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

Comparing libeio/eio.c (file contents):
Revision 1.9 by root, Mon May 12 00:31:43 2008 UTC vs.
Revision 1.69 by root, Fri Jun 10 06:50:42 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
46#ifdef EIO_STACKSIZE
47# define XTHREAD_STACKSIZE EIO_STACKSIZE
48#endif
2#include "xthread.h" 49#include "xthread.h"
3 50
4#include <errno.h> 51#include <errno.h>
5#include <stddef.h> 52#include <stddef.h>
6#include <stdlib.h> 53#include <stdlib.h>
7#include <string.h> 54#include <string.h>
8#include <errno.h> 55#include <errno.h>
9#include <sys/types.h> 56#include <sys/types.h>
10#include <sys/stat.h> 57#include <sys/stat.h>
58#include <sys/statvfs.h>
11#include <limits.h> 59#include <limits.h>
12#include <fcntl.h> 60#include <fcntl.h>
13#include <assert.h> 61#include <assert.h>
14 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes form stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
15#ifndef EIO_FINISH 69#ifndef EIO_FINISH
16# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 70# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17#endif 71#endif
18 72
19#ifndef EIO_DESTROY 73#ifndef EIO_DESTROY
25#endif 79#endif
26 80
27#ifdef _WIN32 81#ifdef _WIN32
28 82
29 /*doh*/ 83 /*doh*/
30
31#else 84#else
32 85
33# include "config.h"
34# include <sys/time.h> 86# include <sys/time.h>
35# include <sys/select.h> 87# include <sys/select.h>
36# include <unistd.h> 88# include <unistd.h>
37# include <utime.h> 89# include <utime.h>
38# include <signal.h> 90# include <signal.h>
39# include <dirent.h> 91# include <dirent.h>
40 92
93#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94# include <sys/mman.h>
95#endif
96
97/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99# define _DIRENT_HAVE_D_TYPE /* sigh */
100# define D_INO(de) (de)->d_fileno
101# define D_NAMLEN(de) (de)->d_namlen
102# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103# define D_INO(de) (de)->d_ino
104# endif
105
106#ifdef _D_EXACT_NAMLEN
107# undef D_NAMLEN
108# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109#endif
110
111# ifdef _DIRENT_HAVE_D_TYPE
112# define D_TYPE(de) (de)->d_type
113# endif
114
41# ifndef EIO_STRUCT_DIRENT 115# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 116# define EIO_STRUCT_DIRENT struct dirent
43# endif 117# endif
44 118
45#endif 119#endif
46 120
47#if HAVE_SENDFILE 121#if HAVE_SENDFILE
48# if __linux 122# if __linux
49# include <sys/sendfile.h> 123# include <sys/sendfile.h>
50# elif __freebsd 124# elif __FreeBSD__ || defined __APPLE__
51# include <sys/socket.h> 125# include <sys/socket.h>
52# include <sys/uio.h> 126# include <sys/uio.h>
53# elif __hpux 127# elif __hpux
54# include <sys/socket.h> 128# include <sys/socket.h>
55# elif __solaris /* not yet */ 129# elif __solaris
56# include <sys/sendfile.h> 130# include <sys/sendfile.h>
57# else 131# else
58# error sendfile support requested but not available 132# error sendfile support requested but not available
59# endif 133# endif
60#endif 134#endif
61 135
62/* number of seconds after which an idle threads exit */ 136#ifndef D_TYPE
63#define IDLE_TIMEOUT 10 137# define D_TYPE(de) 0
138#endif
139#ifndef D_INO
140# define D_INO(de) 0
141#endif
142#ifndef D_NAMLEN
143# define D_NAMLEN(de) strlen ((de)->d_name)
144#endif
64 145
65/* used for struct dirent, AIX doesn't provide it */ 146/* used for struct dirent, AIX doesn't provide it */
66#ifndef NAME_MAX 147#ifndef NAME_MAX
67# define NAME_MAX 4096 148# define NAME_MAX 4096
68#endif 149#endif
69 150
151/* used for readlink etc. */
152#ifndef PATH_MAX
153# define PATH_MAX 4096
154#endif
155
70/* buffer size for various temporary buffers */ 156/* buffer size for various temporary buffers */
71#define EIO_BUFSIZE 65536 157#define EIO_BUFSIZE 65536
72 158
73#define dBUF \ 159#define dBUF \
74 char *eio_buf; \ 160 char *eio_buf; \
75 X_LOCK (wrklock); \ 161 ETP_WORKER_LOCK (self); \
76 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 162 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77 X_UNLOCK (wrklock); \ 163 ETP_WORKER_UNLOCK (self); \
78 errno = ENOMEM; \ 164 errno = ENOMEM; \
79 if (!eio_buf) \ 165 if (!eio_buf) \
80 return -1; 166 return -1;
81 167
82#define EIO_TICKS ((1000000 + 1023) >> 10) 168#define EIO_TICKS ((1000000 + 1023) >> 10)
83 169
170/*****************************************************************************/
171
172#if __GNUC__ >= 3
173# define expect(expr,value) __builtin_expect ((expr),(value))
174#else
175# define expect(expr,value) (expr)
176#endif
177
178#define expect_false(expr) expect ((expr) != 0, 0)
179#define expect_true(expr) expect ((expr) != 0, 1)
180
181/*****************************************************************************/
182
183#define ETP_PRI_MIN EIO_PRI_MIN
184#define ETP_PRI_MAX EIO_PRI_MAX
185
186struct etp_worker;
187
188#define ETP_REQ eio_req
189#define ETP_DESTROY(req) eio_destroy (req)
190static int eio_finish (eio_req *req);
191#define ETP_FINISH(req) eio_finish (req)
192static void eio_execute (struct etp_worker *self, eio_req *req);
193#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194
195#define ETP_WORKER_CLEAR(req) \
196 if (wrk->dbuf) \
197 { \
198 free (wrk->dbuf); \
199 wrk->dbuf = 0; \
200 } \
201 \
202 if (wrk->dirp) \
203 { \
204 closedir (wrk->dirp); \
205 wrk->dirp = 0; \
206 }
207
208#define ETP_WORKER_COMMON \
209 void *dbuf; \
210 DIR *dirp;
211
212/*****************************************************************************/
213
214#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215
216/* calculate time difference in ~1/EIO_TICKS of a second */
217static int tvdiff (struct timeval *tv1, struct timeval *tv2)
218{
219 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
221}
222
223static unsigned int started, idle, wanted = 4;
224
84static void (*want_poll_cb) (void); 225static void (*want_poll_cb) (void);
85static void (*done_poll_cb) (void); 226static void (*done_poll_cb) (void);
86 227
87static unsigned int max_poll_time = 0; 228static unsigned int max_poll_time; /* reslock */
88static unsigned int max_poll_reqs = 0; 229static unsigned int max_poll_reqs; /* reslock */
89 230
90/* calculcate time difference in ~1/EIO_TICKS of a second */ 231static volatile unsigned int nreqs; /* reqlock */
91static int tvdiff (struct timeval *tv1, struct timeval *tv2) 232static volatile unsigned int nready; /* reqlock */
92{ 233static volatile unsigned int npending; /* reqlock */
93 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 234static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
94 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 235static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
95}
96 236
97static unsigned int started, idle, wanted = 4; 237static xmutex_t wrklock;
238static xmutex_t reslock;
239static xmutex_t reqlock;
240static xcond_t reqwait;
241
242#if !HAVE_PREADWRITE
243/*
244 * make our pread/pwrite emulation safe against themselves, but not against
245 * normal read/write by using a mutex. slows down execution a lot,
246 * but that's your problem, not mine.
