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

Comparing libeio/eio.c (file contents):
Revision 1.4 by root, Sun May 11 00:10:15 2008 UTC vs.
Revision 1.81 by root, Tue Jul 5 20:34:42 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#include <stddef.h> 53#include <stddef.h>
6#include <stdlib.h> 54#include <stdlib.h>
7#include <string.h> 55#include <string.h>
8#include <errno.h> 56#include <errno.h>
9#include <sys/types.h> 57#include <sys/types.h>
10#include <sys/stat.h> 58#include <sys/stat.h>
59#include <sys/statvfs.h>
11#include <limits.h> 60#include <limits.h>
12#include <fcntl.h> 61#include <fcntl.h>
13#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);
14 75
15#ifndef EIO_FINISH 76#ifndef EIO_FINISH
16# 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
17#endif 78#endif
18 79
… …
25#endif 86#endif
26 87
27#ifdef _WIN32 88#ifdef _WIN32
28 89
29 /*doh*/ 90 /*doh*/
30
31#else 91#else
32 92
33# include "config.h"
34# include <sys/time.h> 93# include <sys/time.h>
35# include <sys/select.h> 94# include <sys/select.h>
36# include <unistd.h> 95# include <unistd.h>
37# include <utime.h> 96# include <utime.h>
38# include <signal.h> 97# include <signal.h>
39# include <dirent.h> 98# include <dirent.h>
40 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
121
41# ifndef EIO_STRUCT_DIRENT 122# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 123# define EIO_STRUCT_DIRENT struct dirent
43# endif 124# endif
44 125
45#endif 126#endif
46 127
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif
50
51#if HAVE_SENDFILE 128#if HAVE_SENDFILE
52# if __linux 129# if __linux
53# include <sys/sendfile.h> 130# include <sys/sendfile.h>
54# elif __freebsd 131# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 132# include <sys/socket.h>
56# include <sys/uio.h> 133# include <sys/uio.h>
57# elif __hpux 134# elif __hpux
58# include <sys/socket.h> 135# include <sys/socket.h>
59# elif __solaris /* not yet */ 136# elif __solaris
60# include <sys/sendfile.h> 137# include <sys/sendfile.h>
61# else 138# else
62# error sendfile support requested but not available 139# error sendfile support requested but not available
63# endif 140# endif
64#endif 141#endif
65 142
66/* number of seconds after which an idle threads exit */ 143#ifndef D_TYPE
67#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
68 152
69/* used for struct dirent, AIX doesn't provide it */ 153/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 154#ifndef NAME_MAX
71# define NAME_MAX 4096 155# define NAME_MAX 4096
72#endif 156#endif
73 157
158/* used for readlink etc. */
159#ifndef PATH_MAX
160# define PATH_MAX 4096
161#endif
162
74/* buffer size for various temporary buffers */ 163/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 164#define EIO_BUFSIZE 65536
76 165
77#define dBUF \ 166#define dBUF \
78 char *eio_buf; \ 167 char *eio_buf; \
79 X_LOCK (wrklock); \ 168 ETP_WORKER_LOCK (self); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 169 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 170 ETP_WORKER_UNLOCK (self); \
171 errno = ENOMEM; \
82 if (!eio_buf) \ 172 if (!eio_buf) \
83 return -1; 173 return -1;
84 174
85#define EIO_TICKS ((1000000 + 1023) >> 10) 175#define EIO_TICKS ((1000000 + 1023) >> 10)
86 176
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
87static void (*want_poll_cb) (void); 220static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 221static void (*done_poll_cb) (void);
89 222
90static unsigned int max_poll_time = 0; 223static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 224static unsigned int max_poll_reqs; /* reslock */
92 225
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 226static volatile unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 227static volatile unsigned int nready; /* reqlock */
95{ 228static volatile unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 229static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 230static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 231
100static 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)
101 263
102/* worker threads management */ 264/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 265
105typedef struct worker { 266static void ecb_cold
106 /* locked by wrklock */
107 struct worker *prev, *next;
108
109 thread_t tid;
110
111 /* locked by reslock, reqlock or wrklock */
112 eio_req *req; /* currently processed request */
113 void *dbuf;
114 DIR *dirp;
115} worker;
116
117static worker wrk_first = { &wrk_first, &wrk_first, 0 };
118
119static void worker_clear (worker *wrk) 267etp_worker_clear (etp_worker *wrk)
120{ 268{
121 if (wrk->dirp) 269 ETP_WORKER_CLEAR (wrk);
122 {
123 closedir (wrk->dirp);
124 wrk->dirp = 0;
125 }
126
127 if (wrk->dbuf)
128 {
129 free (wrk->dbuf);
130 wrk->dbuf = 0;
131 }
132} 270}
133 271
134static void worker_free (worker *wrk) 272static void ecb_cold
273etp_worker_free (etp_worker *wrk)
135{ 274{
136 wrk->next->prev = wrk->prev; 275 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 276 wrk->prev->next = wrk->next;
138 277
139 free (wrk); 278 free (wrk);
140} 279}
141 280
142static volatile unsigned int nreqs, nready, npending; 281static unsigned int
143static volatile unsigned int max_idle = 4; 282etp_nreqs (void)
144
145static mutex_t reslock = X_MUTEX_INIT;
146static mutex_t reqlock = X_MUTEX_INIT;
147static cond_t reqwait = X_COND_INIT;
148
149unsigned int eio_nreqs (void)
150{ 283{
284 int retval;
285 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
286 retval = nreqs;
287 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 288 return retval;
152} 289}
153 290
154unsigned int eio_nready (void) 291static unsigned int
292etp_nready (void)
155{ 293{
156 unsigned int retval; 294 unsigned int retval;
157 295
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 296 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 297 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 298 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 299
162 return retval; 300 return retval;
163} 301}
164 302
165unsigned int eio_npending (void) 303static unsigned int
304etp_npending (void)
166{ 305{
167 unsigned int retval; 306 unsigned int retval;
168 307
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 308 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 309 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 310 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 311
173 return retval; 312 return retval;
174} 313}
175 314
176unsigned int eio_nthreads (void) 315static unsigned int
316etp_nthreads (void)
177{ 317{
178 unsigned int retval; 318 unsigned int retval;
179 319
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 320 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 321 retval = started;
… …
188 * a somewhat faster data structure might be nice, but 328 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 329 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 330 * per shift, the most expensive operation.
191 */ 331 */
192typedef struct { 332typedef struct {
193 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 */
194 int size; 334 int size;
195} reqq; 335} etp_reqq;
196 336
197static reqq req_queue; 337static etp_reqq req_queue;
198static reqq res_queue; 338static etp_reqq res_queue;
199 339
200static int reqq_push (reqq *q, eio_req *req) 340static int ecb_noinline
341reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 342{
202 int pri = req->pri; 343 int pri = req->pri;
203 req->next = 0; 344 req->next = 0;
204 345
205 if (q->qe[pri]) 346 if (q->qe[pri])
… …
211 q->qe[pri] = q->qs[pri] = req; 352 q->qe[pri] = q->qs[pri] = req;
212 353
213 return q->size++; 354 return q->size++;
214} 355}
215 356
216static eio_req *reqq_shift (reqq *q) 357static ETP_REQ * ecb_noinline
358reqq_shift (etp_reqq *q)
217{ 359{
218 int pri; 360 int pri;
219 361
220 if (!q->size) 362 if (!q->size)
221 return 0; 363 return 0;
222 364
223 --q->size; 365 --q->size;
224 366
225 for (pri = EIO_NUM_PRI; pri--; ) 367 for (pri = ETP_NUM_PRI; pri--; )
226 { 368 {
227 eio_req *req = q->qs[pri]; 369 eio_req *req = q->qs[pri];
228 370
229 if (req) 371 if (req)
230 { 372 {
… …
236 } 378 }
237 379
238 abort (); 380 abort ();
239} 381}
240 382
241static void grp_try_feed (eio_req *grp) 383static void ecb_cold
384etp_thread_init (void)
242{ 385{
243 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 386 X_MUTEX_CREATE (wrklock);
387 X_MUTEX_CREATE (reslock);
388 X_MUTEX_CREATE (reqlock);
389 X_COND_CREATE (reqwait);
390}
391
392static void ecb_cold
393etp_atfork_prepare (void)
394{
395 X_LOCK (wrklock);
396 X_LOCK (reqlock);
397 X_LOCK (reslock);
398#if !HAVE_PREADWRITE
399 X_LOCK (preadwritelock);
400#endif
401}
402
403static void ecb_cold
404etp_atfork_parent (void)
405{
406#if !HAVE_PREADWRITE
407 X_UNLOCK (preadwritelock);
408#endif
409 X_UNLOCK (reslock);
410 X_UNLOCK (reqlock);
411 X_UNLOCK (wrklock);
412}
413
414static void ecb_cold
415etp_atfork_child (void)
416{
417 ETP_REQ *prv;
418
419 while ((prv = reqq_shift (&req_queue)))
