1 |
/* |
2 |
* libeio implementation |
3 |
* |
4 |
* Copyright (c) 2007,2008,2009 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 |
#include "eio.h" |
41 |
|
42 |
#ifdef EIO_STACKSIZE |
43 |
# define XTHREAD_STACKSIZE EIO_STACKSIZE |
44 |
#endif |
45 |
#include "xthread.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 |
#ifndef EIO_FINISH |
59 |
# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
60 |
#endif |
61 |
|
62 |
#ifndef EIO_DESTROY |
63 |
# define EIO_DESTROY(req) do { if ((req)->destroy) (req)->destroy (req); } while (0) |
64 |
#endif |
65 |
|
66 |
#ifndef EIO_FEED |
67 |
# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) |
68 |
#endif |
69 |
|
70 |
#ifdef _WIN32 |
71 |
|
72 |
/*doh*/ |
73 |
#else |
74 |
|
75 |
# include "config.h" |
76 |
# include <sys/time.h> |
77 |
# include <sys/select.h> |
78 |
# include <unistd.h> |
79 |
# include <utime.h> |
80 |
# include <signal.h> |
81 |
# include <dirent.h> |
82 |
|
83 |
/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
84 |
# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) |
85 |
# define _DIRENT_HAVE_D_TYPE /* sigh */ |
86 |
# define D_INO(de) (de)->d_fileno |
87 |
# define D_NAMLEN(de) (de)->d_namlen |
88 |
# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 |
89 |
# define D_INO(de) (de)->d_ino |
90 |
# endif |
91 |
|
92 |
#ifdef _D_EXACT_NAMLEN |
93 |
# undef D_NAMLEN |
94 |
# define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
95 |
#endif |
96 |
|
97 |
# ifdef _DIRENT_HAVE_D_TYPE |
98 |
# define D_TYPE(de) (de)->d_type |
99 |
# endif |
100 |
|
101 |
# ifndef EIO_STRUCT_DIRENT |
102 |
# define EIO_STRUCT_DIRENT struct dirent |
103 |
# endif |
104 |
|
105 |
#endif |
106 |
|
107 |
#if HAVE_SENDFILE |
108 |
# if __linux |
109 |
# include <sys/sendfile.h> |
110 |
# elif __freebsd |
111 |
# include <sys/socket.h> |
112 |
# include <sys/uio.h> |
113 |
# elif __hpux |
114 |
# include <sys/socket.h> |
115 |
# elif __solaris /* not yet */ |
116 |
# include <sys/sendfile.h> |
117 |
# else |
118 |
# error sendfile support requested but not available |
119 |
# endif |
120 |
#endif |
121 |
|
122 |
#ifndef D_TYPE |
123 |
# define D_TYPE(de) 0 |
124 |
#endif |
125 |
#ifndef D_INO |
126 |
# define D_INO(de) 0 |
127 |
#endif |
128 |
#ifndef D_NAMLEN |
129 |
# define D_NAMLEN(de) strlen ((de)->d_name) |
130 |
#endif |
131 |
|
132 |
/* number of seconds after which an idle threads exit */ |
133 |
#define IDLE_TIMEOUT 10 |
134 |
|
135 |
/* used for struct dirent, AIX doesn't provide it */ |
136 |
#ifndef NAME_MAX |
137 |
# define NAME_MAX 4096 |
138 |
#endif |
139 |
|
140 |
/* buffer size for various temporary buffers */ |
141 |
#define EIO_BUFSIZE 65536 |
142 |
|
143 |
#define dBUF \ |
144 |
char *eio_buf; \ |
145 |
ETP_WORKER_LOCK (self); \ |
146 |
self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ |
147 |
ETP_WORKER_UNLOCK (self); \ |
148 |
errno = ENOMEM; \ |
149 |
if (!eio_buf) \ |
150 |
return -1; |
151 |
|
152 |
#define EIO_TICKS ((1000000 + 1023) >> 10) |
153 |
|
154 |
/*****************************************************************************/ |
155 |
|
156 |
#if __GNUC__ >= 3 |
157 |
# define expect(expr,value) __builtin_expect ((expr),(value)) |
158 |
#else |
159 |
# define expect(expr,value) (expr) |
160 |
#endif |
161 |
|
162 |
#define expect_false(expr) expect ((expr) != 0, 0) |
163 |
#define expect_true(expr) expect ((expr) != 0, 1) |
164 |
|
165 |
/*****************************************************************************/ |
166 |
|
167 |
#define ETP_PRI_MIN EIO_PRI_MIN |
168 |
#define ETP_PRI_MAX EIO_PRI_MAX |
169 |
|
170 |
struct etp_worker; |
171 |
|
172 |
#define ETP_REQ eio_req |
173 |
#define ETP_DESTROY(req) eio_destroy (req) |
174 |
static int eio_finish (eio_req *req); |
175 |
#define ETP_FINISH(req) eio_finish (req) |
176 |
static void eio_execute (struct etp_worker *self, eio_req *req); |
177 |
#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) |
178 |
|
179 |
#define ETP_WORKER_CLEAR(req) \ |
180 |
if (wrk->dbuf) \ |
181 |
{ \ |
182 |
free (wrk->dbuf); \ |
183 |
wrk->dbuf = 0; \ |
184 |
} \ |
185 |
\ |
186 |
if (wrk->dirp) \ |
187 |
{ \ |
188 |
closedir (wrk->dirp); \ |
189 |
wrk->dirp = 0; \ |
190 |
} |
191 |
|
192 |
#define ETP_WORKER_COMMON \ |
193 |
void *dbuf; \ |
194 |
DIR *dirp; |
195 |
|
196 |
/*****************************************************************************/ |
197 |
|
198 |
#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
199 |
|
200 |
/* calculcate time difference in ~1/EIO_TICKS of a second */ |
201 |
static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
202 |
{ |
203 |
return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
204 |
+ ((tv2->tv_usec - tv1->tv_usec) >> 10); |
205 |
} |
206 |
|
207 |
static unsigned int started, idle, wanted = 4; |
208 |
|
209 |
static void (*want_poll_cb) (void); |
210 |
static void (*done_poll_cb) (void); |
211 |
|
212 |
static unsigned int max_poll_time; /* reslock */ |
213 |
static unsigned int max_poll_reqs; /* reslock */ |
214 |
|
215 |
static volatile unsigned int nreqs; /* reqlock */ |
216 |
static volatile unsigned int nready; /* reqlock */ |
217 |
static volatile unsigned int npending; /* reqlock */ |
218 |
static volatile unsigned int max_idle = 4; |
219 |
|
220 |
static mutex_t wrklock = X_MUTEX_INIT; |
221 |
static mutex_t reslock = X_MUTEX_INIT; |
222 |
static mutex_t reqlock = X_MUTEX_INIT; |
223 |
static cond_t reqwait = X_COND_INIT; |
224 |
|
225 |
#if !HAVE_PREADWRITE |
226 |
/* |
227 |
* make our pread/pwrite emulation safe against themselves, but not against |
228 |
* normal read/write by using a mutex. slows down execution a lot, |
229 |
* but that's your problem, not mine. |
230 |
*/ |
231 |
static mutex_t preadwritelock = X_MUTEX_INIT; |
232 |
#endif |
233 |
|
234 |
typedef struct etp_worker |
