ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libeio/eio.c
(Generate patch)

Comparing libeio/eio.c (file contents):
Revision 1.2 by root, Sat May 10 23:33:40 2008 UTC vs.
Revision 1.85 by root, Wed Jul 13 00:53:03 2011 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 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
1#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
2#include "xthread.h" 50#include "xthread.h"
3 51
4#include <errno.h> 52#include <errno.h>
5
6#include "EXTERN.h"
7#include "perl.h"
8#include "XSUB.h"
9
10#include <stddef.h> 53#include <stddef.h>
11#include <stdlib.h> 54#include <stdlib.h>
55#include <string.h>
12#include <errno.h> 56#include <errno.h>
13#include <sys/types.h> 57#include <sys/types.h>
14#include <sys/stat.h> 58#include <sys/stat.h>
59#include <sys/statvfs.h>
15#include <limits.h> 60#include <limits.h>
16#include <fcntl.h> 61#include <fcntl.h>
17#include <sched.h> 62#include <assert.h>
63
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H
67# include <stdint.h>
68#endif
69
70#ifndef ECANCELED
71# define ECANCELED EDOM
72#endif
73
74static void eio_destroy (eio_req *req);
18 75
19#ifndef EIO_FINISH 76#ifndef EIO_FINISH
20# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 77# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
21#endif 78#endif
22 79
… …
29#endif 86#endif
30 87
31#ifdef _WIN32 88#ifdef _WIN32
32 89
33 /*doh*/ 90 /*doh*/
34
35#else 91#else
36 92
37# include "config.h"
38# include <sys/time.h> 93# include <sys/time.h>
39# include <sys/select.h> 94# include <sys/select.h>
40# include <unistd.h> 95# include <unistd.h>
41# include <utime.h> 96# include <utime.h>
42# include <signal.h> 97# include <signal.h>
98# include <dirent.h>
99
100#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
101# include <sys/mman.h>
102#endif
103
104/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
105# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
106# define _DIRENT_HAVE_D_TYPE /* sigh */
107# define D_INO(de) (de)->d_fileno
108# define D_NAMLEN(de) (de)->d_namlen
109# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
110# define D_INO(de) (de)->d_ino
111# endif
112
113#ifdef _D_EXACT_NAMLEN
114# undef D_NAMLEN
115# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
116#endif
117
118# ifdef _DIRENT_HAVE_D_TYPE
119# define D_TYPE(de) (de)->d_type
120# endif
43 121
44# ifndef EIO_STRUCT_DIRENT 122# ifndef EIO_STRUCT_DIRENT
45# define EIO_STRUCT_DIRENT struct dirent 123# define EIO_STRUCT_DIRENT struct dirent
46# endif 124# endif
47 125
48#endif 126#endif
49 127
50# ifndef EIO_STRUCT_STAT
51# define EIO_STRUCT_STAT struct stat
52# endif
53
54#if HAVE_SENDFILE 128#if HAVE_SENDFILE
55# if __linux 129# if __linux
56# include <sys/sendfile.h> 130# include <sys/sendfile.h>
57# elif __freebsd 131# elif __FreeBSD__ || defined __APPLE__
58# include <sys/socket.h> 132# include <sys/socket.h>
59# include <sys/uio.h> 133# include <sys/uio.h>
60# elif __hpux 134# elif __hpux
61# include <sys/socket.h> 135# include <sys/socket.h>
62# elif __solaris /* not yet */ 136# elif __solaris
63# include <sys/sendfile.h> 137# include <sys/sendfile.h>
64# else 138# else
65# error sendfile support requested but not available 139# error sendfile support requested but not available
66# endif 140# endif
67#endif 141#endif
68 142
69/* number of seconds after which an idle threads exit */ 143#ifndef D_TYPE
70#define IDLE_TIMEOUT 10 144# define D_TYPE(de) 0
145#endif
146#ifndef D_INO
147# define D_INO(de) 0
148#endif
149#ifndef D_NAMLEN
150# define D_NAMLEN(de) strlen ((de)->d_name)
151#endif
71 152
72/* used for struct dirent, AIX doesn't provide it */ 153/* used for struct dirent, AIX doesn't provide it */
73#ifndef NAME_MAX 154#ifndef NAME_MAX
74# define NAME_MAX 4096 155# define NAME_MAX 4096
75#endif 156#endif
76 157
158/* used for readlink etc. */
159#ifndef PATH_MAX
160# define PATH_MAX 4096
161#endif
162
77/* buffer size for various temporary buffers */ 163/* buffer size for various temporary buffers */
78#define EIO_BUFSIZE 65536 164#define EIO_BUFSIZE 65536
79 165
80#define dBUF \ 166#define dBUF \
81 char *eio_buf; \ 167 char *eio_buf; \
82 X_LOCK (wrklock); \ 168 ETP_WORKER_LOCK (self); \
83 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 169 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
84 X_UNLOCK (wrklock); \ 170 ETP_WORKER_UNLOCK (self); \
171 errno = ENOMEM; \
85 if (!eio_buf) \ 172 if (!eio_buf) \
86 return -1; 173 return -1;
87 174
88#define EIO_TICKS ((1000000 + 1023) >> 10) 175#define EIO_TICKS ((1000000 + 1023) >> 10)
89 176
177#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX
179
180struct etp_worker;
181
182#define ETP_REQ eio_req
183#define ETP_DESTROY(req) eio_destroy (req)
184static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
188
189#define ETP_WORKER_CLEAR(req) \
190 if (wrk->dbuf) \
191 { \
192 free (wrk->dbuf); \
193 wrk->dbuf = 0; \
194 } \
195 \
196 if (wrk->dirp) \
197 { \
198 closedir (wrk->dirp); \
199 wrk->dirp = 0; \
200 }
201
202#define ETP_WORKER_COMMON \
203 void *dbuf; \
204 DIR *dirp;
205
206/*****************************************************************************/
207
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209
210/* calculate time difference in ~1/EIO_TICKS of a second */
211ecb_inline int
212tvdiff (struct timeval *tv1, struct timeval *tv2)
213{
214 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
215 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
216}
217
218static unsigned int started, idle, wanted = 4;
219
90static void (*want_poll_cb) (void); 220static void (*want_poll_cb) (void);
91static void (*done_poll_cb) (void); 221static void (*done_poll_cb) (void);
92 222
93static unsigned int max_poll_time = 0; 223static unsigned int max_poll_time; /* reslock */
94static unsigned int max_poll_reqs = 0; 224static unsigned int max_poll_reqs; /* reslock */
95 225
96/* calculcate time difference in ~1/EIO_TICKS of a second */ 226static volatile unsigned int nreqs; /* reqlock */
97static int tvdiff (struct timeval *tv1, struct timeval *tv2) 227static volatile unsigned int nready; /* reqlock */
98{ 228static volatile unsigned int npending; /* reqlock */
99 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 229static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
100 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 230static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
101}
102 231
103static unsigned int started, idle, wanted; 232static xmutex_t wrklock;
233static xmutex_t reslock;
234static xmutex_t reqlock;
235static xcond_t reqwait;
236
237#if !HAVE_PREADWRITE
238/*
239 * make our pread/pwrite emulation safe against themselves, but not against
240 * normal read/write by using a mutex. slows down execution a lot,
241 * but that's your problem, not mine.
242 */
243static xmutex_t preadwritelock = X_MUTEX_INIT;
244#endif
245
246typedef struct etp_worker
247{
248 /* locked by wrklock */
249 struct etp_worker *prev, *next;
250
251 xthread_t tid;
252
253 /* locked by reslock, reqlock or wrklock */
254 ETP_REQ *req; /* currently processed request */
255
256 ETP_WORKER_COMMON
257} etp_worker;
258
259static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
260
261#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
262#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
104 263
105/* worker threads management */ 264/* worker threads management */
106static mutex_t wrklock = X_MUTEX_INIT;
107 265
108typedef struct worker { 266static void ecb_cold
109 /* locked by wrklock */
110 struct worker *prev, *next;
111
112 thread_t tid;
113
114 /* locked by reslock, reqlock or wrklock */
115 eio_req *req; /* currently processed request */
116 void *dbuf;
117 DIR *dirp;
118} worker;
119
120static worker wrk_first = { &wrk_first, &wrk_first, 0 };
121
122static void worker_clear (worker *wrk) 267etp_worker_clear (etp_worker *wrk)
123{ 268{
124 if (wrk->dirp) 269 ETP_WORKER_CLEAR (wrk);
125 {
126 closedir (wrk->dirp);
127 wrk->dirp = 0;
128 }
129
130 if (wrk->dbuf)
131 {
132 free (wrk->dbuf);
133 wrk->dbuf = 0;
134 }
135} 270}
136 271
137static void worker_free (worker *wrk) 272static void ecb_cold
273etp_worker_free (etp_worker *wrk)
138{ 274{
139 wrk->next->prev = wrk->prev; 275 wrk->next->prev = wrk->prev;
140 wrk->prev->next = wrk->next; 276 wrk->prev->next = wrk->next;
141 277
142 free (wrk); 278 free (wrk);
143} 279}
144 280
145static volatile unsigned int nreqs, nready, npending; 281static unsigned int
146static volatile unsigned int max_idle = 4; 282etp_nreqs (void)
147
148static mutex_t reslock = X_MUTEX_INIT;
149static mutex_t reqlock = X_MUTEX_INIT;
150static cond_t reqwait = X_COND_INIT;
151
152unsigned int eio_nreqs (void)
153{ 283{
284 int retval;
285 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
286 retval = nreqs;
287 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 288 return retval;
155} 289}
156 290
157unsigned int eio_nready (void) 291static unsigned int
292etp_nready (void)
158{ 293{
159 unsigned int retval; 294 unsigned int retval;
160 295
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 296 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 297 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 298 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 299
165 return retval; 300 return retval;
166} 301}
167 302
168unsigned int eio_npending (void) 303static unsigned int
304etp_npending (void)
169{ 305{
170 unsigned int retval; 306 unsigned int retval;
171 307
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 308 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 309 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 310 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 311
176 return retval; 312 return retval;
177} 313}
178 314
179unsigned int eio_nthreads (void) 315static unsigned int
316etp_nthreads (void)
180{ 317{
181 unsigned int retval; 318 unsigned int retval;
182 319
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 320 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 321 retval = started;
… …
191 * a somewhat faster data structure might be nice, but 328 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 329 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 330 * per shift, the most expensive operation.
