1 |
/* |
2 |
* libeio implementation |
3 |
* |
4 |
* Copyright (c) 2007,2008,2009,2010,2011,2012 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 |
#ifdef EIO_STACKSIZE |
48 |
# define X_STACKSIZE EIO_STACKSIZE |
49 |
#endif |
50 |
#include "xthread.h" |
51 |
|
52 |
#include <errno.h> |
53 |
#include <stddef.h> |
54 |
#include <stdlib.h> |
55 |
#include <string.h> |
56 |
#include <errno.h> |
57 |
#include <sys/types.h> |
58 |
#include <sys/stat.h> |
59 |
#include <limits.h> |
60 |
#include <fcntl.h> |
61 |
#include <assert.h> |
62 |
|
63 |
/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ |
64 |
/* intptr_t only comes from stdint.h, says idiot openbsd coder */ |
65 |
#if HAVE_STDINT_H |
66 |
# include <stdint.h> |
67 |
#endif |
68 |
|
69 |
#ifndef ECANCELED |
70 |
# define ECANCELED EDOM |
71 |
#endif |
72 |
#ifndef ELOOP |
73 |
# define ELOOP EDOM |
74 |
#endif |
75 |
|
76 |
#if !defined(ENOTSOCK) && defined(WSAENOTSOCK) |
77 |
# define ENOTSOCK WSAENOTSOCK |
78 |
#endif |
79 |
|
80 |
static void eio_destroy (eio_req *req); |
81 |
|
82 |
#ifndef EIO_FINISH |
83 |
# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
84 |
#endif |
85 |
|
86 |
#ifndef EIO_DESTROY |
87 |
# define EIO_DESTROY(req) do { if ((req)->destroy) (req)->destroy (req); } while (0) |
88 |
#endif |
89 |
|
90 |
#ifndef EIO_FEED |
91 |
# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) |
92 |
#endif |
93 |
|
94 |
#ifndef EIO_FD_TO_WIN32_HANDLE |
95 |
# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd) |
96 |
#endif |
97 |
#ifndef EIO_WIN32_HANDLE_TO_FD |
98 |
# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0) |
99 |
#endif |
100 |
|
101 |
#define EIO_ERRNO(errval,retval) ((errno = errval), retval) |
102 |
|
103 |
#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1) |
104 |
|
105 |
#ifdef _WIN32 |
106 |
|
107 |
#undef PAGESIZE |
108 |
#define PAGESIZE 4096 /* GetSystemInfo? */ |
109 |
|
110 |
/* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */ |
111 |
|
112 |
#ifdef EIO_STRUCT_STATI64 |
113 |
/* look at perl's non-sloppy stat */ |
114 |
#define stat(path,buf) _stati64 (path,buf) |
115 |
#define fstat(fd,buf) _fstati64 (fd,buf) |
116 |
#endif |
117 |
#define lstat(path,buf) stat (path,buf) |
118 |
#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1)) |
119 |
#define mkdir(path,mode) _mkdir (path) |
120 |
#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1)) |
121 |
|
122 |
#define chmod(path,mode) _chmod (path, mode) |
123 |
#define dup(fd) _dup (fd) |
124 |
#define dup2(fd1,fd2) _dup2 (fd1, fd2) |
125 |
|
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 |
if (CreateSymbolicLink (neu, old, 1)) |
209 |
return 0; |
210 |
|
211 |
if (CreateSymbolicLink (neu, old, 0)) |
212 |
return 0; |
213 |
#endif |
214 |
|
215 |
return EIO_ERRNO (ENOENT, -1); |
216 |
} |
217 |
|
218 |
/* POSIX API only */ |
219 |
#define CreateHardLink(neu,old,flags) 0 |
220 |
#define CreateSymbolicLink(neu,old,flags) 0 |
221 |
|
222 |
struct statvfs |
223 |
{ |
224 |
int dummy; |
225 |
}; |
226 |
|
227 |
#define DT_DIR EIO_DT_DIR |
228 |
#define DT_REG EIO_DT_REG |
229 |
#define D_NAME(entp) entp.cFileName |
230 |
#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG) |
231 |
|
232 |
#else |
233 |
|
234 |
#include <sys/time.h> |
235 |
#include <sys/select.h> |
236 |
#include <unistd.h> |
237 |
#include <signal.h> |
238 |
#include <dirent.h> |
239 |
|
240 |
#if ANDROID |
241 |
#include <sys/vfs.h> |
242 |
#define statvfs statfs |
243 |
#define fstatvfs fstatfs |
244 |
#include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */ |
245 |
#else |
246 |
#include <sys/statvfs.h> |
247 |
#endif |
248 |
|
249 |
#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES |
250 |
#include <sys/mman.h> |
251 |
#endif |
252 |
|
253 |
#define D_NAME(entp) entp->d_name |
254 |
|
255 |
/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
256 |
#if __FreeBSD__ || __NetBSD__ || __OpenBSD__ |
257 |
#define _DIRENT_HAVE_D_TYPE /* sigh */ |
258 |
#define D_INO(de) (de)->d_fileno |
259 |
#define D_NAMLEN(de) (de)->d_namlen |
260 |
#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
261 |
#define D_INO(de) (de)->d_ino |
262 |
#endif |
263 |
|
264 |
#ifdef _D_EXACT_NAMLEN |
265 |
#undef D_NAMLEN |
266 |
#define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
267 |
#endif |
268 |
|
269 |
#ifdef _DIRENT_HAVE_D_TYPE |
270 |
#define D_TYPE(de) (de)->d_type |
271 |
#endif |
272 |
|
273 |
#ifndef EIO_STRUCT_DIRENT |
274 |
#define EIO_STRUCT_DIRENT struct dirent |
275 |
#endif |
276 |
|
277 |
#endif |
278 |
|
279 |
#if HAVE_UTIMES |
280 |
# include <utime.h> |
281 |
#endif |
282 |
|
283 |
#if HAVE_SYS_SYSCALL_H |
284 |
# include <sys/syscall.h> |
285 |
#endif |
286 |
|
287 |
#if HAVE_SYS_PRCTL_H |
288 |
# include <sys/prctl.h> |
289 |
#endif |
290 |
|
291 |
#if HAVE_SENDFILE |
292 |
# if __linux |
293 |
# include <sys/sendfile.h> |
294 |
# elif __FreeBSD__ || defined __APPLE__ |
295 |
# include <sys/socket.h> |
296 |
# include <sys/uio.h> |
297 |
# elif __hpux |
298 |
# include <sys/socket.h> |
299 |
# elif __solaris |
300 |
# include <sys/sendfile.h> |
301 |
# else |
302 |
# error sendfile support requested but not available |
303 |
# endif |
304 |
#endif |
305 |
|
306 |
#ifndef D_TYPE |
307 |
# define D_TYPE(de) 0 |
308 |
#endif |
309 |
#ifndef D_INO |
310 |
# define D_INO(de) 0 |
311 |
#endif |
312 |
#ifndef D_NAMLEN |
313 |
# define D_NAMLEN(entp) strlen (D_NAME (entp)) |
314 |
#endif |
315 |
|
316 |
/* used for struct dirent, AIX doesn't provide it */ |
317 |
#ifndef NAME_MAX |
318 |
# define NAME_MAX 4096 |
319 |
#endif |
320 |
|
321 |
/* used for readlink etc. */ |
322 |
#ifndef PATH_MAX |
323 |
# define PATH_MAX 4096 |
324 |
#endif |
325 |
|
326 |
/* buffer size for various temporary buffers */ |
327 |
#define EIO_BUFSIZE 65536 |
328 |
|
329 |
#define dBUF \ |
330 |
char *eio_buf = malloc (EIO_BUFSIZE); \ |
331 |
errno = ENOMEM; \ |
332 |
if (!eio_buf) \ |
333 |
return -1 |
334 |
|
335 |
#define FUBd \ |
336 |
free (eio_buf) |
337 |
|
338 |
#define EIO_TICKS ((1000000 + 1023) >> 10) |
339 |
|
340 |
/*****************************************************************************/ |
341 |
|
342 |
struct tmpbuf |
343 |
{ |
344 |
void *ptr; |
345 |
int len; |
346 |
}; |
347 |
|
348 |
static void * |
349 |
tmpbuf_get (struct tmpbuf *buf, int len) |
350 |
{ |
351 |
if (buf->len < len) |
352 |
{ |
353 |
free (buf->ptr); |
354 |
buf->ptr = malloc (buf->len = len); |
355 |
} |
356 |
|
357 |
return buf->ptr; |
358 |
} |
359 |
|
360 |
struct tmpbuf; |
361 |
|
362 |
#if _POSIX_VERSION >= 200809L |
363 |
#define HAVE_AT 1 |
364 |
#define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD) |
365 |
#ifndef O_SEARCH |
366 |
#define O_SEARCH O_RDONLY |
367 |
#endif |
368 |
#else |
369 |
#define HAVE_AT 0 |
370 |
static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path); |
371 |
#endif |
372 |
|
373 |
struct eio_pwd |
374 |
{ |
375 |
#if HAVE_AT |
376 |
int fd; |
377 |
#endif |
378 |
int len; |
379 |
char str[1]; /* actually, a 0-terminated canonical path */ |
380 |
}; |
381 |
|
382 |
/*****************************************************************************/ |
383 |
|
384 |
#define ETP_PRI_MIN EIO_PRI_MIN |
385 |
#define ETP_PRI_MAX EIO_PRI_MAX |
386 |
|
387 |
struct etp_worker; |
388 |
|
389 |
#define ETP_REQ eio_req |
390 |
#define ETP_DESTROY(req) eio_destroy (req) |
391 |
static int eio_finish (eio_req *req); |
392 |
#define ETP_FINISH(req) eio_finish (req) |
393 |
static void eio_execute (struct etp_worker *self, eio_req *req); |
394 |
#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) |
395 |
|
396 |
/*****************************************************************************/ |
397 |
|
398 |
#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
399 |
|
400 |
/* calculate time difference in ~1/EIO_TICKS of a second */ |
401 |
ecb_inline int |
402 |
tvdiff (struct timeval *tv1, struct timeval *tv2) |
403 |
{ |
404 |
return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
405 |
+ ((tv2->tv_usec - tv1->tv_usec) >> 10); |
406 |
} |
407 |
|
408 |
static unsigned int started, idle, wanted = 4; |
409 |
|
410 |
static void (*want_poll_cb) (void); |
411 |
static void (*done_poll_cb) (void); |
412 |
|
413 |
static unsigned int max_poll_time; /* reslock */ |
414 |
static unsigned int max_poll_reqs; /* reslock */ |
415 |
|
416 |
static unsigned int nreqs; /* reqlock */ |
417 |
static unsigned int nready; /* reqlock */ |
418 |
static unsigned int npending; /* reqlock */ |
419 |
static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ |
420 |
static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ |
421 |
|
422 |
static xmutex_t wrklock; |
423 |
static xmutex_t reslock; |
424 |
static xmutex_t reqlock; |
425 |
static xcond_t reqwait; |
426 |
|
427 |
typedef struct etp_worker |
428 |
{ |
429 |
struct tmpbuf tmpbuf; |
430 |
|
431 |
/* locked by wrklock */ |
432 |
struct etp_worker *prev, *next; |
433 |
|
434 |
xthread_t tid; |
435 |
|
436 |
#ifdef ETP_WORKER_COMMON |
437 |
ETP_WORKER_COMMON |
438 |
#endif |
439 |
} etp_worker; |
440 |
|
441 |
static etp_worker wrk_first; /* NOT etp */ |
442 |
|
443 |
#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) |
444 |
#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) |
445 |
|
446 |
/* worker threads management */ |
447 |
|
448 |
static void |
449 |
etp_worker_clear (etp_worker *wrk) |
450 |
{ |
451 |
} |
452 |
|
453 |
static void ecb_cold |
454 |
etp_worker_free (etp_worker *wrk) |
455 |
{ |
456 |
free (wrk->tmpbuf.ptr); |
457 |
|
458 |
wrk->next->prev = wrk->prev; |
459 |
wrk->prev->next = wrk->next; |
460 |
|
461 |
free (wrk); |
462 |
} |
463 |
|
464 |
static unsigned int |
465 |
etp_nreqs (void) |
466 |
{ |
467 |
int retval; |
468 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
469 |
retval = nreqs; |
470 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
471 |
return retval; |
472 |
} |
473 |
|
474 |
static unsigned int |
475 |
etp_nready (void) |
476 |
{ |
477 |
unsigned int retval; |
478 |
|
479 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
480 |
retval = nready; |
481 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
482 |
|
483 |
return retval; |
484 |
} |
485 |
|
486 |
static unsigned int |
487 |
etp_npending (void) |
488 |
{ |
489 |
unsigned int retval; |
490 |
|
491 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
492 |
retval = npending; |
493 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
494 |
|
495 |
return retval; |
496 |
} |
497 |
|
498 |
static unsigned int |
499 |
etp_nthreads (void) |
500 |
{ |
501 |
unsigned int retval; |
502 |
|
503 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
504 |
retval = started; |
505 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
506 |
|
507 |
return retval; |
508 |
} |
509 |
|
510 |
/* |
511 |
* a somewhat faster data structure might be nice, but |
512 |
* with 8 priorities this actually needs <20 insns |
513 |
* per shift, the most expensive operation. |
514 |
*/ |
515 |
typedef struct { |
516 |
ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */ |
517 |
int size; |
518 |
} etp_reqq; |
519 |
|
520 |
static etp_reqq req_queue; |
521 |
static etp_reqq res_queue; |
522 |
|
523 |
static void ecb_noinline ecb_cold |
524 |
reqq_init (etp_reqq *q) |
525 |
{ |
526 |
int pri; |
527 |
|
528 |
for (pri = 0; pri < ETP_NUM_PRI; ++pri) |
529 |
q->qs[pri] = q->qe[pri] = 0; |
530 |
|
531 |
q->size = 0; |
532 |
} |
533 |
|
534 |
static int ecb_noinline |
535 |
reqq_push (etp_reqq *q, ETP_REQ *req) |
536 |
{ |
537 |
int pri = req->pri; |
538 |
req->next = 0; |
539 |
|
540 |
if (q->qe[pri]) |
541 |
{ |
542 |
q->qe[pri]->next = req; |
543 |
q->qe[pri] = req; |
544 |
} |
545 |
else |
546 |
q->qe[pri] = q->qs[pri] = req; |
547 |
|
548 |
return q->size++; |
549 |
} |
550 |
|
551 |
static ETP_REQ * ecb_noinline |
552 |
reqq_shift (etp_reqq *q) |
553 |
{ |
554 |
int pri; |
555 |
|
556 |
if (!q->size) |
557 |
return 0; |
558 |
|
559 |
--q->size; |
560 |
|
561 |
for (pri = ETP_NUM_PRI; pri--; ) |
562 |
{ |
563 |
eio_req *req = q->qs[pri]; |
564 |
|
565 |
if (req) |
566 |
{ |
567 |
if (!(q->qs[pri] = (eio_req *)req->next)) |
568 |
q->qe[pri] = 0; |
569 |
|
570 |
return req; |
571 |
} |
572 |
} |
573 |
|
574 |
abort (); |
575 |
} |
576 |
|
577 |
static int ecb_cold |
578 |
etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
579 |
{ |
580 |
X_MUTEX_CREATE (wrklock); |
581 |
X_MUTEX_CREATE (reslock); |
582 |
X_MUTEX_CREATE (reqlock); |
583 |
X_COND_CREATE (reqwait); |
584 |
|
585 |
reqq_init (&req_queue); |
586 |
reqq_init (&res_queue); |
587 |
|
588 |
wrk_first.next = |
589 |
wrk_first.prev = &wrk_first; |
590 |
|
591 |
started = 0; |
592 |
idle = 0; |
593 |
nreqs = 0; |
594 |
nready = 0; |
595 |
npending = 0; |
596 |
|
597 |
want_poll_cb = want_poll; |
598 |
done_poll_cb = done_poll; |
599 |
|
600 |
return 0; |
601 |
} |
602 |
|
603 |
X_THREAD_PROC (etp_proc); |
604 |
|
605 |
static void ecb_cold |
606 |
etp_start_thread (void) |
607 |
{ |
608 |
etp_worker *wrk = calloc (1, sizeof (etp_worker)); |
609 |
|
610 |
/*TODO*/ |
611 |
assert (("unable to allocate worker thread data", wrk)); |
612 |
|
613 |
X_LOCK (wrklock); |
614 |
|
615 |
if (xthread_create (&wrk->tid, etp_proc, (void *)wrk)) |
616 |
{ |
617 |
wrk->prev = &wrk_first; |
618 |
wrk->next = wrk_first.next; |
619 |
wrk_first.next->prev = wrk; |
620 |
wrk_first.next = wrk; |
621 |
++started; |
622 |
} |
623 |
else |
624 |
free (wrk); |
625 |
|
626 |
X_UNLOCK (wrklock); |
627 |
} |
628 |
|
629 |
static void |
630 |
etp_maybe_start_thread (void) |
631 |
{ |
632 |
if (ecb_expect_true (etp_nthreads () >= wanted)) |
633 |
return; |
634 |
|
635 |
/* todo: maybe use idle here, but might be less exact */ |
636 |
if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
637 |
return; |
638 |
|
639 |
etp_start_thread (); |
640 |
} |
641 |
|
642 |
static void ecb_cold |
643 |
etp_end_thread (void) |
644 |
{ |
645 |
eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */ |
646 |
|
647 |
req->type = -1; |
648 |
req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
649 |
|
650 |
X_LOCK (reqlock); |
651 |
reqq_push (&req_queue, req); |
652 |
X_COND_SIGNAL (reqwait); |
653 |
X_UNLOCK (reqlock); |
654 |
|
655 |
X_LOCK (wrklock); |
656 |
--started; |
657 |
X_UNLOCK (wrklock); |
658 |
} |
659 |
|
660 |
static int |
661 |
etp_poll (void) |
662 |
{ |
663 |
unsigned int maxreqs; |
664 |
unsigned int maxtime; |
665 |
struct timeval tv_start, tv_now; |
666 |
|
667 |
X_LOCK (reslock); |
668 |
maxreqs = max_poll_reqs; |
669 |
maxtime = max_poll_time; |
670 |
X_UNLOCK (reslock); |
671 |
|
672 |
if (maxtime) |
673 |
gettimeofday (&tv_start, 0); |
674 |
|
675 |
for (;;) |
676 |
{ |
677 |
ETP_REQ *req; |
678 |
|
679 |
etp_maybe_start_thread (); |
680 |
|
681 |
X_LOCK (reslock); |
682 |
req = reqq_shift (&res_queue); |
683 |
|
684 |
if (req) |
685 |
{ |
686 |
--npending; |
687 |
|
688 |
if (!res_queue.size && done_poll_cb) |
689 |
done_poll_cb (); |
690 |
} |
691 |
|
692 |
X_UNLOCK (reslock); |
693 |
|
694 |
if (!req) |
695 |
return 0; |
696 |
|
697 |
X_LOCK (reqlock); |
698 |
--nreqs; |
699 |
X_UNLOCK (reqlock); |
700 |
|
701 |
if (ecb_expect_false (req->type == EIO_GROUP && req->size)) |
702 |
{ |
703 |
req->int1 = 1; /* mark request as delayed */ |
704 |
continue; |
705 |
} |
706 |
else |
707 |
{ |
708 |
int res = ETP_FINISH (req); |
709 |
if (ecb_expect_false (res)) |
710 |
return res; |
711 |
} |
712 |
|
713 |
if (ecb_expect_false (maxreqs && !--maxreqs)) |
714 |
break; |
715 |
|
716 |
if (maxtime) |
717 |
{ |
718 |
gettimeofday (&tv_now, 0); |
719 |
|
720 |
if (tvdiff (&tv_start, &tv_now) >= maxtime) |
721 |
break; |
722 |
} |
723 |
} |
724 |
|
725 |
errno = EAGAIN; |
726 |
return -1; |
727 |
} |
728 |
|
729 |
static void |
730 |
etp_cancel (ETP_REQ *req) |
731 |
{ |
732 |
req->cancelled = 1; |
733 |
|
734 |
eio_grp_cancel (req); |
735 |
} |
736 |
|
737 |
static void |
738 |
etp_submit (ETP_REQ *req) |
739 |
{ |
740 |
req->pri -= ETP_PRI_MIN; |
741 |
|
742 |
if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
743 |
if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
744 |
|
745 |
if (ecb_expect_false (req->type == EIO_GROUP)) |
746 |
{ |
747 |
/* I hope this is worth it :/ */ |
748 |
X_LOCK (reqlock); |
749 |
++nreqs; |
750 |
X_UNLOCK (reqlock); |
751 |
|
752 |
X_LOCK (reslock); |
753 |
|
754 |
++npending; |
755 |
|
756 |
if (!reqq_push (&res_queue, req) && want_poll_cb) |
757 |
want_poll_cb (); |
758 |
|
759 |
X_UNLOCK (reslock); |
760 |
} |
761 |
else |
762 |
{ |
763 |
X_LOCK (reqlock); |
764 |
++nreqs; |
765 |
++nready; |
766 |
reqq_push (&req_queue, req); |
767 |
X_COND_SIGNAL (reqwait); |
768 |
X_UNLOCK (reqlock); |
769 |
|
770 |
etp_maybe_start_thread (); |
771 |
} |
772 |
} |
773 |
|
774 |
static void ecb_cold |
775 |
etp_set_max_poll_time (double nseconds) |
776 |
{ |
777 |
if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
778 |
max_poll_time = nseconds * EIO_TICKS; |
779 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
780 |
} |
781 |
|
782 |
static void ecb_cold |
783 |
etp_set_max_poll_reqs (unsigned int maxreqs) |
784 |
{ |
785 |
if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
786 |
max_poll_reqs = maxreqs; |
787 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
788 |
} |
789 |
|
790 |
static void ecb_cold |
791 |
etp_set_max_idle (unsigned int nthreads) |
792 |
{ |
793 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
794 |
max_idle = nthreads; |
795 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
796 |
} |
797 |
|
798 |
static void ecb_cold |
799 |
etp_set_idle_timeout (unsigned int seconds) |
800 |
{ |
801 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
802 |
idle_timeout = seconds; |
803 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
804 |
} |
805 |
|
806 |
static void ecb_cold |
807 |
etp_set_min_parallel (unsigned int nthreads) |
808 |
{ |
809 |
if (wanted < nthreads) |
810 |
wanted = nthreads; |
811 |
} |
812 |
|
813 |
static void ecb_cold |
814 |
etp_set_max_parallel (unsigned int nthreads) |
815 |
{ |
816 |
if (wanted > nthreads) |
817 |
wanted = nthreads; |
818 |
|
819 |
while (started > wanted) |
820 |
etp_end_thread (); |
821 |
} |
822 |
|
823 |
/*****************************************************************************/ |
824 |
|
825 |
static void |
826 |
grp_try_feed (eio_req *grp) |
827 |
{ |
828 |
while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
829 |
{ |
830 |
grp->flags &= ~EIO_FLAG_GROUPADD; |
831 |
|
832 |
EIO_FEED (grp); |
833 |
|
834 |
/* stop if no progress has been made */ |
835 |
if (!(grp->flags & EIO_FLAG_GROUPADD)) |
836 |
{ |
837 |
grp->feed = 0; |
838 |
break; |
839 |
} |
840 |
} |
841 |