235 |
{ |
236 |
/* locked by wrklock */ |
237 |
struct etp_worker *prev, *next; |
238 |
|
239 |
thread_t tid; |
240 |
|
241 |
/* locked by reslock, reqlock or wrklock */ |
242 |
ETP_REQ *req; /* currently processed request */ |
243 |
|
244 |
ETP_WORKER_COMMON |
245 |
} etp_worker; |
246 |
|
247 |
static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ |
248 |
|
249 |
#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) |
250 |
#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) |
251 |
|
252 |
/* worker threads management */ |
253 |
|
254 |
static void etp_worker_clear (etp_worker *wrk) |
255 |
{ |
256 |
ETP_WORKER_CLEAR (wrk); |
257 |
} |
258 |
|
259 |
static void etp_worker_free (etp_worker *wrk) |
260 |
{ |
261 |
wrk->next->prev = wrk->prev; |
262 |
wrk->prev->next = wrk->next; |
263 |
|
264 |
free (wrk); |
265 |
} |
266 |
|
267 |
static unsigned int etp_nreqs (void) |
268 |
{ |
269 |
int retval; |
270 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
271 |
retval = nreqs; |
272 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
273 |
return retval; |
274 |
} |
275 |
|
276 |
static unsigned int etp_nready (void) |
277 |
{ |
278 |
unsigned int retval; |
279 |
|
280 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
281 |
retval = nready; |
282 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
283 |
|
284 |
return retval; |
285 |
} |
286 |
|
287 |
static unsigned int etp_npending (void) |
288 |
{ |
289 |
unsigned int retval; |
290 |
|
291 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
292 |
retval = npending; |
293 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
294 |
|
295 |
return retval; |
296 |
} |
297 |
|
298 |
static unsigned int etp_nthreads (void) |
299 |
{ |
300 |
unsigned int retval; |
301 |
|
302 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
303 |
retval = started; |
304 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
305 |
|
306 |
return retval; |
307 |
} |
308 |
|
309 |
/* |
310 |
* a somewhat faster data structure might be nice, but |
311 |
* with 8 priorities this actually needs <20 insns |
312 |
* per shift, the most expensive operation. |
313 |
*/ |
314 |
typedef struct { |
315 |
ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */ |
316 |
int size; |
317 |
} etp_reqq; |
318 |
|
319 |
static etp_reqq req_queue; |
320 |
static etp_reqq res_queue; |
321 |
|
322 |
static int reqq_push (etp_reqq *q, ETP_REQ *req) |
323 |
{ |
324 |
int pri = req->pri; |
325 |
req->next = 0; |
326 |
|
327 |
if (q->qe[pri]) |
328 |
{ |
329 |
q->qe[pri]->next = req; |
330 |
q->qe[pri] = req; |
331 |
} |
332 |
else |
333 |
q->qe[pri] = q->qs[pri] = req; |
334 |
|
335 |
return q->size++; |
336 |
} |
337 |
|
338 |
static ETP_REQ *reqq_shift (etp_reqq *q) |
339 |
{ |
340 |
int pri; |
341 |
|
342 |
if (!q->size) |
343 |
return 0; |
344 |
|
345 |
--q->size; |
346 |
|
347 |
for (pri = ETP_NUM_PRI; pri--; ) |
348 |
{ |
349 |
eio_req *req = q->qs[pri]; |
350 |
|
351 |
if (req) |
352 |
{ |
353 |
if (!(q->qs[pri] = (eio_req *)req->next)) |
354 |
q->qe[pri] = 0; |
355 |
|
356 |
return req; |
357 |
} |
358 |
} |
359 |
|
360 |
abort (); |
361 |
} |
362 |
|
363 |
static void etp_atfork_prepare (void) |
364 |
{ |
365 |
X_LOCK (wrklock); |
366 |
X_LOCK (reqlock); |
367 |
X_LOCK (reslock); |
368 |
#if !HAVE_PREADWRITE |
369 |
X_LOCK (preadwritelock); |
370 |
#endif |
371 |
} |
372 |
|
373 |
static void etp_atfork_parent (void) |
374 |
{ |
375 |
#if !HAVE_PREADWRITE |
376 |
X_UNLOCK (preadwritelock); |
377 |
#endif |
378 |
X_UNLOCK (reslock); |
379 |
X_UNLOCK (reqlock); |
380 |
X_UNLOCK (wrklock); |
381 |
} |
382 |
|
383 |
static void etp_atfork_child (void) |
384 |
{ |
385 |
ETP_REQ *prv; |
386 |
|
387 |
while ((prv = reqq_shift (&req_queue))) |
388 |
ETP_DESTROY (prv); |
389 |
|
390 |
while ((prv = reqq_shift (&res_queue))) |
391 |
ETP_DESTROY (prv); |
392 |
|
393 |
while (wrk_first.next != &wrk_first) |
394 |
{ |
395 |
etp_worker *wrk = wrk_first.next; |
396 |
|
397 |
if (wrk->req) |
398 |
ETP_DESTROY (wrk->req); |
399 |
|
400 |
etp_worker_clear (wrk); |
401 |
etp_worker_free (wrk); |
402 |
} |
403 |
|
404 |
started = 0; |
405 |
idle = 0; |
406 |
nreqs = 0; |
407 |
nready = 0; |
408 |
npending = 0; |
409 |
|
410 |
etp_atfork_parent (); |
411 |
} |
412 |
|
413 |
static void |
414 |
etp_once_init (void) |
415 |
{ |
416 |
X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
417 |
} |
418 |
|
419 |
static int |
420 |
etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
421 |
{ |
422 |
static pthread_once_t doinit = PTHREAD_ONCE_INIT; |
423 |
|
424 |
pthread_once (&doinit, etp_once_init); |
425 |
|
426 |
want_poll_cb = want_poll; |
427 |
done_poll_cb = done_poll; |
428 |
|
429 |
return 0; |
430 |
} |
431 |
|
432 |
X_THREAD_PROC (etp_proc); |
433 |
|
434 |
static void etp_start_thread (void) |
435 |
{ |
436 |
etp_worker *wrk = calloc (1, sizeof (etp_worker)); |
437 |
|
438 |
/*TODO*/ |
439 |
assert (("unable to allocate worker thread data", wrk)); |
440 |
|
441 |
X_LOCK (wrklock); |
442 |
|
443 |
if (thread_create (&wrk->tid, etp_proc, (void *)wrk)) |
444 |
{ |
445 |
wrk->prev = &wrk_first; |
446 |
wrk->next = wrk_first.next; |
447 |
wrk_first.next->prev = wrk; |
448 |
wrk_first.next = wrk; |
449 |
++started; |
450 |
} |
451 |
else |
452 |
free (wrk); |
453 |
|
454 |
X_UNLOCK (wrklock); |
455 |
} |
456 |
|
457 |
static void etp_maybe_start_thread (void) |
458 |
{ |
459 |
if (expect_true (etp_nthreads () >= wanted)) |
460 |
return; |
461 |
|
462 |
/* todo: maybe use idle here, but might be less exact */ |
463 |
if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
464 |
return; |
465 |
|
466 |
etp_start_thread (); |
467 |
} |
468 |
|
469 |
static void etp_end_thread (void) |
470 |
{ |
471 |
eio_req *req = calloc (1, sizeof (eio_req)); |
472 |
|
473 |
req->type = -1; |