194 */ 331 */
195typedef struct { 332typedef struct {
196 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 333 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197 int size; 334 int size;
198} reqq; 335} etp_reqq;
199 336
200static reqq req_queue; 337static etp_reqq req_queue;
201static reqq res_queue; 338static etp_reqq res_queue;
202 339
203static int reqq_push (reqq *q, eio_req *req) 340static int ecb_noinline
341reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 342{
205 int pri = req->pri; 343 int pri = req->pri;
206 req->next = 0; 344 req->next = 0;
207 345
208 if (q->qe[pri]) 346 if (q->qe[pri])
… …
214 q->qe[pri] = q->qs[pri] = req; 352 q->qe[pri] = q->qs[pri] = req;
215 353
216 return q->size++; 354 return q->size++;
217} 355}
218 356
219static eio_req *reqq_shift (reqq *q) 357static ETP_REQ * ecb_noinline
358reqq_shift (etp_reqq *q)
220{ 359{
221 int pri; 360 int pri;
222 361
223 if (!q->size) 362 if (!q->size)
224 return 0; 363 return 0;
225 364
226 --q->size; 365 --q->size;
227 366
228 for (pri = EIO_NUM_PRI; pri--; ) 367 for (pri = ETP_NUM_PRI; pri--; )
229 { 368 {
230 eio_req *req = q->qs[pri]; 369 eio_req *req = q->qs[pri];
231 370
232 if (req) 371 if (req)
233 { 372 {
… …
239 } 378 }
240 379
241 abort (); 380 abort ();
242} 381}
243 382
244static void grp_try_feed (eio_req *grp) 383static void ecb_cold
384etp_thread_init (void)
245{ 385{
246 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 386#if !HAVE_PREADWRITE
387 X_MUTEX_CREATE (preadwritelock);
388#endif
389 X_MUTEX_CREATE (wrklock);
390 X_MUTEX_CREATE (reslock);
391 X_MUTEX_CREATE (reqlock);
392 X_COND_CREATE (reqwait);
393}
394
395static void ecb_cold
396etp_atfork_prepare (void)
397{
398}
399
400static void ecb_cold
401etp_atfork_parent (void)
402{
403}
404
405static void ecb_cold
406etp_atfork_child (void)
407{
408 ETP_REQ *prv;
409
410 while ((prv = reqq_shift (&req_queue)))
411 ETP_DESTROY (prv);
412
413 while ((prv = reqq_shift (&res_queue)))
414 ETP_DESTROY (prv);
415
416 while (wrk_first.next != &wrk_first)
247 { 417 {
248 int old_len = grp->size; 418 etp_worker *wrk = wrk_first.next;
249 419
250 EIO_FEED (grp); 420 if (wrk->req)
421 ETP_DESTROY (wrk->req);
251 422
252 /* stop if no progress has been made */ 423 etp_worker_clear (wrk);
253 if (old_len == grp->size) 424 etp_worker_free (wrk);
254 {
255 grp->feed = 0;
256 break;
257 }
258 } 425 }
259}
260 426
261static int eio_finish (eio_req *req); 427 started = 0;
428 idle = 0;
429 nreqs = 0;
430 nready = 0;
431 npending = 0;
262 432
263static int grp_dec (eio_req *grp) 433 etp_thread_init ();
264{ 434}
265 --grp->size;
266 435
267 /* call feeder, if applicable */ 436static void ecb_cold
268 grp_try_feed (grp); 437etp_once_init (void)
438{
439 etp_thread_init ();
440 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
441}
269 442
270 /* finish, if done */ 443static int ecb_cold
271 if (!grp->size && grp->int1) 444etp_init (void (*want_poll)(void), void (*done_poll)(void))
272 return eio_finish (grp); 445{
273 else 446 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
447
448 pthread_once (&doinit, etp_once_init);
449
450 want_poll_cb = want_poll;
451 done_poll_cb = done_poll;
452
274 return 0; 453 return 0;
275} 454}
276 455
277void eio_destroy (eio_req *req)
278{
279 if ((req)->flags & EIO_FLAG_PTR2_FREE)
280 free (req->ptr2);
281
282 EIO_DESTROY (req);
283}
284
285static int eio_finish (eio_req *req)
286{
287 int res = EIO_FINISH (req);
288
289 if (req->grp)
290 {
291 int res2;
292 eio_req *grp = req->grp;
293
294 /* unlink request */
295 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
296 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
297
298 if (grp->grp_first == req)
299 grp->grp_first = req->grp_next;
300
301 res2 = grp_dec (grp);
302
303 if (!res && res2)
304 res = res2;
305 }
306
307 eio_destroy (req);
308
309 return res;
310}
311
312void eio_grp_cancel (eio_req *grp)
313{
314 for (grp = grp->grp_first; grp; grp = grp->grp_next)
315 eio_cancel (grp);
316}
317
318void eio_cancel (eio_req *req)
319{
320 req->flags |= EIO_FLAG_CANCELLED;
321
322 eio_grp_cancel (req);
323}
324
325X_THREAD_PROC (eio_proc); 456X_THREAD_PROC (etp_proc);
326 457
327static void start_thread (void) 458static void ecb_cold
459etp_start_thread (void)
328{ 460{
329 worker *wrk = calloc (1, sizeof (worker)); 461 etp_worker *wrk = calloc (1, sizeof (etp_worker));
330 462
331 if (!wrk) 463 /*TODO*/
332 croak ("unable to allocate worker thread data"); 464 assert (("unable to allocate worker thread data", wrk));
333 465
334 X_LOCK (wrklock); 466 X_LOCK (wrklock);
335 467
336 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 468 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
337 { 469 {
338 wrk->prev = &wrk_first; 470 wrk->prev = &wrk_first;
339 wrk->next = wrk_first.next; 471 wrk->next = wrk_first.next;
340 wrk_first.next->prev = wrk; 472 wrk_first.next->prev = wrk;
341 wrk_first.next = wrk; 473 wrk_first.next = wrk;
… …
345 free (wrk); 477 free (wrk);
346 478
347 X_UNLOCK (wrklock); 479 X_UNLOCK (wrklock);
348} 480}
349 481
482static void
350static void maybe_start_thread (void) 483etp_maybe_start_thread (void)
351{ 484{
352 if (eio_nthreads () >= wanted) 485 if (ecb_expect_true (etp_nthreads () >= wanted))
353 return; 486 return;
354 487
355 /* todo: maybe use idle here, but might be less exact */ 488 /* todo: maybe use idle here, but might be less exact */
356 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 489 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
357 return; 490 return;
358 491
359 start_thread (); 492 etp_start_thread ();
360} 493}
361 494
362void eio_submit (eio_req *req) 495static void ecb_cold
363{ 496etp_end_thread (void)
364 ++nreqs;
365
366 X_LOCK (reqlock);
367 ++nready;
368 reqq_push (&req_queue, req);
369 X_COND_SIGNAL (reqwait);
370 X_UNLOCK (reqlock);
371
372 maybe_start_thread ();
373}
374
375static void end_thread (void)
376{ 497{
377 eio_req *req = calloc (1, sizeof (eio_req)); 498 eio_req *req = calloc (1, sizeof (eio_req));
378 499
379 req->type = EIO_QUIT; 500 req->type = -1;
380 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 501 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381 502
382 X_LOCK (reqlock); 503 X_LOCK (reqlock);
383 reqq_push (&req_queue, req); 504 reqq_push (&req_queue, req);
384 X_COND_SIGNAL (reqwait); 505 X_COND_SIGNAL (reqwait);
385 X_UNLOCK (reqlock); 506 X_UNLOCK (reqlock);
… …
387 X_LOCK (wrklock); 508 X_LOCK (wrklock);
388 --started; 509 --started;
389 X_UNLOCK (wrklock); 510 X_UNLOCK (wrklock);
390} 511}
391 512
392void eio_set_max_poll_time (double nseconds) 513static int
393{
394 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
395 max_poll_time = nseconds;
396 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397}
398
399void eio_set_max_poll_reqs (unsigned int maxreqs)
400{
401 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
402 max_poll_reqs = maxreqs;
403 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404}
405
406void eio_set_max_idle (unsigned int nthreads)
407{
408 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
409 max_idle = nthreads <= 0 ? 1 : nthreads;
410 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411}
412
413void eio_set_min_parallel (unsigned int nthreads)
414{
415 if (wanted < nthreads)
416 wanted = nthreads;
417}
418
419void eio_set_max_parallel (unsigned int nthreads)
420{
421 if (wanted > nthreads)
422 wanted = nthreads;
423
424 while (started > wanted)
425 end_thread ();
426}
427
428int eio_poll (void) 514etp_poll (void)
429{ 515{
430 int maxreqs = max_poll_reqs; 516 unsigned int maxreqs;
517 unsigned int maxtime;
431 struct timeval tv_start, tv_now; 518 struct timeval tv_start, tv_now;
432 eio_req *req;
433 519
520 X_LOCK (reslock);
521 maxreqs = max_poll_reqs;
522 maxtime = max_poll_time;
523 X_UNLOCK (reslock);
524
434 if (max_poll_time) 525 if (maxtime)
435 gettimeofday (&tv_start, 0); 526 gettimeofday (&tv_start, 0);
436 527
437 for (;;) 528 for (;;)
438 { 529 {
530 ETP_REQ *req;
531
439 maybe_start_thread (); 532 etp_maybe_start_thread ();
440 533
441 X_LOCK (reslock); 534 X_LOCK (reslock);
442 req = reqq_shift (&res_queue); 535 req = reqq_shift (&res_queue);
443 536
444 if (req) 537 if (req)
… …
452 X_UNLOCK (reslock); 545 X_UNLOCK (reslock);
453 546
454 if (!req) 547 if (!req)
455 return 0; 548 return 0;
456 549
550 X_LOCK (reqlock);
457 --nreqs; 551 --nreqs;
552 X_UNLOCK (reqlock);
458 553
459 if (req->type == EIO_GROUP && req->size) 554 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
460 { 555 {
461 req->int1 = 1; /* mark request as delayed */ 556 req->int1 = 1; /* mark request as delayed */
462 continue; 557 continue;
463 } 558 }
464 else 559 else
465 { 560 {
466 int res = eio_finish (req); 561 int res = ETP_FINISH (req);
467 if (res) 562 if (ecb_expect_false (res))
468 return res; 563 return res;
469 } 564 }
470 565
471 if (maxreqs && !--maxreqs) 566 if (ecb_expect_false (maxreqs && !--maxreqs))
472 break; 567 break;
473 568
474 if (max_poll_time) 569 if (maxtime)
475 { 570 {
476 gettimeofday (&tv_now, 0); 571 gettimeofday (&tv_now, 0);
477 572
478 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 573 if (tvdiff (&tv_start, &tv_now) >= maxtime)
479 break; 574 break;
480 } 575 }
481 } 576 }
482 577
483 errno = EAGAIN; 578 errno = EAGAIN;
484 return -1; 579 return -1;
485} 580}
486 581
582static void
583etp_cancel (ETP_REQ *req)
584{
585 req->cancelled = 1;
586
587 eio_grp_cancel (req);
588}
589
590static void
591etp_submit (ETP_REQ *req)
592{
593 req->pri -= ETP_PRI_MIN;
594
595 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
596 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
597
598 if (ecb_expect_false (req->type == EIO_GROUP))
599 {
600 /* I hope this is worth it :/ */
601 X_LOCK (reqlock);
602 ++nreqs;
603 X_UNLOCK (reqlock);
604
605 X_LOCK (reslock);
606
607 ++npending;
608
609 if (!reqq_push (&res_queue, req) && want_poll_cb)
610 want_poll_cb ();
611
612 X_UNLOCK (reslock);
613 }
614 else
615 {
616 X_LOCK (reqlock);
617 ++nreqs;
618 ++nready;
619 reqq_push (&req_queue, req);
620 X_COND_SIGNAL (reqwait);
621 X_UNLOCK (reqlock);
622
623 etp_maybe_start_thread ();
624 }
625}
626
627static void ecb_cold
628etp_set_max_poll_time (double nseconds)