420 ETP_DESTROY (prv);
421
422 while ((prv = reqq_shift (&res_queue)))
423 ETP_DESTROY (prv);
424
425 while (wrk_first.next != &wrk_first)
244 { 426 {
245 int old_len = grp->size; 427 etp_worker *wrk = wrk_first.next;
246 428
247 EIO_FEED (grp); 429 if (wrk->req)
430 ETP_DESTROY (wrk->req);
248 431
249 /* stop if no progress has been made */ 432 etp_worker_clear (wrk);
250 if (old_len == grp->size) 433 etp_worker_free (wrk);
251 {
252 grp->feed = 0;
253 break;
254 }
255 } 434 }
256}
257 435
258static int eio_finish (eio_req *req); 436 started = 0;
437 idle = 0;
438 nreqs = 0;
439 nready = 0;
440 npending = 0;
259 441
260static int grp_dec (eio_req *grp) 442 etp_thread_init ();
261{ 443}
262 --grp->size;
263 444
264 /* call feeder, if applicable */ 445static void ecb_cold
265 grp_try_feed (grp); 446etp_once_init (void)
447{
448 etp_thread_init ();
449 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
450}
266 451
267 /* finish, if done */ 452static int ecb_cold
268 if (!grp->size && grp->int1) 453etp_init (void (*want_poll)(void), void (*done_poll)(void))
269 return eio_finish (grp); 454{
270 else 455 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
456
457 pthread_once (&doinit, etp_once_init);
458
459 want_poll_cb = want_poll;
460 done_poll_cb = done_poll;
461
271 return 0; 462 return 0;
272} 463}
273 464
274void eio_destroy (eio_req *req)
275{
276 if ((req)->flags & EIO_FLAG_PTR2_FREE)
277 free (req->ptr2);
278
279 EIO_DESTROY (req);
280}
281
282static int eio_finish (eio_req *req)
283{
284 int res = EIO_FINISH (req);
285
286 if (req->grp)
287 {
288 int res2;
289 eio_req *grp = req->grp;
290
291 /* unlink request */
292 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
293 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
294
295 if (grp->grp_first == req)
296 grp->grp_first = req->grp_next;
297
298 res2 = grp_dec (grp);
299
300 if (!res && res2)
301 res = res2;
302 }
303
304 eio_destroy (req);
305
306 return res;
307}
308
309void eio_grp_cancel (eio_req *grp)
310{
311 for (grp = grp->grp_first; grp; grp = grp->grp_next)
312 eio_cancel (grp);
313}
314
315void eio_cancel (eio_req *req)
316{
317 req->flags |= EIO_FLAG_CANCELLED;
318
319 eio_grp_cancel (req);
320}
321
322X_THREAD_PROC (eio_proc); 465X_THREAD_PROC (etp_proc);
323 466
324static void start_thread (void) 467static void ecb_cold
468etp_start_thread (void)
325{ 469{
326 worker *wrk = calloc (1, sizeof (worker)); 470 etp_worker *wrk = calloc (1, sizeof (etp_worker));
327 471
328 if (!wrk) 472 /*TODO*/
329 croak ("unable to allocate worker thread data"); 473 assert (("unable to allocate worker thread data", wrk));
330 474
331 X_LOCK (wrklock); 475 X_LOCK (wrklock);
332 476
333 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 477 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334 { 478 {
335 wrk->prev = &wrk_first; 479 wrk->prev = &wrk_first;
336 wrk->next = wrk_first.next; 480 wrk->next = wrk_first.next;
337 wrk_first.next->prev = wrk; 481 wrk_first.next->prev = wrk;
338 wrk_first.next = wrk; 482 wrk_first.next = wrk;
… …
342 free (wrk); 486 free (wrk);
343 487
344 X_UNLOCK (wrklock); 488 X_UNLOCK (wrklock);
345} 489}
346 490
491static void
347static void maybe_start_thread (void) 492etp_maybe_start_thread (void)
348{ 493{
349 if (eio_nthreads () >= wanted) 494 if (ecb_expect_true (etp_nthreads () >= wanted))
350 return; 495 return;
351 496
352 /* todo: maybe use idle here, but might be less exact */ 497 /* todo: maybe use idle here, but might be less exact */
353 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 498 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354 return; 499 return;
355 500
356 start_thread (); 501 etp_start_thread ();
357} 502}
358 503
359void eio_submit (eio_req *req) 504static void ecb_cold
360{ 505etp_end_thread (void)
361 ++nreqs;
362
363 X_LOCK (reqlock);
364 ++nready;
365 reqq_push (&req_queue, req);
366 X_COND_SIGNAL (reqwait);
367 X_UNLOCK (reqlock);
368
369 maybe_start_thread ();
370}
371
372static void end_thread (void)
373{ 506{
374 eio_req *req = calloc (1, sizeof (eio_req)); 507 eio_req *req = calloc (1, sizeof (eio_req));
375 508
376 req->type = EIO_QUIT; 509 req->type = -1;
377 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 510 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
378 511
379 X_LOCK (reqlock); 512 X_LOCK (reqlock);
380 reqq_push (&req_queue, req); 513 reqq_push (&req_queue, req);
381 X_COND_SIGNAL (reqwait); 514 X_COND_SIGNAL (reqwait);
382 X_UNLOCK (reqlock); 515 X_UNLOCK (reqlock);
… …
384 X_LOCK (wrklock); 517 X_LOCK (wrklock);
385 --started; 518 --started;
386 X_UNLOCK (wrklock); 519 X_UNLOCK (wrklock);
387} 520}
388 521
389void eio_set_max_poll_time (double nseconds) 522static int
390{
391 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
392 max_poll_time = nseconds;
393 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
394}
395
396void eio_set_max_poll_reqs (unsigned int maxreqs)
397{
398 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
399 max_poll_reqs = maxreqs;
400 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
401}
402
403void eio_set_max_idle (unsigned int nthreads)
404{
405 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
406 max_idle = nthreads <= 0 ? 1 : nthreads;
407 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
408}
409
410void eio_set_min_parallel (unsigned int nthreads)
411{
412 if (wanted < nthreads)
413 wanted = nthreads;
414}
415
416void eio_set_max_parallel (unsigned int nthreads)
417{
418 if (wanted > nthreads)
419 wanted = nthreads;
420
421 while (started > wanted)
422 end_thread ();
423}
424
425int eio_poll (void) 523etp_poll (void)
426{ 524{
427 int maxreqs = max_poll_reqs; 525 unsigned int maxreqs;
526 unsigned int maxtime;
428 struct timeval tv_start, tv_now; 527 struct timeval tv_start, tv_now;
429 eio_req *req;
430 528
529 X_LOCK (reslock);
530 maxreqs = max_poll_reqs;
531 maxtime = max_poll_time;
532 X_UNLOCK (reslock);
533
431 if (max_poll_time) 534 if (maxtime)
432 gettimeofday (&tv_start, 0); 535 gettimeofday (&tv_start, 0);
433 536
434 for (;;) 537 for (;;)
435 { 538 {
539 ETP_REQ *req;
540
436 maybe_start_thread (); 541 etp_maybe_start_thread ();
437 542
438 X_LOCK (reslock); 543 X_LOCK (reslock);
439 req = reqq_shift (&res_queue); 544 req = reqq_shift (&res_queue);
440 545
441 if (req) 546 if (req)
… …
449 X_UNLOCK (reslock); 554 X_UNLOCK (reslock);
450 555
451 if (!req) 556 if (!req)
452 return 0; 557 return 0;
453 558
559 X_LOCK (reqlock);
454 --nreqs; 560 --nreqs;
561 X_UNLOCK (reqlock);
455 562
456 if (req->type == EIO_GROUP && req->size) 563 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
457 { 564 {
458 req->int1 = 1; /* mark request as delayed */ 565 req->int1 = 1; /* mark request as delayed */
459 continue; 566 continue;
460 } 567 }
461 else 568 else
462 { 569 {
463 int res = eio_finish (req); 570 int res = ETP_FINISH (req);
464 if (res) 571 if (ecb_expect_false (res))
465 return res; 572 return res;
466 } 573 }
467 574
468 if (maxreqs && !--maxreqs) 575 if (ecb_expect_false (maxreqs && !--maxreqs))
469 break; 576 break;
470 577
471 if (max_poll_time) 578 if (maxtime)
472 { 579 {
473 gettimeofday (&tv_now, 0); 580 gettimeofday (&tv_now, 0);
474 581
475 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 582 if (tvdiff (&tv_start, &tv_now) >= maxtime)
476 break; 583 break;
477 } 584 }
478 } 585 }
479 586
480 errno = EAGAIN; 587 errno = EAGAIN;
481 return -1; 588 return -1;
482} 589}
483 590
591static void
592etp_cancel (ETP_REQ *req)
593{
594 req->cancelled = 1;
595
596 eio_grp_cancel (req);
597}
598
599static void
600etp_submit (ETP_REQ *req)
601{
602 req->pri -= ETP_PRI_MIN;
603
604 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
605 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
606
607 if (ecb_expect_false (req->type == EIO_GROUP))
608 {
609 /* I hope this is worth it :/ */
610 X_LOCK (reqlock);
611 ++nreqs;
612 X_UNLOCK (reqlock);
613
614 X_LOCK (reslock);
615
616 ++npending;
617
618 if (!reqq_push (&res_queue, req) && want_poll_cb)
619 want_poll_cb ();
620
621 X_UNLOCK (reslock);
622 }
623 else
624 {
625 X_LOCK (reqlock);
626 ++nreqs;
627 ++nready;
628 reqq_push (&req_queue, req);
629 X_COND_SIGNAL (reqwait);
630 X_UNLOCK (reqlock);
631
632 etp_maybe_start_thread ();
633 }
634}
635
636static void ecb_cold
637etp_set_max_poll_time (double nseconds)
638{
639 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
640 max_poll_time = nseconds * EIO_TICKS;
641 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
642}
643
644static void ecb_cold
645etp_set_max_poll_reqs (unsigned int maxreqs)
646{
647 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
648 max_poll_reqs = maxreqs;
649 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
650}
651
652static void ecb_cold
653etp_set_max_idle (unsigned int nthreads)
654{
655 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
656 max_idle = nthreads;
657 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
658}
659
660static void ecb_cold