474 |
req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
475 |
|
476 |
X_LOCK (reqlock); |
477 |
reqq_push (&req_queue, req); |
478 |
X_COND_SIGNAL (reqwait); |
479 |
X_UNLOCK (reqlock); |
480 |
|
481 |
X_LOCK (wrklock); |
482 |
--started; |
483 |
X_UNLOCK (wrklock); |
484 |
} |
485 |
|
486 |
static int etp_poll (void) |
487 |
{ |
488 |
unsigned int maxreqs; |
489 |
unsigned int maxtime; |
490 |
struct timeval tv_start, tv_now; |
491 |
|
492 |
X_LOCK (reslock); |
493 |
maxreqs = max_poll_reqs; |
494 |
maxtime = max_poll_time; |
495 |
X_UNLOCK (reslock); |
496 |
|
497 |
if (maxtime) |
498 |
gettimeofday (&tv_start, 0); |
499 |
|
500 |
for (;;) |
501 |
{ |
502 |
ETP_REQ *req; |
503 |
|
504 |
etp_maybe_start_thread (); |
505 |
|
506 |
X_LOCK (reslock); |
507 |
req = reqq_shift (&res_queue); |
508 |
|
509 |
if (req) |
510 |
{ |
511 |
--npending; |
512 |
|
513 |
if (!res_queue.size && done_poll_cb) |
514 |
done_poll_cb (); |
515 |
} |
516 |
|
517 |
X_UNLOCK (reslock); |
518 |
|
519 |
if (!req) |
520 |
return 0; |
521 |
|
522 |
X_LOCK (reqlock); |
523 |
--nreqs; |
524 |
X_UNLOCK (reqlock); |
525 |
|
526 |
if (expect_false (req->type == EIO_GROUP && req->size)) |
527 |
{ |
528 |
req->int1 = 1; /* mark request as delayed */ |
529 |
continue; |
530 |
} |
531 |
else |
532 |
{ |
533 |
int res = ETP_FINISH (req); |
534 |
if (expect_false (res)) |
535 |
return res; |
536 |
} |
537 |
|
538 |
if (expect_false (maxreqs && !--maxreqs)) |
539 |
break; |
540 |
|
541 |
if (maxtime) |
542 |
{ |
543 |
gettimeofday (&tv_now, 0); |
544 |
|
545 |
if (tvdiff (&tv_start, &tv_now) >= maxtime) |
546 |
break; |
547 |
} |
548 |
} |
549 |
|
550 |
errno = EAGAIN; |
551 |
return -1; |
552 |
} |
553 |
|
554 |
static void etp_cancel (ETP_REQ *req) |
555 |
{ |
556 |
X_LOCK (wrklock); |
557 |
req->flags |= EIO_FLAG_CANCELLED; |
558 |
X_UNLOCK (wrklock); |
559 |
|
560 |
eio_grp_cancel (req); |
561 |
} |
562 |
|
563 |
static void etp_submit (ETP_REQ *req) |
564 |
{ |
565 |
req->pri -= ETP_PRI_MIN; |
566 |
|
567 |
if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
568 |
if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
569 |
|
570 |
if (expect_false (req->type == EIO_GROUP)) |
571 |
{ |
572 |
/* I hope this is worth it :/ */ |
573 |
X_LOCK (reqlock); |
574 |
++nreqs; |
575 |
X_UNLOCK (reqlock); |
576 |
|
577 |
X_LOCK (reslock); |
578 |
|
579 |
++npending; |
580 |
|
581 |
if (!reqq_push (&res_queue, req) && want_poll_cb) |
582 |
want_poll_cb (); |
583 |
|
584 |
X_UNLOCK (reslock); |
585 |
} |
586 |
else |
587 |
{ |
588 |
X_LOCK (reqlock); |
589 |
++nreqs; |
590 |
++nready; |
591 |
reqq_push (&req_queue, req); |
592 |
X_COND_SIGNAL (reqwait); |
593 |
X_UNLOCK (reqlock); |
594 |
|
595 |
etp_maybe_start_thread (); |
596 |
} |
597 |
} |
598 |
|
599 |
static void etp_set_max_poll_time (double nseconds) |
600 |
{ |
601 |
if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
602 |
max_poll_time = nseconds; |
603 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
604 |
} |
605 |
|
606 |
static void etp_set_max_poll_reqs (unsigned int maxreqs) |
607 |
{ |
608 |
if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
609 |
max_poll_reqs = maxreqs; |
610 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
611 |
} |
612 |
|
613 |
static void etp_set_max_idle (unsigned int nthreads) |
614 |
{ |
615 |
if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
616 |
max_idle = nthreads <= 0 ? 1 : nthreads; |
617 |
if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
618 |
} |
619 |
|
620 |
static void etp_set_min_parallel (unsigned int nthreads) |
621 |
{ |
622 |
if (wanted < nthreads) |
623 |
wanted = nthreads; |
624 |
} |
625 |
|
626 |
static void etp_set_max_parallel (unsigned int nthreads) |
627 |
{ |
628 |
if (wanted > nthreads) |
629 |
wanted = nthreads; |
630 |
|
631 |
while (started > wanted) |
632 |
etp_end_thread (); |
633 |
} |
634 |
|
635 |
/*****************************************************************************/ |
636 |
|
637 |
static void grp_try_feed (eio_req *grp) |
638 |
{ |
639 |
while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
640 |
{ |
641 |
grp->flags &= ~EIO_FLAG_GROUPADD; |
642 |
|
643 |
EIO_FEED (grp); |
644 |
|
645 |
/* stop if no progress has been made */ |
646 |
if (!(grp->flags & EIO_FLAG_GROUPADD)) |
647 |
{ |
648 |
grp->feed = 0; |
649 |
break; |
650 |
} |
651 |
} |
652 |
} |
653 |
|
654 |
static int grp_dec (eio_req *grp) |
655 |
{ |
656 |
--grp->size; |
657 |
|
658 |
/* call feeder, if applicable */ |
659 |
grp_try_feed (grp); |
660 |
|
661 |
/* finish, if done */ |
662 |
if (!grp->size && grp->int1) |
663 |
return eio_finish (grp); |
664 |
else |
665 |
return 0; |
666 |
} |
667 |
|
668 |
void eio_destroy (eio_req *req) |
669 |
{ |
670 |
if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); |
671 |
if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); |
672 |
|
673 |
EIO_DESTROY (req); |
674 |
} |
675 |
|
676 |
static int eio_finish (eio_req *req) |
677 |
{ |
678 |
int res = EIO_FINISH (req); |
679 |
|
680 |
if (req->grp) |
681 |
{ |
682 |
int res2; |
683 |
eio_req *grp = req->grp; |
684 |
|
685 |
/* unlink request */ |
686 |
if (req->grp_next) req->grp_next->grp_prev = req->grp_prev; |
687 |
if (req->grp_prev) req->grp_prev->grp_next = req->grp_next; |
688 |
|
689 |
if (grp->grp_first == req) |
690 |
grp->grp_first = req->grp_next; |
691 |
|
692 |
res2 = grp_dec (grp); |
693 |
|
694 |
if (!res && res2) |
695 |
res = res2; |
696 |
} |
697 |
|
698 |
eio_destroy (req); |
699 |
|
700 |
return res; |
701 |
} |
702 |
|
703 |
void eio_grp_cancel (eio_req *grp) |
704 |
{ |
705 |
for (grp = grp->grp_first; grp; grp = grp->grp_next) |
706 |
eio_cancel (grp); |
707 |
} |
708 |
|
709 |
void eio_cancel (eio_req *req) |
710 |
{ |
711 |
etp_cancel (req); |
712 |
} |
713 |
|
714 |
void eio_submit (eio_req *req) |