629{
630 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
631 max_poll_time = nseconds * EIO_TICKS;
632 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
633}
634
635static void ecb_cold
636etp_set_max_poll_reqs (unsigned int maxreqs)
637{
638 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
639 max_poll_reqs = maxreqs;
640 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
641}
642
643static void ecb_cold
644etp_set_max_idle (unsigned int nthreads)
645{
646 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
647 max_idle = nthreads;
648 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
649}
650
651static void ecb_cold
652etp_set_idle_timeout (unsigned int seconds)
653{
654 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
655 idle_timeout = seconds;
656 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
657}
658
659static void ecb_cold
660etp_set_min_parallel (unsigned int nthreads)
661{
662 if (wanted < nthreads)
663 wanted = nthreads;
664}
665
666static void ecb_cold
667etp_set_max_parallel (unsigned int nthreads)
668{
669 if (wanted > nthreads)
670 wanted = nthreads;
671
672 while (started > wanted)
673 etp_end_thread ();
674}
675
676/*****************************************************************************/
677
678static void
679grp_try_feed (eio_req *grp)
680{
681 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
682 {
683 grp->flags &= ~EIO_FLAG_GROUPADD;
684
685 EIO_FEED (grp);
686
687 /* stop if no progress has been made */
688 if (!(grp->flags & EIO_FLAG_GROUPADD))
689 {
690 grp->feed = 0;
691 break;
692 }
693 }
694}
695
696static int
697grp_dec (eio_req *grp)
698{
699 --grp->size;
700
701 /* call feeder, if applicable */
702 grp_try_feed (grp);
703
704 /* finish, if done */
705 if (!grp->size && grp->int1)
706 return eio_finish (grp);
707 else
708 return 0;
709}
710
711static void
712eio_destroy (eio_req *req)
713{
714 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
715 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
716
717 EIO_DESTROY (req);
718}
719
720static int
721eio_finish (eio_req *req)
722{
723 int res = EIO_FINISH (req);
724
725 if (req->grp)
726 {
727 int res2;
728 eio_req *grp = req->grp;
729
730 /* unlink request */
731 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
732 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
733
734 if (grp->grp_first == req)
735 grp->grp_first = req->grp_next;
736
737 res2 = grp_dec (grp);
738
739 if (!res)
740 res = res2;
741 }
742
743 eio_destroy (req);
744
745 return res;
746}
747
748void
749eio_grp_cancel (eio_req *grp)
750{
751 for (grp = grp->grp_first; grp; grp = grp->grp_next)
752 eio_cancel (grp);
753}
754
755void
756eio_cancel (eio_req *req)
757{
758 etp_cancel (req);
759}
760
761void
762eio_submit (eio_req *req)
763{
764 etp_submit (req);
765}
766
767unsigned int
768eio_nreqs (void)
769{
770 return etp_nreqs ();
771}
772
773unsigned int
774eio_nready (void)
775{
776 return etp_nready ();
777}
778
779unsigned int
780eio_npending (void)
781{
782 return etp_npending ();
783}
784
785unsigned int ecb_cold
786eio_nthreads (void)
787{
788 return etp_nthreads ();
789}
790
791void ecb_cold
792eio_set_max_poll_time (double nseconds)
793{
794 etp_set_max_poll_time (nseconds);
795}
796
797void ecb_cold
798eio_set_max_poll_reqs (unsigned int maxreqs)
799{
800 etp_set_max_poll_reqs (maxreqs);
801}
802
803void ecb_cold
804eio_set_max_idle (unsigned int nthreads)
805{
806 etp_set_max_idle (nthreads);
807}
808
809void ecb_cold
810eio_set_idle_timeout (unsigned int seconds)
811{
812 etp_set_idle_timeout (seconds);
813}
814
815void ecb_cold
816eio_set_min_parallel (unsigned int nthreads)
817{
818 etp_set_min_parallel (nthreads);
819}
820
821void ecb_cold
822eio_set_max_parallel (unsigned int nthreads)
823{
824 etp_set_max_parallel (nthreads);
825}
826
827int eio_poll (void)
828{
829 return etp_poll ();
830}
831
487/*****************************************************************************/ 832/*****************************************************************************/
488/* work around various missing functions */ 833/* work around various missing functions */
489 834
490#if !HAVE_PREADWRITE 835#if !HAVE_PREADWRITE
836# undef pread
837# undef pwrite
491# define pread aio_pread 838# define pread eio__pread
492# define pwrite aio_pwrite 839# define pwrite eio__pwrite
493 840
494/* 841static ssize_t
495 * make our pread/pwrite safe against themselves, but not against 842eio__pread (int fd, void *buf, size_t count, off_t offset)
496 * normal read/write by using a mutex. slows down execution a lot,
497 * but that's your problem, not mine.
498 */
499static mutex_t preadwritelock = X_MUTEX_INIT;
500
501static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
502{ 843{
503 ssize_t res; 844 ssize_t res;
504 off_t ooffset; 845 off_t ooffset;
505 846
506 X_LOCK (preadwritelock); 847 X_LOCK (preadwritelock);
… …
511 X_UNLOCK (preadwritelock); 852 X_UNLOCK (preadwritelock);
512 853
513 return res; 854 return res;
514} 855}
515 856
516static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 857static ssize_t
858eio__pwrite (int fd, void *buf, size_t count, off_t offset)
517{ 859{
518 ssize_t res; 860 ssize_t res;
519 off_t ooffset; 861 off_t ooffset;
520 862
521 X_LOCK (preadwritelock); 863 X_LOCK (preadwritelock);
522 ooffset = lseek (fd, 0, SEEK_CUR); 864 ooffset = lseek (fd, 0, SEEK_CUR);
523 lseek (fd, offset, SEEK_SET); 865 lseek (fd, offset, SEEK_SET);
524 res = write (fd, buf, count); 866 res = write (fd, buf, count);
525 lseek (fd, offset, SEEK_SET); 867 lseek (fd, ooffset, SEEK_SET);
526 X_UNLOCK (preadwritelock); 868 X_UNLOCK (preadwritelock);
527 869
528 return res; 870 return res;
529} 871}
530#endif 872#endif
531 873
532#ifndef HAVE_FUTIMES 874#ifndef HAVE_UTIMES
533 875
876# undef utimes
534# define utimes(path,times) aio_utimes (path, times) 877# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times)
536 878
879static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 880eio__utimes (const char *filename, const struct timeval times[2])
538{ 881{
539 if (times) 882 if (times)
540 { 883 {
541 struct utimbuf buf; 884 struct utimbuf buf;
542 885
… …
547 } 890 }
548 else 891 else
549 return utime (filename, 0); 892 return utime (filename, 0);
550} 893}
551 894
895#endif
896
897#ifndef HAVE_FUTIMES
898
899# undef futimes
900# define futimes(fd,times) eio__futimes (fd, times)
901
902static int
552static int aio_futimes (int fd, const struct timeval tv[2]) 903eio__futimes (int fd, const struct timeval tv[2])
553{ 904{
554 errno = ENOSYS; 905 errno = ENOSYS;
555 return -1; 906 return -1;
556} 907}
557 908
558#endif 909#endif
559 910
560#if !HAVE_FDATASYNC 911#if !HAVE_FDATASYNC
912# undef fdatasync
561# define fdatasync fsync 913# define fdatasync(fd) fsync (fd)
562#endif 914#endif
915
916/* sync_file_range always needs emulation */
917static int
918eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
919{
920#if HAVE_SYNC_FILE_RANGE
921 int res;
922
923 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
924 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
925 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
926 {
927 flags = 0
928 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
929 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
930 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
931 }
932
933 res = sync_file_range (fd, offset, nbytes, flags);
934
935 if (!res || errno != ENOSYS)
936 return res;
937#endif
938
939 /* even though we could play tricks with the flags, it's better to always
940 * call fdatasync, as that matches the expectation of its users best */
941 return fdatasync (fd);
942}
943
944static int
945eio__fallocate (int fd, int mode, off_t offset, size_t len)
946{
947#if HAVE_FALLOCATE
948 return fallocate (fd, mode, offset, len);
949#else
950 errno = ENOSYS;
951 return -1;
952#endif
953}
563 954
564#if !HAVE_READAHEAD 955#if !HAVE_READAHEAD
956# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 957# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 958
959static ssize_t
567static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 960eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568{ 961{
569 size_t todo = count; 962 size_t todo = count;
570 dBUF; 963 dBUF;
571 964
572 while (todo > 0) 965 while (todo > 0)
573 { 966 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 967 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 968
576 pread (fd, aio_buf, len, offset); 969 pread (fd, eio_buf, len, offset);
577 offset += len; 970 offset += len;
578 todo -= len; 971 todo -= len;
579 } 972 }
580 973
581 errno = 0; 974 errno = 0;
582 return count; 975 return count;
583} 976}
584 977
585#endif 978#endif
586 979
587#if !HAVE_READDIR_R
588# define readdir_r aio_readdir_r
589
590static mutex_t readdirlock = X_MUTEX_INIT;
591
592static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
593{
594 X_DIRENT *e;
595 int errorno;
596
597 X_LOCK (readdirlock);
598
599 e = readdir (dirp);
600 errorno = errno;
601
602 if (e)
603 {
604 *res = ent;
605 strcpy (ent->d_name, e->d_name);
606 }
607 else
608 *res = 0;
609
610 X_UNLOCK (readdirlock);
611
612 errno = errorno;
613 return e ? 0 : -1;
614}
615#endif
616
617/* sendfile always needs emulation */ 980/* sendfile always needs emulation */
981static ssize_t
618static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 982eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
619{ 983{
984 ssize_t written = 0;
620 ssize_t res; 985 ssize_t res;
621 986
622 if (!count) 987 if (!count)
623 return 0; 988 return 0;
624 989
990 for (;;)
991 {
992#ifdef __APPLE__
993# undef HAVE_SENDFILE /* broken, as everything on os x */
994#endif
625#if HAVE_SENDFILE 995#if HAVE_SENDFILE
626# if __linux 996# if __linux
997 off_t soffset = offset;
627 res = sendfile (ofd, ifd, &offset, count); 998 res = sendfile (ofd, ifd, &soffset, count);
628 999
629# elif __freebsd 1000# elif __FreeBSD__
630 /* 1001 /*
631 * Of course, the freebsd sendfile is a dire hack with no thoughts 1002 * Of course, the freebsd sendfile is a dire hack with no thoughts
632 * wasted on making it similar to other I/O functions. 1003 * wasted on making it similar to other I/O functions.