661etp_set_idle_timeout (unsigned int seconds)
662{
663 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
664 idle_timeout = seconds;
665 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
666}
667
668static void ecb_cold
669etp_set_min_parallel (unsigned int nthreads)
670{
671 if (wanted < nthreads)
672 wanted = nthreads;
673}
674
675static void ecb_cold
676etp_set_max_parallel (unsigned int nthreads)
677{
678 if (wanted > nthreads)
679 wanted = nthreads;
680
681 while (started > wanted)
682 etp_end_thread ();
683}
684
685/*****************************************************************************/
686
687static void
688grp_try_feed (eio_req *grp)
689{
690 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
691 {
692 grp->flags &= ~EIO_FLAG_GROUPADD;
693
694 EIO_FEED (grp);
695
696 /* stop if no progress has been made */
697 if (!(grp->flags & EIO_FLAG_GROUPADD))
698 {
699 grp->feed = 0;
700 break;
701 }
702 }
703}
704
705static int
706grp_dec (eio_req *grp)
707{
708 --grp->size;
709
710 /* call feeder, if applicable */
711 grp_try_feed (grp);
712
713 /* finish, if done */
714 if (!grp->size && grp->int1)
715 return eio_finish (grp);
716 else
717 return 0;
718}
719
720static void
721eio_destroy (eio_req *req)
722{
723 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
724 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
725
726 EIO_DESTROY (req);
727}
728
729static int
730eio_finish (eio_req *req)
731{
732 int res = EIO_FINISH (req);
733
734 if (req->grp)
735 {
736 int res2;
737 eio_req *grp = req->grp;
738
739 /* unlink request */
740 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
741 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
742
743 if (grp->grp_first == req)
744 grp->grp_first = req->grp_next;
745
746 res2 = grp_dec (grp);
747
748 if (!res)
749 res = res2;
750 }
751
752 eio_destroy (req);
753
754 return res;
755}
756
757void
758eio_grp_cancel (eio_req *grp)
759{
760 for (grp = grp->grp_first; grp; grp = grp->grp_next)
761 eio_cancel (grp);
762}
763
764void
765eio_cancel (eio_req *req)
766{
767 etp_cancel (req);
768}
769
770void
771eio_submit (eio_req *req)
772{
773 etp_submit (req);
774}
775
776unsigned int
777eio_nreqs (void)
778{
779 return etp_nreqs ();
780}
781
782unsigned int
783eio_nready (void)
784{
785 return etp_nready ();
786}
787
788unsigned int
789eio_npending (void)
790{
791 return etp_npending ();
792}
793
794unsigned int ecb_cold
795eio_nthreads (void)
796{
797 return etp_nthreads ();
798}
799
800void ecb_cold
801eio_set_max_poll_time (double nseconds)
802{
803 etp_set_max_poll_time (nseconds);
804}
805
806void ecb_cold
807eio_set_max_poll_reqs (unsigned int maxreqs)
808{
809 etp_set_max_poll_reqs (maxreqs);
810}
811
812void ecb_cold
813eio_set_max_idle (unsigned int nthreads)
814{
815 etp_set_max_idle (nthreads);
816}
817
818void ecb_cold
819eio_set_idle_timeout (unsigned int seconds)
820{
821 etp_set_idle_timeout (seconds);
822}
823
824void ecb_cold
825eio_set_min_parallel (unsigned int nthreads)
826{
827 etp_set_min_parallel (nthreads);
828}
829
830void ecb_cold
831eio_set_max_parallel (unsigned int nthreads)
832{
833 etp_set_max_parallel (nthreads);
834}
835
836int eio_poll (void)
837{
838 return etp_poll ();
839}
840
484/*****************************************************************************/ 841/*****************************************************************************/
485/* work around various missing functions */ 842/* work around various missing functions */
486 843
487#if !HAVE_PREADWRITE 844#if !HAVE_PREADWRITE
845# undef pread
846# undef pwrite
488# define pread aio_pread 847# define pread eio__pread
489# define pwrite aio_pwrite 848# define pwrite eio__pwrite
490 849
491/* 850static ssize_t
492 * make our pread/pwrite safe against themselves, but not against 851eio__pread (int fd, void *buf, size_t count, off_t offset)
493 * normal read/write by using a mutex. slows down execution a lot,
494 * but that's your problem, not mine.
495 */
496static mutex_t preadwritelock = X_MUTEX_INIT;
497
498static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
499{ 852{
500 ssize_t res; 853 ssize_t res;
501 off_t ooffset; 854 off_t ooffset;
502 855
503 X_LOCK (preadwritelock); 856 X_LOCK (preadwritelock);
… …
508 X_UNLOCK (preadwritelock); 861 X_UNLOCK (preadwritelock);
509 862
510 return res; 863 return res;
511} 864}
512 865
513static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 866static ssize_t
867eio__pwrite (int fd, void *buf, size_t count, off_t offset)
514{ 868{
515 ssize_t res; 869 ssize_t res;
516 off_t ooffset; 870 off_t ooffset;
517 871
518 X_LOCK (preadwritelock); 872 X_LOCK (preadwritelock);
519 ooffset = lseek (fd, 0, SEEK_CUR); 873 ooffset = lseek (fd, 0, SEEK_CUR);
520 lseek (fd, offset, SEEK_SET); 874 lseek (fd, offset, SEEK_SET);
521 res = write (fd, buf, count); 875 res = write (fd, buf, count);
522 lseek (fd, offset, SEEK_SET); 876 lseek (fd, ooffset, SEEK_SET);
523 X_UNLOCK (preadwritelock); 877 X_UNLOCK (preadwritelock);
524 878
525 return res; 879 return res;
526} 880}
527#endif 881#endif
528 882
529#ifndef HAVE_FUTIMES 883#ifndef HAVE_UTIMES
530 884
885# undef utimes
531# define utimes(path,times) aio_utimes (path, times) 886# define utimes(path,times) eio__utimes (path, times)
532# define futimes(fd,times) aio_futimes (fd, times)
533 887
888static int
534static int aio_utimes (const char *filename, const struct timeval times[2]) 889eio__utimes (const char *filename, const struct timeval times[2])
535{ 890{
536 if (times) 891 if (times)
537 { 892 {
538 struct utimbuf buf; 893 struct utimbuf buf;
539 894
… …
544 } 899 }
545 else 900 else
546 return utime (filename, 0); 901 return utime (filename, 0);
547} 902}
548 903
904#endif
905
906#ifndef HAVE_FUTIMES
907
908# undef futimes
909# define futimes(fd,times) eio__futimes (fd, times)
910
911static int
549static int aio_futimes (int fd, const struct timeval tv[2]) 912eio__futimes (int fd, const struct timeval tv[2])
550{ 913{
551 errno = ENOSYS; 914 errno = ENOSYS;
552 return -1; 915 return -1;
553} 916}
554 917
555#endif 918#endif
556 919
557#if !HAVE_FDATASYNC 920#if !HAVE_FDATASYNC
921# undef fdatasync
558# define fdatasync fsync 922# define fdatasync(fd) fsync (fd)
559#endif 923#endif
924
925/* sync_file_range always needs emulation */
926static int
927eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
928{
929#if HAVE_SYNC_FILE_RANGE
930 int res;
931
932 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
933 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
934 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
935 {
936 flags = 0
937 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
938 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
939 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
940 }
941
942 res = sync_file_range (fd, offset, nbytes, flags);
943
944 if (!res || errno != ENOSYS)
945 return res;
946#endif
947
948 /* even though we could play tricks with the flags, it's better to always
949 * call fdatasync, as that matches the expectation of its users best */
950 return fdatasync (fd);
951}
560 952
561#if !HAVE_READAHEAD 953#if !HAVE_READAHEAD
954# undef readahead
562# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 955# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
563 956
957static ssize_t
564static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 958eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
565{ 959{
566 size_t todo = count; 960 size_t todo = count;
567 dBUF; 961 dBUF;
568 962
569 while (todo > 0) 963 while (todo > 0)
… …
579 return count; 973 return count;
580} 974}
581 975
582#endif 976#endif
583 977
584#if !HAVE_READDIR_R
585# define readdir_r aio_readdir_r
586
587static mutex_t readdirlock = X_MUTEX_INIT;
588
589static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
590{
591 X_DIRENT *e;
592 int errorno;
593
594 X_LOCK (readdirlock);
595
596 e = readdir (dirp);
597 errorno = errno;
598
599 if (e)
600 {
601 *res = ent;
602 strcpy (ent->d_name, e->d_name);
603 }
604 else
605 *res = 0;
606
607 X_UNLOCK (readdirlock);
608
609 errno = errorno;
610 return e ? 0 : -1;
611}
612#endif
613
614/* sendfile always needs emulation */ 978/* sendfile always needs emulation */
979static ssize_t
615static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 980eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
616{ 981{
982 ssize_t written = 0;
617 ssize_t res; 983 ssize_t res;
618 984
619 if (!count) 985 if (!count)
620 return 0; 986 return 0;
621 987
988 for (;;)
989 {
622#if HAVE_SENDFILE 990#if HAVE_SENDFILE
623# if __linux 991# if __linux
992 off_t soffset = offset;
624 res = sendfile (ofd, ifd, &offset, count); 993 res = sendfile (ofd, ifd, &soffset, count);
625 994
626# elif __freebsd 995# elif __FreeBSD__
627 /* 996 /*
628 * Of course, the freebsd sendfile is a dire hack with no thoughts 997 * Of course, the freebsd sendfile is a dire hack with no thoughts
629 * wasted on making it similar to other I/O functions. 998 * wasted on making it similar to other I/O functions.