715 |
{ |
716 |
etp_submit (req); |
717 |
} |
718 |
|
719 |
unsigned int eio_nreqs (void) |
720 |
{ |
721 |
return etp_nreqs (); |
722 |
} |
723 |
|
724 |
unsigned int eio_nready (void) |
725 |
{ |
726 |
return etp_nready (); |
727 |
} |
728 |
|
729 |
unsigned int eio_npending (void) |
730 |
{ |
731 |
return etp_npending (); |
732 |
} |
733 |
|
734 |
unsigned int eio_nthreads (void) |
735 |
{ |
736 |
return etp_nthreads (); |
737 |
} |
738 |
|
739 |
void eio_set_max_poll_time (double nseconds) |
740 |
{ |
741 |
etp_set_max_poll_time (nseconds); |
742 |
} |
743 |
|
744 |
void eio_set_max_poll_reqs (unsigned int maxreqs) |
745 |
{ |
746 |
etp_set_max_poll_reqs (maxreqs); |
747 |
} |
748 |
|
749 |
void eio_set_max_idle (unsigned int nthreads) |
750 |
{ |
751 |
etp_set_max_idle (nthreads); |
752 |
} |
753 |
|
754 |
void eio_set_min_parallel (unsigned int nthreads) |
755 |
{ |
756 |
etp_set_min_parallel (nthreads); |
757 |
} |
758 |
|
759 |
void eio_set_max_parallel (unsigned int nthreads) |
760 |
{ |
761 |
etp_set_max_parallel (nthreads); |
762 |
} |
763 |
|
764 |
int eio_poll (void) |
765 |
{ |
766 |
return etp_poll (); |
767 |
} |
768 |
|
769 |
/*****************************************************************************/ |
770 |
/* work around various missing functions */ |
771 |
|
772 |
#if !HAVE_PREADWRITE |
773 |
# undef pread |
774 |
# undef pwrite |
775 |
# define pread eio__pread |
776 |
# define pwrite eio__pwrite |
777 |
|
778 |
static ssize_t |
779 |
eio__pread (int fd, void *buf, size_t count, off_t offset) |
780 |
{ |
781 |
ssize_t res; |
782 |
off_t ooffset; |
783 |
|
784 |
X_LOCK (preadwritelock); |
785 |
ooffset = lseek (fd, 0, SEEK_CUR); |
786 |
lseek (fd, offset, SEEK_SET); |
787 |
res = read (fd, buf, count); |
788 |
lseek (fd, ooffset, SEEK_SET); |
789 |
X_UNLOCK (preadwritelock); |
790 |
|
791 |
return res; |
792 |
} |
793 |
|
794 |
static ssize_t |
795 |
eio__pwrite (int fd, void *buf, size_t count, off_t offset) |
796 |
{ |
797 |
ssize_t res; |
798 |
off_t ooffset; |
799 |
|
800 |
X_LOCK (preadwritelock); |
801 |
ooffset = lseek (fd, 0, SEEK_CUR); |
802 |
lseek (fd, offset, SEEK_SET); |
803 |
res = write (fd, buf, count); |
804 |
lseek (fd, ooffset, SEEK_SET); |
805 |
X_UNLOCK (preadwritelock); |
806 |
|
807 |
return res; |
808 |
} |
809 |
#endif |
810 |
|
811 |
#ifndef HAVE_FUTIMES |
812 |
|
813 |
# undef utimes |
814 |
# undef futimes |
815 |
# define utimes(path,times) eio__utimes (path, times) |
816 |
# define futimes(fd,times) eio__futimes (fd, times) |
817 |
|
818 |
static int |
819 |
eio__utimes (const char *filename, const struct timeval times[2]) |
820 |
{ |
821 |
if (times) |
822 |
{ |
823 |
struct utimbuf buf; |
824 |
|
825 |
buf.actime = times[0].tv_sec; |
826 |
buf.modtime = times[1].tv_sec; |
827 |
|
828 |
return utime (filename, &buf); |
829 |
} |
830 |
else |
831 |
return utime (filename, 0); |
832 |
} |
833 |
|
834 |
static int eio__futimes (int fd, const struct timeval tv[2]) |
835 |
{ |
836 |
errno = ENOSYS; |
837 |
return -1; |
838 |
} |
839 |
|
840 |
#endif |
841 |
|
842 |
#if !HAVE_FDATASYNC |
843 |
# undef fdatasync |
844 |
# define fdatasync(fd) fsync (fd) |
845 |
#endif |
846 |
|
847 |
/* sync_file_range always needs emulation */ |
848 |
int |
849 |
eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
850 |
{ |
851 |
#if HAVE_SYNC_FILE_RANGE |
852 |
int res; |
853 |
|
854 |
if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE |
855 |
|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE |
856 |
|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER) |
857 |
{ |
858 |
flags = 0 |
859 |
| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0) |
860 |
| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0) |
861 |
| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); |
862 |
} |
863 |
|
864 |
res = sync_file_range (fd, offset, nbytes, flags); |
865 |
|
866 |
if (!res || errno != ENOSYS) |
867 |
return res; |
868 |
#endif |
869 |
|
870 |
/* even though we could play tricks with the flags, it's better to always |
871 |
* call fdatasync, as thta matches the expectation of it's users best */ |
872 |
return fdatasync (fd); |
873 |
} |
874 |
|
875 |
#if !HAVE_READAHEAD |
876 |
# undef readahead |
877 |
# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
878 |
|
879 |
static ssize_t |
880 |
eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
881 |
{ |
882 |
size_t todo = count; |
883 |
dBUF; |
884 |
|
885 |
while (todo > 0) |
886 |
{ |
887 |
size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; |
888 |
|
889 |
pread (fd, eio_buf, len, offset); |
890 |
offset += len; |
891 |
todo -= len; |
892 |
} |
893 |
|
894 |
errno = 0; |
895 |
return count; |
896 |
} |
897 |
|
898 |
#endif |
899 |
|
900 |
/* sendfile always needs emulation */ |
901 |
static ssize_t |
902 |
eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
903 |
{ |
904 |
ssize_t res; |
905 |
|
906 |
if (!count) |
907 |
return 0; |
908 |
|
909 |
#if HAVE_SENDFILE |
910 |
# if __linux |
911 |
res = sendfile (ofd, ifd, &offset, count); |
912 |
|
913 |
# elif __freebsd |
914 |
/* |
915 |
* Of course, the freebsd sendfile is a dire hack with no thoughts |
916 |
* wasted on making it similar to other I/O functions. |
917 |
*/ |
918 |
{ |
919 |
off_t sbytes; |
920 |
res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
921 |
|
922 |
if (res < 0 && sbytes) |
923 |
/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
924 |
res = sbytes; |
925 |
} |
926 |
|
927 |
# elif __hpux |
928 |
res = sendfile (ofd, ifd, offset, count, 0, 0); |
929 |
|
930 |
# elif __solaris |
931 |
{ |
932 |
struct sendfilevec vec; |
933 |
size_t sbytes; |
934 |
|
935 |
vec.sfv_fd = ifd; |
936 |
vec.sfv_flag = 0; |
937 |
vec.sfv_off = offset; |
938 |
vec.sfv_len = count; |
939 |
|
940 |