633 */ 1004 */
634 {
635 off_t sbytes; 1005 off_t sbytes;
636 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1006 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637 1007
638 if (res < 0 && sbytes) 1008 #if 0 /* according to the manpage, this is correct, but broken behaviour */
639 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1009 /* freebsd' sendfile will return 0 on success */
1010 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1011 /* not on e.g. EIO or EPIPE - sounds broken */
1012 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
640 res = sbytes; 1013 res = sbytes;
641 } 1014 #endif
1015
1016 /* according to source inspection, this is correct, and useful behaviour */
1017 if (sbytes)
1018 res = sbytes;
1019
1020# elif defined (__APPLE__)
1021 off_t sbytes = count;
1022 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1023
1024 /* according to the manpage, sbytes is always valid */
1025 if (sbytes)
1026 res = sbytes;
642 1027
643# elif __hpux 1028# elif __hpux
644 res = sendfile (ofd, ifd, offset, count, 0, 0); 1029 res = sendfile (ofd, ifd, offset, count, 0, 0);
645 1030
646# elif __solaris 1031# elif __solaris
647 {
648 struct sendfilevec vec; 1032 struct sendfilevec vec;
649 size_t sbytes; 1033 size_t sbytes;
650 1034
651 vec.sfv_fd = ifd; 1035 vec.sfv_fd = ifd;
652 vec.sfv_flag = 0; 1036 vec.sfv_flag = 0;
653 vec.sfv_off = offset; 1037 vec.sfv_off = offset;
654 vec.sfv_len = count; 1038 vec.sfv_len = count;
655 1039
656 res = sendfilev (ofd, &vec, 1, &sbytes); 1040 res = sendfilev (ofd, &vec, 1, &sbytes);
657 1041
658 if (res < 0 && sbytes) 1042 if (res < 0 && sbytes)
659 res = sbytes; 1043 res = sbytes;
660 }
661 1044
662# endif 1045# endif
1046
1047#elif defined (_WIN32)
1048 /* does not work, just for documentation of what would need to be done */
1049 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1050 /* libeio guarantees that the file offset does not change, and windows */
1051 /* has no way to get an independent handle to the same file description */
1052 HANDLE h = TO_SOCKET (ifd);
1053 SetFilePointer (h, offset, 0, FILE_BEGIN);
1054 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1055
663#else 1056#else
664 res = -1; 1057 res = -1;
665 errno = ENOSYS; 1058 errno = ENOSYS;
666#endif 1059#endif
667 1060
1061 /* we assume sendfile can copy at least 128mb in one go */
1062 if (res <= 128 * 1024 * 1024)
1063 {
1064 if (res > 0)
1065 written += res;
1066
1067 if (written)
1068 return written;
1069
1070 break;
1071 }
1072 else
1073 {
1074 /* if we requested more, then probably the kernel was lazy */
1075 written += res;
1076 offset += res;
1077 count -= res;
1078
1079 if (!count)
1080 return written;
1081 }
1082 }
1083
668 if (res < 0 1084 if (res < 0
669 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1085 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1086 /* BSDs */
1087#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1088 || errno == ENOTSUP
1089#endif
1090 || errno == EOPNOTSUPP /* BSDs */
670#if __solaris 1091#if __solaris
671 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1092 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
672#endif 1093#endif
673 ) 1094 )
674 ) 1095 )
… …
705 } 1126 }
706 1127
707 return res; 1128 return res;
708} 1129}
709 1130
1131#ifdef PAGESIZE
1132# define eio_pagesize() PAGESIZE
1133#else
1134static intptr_t
1135eio_pagesize (void)
1136{
1137 static intptr_t page;
1138
1139 if (!page)
1140 page = sysconf (_SC_PAGESIZE);
1141
1142 return page;
1143}
1144#endif
1145
1146static void
1147eio_page_align (void **addr, size_t *length)
1148{
1149 intptr_t mask = eio_pagesize () - 1;
1150
1151 /* round down addr */
1152 intptr_t adj = mask & (intptr_t)*addr;
1153
1154 *addr = (void *)((intptr_t)*addr - adj);
1155 *length += adj;
1156
1157 /* round up length */
1158 *length = (*length + mask) & ~mask;
1159}
1160
1161#if !_POSIX_MEMLOCK
1162# define eio__mlockall(a) ((errno = ENOSYS), -1)
1163#else
1164
1165static int
1166eio__mlockall (int flags)
1167{
1168 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1169 extern int mallopt (int, int);
1170 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1171 #endif
1172
1173 if (EIO_MCL_CURRENT != MCL_CURRENT
1174 || EIO_MCL_FUTURE != MCL_FUTURE)
1175 {
1176 flags = 0
1177 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1178 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1179 }
1180
1181 return mlockall (flags);
1182}
1183#endif
1184
1185#if !_POSIX_MEMLOCK_RANGE
1186# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1187#else
1188
1189static int
1190eio__mlock (void *addr, size_t length)
1191{
1192 eio_page_align (&addr, &length);
1193
1194 return mlock (addr, length);
1195}
1196
1197#endif
1198
1199#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1200# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1201#else
1202
1203static int
1204eio__msync (void *mem, size_t len, int flags)
1205{
1206 eio_page_align (&mem, &len);
1207
1208 if (EIO_MS_ASYNC != MS_SYNC
1209 || EIO_MS_INVALIDATE != MS_INVALIDATE
1210 || EIO_MS_SYNC != MS_SYNC)
1211 {
1212 flags = 0
1213 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1214 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1215 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1216 }
1217
1218 return msync (mem, len, flags);
1219}
1220
1221#endif
1222
1223static int
1224eio__mtouch (eio_req *req)
1225{
1226 void *mem = req->ptr2;
1227 size_t len = req->size;
1228 int flags = req->int1;
1229
1230 eio_page_align (&mem, &len);
1231
1232 {
1233 intptr_t addr = (intptr_t)mem;
1234 intptr_t end = addr + len;
1235 intptr_t page = eio_pagesize ();
1236
1237 if (addr < end)
1238 if (flags & EIO_MT_MODIFY) /* modify */
1239 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1240 else
1241 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1242 }
1243
1244 return 0;
1245}
1246
1247/*****************************************************************************/
1248/* requests implemented outside eio_execute, because they are so large */
1249
1250static void
1251eio__realpath (eio_req *req, etp_worker *self)
1252{
1253 char *rel = req->ptr1;
1254 char *res;
1255 char *tmp1, *tmp2;
1256#if SYMLOOP_MAX > 32
1257 int symlinks = SYMLOOP_MAX;
1258#else
1259 int symlinks = 32;
1260#endif
1261
1262 req->result = -1;
1263
1264 errno = EINVAL;
1265 if (!rel)
1266 return;
1267
1268 errno = ENOENT;
1269 if (!*rel)
1270 return;
1271
1272 if (!req->ptr2)
1273 {
1274 X_LOCK (wrklock);
1275 req->flags |= EIO_FLAG_PTR2_FREE;
1276 X_UNLOCK (wrklock);
1277 req->ptr2 = malloc (PATH_MAX * 3);
1278
1279 errno = ENOMEM;
1280 if (!req->ptr2)
1281 return;
1282 }
1283
1284 res = req->ptr2;
1285 tmp1 = res + PATH_MAX;
1286 tmp2 = tmp1 + PATH_MAX;
1287
1288#if 0 /* disabled, the musl way to do things is just too racy */
1289#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1290 /* on linux we may be able to ask the kernel */
1291 {
1292 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1293
1294 if (fd >= 0)
1295 {
1296 sprintf (tmp1, "/proc/self/fd/%d", fd);
1297 req->result = readlink (tmp1, res, PATH_MAX);
1298 close (fd);
1299
1300 /* here we should probably stat the open file and the disk file, to make sure they still match */
1301
1302 if (req->result > 0)
1303 goto done;
1304 }
1305 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1306 return;
1307 }
1308#endif
1309#endif
1310
1311 if (*rel != '/')
1312 {
1313 if (!getcwd (res, PATH_MAX))
1314 return;
1315
1316 if (res [1]) /* only use if not / */
1317 res += strlen (res);
1318 }
1319
1320 while (*rel)
1321 {
1322 ssize_t len, linklen;
1323 char *beg = rel;
1324
1325 while (*rel && *rel != '/')
1326 ++rel;
1327
1328 len = rel - beg;
1329
1330 if (!len) /* skip slashes */
1331 {
1332 ++rel;
1333 continue;
1334 }
1335
1336 if (beg [0] == '.')