247 */
248static xmutex_t preadwritelock = X_MUTEX_INIT;
249#endif
250
251typedef struct etp_worker
252{
253 /* locked by wrklock */
254 struct etp_worker *prev, *next;
255
256 xthread_t tid;
257
258 /* locked by reslock, reqlock or wrklock */
259 ETP_REQ *req; /* currently processed request */
260
261 ETP_WORKER_COMMON
262} etp_worker;
263
264static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
265
266#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
98 268
99/* worker threads management */ 269/* worker threads management */
100static mutex_t wrklock = X_MUTEX_INIT;
101 270
102typedef struct worker
103{
104 /* locked by wrklock */
105 struct worker *prev, *next;
106
107 thread_t tid;
108
109 /* locked by reslock, reqlock or wrklock */
110 eio_req *req; /* currently processed request */
111 void *dbuf;
112 DIR *dirp;
113} worker;
114
115static worker wrk_first = { &wrk_first, &wrk_first, 0 };
116
117static void worker_clear (worker *wrk) 271static void etp_worker_clear (etp_worker *wrk)
118{ 272{
119 if (wrk->dirp) 273 ETP_WORKER_CLEAR (wrk);
120 {
121 closedir (wrk->dirp);
122 wrk->dirp = 0;
123 }
124
125 if (wrk->dbuf)
126 {
127 free (wrk->dbuf);
128 wrk->dbuf = 0;
129 }
130} 274}
131 275
132static void worker_free (worker *wrk) 276static void etp_worker_free (etp_worker *wrk)
133{ 277{
134 wrk->next->prev = wrk->prev; 278 wrk->next->prev = wrk->prev;
135 wrk->prev->next = wrk->next; 279 wrk->prev->next = wrk->next;
136 280
137 free (wrk); 281 free (wrk);
138} 282}
139 283
140static volatile unsigned int nreqs, nready, npending;
141static volatile unsigned int max_idle = 4;
142
143static mutex_t reslock = X_MUTEX_INIT;
144static mutex_t reqlock = X_MUTEX_INIT;
145static cond_t reqwait = X_COND_INIT;
146
147unsigned int eio_nreqs (void) 284static unsigned int etp_nreqs (void)
148{ 285{
286 int retval;
287 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288 retval = nreqs;
289 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
149 return nreqs; 290 return retval;
150} 291}
151 292
152unsigned int eio_nready (void) 293static unsigned int etp_nready (void)
153{ 294{
154 unsigned int retval; 295 unsigned int retval;
155 296
156 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 297 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157 retval = nready; 298 retval = nready;
158 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 299 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159 300
160 return retval; 301 return retval;
161} 302}
162 303
163unsigned int eio_npending (void) 304static unsigned int etp_npending (void)
164{ 305{
165 unsigned int retval; 306 unsigned int retval;
166 307
167 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 308 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168 retval = npending; 309 retval = npending;
169 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 310 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170 311
171 return retval; 312 return retval;
172} 313}
173 314
174unsigned int eio_nthreads (void) 315static unsigned int etp_nthreads (void)
175{ 316{
176 unsigned int retval; 317 unsigned int retval;
177 318
178 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 319 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179 retval = started; 320 retval = started;
186 * a somewhat faster data structure might be nice, but 327 * a somewhat faster data structure might be nice, but
187 * with 8 priorities this actually needs <20 insns 328 * with 8 priorities this actually needs <20 insns
188 * per shift, the most expensive operation. 329 * per shift, the most expensive operation.
189 */ 330 */
190typedef struct { 331typedef struct {
191 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 332 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
192 int size; 333 int size;
193} reqq; 334} etp_reqq;
194 335
195static reqq req_queue; 336static etp_reqq req_queue;
196static reqq res_queue; 337static etp_reqq res_queue;
197 338
198static int reqq_push (reqq *q, eio_req *req) 339static int reqq_push (etp_reqq *q, ETP_REQ *req)
199{ 340{
200 int pri = req->pri; 341 int pri = req->pri;
201 req->next = 0; 342 req->next = 0;
202 343
203 if (q->qe[pri]) 344 if (q->qe[pri])
209 q->qe[pri] = q->qs[pri] = req; 350 q->qe[pri] = q->qs[pri] = req;
210 351
211 return q->size++; 352 return q->size++;
212} 353}
213 354
214static eio_req *reqq_shift (reqq *q) 355static ETP_REQ *reqq_shift (etp_reqq *q)
215{ 356{
216 int pri; 357 int pri;
217 358
218 if (!q->size) 359 if (!q->size)
219 return 0; 360 return 0;
220 361
221 --q->size; 362 --q->size;
222 363
223 for (pri = EIO_NUM_PRI; pri--; ) 364 for (pri = ETP_NUM_PRI; pri--; )
224 { 365 {
225 eio_req *req = q->qs[pri]; 366 eio_req *req = q->qs[pri];
226 367
227 if (req) 368 if (req)
228 { 369 {
234 } 375 }
235 376
236 abort (); 377 abort ();
237} 378}
238 379
239static void grp_try_feed (eio_req *grp) 380static void etp_thread_init (void)
240{ 381{
241 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 382 X_MUTEX_CREATE (wrklock);
383 X_MUTEX_CREATE (reslock);
384 X_MUTEX_CREATE (reqlock);
385 X_COND_CREATE (reqwait);
386}
387
388static void etp_atfork_prepare (void)
389{
390 X_LOCK (wrklock);
391 X_LOCK (reqlock);
392 X_LOCK (reslock);
393#if !HAVE_PREADWRITE
394 X_LOCK (preadwritelock);
395#endif
396}
397
398static void etp_atfork_parent (void)
399{
400#if !HAVE_PREADWRITE
401 X_UNLOCK (preadwritelock);
402#endif
403 X_UNLOCK (reslock);
404 X_UNLOCK (reqlock);
405 X_UNLOCK (wrklock);
406}
407
408static void etp_atfork_child (void)
409{
410 ETP_REQ *prv;
411
412 while ((prv = reqq_shift (&req_queue)))
413 ETP_DESTROY (prv);
414
415 while ((prv = reqq_shift (&res_queue)))
416 ETP_DESTROY (prv);
417
418 while (wrk_first.next != &wrk_first)
242 { 419 {
243 int old_len = grp->size; 420 etp_worker *wrk = wrk_first.next;
244 421
245 EIO_FEED (grp); 422 if (wrk->req)
423 ETP_DESTROY (wrk->req);
246 424
247 /* stop if no progress has been made */ 425 etp_worker_clear (wrk);
248 if (old_len == grp->size) 426 etp_worker_free (wrk);
249 {
250 grp->feed = 0;
251 break;
252 }
253 } 427 }
254}
255 428
256static int eio_finish (eio_req *req); 429 started = 0;
430 idle = 0;
431 nreqs = 0;
432 nready = 0;
433 npending = 0;
257 434
258static int grp_dec (eio_req *grp) 435 etp_thread_init ();
259{ 436}
260 --grp->size;
261 437
262 /* call feeder, if applicable */ 438static void
263 grp_try_feed (grp); 439etp_once_init (void)
440{
441 etp_thread_init ();
442 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
443}
264 444
265 /* finish, if done */ 445static int
266 if (!grp->size && grp->int1) 446etp_init (void (*want_poll)(void), void (*done_poll)(void))
267 return eio_finish (grp); 447{
268 else 448 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
449
450 pthread_once (&doinit, etp_once_init);
451
452 want_poll_cb = want_poll;
453 done_poll_cb = done_poll;
454
269 return 0; 455 return 0;
270} 456}
271 457
272void eio_destroy (eio_req *req)
273{
274 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
275 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
276
277 EIO_DESTROY (req);
278}
279
280static int eio_finish (eio_req *req)
281{
282 int res = EIO_FINISH (req);
283
284 if (req->grp)
285 {
286 int res2;
287 eio_req *grp = req->grp;
288
289 /* unlink request */
290 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
291 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
292
293 if (grp->grp_first == req)
294 grp->grp_first = req->grp_next;
295
296 res2 = grp_dec (grp);
297
298 if (!res && res2)
299 res = res2;
300 }
301
302 eio_destroy (req);
303
304 return res;
305}
306
307void eio_grp_cancel (eio_req *grp)
308{
309 for (grp = grp->grp_first; grp; grp = grp->grp_next)
310 eio_cancel (grp);
311}
312
313void eio_cancel (eio_req *req)
314{
315 X_LOCK (wrklock);
316 req->flags |= EIO_FLAG_CANCELLED;
317 X_UNLOCK (wrklock);
318
319 eio_grp_cancel (req);
320}
321
322X_THREAD_PROC (eio_proc); 458X_THREAD_PROC (etp_proc);
323 459
324static void start_thread (void) 460static void etp_start_thread (void)
325{ 461{
326 worker *wrk = calloc (1, sizeof (worker)); 462 etp_worker *wrk = calloc (1, sizeof (etp_worker));
327 463
328 /*TODO*/ 464 /*TODO*/
329 assert (("unable to allocate worker thread data", wrk)); 465 assert (("unable to allocate worker thread data", wrk));
330 466
331 X_LOCK (wrklock); 467 X_LOCK (wrklock);
332 468
333 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 469 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334 { 470 {
335 wrk->prev = &wrk_first; 471 wrk->prev = &wrk_first;
336 wrk->next = wrk_first.next; 472 wrk->next = wrk_first.next;
337 wrk_first.next->prev = wrk; 473 wrk_first.next->prev = wrk;
338 wrk_first.next = wrk; 474 wrk_first.next = wrk;
342 free (wrk); 478 free (wrk);
343 479
344 X_UNLOCK (wrklock); 480 X_UNLOCK (wrklock);
345} 481}
346 482
347static void maybe_start_thread (void) 483static void etp_maybe_start_thread (void)
348{ 484{
349 if (eio_nthreads () >= wanted) 485 if (expect_true (etp_nthreads () >= wanted))
350 return; 486 return;
351 487
352 /* todo: maybe use idle here, but might be less exact */ 488 /* todo: maybe use idle here, but might be less exact */
353 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 489 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354 return; 490 return;
355 491
356 start_thread (); 492 etp_start_thread ();
357} 493}
358 494
359void eio_submit (eio_req *req)
360{
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) 495static void etp_end_thread (void)
373{ 496{
374 eio_req *req = calloc (1, sizeof (eio_req)); 497 eio_req *req = calloc (1, sizeof (eio_req));
375 498
376 req->type = EIO_QUIT; 499 req->type = -1;
377 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 500 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
378 501
379 X_LOCK (reqlock); 502 X_LOCK (reqlock);
380 reqq_push (&req_queue, req); 503 reqq_push (&req_queue, req);
381 X_COND_SIGNAL (reqwait); 504 X_COND_SIGNAL (reqwait);
382 X_UNLOCK (reqlock); 505 X_UNLOCK (reqlock);
384 X_LOCK (wrklock); 507 X_LOCK (wrklock);
385 --started; 508 --started;
386 X_UNLOCK (wrklock); 509 X_UNLOCK (wrklock);
387} 510}
388 511
389void eio_set_max_poll_time (double nseconds)
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) 512static int etp_poll (void)
426{ 513{
427 int maxreqs = max_poll_reqs; 514 unsigned int maxreqs;
515 unsigned int maxtime;
428 struct timeval tv_start, tv_now; 516 struct timeval tv_start, tv_now;
429 eio_req *req;
430 517
518 X_LOCK (reslock);
519 maxreqs = max_poll_reqs;
520 maxtime = max_poll_time;
521 X_UNLOCK (reslock);
522
431 if (max_poll_time) 523 if (maxtime)
432 gettimeofday (&tv_start, 0); 524 gettimeofday (&tv_start, 0);
433 525
434 for (;;) 526 for (;;)
435 { 527 {
528 ETP_REQ *req;
529
436 maybe_start_thread (); 530 etp_maybe_start_thread ();
437 531
438 X_LOCK (reslock); 532 X_LOCK (reslock);
439 req = reqq_shift (&res_queue); 533 req = reqq_shift (&res_queue);
440 534
441 if (req) 535 if (req)
449 X_UNLOCK (reslock); 543 X_UNLOCK (reslock);
450 544
451 if (!req) 545 if (!req)
452 return 0; 546 return 0;
453 547
548 X_LOCK (reqlock);
454 --nreqs; 549 --nreqs;
550 X_UNLOCK (reqlock);
455 551
456 if (req->type == EIO_GROUP && req->size) 552 if (expect_false (req->type == EIO_GROUP && req->size))
457 { 553 {
458 req->int1 = 1; /* mark request as delayed */ 554 req->int1 = 1; /* mark request as delayed */
459 continue; 555 continue;
460 } 556 }
461 else 557 else
462 { 558 {
463 int res = eio_finish (req); 559 int res = ETP_FINISH (req);
464 if (res) 560 if (expect_false (res))
465 return res; 561 return res;
466 } 562 }
467 563
468 if (maxreqs && !--maxreqs) 564 if (expect_false (maxreqs && !--maxreqs))
469 break; 565 break;
470 566
471 if (max_poll_time) 567 if (maxtime)
472 { 568 {
473 gettimeofday (&tv_now, 0); 569 gettimeofday (&tv_now, 0);
474 570
475 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 571 if (tvdiff (&tv_start, &tv_now) >= maxtime)
476 break; 572 break;
477 } 573 }
478 } 574 }
479 575
480 errno = EAGAIN; 576 errno = EAGAIN;
481 return -1; 577 return -1;
482} 578}
483 579
580static void etp_cancel (ETP_REQ *req)
581{
582 X_LOCK (wrklock);
583 req->flags |= EIO_FLAG_CANCELLED;
584 X_UNLOCK (wrklock);
585
586 eio_grp_cancel (req);
587}
588
589static void etp_submit (ETP_REQ *req)
590{
591 req->pri -= ETP_PRI_MIN;
592
593 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
594 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595
596 if (expect_false (req->type == EIO_GROUP))
597 {
598 /* I hope this is worth it :/ */
599 X_LOCK (reqlock);
600 ++nreqs;
601 X_UNLOCK (reqlock);
602
603 X_LOCK (reslock);
604
605 ++npending;
606
607 if (!reqq_push (&res_queue, req) && want_poll_cb)
608 want_poll_cb ();
609
610 X_UNLOCK (reslock);
611 }
612 else
613 {
614 X_LOCK (reqlock);
615 ++nreqs;
616 ++nready;
617 reqq_push (&req_queue, req);
618 X_COND_SIGNAL (reqwait);
619 X_UNLOCK (reqlock);
620
621 etp_maybe_start_thread ();
622 }
623}
624
625static void etp_set_max_poll_time (double nseconds)
626{
627 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628 max_poll_time = nseconds * EIO_TICKS;
629 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630}
631
632static void etp_set_max_poll_reqs (unsigned int maxreqs)
633{
634 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635 max_poll_reqs = maxreqs;
636 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637}
638
639static void etp_set_max_idle (unsigned int nthreads)
640{
641 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642 max_idle = nthreads;
643 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644}
645
646static void etp_set_idle_timeout (unsigned int seconds)
647{
648 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649 idle_timeout = seconds;
650 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651}
652
653static void etp_set_min_parallel (unsigned int nthreads)
654{
655 if (wanted < nthreads)
656 wanted = nthreads;
657}
658
659static void etp_set_max_parallel (unsigned int nthreads)
660{
661 if (wanted > nthreads)
662 wanted = nthreads;
663
664 while (started > wanted)
665 etp_end_thread ();
666}
667
668/*****************************************************************************/
669
670static void grp_try_feed (eio_req *grp)
671{
672 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673 {
674 grp->flags &= ~EIO_FLAG_GROUPADD;
675
676 EIO_FEED (grp);
677
678 /* stop if no progress has been made */
679 if (!(grp->flags & EIO_FLAG_GROUPADD))
680 {
681 grp->feed = 0;
682 break;
683 }
684 }
685}
686
687static int grp_dec (eio_req *grp)
688{
689 --grp->size;
690
691 /* call feeder, if applicable */
692 grp_try_feed (grp);
693
694 /* finish, if done */
695 if (!grp->size && grp->int1)
696 return eio_finish (grp);
697 else
698 return 0;
699}
700
701void eio_destroy (eio_req *req)
702{
703 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705
706 EIO_DESTROY (req);
707}
708
709static int eio_finish (eio_req *req)
710{
711 int res = EIO_FINISH (req);
712
713 if (req->grp)
714 {
715 int res2;
716 eio_req *grp = req->grp;
717
718 /* unlink request */
719 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
720 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
721
722 if (grp->grp_first == req)
723 grp->grp_first = req->grp_next;
724
725 res2 = grp_dec (grp);
726
727 if (!res && res2)
728 res = res2;
729 }
730
731 eio_destroy (req);
732
733 return res;
734}
735
736void eio_grp_cancel (eio_req *grp)
737{
738 for (grp = grp->grp_first; grp; grp = grp->grp_next)
739 eio_cancel (grp);
740}
741
742void eio_cancel (eio_req *req)
743{
744 etp_cancel (req);
745}
746
747void eio_submit (eio_req *req)
748{
749 etp_submit (req);
750}
751
752unsigned int eio_nreqs (void)
753{
754 return etp_nreqs ();
755}
756
757unsigned int eio_nready (void)
758{
759 return etp_nready ();
760}
761
762unsigned int eio_npending (void)
763{
764 return etp_npending ();
765}
766
767unsigned int eio_nthreads (void)
768{
769 return etp_nthreads ();
770}
771
772void eio_set_max_poll_time (double nseconds)
773{
774 etp_set_max_poll_time (nseconds);
775}
776
777void eio_set_max_poll_reqs (unsigned int maxreqs)
778{
779 etp_set_max_poll_reqs (maxreqs);
780}
781
782void eio_set_max_idle (unsigned int nthreads)
783{
784 etp_set_max_idle (nthreads);
785}
786
787void eio_set_idle_timeout (unsigned int seconds)
788{
789 etp_set_idle_timeout (seconds);
790}
791
792void eio_set_min_parallel (unsigned int nthreads)
793{
794 etp_set_min_parallel (nthreads);
795}
796
797void eio_set_max_parallel (unsigned int nthreads)
798{
799 etp_set_max_parallel (nthreads);
800}
801
802int eio_poll (void)
803{
804 return etp_poll ();
805}
806
484/*****************************************************************************/ 807/*****************************************************************************/
485/* work around various missing functions */ 808/* work around various missing functions */
486 809
487#if !HAVE_PREADWRITE 810#if !HAVE_PREADWRITE
811# undef pread
812# undef pwrite
488# define pread eio__pread 813# define pread eio__pread