res = sendfilev (ofd, &vec, 1, &sbytes); |
941 |
|
942 |
if (res < 0 && sbytes) |
943 |
res = sbytes; |
944 |
} |
945 |
|
946 |
# endif |
947 |
#else |
948 |
res = -1; |
949 |
errno = ENOSYS; |
950 |
#endif |
951 |
|
952 |
if (res < 0 |
953 |
&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
954 |
#if __solaris |
955 |
|| errno == EAFNOSUPPORT || errno == EPROTOTYPE |
956 |
#endif |
957 |
) |
958 |
) |
959 |
{ |
960 |
/* emulate sendfile. this is a major pain in the ass */ |
961 |
dBUF; |
962 |
|
963 |
res = 0; |
964 |
|
965 |
while (count) |
966 |
{ |
967 |
ssize_t cnt; |
968 |
|
969 |
cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); |
970 |
|
971 |
if (cnt <= 0) |
972 |
{ |
973 |
if (cnt && !res) res = -1; |
974 |
break; |
975 |
} |
976 |
|
977 |
cnt = write (ofd, eio_buf, cnt); |
978 |
|
979 |
if (cnt <= 0) |
980 |
{ |
981 |
if (cnt && !res) res = -1; |
982 |
break; |
983 |
} |
984 |
|
985 |
offset += cnt; |
986 |
res += cnt; |
987 |
count -= cnt; |
988 |
} |
989 |
} |
990 |
|
991 |
return res; |
992 |
} |
993 |
|
994 |
static int |
995 |
eio_dent_cmp (const void *a_, const void *b_) |
996 |
{ |
997 |
const eio_dirent *a = (const eio_dirent *)a_; |
998 |
const eio_dirent *b = (const eio_dirent *)b_; |
999 |
|
1000 |
return (int)b->score - (int)a->score ? (int)b->score - (int)a->score |
1001 |
: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */ |
1002 |
} |
1003 |
|
1004 |
/* read a full directory */ |
1005 |
static void |
1006 |
eio__scandir (eio_req *req, etp_worker *self) |
1007 |
{ |
1008 |
DIR *dirp; |
1009 |
EIO_STRUCT_DIRENT *entp; |
1010 |
unsigned char *name, *names; |
1011 |
int namesalloc = 4096; |
1012 |
int namesoffs = 0; |
1013 |
int flags = req->int1; |
1014 |
eio_dirent *dents = 0; |
1015 |
int dentalloc = 128; |
1016 |
int dentoffs = 0; |
1017 |
|
1018 |
req->result = -1; |
1019 |
|
1020 |
if (!(flags & EIO_READDIR_DENTS)) |
1021 |
flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1022 |
|
1023 |
X_LOCK (wrklock); |
1024 |
/* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1025 |
self->dirp = dirp = opendir (req->ptr1); |
1026 |
req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1027 |
req->ptr1 = names = malloc (namesalloc); |
1028 |
req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1029 |
X_UNLOCK (wrklock); |
1030 |
|
1031 |
if (dirp && names && (!flags || dents)) |
1032 |
for (;;) |
1033 |
{ |
1034 |
errno = 0; |
1035 |
entp = readdir (dirp); |
1036 |
|
1037 |
if (!entp) |
1038 |
{ |
1039 |
if (errno) |
1040 |
break; |
1041 |
|
1042 |
/* sort etc. */ |
1043 |
req->int1 = flags; |
1044 |
req->result = dentoffs; |
1045 |
|
1046 |
if (dents) |
1047 |
{ |
1048 |
eio_dirent *ent = dents + dentoffs; |
1049 |
|
1050 |
while (ent > dents) |
1051 |
(--ent)->name = names + (size_t)ent->name; |
1052 |
} |
1053 |
|
1054 |
if (flags & EIO_READDIR_STAT_ORDER |
1055 |
|| !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) |
1056 |
{ |
1057 |
/* pray your qsort doesn't use quicksort */ |
1058 |
qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */ |
1059 |
} |
1060 |
else if (flags & EIO_READDIR_DIRS_FIRST) |
1061 |
{ |
1062 |
/* in this case, all is known, and we just put dirs first and sort them */ |
1063 |
eio_dirent *ent = dents + dentoffs; |
1064 |
eio_dirent *dir = dents; |
1065 |
|
1066 |
/* now move dirs to the front, and non-dirs to the back */ |
1067 |
/* by walkign from both sides and swapping if necessary */ |
1068 |
while (ent > dir) |
1069 |
{ |
1070 |
if (dir->type == DT_DIR) |
1071 |
++dir; |
1072 |
else |
1073 |
{ |
1074 |
--ent; |
1075 |
|
1076 |
if (ent->type == DT_DIR) |
1077 |
{ |
1078 |
eio_dirent tmp = *dir; |
1079 |
*dir = *ent; |
1080 |
*ent = tmp; |
1081 |
|
1082 |
++dir; |
1083 |
} |
1084 |
} |
1085 |
} |
1086 |
|
1087 |
/* now sort the dirs only */ |
1088 |
qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp); |
1089 |
} |
1090 |
|
1091 |
/* only provide the names array unless DENTS is specified */ |
1092 |
if (!(flags & EIO_READDIR_DENTS)) |
1093 |
{ |
1094 |
X_LOCK (wrklock); |
1095 |
assert (!dents); |
1096 |
req->ptr1 = 0; |
1097 |
req->ptr2 = names; |
1098 |
X_UNLOCK (wrklock); |
1099 |
} |
1100 |
|
1101 |
break; |
1102 |
} |
1103 |
|
1104 |
/* now add the entry to our list(s) */ |
1105 |
name = entp->d_name; |
1106 |
|
1107 |
/* skip . and .. entries */ |
1108 |
if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1109 |
{ |
1110 |
int len = D_NAMLEN (entp) + 1; |
1111 |
|
1112 |
while (expect_false (namesoffs + len > namesalloc)) |
1113 |
{ |
1114 |
namesalloc *= 2; |
1115 |
X_LOCK (wrklock); |
1116 |
req->ptr1 = names = realloc (names, namesalloc); |
1117 |
X_UNLOCK (wrklock); |
1118 |
|
1119 |
if (!names) |
1120 |
break; |
1121 |
} |
1122 |
|
1123 |
memcpy (names + namesoffs, name, len); |
1124 |
|
1125 |
if (dents) |
1126 |
{ |
1127 |
struct eio_dirent *ent; |
1128 |
|
1129 |
if (expect_false (dentoffs == dentalloc)) |
1130 |
{ |
1131 |
dentalloc *= 2; |
1132 |
X_LOCK (wrklock); |
1133 |
req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1134 |
X_UNLOCK (wrklock); |
1135 |
|
1136 |
if (!dents) |
1137 |
break; |
1138 |
} |
1139 |
|
1140 |
ent = dents + dentoffs; |
1141 |
|
1142 |
ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ |
1143 |
ent->namelen = len - 1; |
1144 |
ent->inode = D_INO (entp); |
1145 |
|
1146 |
switch (D_TYPE (entp)) |
1147 |
{ |
1148 |
default: |
1149 |
ent->type = EIO_DT_UNKNOWN; |
1150 |
flags |= EIO_READDIR_FOUND_UNKNOWN; |
1151 |
break; |
1152 |
|
1153 |
#ifdef DT_FIFO |
1154 |
case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
1155 |
#endif |
1156 |
#ifdef DT_CHR |
1157 |
case DT_CHR: ent->type = EIO_DT_CHR; break; |
1158 |
#endif |
1159 |
#ifdef DT_MPC |
1160 |
case DT_MPC: ent->type = EIO_DT_MPC; break; |
1161 |
#endif |
1162 |
#ifdef DT_DIR |
1163 |
case DT_DIR: ent->type = EIO_DT_DIR; break; |