1337 {
1338 if (len == 1)
1339 continue; /* . - nop */
1340
1341 if (beg [1] == '.' && len == 2)
1342 {
1343 /* .. - back up one component, if possible */
1344
1345 while (res != req->ptr2)
1346 if (*--res == '/')
1347 break;
1348
1349 continue;
1350 }
1351 }
1352
1353 errno = ENAMETOOLONG;
1354 if (res + 1 + len + 1 >= tmp1)
1355 return;
1356
1357 /* copy one component */
1358 *res = '/';
1359 memcpy (res + 1, beg, len);
1360
1361 /* zero-terminate, for readlink */
1362 res [len + 1] = 0;
1363
1364 /* now check if it's a symlink */
1365 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1366
1367 if (linklen < 0)
1368 {
1369 if (errno != EINVAL)
1370 return;
1371
1372 /* it's a normal directory. hopefully */
1373 res += len + 1;
1374 }
1375 else
1376 {
1377 /* yay, it was a symlink - build new path in tmp2 */
1378 int rellen = strlen (rel);
1379
1380 errno = ENAMETOOLONG;
1381 if (linklen + 1 + rellen >= PATH_MAX)
1382 return;
1383
1384 errno = ELOOP;
1385 if (!--symlinks)
1386 return;
1387
1388 if (*tmp1 == '/')
1389 res = req->ptr2; /* symlink resolves to an absolute path */
1390
1391 /* we need to be careful, as rel might point into tmp2 already */
1392 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1393 tmp2 [linklen] = '/';
1394 memcpy (tmp2, tmp1, linklen);
1395
1396 rel = tmp2;
1397 }
1398 }
1399
1400 /* special case for the lone root path */
1401 if (res == req->ptr2)
1402 *res++ = '/';
1403
1404 req->result = res - (char *)req->ptr2;
1405
1406done:
1407 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1408}
1409
1410static signed char
1411eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1412{
1413 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1414 : a->inode < b->inode ? -1
1415 : a->inode > b->inode ? 1
1416 : 0;
1417}
1418
1419#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1420
1421#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1422#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1423
1424static void
1425eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1426{
1427 unsigned char bits [9 + sizeof (ino_t) * 8];
1428 unsigned char *bit = bits;
1429
1430 assert (CHAR_BIT == 8);
1431 assert (sizeof (eio_dirent) * 8 < 256);
1432 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1433 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1434
1435 if (size <= EIO_SORT_FAST)
1436 return;
1437
1438 /* first prepare an array of bits to test in our radix sort */
1439 /* try to take endianness into account, as well as differences in ino_t sizes */
1440 /* inode_bits must contain all inodes ORed together */
1441 /* which is used to skip bits that are 0 everywhere, which is very common */
1442 {
1443 ino_t endianness;
1444 int i, j;
1445
1446 /* we store the byte offset of byte n into byte n of "endianness" */
1447 for (i = 0; i < sizeof (ino_t); ++i)
1448 ((unsigned char *)&endianness)[i] = i;
1449
1450 *bit++ = 0;
1451
1452 for (i = 0; i < sizeof (ino_t); ++i)
1453 {
1454 /* shifting off the byte offsets out of "endianness" */
1455 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1456 endianness >>= 8;
1457
1458 for (j = 0; j < 8; ++j)
1459 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1460 *bit++ = offs + j;
1461 }
1462
1463 for (j = 0; j < 8; ++j)
1464 if (score_bits & (1 << j))
1465 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1466 }
1467
1468 /* now actually do the sorting (a variant of MSD radix sort) */
1469 {
1470 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1471 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1472 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1473 int stk_idx = 0;
1474
1475 base_stk [stk_idx] = dents;
1476 end_stk [stk_idx] = dents + size;
1477 bit_stk [stk_idx] = bit - 1;
1478
1479 do
1480 {
1481 base = base_stk [stk_idx];
1482 end = end_stk [stk_idx];
1483 bit = bit_stk [stk_idx];
1484
1485 for (;;)
1486 {
1487 unsigned char O = *bit >> 3;
1488 unsigned char M = 1 << (*bit & 7);
1489
1490 eio_dirent *a = base;
1491 eio_dirent *b = end;
1492
1493 if (b - a < EIO_SORT_CUTOFF)
1494 break;
1495
1496 /* now bit-partition the array on the bit */
1497 /* this ugly asymmetric loop seems to perform much better than typical */
1498 /* partition algos found in the literature */
1499 do
1500 if (!(((unsigned char *)a)[O] & M))
1501 ++a;
1502 else if (!(((unsigned char *)--b)[O] & M))
1503 {
1504 eio_dirent tmp = *a; *a = *b; *b = tmp;
1505 ++a;
1506 }
1507 while (b > a);
1508
1509 /* next bit, or stop, if no bits left in this path */
1510 if (!*--bit)
1511 break;
1512
1513 base_stk [stk_idx] = a;
1514 end_stk [stk_idx] = end;
1515 bit_stk [stk_idx] = bit;
1516 ++stk_idx;
1517
1518 end = a;
1519 }
1520 }
1521 while (stk_idx--);
1522 }
1523}
1524
1525static void
1526eio_dent_insertion_sort (eio_dirent *dents, int size)
1527{
1528 /* first move the smallest element to the front, to act as a sentinel */
1529 {
1530 int i;
1531 eio_dirent *min = dents;
1532
1533 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1534 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1535 if (EIO_DENT_CMP (dents [i], <, *min))
1536 min = &dents [i];
1537
1538 /* swap elements 0 and j (minimum) */
1539 {
1540 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1541 }
1542 }
1543
1544 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1545 {
1546 eio_dirent *i, *j;
1547
1548 for (i = dents + 1; i < dents + size; ++i)
1549 {
1550 eio_dirent value = *i;
1551
1552 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1553 j [1] = j [0];
1554
1555 j [1] = value;
1556 }
1557 }
1558}
1559
1560static void
1561eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1562{
1563 if (size <= 1)
1564 return; /* our insertion sort relies on size > 0 */
1565
1566 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1567 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1568 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1569
1570 /* use an insertion sort at the end, or for small arrays, */
1571 /* as insertion sort is more efficient for small partitions */
1572 eio_dent_insertion_sort (dents, size);
1573}
1574
710/* read a full directory */ 1575/* read a full directory */
1576static void
711static void scandir_ (eio_req *req, worker *self) 1577eio__scandir (eio_req *req, etp_worker *self)
712{ 1578{
713 DIR *dirp; 1579 DIR *dirp;
714 union
715 {
716 EIO_STRUCT_DIRENT d;
717 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
718 } *u;
719 EIO_STRUCT_DIRENT *entp; 1580 EIO_STRUCT_DIRENT *entp;
720 char *name, *names; 1581 char *name, *names;
721 int memlen = 4096; 1582 int namesalloc = 4096;
722 int memofs = 0; 1583 int namesoffs = 0;
1584 int flags = req->int1;
1585 eio_dirent *dents = 0;
1586 int dentalloc = 128;
723 int res = 0; 1587 int dentoffs = 0;
1588 ino_t inode_bits = 0;
1589
1590 req->result = -1;
1591
1592 if (!(flags & EIO_READDIR_DENTS))
1593 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
724 1594
725 X_LOCK (wrklock); 1595 X_LOCK (wrklock);
1596 /* the corresponding closedir is in ETP_WORKER_CLEAR */
726 self->dirp = dirp = opendir (req->ptr1); 1597 self->dirp = dirp = opendir (req->ptr1);
727 self->dbuf = u = malloc (sizeof (*u)); 1598
728 req->flags |= EIO_FLAG_PTR2_FREE; 1599 if (req->flags & EIO_FLAG_PTR1_FREE)
1600 free (req->ptr1);
1601
1602 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1603 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
729 req->ptr2 = names = malloc (memlen); 1604 req->ptr2 = names = malloc (namesalloc);
730 X_UNLOCK (wrklock); 1605 X_UNLOCK (wrklock);
731 1606
732 if (dirp && u && names) 1607 if (dirp && names && (!flags || dents))
733 for (;;) 1608 for (;;)
734 { 1609 {
735 errno = 0; 1610 errno = 0;
736 readdir_r (dirp, &u->d, &entp); 1611 entp = readdir (dirp);
737 1612
738 if (!entp) 1613 if (!entp)
1614 {
1615 if (errno)
1616 break;
1617
1618 /* sort etc. */
1619 req->int1 = flags;
1620 req->result = dentoffs;
1621
1622 if (flags & EIO_READDIR_STAT_ORDER)
1623 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1624 else if (flags & EIO_READDIR_DIRS_FIRST)
1625 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1626 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1627 else
1628 {
1629 /* in this case, all is known, and we just put dirs first and sort them */
1630 eio_dirent *oth = dents + dentoffs;
1631 eio_dirent *dir = dents;
1632
1633 /* now partition dirs to the front, and non-dirs to the back */
1634 /* by walking from both sides and swapping if necessary */
1635 while (oth > dir)
1636 {
1637 if (dir->type == EIO_DT_DIR)