630 */ 999 */
631 {
632 off_t sbytes; 1000 off_t sbytes;
633 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1001 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
634 1002
635 if (res < 0 && sbytes) 1003 #if 0 /* according to the manpage, this is correct, but broken behaviour */
636 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1004 /* freebsd' sendfile will return 0 on success */
1005 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1006 /* not on e.g. EIO or EPIPE - sounds broken */
1007 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
637 res = sbytes; 1008 res = sbytes;
638 } 1009 #endif
1010
1011 /* according to source inspection, this is correct, and useful behaviour */
1012 if (sbytes)
1013 res = sbytes;
1014
1015# elif defined (__APPLE__) && 0 /* broken, as everything on os x */
1016 off_t sbytes = count;
1017 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1018
1019 /* according to the manpage, sbytes is always valid */
1020 if (sbytes)
1021 res = sbytes;
639 1022
640# elif __hpux 1023# elif __hpux
641 res = sendfile (ofd, ifd, offset, count, 0, 0); 1024 res = sendfile (ofd, ifd, offset, count, 0, 0);
642 1025
643# elif __solaris 1026# elif __solaris
644 {
645 struct sendfilevec vec; 1027 struct sendfilevec vec;
646 size_t sbytes; 1028 size_t sbytes;
647 1029
648 vec.sfv_fd = ifd; 1030 vec.sfv_fd = ifd;
649 vec.sfv_flag = 0; 1031 vec.sfv_flag = 0;
650 vec.sfv_off = offset; 1032 vec.sfv_off = offset;
651 vec.sfv_len = count; 1033 vec.sfv_len = count;
652 1034
653 res = sendfilev (ofd, &vec, 1, &sbytes); 1035 res = sendfilev (ofd, &vec, 1, &sbytes);
654 1036
655 if (res < 0 && sbytes) 1037 if (res < 0 && sbytes)
656 res = sbytes; 1038 res = sbytes;
657 }
658 1039
659# endif 1040# endif
1041
1042#elif defined (_WIN32)
1043 /* does not work, just for documentation of what would need to be done */
1044 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1045 /* libeio guarantees that the file offset does not change, and windows */
1046 /* has no way to get an independent handle to the same file description */
1047 HANDLE h = TO_SOCKET (ifd);
1048 SetFilePointer (h, offset, 0, FILE_BEGIN);
1049 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1050
660#else 1051#else
661 res = -1; 1052 res = -1;
662 errno = ENOSYS; 1053 errno = ENOSYS;
663#endif 1054#endif
664 1055
1056 /* we assume sendfile can copy at least 128mb in one go */
1057 if (res <= 128 * 1024 * 1024)
1058 {
1059 if (res > 0)
1060 written += res;
1061
1062 if (written)
1063 return written;
1064
1065 break;
1066 }
1067 else
1068 {
1069 /* if we requested more, then probably the kernel was lazy */
1070 written += res;
1071 offset += res;
1072 count -= res;
1073
1074 if (!count)
1075 return written;
1076 }
1077 }
1078
665 if (res < 0 1079 if (res < 0
666 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1080 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1081 /* BSDs */
1082#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1083 || errno == ENOTSUP
1084#endif
1085 || errno == EOPNOTSUPP /* BSDs */
667#if __solaris 1086#if __solaris
668 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1087 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
669#endif 1088#endif
670 ) 1089 )
671 ) 1090 )
… …
702 } 1121 }
703 1122
704 return res; 1123 return res;
705} 1124}
706 1125
1126#ifdef PAGESIZE
1127# define eio_pagesize() PAGESIZE
1128#else
1129static intptr_t
1130eio_pagesize (void)
1131{
1132 static intptr_t page;
1133
1134 if (!page)
1135 page = sysconf (_SC_PAGESIZE);
1136
1137 return page;
1138}
1139#endif
1140
1141static void
1142eio_page_align (void **addr, size_t *length)
1143{
1144 intptr_t mask = eio_pagesize () - 1;
1145
1146 /* round down addr */
1147 intptr_t adj = mask & (intptr_t)*addr;
1148
1149 *addr = (void *)((intptr_t)*addr - adj);
1150 *length += adj;
1151
1152 /* round up length */
1153 *length = (*length + mask) & ~mask;
1154}
1155
1156#if !_POSIX_MEMLOCK
1157# define eio__mlockall(a) ((errno = ENOSYS), -1)
1158#else
1159
1160static int
1161eio__mlockall (int flags)
1162{
1163 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1164 extern int mallopt (int, int);
1165 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1166 #endif
1167
1168 if (EIO_MCL_CURRENT != MCL_CURRENT
1169 || EIO_MCL_FUTURE != MCL_FUTURE)
1170 {
1171 flags = 0
1172 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1173 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1174 }
1175
1176 return mlockall (flags);
1177}
1178#endif
1179
1180#if !_POSIX_MEMLOCK_RANGE
1181# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1182#else
1183
1184static int
1185eio__mlock (void *addr, size_t length)
1186{
1187 eio_page_align (&addr, &length);
1188
1189 return mlock (addr, length);
1190}
1191
1192#endif
1193
1194#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1195# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1196#else
1197
1198static int
1199eio__msync (void *mem, size_t len, int flags)
1200{
1201 eio_page_align (&mem, &len);
1202
1203 if (EIO_MS_ASYNC != MS_SYNC
1204 || EIO_MS_INVALIDATE != MS_INVALIDATE
1205 || EIO_MS_SYNC != MS_SYNC)
1206 {
1207 flags = 0
1208 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1209 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1210 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1211 }
1212
1213 return msync (mem, len, flags);
1214}
1215
1216#endif
1217
1218static int
1219eio__mtouch (eio_req *req)
1220{
1221 void *mem = req->ptr2;
1222 size_t len = req->size;
1223 int flags = req->int1;
1224
1225 eio_page_align (&mem, &len);
1226
1227 {
1228 intptr_t addr = (intptr_t)mem;
1229 intptr_t end = addr + len;
1230 intptr_t page = eio_pagesize ();
1231
1232 if (addr < end)
1233 if (flags & EIO_MT_MODIFY) /* modify */
1234 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1235 else
1236 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1237 }
1238
1239 return 0;
1240}
1241
1242/*****************************************************************************/
1243/* requests implemented outside eio_execute, because they are so large */
1244
1245static void
1246eio__realpath (eio_req *req, etp_worker *self)
1247{
1248 char *rel = req->ptr1;
1249 char *res;
1250 char *tmp1, *tmp2;
1251#if SYMLOOP_MAX > 32
1252 int symlinks = SYMLOOP_MAX;
1253#else
1254 int symlinks = 32;
1255#endif
1256
1257 req->result = -1;
1258
1259 errno = EINVAL;
1260 if (!rel)
1261 return;
1262
1263 errno = ENOENT;
1264 if (!*rel)
1265 return;
1266
1267 if (!req->ptr2)
1268 {
1269 X_LOCK (wrklock);
1270 req->flags |= EIO_FLAG_PTR2_FREE;
1271 X_UNLOCK (wrklock);
1272 req->ptr2 = malloc (PATH_MAX * 3);
1273
1274 errno = ENOMEM;
1275 if (!req->ptr2)
1276 return;
1277 }
1278
1279 res = req->ptr2;
1280 tmp1 = res + PATH_MAX;
1281 tmp2 = tmp1 + PATH_MAX;
1282
1283#if 0 /* disabled, the musl way to do things is just too racy */
1284#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1285 /* on linux we may be able to ask the kernel */
1286 {
1287 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1288
1289 if (fd >= 0)
1290 {
1291 sprintf (tmp1, "/proc/self/fd/%d", fd);
1292 req->result = readlink (tmp1, res, PATH_MAX);
1293 close (fd);
1294
1295 /* here we should probably stat the open file and the disk file, to make sure they still match */
1296
1297 if (req->result > 0)
1298 goto done;
1299 }
1300 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1301 return;
1302 }
1303#endif
1304#endif
1305
1306 if (*rel != '/')
1307 {
1308 if (!getcwd (res, PATH_MAX))
1309 return;
1310
1311 if (res [1]) /* only use if not / */
1312 res += strlen (res);
1313 }
1314
1315 while (*rel)
1316 {
1317 ssize_t len, linklen;
1318 char *beg = rel;
1319
1320 while (*rel && *rel != '/')
1321 ++rel;
1322
1323 len = rel - beg;
1324
1325 if (!len) /* skip slashes */
1326 {
1327 ++rel;
1328 continue;
1329 }
1330
1331 if (beg [0] == '.')
1332 {
1333 if (len == 1)
1334 continue; /* . - nop */
1335
1336 if (beg [1] == '.' && len == 2)
1337 {
1338 /* .. - back up one component, if possible */
1339
1340 while (res != req->ptr2)
1341 if (*--res == '/')
1342 break;
1343
1344 continue;
1345 }
1346 }
1347
1348 errno = ENAMETOOLONG;
1349 if (res + 1 + len + 1 >= tmp1)
1350 return;
1351
1352 /* copy one component */
1353 *res = '/';
1354 memcpy (res + 1, beg, len);
1355
1356 /* zero-terminate, for readlink */
1357 res [len + 1] = 0;
1358
1359 /* now check if it's a symlink */
1360 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1361
1362 if (linklen < 0)
1363 {
1364 if (errno != EINVAL)
1365 return;
1366
1367 /* it's a normal directory. hopefully */
1368 res += len + 1;
1369 }
1370 else
1371 {
1372 /* yay, it was a symlink - build new path in tmp2 */
1373 int rellen = strlen (rel);
1374
1375 errno = ENAMETOOLONG;
1376 if (linklen + 1 + rellen >= PATH_MAX)
1377 return;
1378
1379 errno = ELOOP;
1380 if (!--symlinks)
1381 return;
1382
1383 if (*tmp1 == '/')
1384 res = req->ptr2; /* symlink resolves to an absolute path */
1385
1386 /* we need to be careful, as rel might point into tmp2 already */
1387 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1388 tmp2 [linklen] = '/';
1389 memcpy (tmp2, tmp1, linklen);
1390
1391 rel = tmp2;
1392 }
1393 }
1394
1395 /* special case for the lone root path */
1396 if (res == req->ptr2)
1397 *res++ = '/';
1398
1399 req->result = res - (char *)req->ptr2;
1400
1401done:
1402 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1403}
1404
1405static signed char
1406eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1407{
1408 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1409 : a->inode < b->inode ? -1
1410 : a->inode > b->inode ? 1
1411 : 0;
1412}
1413
1414#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1415
1416#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1417#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1418
1419static void
1420eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1421{
1422 unsigned char bits [9 + sizeof (ino_t) * 8];
1423 unsigned char *bit = bits;
1424
1425 assert (CHAR_BIT == 8);
1426 assert (sizeof (eio_dirent) * 8 < 256);
1427 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1428 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1429
1430 if (size <= EIO_SORT_FAST)
1431 return;
1432
1433 /* first prepare an array of bits to test in our radix sort */
1434 /* try to take endianness into account, as well as differences in ino_t sizes */
1435 /* inode_bits must contain all inodes ORed together */
1436 /* which is used to skip bits that are 0 everywhere, which is very common */
1437 {
1438 ino_t endianness;
1439 int i, j;
1440
1441 /* we store the byte offset of byte n into byte n of "endianness" */
1442 for (i = 0; i < sizeof (ino_t); ++i)
1443 ((unsigned char *)&endianness)[i] = i;
1444
1445 *bit++ = 0;
1446
1447 for (i = 0; i < sizeof (ino_t); ++i)
1448 {
1449 /* shifting off the byte offsets out of "endianness" */
1450 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1451 endianness >>= 8;
1452
1453 for (j = 0; j < 8; ++j)
1454 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1455 *bit++ = offs + j;
1456 }
1457
1458 for (j = 0; j < 8; ++j)
1459 if (score_bits & (1 << j))
1460 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1461 }
1462
1463 /* now actually do the sorting (a variant of MSD radix sort) */
1464 {
1465 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1466 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1467 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1468 int stk_idx = 0;
1469
1470 base_stk [stk_idx] = dents;
1471 end_stk [stk_idx] = dents + size;
1472 bit_stk [stk_idx] = bit - 1;
1473
1474 do
1475 {
1476 base = base_stk [stk_idx];
1477 end = end_stk [stk_idx];
1478 bit = bit_stk [stk_idx];
1479
1480 for (;;)
1481 {
1482 unsigned char O = *bit >> 3;
1483 unsigned char M = 1 << (*bit & 7);
1484
1485 eio_dirent *a = base;
1486 eio_dirent *b = end;
1487
1488 if (b - a < EIO_SORT_CUTOFF)
1489 break;
1490
1491 /* now bit-partition the array on the bit */
1492 /* this ugly asymmetric loop seems to perform much better than typical */
1493 /* partition algos found in the literature */
1494 do
1495 if (!(((unsigned char *)a)[O] & M))
1496 ++a;
1497 else if (!(((unsigned char *)--b)[O] & M))
1498 {
1499 eio_dirent tmp = *a; *a = *b; *b = tmp;
1500 ++a;
1501 }
1502 while (b > a);
1503
1504 /* next bit, or stop, if no bits left in this path */
1505 if (!*--bit)
1506 break;
1507
1508 base_stk [stk_idx] = a;
1509 end_stk [stk_idx] = end;
1510 bit_stk [stk_idx] = bit;
1511 ++stk_idx;
1512
1513 end = a;
1514 }
1515 }
1516 while (stk_idx--);
1517 }
1518}
1519
1520static void
1521eio_dent_insertion_sort (eio_dirent *dents, int size)
1522{
1523 /* first move the smallest element to the front, to act as a sentinel */
1524 {
1525 int i;
1526 eio_dirent *min = dents;
1527
1528 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1529 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1530 if (EIO_DENT_CMP (dents [i], <, *min))
1531 min = &dents [i];
1532
1533 /* swap elements 0 and j (minimum) */
1534 {
1535 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1536 }
1537 }
1538
1539 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1540 {
1541 eio_dirent *i, *j;
1542
1543 for (i = dents + 1; i < dents + size; ++i)
1544 {
1545 eio_dirent value = *i;
1546
1547 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1548 j [1] = j [0];
1549
1550 j [1] = value;
1551 }
1552 }
1553}
1554
1555static void
1556eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1557{
1558 if (size <= 1)
1559 return; /* our insertion sort relies on size > 0 */
1560
1561 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1562 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1563 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1564
1565 /* use an insertion sort at the end, or for small arrays, */
1566 /* as insertion sort is more efficient for small partitions */
1567 eio_dent_insertion_sort (dents, size);
1568}
1569
707/* read a full directory */ 1570/* read a full directory */
1571static void
708static void scandir_ (eio_req *req, worker *self) 1572eio__scandir (eio_req *req, etp_worker *self)
709{ 1573{
710 DIR *dirp; 1574 DIR *dirp;
711 union
712 {
713 EIO_STRUCT_DIRENT d;
714 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
715 } *u;
716 EIO_STRUCT_DIRENT *entp; 1575 EIO_STRUCT_DIRENT *entp;
717 char *name, *names; 1576 char *name, *names;
718 int memlen = 4096; 1577 int namesalloc = 4096;
719 int memofs = 0; 1578 int namesoffs = 0;
1579 int flags = req->int1;
1580 eio_dirent *dents = 0;
1581 int dentalloc = 128;
720 int res = 0; 1582 int dentoffs = 0;
1583 ino_t inode_bits = 0;
1584
1585 req->result = -1;
1586
1587 if (!(flags & EIO_READDIR_DENTS))
1588 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
721 1589
722 X_LOCK (wrklock); 1590 X_LOCK (wrklock);
1591 /* the corresponding closedir is in ETP_WORKER_CLEAR */
723 self->dirp = dirp = opendir (req->ptr1); 1592 self->dirp = dirp = opendir (req->ptr1);
724 self->dbuf = u = malloc (sizeof (*u)); 1593
725 req->flags |= EIO_FLAG_PTR2_FREE; 1594 if (req->flags & EIO_FLAG_PTR1_FREE)
1595 free (req->ptr1);
1596
1597 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1598 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
726 req->ptr2 = names = malloc (memlen); 1599 req->ptr2 = names = malloc (namesalloc);
727 X_UNLOCK (wrklock); 1600 X_UNLOCK (wrklock);
728 1601
729 if (dirp && u && names) 1602 if (dirp && names && (!flags || dents))
730 for (;;) 1603 for (;;)
731 { 1604 {
732 errno = 0; 1605 errno = 0;
733 readdir_r (dirp, &u->d, &entp); 1606 entp = readdir (dirp);
734 1607
735 if (!entp) 1608 if (!entp)
1609 {
1610 if (errno)
1611 break;
1612
1613 /* sort etc. */
1614 req->int1 = flags;
1615 req->result = dentoffs;
1616
1617 if (flags & EIO_READDIR_STAT_ORDER)
1618 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1619 else if (flags & EIO_READDIR_DIRS_FIRST)
1620 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1621 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1622 else
1623 {
1624 /* in this case, all is known, and we just put dirs first and sort them */
1625 eio_dirent *oth = dents + dentoffs;
1626 eio_dirent *dir = dents;
1627
1628 /* now partition dirs to the front, and non-dirs to the back */
1629 /* by walking from both sides and swapping if necessary */
1630 while (oth > dir)
1631 {
1632 if (dir->type == EIO_DT_DIR)
1633 ++dir;
1634 else if ((--oth)->type == EIO_DT_DIR)
1635 {
1636 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1637
1638 ++dir;
1639 }
1640 }
1641
1642 /* now sort the dirs only (dirs all have the same score) */
1643 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1644 }
1645
736 break; 1646 break;
1647 }
737 1648
1649 /* now add the entry to our list(s) */
738 name = entp->d_name; 1650 name = entp->d_name;
739 1651
1652 /* skip . and .. entries */
740 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1653 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
741 { 1654 {
742 int len = strlen (name) + 1; 1655 int len = D_NAMLEN (entp) + 1;
743 1656
744 res++; 1657 while (ecb_expect_false (namesoffs + len > namesalloc))
745
746 while (memofs + len > memlen)
747 { 1658 {
748 memlen *= 2; 1659 namesalloc *= 2;
749 X_LOCK (wrklock); 1660 X_LOCK (wrklock);
750 req->ptr2 = names = realloc (names, memlen); 1661 req->ptr2 = names = realloc (names, namesalloc);
751 X_UNLOCK (wrklock); 1662 X_UNLOCK (wrklock);
752 1663
753 if (!names) 1664 if (!names)
754 break; 1665 break;
755 } 1666 }
756 1667
757 memcpy (names + memofs, name, len); 1668 memcpy (names + namesoffs, name, len);
1669
1670 if (dents)
1671 {
1672 struct eio_dirent *ent;
1673
1674 if (ecb_expect_false (dentoffs == dentalloc))
1675 {
1676 dentalloc *= 2;
1677 X_LOCK (wrklock);
1678 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1679 X_UNLOCK (wrklock);
1680
1681 if (!dents)
1682 break;
1683 }
1684
1685 ent = dents + dentoffs;
1686
1687 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1688 ent->namelen = len - 1;
1689 ent->inode = D_INO (entp);
1690
1691 inode_bits |= ent->inode;
1692
1693 switch (D_TYPE (entp))
1694 {
1695 default:
1696 ent->type = EIO_DT_UNKNOWN;
1697 flags |= EIO_READDIR_FOUND_UNKNOWN;
1698 break;
1699
1700 #ifdef DT_FIFO
1701 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1702 #endif
1703 #ifdef DT_CHR
1704 case DT_CHR: ent->type = EIO_DT_CHR; break;
1705 #endif
1706 #ifdef DT_MPC
1707 case DT_MPC: ent->type = EIO_DT_MPC; break;
1708 #endif
1709 #ifdef DT_DIR
1710 case DT_DIR: ent->type = EIO_DT_DIR; break;
1711 #endif
1712 #ifdef DT_NAM
1713 case DT_NAM: ent->type = EIO_DT_NAM; break;
1714 #endif
1715 #ifdef DT_BLK
1716 case DT_BLK: ent->type = EIO_DT_BLK; break;
1717 #endif
1718 #ifdef DT_MPB
1719 case DT_MPB: ent->type = EIO_DT_MPB; break;
1720 #endif
1721 #ifdef DT_REG
1722 case DT_REG: ent->type = EIO_DT_REG; break;
1723 #endif
1724 #ifdef DT_NWK
1725 case DT_NWK: ent->type = EIO_DT_NWK; break;
1726 #endif
1727 #ifdef DT_CMP
1728 case DT_CMP: ent->type = EIO_DT_CMP; break;
1729 #endif
1730 #ifdef DT_LNK
1731 case DT_LNK: ent->type = EIO_DT_LNK; break;
1732 #endif
1733 #ifdef DT_SOCK
1734 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1735 #endif
1736 #ifdef DT_DOOR
1737 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1738 #endif
1739 #ifdef DT_WHT
1740 case DT_WHT: ent->type = EIO_DT_WHT; break;
1741 #endif
1742 }
1743
1744 ent->score = 7;
1745
1746 if (flags & EIO_READDIR_DIRS_FIRST)
1747 {
1748 if (ent->type == EIO_DT_UNKNOWN)
1749 {
1750 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1751 ent->score = 1;
1752 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1753 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1754 }
1755 else if (ent->type == EIO_DT_DIR)
1756 ent->score = 0;
1757 }
1758 }
1759
758 memofs += len; 1760 namesoffs += len;
1761 ++dentoffs;
1762 }
1763
1764 if (EIO_CANCELLED (req))
1765 {
1766 errno = ECANCELED;
1767 break;
759 } 1768 }
760 } 1769 }
761
762 if (errno)
763 res = -1;
764
765 req->result = res;
766} 1770}
767 1771
768/*****************************************************************************/ 1772/*****************************************************************************/
769 1773
1774#define ALLOC(len) \
1775 if (!req->ptr2) \
1776 { \
1777 X_LOCK (wrklock); \
1778 req->flags |= EIO_FLAG_PTR2_FREE; \
1779 X_UNLOCK (wrklock); \
1780 req->ptr2 = malloc (len); \
1781 if (!req->ptr2) \
1782 { \
1783 errno = ENOMEM; \
1784 req->result = -1; \
1785 break; \
1786 } \
1787 }
1788
770X_THREAD_PROC (eio_proc) 1789X_THREAD_PROC (etp_proc)
771{ 1790{
772 eio_req *req; 1791 ETP_REQ *req;
773 struct timespec ts; 1792 struct timespec ts;
774 worker *self = (worker *)thr_arg; 1793 etp_worker *self = (etp_worker *)thr_arg;
775 1794
776 /* try to distribute timeouts somewhat randomly */ 1795 /* try to distribute timeouts somewhat randomly */
777 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1796 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
778 1797
779 for (;;) 1798 for (;;)
780 { 1799 {
781 ts.tv_sec = time (0) + IDLE_TIMEOUT;
782
783 X_LOCK (reqlock); 1800 X_LOCK (reqlock);
784 1801
785 for (;;) 1802 for (;;)
786 { 1803 {
787 self->req = req = reqq_shift (&req_queue); 1804 self->req = req = reqq_shift (&req_queue);
… …
789 if (req) 1806 if (req)
790 break; 1807 break;
791 1808
792 ++idle; 1809 ++idle;
793 1810
1811 ts.tv_sec = time (0) + idle_timeout;
794 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1812 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
795 { 1813 {
796 if (idle > max_idle) 1814 if (idle > max_idle)
797 { 1815 {
798 --idle; 1816 --idle;
… …
803 goto quit; 1821 goto quit;
804 } 1822 }
805 1823
806 /* we are allowed to idle, so do so without any timeout */ 1824 /* we are allowed to idle, so do so without any timeout */
807 X_COND_WAIT (reqwait, reqlock); 1825 X_COND_WAIT (reqwait, reqlock);
808 ts.tv_sec = time (0) + IDLE_TIMEOUT;
809 } 1826 }
810 1827
811 --idle; 1828 --idle;
812 } 1829 }
813 1830
814 --nready; 1831 --nready;
815 1832
816 X_UNLOCK (reqlock); 1833 X_UNLOCK (reqlock);
817 1834
818 errno = 0; /* strictly unnecessary */
819
820 if (!EIO_CANCELLED (req))
821 switch (req->type) 1835 if (req->type < 0)
822 {
823 case EIO_READ: req->result = req->offs >= 0
824 ? pread (req->int1, req->ptr2, req->size, req->offs)
825 : read (req->int1, req->ptr2, req->size); break;
826 case EIO_WRITE: req->result = req->offs >= 0
827 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
828 : write (req->int1, req->ptr2, req->size); break;
829
830 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
831 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
832
833 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
834 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
835 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
836
837 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
838 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
839 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
840 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
841 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
842 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
843
844 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
845 case EIO_CLOSE: req->result = close (req->int1); break;
846 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
847 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
848 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
849 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
850 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
851 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
852 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
853 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
854 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
855
856 case EIO_SYNC: req->result = 0; sync (); break;
857 case EIO_FSYNC: req->result = fsync (req->int1); break;
858 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
859
860 case EIO_READDIR: scandir_ (req, self); break;
861
862 case EIO_BUSY:
863#ifdef _WIN32
864 Sleep (req->nv1 * 1000.);
865#else
866 {
867 struct timeval tv;
868
869 tv.tv_sec = req->nv1;
870 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
871
872 req->result = select (0, 0, 0, 0, &tv);
873 }
874#endif
875 break;
876
877 case EIO_UTIME:
878 case EIO_FUTIME:
879 {
880 struct timeval tv[2];
881 struct timeval *times;
882
883 if (req->nv1 != -1. || req->nv2 != -1.)
884 {
885 tv[0].tv_sec = req->nv1;
886 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
887 tv[1].tv_sec = req->nv2;
888 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
889
890 times = tv;
891 }
892 else
893 times = 0;
894
895
896 req->result = req->type == EIO_FUTIME
897 ? futimes (req->int1, times)
898 : utimes (req->ptr1, times);
899 }
900
901 case EIO_GROUP:
902 case EIO_NOP:
903 break;
904
905 case EIO_QUIT:
906 goto quit; 1836 goto quit;
907 1837
908 default: 1838 ETP_EXECUTE (self, req);
909 req->result = -1;
910 break;
911 }
912
913 req->errorno = errno;
914 1839
915 X_LOCK (reslock); 1840 X_LOCK (reslock);
916 1841
917 ++npending; 1842 ++npending;
918 1843
919 if (!reqq_push (&res_queue, req) && want_poll_cb) 1844 if (!reqq_push (&res_queue, req) && want_poll_cb)
920 want_poll_cb (); 1845 want_poll_cb ();
921 1846
922 self->req = 0; 1847 self->req = 0;
923 worker_clear (self); 1848 etp_worker_clear (self);
924 1849
925 X_UNLOCK (reslock); 1850 X_UNLOCK (reslock);
926 } 1851 }
927 1852
928quit: 1853quit:
929 X_LOCK (wrklock); 1854 X_LOCK (wrklock);
930 worker_free (self); 1855 etp_worker_free (self);
931 X_UNLOCK (wrklock); 1856 X_UNLOCK (wrklock);
932 1857
933 return 0; 1858 return 0;
934} 1859}
935 1860
936/*****************************************************************************/ 1861/*****************************************************************************/
937 1862
938static void eio_atfork_prepare (void) 1863int ecb_cold
1864eio_init (void (*want_poll)(void), void (*done_poll)(void))
939{ 1865{
940 X_LOCK (wrklock); 1866 return etp_init (want_poll, done_poll);
941 X_LOCK (reqlock);
942 X_LOCK (reslock);
943#if !HAVE_PREADWRITE
944 X_LOCK (preadwritelock);
945#endif
946#if !HAVE_READDIR_R
947 X_LOCK (readdirlock);
948#endif
949} 1867}
950 1868
951static void eio_atfork_parent (void) 1869ecb_inline void
1870eio_api_destroy (eio_req *req)
952{ 1871{
953#if !HAVE_READDIR_R 1872 free (req);
954 X_UNLOCK (readdirlock);
955#endif
956#if !HAVE_PREADWRITE
957 X_UNLOCK (preadwritelock);
958#endif
959 X_UNLOCK (reslock);
960 X_UNLOCK (reqlock);
961 X_UNLOCK (wrklock);
962} 1873}
963 1874
964static void eio_atfork_child (void) 1875#define REQ(rtype) \
965{ 1876 eio_req *req; \
966 eio_req *prv; 1877 \
1878 req = (eio_req *)calloc (1, sizeof *req); \
1879 if (!req) \
1880 return 0; \
1881 \
1882 req->type = rtype; \
1883 req->pri = pri; \
1884 req->finish = cb; \
1885 req->data = data; \
1886 req->destroy = eio_api_destroy;
967 1887
968 while (prv = reqq_shift (&req_queue)) 1888#define SEND eio_submit (req); return req
969 eio_destroy (prv);
970 1889
971 while (prv = reqq_shift (&res_queue)) 1890#define PATH \
972 eio_destroy (prv); 1891 req->flags |= EIO_FLAG_PTR1_FREE; \
1892 req->ptr1 = strdup (path); \
1893 if (!req->ptr1) \
1894 { \
1895 eio_api_destroy (req); \
1896 return 0; \
1897 }
973 1898
974 while (wrk_first.next != &wrk_first) 1899static void
1900eio_execute (etp_worker *self, eio_req *req)
1901{
1902 if (ecb_expect_false (EIO_CANCELLED (req)))
975 { 1903 {
976 worker *wrk = wrk_first.next; 1904 req->result = -1;
1905 req->errorno = ECANCELED;
1906 return;
1907 }
977 1908
978 if (wrk->req) 1909 switch (req->type)
979 eio_destroy (wrk->req);
980
981 worker_clear (wrk);
982 worker_free (wrk);
983 } 1910 {
1911 case EIO_READ: ALLOC (req->size);
1912 req->result = req->offs >= 0
1913 ? pread (req->int1, req->ptr2, req->size, req->offs)
1914 : read (req->int1, req->ptr2, req->size); break;
1915 case EIO_WRITE: req->result = req->offs >= 0
1916 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1917 : write (req->int1, req->ptr2, req->size); break;
984 1918
985 started = 0; 1919 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
986 idle = 0; 1920 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
987 nreqs = 0;
988 nready = 0;
989 npending = 0;
990 1921
991 eio_atfork_parent (); 1922 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
992} 1923 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1924 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1925 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1926 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1927 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
993 1928
994int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1929 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
995{ 1930 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
996 want_poll_cb = want_poll; 1931 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
997 done_poll_cb = done_poll; 1932 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
998 1933
1934 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1935 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1936 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1937 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1938 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1939 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1940
1941 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1942 case EIO_CLOSE: req->result = close (req->int1); break;
1943 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1944 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1945 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1946 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1947 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1948 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1949 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1950 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1951
1952 case EIO_REALPATH: eio__realpath (req, self); break;
1953
1954 case EIO_READLINK: ALLOC (PATH_MAX);
1955 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1956
1957 case EIO_SYNC: req->result = 0; sync (); break;
1958 case EIO_FSYNC: req->result = fsync (req->int1); break;
1959 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1960 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1961 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1962 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1963 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1964 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1965
1966 case EIO_READDIR: eio__scandir (req, self); break;
1967
1968 case EIO_BUSY:
999#ifdef _WIN32 1969#ifdef _WIN32
1000 X_MUTEX_CHECK (wrklock); 1970 Sleep (req->nv1 * 1e3);
1001 X_MUTEX_CHECK (reslock); 1971#else
1002 X_MUTEX_CHECK (reqlock); 1972 {
1003 X_MUTEX_CHECK (reqwait); 1973 struct timeval tv;
1004 X_MUTEX_CHECK (preadwritelock);
1005 X_MUTEX_CHECK (readdirlock);
1006 1974
1007 X_COND_CHECK (reqwait); 1975 tv.tv_sec = req->nv1;
1008#endif 1976 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1009 1977
1010 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child); 1978 req->result = select (0, 0, 0, 0, &tv);
1011} 1979 }
1980#endif
1981 break;
1012 1982
1013#if 0 1983 case EIO_UTIME:
1984 case EIO_FUTIME:
1985 {
1986 struct timeval tv[2];
1987 struct timeval *times;
1014 1988
1989 if (req->nv1 != -1. || req->nv2 != -1.)
1990 {
1991 tv[0].tv_sec = req->nv1;
1992 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1993 tv[1].tv_sec = req->nv2;
1994 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1995
1996 times = tv;
1997 }
1998 else
1999 times = 0;
2000
2001 req->result = req->type == EIO_FUTIME
2002 ? futimes (req->int1, times)
2003 : utimes (req->ptr1, times);
2004 }
2005 break;
2006
2007 case EIO_GROUP:
2008 abort (); /* handled in eio_request */
2009
2010 case EIO_NOP:
2011 req->result = 0;
2012 break;
2013
2014 case EIO_CUSTOM:
2015 req->feed (req);
2016 break;
2017
2018 default:
2019 errno = ENOSYS;
2020 req->result = -1;
2021 break;
2022 }
2023
2024 req->errorno = errno;
2025}
2026
2027#ifndef EIO_NO_WRAPPERS
2028
2029eio_req *eio_nop (int pri, eio_cb cb, void *data)
2030{
2031 REQ (EIO_NOP); SEND;
2032}
2033
2034eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2035{
2036 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2037}
2038
1015eio_req *eio_fsync (int fd, eio_cb cb); 2039eio_req *eio_sync (int pri, eio_cb cb, void *data)
2040{
2041 REQ (EIO_SYNC); SEND;
2042}
2043
2044eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2045{
2046 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2047}
2048
2049eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2050{
2051 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2052}
2053
2054eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2055{
2056 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2057}
2058
2059eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2060{
2061 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2062}
2063
2064eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2065{
2066 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2067}
2068
2069eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2070{
2071 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2072}
2073
1016eio_req *eio_fdatasync (int fd, eio_cb cb); 2074eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1017eio_req *eio_dupclose (int fd, eio_cb cb); 2075{
2076 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2077}
2078
2079eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2080{
2081 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2082}
2083
1018eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb); 2084eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1019eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2085{
1020eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2086 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1021eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2087}
2088
2089eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2090{
2091 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2092}
2093
2094eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2095{
2096 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2097}
2098
2099eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2100{
2101 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2102}
2103
2104eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2105{
2106 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2107}
2108
1022eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb); 2109eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2110{
2111 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2112}
2113
1023eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb); 2114eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2115{
2116 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2117}
2118
1024eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb); 2119eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2120{
2121 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2122}
2123
1025eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb); 2124eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
2125{
2126 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2127}
2128
1026eio_req *eio_dup2 (int fd, int fd2, eio_cb cb); 2129eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2130{
2131 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2132}
2133
1027eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb); 2134eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2135{
2136 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2137}
2138
1028eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb); 2139eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1029eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2140{
1030eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2141 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1031eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2142}
2143
1032eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb); 2144eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2145{
2146 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2147}
2148
1033eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb); 2149eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1034eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb); 2150{
1035eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb); 2151 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2152}
2153
1036eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb); 2154eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
2155{
2156 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2157}
2158
2159eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2160{
2161 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2162}
2163
2164eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2165{
2166 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2167}
2168
2169static eio_req *
2170eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2171{
2172 REQ (type); PATH; SEND;
2173}
2174
2175eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2176{
2177 return eio__1path (EIO_READLINK, path, pri, cb, data);
2178}
2179
2180eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2181{
2182 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2183}
2184
2185eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2186{
2187 return eio__1path (EIO_STAT, path, pri, cb, data);
2188}
2189
2190eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2191{
2192 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2193}
2194
2195eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2196{
2197 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2198}
2199
1037eio_req *eio_unlink (const char *path, eio_cb cb); 2200eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2201{
2202 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2203}
2204
1038eio_req *eio_rmdir (const char *path, eio_cb cb); 2205eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1039eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2206{
2207 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2208}
2209
2210eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2211{
2212 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2213}
2214
1040eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb); 2215eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1041eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */ 2216{
1042eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */ 2217 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2218}
2219
2220static eio_req *
2221eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2222{
2223 REQ (type); PATH;
2224
2225 req->flags |= EIO_FLAG_PTR2_FREE;
2226 req->ptr2 = strdup (new_path);
2227 if (!req->ptr2)
2228 {
2229 eio_api_destroy (req);
2230 return 0;
2231 }
2232
2233 SEND;
2234}
2235
2236eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2237{
2238 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2239}
2240
2241eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2242{
2243 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2244}
2245
2246eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2247{
2248 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2249}
2250
2251eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2252{
2253 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2254}
2255
2256#endif
2257
2258eio_req *eio_grp (eio_cb cb, void *data)
2259{
2260 const int pri = EIO_PRI_MAX;
2261
2262 REQ (EIO_GROUP); SEND;
2263}
2264
2265#undef REQ
2266#undef PATH
2267#undef SEND
2268
2269/*****************************************************************************/
2270/* grp functions */
2271
1043void 2272void
1044aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1045 PROTOTYPE: $$$;$
1046 PPCODE:
1047{
1048 dREQ;
1049
1050 req->type = EIO_OPEN;
1051 req->sv1 = newSVsv (pathname);
1052 req->ptr1 = SvPVbyte_nolen (req->sv1);
1053 req->int1 = flags;
1054 req->int2 = mode;
1055
1056 EIO_SEND;
1057}
1058
1059void
1060aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1061 PROTOTYPE: $;$
1062 ALIAS:
1063 aio_fsync = EIO_FSYNC
1064 aio_fdatasync = EIO_FDATASYNC
1065 PPCODE:
1066{
1067 dREQ;
1068
1069 req->type = ix;
1070 req->sv1 = newSVsv (fh);
1071 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1072
1073 EIO_SEND (req);
1074}
1075
1076void
1077aio_close (SV *fh, SV *callback=&PL_sv_undef)
1078 PROTOTYPE: $;$
1079 PPCODE:
1080{
1081 dREQ;
1082
1083 req->type = EIO_CLOSE;
1084 req->sv1 = newSVsv (fh);
1085 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1086
1087 EIO_SEND (req);
1088}
1089
1090void
1091aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1092 ALIAS:
1093 aio_read = EIO_READ
1094 aio_write = EIO_WRITE
1095 PROTOTYPE: $$$$$;$
1096 PPCODE:
1097{
1098 STRLEN svlen;
1099 char *svptr = SvPVbyte (data, svlen);
1100 UV len = SvUV (length);
1101
1102 SvUPGRADE (data, SVt_PV);
1103 SvPOK_on (data);
1104
1105 if (dataoffset < 0)
1106 dataoffset += svlen;
1107
1108 if (dataoffset < 0 || dataoffset > svlen)
1109 croak ("dataoffset outside of data scalar");
1110
1111 if (ix == EIO_WRITE)
1112 {
1113 /* write: check length and adjust. */
1114 if (!SvOK (length) || len + dataoffset > svlen)
1115 len = svlen - dataoffset;
1116 }
1117 else
1118 {
1119 /* read: grow scalar as necessary */
1120 svptr = SvGROW (data, len + dataoffset + 1);
1121 }
1122
1123 if (len < 0)
1124 croak ("length must not be negative");
1125
1126 {
1127 dREQ;
1128
1129 req->type = ix;
1130 req->sv1 = newSVsv (fh);
1131 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1132 : IoOFP (sv_2io (fh)));
1133 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1134 req->size = len;
1135 req->sv2 = SvREFCNT_inc (data);
1136 req->ptr2 = (char *)svptr + dataoffset;
1137 req->stroffset = dataoffset;
1138
1139 if (!SvREADONLY (data))
1140 {
1141 SvREADONLY_on (data);
1142 req->flags |= FLAG_SV2_RO_OFF;
1143 }
1144
1145 EIO_SEND;
1146 }
1147}
1148
1149void
1150aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1151 PROTOTYPE: $$;$
1152 PPCODE:
1153{
1154 SV *data;
1155 dREQ;
1156
1157 data = newSV (NAME_MAX);
1158 SvPOK_on (data);
1159
1160 req->type = EIO_READLINK;
1161 req->sv1 = newSVsv (path);
1162 req->ptr1 = SvPVbyte_nolen (req->sv1);
1163 req->sv2 = data;
1164 req->ptr2 = SvPVbyte_nolen (data);
1165
1166 EIO_SEND;
1167}
1168
1169void
1170aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1171 PROTOTYPE: $$$$;$
1172 PPCODE:
1173{
1174 dREQ;
1175
1176 req->type = EIO_SENDFILE;
1177 req->sv1 = newSVsv (out_fh);
1178 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1179 req->sv2 = newSVsv (in_fh);
1180 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1181 req->offs = SvVAL64 (in_offset);
1182 req->size = length;
1183
1184 EIO_SEND;
1185}
1186
1187void
1188aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1189 PROTOTYPE: $$$;$
1190 PPCODE:
1191{
1192 dREQ;
1193
1194 req->type = EIO_READAHEAD;
1195 req->sv1 = newSVsv (fh);
1196 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1197 req->offs = SvVAL64 (offset);
1198 req->size = length;
1199
1200 EIO_SEND;
1201}
1202
1203void
1204aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1205 ALIAS:
1206 aio_stat = EIO_STAT
1207 aio_lstat = EIO_LSTAT
1208 PPCODE:
1209{
1210 dREQ;
1211
1212 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1213 if (!req->ptr2)
1214 {
1215 req_destroy (req);
1216 croak ("out of memory during aio_stat statdata allocation");
1217 }
1218
1219 req->flags |= FLAG_PTR2_FREE;
1220 req->sv1 = newSVsv (fh_or_path);
1221
1222 if (SvPOK (fh_or_path))
1223 {
1224 req->type = ix;
1225 req->ptr1 = SvPVbyte_nolen (req->sv1);
1226 }
1227 else
1228 {
1229 req->type = EIO_FSTAT;
1230 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1231 }
1232
1233 EIO_SEND;
1234}
1235
1236void
1237aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1238 PPCODE:
1239{
1240 dREQ;
1241
1242 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1243 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1244 req->sv1 = newSVsv (fh_or_path);
1245
1246 if (SvPOK (fh_or_path))
1247 {
1248 req->type = EIO_UTIME;
1249 req->ptr1 = SvPVbyte_nolen (req->sv1);
1250 }
1251 else
1252 {
1253 req->type = EIO_FUTIME;
1254 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1255 }
1256
1257 EIO_SEND;
1258}
1259
1260void
1261aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1262 PPCODE:
1263{
1264 dREQ;
1265
1266 req->sv1 = newSVsv (fh_or_path);
1267 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1268
1269 if (SvPOK (fh_or_path))
1270 {
1271 req->type = EIO_TRUNCATE;
1272 req->ptr1 = SvPVbyte_nolen (req->sv1);
1273 }
1274 else
1275 {
1276 req->type = EIO_FTRUNCATE;
1277 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1278 }
1279
1280 EIO_SEND;
1281}
1282
1283void
1284aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1285 ALIAS:
1286 aio_chmod = EIO_CHMOD
1287 aio_fchmod = EIO_FCHMOD
1288 aio_mkdir = EIO_MKDIR
1289 PPCODE:
1290{
1291 dREQ;
1292
1293 req->type = type;
1294 req->int2 = mode;
1295 req->sv1 = newSVsv (fh_or_path);
1296
1297 if (ix == EIO_FCHMOD)
1298 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1299 else
1300 req->ptr1 = SvPVbyte_nolen (req->sv1);
1301
1302 EIO_SEND;
1303}
1304
1305void
1306aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1307 PPCODE:
1308{
1309 dREQ;
1310
1311 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1312 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1313 req->sv1 = newSVsv (fh_or_path);
1314
1315 if (SvPOK (fh_or_path))
1316 {
1317 req->type = EIO_CHOWN;
1318 req->ptr1 = SvPVbyte_nolen (req->sv1);
1319 }
1320 else
1321 {
1322 req->type = EIO_FCHOWN;
1323 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1324 }
1325
1326 EIO_SEND;
1327}
1328
1329void
1330aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1331 ALIAS:
1332 aio_unlink = EIO_UNLINK
1333 aio_rmdir = EIO_RMDIR
1334 aio_readdir = EIO_READDIR
1335 PPCODE:
1336{
1337 dREQ;
1338
1339 req->type = ix;
1340 req->sv1 = newSVsv (pathname);
1341 req->ptr1 = SvPVbyte_nolen (req->sv1);
1342
1343 EIO_SEND;
1344}
1345
1346void
1347aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1348 ALIAS:
1349 aio_link = EIO_LINK
1350 aio_symlink = EIO_SYMLINK
1351 aio_rename = EIO_RENAME
1352 PPCODE:
1353{
1354 dREQ;
1355
1356 req->type = ix;
1357 req->sv1 = newSVsv (oldpath);
1358 req->ptr1 = SvPVbyte_nolen (req->sv1);
1359 req->sv2 = newSVsv (newpath);
1360 req->ptr2 = SvPVbyte_nolen (req->sv2);
1361
1362 EIO_SEND;
1363}
1364
1365void
1366aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1367 PPCODE:
1368{
1369 dREQ;
1370
1371 req->type = EIO_MKNOD;
1372 req->sv1 = newSVsv (pathname);
1373 req->ptr1 = SvPVbyte_nolen (req->sv1);
1374 req->int2 = (mode_t)mode;
1375 req->offs = dev;
1376
1377 EIO_SEND;
1378}
1379
1380void
1381aio_busy (double delay, SV *callback=&PL_sv_undef)
1382 PPCODE:
1383{
1384 dREQ;
1385
1386 req->type = EIO_BUSY;
1387 req->nv1 = delay < 0. ? 0. : delay;
1388
1389 EIO_SEND;
1390}
1391
1392void
1393aio_group (SV *callback=&PL_sv_undef)
1394 PROTOTYPE: ;$
1395 PPCODE:
1396{
1397 dREQ;
1398
1399 req->type = EIO_GROUP;
1400
1401 req_send (req);
1402 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1403}
1404
1405void
1406aio_nop (SV *callback=&PL_sv_undef)
1407 ALIAS:
1408 aio_nop = EIO_NOP
1409 aio_sync = EIO_SYNC
1410 PPCODE:
1411{
1412 dREQ;
1413
1414#endif
1415
1416void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2273eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1417{ 2274{
1418 grp->int2 = limit; 2275 grp->int2 = limit;
1419 grp->feed = feed; 2276 grp->feed = feed;
1420 2277
1421 grp_try_feed (grp); 2278 grp_try_feed (grp);
1422} 2279}
1423 2280
2281void
1424void eio_grp_limit (eio_req *grp, int limit) 2282eio_grp_limit (eio_req *grp, int limit)
1425{ 2283{
1426 grp->int2 = limit; 2284 grp->int2 = limit;
1427 2285
1428 grp_try_feed (grp); 2286 grp_try_feed (grp);
1429} 2287}
1430 2288
2289void
1431void eio_grp_add (eio_req *grp, eio_req *req) 2290eio_grp_add (eio_req *grp, eio_req *req)
1432{ 2291{
1433 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2292 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2293
2294 grp->flags |= EIO_FLAG_GROUPADD;
1434 2295
1435 ++grp->size; 2296 ++grp->size;
1436 req->grp = grp; 2297 req->grp = grp;
1437 2298
1438 req->grp_prev = 0; 2299 req->grp_prev = 0;
… …
1442 grp->grp_first->grp_prev = req; 2303 grp->grp_first->grp_prev = req;
1443 2304
1444 grp->grp_first = req; 2305 grp->grp_first = req;
1445} 2306}
1446 2307
2308/*****************************************************************************/
2309/* misc garbage */
1447 2310
2311ssize_t
2312eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2313{
2314 etp_worker wrk;
2315 ssize_t ret;
2316
2317 wrk.dbuf = 0;
2318
2319 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2320
2321 if (wrk.dbuf)
2322 free (wrk.dbuf);
2323
2324 return ret;
2325}
2326

Diff Legend

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