489# define pwrite eio__pwrite 814# define pwrite eio__pwrite
490
491/*
492 * make our pread/pwrite safe against themselves, but not against
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;
497 815
498static ssize_t 816static ssize_t
499eio__pread (int fd, void *buf, size_t count, off_t offset) 817eio__pread (int fd, void *buf, size_t count, off_t offset)
500{ 818{
501 ssize_t res; 819 ssize_t res;
519 837
520 X_LOCK (preadwritelock); 838 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR); 839 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET); 840 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count); 841 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET); 842 lseek (fd, ooffset, SEEK_SET);
525 X_UNLOCK (preadwritelock); 843 X_UNLOCK (preadwritelock);
526 844
527 return res; 845 return res;
528} 846}
529#endif 847#endif
530 848
531#ifndef HAVE_FUTIMES 849#ifndef HAVE_UTIMES
532 850
851# undef utimes
533# define utimes(path,times) eio__utimes (path, times) 852# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) eio__futimes (fd, times)
535 853
536static int 854static int
537eio__utimes (const char *filename, const struct timeval times[2]) 855eio__utimes (const char *filename, const struct timeval times[2])
538{ 856{
539 if (times) 857 if (times)
547 } 865 }
548 else 866 else
549 return utime (filename, 0); 867 return utime (filename, 0);
550} 868}
551 869
870#endif
871
872#ifndef HAVE_FUTIMES
873
874# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times)
876
552static int eio__futimes (int fd, const struct timeval tv[2]) 877static int eio__futimes (int fd, const struct timeval tv[2])
553{ 878{
554 errno = ENOSYS; 879 errno = ENOSYS;
555 return -1; 880 return -1;
556} 881}
557 882
558#endif 883#endif
559 884
560#if !HAVE_FDATASYNC 885#if !HAVE_FDATASYNC
886# undef fdatasync
561# define fdatasync fsync 887# define fdatasync(fd) fsync (fd)
562#endif 888#endif
889
890/* sync_file_range always needs emulation */
891int
892eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893{
894#if HAVE_SYNC_FILE_RANGE
895 int res;
896
897 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
898 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
899 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
900 {
901 flags = 0
902 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
903 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
904 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
905 }
906
907 res = sync_file_range (fd, offset, nbytes, flags);
908
909 if (!res || errno != ENOSYS)
910 return res;
911#endif
912
913 /* even though we could play tricks with the flags, it's better to always
914 * call fdatasync, as that matches the expectation of its users best */
915 return fdatasync (fd);
916}
563 917
564#if !HAVE_READAHEAD 918#if !HAVE_READAHEAD
919# undef readahead
565# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 920# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 921
567static ssize_t 922static ssize_t
568eio__readahead (int fd, off_t offset, size_t count, worker *self) 923eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
569{ 924{
570 size_t todo = count; 925 size_t todo = count;
571 dBUF; 926 dBUF;
572 927
573 while (todo > 0) 928 while (todo > 0)
583 return count; 938 return count;
584} 939}
585 940
586#endif 941#endif
587 942
588#if !HAVE_READDIR_R
589# define readdir_r eio__readdir_r
590
591static mutex_t readdirlock = X_MUTEX_INIT;
592
593static int
594eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
595{
596 EIO_STRUCT_DIRENT *e;
597 int errorno;
598
599 X_LOCK (readdirlock);
600
601 e = readdir (dirp);
602 errorno = errno;
603
604 if (e)
605 {
606 *res = ent;
607 strcpy (ent->d_name, e->d_name);
608 }
609 else
610 *res = 0;
611
612 X_UNLOCK (readdirlock);
613
614 errno = errorno;
615 return e ? 0 : -1;
616}
617#endif
618
619/* sendfile always needs emulation */ 943/* sendfile always needs emulation */
620static ssize_t 944static ssize_t
621eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self) 945eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
622{ 946{
947 ssize_t written = 0;
623 ssize_t res; 948 ssize_t res;
624 949
625 if (!count) 950 if (!count)
626 return 0; 951 return 0;
627 952
953 for (;;)
954 {
628#if HAVE_SENDFILE 955#if HAVE_SENDFILE
629# if __linux 956# if __linux
957 off_t soffset = offset;
630 res = sendfile (ofd, ifd, &offset, count); 958 res = sendfile (ofd, ifd, &soffset, count);
631 959
632# elif __freebsd 960# elif __FreeBSD__
633 /* 961 /*
634 * Of course, the freebsd sendfile is a dire hack with no thoughts 962 * Of course, the freebsd sendfile is a dire hack with no thoughts
635 * wasted on making it similar to other I/O functions. 963 * wasted on making it similar to other I/O functions.
636 */ 964 */
637 {
638 off_t sbytes; 965 off_t sbytes;
639 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 966 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
640 967
641 if (res < 0 && sbytes) 968 #if 0 /* according to the manpage, this is correct, but broken behaviour */
642 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 969 /* freebsd' sendfile will return 0 on success */
970 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
971 /* not on e.g. EIO or EPIPE - sounds broken */
972 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
643 res = sbytes; 973 res = sbytes;
644 } 974 #endif
975
976 /* according to source inspection, this is correct, and useful behaviour */
977 if (sbytes)
978 res = sbytes;
979
980# elif defined (__APPLE__)
981 off_t sbytes = count;
982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983
984 /* according to the manpage, sbytes is always valid */
985 if (sbytes)
986 res = sbytes;
645 987
646# elif __hpux 988# elif __hpux
647 res = sendfile (ofd, ifd, offset, count, 0, 0); 989 res = sendfile (ofd, ifd, offset, count, 0, 0);
648 990
649# elif __solaris 991# elif __solaris
650 {
651 struct sendfilevec vec; 992 struct sendfilevec vec;
652 size_t sbytes; 993 size_t sbytes;
653 994
654 vec.sfv_fd = ifd; 995 vec.sfv_fd = ifd;
655 vec.sfv_flag = 0; 996 vec.sfv_flag = 0;
656 vec.sfv_off = offset; 997 vec.sfv_off = offset;
657 vec.sfv_len = count; 998 vec.sfv_len = count;
658 999
659 res = sendfilev (ofd, &vec, 1, &sbytes); 1000 res = sendfilev (ofd, &vec, 1, &sbytes);
660 1001
661 if (res < 0 && sbytes) 1002 if (res < 0 && sbytes)
662 res = sbytes; 1003 res = sbytes;
663 }
664 1004
665# endif 1005# endif
1006
1007#elif defined (_WIN32)
1008 /* does not work, just for documentation of what would need to be done */
1009 HANDLE h = TO_SOCKET (ifd);
1010 SetFilePointer (h, offset, 0, FILE_BEGIN);
1011 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1012
666#else 1013#else
667 res = -1; 1014 res = -1;
668 errno = ENOSYS; 1015 errno = ENOSYS;
669#endif 1016#endif
670 1017
1018 /* we assume sendfile can copy at least 128mb in one go */
1019 if (res <= 128 * 1024 * 1024)
1020 {
1021 if (res > 0)
1022 written += res;
1023
1024 if (written)
1025 return written;
1026
1027 break;
1028 }
1029 else
1030 {
1031 /* if we requested more, then probably the kernel was lazy */
1032 written += res;
1033 offset += res;
1034 count -= res;
1035
1036 if (!count)
1037 return written;
1038 }
1039 }
1040
671 if (res < 0 1041 if (res < 0
672 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1042 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1043 /* BSDs */
1044#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1045 || errno == ENOTSUP
1046#endif
1047 || errno == EOPNOTSUPP /* BSDs */
673#if __solaris 1048#if __solaris
674 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1049 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
675#endif 1050#endif
676 ) 1051 )
677 ) 1052 )
708 } 1083 }
709 1084
710 return res; 1085 return res;
711} 1086}
712 1087
1088static signed char
1089eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1090{
1091 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1092 : a->inode < b->inode ? -1
1093 : a->inode > b->inode ? 1
1094 : 0;
1095}
1096
1097#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1098
1099#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1100#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1101
1102static void
1103eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1104{
1105 unsigned char bits [9 + sizeof (ino_t) * 8];
1106 unsigned char *bit = bits;
1107
1108 assert (CHAR_BIT == 8);
1109 assert (sizeof (eio_dirent) * 8 < 256);
1110 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1111 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1112
1113 if (size <= EIO_SORT_FAST)
1114 return;
1115
1116 /* first prepare an array of bits to test in our radix sort */
1117 /* try to take endianness into account, as well as differences in ino_t sizes */
1118 /* inode_bits must contain all inodes ORed together */
1119 /* which is used to skip bits that are 0 everywhere, which is very common */
1120 {
1121 ino_t endianness;
1122 int i, j;
1123
1124 /* we store the byte offset of byte n into byte n of "endianness" */
1125 for (i = 0; i < sizeof (ino_t); ++i)
1126 ((unsigned char *)&endianness)[i] = i;
1127
1128 *bit++ = 0;
1129
1130 for (i = 0; i < sizeof (ino_t); ++i)
1131 {
1132 /* shifting off the byte offsets out of "endianness" */
1133 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1134 endianness >>= 8;
1135
1136 for (j = 0; j < 8; ++j)
1137 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1138 *bit++ = offs + j;
1139 }
1140
1141 for (j = 0; j < 8; ++j)
1142 if (score_bits & (1 << j))
1143 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1144 }
1145
1146 /* now actually do the sorting (a variant of MSD radix sort) */
1147 {
1148 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1149 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1150 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1151 int stk_idx = 0;
1152
1153 base_stk [stk_idx] = dents;
1154 end_stk [stk_idx] = dents + size;
1155 bit_stk [stk_idx] = bit - 1;
1156
1157 do
1158 {
1159 base = base_stk [stk_idx];
1160 end = end_stk [stk_idx];
1161 bit = bit_stk [stk_idx];
1162
1163 for (;;)
1164 {
1165 unsigned char O = *bit >> 3;
1166 unsigned char M = 1 << (*bit & 7);
1167
1168 eio_dirent *a = base;
1169 eio_dirent *b = end;
1170
1171 if (b - a < EIO_SORT_CUTOFF)
1172 break;
1173
1174 /* now bit-partition the array on the bit */
1175 /* this ugly asymmetric loop seems to perform much better than typical */
1176 /* partition algos found in the literature */
1177 do
1178 if (!(((unsigned char *)a)[O] & M))
1179 ++a;
1180 else if (!(((unsigned char *)--b)[O] & M))
1181 {
1182 eio_dirent tmp = *a; *a = *b; *b = tmp;
1183 ++a;
1184 }
1185 while (b > a);
1186
1187 /* next bit, or stop, if no bits left in this path */
1188 if (!*--bit)
1189 break;
1190
1191 base_stk [stk_idx] = a;
1192 end_stk [stk_idx] = end;
1193 bit_stk [stk_idx] = bit;
1194 ++stk_idx;
1195
1196 end = a;
1197 }
1198 }
1199 while (stk_idx--);
1200 }
1201}
1202
1203static void
1204eio_dent_insertion_sort (eio_dirent *dents, int size)
1205{
1206 /* first move the smallest element to the front, to act as a sentinel */
1207 {
1208 int i;
1209 eio_dirent *min = dents;
1210
1211 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1212 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1213 if (EIO_DENT_CMP (dents [i], <, *min))
1214 min = &dents [i];
1215
1216 /* swap elements 0 and j (minimum) */
1217 {
1218 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1219 }
1220 }
1221
1222 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1223 {
1224 eio_dirent *i, *j;
1225
1226 for (i = dents + 1; i < dents + size; ++i)
1227 {
1228 eio_dirent value = *i;
1229
1230 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1231 j [1] = j [0];
1232
1233 j [1] = value;
1234 }
1235 }
1236}
1237
1238static void
1239eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1240{
1241 if (size <= 1)
1242 return; /* our insertion sort relies on size > 0 */
1243
1244 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1245 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1246 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1247
1248 /* use an insertion sort at the end, or for small arrays, */
1249 /* as insertion sort is more efficient for small partitions */
1250 eio_dent_insertion_sort (dents, size);
1251}
1252
713/* read a full directory */ 1253/* read a full directory */
714static void 1254static void
715eio__scandir (eio_req *req, worker *self) 1255eio__scandir (eio_req *req, etp_worker *self)
716{ 1256{
717 DIR *dirp; 1257 DIR *dirp;
718 union
719 {
720 EIO_STRUCT_DIRENT d;
721 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
722 } *u;
723 EIO_STRUCT_DIRENT *entp; 1258 EIO_STRUCT_DIRENT *entp;
724 char *name, *names; 1259 char *name, *names;
725 int memlen = 4096; 1260 int namesalloc = 4096;
726 int memofs = 0; 1261 int namesoffs = 0;
1262 int flags = req->int1;
1263 eio_dirent *dents = 0;
1264 int dentalloc = 128;
727 int res = 0; 1265 int dentoffs = 0;
1266 ino_t inode_bits = 0;
1267
1268 req->result = -1;
1269
1270 if (!(flags & EIO_READDIR_DENTS))
1271 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
728 1272
729 X_LOCK (wrklock); 1273 X_LOCK (wrklock);
1274 /* the corresponding closedir is in ETP_WORKER_CLEAR */
730 self->dirp = dirp = opendir (req->ptr1); 1275 self->dirp = dirp = opendir (req->ptr1);
731 self->dbuf = u = malloc (sizeof (*u)); 1276
732 req->flags |= EIO_FLAG_PTR2_FREE; 1277 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1278 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
733 req->ptr2 = names = malloc (memlen); 1279 req->ptr2 = names = malloc (namesalloc);
734 X_UNLOCK (wrklock); 1280 X_UNLOCK (wrklock);
735 1281
736 if (dirp && u && names) 1282 if (dirp && names && (!flags || dents))
737 for (;;) 1283 for (;;)
738 { 1284 {
739 errno = 0; 1285 errno = 0;
740 readdir_r (dirp, &u->d, &entp); 1286 entp = readdir (dirp);
741 1287
742 if (!entp) 1288 if (!entp)
1289 {
1290 if (errno)
1291 break;
1292
1293 /* sort etc. */
1294 req->int1 = flags;
1295 req->result = dentoffs;
1296
1297 if (flags & EIO_READDIR_STAT_ORDER)
1298 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1299 else if (flags & EIO_READDIR_DIRS_FIRST)
1300 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1301 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1302 else
1303 {
1304 /* in this case, all is known, and we just put dirs first and sort them */
1305 eio_dirent *oth = dents + dentoffs;
1306 eio_dirent *dir = dents;
1307
1308 /* now partition dirs to the front, and non-dirs to the back */
1309 /* by walking from both sides and swapping if necessary */
1310 while (oth > dir)
1311 {
1312 if (dir->type == EIO_DT_DIR)
1313 ++dir;
1314 else if ((--oth)->type == EIO_DT_DIR)
1315 {
1316 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1317
1318 ++dir;
1319 }
1320 }
1321
1322 /* now sort the dirs only (dirs all have the same score) */
1323 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1324 }
1325
743 break; 1326 break;
1327 }
744 1328
1329 /* now add the entry to our list(s) */
745 name = entp->d_name; 1330 name = entp->d_name;
746 1331
1332 /* skip . and .. entries */
747 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1333 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
748 { 1334 {
749 int len = strlen (name) + 1; 1335 int len = D_NAMLEN (entp) + 1;
750 1336
751 res++; 1337 while (expect_false (namesoffs + len > namesalloc))
752
753 while (memofs + len > memlen)
754 { 1338 {
755 memlen *= 2; 1339 namesalloc *= 2;
756 X_LOCK (wrklock); 1340 X_LOCK (wrklock);
757 req->ptr2 = names = realloc (names, memlen); 1341 req->ptr2 = names = realloc (names, namesalloc);
758 X_UNLOCK (wrklock); 1342 X_UNLOCK (wrklock);
759 1343
760 if (!names) 1344 if (!names)
761 break; 1345 break;
762 } 1346 }
763 1347
764 memcpy (names + memofs, name, len); 1348 memcpy (names + namesoffs, name, len);
1349
1350 if (dents)
1351 {
1352 struct eio_dirent *ent;
1353
1354 if (expect_false (dentoffs == dentalloc))
1355 {
1356 dentalloc *= 2;
1357 X_LOCK (wrklock);
1358 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1359 X_UNLOCK (wrklock);
1360
1361 if (!dents)
1362 break;
1363 }
1364
1365 ent = dents + dentoffs;
1366
1367 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1368 ent->namelen = len - 1;
1369 ent->inode = D_INO (entp);
1370
1371 inode_bits |= ent->inode;
1372
1373 switch (D_TYPE (entp))
1374 {
1375 default:
1376 ent->type = EIO_DT_UNKNOWN;
1377 flags |= EIO_READDIR_FOUND_UNKNOWN;
1378 break;
1379
1380 #ifdef DT_FIFO
1381 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1382 #endif
1383 #ifdef DT_CHR
1384 case DT_CHR: ent->type = EIO_DT_CHR; break;
1385 #endif
1386 #ifdef DT_MPC
1387 case DT_MPC: ent->type = EIO_DT_MPC; break;
1388 #endif
1389 #ifdef DT_DIR
1390 case DT_DIR: ent->type = EIO_DT_DIR; break;
1391 #endif
1392 #ifdef DT_NAM
1393 case DT_NAM: ent->type = EIO_DT_NAM; break;
1394 #endif
1395 #ifdef DT_BLK
1396 case DT_BLK: ent->type = EIO_DT_BLK; break;
1397 #endif
1398 #ifdef DT_MPB
1399 case DT_MPB: ent->type = EIO_DT_MPB; break;
1400 #endif
1401 #ifdef DT_REG
1402 case DT_REG: ent->type = EIO_DT_REG; break;
1403 #endif
1404 #ifdef DT_NWK
1405 case DT_NWK: ent->type = EIO_DT_NWK; break;
1406 #endif
1407 #ifdef DT_CMP
1408 case DT_CMP: ent->type = EIO_DT_CMP; break;
1409 #endif
1410 #ifdef DT_LNK
1411 case DT_LNK: ent->type = EIO_DT_LNK; break;
1412 #endif
1413 #ifdef DT_SOCK
1414 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1415 #endif
1416 #ifdef DT_DOOR
1417 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1418 #endif
1419 #ifdef DT_WHT
1420 case DT_WHT: ent->type = EIO_DT_WHT; break;
1421 #endif
1422 }
1423
1424 ent->score = 7;
1425
1426 if (flags & EIO_READDIR_DIRS_FIRST)
1427 {
1428 if (ent->type == EIO_DT_UNKNOWN)
1429 {
1430 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1431 ent->score = 1;
1432 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1433 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1434 }
1435 else if (ent->type == EIO_DT_DIR)
1436 ent->score = 0;
1437 }
1438 }
1439
765 memofs += len; 1440 namesoffs += len;
1441 ++dentoffs;
1442 }
1443
1444 if (EIO_CANCELLED (req))
1445 {
1446 errno = ECANCELED;
1447 break;
766 } 1448 }
767 } 1449 }
1450}
768 1451
769 if (errno) 1452#ifdef PAGESIZE
770 res = -1; 1453# define eio_pagesize() PAGESIZE
1454#else
1455static intptr_t
1456eio_pagesize (void)
1457{
1458 static intptr_t page;
1459
1460 if (!page)
1461 page = sysconf (_SC_PAGESIZE);
1462
1463 return page;
1464}
1465#endif
1466
1467static void
1468eio_page_align (void **addr, size_t *length)
1469{
1470 intptr_t mask = eio_pagesize () - 1;
1471
1472 /* round down addr */
1473 intptr_t adj = mask & (intptr_t)*addr;
1474
1475 *addr = (void *)((intptr_t)*addr - adj);
1476 *length += adj;
1477
1478 /* round up length */
1479 *length = (*length + mask) & ~mask;
1480}
1481
1482#if !_POSIX_MEMLOCK
1483# define eio__mlockall(a) ((errno = ENOSYS), -1)
1484#else
1485
1486static int
1487eio__mlockall (int flags)
1488{
1489 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1490 extern int mallopt (int, int);
1491 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1492 #endif
1493
1494 if (EIO_MCL_CURRENT != MCL_CURRENT
1495 || EIO_MCL_FUTURE != MCL_FUTURE)
1496 {
1497 flags = 0
1498 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1499 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1500 }
1501
1502 return mlockall (flags);
1503}
1504#endif
1505
1506#if !_POSIX_MEMLOCK_RANGE
1507# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1508#else
1509
1510static int
1511eio__mlock (void *addr, size_t length)
1512{
1513 eio_page_align (&addr, &length);
1514
1515 return mlock (addr, length);
1516}
1517
1518#endif
1519
1520#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1521# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1522#else
1523
1524int
1525eio__msync (void *mem, size_t len, int flags)
1526{
1527 eio_page_align (&mem, &len);
1528
1529 if (EIO_MS_ASYNC != MS_SYNC
1530 || EIO_MS_INVALIDATE != MS_INVALIDATE
1531 || EIO_MS_SYNC != MS_SYNC)
1532 {
1533 flags = 0
1534 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1535 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1536 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1537 }
1538
1539 return msync (mem, len, flags);
1540}
1541
1542#endif
1543
1544int
1545eio__mtouch (eio_req *req)
1546{
1547 void *mem = req->ptr2;
1548 size_t len = req->size;
1549 int flags = req->int1;
1550
1551 eio_page_align (&mem, &len);
1552
771 1553 {
772 req->result = res; 1554 intptr_t addr = (intptr_t)mem;
1555 intptr_t end = addr + len;
1556 intptr_t page = eio_pagesize ();
1557
1558 if (addr < end)
1559 if (flags & EIO_MT_MODIFY) /* modify */
1560 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1561 else
1562 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1563 }
1564
1565 return 0;
773} 1566}
774 1567
775/*****************************************************************************/ 1568/*****************************************************************************/
776 1569
777#define ALLOC(len) \ 1570#define ALLOC(len) \
787 req->result = -1; \ 1580 req->result = -1; \
788 break; \ 1581 break; \
789 } \ 1582 } \
790 } 1583 }
791 1584
792X_THREAD_PROC (eio_proc) 1585X_THREAD_PROC (etp_proc)
793{ 1586{
794 eio_req *req; 1587 ETP_REQ *req;
795 struct timespec ts; 1588 struct timespec ts;
796 worker *self = (worker *)thr_arg; 1589 etp_worker *self = (etp_worker *)thr_arg;
797 1590
798 /* try to distribute timeouts somewhat randomly */ 1591 /* try to distribute timeouts somewhat randomly */
799 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1592 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
800 1593
801 for (;;) 1594 for (;;)
802 { 1595 {
803 ts.tv_sec = time (0) + IDLE_TIMEOUT;
804
805 X_LOCK (reqlock); 1596 X_LOCK (reqlock);
806 1597
807 for (;;) 1598 for (;;)
808 { 1599 {
809 self->req = req = reqq_shift (&req_queue); 1600 self->req = req = reqq_shift (&req_queue);
811 if (req) 1602 if (req)
812 break; 1603 break;
813 1604
814 ++idle; 1605 ++idle;
815 1606
1607 ts.tv_sec = time (0) + idle_timeout;
816 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1608 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
817 { 1609 {
818 if (idle > max_idle) 1610 if (idle > max_idle)
819 { 1611 {
820 --idle; 1612 --idle;
825 goto quit; 1617 goto quit;
826 } 1618 }
827 1619
828 /* we are allowed to idle, so do so without any timeout */ 1620 /* we are allowed to idle, so do so without any timeout */
829 X_COND_WAIT (reqwait, reqlock); 1621 X_COND_WAIT (reqwait, reqlock);
830 ts.tv_sec = time (0) + IDLE_TIMEOUT;
831 } 1622 }
832 1623
833 --idle; 1624 --idle;
834 } 1625 }
835 1626
836 --nready; 1627 --nready;
837 1628
838 X_UNLOCK (reqlock); 1629 X_UNLOCK (reqlock);
839 1630
840 errno = 0; /* strictly unnecessary */ 1631 if (req->type < 0)
1632 goto quit;
841 1633
842 if (!EIO_CANCELLED (req)) 1634 if (!EIO_CANCELLED (req))
843 switch (req->type) 1635 ETP_EXECUTE (self, req);
844 {
845 case EIO_READ: ALLOC (req->size);
846 req->result = req->offs >= 0
847 ? pread (req->int1, req->ptr2, req->size, req->offs)
848 : read (req->int1, req->ptr2, req->size); break;
849 case EIO_WRITE: req->result = req->offs >= 0
850 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
851 : write (req->int1, req->ptr2, req->size); break;
852
853 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
854 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
855
856 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
857 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
858 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
859 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
860 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
861 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
862
863 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
864 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
865 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
866 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
867 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
868 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
869
870 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
871 case EIO_CLOSE: req->result = close (req->int1); break;
872 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
873 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
874 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
875 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
876 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
877 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
878 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
879 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
880
881 case EIO_READLINK: ALLOC (NAME_MAX);
882 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
883
884 case EIO_SYNC: req->result = 0; sync (); break;
885 case EIO_FSYNC: req->result = fsync (req->int1); break;
886 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
887
888 case EIO_READDIR: eio__scandir (req, self); break;
889
890 case EIO_BUSY:
891#ifdef _WIN32
892 Sleep (req->nv1 * 1000.);
893#else
894 {
895 struct timeval tv;
896
897 tv.tv_sec = req->nv1;
898 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
899
900 req->result = select (0, 0, 0, 0, &tv);
901 }
902#endif
903 break;
904
905 case EIO_UTIME:
906 case EIO_FUTIME:
907 {
908 struct timeval tv[2];
909 struct timeval *times;
910
911 if (req->nv1 != -1. || req->nv2 != -1.)
912 {
913 tv[0].tv_sec = req->nv1;
914 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
915 tv[1].tv_sec = req->nv2;
916 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
917
918 times = tv;
919 }
920 else
921 times = 0;
922
923
924 req->result = req->type == EIO_FUTIME
925 ? futimes (req->int1, times)
926 : utimes (req->ptr1, times);
927 }
928
929 case EIO_GROUP:
930 case EIO_NOP:
931 req->result = 0;
932 break;
933
934 case EIO_QUIT:
935 goto quit;
936
937 default:
938 req->result = -1;
939 break;
940 }
941
942 req->errorno = errno;
943 1636
944 X_LOCK (reslock); 1637 X_LOCK (reslock);
945 1638
946 ++npending; 1639 ++npending;
947 1640
948 if (!reqq_push (&res_queue, req) && want_poll_cb) 1641 if (!reqq_push (&res_queue, req) && want_poll_cb)
949 want_poll_cb (); 1642 want_poll_cb ();
950 1643
951 self->req = 0; 1644 self->req = 0;
952 worker_clear (self); 1645 etp_worker_clear (self);
953 1646
954 X_UNLOCK (reslock); 1647 X_UNLOCK (reslock);
955 } 1648 }
956 1649
957quit: 1650quit:
958 X_LOCK (wrklock); 1651 X_LOCK (wrklock);
959 worker_free (self); 1652 etp_worker_free (self);
960 X_UNLOCK (wrklock); 1653 X_UNLOCK (wrklock);
961 1654
962 return 0; 1655 return 0;
963} 1656}
964 1657
965/*****************************************************************************/ 1658/*****************************************************************************/
966 1659
967static void eio_atfork_prepare (void)
968{
969 X_LOCK (wrklock);
970 X_LOCK (reqlock);
971 X_LOCK (reslock);
972#if !HAVE_PREADWRITE
973 X_LOCK (preadwritelock);
974#endif
975#if !HAVE_READDIR_R
976 X_LOCK (readdirlock);
977#endif
978}
979
980static void eio_atfork_parent (void)
981{
982#if !HAVE_READDIR_R
983 X_UNLOCK (readdirlock);
984#endif
985#if !HAVE_PREADWRITE
986 X_UNLOCK (preadwritelock);
987#endif
988 X_UNLOCK (reslock);
989 X_UNLOCK (reqlock);
990 X_UNLOCK (wrklock);
991}
992
993static void eio_atfork_child (void)
994{
995 eio_req *prv;
996
997 while (prv = reqq_shift (&req_queue))
998 eio_destroy (prv);
999
1000 while (prv = reqq_shift (&res_queue))
1001 eio_destroy (prv);
1002
1003 while (wrk_first.next != &wrk_first)
1004 {
1005 worker *wrk = wrk_first.next;
1006
1007 if (wrk->req)
1008 eio_destroy (wrk->req);
1009
1010 worker_clear (wrk);
1011 worker_free (wrk);
1012 }
1013
1014 started = 0;
1015 idle = 0;
1016 nreqs = 0;
1017 nready = 0;
1018 npending = 0;
1019
1020 eio_atfork_parent ();
1021}
1022
1023int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1660int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1024{ 1661{
1025 want_poll_cb = want_poll; 1662 return etp_init (want_poll, done_poll);
1026 done_poll_cb = done_poll;
1027
1028#ifdef _WIN32
1029 X_MUTEX_CHECK (wrklock);
1030 X_MUTEX_CHECK (reslock);
1031 X_MUTEX_CHECK (reqlock);
1032 X_MUTEX_CHECK (reqwait);
1033 X_MUTEX_CHECK (preadwritelock);
1034 X_MUTEX_CHECK (readdirlock);
1035
1036 X_COND_CHECK (reqwait);
1037#endif
1038
1039 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1040} 1663}
1041 1664
1042static void eio_api_destroy (eio_req *req) 1665static void eio_api_destroy (eio_req *req)
1043{ 1666{
1044 free (req); 1667 free (req);
1049 \ 1672 \
1050 req = (eio_req *)calloc (1, sizeof *req); \ 1673 req = (eio_req *)calloc (1, sizeof *req); \
1051 if (!req) \ 1674 if (!req) \
1052 return 0; \ 1675 return 0; \
1053 \ 1676 \
1054 req->type = rtype; \ 1677 req->type = rtype; \
1055 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 1678 req->pri = pri; \
1056 req->finish = cb; \ 1679 req->finish = cb; \
1680 req->data = data; \
1057 req->destroy = eio_api_destroy; 1681 req->destroy = eio_api_destroy;
1058 1682
1059#define SEND eio_submit (req); return req 1683#define SEND eio_submit (req); return req
1060 1684
1061#define PATH \ 1685#define PATH \
1065 { \ 1689 { \
1066 eio_api_destroy (req); \ 1690 eio_api_destroy (req); \
1067 return 0; \ 1691 return 0; \
1068 } 1692 }
1069 1693
1070eio_req *eio_nop (eio_cb cb) 1694static void eio_execute (etp_worker *self, eio_req *req)
1695{
1696 switch (req->type)
1697 {
1698 case EIO_READ: ALLOC (req->size);
1699 req->result = req->offs >= 0
1700 ? pread (req->int1, req->ptr2, req->size, req->offs)
1701 : read (req->int1, req->ptr2, req->size); break;
1702 case EIO_WRITE: req->result = req->offs >= 0
1703 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1704 : write (req->int1, req->ptr2, req->size); break;
1705
1706 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1707 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1708
1709 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1710 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1711 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1712 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1713 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1714 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1715
1716 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1717 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1718 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1719 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1720
1721 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1722 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1723 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1724 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1725 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1726 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1727
1728 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1729 case EIO_CLOSE: req->result = close (req->int1); break;
1730 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1731 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1732 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1733 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1734 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1735 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1736 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1737 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1738
1739 case EIO_READLINK: ALLOC (PATH_MAX);
1740 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1741
1742 case EIO_SYNC: req->result = 0; sync (); break;
1743 case EIO_FSYNC: req->result = fsync (req->int1); break;
1744 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1745 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1746 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1747 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1748 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1749 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1750
1751 case EIO_READDIR: eio__scandir (req, self); break;
1752
1753 case EIO_BUSY:
1754#ifdef _WIN32
1755 Sleep (req->nv1 * 1e3);
1756#else
1757 {
1758 struct timeval tv;
1759
1760 tv.tv_sec = req->nv1;
1761 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1762
1763 req->result = select (0, 0, 0, 0, &tv);
1764 }
1765#endif
1766 break;
1767
1768 case EIO_UTIME:
1769 case EIO_FUTIME:
1770 {
1771 struct timeval tv[2];
1772 struct timeval *times;
1773
1774 if (req->nv1 != -1. || req->nv2 != -1.)