1638 ++dir;
1639 else if ((--oth)->type == EIO_DT_DIR)
1640 {
1641 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1642
1643 ++dir;
1644 }
1645 }
1646
1647 /* now sort the dirs only (dirs all have the same score) */
1648 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1649 }
1650
739 break; 1651 break;
1652 }
740 1653
1654 /* now add the entry to our list(s) */
741 name = entp->d_name; 1655 name = entp->d_name;
742 1656
1657 /* skip . and .. entries */
743 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1658 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
744 { 1659 {
745 int len = strlen (name) + 1; 1660 int len = D_NAMLEN (entp) + 1;
746 1661
747 res++; 1662 while (ecb_expect_false (namesoffs + len > namesalloc))
748
749 while (memofs + len > memlen)
750 { 1663 {
751 memlen *= 2; 1664 namesalloc *= 2;
752 X_LOCK (wrklock); 1665 X_LOCK (wrklock);
753 req->ptr2 = names = realloc (names, memlen); 1666 req->ptr2 = names = realloc (names, namesalloc);
754 X_UNLOCK (wrklock); 1667 X_UNLOCK (wrklock);
755 1668
756 if (!names) 1669 if (!names)
757 break; 1670 break;
758 } 1671 }
759 1672
760 memcpy (names + memofs, name, len); 1673 memcpy (names + namesoffs, name, len);
1674
1675 if (dents)
1676 {
1677 struct eio_dirent *ent;
1678
1679 if (ecb_expect_false (dentoffs == dentalloc))
1680 {
1681 dentalloc *= 2;
1682 X_LOCK (wrklock);
1683 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1684 X_UNLOCK (wrklock);
1685
1686 if (!dents)
1687 break;
1688 }
1689
1690 ent = dents + dentoffs;
1691
1692 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1693 ent->namelen = len - 1;
1694 ent->inode = D_INO (entp);
1695
1696 inode_bits |= ent->inode;
1697
1698 switch (D_TYPE (entp))
1699 {
1700 default:
1701 ent->type = EIO_DT_UNKNOWN;
1702 flags |= EIO_READDIR_FOUND_UNKNOWN;
1703 break;
1704
1705 #ifdef DT_FIFO
1706 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1707 #endif
1708 #ifdef DT_CHR
1709 case DT_CHR: ent->type = EIO_DT_CHR; break;
1710 #endif
1711 #ifdef DT_MPC
1712 case DT_MPC: ent->type = EIO_DT_MPC; break;
1713 #endif
1714 #ifdef DT_DIR
1715 case DT_DIR: ent->type = EIO_DT_DIR; break;
1716 #endif
1717 #ifdef DT_NAM
1718 case DT_NAM: ent->type = EIO_DT_NAM; break;
1719 #endif
1720 #ifdef DT_BLK
1721 case DT_BLK: ent->type = EIO_DT_BLK; break;
1722 #endif
1723 #ifdef DT_MPB
1724 case DT_MPB: ent->type = EIO_DT_MPB; break;
1725 #endif
1726 #ifdef DT_REG
1727 case DT_REG: ent->type = EIO_DT_REG; break;
1728 #endif
1729 #ifdef DT_NWK
1730 case DT_NWK: ent->type = EIO_DT_NWK; break;
1731 #endif
1732 #ifdef DT_CMP
1733 case DT_CMP: ent->type = EIO_DT_CMP; break;
1734 #endif
1735 #ifdef DT_LNK
1736 case DT_LNK: ent->type = EIO_DT_LNK; break;
1737 #endif
1738 #ifdef DT_SOCK
1739 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1740 #endif
1741 #ifdef DT_DOOR
1742 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1743 #endif
1744 #ifdef DT_WHT
1745 case DT_WHT: ent->type = EIO_DT_WHT; break;
1746 #endif
1747 }
1748
1749 ent->score = 7;
1750
1751 if (flags & EIO_READDIR_DIRS_FIRST)
1752 {
1753 if (ent->type == EIO_DT_UNKNOWN)
1754 {
1755 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1756 ent->score = 1;
1757 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1758 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1759 }
1760 else if (ent->type == EIO_DT_DIR)
1761 ent->score = 0;
1762 }
1763 }
1764
761 memofs += len; 1765 namesoffs += len;
1766 ++dentoffs;
1767 }
1768
1769 if (EIO_CANCELLED (req))
1770 {
1771 errno = ECANCELED;
1772 break;
762 } 1773 }
763 } 1774 }
764
765 if (errno)
766 res = -1;
767
768 req->result = res;
769} 1775}
770 1776
771/*****************************************************************************/ 1777/*****************************************************************************/
772 1778
1779#define ALLOC(len) \
1780 if (!req->ptr2) \
1781 { \
1782 X_LOCK (wrklock); \
1783 req->flags |= EIO_FLAG_PTR2_FREE; \
1784 X_UNLOCK (wrklock); \
1785 req->ptr2 = malloc (len); \
1786 if (!req->ptr2) \
1787 { \
1788 errno = ENOMEM; \
1789 req->result = -1; \
1790 break; \
1791 } \
1792 }
1793
773X_THREAD_PROC (eio_proc) 1794X_THREAD_PROC (etp_proc)
774{ 1795{
775 eio_req *req; 1796 ETP_REQ *req;
776 struct timespec ts; 1797 struct timespec ts;
777 worker *self = (worker *)thr_arg; 1798 etp_worker *self = (etp_worker *)thr_arg;
1799 int timeout;
778 1800
779 /* try to distribute timeouts somewhat randomly */ 1801 /* try to distribute timeouts somewhat evenly */
780 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1802 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781 1803
782 for (;;) 1804 for (;;)
783 { 1805 {
784 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1806 ts.tv_sec = 0;
785 1807
786 X_LOCK (reqlock); 1808 X_LOCK (reqlock);
787 1809
788 for (;;) 1810 for (;;)
789 { 1811 {
790 self->req = req = reqq_shift (&req_queue); 1812 self->req = req = reqq_shift (&req_queue);
791 1813
792 if (req) 1814 if (req)
793 break; 1815 break;
794 1816
1817 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1818 {
1819 X_UNLOCK (reqlock);
1820 X_LOCK (wrklock);
1821 --started;
1822 X_UNLOCK (wrklock);
1823 goto quit;
1824 }
1825
795 ++idle; 1826 ++idle;
796 1827
797 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1828 if (idle <= max_idle)
1829 /* we are allowed to idle, so do so without any timeout */
1830 X_COND_WAIT (reqwait, reqlock);
1831 else
798 { 1832 {
799 if (idle > max_idle) 1833 /* initialise timeout once */
800 { 1834 if (!ts.tv_sec)
801 --idle; 1835 ts.tv_sec = time (0) + idle_timeout;
802 X_UNLOCK (reqlock);
803 X_LOCK (wrklock);
804 --started;
805 X_UNLOCK (wrklock);
806 goto quit;
807 }
808 1836
809 /* we are allowed to idle, so do so without any timeout */
810 X_COND_WAIT (reqwait, reqlock); 1837 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
811 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1838 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
812 } 1839 }
813 1840
814 --idle; 1841 --idle;
815 } 1842 }
816 1843
817 --nready; 1844 --nready;
818 1845
819 X_UNLOCK (reqlock); 1846 X_UNLOCK (reqlock);
820 1847
821 errno = 0; /* strictly unnecessary */
822
823 if (!EIO_CANCELLED (req))
824 switch (req->type) 1848 if (req->type < 0)
825 {
826 case EIO_READ: req->result = req->offs >= 0
827 ? pread (req->int1, req->ptr2, req->size, req->offs)
828 : read (req->int1, req->ptr2, req->size); break;
829 case EIO_WRITE: req->result = req->offs >= 0
830 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
831 : write (req->int1, req->ptr2, req->size); break;
832
833 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
834 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
835
836 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
837 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
838 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
839
840 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
841 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
842 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
843 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
844 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
845 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
846
847 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
848 case EIO_CLOSE: req->result = close (req->int1); break;
849 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
850 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
851 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
852 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
853 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
854 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
855 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
856 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
857 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
858
859 case EIO_SYNC: req->result = 0; sync (); break;
860 case EIO_FSYNC: req->result = fsync (req->int1); break;
861 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
862
863 case EIO_READDIR: scandir_ (req, self); break;
864
865 case EIO_BUSY:
866#ifdef _WIN32
867 Sleep (req->nv1 * 1000.);
868#else
869 {
870 struct timeval tv;
871
872 tv.tv_sec = req->nv1;
873 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
874
875 req->result = select (0, 0, 0, 0, &tv);
876 }
877#endif
878 break;
879
880 case EIO_UTIME:
881 case EIO_FUTIME:
882 {
883 struct timeval tv[2];
884 struct timeval *times;
885
886 if (req->nv1 != -1. || req->nv2 != -1.)
887 {
888 tv[0].tv_sec = req->nv1;
889 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
890 tv[1].tv_sec = req->nv2;
891 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
892
893 times = tv;
894 }
895 else
896 times = 0;
897
898
899 req->result = req->type == EIO_FUTIME
900 ? futimes (req->int1, times)
901 : utimes (req->ptr1, times);
902 }
903
904 case EIO_GROUP:
905 case EIO_NOP:
906 break;
907
908 case EIO_QUIT:
909 goto quit; 1849 goto quit;
910 1850
911 default: 1851 ETP_EXECUTE (self, req);
912 req->result = -1;
913 break;
914 }
915
916 req->errorno = errno;
917 1852
918 X_LOCK (reslock); 1853 X_LOCK (reslock);
919 1854
920 ++npending; 1855 ++npending;
921 1856
922 if (!reqq_push (&res_queue, req) && want_poll_cb) 1857 if (!reqq_push (&res_queue, req) && want_poll_cb)
923 want_poll_cb (); 1858 want_poll_cb ();
924 1859
925 self->req = 0; 1860 self->req = 0;
926 worker_clear (self); 1861 etp_worker_clear (self);
927 1862
928 X_UNLOCK (reslock); 1863 X_UNLOCK (reslock);
929 } 1864 }
930 1865
931quit: 1866quit:
932 X_LOCK (wrklock); 1867 X_LOCK (wrklock);
933 worker_free (self); 1868 etp_worker_free (self);
934 X_UNLOCK (wrklock); 1869 X_UNLOCK (wrklock);
935 1870
936 return 0; 1871 return 0;
937} 1872}
938 1873
939/*****************************************************************************/ 1874/*****************************************************************************/
940 1875
941static void eio_atfork_prepare (void) 1876int ecb_cold
1877eio_init (void (*want_poll)(void), void (*done_poll)(void))
942{ 1878{
943 X_LOCK (wrklock); 1879 return etp_init (want_poll, done_poll);
944 X_LOCK (reqlock);
945 X_LOCK (reslock);
946#if !HAVE_PREADWRITE
947 X_LOCK (preadwritelock);
948#endif
949#if !HAVE_READDIR_R
950 X_LOCK (readdirlock);
951#endif
952} 1880}
953 1881
954static void eio_atfork_parent (void) 1882ecb_inline void
1883eio_api_destroy (eio_req *req)
955{ 1884{
956#if !HAVE_READDIR_R 1885 free (req);
957 X_UNLOCK (readdirlock);
958#endif
959#if !HAVE_PREADWRITE
960 X_UNLOCK (preadwritelock);
961#endif
962 X_UNLOCK (reslock);
963 X_UNLOCK (reqlock);
964 X_UNLOCK (wrklock);
965} 1886}
966 1887
967static void eio_atfork_child (void) 1888#define REQ(rtype) \
968{ 1889 eio_req *req; \
969 eio_req *prv; 1890 \
1891 req = (eio_req *)calloc (1, sizeof *req); \
1892 if (!req) \
1893 return 0; \
1894 \
1895 req->type = rtype; \
1896 req->pri = pri; \
1897 req->finish = cb; \
1898 req->data = data; \
1899 req->destroy = eio_api_destroy;
970 1900
971 while (prv = reqq_shift (&req_queue)) 1901#define SEND eio_submit (req); return req
972 eio_destroy (prv);
973 1902
974 while (prv = reqq_shift (&res_queue)) 1903#define PATH \
975 eio_destroy (prv); 1904 req->flags |= EIO_FLAG_PTR1_FREE; \
1905 req->ptr1 = strdup (path); \
1906 if (!req->ptr1) \
1907 { \
1908 eio_api_destroy (req); \
1909 return 0; \
1910 }
976 1911
977 while (wrk_first.next != &wrk_first) 1912static void
1913eio_execute (etp_worker *self, eio_req *req)
1914{
1915 if (ecb_expect_false (EIO_CANCELLED (req)))
978 { 1916 {
979 worker *wrk = wrk_first.next; 1917 req->result = -1;
1918 req->errorno = ECANCELED;
1919 return;
1920 }
980 1921
981 if (wrk->req) 1922 switch (req->type)
982 eio_destroy (wrk->req);
983
984 worker_clear (wrk);
985 worker_free (wrk);
986 } 1923 {
1924 case EIO_READ: ALLOC (req->size);
1925 req->result = req->offs >= 0
1926 ? pread (req->int1, req->ptr2, req->size, req->offs)
1927 : read (req->int1, req->ptr2, req->size); break;
1928 case EIO_WRITE: req->result = req->offs >= 0
1929 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1930 : write (req->int1, req->ptr2, req->size); break;
987 1931
988 started = 0; 1932 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
989 idle = 0; 1933 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
990 nreqs = 0;
991 nready = 0;
992 npending = 0;
993 1934
994 eio_atfork_parent (); 1935 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
995} 1936 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1937 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1938 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1939 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1940 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
996 1941
997int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1942 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
998{ 1943 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
999 want_poll_cb = want_poll; 1944 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1000 done_poll_cb = done_poll; 1945 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1001 1946
1947 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1948 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1949 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1950 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1951 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1952 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1953
1954 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1955 case EIO_CLOSE: req->result = close (req->int1); break;
1956 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1957 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1958 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1959 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1960 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1961 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1962 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1963 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1964
1965 case EIO_REALPATH: eio__realpath (req, self); break;
1966
1967 case EIO_READLINK: ALLOC (PATH_MAX);
1968 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1969
1970 case EIO_SYNC: req->result = 0; sync (); break;
1971 case EIO_FSYNC: req->result = fsync (req->int1); break;
1972 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1973 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1974 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1975 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1976 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1977 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1978 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1979
1980 case EIO_READDIR: eio__scandir (req, self); break;
1981
1982 case EIO_BUSY:
1002#ifdef _WIN32 1983#ifdef _WIN32
1003 X_MUTEX_CHECK (wrklock); 1984 Sleep (req->nv1 * 1e3);
1004 X_MUTEX_CHECK (reslock); 1985#else
1005 X_MUTEX_CHECK (reqlock); 1986 {
1006 X_MUTEX_CHECK (reqwait); 1987 struct timeval tv;
1007 X_MUTEX_CHECK (preadwritelock);
1008 X_MUTEX_CHECK (readdirlock);
1009 1988
1010 X_COND_CHECK (reqwait); 1989 tv.tv_sec = req->nv1;
1011#endif 1990 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1012 1991
1013 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child); 1992 req->result = select (0, 0, 0, 0, &tv);
1014} 1993 }
1994#endif
1995 break;
1015 1996
1016#if 0 1997 case EIO_UTIME:
1998 case EIO_FUTIME:
1999 {
2000 struct timeval tv[2];
2001 struct timeval *times;
1017 2002
2003 if (req->nv1 != -1. || req->nv2 != -1.)
2004 {
2005 tv[0].tv_sec = req->nv1;
2006 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2007 tv[1].tv_sec = req->nv2;
2008 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2009
2010 times = tv;
2011 }
2012 else
2013 times = 0;
2014
2015 req->result = req->type == EIO_FUTIME
2016 ? futimes (req->int1, times)
2017 : utimes (req->ptr1, times);
2018 }
2019 break;
2020
2021 case EIO_GROUP:
2022 abort (); /* handled in eio_request */
2023
2024 case EIO_NOP:
2025 req->result = 0;
2026 break;
2027
2028 case EIO_CUSTOM:
2029 req->feed (req);
2030 break;
2031
2032 default:
2033 errno = ENOSYS;
2034 req->result = -1;
2035 break;
2036 }
2037
2038 req->errorno = errno;
2039}
2040
2041#ifndef EIO_NO_WRAPPERS
2042
2043eio_req *eio_nop (int pri, eio_cb cb, void *data)
2044{
2045 REQ (EIO_NOP); SEND;
2046}
2047
2048eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2049{
2050 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2051}
2052
1018eio_req *eio_fsync (int fd, eio_cb cb); 2053eio_req *eio_sync (int pri, eio_cb cb, void *data)
2054{
2055 REQ (EIO_SYNC); SEND;
2056}
2057
2058eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2059{
2060 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2061}
2062
2063eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2064{
2065 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2066}
2067
2068eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2069{
2070 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2071}
2072
2073eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2074{
2075 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2076}
2077
2078eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2079{
2080 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2081}
2082
2083eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2084{
2085 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2086}
2087
2088eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2089{
2090 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2091}
2092
1019eio_req *eio_fdatasync (int fd, eio_cb cb); 2093eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1020eio_req *eio_dupclose (int fd, eio_cb cb); 2094{
2095 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2096}
2097
2098eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2099{
2100 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2101}
2102
1021eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb); 2103eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1022eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2104{
1023eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2105 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1024eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2106}
2107
2108eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2109{
2110 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2111}
2112
2113eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2114{
2115 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2116}
2117
2118eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2119{
2120 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2121}
2122
2123eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2124{
2125 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2126}
2127
1025eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb); 2128eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2129{
2130 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2131}
2132
1026eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb); 2133eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2134{
2135 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2136}
2137
1027eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb); 2138eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2139{
2140 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2141}
2142
1028eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb); 2143eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
2144{
2145 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2146}
2147
1029eio_req *eio_dup2 (int fd, int fd2, eio_cb cb); 2148eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2149{
2150 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2151}
2152
1030eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb); 2153eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2154{
2155 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2156}
2157
1031eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb); 2158eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1032eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2159{
1033eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2160 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1034eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2161}
2162
1035eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb); 2163eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2164{
2165 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2166}
2167
1036eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb); 2168eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1037eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb); 2169{
1038eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb); 2170 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2171}
2172