1164 |
#endif |
1165 |
#ifdef DT_NAM |
1166 |
case DT_NAM: ent->type = EIO_DT_NAM; break; |
1167 |
#endif |
1168 |
#ifdef DT_BLK |
1169 |
case DT_BLK: ent->type = EIO_DT_BLK; break; |
1170 |
#endif |
1171 |
#ifdef DT_MPB |
1172 |
case DT_MPB: ent->type = EIO_DT_MPB; break; |
1173 |
#endif |
1174 |
#ifdef DT_REG |
1175 |
case DT_REG: ent->type = EIO_DT_REG; break; |
1176 |
#endif |
1177 |
#ifdef DT_NWK |
1178 |
case DT_NWK: ent->type = EIO_DT_NWK; break; |
1179 |
#endif |
1180 |
#ifdef DT_CMP |
1181 |
case DT_CMP: ent->type = EIO_DT_CMP; break; |
1182 |
#endif |
1183 |
#ifdef DT_LNK |
1184 |
case DT_LNK: ent->type = EIO_DT_LNK; break; |
1185 |
#endif |
1186 |
#ifdef DT_SOCK |
1187 |
case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
1188 |
#endif |
1189 |
#ifdef DT_DOOR |
1190 |
case DT_DOOR: ent->type = EIO_DT_DOOR; break; |
1191 |
#endif |
1192 |
#ifdef DT_WHT |
1193 |
case DT_WHT: ent->type = EIO_DT_WHT; break; |
1194 |
#endif |
1195 |
} |
1196 |
|
1197 |
ent->score = 0; |
1198 |
|
1199 |
if (flags & EIO_READDIR_DIRS_FIRST) |
1200 |
{ |
1201 |
if (ent->type == EIO_DT_UNKNOWN) |
1202 |
{ |
1203 |
if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
1204 |
ent->score = 98; |
1205 |
else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
1206 |
ent->score = len <= 2 ? len + 6 : len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */ |
1207 |
} |
1208 |
else if (ent->type == EIO_DT_DIR) |
1209 |
ent->score = 100; |
1210 |
} |
1211 |
} |
1212 |
|
1213 |
namesoffs += len; |
1214 |
++dentoffs; |
1215 |
} |
1216 |
|
1217 |
if (EIO_CANCELLED (req)) |
1218 |
{ |
1219 |
errno = ECANCELED; |
1220 |
break; |
1221 |
} |
1222 |
} |
1223 |
} |
1224 |
|
1225 |
#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1226 |
# undef msync |
1227 |
# define msync(a,b,c) ((errno = ENOSYS), -1) |
1228 |
#endif |
1229 |
|
1230 |
int |
1231 |
eio__mtouch (void *mem, size_t len, int flags) |
1232 |
{ |
1233 |
intptr_t addr = (intptr_t)mem; |
1234 |
intptr_t end = addr + len; |
1235 |
#ifdef PAGESIZE |
1236 |
const intptr_t page = PAGESIZE; |
1237 |
#else |
1238 |
static intptr_t page; |
1239 |
|
1240 |
if (!page) |
1241 |
page = sysconf (_SC_PAGESIZE); |
1242 |
#endif |
1243 |
|
1244 |
addr &= ~(page - 1); /* assume page size is always a power of two */ |
1245 |
|
1246 |
if (addr < end) |
1247 |
if (flags) /* modify */ |
1248 |
do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1249 |
else |
1250 |
do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1251 |
|
1252 |
return 0; |
1253 |
} |
1254 |
|
1255 |
/*****************************************************************************/ |
1256 |
|
1257 |
#define ALLOC(len) \ |
1258 |
if (!req->ptr2) \ |
1259 |
{ \ |
1260 |
X_LOCK (wrklock); \ |
1261 |
req->flags |= EIO_FLAG_PTR2_FREE; \ |
1262 |
X_UNLOCK (wrklock); \ |
1263 |
req->ptr2 = malloc (len); \ |
1264 |
if (!req->ptr2) \ |
1265 |
{ \ |
1266 |
errno = ENOMEM; \ |
1267 |
req->result = -1; \ |
1268 |
break; \ |
1269 |
} \ |
1270 |
} |
1271 |
|
1272 |
X_THREAD_PROC (etp_proc) |
1273 |
{ |
1274 |
ETP_REQ *req; |
1275 |
struct timespec ts; |
1276 |
etp_worker *self = (etp_worker *)thr_arg; |
1277 |
|
1278 |
/* try to distribute timeouts somewhat randomly */ |
1279 |
ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
1280 |
|
1281 |
for (;;) |
1282 |
{ |
1283 |
X_LOCK (reqlock); |
1284 |
|
1285 |
for (;;) |
1286 |
{ |
1287 |
self->req = req = reqq_shift (&req_queue); |
1288 |
|
1289 |
if (req) |
1290 |
break; |
1291 |
|
1292 |
++idle; |
1293 |
|
1294 |
ts.tv_sec = time (0) + IDLE_TIMEOUT; |
1295 |
if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1296 |
{ |
1297 |
if (idle > max_idle) |
1298 |
{ |
1299 |
--idle; |
1300 |
X_UNLOCK (reqlock); |
1301 |
X_LOCK (wrklock); |
1302 |
--started; |
1303 |
X_UNLOCK (wrklock); |
1304 |
goto quit; |
1305 |
} |
1306 |
|
1307 |
/* we are allowed to idle, so do so without any timeout */ |
1308 |
X_COND_WAIT (reqwait, reqlock); |
1309 |
} |
1310 |
|
1311 |
--idle; |
1312 |
} |
1313 |
|
1314 |
--nready; |
1315 |
|
1316 |
X_UNLOCK (reqlock); |
1317 |
|
1318 |
if (req->type < 0) |
1319 |
goto quit; |
1320 |
|
1321 |
if (!EIO_CANCELLED (req)) |
1322 |
ETP_EXECUTE (self, req); |
1323 |
|
1324 |
X_LOCK (reslock); |
1325 |
|
1326 |
++npending; |
1327 |
|
1328 |
if (!reqq_push (&res_queue, req) && want_poll_cb) |
1329 |
want_poll_cb (); |
1330 |
|
1331 |
self->req = 0; |
1332 |
etp_worker_clear (self); |
1333 |
|
1334 |
X_UNLOCK (reslock); |
1335 |
} |
1336 |
|
1337 |
quit: |
1338 |
X_LOCK (wrklock); |
1339 |
etp_worker_free (self); |
1340 |
X_UNLOCK (wrklock); |
1341 |
|
1342 |
return 0; |
1343 |
} |
1344 |
|
1345 |
/*****************************************************************************/ |
1346 |
|
1347 |
int eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1348 |
{ |
1349 |
return etp_init (want_poll, done_poll); |
1350 |
} |
1351 |
|
1352 |
static void eio_api_destroy (eio_req *req) |
1353 |
{ |
1354 |
free (req); |
1355 |
} |
1356 |
|
1357 |
#define REQ(rtype) \ |
1358 |
eio_req *req; \ |
1359 |
\ |
1360 |
req = (eio_req *)calloc (1, sizeof *req); \ |
1361 |
if (!req) \ |
1362 |
return 0; \ |
1363 |
\ |
1364 |
req->type = rtype; \ |
1365 |
req->pri = pri; \ |
1366 |
req->finish = cb; \ |
1367 |
req->data = data; \ |
1368 |
req->destroy = eio_api_destroy; |
1369 |
|
1370 |
#define SEND eio_submit (req); return req |
1371 |
|
1372 |
#define PATH \ |
1373 |
req->flags |= EIO_FLAG_PTR1_FREE; \ |
1374 |
req->ptr1 = strdup (path); \ |
1375 |
if (!req->ptr1) \ |
1376 |
{ \ |
1377 |
eio_api_destroy (req); \ |
1378 |
return 0; \ |
1379 |
} |
1380 |
|
1381 |
static void eio_execute (etp_worker *self, eio_req *req) |
1382 |
{ |
1383 |
errno = 0; |
1384 |
|
1385 |
switch (req->type) |
1386 |
{ |
1387 |
case EIO_READ: ALLOC (req->size); |
1388 |
req->result = req->offs >= 0 |
1389 |
? pread (req->int1, req->ptr2, req->size, req->offs) |
1390 |
: read (req->int1, req->ptr2, req->size); break; |
1391 |
case EIO_WRITE: req->result = req->offs >= 0 |
1392 |
? pwrite (req->int1, req->ptr2, req->size, req->offs) |
1393 |
: write (req->int1, req->ptr2, req->size); break; |
1394 |
|
1395 |
case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; |
1396 |
case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; |
1397 |
|
1398 |
case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1399 |
req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1400 |
case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1401 |
req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1402 |
case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1403 |
req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1404 |
|
1405 |
case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1406 |
case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1407 |
case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1408 |
case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1409 |
case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
1410 |
case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; |
1411 |
|
1412 |
case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; |
1413 |
case EIO_CLOSE: req->result = close (req->int1); break; |
1414 |
case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; |
1415 |
case EIO_UNLINK: req->result = unlink (req->ptr1); break; |
1416 |
case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1417 |
case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1418 |
case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1419 |
case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1420 |
case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1421 |
case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
1422 |
|
1423 |
case EIO_READLINK: ALLOC (NAME_MAX); |
1424 |
req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1425 |
|
1426 |
case EIO_SYNC: req->result = 0; sync (); break; |
1427 |
case EIO_FSYNC: req->result = fsync (req->int1); break; |
1428 |
case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1429 |
case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
1430 |
case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1431 |
case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1432 |
|
1433 |
case EIO_READDIR: eio__scandir (req, self); break; |
1434 |
|
1435 |
case EIO_BUSY: |
1436 |
#ifdef _WIN32 |
1437 |
Sleep (req->nv1 * 1000.); |
1438 |
#else |
1439 |
{ |
1440 |
struct timeval tv; |
1441 |
|
1442 |
tv.tv_sec = req->nv1; |
1443 |
tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
1444 |
|
1445 |
req->result = select (0, 0, 0, 0, &tv); |
1446 |
} |
1447 |
#endif |
1448 |
break; |
1449 |
|
1450 |
case EIO_UTIME: |
1451 |
case EIO_FUTIME: |
1452 |
{ |
1453 |
struct timeval tv[2]; |
1454 |
struct timeval *times; |
1455 |
|
1456 |
if (req->nv1 != -1. || req->nv2 != -1.) |
1457 |
{ |
1458 |
tv[0].tv_sec = req->nv1; |
1459 |
tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; |
1460 |
tv[1].tv_sec = req->nv2; |
1461 |
tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; |
1462 |
|
1463 |
times = tv; |
1464 |
} |
1465 |
else |
1466 |
times = 0; |
1467 |
|
1468 |
|
1469 |
req->result = req->type == EIO_FUTIME |
1470 |
? futimes (req->int1, times) |
1471 |
: utimes (req->ptr1, times); |
1472 |
} |
1473 |
break; |
1474 |
|
1475 |
case EIO_GROUP: |
1476 |
abort (); /* handled in eio_request */ |
1477 |
|
1478 |
case EIO_NOP: |
1479 |
req->result = 0; |
1480 |
break; |
1481 |
|
1482 |
case EIO_CUSTOM: |
1483 |
((void (*)(eio_req *))req->feed) (req); |
1484 |
break; |
1485 |
|
1486 |
default: |
1487 |
req->result = -1; |
1488 |
break; |
1489 |
} |
1490 |
|
1491 |
req->errorno = errno; |
1492 |
} |
1493 |
|
1494 |
#ifndef EIO_NO_WRAPPERS |
1495 |
|
1496 |
eio_req *eio_nop (int pri, eio_cb cb, void *data) |
1497 |
{ |
1498 |
REQ (EIO_NOP); SEND; |
1499 |
} |
1500 |
|
1501 |
eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data) |
1502 |
{ |
1503 |
REQ (EIO_BUSY); req->nv1 = delay; SEND; |
1504 |
} |
1505 |
|
1506 |
eio_req *eio_sync (int pri, eio_cb cb, void *data) |
1507 |
{ |
1508 |
REQ (EIO_SYNC); SEND; |
1509 |
} |
1510 |
|
1511 |
eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1512 |
{ |
1513 |
REQ (EIO_FSYNC); req->int1 = fd; SEND; |
1514 |
} |
1515 |
|
1516 |
eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1517 |
{ |
1518 |
REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1519 |
} |
1520 |
|
1521 |
eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1522 |
{ |
1523 |
REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1524 |
} |
1525 |
|
1526 |
eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
1527 |
{ |
1528 |
REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
1529 |
} |
1530 |
|
1531 |
eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1532 |
{ |
1533 |
REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1534 |
} |
1535 |
|
1536 |
eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) |
1537 |
{ |
1538 |
REQ (EIO_CLOSE); req->int1 = fd; SEND; |
1539 |
} |
1540 |
|
1541 |
eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) |
1542 |
{ |
1543 |
REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; |
1544 |
} |
1545 |
|
1546 |
eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) |
1547 |
{ |
1548 |
REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; |
1549 |
} |
1550 |
|
1551 |
eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) |
1552 |
{ |
1553 |
REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; |
1554 |
} |
1555 |
|
1556 |
eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1557 |
{ |
1558 |
REQ (EIO_FSTAT); req->int1 = fd; SEND; |
1559 |
} |
1560 |
|
1561 |
eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1562 |
{ |
1563 |
REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
1564 |
} |
1565 |
|
1566 |
eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data) |
1567 |
{ |
1568 |
REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; |
1569 |
} |
1570 |
|
1571 |
eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) |
1572 |
{ |
1573 |
REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; |
1574 |
} |
1575 |
|
1576 |
eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) |
1577 |
{ |
1578 |
REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
1579 |
} |
1580 |
|
1581 |
eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) |
1582 |
{ |
1583 |
REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND; |
1584 |
} |
1585 |
|
1586 |
eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data) |
1587 |
{ |
1588 |
REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; |
1589 |
} |
1590 |
|
1591 |
eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data) |
1592 |
{ |
1593 |
REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; |
1594 |
} |
1595 |
|
1596 |
eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data) |
1597 |
{ |
1598 |
REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; |
1599 |
} |
1600 |
|
1601 |
eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) |
1602 |
{ |
1603 |
REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; |
1604 |
} |
1605 |
|
1606 |
eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) |
1607 |
{ |
1608 |
REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
1609 |
} |
1610 |
|
1611 |
eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) |
1612 |
{ |
1613 |
REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND; |
1614 |
} |
1615 |
|
1616 |
eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data) |
1617 |
{ |
1618 |
REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND; |
1619 |
} |
1620 |
|
1621 |
static eio_req * |
1622 |
eio__1path (int type, const char *path, int pri, eio_cb cb, void *data) |
1623 |
{ |
1624 |
REQ (type); PATH; SEND; |
1625 |
} |
1626 |
|
1627 |
eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) |
1628 |
{ |
1629 |
return eio__1path (EIO_READLINK, path, pri, cb, data); |
1630 |
} |
1631 |
|
1632 |
eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) |
1633 |
{ |
1634 |
return eio__1path (EIO_STAT, path, pri, cb, data); |
1635 |
} |
1636 |
|
1637 |
eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1638 |
{ |
1639 |
return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1640 |
} |
1641 |
|
1642 |
eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1643 |
{ |
1644 |
return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1645 |
} |
1646 |
|
1647 |
eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
1648 |
{ |
1649 |
return eio__1path (EIO_RMDIR, path, pri, cb, data); |
1650 |
} |
1651 |
|
1652 |
eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data) |
1653 |
{ |
1654 |
REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1655 |
} |
1656 |
|
1657 |
eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1658 |
{ |
1659 |
REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
1660 |
} |
1661 |
|
1662 |
static eio_req * |
1663 |
eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1664 |
{ |
1665 |
REQ (type); PATH; |
1666 |
|
1667 |
req->flags |= EIO_FLAG_PTR2_FREE; |
1668 |
req->ptr2 = strdup (new_path); |
1669 |
if (!req->ptr2) |
1670 |
{ |
1671 |
eio_api_destroy (req); |
1672 |
return 0; |
1673 |
} |
1674 |
|
1675 |
SEND; |
1676 |
} |
1677 |
|
1678 |
eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1679 |
{ |
1680 |
return eio__2path (EIO_LINK, path, new_path, pri, cb, data); |
1681 |
} |
1682 |
|
1683 |
eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1684 |
{ |
1685 |
return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data); |
1686 |
} |
1687 |
|
1688 |
eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1689 |
{ |
1690 |
return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1691 |
} |
1692 |
|
1693 |
eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
1694 |
{ |
1695 |
REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; |
1696 |
} |
1697 |
|
1698 |
#endif |
1699 |
|
1700 |
eio_req *eio_grp (eio_cb cb, void *data) |
1701 |
{ |
1702 |
const int pri = EIO_PRI_MAX; |
1703 |
|
1704 |
REQ (EIO_GROUP); SEND; |
1705 |
} |
1706 |
|
1707 |
#undef REQ |
1708 |
#undef PATH |
1709 |
#undef SEND |
1710 |
|
1711 |
/*****************************************************************************/ |
1712 |
/* grp functions */ |
1713 |
|
1714 |
void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) |
1715 |
{ |
1716 |
grp->int2 = limit; |
1717 |
grp->feed = feed; |
1718 |
|
1719 |
grp_try_feed (grp); |
1720 |
} |
1721 |
|
1722 |
void eio_grp_limit (eio_req *grp, int limit) |
1723 |
{ |
1724 |
grp->int2 = limit; |
1725 |
|
1726 |
grp_try_feed (grp); |
1727 |
} |
1728 |
|
1729 |
void eio_grp_add (eio_req *grp, eio_req *req) |
1730 |
{ |
1731 |
assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
1732 |
|
1733 |
grp->flags |= EIO_FLAG_GROUPADD; |
1734 |
|
1735 |
++grp->size; |
1736 |
req->grp = grp; |
1737 |
|
1738 |
req->grp_prev = 0; |
1739 |
req->grp_next = grp->grp_first; |
1740 |
|
1741 |
if (grp->grp_first) |
1742 |
grp->grp_first->grp_prev = req; |
1743 |
|
1744 |
grp->grp_first = req; |
1745 |
} |
1746 |
|
1747 |
/*****************************************************************************/ |
1748 |
/* misc garbage */ |
1749 |
|
1750 |
ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1751 |
{ |
1752 |
etp_worker wrk; |
1753 |
|
1754 |
wrk.dbuf = 0; |
1755 |
|
1756 |
eio__sendfile (ofd, ifd, offset, count, &wrk); |
1757 |
|
1758 |
if (wrk.dbuf) |
1759 |
free (wrk.dbuf); |
1760 |
} |
1761 |
|