1775 {
1776 tv[0].tv_sec = req->nv1;
1777 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1778 tv[1].tv_sec = req->nv2;
1779 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1780
1781 times = tv;
1782 }
1783 else
1784 times = 0;
1785
1786 req->result = req->type == EIO_FUTIME
1787 ? futimes (req->int1, times)
1788 : utimes (req->ptr1, times);
1789 }
1790 break;
1791
1792 case EIO_GROUP:
1793 abort (); /* handled in eio_request */
1794
1795 case EIO_NOP:
1796 req->result = 0;
1797 break;
1798
1799 case EIO_CUSTOM:
1800 req->feed (req);
1801 break;
1802
1803 default:
1804 errno = ENOSYS;
1805 req->result = -1;
1806 break;
1807 }
1808
1809 req->errorno = errno;
1810}
1811
1812#ifndef EIO_NO_WRAPPERS
1813
1814eio_req *eio_nop (int pri, eio_cb cb, void *data)
1071{ 1815{
1072 REQ (EIO_NOP); SEND; 1816 REQ (EIO_NOP); SEND;
1073} 1817}
1074 1818
1075eio_req *eio_busy (double delay, eio_cb cb) 1819eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1076{ 1820{
1077 REQ (EIO_BUSY); req->nv1 = delay; SEND; 1821 REQ (EIO_BUSY); req->nv1 = delay; SEND;
1078} 1822}
1079 1823
1080eio_req *eio_sync (eio_cb cb) 1824eio_req *eio_sync (int pri, eio_cb cb, void *data)
1081{ 1825{
1082 REQ (EIO_SYNC); SEND; 1826 REQ (EIO_SYNC); SEND;
1083} 1827}
1084 1828
1085eio_req *eio_fsync (int fd, eio_cb cb) 1829eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1086{ 1830{
1087 REQ (EIO_FSYNC); req->int1 = fd; SEND; 1831 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1088} 1832}
1089 1833
1834eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1835{
1836 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1837}
1838
1839eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1840{
1841 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1842}
1843
1844eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1845{
1846 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1847}
1848
1849eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1850{
1851 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1852}
1853
1854eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1855{
1856 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1857}
1858
1090eio_req *eio_fdatasync (int fd, eio_cb cb) 1859eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1091{ 1860{
1092 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1861 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1093} 1862}
1094 1863
1095eio_req *eio_close (int fd, eio_cb cb) 1864eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1096{ 1865{
1097 REQ (EIO_CLOSE); req->int1 = fd; SEND; 1866 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1098} 1867}
1099 1868
1100eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb) 1869eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1101{ 1870{
1102 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 1871 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1103} 1872}
1104 1873
1105eio_req *eio_read (int fd, void *data, size_t length, off_t offset, eio_cb cb) 1874eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1106{ 1875{
1107 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 1876 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1108} 1877}
1109 1878
1110eio_req *eio_write (int fd, void *data, size_t length, off_t offset, eio_cb cb) 1879eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1111{ 1880{
1112 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 1881 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1113} 1882}
1114 1883
1115eio_req *eio_fstat (int fd, eio_cb cb) 1884eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1116{ 1885{
1117 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1886 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1118} 1887}
1119 1888
1889eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1890{
1891 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1892}
1893
1120eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb) 1894eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1121{ 1895{
1122 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1896 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1123} 1897}
1124 1898
1125eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb) 1899eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1126{ 1900{
1127 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 1901 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1128} 1902}
1129 1903
1130eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb) 1904eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1131{ 1905{
1132 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 1906 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1133} 1907}
1134 1908
1135eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb) 1909eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1136{ 1910{
1137 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 1911 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1138} 1912}
1139 1913
1140eio_req *eio_dup2 (int fd, int fd2, eio_cb cb) 1914eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1141{ 1915{
1142 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND; 1916 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1143} 1917}
1144 1918
1145eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb) 1919eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1146{ 1920{
1147 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 1921 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1148} 1922}
1149 1923
1150eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb) 1924eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1151{ 1925{
1152 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 1926 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1153} 1927}
1154 1928
1155eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb) 1929eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1156{ 1930{
1157 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 1931 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1158} 1932}
1159 1933
1160eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb) 1934eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1161{ 1935{
1162 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 1936 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1163} 1937}
1164 1938
1165eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 1939eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1166{ 1940{
1167 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 1941 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1168} 1942}
1169 1943
1170eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb) 1944eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1171{ 1945{
1172 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND; 1946 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1173} 1947}
1174 1948
1175eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb) 1949eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1176{ 1950{
1177 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND; 1951 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1178} 1952}
1179 1953
1180static eio_req * 1954static eio_req *
1181eio__1path (int type, const char *path, eio_cb cb) 1955eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1182{ 1956{
1183 REQ (type); PATH; SEND; 1957 REQ (type); PATH; SEND;
1184} 1958}
1185 1959
1186eio_req *eio_readlink (const char *path, eio_cb cb) 1960eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1187{ 1961{
1188 return eio__1path (EIO_READLINK, path, cb); 1962 return eio__1path (EIO_READLINK, path, pri, cb, data);
1189} 1963}
1190 1964
1191eio_req *eio_stat (const char *path, eio_cb cb) 1965eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1192{ 1966{
1193 return eio__1path (EIO_STAT, path, cb); 1967 return eio__1path (EIO_STAT, path, pri, cb, data);
1194} 1968}
1195 1969
1196eio_req *eio_lstat (const char *path, eio_cb cb) 1970eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1197{ 1971{
1198 return eio__1path (EIO_LSTAT, path, cb); 1972 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1199} 1973}
1200 1974
1975eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1976{
1977 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1978}
1979
1201eio_req *eio_unlink (const char *path, eio_cb cb) 1980eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1202{ 1981{
1203 return eio__1path (EIO_UNLINK, path, cb); 1982 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1204} 1983}
1205 1984
1206eio_req *eio_rmdir (const char *path, eio_cb cb) 1985eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1207{ 1986{
1208 return eio__1path (EIO_RMDIR, path, cb); 1987 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1209} 1988}
1210 1989
1211eio_req *eio_readdir (const char *path, eio_cb cb) 1990eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1212{ 1991{
1213 return eio__1path (EIO_READDIR, path, cb); 1992 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1214} 1993}
1215 1994
1216eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb) 1995eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1217{ 1996{
1218 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1997 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1219} 1998}
1220 1999
1221static eio_req * 2000static eio_req *
1222eio__2path (int type, const char *path, const char *new_path, eio_cb cb) 2001eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1223{ 2002{
1224 REQ (type); PATH; 2003 REQ (type); PATH;
1225 2004
1226 req->flags |= EIO_FLAG_PTR2_FREE; 2005 req->flags |= EIO_FLAG_PTR2_FREE;
1227 req->ptr2 = strdup (new_path); 2006 req->ptr2 = strdup (new_path);
1232 } 2011 }
1233 2012
1234 SEND; 2013 SEND;
1235} 2014}
1236 2015
1237eio_req *eio_link (const char *path, const char *new_path, eio_cb cb) 2016eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1238{ 2017{
1239 return eio__2path (EIO_LINK, path, new_path, cb); 2018 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
1240} 2019}
1241 2020
1242eio_req *eio_symlink (const char *path, const char *new_path, eio_cb cb) 2021eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1243{ 2022{
1244 return eio__2path (EIO_SYMLINK, path, new_path, cb); 2023 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
1245} 2024}
1246 2025
1247eio_req *eio_rename (const char *path, const char *new_path, eio_cb cb) 2026eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1248{ 2027{
1249 return eio__2path (EIO_RENAME, path, new_path, cb); 2028 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2029}
2030
2031eio_req *eio_custom (void (*)(eio_req *) execute, int pri, eio_cb cb, void *data);
2032{
2033 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2034}
2035
2036#endif
2037
2038eio_req *eio_grp (eio_cb cb, void *data)
2039{
2040 const int pri = EIO_PRI_MAX;
2041
2042 REQ (EIO_GROUP); SEND;
1250} 2043}
1251 2044
1252#undef REQ 2045#undef REQ
1253#undef PATH 2046#undef PATH
1254#undef SEND 2047#undef SEND
1273 2066
1274void eio_grp_add (eio_req *grp, eio_req *req) 2067void eio_grp_add (eio_req *grp, eio_req *req)
1275{ 2068{
1276 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2069 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1277 2070
2071 grp->flags |= EIO_FLAG_GROUPADD;
2072
1278 ++grp->size; 2073 ++grp->size;
1279 req->grp = grp; 2074 req->grp = grp;
1280 2075
1281 req->grp_prev = 0; 2076 req->grp_prev = 0;
1282 req->grp_next = grp->grp_first; 2077 req->grp_next = grp->grp_first;
1290/*****************************************************************************/ 2085/*****************************************************************************/
1291/* misc garbage */ 2086/* misc garbage */
1292 2087
1293ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2088ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1294{ 2089{
1295 worker wrk; 2090 etp_worker wrk;
2091 ssize_t ret;
1296 2092
1297 wrk.dbuf = 0; 2093 wrk.dbuf = 0;
1298 2094
1299 eio__sendfile (ofd, ifd, offset, count, &wrk); 2095 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1300 2096
1301 if (wrk.dbuf) 2097 if (wrk.dbuf)
1302 free (wrk.dbuf); 2098 free (wrk.dbuf);
1303}
1304 2099
2100 return ret;
2101}
2102

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines