 /* complications arise because "." cannot be renamed, so we might have to expand */
1914 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1915 WD2FD ((eio_wd)req->int3),
1916 req->ptr2,
1917 req->int2
1918 );
1919 break;
1920 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1921 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1922 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1923 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1924 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1925 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1926 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1927 if (req->result == EIO_PATH_MAX)
1928 {
1929 req->result = -1;
1930 errno = ENAMETOOLONG;
1931 }
1932 break;
1933 case EIO_UTIME:
1934 case EIO_FUTIME:
1935 {
1936 struct timespec ts[2];
1937 struct timespec *times;
1938
1939 if (req->nv1 != -1. || req->nv2 != -1.)
1940 {
1941 ts[0].tv_sec = req->nv1;
1942 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1943 ts[1].tv_sec = req->nv2;
1944 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1945
1946 times = ts;
1947 }
1948 else
1949 times = 0;
1950
1951 req->result = req->type == EIO_FUTIME
1952 ? futimens (req->int1, times)
1953 : utimensat (dirfd, req->ptr1, times, 0);
1954 }
1955 break;
1956
1957 #else
1958
1959 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1960 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1961 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1962 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1963 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1964 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1965 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1966 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1967 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1968
1969 case EIO_UNLINK: req->result = unlink (path ); break;
1970 case EIO_RMDIR: req->result = rmdir (path ); break;
1971 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1972 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
1973 case EIO_LINK: req->result = link (path , req->ptr2); break;
1974 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1975 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1976 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1977 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1978 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1979 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
1980 if (req->result == EIO_PATH_MAX)
1981 {
1982 req->result = -1;
1983 errno = ENAMETOOLONG;
1984 }
1985 break;
1986
1987 case EIO_UTIME:
1988 case EIO_FUTIME:
1989 {
1990 struct timeval tv[2];
1991 struct timeval *times;
1992
1993 if (req->nv1 != -1. || req->nv2 != -1.)
1994 {
1995 tv[0].tv_sec = req->nv1;
1996 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1997 tv[1].tv_sec = req->nv2;
1998 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1999
2000 times = tv;
2001 }
2002 else
2003 times = 0;
2004
2005 req->result = req->type == EIO_FUTIME
2006 ? futimes (req->int1, times)
2007 : utimes (req->ptr1, times);
2008 }
2009 break;
2010
2011 #endif
2012
2013 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2014 {
2015 ALLOC (req->result);
2016 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2017 }
2018 break;
2019
2020 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2021 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2022
2023 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2024 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2025
2026 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2027 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2028 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2029
2030 case EIO_CLOSE: req->result = eio__close (req->int1); break;
2031 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2032 case EIO_SYNC: req->result = 0; sync (); break;
2033 case EIO_FSYNC: req->result = fsync (req->int1); break;
2034 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2035 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2036 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2037 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2038 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2039 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2040 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2041 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2042
2043 case EIO_READDIR: eio__scandir (req, self); break;
2044
2045 case EIO_BUSY:
2046 #ifdef _WIN32
2047 Sleep (req->nv1 * 1e3);
2048 #else
2049 {
2050 struct timeval tv;
2051
2052 tv.tv_sec = req->nv1;
2053 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2054
2055 req->result = select (0, 0, 0, 0, &tv);
2056 }
2057 #endif
2058 break;
2059
2060 #if 0
2061 case EIO_GROUP:
2062 abort (); /* handled in eio_request */
2063 #endif
2064
2065 case EIO_NOP:
2066 req->result = 0;
2067 break;
2068
2069 case EIO_CUSTOM:
2070 req->feed (req);
2071 break;
2072
2073 default:
2074 req->result = EIO_ENOSYS ();
2075 break;
2076 }
2077
2078 alloc_fail:
2079 req->errorno = errno;
2080 }
2081
2082 #ifndef EIO_NO_WRAPPERS
2083
2084 eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2085 {
2086 REQ (EIO_WD_OPEN); PATH; SEND;
2087 }
2088
2089 eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2090 {
2091 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2092 }
2093
2094 eio_req *eio_nop (int pri, eio_cb cb, void *data)
2095 {
2096 REQ (EIO_NOP); SEND;
2097 }
2098
2099 eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2100 {
2101 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2102 }
2103
2104 eio_req *eio_sync (int pri, eio_cb cb, void *data)
2105 {
2106 REQ (EIO_SYNC); SEND;
2107 }
2108
2109 eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2110 {
2111 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2112 }
2113
2114 eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2115 {
2116 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2117 }
2118
2119 eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2120 {
2121 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2122 }
2123
2124 eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2125 {
2126 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2127 }
2128
2129 eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2130 {
2131 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2132 }
2133
2134 eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2135 {
2136 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2137 }
2138
2139 eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2140 {
2141 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2142 }
2143
2144 eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2145 {
2146 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2147 }
2148
2149 eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2150 {
2151 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2152 }
2153
2154 eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2155 {
2156 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2157 }
2158
2159 eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2160 {
2161 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2162 }
2163
2164 eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2165 {
2166 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2167 }
2168
2169 eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2170 {
2171 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2172 }
2173
2174 eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2175 {
2176 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2177 }
2178
2179 eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2180 {
2181 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2182 }
2183
2184 eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2185 {
2186 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2187 }
2188
2189 eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2190 {
2191 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2192 }
2193
2194 eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2195 {
2196 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2197 }
2198
2199 eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2200 {
2201 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2202 }
2203
2204 eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2205 {
2206 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2207 }
2208
2209 eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2210 {
2211 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2212 }
2213
2214 eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2215 {
2216 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2217 }
2218
2219 eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2220 {
2221 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2222 }
2223
2224 eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2225 {
2226 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2227 }
2228
2229 eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
2230 {
2231 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
2232 }
2233
2234 eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2235 {
2236 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2237 }
2238
2239 eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2240 {
2241 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2242 }
2243
2244 eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2245 {
2246 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2247 }
2248
2249 eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2250 {
2251 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2252 }
2253
2254 eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2255 {
2256 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2257 }
2258
2259 static eio_req *
2260 eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2261 {
2262 REQ (type); PATH; SEND;
2263 }
2264
2265 eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2266 {
2267 return eio__1path (EIO_READLINK, path, pri, cb, data);
2268 }
2269
2270 eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2271 {
2272 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2273 }
2274
2275 eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2276 {
2277 return eio__1path (EIO_STAT, path, pri, cb, data);
2278 }
2279
2280 eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2281 {
2282 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2283 }
2284
2285 eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2286 {
2287 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2288 }
2289
2290 eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2291 {
2292 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2293 }
2294
2295 eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
2296 {
2297 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2298 }
2299
2300 eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2301 {
2302 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2303 }
2304
2305 eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
2306 {
2307 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2308 }
2309
2310 static eio_req *
2311 eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2312 {
2313 REQ (type); PATH;
2314
2315 req->flags |= EIO_FLAG_PTR2_FREE;
2316 req->ptr2 = strdup (new_path);
2317 if (!req->ptr2)
2318 {
2319 eio_api_destroy (req);
2320 return 0;
2321 }
2322
2323 SEND;
2324 }
2325
2326 eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2327 {
2328 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2329 }
2330
2331 eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2332 {
2333 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2334 }
2335
2336 eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2337 {
2338 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2339 }
2340
2341 eio_req *eio_slurp (const char *path, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2342 {
2343 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2344 }
2345
2346 eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2347 {
2348 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2349 }
2350
2351 #endif
2352
2353 eio_req *eio_grp (eio_cb cb, void *data)
2354 {
2355 const int pri = EIO_PRI_MAX;
2356
2357 REQ (EIO_GROUP); SEND;
2358 }
2359
2360 #undef REQ
2361 #undef PATH
2362 #undef SEND
2363
2364 /*****************************************************************************/
2365 /* grp functions */
2366
2367 void
2368 eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2369 {
2370 grp->int2 = limit;
2371 grp->feed = feed;
2372
2373 grp_try_feed (grp);
2374 }
2375
2376 void
2377 eio_grp_limit (eio_req *grp, int limit)
2378 {
2379 grp->int2 = limit;
2380
2381 grp_try_feed (grp);
2382 }
2383
2384 void
2385 eio_grp_add (eio_req *grp, eio_req *req)
2386 {
2387 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2388
2389 grp->flags |= ETP_FLAG_GROUPADD;
2390
2391 ++grp->size;
2392 req->grp = grp;
2393
2394 req->grp_prev = 0;
2395 req->grp_next = grp->grp_first;
2396
2397 if (grp->grp_first)
2398 grp->grp_first->grp_prev = req;
2399
2400 grp->grp_first = req;
2401 }
2402
2403 /*****************************************************************************/
2404 /* misc garbage */
2405
2406 eio_ssize_t
2407 eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2408 {
2409 return eio__sendfile (ofd, ifd, offset, count);
2410 }
2411