} |
842 |
|
843 |
static int |
844 |
grp_dec (eio_req *grp) |
845 |
{ |
846 |
--grp->size; |
847 |
|
848 |
/* call feeder, if applicable */ |
849 |
grp_try_feed (grp); |
850 |
|
851 |
/* finish, if done */ |
852 |
if (!grp->size && grp->int1) |
853 |
return eio_finish (grp); |
854 |
else |
855 |
return 0; |
856 |
} |
857 |
|
858 |
static void |
859 |
eio_destroy (eio_req *req) |
860 |
{ |
861 |
if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); |
862 |
if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); |
863 |
|
864 |
EIO_DESTROY (req); |
865 |
} |
866 |
|
867 |
static int |
868 |
eio_finish (eio_req *req) |
869 |
{ |
870 |
int res = EIO_FINISH (req); |
871 |
|
872 |
if (req->grp) |
873 |
{ |
874 |
int res2; |
875 |
eio_req *grp = req->grp; |
876 |
|
877 |
/* unlink request */ |
878 |
if (req->grp_next) req->grp_next->grp_prev = req->grp_prev; |
879 |
if (req->grp_prev) req->grp_prev->grp_next = req->grp_next; |
880 |
|
881 |
if (grp->grp_first == req) |
882 |
grp->grp_first = req->grp_next; |
883 |
|
884 |
res2 = grp_dec (grp); |
885 |
|
886 |
if (!res) |
887 |
res = res2; |
888 |
} |
889 |
|
890 |
eio_destroy (req); |
891 |
|
892 |
return res; |
893 |
} |
894 |
|
895 |
void |
896 |
eio_grp_cancel (eio_req *grp) |
897 |
{ |
898 |
for (grp = grp->grp_first; grp; grp = grp->grp_next) |
899 |
eio_cancel (grp); |
900 |
} |
901 |
|
902 |
void |
903 |
eio_cancel (eio_req *req) |
904 |
{ |
905 |
etp_cancel (req); |
906 |
} |
907 |
|
908 |
void |
909 |
eio_submit (eio_req *req) |
910 |
{ |
911 |
etp_submit (req); |
912 |
} |
913 |
|
914 |
unsigned int |
915 |
eio_nreqs (void) |
916 |
{ |
917 |
return etp_nreqs (); |
918 |
} |
919 |
|
920 |
unsigned int |
921 |
eio_nready (void) |
922 |
{ |
923 |
return etp_nready (); |
924 |
} |
925 |
|
926 |
unsigned int |
927 |
eio_npending (void) |
928 |
{ |
929 |
return etp_npending (); |
930 |
} |
931 |
|
932 |
unsigned int ecb_cold |
933 |
eio_nthreads (void) |
934 |
{ |
935 |
return etp_nthreads (); |
936 |
} |
937 |
|
938 |
void ecb_cold |
939 |
eio_set_max_poll_time (double nseconds) |
940 |
{ |
941 |
etp_set_max_poll_time (nseconds); |
942 |
} |
943 |
|
944 |
void ecb_cold |
945 |
eio_set_max_poll_reqs (unsigned int maxreqs) |
946 |
{ |
947 |
etp_set_max_poll_reqs (maxreqs); |
948 |
} |
949 |
|
950 |
void ecb_cold |
951 |
eio_set_max_idle (unsigned int nthreads) |
952 |
{ |
953 |
etp_set_max_idle (nthreads); |
954 |
} |
955 |
|
956 |
void ecb_cold |
957 |
eio_set_idle_timeout (unsigned int seconds) |
958 |
{ |
959 |
etp_set_idle_timeout (seconds); |
960 |
} |
961 |
|
962 |
void ecb_cold |
963 |
eio_set_min_parallel (unsigned int nthreads) |
964 |
{ |
965 |
etp_set_min_parallel (nthreads); |
966 |
} |
967 |
|
968 |
void ecb_cold |
969 |
eio_set_max_parallel (unsigned int nthreads) |
970 |
{ |
971 |
etp_set_max_parallel (nthreads); |
972 |
} |
973 |
|
974 |
int eio_poll (void) |
975 |
{ |
976 |
return etp_poll (); |
977 |
} |
978 |
|
979 |
/*****************************************************************************/ |
980 |
/* work around various missing functions */ |
981 |
|
982 |
#ifndef HAVE_UTIMES |
983 |
|
984 |
# undef utimes |
985 |
# define utimes(path,times) eio__utimes (path, times) |
986 |
|
987 |
static int |
988 |
eio__utimes (const char *filename, const struct timeval times[2]) |
989 |
{ |
990 |
if (times) |
991 |
{ |
992 |
struct utimbuf buf; |
993 |
|
994 |
buf.actime = times[0].tv_sec; |
995 |
buf.modtime = times[1].tv_sec; |
996 |
|
997 |
return utime (filename, &buf); |
998 |
} |
999 |
else |
1000 |
return utime (filename, 0); |
1001 |
} |
1002 |
|
1003 |
#endif |
1004 |
|
1005 |
#ifndef HAVE_FUTIMES |
1006 |
|
1007 |
# undef futimes |
1008 |
# define futimes(fd,times) eio__futimes (fd, times) |
1009 |
|
1010 |
static int |
1011 |
eio__futimes (int fd, const struct timeval tv[2]) |
1012 |
{ |
1013 |
errno = ENOSYS; |
1014 |
return -1; |
1015 |
} |
1016 |
|
1017 |
#endif |
1018 |
|
1019 |
#if !HAVE_FDATASYNC |
1020 |
# undef fdatasync |
1021 |
# define fdatasync(fd) fsync (fd) |
1022 |
#endif |
1023 |
|
1024 |
static int |
1025 |
eio__syncfs (int fd) |
1026 |
{ |
1027 |
int res; |
1028 |
|
1029 |
#if HAVE_SYS_SYNCFS |
1030 |
res = (int)syscall (__NR_syncfs, (int)(fd)); |
1031 |
#else |
1032 |
res = EIO_ENOSYS (); |
1033 |
#endif |
1034 |
|
1035 |
if (res < 0 && errno == ENOSYS && fd >= 0) |
1036 |
sync (); |
1037 |
|
1038 |
return res; |
1039 |
} |
1040 |
|
1041 |
/* sync_file_range always needs emulation */ |
1042 |
static int |
1043 |
eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
1044 |
{ |
1045 |
#if HAVE_SYNC_FILE_RANGE |
1046 |
int res; |
1047 |
|
1048 |
if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE |
1049 |
|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE |
1050 |
|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER) |
1051 |
{ |
1052 |
flags = 0 |
1053 |
| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0) |
1054 |
| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0) |
1055 |
| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); |
1056 |
} |
1057 |
|
1058 |
res = sync_file_range (fd, offset, nbytes, flags); |
1059 |
|
1060 |
if (!res || errno != ENOSYS) |
1061 |
return res; |
1062 |
#endif |
1063 |
|
1064 |
/* even though we could play tricks with the flags, it's better to always |
1065 |
* call fdatasync, as that matches the expectation of its users best */ |
1066 |
return fdatasync (fd); |
1067 |
} |
1068 |
|
1069 |
static int |
1070 |
eio__fallocate (int fd, int mode, off_t offset, size_t len) |
1071 |
{ |
1072 |
#if HAVE_LINUX_FALLOCATE |
1073 |
return fallocate (fd, mode, offset, len); |
1074 |
#else |
1075 |
return EIO_ENOSYS (); |
1076 |
#endif |
1077 |
} |
1078 |
|
1079 |
#if !HAVE_READAHEAD |
1080 |
# undef readahead |
1081 |
# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
1082 |
|
1083 |
static eio_ssize_t |
1084 |
eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
1085 |
{ |
1086 |
size_t todo = count; |
1087 |
dBUF; |
1088 |
|
1089 |
while (todo > 0) |
1090 |
{ |
1091 |
size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; |
1092 |
|
1093 |
pread (fd, eio_buf, len, offset); |
1094 |
offset += len; |
1095 |
todo -= len; |
1096 |
} |
1097 |
|
1098 |
FUBd; |
1099 |
|
1100 |
/* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */ |
1101 |
/* but not for e.g. EIO or eof, so we also never fail */ |
1102 |
return 0; |
1103 |
} |
1104 |
|
1105 |
#endif |
1106 |
|
1107 |
/* sendfile always needs emulation */ |
1108 |
static eio_ssize_t |
1109 |
eio__sendfile (int ofd, int ifd, off_t offset, size_t count) |
1110 |
{ |
1111 |
eio_ssize_t written = 0; |
1112 |
eio_ssize_t res; |
1113 |
|
1114 |
if (!count) |
1115 |
return 0; |
1116 |
|
1117 |
for (;;) |
1118 |
{ |
1119 |
#ifdef __APPLE__ |
1120 |
# undef HAVE_SENDFILE /* broken, as everything on os x */ |
1121 |
#endif |
1122 |
#if HAVE_SENDFILE |
1123 |
# if __linux |
1124 |
off_t soffset = offset; |
1125 |
res = sendfile (ofd, ifd, &soffset, count); |
1126 |
|
1127 |
# elif __FreeBSD__ |
1128 |
/* |
1129 |
* Of course, the freebsd sendfile is a dire hack with no thoughts |
1130 |
* wasted on making it similar to other I/O functions. |
1131 |
*/ |
1132 |
off_t sbytes; |
1133 |
res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
1134 |
|
1135 |
#if 0 /* according to the manpage, this is correct, but broken behaviour */ |
1136 |
/* freebsd' sendfile will return 0 on success */ |
1137 |
/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
1138 |
/* not on e.g. EIO or EPIPE - sounds broken */ |
1139 |
if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
1140 |
res = sbytes; |
1141 |
#endif |
1142 |
|
1143 |
/* according to source inspection, this is correct, and useful behaviour */ |
1144 |
if (sbytes) |
1145 |
res = sbytes; |
1146 |
|
1147 |
# elif defined __APPLE__ |
1148 |
off_t sbytes = count; |
1149 |
res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
1150 |
|
1151 |
/* according to the manpage, sbytes is always valid */ |
1152 |
if (sbytes) |
1153 |
res = sbytes; |
1154 |
|
1155 |
# elif __hpux |
1156 |
res = sendfile (ofd, ifd, offset, count, 0, 0); |
1157 |
|
1158 |
# elif __solaris |
1159 |
struct sendfilevec vec; |
1160 |
size_t sbytes; |
1161 |
|
1162 |
vec.sfv_fd = ifd; |
1163 |
vec.sfv_flag = 0; |
1164 |
vec.sfv_off = offset; |
1165 |
vec.sfv_len = count; |
1166 |
|
1167 |
res = sendfilev (ofd, &vec, 1, &sbytes); |
1168 |
|
1169 |
if (res < 0 && sbytes) |
1170 |
res = sbytes; |
1171 |
|
1172 |
# endif |
1173 |
|
1174 |
#elif defined (_WIN32) && 0 |
1175 |
/* does not work, just for documentation of what would need to be done */ |
1176 |
/* actually, cannot be done like this, as TransmitFile changes the file offset, */ |
1177 |
/* libeio guarantees that the file offset does not change, and windows */ |
1178 |
/* has no way to get an independent handle to the same file description */ |
1179 |
HANDLE h = TO_SOCKET (ifd); |
1180 |
SetFilePointer (h, offset, 0, FILE_BEGIN); |
1181 |
res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
1182 |
|
1183 |
#else |
1184 |
res = EIO_ENOSYS (); |
1185 |
#endif |
1186 |
|
1187 |
/* we assume sendfile can copy at least 128mb in one go */ |
1188 |
if (res <= 128 * 1024 * 1024) |
1189 |
{ |
1190 |
if (res > 0) |
1191 |
written += res; |
1192 |
|
1193 |
if (written) |
1194 |
return written; |
1195 |
|
1196 |
break; |
1197 |
} |
1198 |
else |
1199 |
{ |
1200 |
/* if we requested more, then probably the kernel was lazy */ |
1201 |
written += res; |
1202 |
offset += res; |
1203 |
count -= res; |
1204 |
|
1205 |
if (!count) |
1206 |
return written; |
1207 |
} |
1208 |
} |
1209 |
|
1210 |
if (res < 0 |
1211 |
&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
1212 |
/* BSDs */ |
1213 |
#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
1214 |
|| errno == ENOTSUP |
1215 |
#endif |
1216 |
#ifdef EOPNOTSUPP /* windows */ |
1217 |
|| errno == EOPNOTSUPP /* BSDs */ |
1218 |
#endif |
1219 |
#if __solaris |
1220 |
|| errno == EAFNOSUPPORT || errno == EPROTOTYPE |
1221 |
#endif |
1222 |
) |
1223 |
) |
1224 |
{ |
1225 |
/* emulate sendfile. this is a major pain in the ass */ |
1226 |
dBUF; |
1227 |
|
1228 |
res = 0; |
1229 |
|
1230 |
while (count) |
1231 |
{ |
1232 |
eio_ssize_t cnt; |
1233 |
|
1234 |
cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); |
1235 |
|
1236 |
if (cnt <= 0) |
1237 |
{ |
1238 |
if (cnt && !res) res = -1; |
1239 |
break; |
1240 |
} |
1241 |
|
1242 |
cnt = write (ofd, eio_buf, cnt); |
1243 |
|
1244 |
if (cnt <= 0) |
1245 |
{ |
1246 |
if (cnt && !res) res = -1; |
1247 |
break; |
1248 |
} |
1249 |
|
1250 |
offset += cnt; |
1251 |
res += cnt; |
1252 |
count -= cnt; |
1253 |
} |
1254 |
|
1255 |
FUBd; |
1256 |
} |
1257 |
|
1258 |
return res; |
1259 |
} |
1260 |
|
1261 |
#ifdef PAGESIZE |
1262 |
# define eio_pagesize() PAGESIZE |
1263 |
#else |
1264 |
static intptr_t |
1265 |
eio_pagesize (void) |
1266 |
{ |
1267 |
static intptr_t page; |
1268 |
|
1269 |
if (!page) |
1270 |
page = sysconf (_SC_PAGESIZE); |
1271 |
|
1272 |
return page; |
1273 |
} |
1274 |
#endif |
1275 |
|
1276 |
static void |
1277 |
eio_page_align (void **addr, size_t *length) |
1278 |
{ |
1279 |
intptr_t mask = eio_pagesize () - 1; |
1280 |
|
1281 |
/* round down addr */ |
1282 |
intptr_t adj = mask & (intptr_t)*addr; |
1283 |
|
1284 |
*addr = (void *)((intptr_t)*addr - adj); |
1285 |
*length += adj; |
1286 |
|
1287 |
/* round up length */ |
1288 |
*length = (*length + mask) & ~mask; |
1289 |
} |
1290 |
|
1291 |
#if !_POSIX_MEMLOCK |
1292 |
# define eio__mlockall(a) EIO_ENOSYS () |
1293 |
#else |
1294 |
|
1295 |
static int |
1296 |
eio__mlockall (int flags) |
1297 |
{ |
1298 |
#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 |
1299 |
extern int mallopt (int, int); |
1300 |
mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ |
1301 |
#endif |
1302 |
|
1303 |
if (EIO_MCL_CURRENT != MCL_CURRENT |
1304 |
|| EIO_MCL_FUTURE != MCL_FUTURE) |
1305 |
{ |
1306 |
flags = 0 |
1307 |
| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0) |
1308 |
| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0); |
1309 |
} |
1310 |
|
1311 |
return mlockall (flags); |
1312 |
} |
1313 |
#endif |
1314 |
|
1315 |
#if !_POSIX_MEMLOCK_RANGE |
1316 |
# define eio__mlock(a,b) EIO_ENOSYS () |
1317 |
#else |
1318 |
|
1319 |
static int |
1320 |
eio__mlock (void *addr, size_t length) |
1321 |
{ |
1322 |
eio_page_align (&addr, &length); |
1323 |
|
1324 |
return mlock (addr, length); |
1325 |
} |
1326 |
|
1327 |
#endif |
1328 |
|
1329 |
#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1330 |
# define eio__msync(a,b,c) EIO_ENOSYS () |
1331 |
#else |
1332 |
|
1333 |
static int |
1334 |
eio__msync (void *mem, size_t len, int flags) |
1335 |
{ |
1336 |
eio_page_align (&mem, &len); |
1337 |
|
1338 |
if (EIO_MS_ASYNC != MS_SYNC |
1339 |
|| EIO_MS_INVALIDATE != MS_INVALIDATE |
1340 |
|| EIO_MS_SYNC != MS_SYNC) |
1341 |
{ |
1342 |
flags = 0 |
1343 |
| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
1344 |
| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
1345 |
| (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
1346 |
} |
1347 |
|
1348 |
return msync (mem, len, flags); |
1349 |
} |
1350 |
|
1351 |
#endif |
1352 |
|
1353 |
static int |
1354 |
eio__mtouch (eio_req *req) |
1355 |
{ |
1356 |
void *mem = req->ptr2; |
1357 |
size_t len = req->size; |
1358 |
int flags = req->int1; |
1359 |
|
1360 |
eio_page_align (&mem, &len); |
1361 |
|
1362 |
{ |
1363 |
intptr_t addr = (intptr_t)mem; |
1364 |
intptr_t end = addr + len; |
1365 |
intptr_t page = eio_pagesize (); |
1366 |
|
1367 |
if (addr < end) |
1368 |
if (flags & EIO_MT_MODIFY) /* modify */ |
1369 |
do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
1370 |
else |
1371 |
do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
1372 |
} |
1373 |
|
1374 |
return 0; |
1375 |
} |
1376 |
|
1377 |
/*****************************************************************************/ |
1378 |
/* requests implemented outside eio_execute, because they are so large */ |
1379 |
|
1380 |
static void |
1381 |
eio__lseek (eio_req *req) |
1382 |
{ |
1383 |
/* this usually gets optimised away completely, or your compiler sucks, */ |
1384 |
/* or the whence constants really are not 0, 1, 2 */ |
1385 |
int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET |
1386 |
: req->int2 == EIO_SEEK_CUR ? SEEK_CUR |
1387 |
: req->int2 == EIO_SEEK_END ? SEEK_END |
1388 |
: req->int2; |
1389 |
|
1390 |
req->offs = lseek (req->int1, req->offs, whence); |
1391 |
req->result = req->offs == (off_t)-1 ? -1 : 0; |
1392 |
} |
1393 |
|
1394 |
/* result will always end up in tmpbuf, there is always space for adding a 0-byte */ |
1395 |
static int |
1396 |
eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path) |
1397 |
{ |
1398 |
const char *rel = path; |
1399 |
char *res; |
1400 |
char *tmp1, *tmp2; |
1401 |
#if SYMLOOP_MAX > 32 |
1402 |
int symlinks = SYMLOOP_MAX; |
1403 |
#else |
1404 |
int symlinks = 32; |
1405 |
#endif |
1406 |
|
1407 |
errno = EINVAL; |
1408 |
if (!rel) |
1409 |
return -1; |
1410 |
|
1411 |
errno = ENOENT; |
1412 |
if (!*rel) |
1413 |
return -1; |
1414 |
|
1415 |
res = tmpbuf_get (tmpbuf, PATH_MAX * 3); |
1416 |
tmp1 = res + PATH_MAX; |
1417 |
tmp2 = tmp1 + PATH_MAX; |
1418 |
|
1419 |
#if 0 /* disabled, the musl way to do things is just too racy */ |
1420 |
#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) |
1421 |
/* on linux we may be able to ask the kernel */ |
1422 |
{ |
1423 |
int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME); |
1424 |
|
1425 |
if (fd >= 0) |
1426 |
{ |
1427 |
sprintf (tmp1, "/proc/self/fd/%d", fd); |
1428 |
req->result = readlink (tmp1, res, PATH_MAX); |
1429 |
/* here we should probably stat the open file and the disk file, to make sure they still match */ |
1430 |
close (fd); |
1431 |
|
1432 |
if (req->result > 0) |
1433 |
goto done; |
1434 |
} |
1435 |
else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) |
1436 |
return -1; |
1437 |
} |
1438 |
#endif |
1439 |
#endif |
1440 |
|
1441 |
if (*rel != '/') |
1442 |
{ |
1443 |
int len; |
1444 |
|
1445 |
errno = ENOENT; |
1446 |
if (wd == EIO_INVALID_WD) |
1447 |
return -1; |
1448 |
|
1449 |
if (wd == EIO_CWD) |
1450 |
{ |
1451 |
if (!getcwd (res, PATH_MAX)) |
1452 |
return -1; |
1453 |
|
1454 |
len = strlen (res); |
1455 |
} |
1456 |
else |
1457 |
memcpy (res, wd->str, len = wd->len); |
1458 |
|
1459 |
if (res [1]) /* only use if not / */ |
1460 |
res += len; |
1461 |
} |
1462 |
|
1463 |
while (*rel) |
1464 |
{ |
1465 |
eio_ssize_t len, linklen; |
1466 |
const char *beg = rel; |
1467 |
|
1468 |
while (*rel && *rel != '/') |
1469 |
++rel; |
1470 |
|
1471 |
len = rel - beg; |
1472 |
|
1473 |
if (!len) /* skip slashes */ |
1474 |
{ |
1475 |
++rel; |
1476 |
continue; |
1477 |
} |
1478 |
|
1479 |
if (beg [0] == '.') |
1480 |
{ |
1481 |
if (len == 1) |
1482 |
continue; /* . - nop */ |
1483 |
|
1484 |
if (beg [1] == '.' && len == 2) |
1485 |
{ |
1486 |
/* .. - back up one component, if possible */ |
1487 |
|
1488 |
while (res != tmpbuf->ptr) |
1489 |
if (*--res == '/') |
1490 |
break; |
1491 |
|
1492 |
continue; |
1493 |
} |
1494 |
} |
1495 |
|
1496 |
errno = ENAMETOOLONG; |
1497 |
if (res + 1 + len + 1 >= tmp1) |
1498 |
return -1; |
1499 |
|
1500 |
/* copy one component */ |
1501 |
*res = '/'; |
1502 |
memcpy (res + 1, beg, len); |
1503 |
|
1504 |
/* zero-terminate, for readlink */ |
1505 |
res [len + 1] = 0; |
1506 |
|
1507 |
/* now check if it's a symlink */ |
1508 |
linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX); |
1509 |
|
1510 |
if (linklen < 0) |
1511 |
{ |
1512 |
if (errno != EINVAL) |
1513 |
return -1; |
1514 |
|
1515 |
/* it's a normal directory. hopefully */ |
1516 |
res += len + 1; |
1517 |
} |
1518 |
else |
1519 |
{ |
1520 |
/* yay, it was a symlink - build new path in tmp2 */ |
1521 |
int rellen = strlen (rel); |
1522 |
|
1523 |
errno = ENAMETOOLONG; |
1524 |
if (linklen + 1 + rellen >= PATH_MAX) |
1525 |
return -1; |
1526 |
|
1527 |
errno = ELOOP; |
1528 |
if (!--symlinks) |
1529 |
return -1; |
1530 |
|
1531 |
if (*tmp1 == '/') |
1532 |
res = tmpbuf->ptr; /* symlink resolves to an absolute path */ |
1533 |
|
1534 |
/* we need to be careful, as rel might point into tmp2 already */ |
1535 |
memmove (tmp2 + linklen + 1, rel, rellen + 1); |
1536 |
tmp2 [linklen] = '/'; |
1537 |
memcpy (tmp2, tmp1, linklen); |
1538 |
|
1539 |
rel = tmp2; |
1540 |
} |
1541 |
} |
1542 |
|
1543 |
/* special case for the lone root path */ |
1544 |
if (res == tmpbuf->ptr) |
1545 |
*res++ = '/'; |
1546 |
|
1547 |
return res - (char *)tmpbuf->ptr; |
1548 |
} |
1549 |
|
1550 |
static signed char |
1551 |
eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
1552 |
{ |
1553 |
return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
1554 |
: a->inode < b->inode ? -1 |
1555 |
: a->inode > b->inode ? 1 |
1556 |
: 0; |
1557 |
} |
1558 |
|
1559 |
#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
1560 |
|
1561 |
#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
1562 |
#define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
1563 |
|
1564 |
static void |
1565 |
eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits) |
1566 |
{ |
1567 |
unsigned char bits [9 + sizeof (eio_ino_t) * 8]; |
1568 |
unsigned char *bit = bits; |
1569 |
|
1570 |
assert (CHAR_BIT == 8); |
1571 |
assert (sizeof (eio_dirent) * 8 < 256); |
1572 |
assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ |
1573 |
assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */ |
1574 |
|
1575 |
if (size <= EIO_SORT_FAST) |
1576 |
return; |
1577 |
|
1578 |
/* first prepare an array of bits to test in our radix sort */ |
1579 |
/* try to take endianness into account, as well as differences in eio_ino_t sizes */ |
1580 |
/* inode_bits must contain all inodes ORed together */ |
1581 |
/* which is used to skip bits that are 0 everywhere, which is very common */ |
1582 |
{ |
1583 |
eio_ino_t endianness; |
1584 |
int i, j; |
1585 |
|
1586 |
/* we store the byte offset of byte n into byte n of "endianness" */ |
1587 |
for (i = 0; i < sizeof (eio_ino_t); ++i) |
1588 |
((unsigned char *)&endianness)[i] = i; |
1589 |
|
1590 |
*bit++ = 0; |
1591 |
|
1592 |
for (i = 0; i < sizeof (eio_ino_t); ++i) |
1593 |
{ |
1594 |
/* shifting off the byte offsets out of "endianness" */ |
1595 |
int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
1596 |
endianness >>= 8; |
1597 |
|
1598 |
for (j = 0; j < 8; ++j) |
1599 |
if (inode_bits & (((eio_ino_t)1) << (i * 8 + j))) |
1600 |
*bit++ = offs + j; |
1601 |
} |
1602 |
|
1603 |
for (j = 0; j < 8; ++j) |
1604 |
if (score_bits & (1 << j)) |
1605 |
*bit++ = offsetof (eio_dirent, score) * 8 + j; |
1606 |
} |
1607 |
|
1608 |
/* now actually do the sorting (a variant of MSD radix sort) */ |
1609 |
{ |
1610 |
eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base; |
1611 |
eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end; |
1612 |
unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8]; |
1613 |
int stk_idx = 0; |
1614 |
|
1615 |
base_stk [stk_idx] = dents; |
1616 |
end_stk [stk_idx] = dents + size; |
1617 |
bit_stk [stk_idx] = bit - 1; |
1618 |
|
1619 |
do |
1620 |
{ |
1621 |
base = base_stk [stk_idx]; |
1622 |
end = end_stk [stk_idx]; |
1623 |
bit = bit_stk [stk_idx]; |
1624 |
|
1625 |
for (;;) |
1626 |
{ |
1627 |
unsigned char O = *bit >> 3; |
1628 |
unsigned char M = 1 << (*bit & 7); |
1629 |
|
1630 |
eio_dirent *a = base; |
1631 |
eio_dirent *b = end; |
1632 |
|
1633 |
if (b - a < EIO_SORT_CUTOFF) |
1634 |
break; |
1635 |
|
1636 |
/* now bit-partition the array on the bit */ |
1637 |
/* this ugly asymmetric loop seems to perform much better than typical */ |
1638 |
/* partition algos found in the literature */ |
1639 |
do |
1640 |
if (!(((unsigned char *)a)[O] & M)) |
1641 |
++a; |
1642 |
else if (!(((unsigned char *)--b)[O] & M)) |
1643 |
{ |
1644 |
eio_dirent tmp = *a; *a = *b; *b = tmp; |
1645 |
++a; |
1646 |
} |
1647 |
while (b > a); |
1648 |
|
1649 |
/* next bit, or stop, if no bits left in this path */ |
1650 |
if (!*--bit) |
1651 |
break; |
1652 |
|
1653 |
base_stk [stk_idx] = a; |
1654 |
end_stk [stk_idx] = end; |
1655 |
bit_stk [stk_idx] = bit; |
1656 |
++stk_idx; |
1657 |
|
1658 |
end = a; |
1659 |
} |
1660 |
} |
1661 |
while (stk_idx--); |
1662 |
} |
1663 |
} |
1664 |
|
1665 |
static void |
1666 |
eio_dent_insertion_sort (eio_dirent *dents, int size) |
1667 |
{ |
1668 |
/* first move the smallest element to the front, to act as a sentinel */ |
1669 |
{ |
1670 |
int i; |
1671 |
eio_dirent *min = dents; |
1672 |
|
1673 |
/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
1674 |
for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
1675 |
if (EIO_DENT_CMP (dents [i], <, *min)) |
1676 |
min = &dents [i]; |
1677 |
|
1678 |
/* swap elements 0 and j (minimum) */ |
1679 |
{ |
1680 |
eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
1681 |
} |
1682 |
} |
1683 |
|
1684 |
/* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
1685 |
{ |
1686 |
eio_dirent *i, *j; |
1687 |
|
1688 |
for (i = dents + 1; i < dents + size; ++i) |
1689 |
{ |
1690 |
eio_dirent value = *i; |
1691 |
|
1692 |
for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
1693 |
j [1] = j [0]; |
1694 |
|
1695 |
j [1] = value; |
1696 |
} |
1697 |
} |
1698 |
} |
1699 |
|
1700 |
static void |
1701 |
eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits) |
1702 |
{ |
1703 |
if (size <= 1) |
1704 |
return; /* our insertion sort relies on size > 0 */ |
1705 |
|
1706 |
/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
1707 |
/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
1708 |
eio_dent_radix_sort (dents, size, score_bits, inode_bits); |
1709 |
|
1710 |
/* use an insertion sort at the end, or for small arrays, */ |
1711 |
/* as insertion sort is more efficient for small partitions */ |
1712 |
eio_dent_insertion_sort (dents, size); |
1713 |
} |
1714 |
|
1715 |
/* read a full directory */ |
1716 |
static void |
1717 |
eio__scandir (eio_req *req, etp_worker *self) |
1718 |
{ |
1719 |
char *name, *names; |
1720 |
int namesalloc = 4096 - sizeof (void *) * 4; |
1721 |
int namesoffs = 0; |
1722 |
int flags = req->int1; |
1723 |
eio_dirent *dents = 0; |
1724 |
int dentalloc = 128; |
1725 |
int dentoffs = 0; |
1726 |
eio_ino_t inode_bits = 0; |
1727 |
#ifdef _WIN32 |
1728 |
HANDLE dirp; |
1729 |
WIN32_FIND_DATA entp; |
1730 |
#else |
1731 |
DIR *dirp; |
1732 |
EIO_STRUCT_DIRENT *entp; |
1733 |
#endif |
1734 |
|
1735 |
req->result = -1; |
1736 |
|
1737 |
if (!(flags & EIO_READDIR_DENTS)) |
1738 |
flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1739 |
|
1740 |
#ifdef _WIN32 |
1741 |
{ |
1742 |
int len = strlen ((const char *)req->ptr1); |
1743 |
char *path = malloc (MAX_PATH); |
1744 |
const char *fmt; |
1745 |
const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1); |
1746 |
|
1747 |
if (!len) |
1748 |
fmt = "./*"; |
1749 |
else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\') |
1750 |
fmt = "%s*"; |
1751 |
else |
1752 |
fmt = "%s/*"; |
1753 |
|
1754 |
_snprintf (path, MAX_PATH, fmt, reqpath); |
1755 |
dirp = FindFirstFile (path, &entp); |
1756 |
free (path); |
1757 |
|
1758 |
if (dirp == INVALID_HANDLE_VALUE) |
1759 |
{ |
1760 |
/* should steal _dosmaperr */ |
1761 |
switch (GetLastError ()) |
1762 |
{ |
1763 |
case ERROR_FILE_NOT_FOUND: |
1764 |
req->result = 0; |
1765 |
break; |
1766 |
|
1767 |
case ERROR_INVALID_NAME: |
1768 |
case ERROR_PATH_NOT_FOUND: |
1769 |
case ERROR_NO_MORE_FILES: |
1770 |
errno = ENOENT; |
1771 |
break; |
1772 |
|
1773 |
case ERROR_NOT_ENOUGH_MEMORY: |
1774 |
errno = ENOMEM; |
1775 |
break; |
1776 |
|
1777 |
default: |
1778 |
errno = EINVAL; |
1779 |
break; |
1780 |
} |
1781 |
|
1782 |
return; |
1783 |
} |
1784 |
} |
1785 |
#else |
1786 |
#if HAVE_AT |
1787 |
if (req->wd) |
1788 |
{ |
1789 |
int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY); |
1790 |
|
1791 |
if (fd < 0) |
1792 |
return; |
1793 |
|
1794 |
dirp = fdopendir (fd); |
1795 |
|
1796 |
if (!dirp) |
1797 |
close (fd); |
1798 |
} |
1799 |
else |
1800 |
dirp = opendir (req->ptr1); |
1801 |
#else |
1802 |
dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1)); |
1803 |
#endif |
1804 |
|
1805 |
if (!dirp) |
1806 |
return; |
1807 |
#endif |
1808 |
|
1809 |
if (req->flags & EIO_FLAG_PTR1_FREE) |
1810 |
free (req->ptr1); |
1811 |
|
1812 |
req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1813 |
req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1814 |
req->ptr2 = names = malloc (namesalloc); |
1815 |
|
1816 |
if (!names || (flags && !dents)) |
1817 |
return; |
1818 |
|
1819 |
for (;;) |
1820 |
{ |
1821 |
int done; |
1822 |
|
1823 |
#ifdef _WIN32 |
1824 |
done = !dirp; |
1825 |
#else |
1826 |
errno = 0; |
1827 |
entp = readdir (dirp); |
1828 |
done = !entp; |
1829 |
#endif |
1830 |
|
1831 |
if (done) |
1832 |
{ |
1833 |
#ifndef _WIN32 |
1834 |
int old_errno = errno; |
1835 |
closedir (dirp); |
1836 |
errno = old_errno; |
1837 |
|
1838 |
if (errno) |
1839 |
break; |
1840 |
#endif |
1841 |
|
1842 |
/* sort etc. */ |
1843 |
req->int1 = flags; |
1844 |
req->result = dentoffs; |
1845 |
|
1846 |
if (flags & EIO_READDIR_STAT_ORDER) |
1847 |
eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); |
1848 |
else if (flags & EIO_READDIR_DIRS_FIRST) |
1849 |
if (flags & EIO_READDIR_FOUND_UNKNOWN) |
1850 |
eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
1851 |
else |
1852 |
{ |
1853 |
/* in this case, all is known, and we just put dirs first and sort them */ |
1854 |
eio_dirent *oth = dents + dentoffs; |
1855 |
eio_dirent *dir = dents; |
1856 |
|
1857 |
/* now partition dirs to the front, and non-dirs to the back */ |
1858 |
/* by walking from both sides and swapping if necessary */ |
1859 |
while (oth > dir) |
1860 |
{ |
1861 |
if (dir->type == EIO_DT_DIR) |
1862 |
++dir; |
1863 |
else if ((--oth)->type == EIO_DT_DIR) |
1864 |
{ |
1865 |
eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
1866 |
|
1867 |
++dir; |
1868 |
} |
1869 |
} |
1870 |
|
1871 |
/* now sort the dirs only (dirs all have the same score) */ |
1872 |
eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1873 |
} |
1874 |
|
1875 |
break; |
1876 |
} |
1877 |
|
1878 |
/* now add the entry to our list(s) */ |
1879 |
name = D_NAME (entp); |
1880 |
|
1881 |
/* skip . and .. entries */ |
1882 |
if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1883 |
{ |
1884 |
int len = D_NAMLEN (entp) + 1; |
1885 |
|
1886 |
while (ecb_expect_false (namesoffs + len > namesalloc)) |
1887 |
{ |
1888 |
namesalloc *= 2; |
1889 |
req->ptr2 = names = realloc (names, namesalloc); |
1890 |
|
1891 |
if (!names) |
1892 |
break; |
1893 |
} |
1894 |
|
1895 |
memcpy (names + namesoffs, name, len); |
1896 |
|
1897 |
if (dents) |
1898 |
{ |
1899 |
struct eio_dirent *ent; |
1900 |
|
1901 |
if (ecb_expect_false (dentoffs == dentalloc)) |
1902 |
{ |
1903 |
dentalloc *= 2; |
1904 |
req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1905 |
|
1906 |
if (!dents) |
1907 |
break; |
1908 |
} |
1909 |
|
1910 |
ent = dents + dentoffs; |
1911 |
|
1912 |
ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
1913 |
ent->namelen = len - 1; |
1914 |
ent->inode = D_INO (entp); |
1915 |
|
1916 |
inode_bits |= ent->inode; |
1917 |
|
1918 |
switch (D_TYPE (entp)) |
1919 |
{ |
1920 |
default: |
1921 |
ent->type = EIO_DT_UNKNOWN; |
1922 |
flags |= EIO_READDIR_FOUND_UNKNOWN; |
1923 |
break; |
1924 |
|
1925 |
#ifdef DT_FIFO |
1926 |
case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
1927 |
#endif |
1928 |
#ifdef DT_CHR |
1929 |
case DT_CHR: ent->type = EIO_DT_CHR; break; |
1930 |
#endif |
1931 |
#ifdef DT_MPC |
1932 |
case DT_MPC: ent->type = EIO_DT_MPC; break; |
1933 |
#endif |
1934 |
#ifdef DT_DIR |
1935 |
case DT_DIR: ent->type = EIO_DT_DIR; break; |
1936 |
#endif |
1937 |
#ifdef DT_NAM |
1938 |
case DT_NAM: ent->type = EIO_DT_NAM; break; |
1939 |
#endif |
1940 |
#ifdef DT_BLK |
1941 |
case DT_BLK: ent->type = EIO_DT_BLK; break; |
1942 |
#endif |
1943 |
#ifdef DT_MPB |
1944 |
case DT_MPB: ent->type = EIO_DT_MPB; break; |
1945 |
#endif |
1946 |
#ifdef DT_REG |
1947 |
case DT_REG: ent->type = EIO_DT_REG; break; |
1948 |
#endif |
1949 |
#ifdef DT_NWK |
1950 |
case DT_NWK: ent->type = EIO_DT_NWK; break; |
1951 |
#endif |
1952 |
#ifdef DT_CMP |
1953 |
case DT_CMP: ent->type = EIO_DT_CMP; break; |
1954 |
#endif |
1955 |
#ifdef DT_LNK |
1956 |
case DT_LNK: ent->type = EIO_DT_LNK; break; |
1957 |
#endif |
1958 |
#ifdef DT_SOCK |
1959 |
case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
1960 |
#endif |
1961 |
#ifdef DT_DOOR |
1962 |
case DT_DOOR: ent->type = EIO_DT_DOOR; break; |
1963 |
#endif |
1964 |
#ifdef DT_WHT |
1965 |
case DT_WHT: ent->type = EIO_DT_WHT; break; |
1966 |
#endif |
1967 |
} |
1968 |
|
1969 |
ent->score = 7; |
1970 |
|
1971 |
if (flags & EIO_READDIR_DIRS_FIRST) |
1972 |
{ |
1973 |
if (ent->type == EIO_DT_UNKNOWN) |
1974 |
{ |
1975 |
if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1976 |
ent->score = 1; |
1977 |
else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */ |
1978 |
ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
1979 |
} |
1980 |
else if (ent->type == EIO_DT_DIR) |
1981 |
ent->score = 0; |
1982 |
} |
1983 |
} |
1984 |
|
1985 |
namesoffs += len; |
1986 |
++dentoffs; |
1987 |
} |
1988 |
|
1989 |
if (EIO_CANCELLED (req)) |
1990 |
{ |
1991 |
errno = ECANCELED; |
1992 |
break; |
1993 |
} |
1994 |
|
1995 |
#ifdef _WIN32 |
1996 |
if (!FindNextFile (dirp, &entp)) |
1997 |
{ |
1998 |
FindClose (dirp); |
1999 |
dirp = 0; |
2000 |
} |
2001 |
#endif |
2002 |
} |
2003 |
} |
2004 |
|
2005 |
/*****************************************************************************/ |
2006 |
/* working directory stuff */ |
2007 |
/* various deficiencies in the posix 2008 api force us to */ |
2008 |
/* keep the absolute path in string form at all times */ |
2009 |
/* fuck yeah. */ |
2010 |
|
2011 |
#if !HAVE_AT |
2012 |
|
2013 |
/* a bit like realpath, but usually faster because it doesn'T have to return */ |
2014 |
/* an absolute or canonical path */ |
2015 |
static const char * |
2016 |
wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path) |
2017 |
{ |
2018 |
if (!wd || *path == '/') |
2019 |
return path; |
2020 |
|
2021 |
if (path [0] == '.' && !path [1]) |
2022 |
return wd->str; |
2023 |
|
2024 |
{ |
2025 |
int l1 = wd->len; |
2026 |
int l2 = strlen (path); |
2027 |
|
2028 |
char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2); |
2029 |
|
2030 |
memcpy (res, wd->str, l1); |
2031 |
res [l1] = '/'; |
2032 |
memcpy (res + l1 + 1, path, l2 + 1); |
2033 |
|
2034 |
return res; |
2035 |
} |
2036 |
} |
2037 |
|
2038 |
#endif |
2039 |
|
2040 |
static eio_wd |
2041 |
eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path) |
2042 |
{ |
2043 |
int fd; |
2044 |
eio_wd res; |
2045 |
int len = eio__realpath (tmpbuf, wd, path); |
2046 |
|
2047 |
if (len < 0) |
2048 |
return EIO_INVALID_WD; |
2049 |
|
2050 |
#if HAVE_AT |
2051 |
fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY); |
2052 |
|
2053 |
if (fd < 0) |
2054 |
return EIO_INVALID_WD; |
2055 |
#endif |
2056 |
|
2057 |
res = malloc (sizeof (*res) + len); /* one extra 0-byte */ |
2058 |
|
2059 |
#if HAVE_AT |
2060 |
res->fd = fd; |
2061 |
#endif |
2062 |
|
2063 |
res->len = len; |
2064 |
memcpy (res->str, tmpbuf->ptr, len); |
2065 |
res->str [len] = 0; |
2066 |
|
2067 |
return res; |
2068 |
} |
2069 |
|
2070 |
eio_wd |
2071 |
eio_wd_open_sync (eio_wd wd, const char *path) |
2072 |
{ |
2073 |
struct tmpbuf tmpbuf = { 0 }; |
2074 |
wd = eio__wd_open_sync (&tmpbuf, wd, path); |
2075 |
free (tmpbuf.ptr); |
2076 |
|
2077 |
return wd; |
2078 |
} |
2079 |
|
2080 |
void |
2081 |
eio_wd_close_sync (eio_wd wd) |
2082 |
{ |
2083 |
if (wd != EIO_INVALID_WD && wd != EIO_CWD) |
2084 |
{ |
2085 |
#if HAVE_AT |
2086 |
close (wd->fd); |
2087 |
#endif |
2088 |
free (wd); |
2089 |
} |
2090 |
} |
2091 |
|
2092 |
#if HAVE_AT |
2093 |
|
2094 |
/* they forgot these */ |
2095 |
|
2096 |
static int |
2097 |
eio__truncateat (int dirfd, const char *path, off_t length) |
2098 |
{ |
2099 |
int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC); |
2100 |
int res; |
2101 |
|
2102 |
if (fd < 0) |
2103 |
return fd; |
2104 |
|
2105 |
res = ftruncate (fd, length); |
2106 |
close (fd); |
2107 |
return res; |
2108 |
} |
2109 |
|
2110 |
static int |
2111 |
eio__statvfsat (int dirfd, const char *path, struct statvfs *buf) |
2112 |
{ |
2113 |
int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC); |
2114 |
int res; |
2115 |
|
2116 |
if (fd < 0) |
2117 |
return fd; |
2118 |
|
2119 |
res = fstatvfs (fd, buf); |
2120 |
close (fd); |
2121 |
return res; |
2122 |
|
2123 |
} |
2124 |
|
2125 |
#endif |
2126 |
|
2127 |
/*****************************************************************************/ |
2128 |
|
2129 |
#define ALLOC(len) \ |
2130 |
if (!req->ptr2) \ |
2131 |
{ \ |
2132 |
X_LOCK (wrklock); \ |
2133 |
req->flags |= EIO_FLAG_PTR2_FREE; \ |
2134 |
X_UNLOCK (wrklock); \ |
2135 |
req->ptr2 = malloc (len); \ |
2136 |
if (!req->ptr2) \ |
2137 |
{ \ |
2138 |
errno = ENOMEM; \ |
2139 |
req->result = -1; \ |
2140 |
break; \ |
2141 |
} \ |
2142 |
} |
2143 |
|
2144 |
static void ecb_noinline ecb_cold |
2145 |
etp_proc_init (void) |
2146 |
{ |
2147 |
#if HAVE_PRCTL_SET_NAME |
2148 |
/* provide a more sensible "thread name" */ |
2149 |
char name[16 + 1]; |
2150 |
const int namelen = sizeof (name) - 1; |
2151 |
int len; |
2152 |
|
2153 |
prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0); |
2154 |
name [namelen] = 0; |
2155 |
len = strlen (name); |
2156 |
strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio"); |
2157 |
prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0); |
2158 |
#endif |
2159 |
} |
2160 |
|
2161 |
X_THREAD_PROC (etp_proc) |
2162 |
{ |
2163 |
ETP_REQ *req; |
2164 |
struct timespec ts; |
2165 |
etp_worker *self = (etp_worker *)thr_arg; |
2166 |
|
2167 |
etp_proc_init (); |
2168 |
|
2169 |
/* try to distribute timeouts somewhat evenly */ |
2170 |
ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
2171 |
|
2172 |
for (;;) |
2173 |
{ |
2174 |
ts.tv_sec = 0; |
2175 |
|
2176 |
X_LOCK (reqlock); |
2177 |
|
2178 |
for (;;) |
2179 |
{ |
2180 |
req = reqq_shift (&req_queue); |
2181 |
|
2182 |
if (req) |
2183 |
break; |
2184 |
|
2185 |
if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */ |
2186 |
{ |
2187 |
X_UNLOCK (reqlock); |
2188 |
X_LOCK (wrklock); |
2189 |
--started; |
2190 |
X_UNLOCK (wrklock); |
2191 |
goto quit; |
2192 |
} |
2193 |
|
2194 |
++idle; |
2195 |
|
2196 |
if (idle <= max_idle) |
2197 |
/* we are allowed to idle, so do so without any timeout */ |
2198 |
X_COND_WAIT (reqwait, reqlock); |
2199 |
else |
2200 |
{ |
2201 |
/* initialise timeout once */ |
2202 |
if (!ts.tv_sec) |
2203 |
ts.tv_sec = time (0) + idle_timeout; |
2204 |
|
2205 |
if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
2206 |
ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */ |
2207 |
} |
2208 |
|
2209 |
--idle; |
2210 |
} |
2211 |
|
2212 |
--nready; |
2213 |
|
2214 |
X_UNLOCK (reqlock); |
2215 |
|
2216 |
if (req->type < 0) |
2217 |
goto quit; |
2218 |
|
2219 |
ETP_EXECUTE (self, req); |
2220 |
|
2221 |
X_LOCK (reslock); |
2222 |
|
2223 |
++npending; |
2224 |
|
2225 |
if (!reqq_push (&res_queue, req) && want_poll_cb) |
2226 |
want_poll_cb (); |
2227 |
|
2228 |
etp_worker_clear (self); |
2229 |
|
2230 |
X_UNLOCK (reslock); |
2231 |
} |
2232 |
|
2233 |
quit: |
2234 |
free (req); |
2235 |
|
2236 |
X_LOCK (wrklock); |
2237 |
etp_worker_free (self); |
2238 |
X_UNLOCK (wrklock); |
2239 |
|
2240 |
return 0; |
2241 |
} |
2242 |
|
2243 |
/*****************************************************************************/ |
2244 |
|
2245 |
int ecb_cold |
2246 |
eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
2247 |
{ |
2248 |
return etp_init (want_poll, done_poll); |
2249 |
} |
2250 |
|
2251 |
ecb_inline void |
2252 |
eio_api_destroy (eio_req *req) |
2253 |
{ |
2254 |
free (req); |
2255 |
} |
2256 |
|
2257 |
#define REQ(rtype) \ |
2258 |
eio_req *req; \ |
2259 |
\ |
2260 |
req = (eio_req *)calloc (1, sizeof *req); \ |
2261 |
if (!req) \ |
2262 |
return 0; \ |
2263 |
\ |
2264 |
req->type = rtype; \ |
2265 |
req->pri = pri; \ |
2266 |
req->finish = cb; \ |
2267 |
req->data = data; \ |
2268 |
req->destroy = eio_api_destroy; |
2269 |
|
2270 |
#define SEND eio_submit (req); return req |
2271 |
|
2272 |
#define PATH \ |
2273 |
req->flags |= EIO_FLAG_PTR1_FREE; \ |
2274 |
req->ptr1 = strdup (path); \ |
2275 |
if (!req->ptr1) \ |
2276 |
{ \ |
2277 |
eio_api_destroy (req); \ |
2278 |
return 0; \ |
2279 |
} |
2280 |
|
2281 |
#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)]) |
2282 |
|
2283 |
static void |
2284 |
eio_execute (etp_worker *self, eio_req *req) |
2285 |
{ |
2286 |
#if HAVE_AT |
2287 |
int dirfd; |
2288 |
#else |
2289 |
const char *path; |
2290 |
#endif |
2291 |
|
2292 |
if (ecb_expect_false (EIO_CANCELLED (req))) |
2293 |
{ |
2294 |
req->result = -1; |
2295 |
req->errorno = ECANCELED; |
2296 |
return; |
2297 |
} |
2298 |
|
2299 |
if (ecb_expect_false (req->wd == EIO_INVALID_WD)) |
2300 |
{ |
2301 |
req->result = -1; |
2302 |
req->errorno = ENOENT; |
2303 |
return; |
2304 |
} |
2305 |
|
2306 |
if (req->type >= EIO_OPEN) |
2307 |
{ |
2308 |
#if HAVE_AT |
2309 |
dirfd = WD2FD (req->wd); |
2310 |
#else |
2311 |
path = wd_expand (&self->tmpbuf, req->wd, req->ptr1); |
2312 |
#endif |
2313 |
} |
2314 |
|
2315 |
switch (req->type) |
2316 |
{ |
2317 |
case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1); |
2318 |
req->result = req->wd == EIO_INVALID_WD ? -1 : 0; |
2319 |
break; |
2320 |
case EIO_WD_CLOSE: req->result = 0; |
2321 |
eio_wd_close_sync (req->wd); break; |
2322 |
|
2323 |
case EIO_SEEK: eio__lseek (req); break; |
2324 |
case EIO_READ: ALLOC (req->size); |
2325 |
req->result = req->offs >= 0 |
2326 |
? pread (req->int1, req->ptr2, req->size, req->offs) |
2327 |
: read (req->int1, req->ptr2, req->size); break; |
2328 |
case EIO_WRITE: req->result = req->offs >= 0 |
2329 |
? pwrite (req->int1, req->ptr2, req->size, req->offs) |
2330 |
: write (req->int1, req->ptr2, req->size); break; |
2331 |
|
2332 |
case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; |
2333 |
case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break; |
2334 |
|
2335 |
#if HAVE_AT |
2336 |
|
2337 |
case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
2338 |
req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break; |
2339 |
case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
2340 |
req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break; |
2341 |
case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break; |
2342 |
case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break; |
2343 |
case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break; |
2344 |
case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break; |
2345 |
|
2346 |
case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break; |
2347 |
case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */ |
2348 |
req->result = req->wd && SINGLEDOT (req->ptr1) |
2349 |
? rmdir (req->wd->str) |
2350 |
: unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break; |
2351 |
case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break; |
2352 |
case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */ |
2353 |
req->result = req->wd && SINGLEDOT (req->ptr1) |
2354 |
? rename (req->wd->str, req->ptr2) |
2355 |
: renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break; |
2356 |
case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break; |
2357 |
case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break; |
2358 |
case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; |
2359 |
case EIO_READLINK: ALLOC (PATH_MAX); |
2360 |
req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break; |
2361 |
case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
2362 |
req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
2363 |
case EIO_UTIME: |
2364 |
case EIO_FUTIME: |
2365 |
{ |
2366 |
struct timespec ts[2]; |
2367 |
struct timespec *times; |
2368 |
|
2369 |
if (req->nv1 != -1. || req->nv2 != -1.) |
2370 |
{ |
2371 |
ts[0].tv_sec = req->nv1; |
2372 |
ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9; |
2373 |
ts[1].tv_sec = req->nv2; |
2374 |
ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9; |
2375 |
|
2376 |
times = ts; |
2377 |
} |
2378 |
else |
2379 |
times = 0; |
2380 |
|
2381 |
req->result = req->type == EIO_FUTIME |
2382 |
? futimens (req->int1, times) |
2383 |
: utimensat (dirfd, req->ptr1, times, 0); |
2384 |
} |
2385 |
break; |
2386 |
|
2387 |
#else |
2388 |
|
2389 |
case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
2390 |
req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break; |
2391 |
case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
2392 |
req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break; |
2393 |
case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break; |
2394 |
case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break; |
2395 |
case EIO_TRUNCATE: req->result = truncate (path , req->offs); break; |
2396 |
case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break; |
2397 |
|
2398 |
case EIO_UNLINK: req->result = unlink (path ); break; |
2399 |
case EIO_RMDIR: req->result = rmdir (path ); break; |
2400 |
case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break; |
2401 |
case EIO_RENAME: req->result = rename (path , req->ptr2); break; |
2402 |
case EIO_LINK: req->result = link (path , req->ptr2); break; |
2403 |
case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break; |
2404 |
case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break; |
2405 |
case EIO_READLINK: ALLOC (PATH_MAX); |
2406 |
req->result = readlink (path, req->ptr2, PATH_MAX); break; |
2407 |
case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
2408 |
req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break; |
2409 |
|
2410 |
case EIO_UTIME: |
2411 |
case EIO_FUTIME: |
2412 |
{ |
2413 |
struct timeval tv[2]; |
2414 |
struct timeval *times; |
2415 |
|
2416 |
if (req->nv1 != -1. || req->nv2 != -1.) |
2417 |
{ |
2418 |
tv[0].tv_sec = req->nv1; |
2419 |
tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6; |
2420 |
tv[1].tv_sec = req->nv2; |
2421 |
tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6; |
2422 |
|
2423 |
times = tv; |
2424 |
} |
2425 |
else |
2426 |
times = 0; |
2427 |
|
2428 |
req->result = req->type == EIO_FUTIME |
2429 |
? futimes (req->int1, times) |
2430 |
: utimes (req->ptr1, times); |
2431 |
} |
2432 |
break; |
2433 |
|
2434 |
#endif |
2435 |
|
2436 |
case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1))) |
2437 |
{ |
2438 |
ALLOC (req->result); |
2439 |
memcpy (req->ptr2, self->tmpbuf.ptr, req->result); |
2440 |
} |
2441 |
break; |
2442 |
|
2443 |
case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
2444 |
req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
2445 |
|
2446 |
case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
2447 |
req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
2448 |
|
2449 |
case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
2450 |
case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
2451 |
case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; |
2452 |
|
2453 |
case EIO_CLOSE: req->result = close (req->int1); break; |
2454 |
case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; |
2455 |
case EIO_SYNC: req->result = 0; sync (); break; |
2456 |
case EIO_FSYNC: req->result = fsync (req->int1); break; |
2457 |
case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
2458 |
case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break; |
2459 |
case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
2460 |
case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
2461 |
case EIO_MTOUCH: req->result = eio__mtouch (req); break; |
2462 |
case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; |
2463 |
case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; |
2464 |
case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; |
2465 |
|
2466 |
case EIO_READDIR: eio__scandir (req, self); break; |
2467 |
|
2468 |
case EIO_BUSY: |
2469 |
#ifdef _WIN32 |
2470 |
Sleep (req->nv1 * 1e3); |
2471 |
#else |
2472 |
{ |
2473 |
struct timeval tv; |
2474 |
|
2475 |
tv.tv_sec = req->nv1; |
2476 |
tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6; |
2477 |
|
2478 |
req->result = select (0, 0, 0, 0, &tv); |
2479 |
} |
2480 |
#endif |
2481 |
break; |
2482 |
|
2483 |
case EIO_GROUP: |
2484 |
abort (); /* handled in eio_request */ |
2485 |
|
2486 |
case EIO_NOP: |
2487 |
req->result = 0; |
2488 |
break; |
2489 |
|
2490 |
case EIO_CUSTOM: |
2491 |
req->feed (req); |
2492 |
break; |
2493 |
|
2494 |
default: |
2495 |
req->result = EIO_ENOSYS (); |
2496 |
break; |
2497 |
} |
2498 |
|
2499 |
req->errorno = errno; |
2500 |
} |
2501 |
|
2502 |
#ifndef EIO_NO_WRAPPERS |
2503 |
|
2504 |
eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data) |
2505 |
{ |
2506 |
REQ (EIO_WD_OPEN); PATH; SEND; |
2507 |
} |
2508 |
|
2509 |
eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data) |
2510 |
{ |
2511 |
REQ (EIO_WD_CLOSE); req->wd = wd; SEND; |
2512 |
} |
2513 |
|
2514 |
eio_req *eio_nop (int pri, eio_cb cb, void *data) |
2515 |
{ |
2516 |
REQ (EIO_NOP); SEND; |
2517 |
} |
2518 |
|
2519 |
eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data) |
2520 |
{ |
2521 |
REQ (EIO_BUSY); req->nv1 = delay; SEND; |
2522 |
} |
2523 |
|
2524 |
eio_req *eio_sync (int pri, eio_cb cb, void *data) |
2525 |
{ |
2526 |
REQ (EIO_SYNC); SEND; |
2527 |
} |
2528 |
|
2529 |
eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
2530 |
{ |
2531 |
REQ (EIO_FSYNC); req->int1 = fd; SEND; |
2532 |
} |
2533 |
|
2534 |
eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
2535 |
{ |
2536 |
REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
2537 |
} |
2538 |
|
2539 |
eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
2540 |
{ |
2541 |
REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
2542 |
} |
2543 |
|
2544 |
eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data) |
2545 |
{ |
2546 |
REQ (EIO_SYNCFS); req->int1 = fd; SEND; |
2547 |
} |
2548 |
|
2549 |
eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
2550 |
{ |
2551 |
REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
2552 |
} |
2553 |
|
2554 |
eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
2555 |
{ |
2556 |
REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
2557 |
} |
2558 |
|
2559 |
eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data) |
2560 |
{ |
2561 |
REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND; |
2562 |
} |
2563 |
|
2564 |
eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) |
2565 |
{ |
2566 |
REQ (EIO_MLOCKALL); req->int1 = flags; SEND; |
2567 |
} |
2568 |
|
2569 |
eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) |
2570 |
{ |
2571 |
REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; |
2572 |
} |
2573 |
|
2574 |
eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) |
2575 |
{ |
2576 |
REQ (EIO_CLOSE); req->int1 = fd; SEND; |
2577 |
} |
2578 |
|
2579 |
eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) |
2580 |
{ |
2581 |
REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; |
2582 |
} |
2583 |
|
2584 |
eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data) |
2585 |
{ |
2586 |
REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND; |
2587 |
} |
2588 |
|
2589 |
eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) |
2590 |
{ |
2591 |
REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; |
2592 |
} |
2593 |
|
2594 |
eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) |
2595 |
{ |
2596 |
REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; |
2597 |
} |
2598 |
|
2599 |
eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
2600 |
{ |
2601 |
REQ (EIO_FSTAT); req->int1 = fd; SEND; |
2602 |
} |
2603 |
|
2604 |
eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data) |
2605 |
{ |
2606 |
REQ (EIO_FSTATVFS); req->int1 = fd; SEND; |
2607 |
} |
2608 |
|
2609 |
eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
2610 |
{ |
2611 |
REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
2612 |
} |
2613 |
|
2614 |
eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data) |
2615 |
{ |
2616 |
REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; |
2617 |
} |
2618 |
|
2619 |
eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) |
2620 |
{ |
2621 |
REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; |
2622 |
} |
2623 |
|
2624 |
eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data) |
2625 |
{ |
2626 |
REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
2627 |
} |
2628 |
|
2629 |
eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) |
2630 |
{ |
2631 |
REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND; |
2632 |
} |
2633 |
|
2634 |
eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data) |
2635 |
{ |
2636 |
REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; |
2637 |
} |
2638 |
|
2639 |
eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data) |
2640 |
{ |
2641 |
REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; |
2642 |
} |
2643 |
|
2644 |
eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data) |
2645 |
{ |
2646 |
REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; |
2647 |
} |
2648 |
|
2649 |
eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) |
2650 |
{ |
2651 |
REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; |
2652 |
} |
2653 |
|
2654 |
eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data) |
2655 |
{ |
2656 |
REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
2657 |
} |
2658 |
|
2659 |
eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) |
2660 |
{ |
2661 |
REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND; |
2662 |
} |
2663 |
|
2664 |
eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data) |
2665 |
{ |
2666 |
REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND; |
2667 |
} |
2668 |
|
2669 |
static eio_req * |
2670 |
eio__1path (int type, const char *path, int pri, eio_cb cb, void *data) |
2671 |
{ |
2672 |
REQ (type); PATH; SEND; |
2673 |
} |
2674 |
|
2675 |
eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) |
2676 |
{ |
2677 |
return eio__1path (EIO_READLINK, path, pri, cb, data); |
2678 |
} |
2679 |
|
2680 |
eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data) |
2681 |
{ |
2682 |
return eio__1path (EIO_REALPATH, path, pri, cb, data); |
2683 |
} |
2684 |
|
2685 |
eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) |
2686 |
{ |
2687 |
return eio__1path (EIO_STAT, path, pri, cb, data); |
2688 |
} |
2689 |
|
2690 |
eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
2691 |
{ |
2692 |
return eio__1path (EIO_LSTAT, path, pri, cb, data); |
2693 |
} |
2694 |
|
2695 |
eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data) |
2696 |
{ |
2697 |
return eio__1path (EIO_STATVFS, path, pri, cb, data); |
2698 |
} |
2699 |
|
2700 |
eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
2701 |
{ |
2702 |
return eio__1path (EIO_UNLINK, path, pri, cb, data); |
2703 |
} |
2704 |
|
2705 |
eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
2706 |
{ |
2707 |
return eio__1path (EIO_RMDIR, path, pri, cb, data); |
2708 |
} |
2709 |
|
2710 |
eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data) |
2711 |
{ |
2712 |
REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
2713 |
} |
2714 |
|
2715 |
eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
2716 |
{ |
2717 |
REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND; |
2718 |
} |
2719 |
|
2720 |
static eio_req * |
2721 |
eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2722 |
{ |
2723 |
REQ (type); PATH; |
2724 |
|
2725 |
req->flags |= EIO_FLAG_PTR2_FREE; |
2726 |
req->ptr2 = strdup (new_path); |
2727 |
if (!req->ptr2) |
2728 |
{ |
2729 |
eio_api_destroy (req); |
2730 |
return 0; |
2731 |
} |
2732 |
|
2733 |
SEND; |
2734 |
} |
2735 |
|
2736 |
eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2737 |
{ |
2738 |
return eio__2path (EIO_LINK, path, new_path, pri, cb, data); |
2739 |
} |
2740 |
|
2741 |
eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2742 |
{ |
2743 |
return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data); |
2744 |
} |
2745 |
|
2746 |
eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2747 |
{ |
2748 |
return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
2749 |
} |
2750 |
|
2751 |
eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data) |
2752 |
{ |
2753 |
REQ (EIO_CUSTOM); req->feed = execute; SEND; |
2754 |
} |
2755 |
|
2756 |
#endif |
2757 |
|
2758 |
eio_req *eio_grp (eio_cb cb, void *data) |
2759 |
{ |
2760 |
const int pri = EIO_PRI_MAX; |
2761 |
|
2762 |
REQ (EIO_GROUP); SEND; |
2763 |
} |
2764 |
|
2765 |
#undef REQ |
2766 |
#undef PATH |
2767 |
#undef SEND |
2768 |
|
2769 |
/*****************************************************************************/ |
2770 |
/* grp functions */ |
2771 |
|
2772 |
void |
2773 |
eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) |
2774 |
{ |
2775 |
grp->int2 = limit; |
2776 |
grp->feed = feed; |
2777 |
|
2778 |
grp_try_feed (grp); |
2779 |
} |
2780 |
|
2781 |
void |
2782 |
eio_grp_limit (eio_req *grp, int limit) |
2783 |
{ |
2784 |
grp->int2 = limit; |
2785 |
|
2786 |
grp_try_feed (grp); |
2787 |
} |
2788 |
|
2789 |
void |
2790 |
eio_grp_add (eio_req *grp, eio_req *req) |
2791 |
{ |
2792 |
assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
2793 |
|
2794 |
grp->flags |= EIO_FLAG_GROUPADD; |
2795 |
|
2796 |
++grp->size; |
2797 |
req->grp = grp; |
2798 |
|
2799 |
req->grp_prev = 0; |
2800 |
req->grp_next = grp->grp_first; |
2801 |
|
2802 |
if (grp->grp_first) |
2803 |
grp->grp_first->grp_prev = req; |
2804 |
|
2805 |
grp->grp_first = req; |
2806 |
} |
2807 |
|
2808 |
/*****************************************************************************/ |
2809 |
/* misc garbage */ |
2810 |
|
2811 |
eio_ssize_t |
2812 |
eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
2813 |
{ |
2814 |
return eio__sendfile (ofd, ifd, offset, count); |
2815 |
} |
2816 |
|