1039eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb); 2173eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
2174{
2175 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2176}
2177
2178eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2179{
2180 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2181}
2182
2183eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2184{
2185 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2186}
2187
2188static eio_req *
2189eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2190{
2191 REQ (type); PATH; SEND;
2192}
2193
2194eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2195{
2196 return eio__1path (EIO_READLINK, path, pri, cb, data);
2197}
2198
2199eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2200{
2201 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2202}
2203
2204eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2205{
2206 return eio__1path (EIO_STAT, path, pri, cb, data);
2207}
2208
2209eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2210{
2211 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2212}
2213
2214eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2215{
2216 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2217}
2218
1040eio_req *eio_unlink (const char *path, eio_cb cb); 2219eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2220{
2221 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2222}
2223
1041eio_req *eio_rmdir (const char *path, eio_cb cb); 2224eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1042eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2225{
2226 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2227}
2228
2229eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2230{
2231 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2232}
2233
1043eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb); 2234eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1044eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */ 2235{
1045eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */ 2236 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2237}
2238
2239static eio_req *
2240eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2241{
2242 REQ (type); PATH;
2243
2244 req->flags |= EIO_FLAG_PTR2_FREE;
2245 req->ptr2 = strdup (new_path);
2246 if (!req->ptr2)
2247 {
2248 eio_api_destroy (req);
2249 return 0;
2250 }
2251
2252 SEND;
2253}
2254
2255eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2256{
2257 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2258}
2259
2260eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2261{
2262 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2263}
2264
2265eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2266{
2267 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2268}
2269
2270eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2271{
2272 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2273}
2274
2275#endif
2276
2277eio_req *eio_grp (eio_cb cb, void *data)
2278{
2279 const int pri = EIO_PRI_MAX;
2280
2281 REQ (EIO_GROUP); SEND;
2282}
2283
2284#undef REQ
2285#undef PATH
2286#undef SEND
2287
2288/*****************************************************************************/
2289/* grp functions */
2290
1046void 2291void
1047aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1048 PROTOTYPE: $$$;$
1049 PPCODE:
1050{
1051 dREQ;
1052
1053 req->type = EIO_OPEN;
1054 req->sv1 = newSVsv (pathname);
1055 req->ptr1 = SvPVbyte_nolen (req->sv1);
1056 req->int1 = flags;
1057 req->int2 = mode;
1058
1059 EIO_SEND;
1060}
1061
1062void
1063aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1064 PROTOTYPE: $;$
1065 ALIAS:
1066 aio_fsync = EIO_FSYNC
1067 aio_fdatasync = EIO_FDATASYNC
1068 PPCODE:
1069{
1070 dREQ;
1071
1072 req->type = ix;
1073 req->sv1 = newSVsv (fh);
1074 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1075
1076 EIO_SEND (req);
1077}
1078
1079void
1080aio_close (SV *fh, SV *callback=&PL_sv_undef)
1081 PROTOTYPE: $;$
1082 PPCODE:
1083{
1084 dREQ;
1085
1086 req->type = EIO_CLOSE;
1087 req->sv1 = newSVsv (fh);
1088 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1089
1090 EIO_SEND (req);
1091}
1092
1093void
1094aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1095 ALIAS:
1096 aio_read = EIO_READ
1097 aio_write = EIO_WRITE
1098 PROTOTYPE: $$$$$;$
1099 PPCODE:
1100{
1101 STRLEN svlen;
1102 char *svptr = SvPVbyte (data, svlen);
1103 UV len = SvUV (length);
1104
1105 SvUPGRADE (data, SVt_PV);
1106 SvPOK_on (data);
1107
1108 if (dataoffset < 0)
1109 dataoffset += svlen;
1110
1111 if (dataoffset < 0 || dataoffset > svlen)
1112 croak ("dataoffset outside of data scalar");
1113
1114 if (ix == EIO_WRITE)
1115 {
1116 /* write: check length and adjust. */
1117 if (!SvOK (length) || len + dataoffset > svlen)
1118 len = svlen - dataoffset;
1119 }
1120 else
1121 {
1122 /* read: grow scalar as necessary */
1123 svptr = SvGROW (data, len + dataoffset + 1);
1124 }
1125
1126 if (len < 0)
1127 croak ("length must not be negative");
1128
1129 {
1130 dREQ;
1131
1132 req->type = ix;
1133 req->sv1 = newSVsv (fh);
1134 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1135 : IoOFP (sv_2io (fh)));
1136 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1137 req->size = len;
1138 req->sv2 = SvREFCNT_inc (data);
1139 req->ptr2 = (char *)svptr + dataoffset;
1140 req->stroffset = dataoffset;
1141
1142 if (!SvREADONLY (data))
1143 {
1144 SvREADONLY_on (data);
1145 req->flags |= FLAG_SV2_RO_OFF;
1146 }
1147
1148 EIO_SEND;
1149 }
1150}
1151
1152void
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{
1157 SV *data;
1158 dREQ;
1159
1160 data = newSV (NAME_MAX);
1161 SvPOK_on (data);
1162
1163 req->type = EIO_READLINK;
1164 req->sv1 = newSVsv (path);
1165 req->ptr1 = SvPVbyte_nolen (req->sv1);
1166 req->sv2 = data;
1167 req->ptr2 = SvPVbyte_nolen (data);
1168
1169 EIO_SEND;
1170}
1171
1172void
1173aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174 PROTOTYPE: $$$$;$
1175 PPCODE:
1176{
1177 dREQ;
1178
1179 req->type = EIO_SENDFILE;
1180 req->sv1 = newSVsv (out_fh);
1181 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182 req->sv2 = newSVsv (in_fh);
1183 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184 req->offs = SvVAL64 (in_offset);
1185 req->size = length;
1186
1187 EIO_SEND;
1188}
1189
1190void
1191aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192 PROTOTYPE: $$$;$
1193 PPCODE:
1194{
1195 dREQ;
1196
1197 req->type = EIO_READAHEAD;
1198 req->sv1 = newSVsv (fh);
1199 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1200 req->offs = SvVAL64 (offset);
1201 req->size = length;
1202
1203 EIO_SEND;
1204}
1205
1206void
1207aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1208 ALIAS:
1209 aio_stat = EIO_STAT
1210 aio_lstat = EIO_LSTAT
1211 PPCODE:
1212{
1213 dREQ;
1214
1215 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1216 if (!req->ptr2)
1217 {
1218 req_destroy (req);
1219 croak ("out of memory during aio_stat statdata allocation");
1220 }
1221
1222 req->flags |= FLAG_PTR2_FREE;
1223 req->sv1 = newSVsv (fh_or_path);
1224
1225 if (SvPOK (fh_or_path))
1226 {
1227 req->type = ix;
1228 req->ptr1 = SvPVbyte_nolen (req->sv1);
1229 }
1230 else
1231 {
1232 req->type = EIO_FSTAT;
1233 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1234 }
1235
1236 EIO_SEND;
1237}
1238
1239void
1240aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1241 PPCODE:
1242{
1243 dREQ;
1244
1245 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1246 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1247 req->sv1 = newSVsv (fh_or_path);
1248
1249 if (SvPOK (fh_or_path))
1250 {
1251 req->type = EIO_UTIME;
1252 req->ptr1 = SvPVbyte_nolen (req->sv1);
1253 }
1254 else
1255 {
1256 req->type = EIO_FUTIME;
1257 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1258 }
1259
1260 EIO_SEND;
1261}
1262
1263void
1264aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1265 PPCODE:
1266{
1267 dREQ;
1268
1269 req->sv1 = newSVsv (fh_or_path);
1270 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1271
1272 if (SvPOK (fh_or_path))
1273 {
1274 req->type = EIO_TRUNCATE;
1275 req->ptr1 = SvPVbyte_nolen (req->sv1);
1276 }
1277 else
1278 {
1279 req->type = EIO_FTRUNCATE;
1280 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1281 }
1282
1283 EIO_SEND;
1284}
1285
1286void
1287aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1288 ALIAS:
1289 aio_chmod = EIO_CHMOD
1290 aio_fchmod = EIO_FCHMOD
1291 aio_mkdir = EIO_MKDIR
1292 PPCODE:
1293{
1294 dREQ;
1295
1296 req->type = type;
1297 req->int2 = mode;
1298 req->sv1 = newSVsv (fh_or_path);
1299
1300 if (ix == EIO_FCHMOD)
1301 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1302 else
1303 req->ptr1 = SvPVbyte_nolen (req->sv1);
1304
1305 EIO_SEND;
1306}
1307
1308void
1309aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1310 PPCODE:
1311{
1312 dREQ;
1313
1314 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1315 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1316 req->sv1 = newSVsv (fh_or_path);
1317
1318 if (SvPOK (fh_or_path))
1319 {
1320 req->type = EIO_CHOWN;
1321 req->ptr1 = SvPVbyte_nolen (req->sv1);
1322 }
1323 else
1324 {
1325 req->type = EIO_FCHOWN;
1326 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1327 }
1328
1329 EIO_SEND;
1330}
1331
1332void
1333aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1334 ALIAS:
1335 aio_unlink = EIO_UNLINK
1336 aio_rmdir = EIO_RMDIR
1337 aio_readdir = EIO_READDIR
1338 PPCODE:
1339{
1340 dREQ;
1341
1342 req->type = ix;
1343 req->sv1 = newSVsv (pathname);
1344 req->ptr1 = SvPVbyte_nolen (req->sv1);
1345
1346 EIO_SEND;
1347}
1348
1349void
1350aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1351 ALIAS:
1352 aio_link = EIO_LINK
1353 aio_symlink = EIO_SYMLINK
1354 aio_rename = EIO_RENAME
1355 PPCODE:
1356{
1357 dREQ;
1358
1359 req->type = ix;
1360 req->sv1 = newSVsv (oldpath);
1361 req->ptr1 = SvPVbyte_nolen (req->sv1);
1362 req->sv2 = newSVsv (newpath);
1363 req->ptr2 = SvPVbyte_nolen (req->sv2);
1364
1365 EIO_SEND;
1366}
1367
1368void
1369aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1370 PPCODE:
1371{
1372 dREQ;
1373
1374 req->type = EIO_MKNOD;
1375 req->sv1 = newSVsv (pathname);
1376 req->ptr1 = SvPVbyte_nolen (req->sv1);
1377 req->int2 = (mode_t)mode;
1378 req->offs = dev;
1379
1380 EIO_SEND;
1381}
1382
1383void
1384aio_busy (double delay, SV *callback=&PL_sv_undef)
1385 PPCODE:
1386{
1387 dREQ;
1388
1389 req->type = EIO_BUSY;
1390 req->nv1 = delay < 0. ? 0. : delay;
1391
1392 EIO_SEND;
1393}
1394
1395void
1396aio_group (SV *callback=&PL_sv_undef)
1397 PROTOTYPE: ;$
1398 PPCODE:
1399{
1400 dREQ;
1401
1402 req->type = EIO_GROUP;
1403
1404 req_send (req);
1405 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1406}
1407
1408void
1409aio_nop (SV *callback=&PL_sv_undef)
1410 ALIAS:
1411 aio_nop = EIO_NOP
1412 aio_sync = EIO_SYNC
1413 PPCODE:
1414{
1415 dREQ;
1416
1417#endif
1418
1419void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2292eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 2293{
1421 grp->int2 = limit; 2294 grp->int2 = limit;
1422 grp->feed = feed; 2295 grp->feed = feed;
1423 2296
1424 grp_try_feed (grp); 2297 grp_try_feed (grp);
1425} 2298}
1426 2299
2300void
1427void eio_grp_limit (eio_req *grp, int limit) 2301eio_grp_limit (eio_req *grp, int limit)
1428{ 2302{
1429 grp->int2 = limit; 2303 grp->int2 = limit;
1430 2304
1431 grp_try_feed (grp); 2305 grp_try_feed (grp);
1432} 2306}
1433 2307
2308void
1434void eio_grp_add (eio_req *grp, eio_req *req) 2309eio_grp_add (eio_req *grp, eio_req *req)
1435{ 2310{
1436 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2311 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2312
2313 grp->flags |= EIO_FLAG_GROUPADD;
1437 2314
1438 ++grp->size; 2315 ++grp->size;
1439 req->grp = grp; 2316 req->grp = grp;
1440 2317
1441 req->grp_prev = 0; 2318 req->grp_prev = 0;
… …
1445 grp->grp_first->grp_prev = req; 2322 grp->grp_first->grp_prev = req;
1446 2323
1447 grp->grp_first = req; 2324 grp->grp_first = req;
1448} 2325}
1449 2326
2327/*****************************************************************************/
2328/* misc garbage */
1450 2329
2330ssize_t
2331eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2332{
2333 etp_worker wrk;
2334 ssize_t ret;
2335
2336 wrk.dbuf = 0;
2337
2338 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2339
2340 if (wrk.dbuf)
2341 free (wrk.dbuf);
2342
2343 return ret;
2344}
2345

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines