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

Comparing libeio/eio.c (file contents):
Revision 1.8 by root, Sun May 11 20:08:37 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# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif
50
51#if HAVE_SENDFILE 121#if HAVE_SENDFILE
52# if __linux 122# if __linux
53# include <sys/sendfile.h> 123# include <sys/sendfile.h>
54# elif __freebsd 124# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 125# include <sys/socket.h>
56# include <sys/uio.h> 126# include <sys/uio.h>
57# elif __hpux 127# elif __hpux
58# include <sys/socket.h> 128# include <sys/socket.h>
59# elif __solaris /* not yet */ 129# elif __solaris
60# include <sys/sendfile.h> 130# include <sys/sendfile.h>
61# else 131# else
62# error sendfile support requested but not available 132# error sendfile support requested but not available
63# endif 133# endif
64#endif 134#endif
65 135
66/* number of seconds after which an idle threads exit */ 136#ifndef D_TYPE
67#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
68 145
69/* used for struct dirent, AIX doesn't provide it */ 146/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 147#ifndef NAME_MAX
71# define NAME_MAX 4096 148# define NAME_MAX 4096
72#endif 149#endif
73 150
151/* used for readlink etc. */
152#ifndef PATH_MAX
153# define PATH_MAX 4096
154#endif
155
74/* buffer size for various temporary buffers */ 156/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 157#define EIO_BUFSIZE 65536
76 158
77#define dBUF \ 159#define dBUF \
78 char *eio_buf; \ 160 char *eio_buf; \
79 X_LOCK (wrklock); \ 161 ETP_WORKER_LOCK (self); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 162 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 163 ETP_WORKER_UNLOCK (self); \
164 errno = ENOMEM; \
82 if (!eio_buf) \ 165 if (!eio_buf) \
83 return -1; 166 return -1;
84 167
85#define EIO_TICKS ((1000000 + 1023) >> 10) 168#define EIO_TICKS ((1000000 + 1023) >> 10)
86 169
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
87static void (*want_poll_cb) (void); 225static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 226static void (*done_poll_cb) (void);
89 227
90static unsigned int max_poll_time = 0; 228static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 229static unsigned int max_poll_reqs; /* reslock */
92 230
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 231static volatile unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 232static volatile unsigned int nready; /* reqlock */
95{ 233static volatile unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 234static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 235static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 236
100static unsigned int started, idle, wanted; 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)
101 268
102/* worker threads management */ 269/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 270
105typedef struct worker {
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) 271static void etp_worker_clear (etp_worker *wrk)
120{ 272{
121 if (wrk->dirp) 273 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} 274}
133 275
134static void worker_free (worker *wrk) 276static void etp_worker_free (etp_worker *wrk)
135{ 277{
136 wrk->next->prev = wrk->prev; 278 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 279 wrk->prev->next = wrk->next;
138 280
139 free (wrk); 281 free (wrk);
140} 282}
141 283
142static volatile unsigned int nreqs, nready, npending;
143static volatile unsigned int max_idle = 4;
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) 284static unsigned int etp_nreqs (void)
150{ 285{
286 int retval;
287 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288 retval = nreqs;
289 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 290 return retval;
152} 291}
153 292
154unsigned int eio_nready (void) 293static unsigned int etp_nready (void)
155{ 294{
156 unsigned int retval; 295 unsigned int retval;
157 296
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 297 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 298 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 299 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 300
162 return retval; 301 return retval;
163} 302}
164 303
165unsigned int eio_npending (void) 304static unsigned int etp_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 etp_nthreads (void)
177{ 316{
178 unsigned int retval; 317 unsigned int retval;
179 318
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 319 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 320 retval = started;
… …
188 * a somewhat faster data structure might be nice, but 327 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 328 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 329 * per shift, the most expensive operation.
191 */ 330 */
192typedef struct { 331typedef struct {
193 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 */
194 int size; 333 int size;
195} reqq; 334} etp_reqq;
196 335
197static reqq req_queue; 336static etp_reqq req_queue;
198static reqq res_queue; 337static etp_reqq res_queue;
199 338
200static int reqq_push (reqq *q, eio_req *req) 339static int reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 340{
202 int pri = req->pri; 341 int pri = req->pri;
203 req->next = 0; 342 req->next = 0;
204 343
205 if (q->qe[pri]) 344 if (q->qe[pri])
… …
211 q->qe[pri] = q->qs[pri] = req; 350 q->qe[pri] = q->qs[pri] = req;
212 351
213 return q->size++; 352 return q->size++;
214} 353}
215 354
216static eio_req *reqq_shift (reqq *q) 355static ETP_REQ *reqq_shift (etp_reqq *q)
217{ 356{
218 int pri; 357 int pri;
219 358
220 if (!q->size) 359 if (!q->size)
221 return 0; 360 return 0;
222 361
223 --q->size; 362 --q->size;
224 363
225 for (pri = EIO_NUM_PRI; pri--; ) 364 for (pri = ETP_NUM_PRI; pri--; )
226 { 365 {
227 eio_req *req = q->qs[pri]; 366 eio_req *req = q->qs[pri];
228 367
229 if (req) 368 if (req)
230 { 369 {
… …
236 } 375 }
237 376
238 abort (); 377 abort ();
239} 378}
240 379
241static void grp_try_feed (eio_req *grp) 380static void etp_thread_init (void)
242{ 381{
243 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)
244 { 419 {
245 int old_len = grp->size; 420 etp_worker *wrk = wrk_first.next;
246 421
247 EIO_FEED (grp); 422 if (wrk->req)
423 ETP_DESTROY (wrk->req);
248 424
249 /* stop if no progress has been made */ 425 etp_worker_clear (wrk);
250 if (old_len == grp->size) 426 etp_worker_free (wrk);
251 {
252 grp->feed = 0;
253 break;
254 }
255 } 427 }
256}
257 428
258static int eio_finish (eio_req *req); 429 started = 0;
430 idle = 0;
431 nreqs = 0;
432 nready = 0;
433 npending = 0;
259 434
260static int grp_dec (eio_req *grp) 435 etp_thread_init ();
261{ 436}
262 --grp->size;
263 437
264 /* call feeder, if applicable */ 438static void
265 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}
266 444
267 /* finish, if done */ 445static int
268 if (!grp->size && grp->int1) 446etp_init (void (*want_poll)(void), void (*done_poll)(void))
269 return eio_finish (grp); 447{
270 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
271 return 0; 455 return 0;
272} 456}
273 457
274void eio_destroy (eio_req *req)
275{
276 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
277 if ((req)->flags & EIO_FLAG_PTR2_FREE) 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 X_LOCK (wrklock);
318 req->flags |= EIO_FLAG_CANCELLED;
319 X_UNLOCK (wrklock);
320
321 eio_grp_cancel (req);
322}
323
324X_THREAD_PROC (eio_proc); 458X_THREAD_PROC (etp_proc);
325 459
326static void start_thread (void) 460static void etp_start_thread (void)
327{ 461{
328 worker *wrk = calloc (1, sizeof (worker)); 462 etp_worker *wrk = calloc (1, sizeof (etp_worker));
329 463
330 /*TODO*/ 464 /*TODO*/
331 assert (("unable to allocate worker thread data", !wrk)); 465 assert (("unable to allocate worker thread data", wrk));
332 466
333 X_LOCK (wrklock); 467 X_LOCK (wrklock);
334 468
335 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 469 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
336 { 470 {
337 wrk->prev = &wrk_first; 471 wrk->prev = &wrk_first;
338 wrk->next = wrk_first.next; 472 wrk->next = wrk_first.next;
339 wrk_first.next->prev = wrk; 473 wrk_first.next->prev = wrk;
340 wrk_first.next = wrk; 474 wrk_first.next = wrk;
… …
344 free (wrk); 478 free (wrk);
345 479
346 X_UNLOCK (wrklock); 480 X_UNLOCK (wrklock);
347} 481}
348 482
349static void maybe_start_thread (void) 483static void etp_maybe_start_thread (void)
350{ 484{
351 if (eio_nthreads () >= wanted) 485 if (expect_true (etp_nthreads () >= wanted))
352 return; 486 return;
353 487
354 /* todo: maybe use idle here, but might be less exact */ 488 /* todo: maybe use idle here, but might be less exact */
355 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 ()))
356 return; 490 return;
357 491
358 start_thread (); 492 etp_start_thread ();
359} 493}
360 494
361void eio_submit (eio_req *req)
362{
363 ++nreqs;
364
365 X_LOCK (reqlock);
366 ++nready;
367 reqq_push (&req_queue, req);
368 X_COND_SIGNAL (reqwait);
369 X_UNLOCK (reqlock);
370
371 maybe_start_thread ();
372}
373
374static void end_thread (void) 495static void etp_end_thread (void)
375{ 496{
376 eio_req *req = calloc (1, sizeof (eio_req)); 497 eio_req *req = calloc (1, sizeof (eio_req));
377 498
378 req->type = EIO_QUIT; 499 req->type = -1;
379 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 500 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
380 501
381 X_LOCK (reqlock); 502 X_LOCK (reqlock);
382 reqq_push (&req_queue, req); 503 reqq_push (&req_queue, req);
383 X_COND_SIGNAL (reqwait); 504 X_COND_SIGNAL (reqwait);
384 X_UNLOCK (reqlock); 505 X_UNLOCK (reqlock);
… …
386 X_LOCK (wrklock); 507 X_LOCK (wrklock);
387 --started; 508 --started;
388 X_UNLOCK (wrklock); 509 X_UNLOCK (wrklock);
389} 510}
390 511
391void eio_set_max_poll_time (double nseconds)
392{
393 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
394 max_poll_time = nseconds;
395 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
396}
397
398void eio_set_max_poll_reqs (unsigned int maxreqs)
399{
400 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
401 max_poll_reqs = maxreqs;
402 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
403}
404
405void eio_set_max_idle (unsigned int nthreads)
406{
407 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
408 max_idle = nthreads <= 0 ? 1 : nthreads;
409 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
410}
411
412void eio_set_min_parallel (unsigned int nthreads)
413{
414 if (wanted < nthreads)
415 wanted = nthreads;
416}
417
418void eio_set_max_parallel (unsigned int nthreads)
419{
420 if (wanted > nthreads)
421 wanted = nthreads;
422
423 while (started > wanted)
424 end_thread ();
425}
426
427int eio_poll (void) 512static int etp_poll (void)
428{ 513{
429 int maxreqs = max_poll_reqs; 514 unsigned int maxreqs;
515 unsigned int maxtime;
430 struct timeval tv_start, tv_now; 516 struct timeval tv_start, tv_now;
431 eio_req *req;
432 517
518 X_LOCK (reslock);
519 maxreqs = max_poll_reqs;
520 maxtime = max_poll_time;
521 X_UNLOCK (reslock);
522
433 if (max_poll_time) 523 if (maxtime)
434 gettimeofday (&tv_start, 0); 524 gettimeofday (&tv_start, 0);
435 525
436 for (;;) 526 for (;;)
437 { 527 {
528 ETP_REQ *req;
529
438 maybe_start_thread (); 530 etp_maybe_start_thread ();
439 531
440 X_LOCK (reslock); 532 X_LOCK (reslock);
441 req = reqq_shift (&res_queue); 533 req = reqq_shift (&res_queue);
442 534
443 if (req) 535 if (req)
… …
451 X_UNLOCK (reslock); 543 X_UNLOCK (reslock);
452 544
453 if (!req) 545 if (!req)
454 return 0; 546 return 0;
455 547
548 X_LOCK (reqlock);
456 --nreqs; 549 --nreqs;
550 X_UNLOCK (reqlock);
457 551
458 if (req->type == EIO_GROUP && req->size) 552 if (expect_false (req->type == EIO_GROUP && req->size))
459 { 553 {
460 req->int1 = 1; /* mark request as delayed */ 554 req->int1 = 1; /* mark request as delayed */
461 continue; 555 continue;
462 } 556 }
463 else 557 else
464 { 558 {
465 int res = eio_finish (req); 559 int res = ETP_FINISH (req);
466 if (res) 560 if (expect_false (res))
467 return res; 561 return res;
468 } 562 }
469 563
470 if (maxreqs && !--maxreqs) 564 if (expect_false (maxreqs && !--maxreqs))
471 break; 565 break;
472 566
473 if (max_poll_time) 567 if (maxtime)
474 { 568 {
475 gettimeofday (&tv_now, 0); 569 gettimeofday (&tv_now, 0);
476 570
477 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 571 if (tvdiff (&tv_start, &tv_now) >= maxtime)
478 break; 572 break;
479 } 573 }
480 } 574 }
481 575
482 errno = EAGAIN; 576 errno = EAGAIN;
483 return -1; 577 return -1;
484} 578}
485 579
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
486/*****************************************************************************/ 807/*****************************************************************************/
487/* work around various missing functions */ 808/* work around various missing functions */
488 809
489#if !HAVE_PREADWRITE 810#if !HAVE_PREADWRITE
811# undef pread
812# undef pwrite
490# define pread aio_pread 813# define pread eio__pread
491# define pwrite aio_pwrite 814# define pwrite eio__pwrite
492 815
493/* 816static ssize_t
494 * make our pread/pwrite safe against themselves, but not against 817eio__pread (int fd, void *buf, size_t count, off_t offset)
495 * normal read/write by using a mutex. slows down execution a lot,
496 * but that's your problem, not mine.
497 */
498static mutex_t preadwritelock = X_MUTEX_INIT;
499
500static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
501{ 818{
502 ssize_t res; 819 ssize_t res;
503 off_t ooffset; 820 off_t ooffset;
504 821
505 X_LOCK (preadwritelock); 822 X_LOCK (preadwritelock);
… …
510 X_UNLOCK (preadwritelock); 827 X_UNLOCK (preadwritelock);
511 828
512 return res; 829 return res;
513} 830}
514 831
515static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 832static ssize_t
833eio__pwrite (int fd, void *buf, size_t count, off_t offset)
516{ 834{
517 ssize_t res; 835 ssize_t res;
518 off_t ooffset; 836 off_t ooffset;
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) aio_utimes (path, times) 852# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) aio_futimes (fd, times)
535 853
854static int
536static int aio_utimes (const char *filename, const struct timeval times[2]) 855eio__utimes (const char *filename, const struct timeval times[2])
537{ 856{
538 if (times) 857 if (times)
539 { 858 {
540 struct utimbuf buf; 859 struct utimbuf buf;
541 860
… …
546 } 865 }
547 else 866 else
548 return utime (filename, 0); 867 return utime (filename, 0);
549} 868}
550 869
870#endif
871
872#ifndef HAVE_FUTIMES
873
874# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times)
876
551static int aio_futimes (int fd, const struct timeval tv[2]) 877static int eio__futimes (int fd, const struct timeval tv[2])
552{ 878{
553 errno = ENOSYS; 879 errno = ENOSYS;
554 return -1; 880 return -1;
555} 881}
556 882
557#endif 883#endif
558 884
559#if !HAVE_FDATASYNC 885#if !HAVE_FDATASYNC
886# undef fdatasync
560# define fdatasync fsync 887# define fdatasync(fd) fsync (fd)
561#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}
562 917
563#if !HAVE_READAHEAD 918#if !HAVE_READAHEAD
919# undef readahead
564# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 920# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565 921
922static ssize_t
566static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 923eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
567{ 924{
568 size_t todo = count; 925 size_t todo = count;
569 dBUF; 926 dBUF;
570 927
571 while (todo > 0) 928 while (todo > 0)
… …
581 return count; 938 return count;
582} 939}
583 940
584#endif 941#endif
585 942
586#if !HAVE_READDIR_R
587# define readdir_r aio_readdir_r
588
589static mutex_t readdirlock = X_MUTEX_INIT;
590
591static int readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
592{
593 EIO_STRUCT_DIRENT *e;
594 int errorno;
595
596 X_LOCK (readdirlock);
597
598 e = readdir (dirp);
599 errorno = errno;
600
601 if (e)
602 {
603 *res = ent;
604 strcpy (ent->d_name, e->d_name);
605 }
606 else
607 *res = 0;
608
609 X_UNLOCK (readdirlock);
610
611 errno = errorno;
612 return e ? 0 : -1;
613}
614#endif
615
616/* sendfile always needs emulation */ 943/* sendfile always needs emulation */
944static ssize_t
617static ssize_t 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)
618{ 946{
947 ssize_t written = 0;
619 ssize_t res; 948 ssize_t res;
620 949
621 if (!count) 950 if (!count)
622 return 0; 951 return 0;
623 952
953 for (;;)
954 {
624#if HAVE_SENDFILE 955#if HAVE_SENDFILE
625# if __linux 956# if __linux
957 off_t soffset = offset;
626 res = sendfile (ofd, ifd, &offset, count); 958 res = sendfile (ofd, ifd, &soffset, count);
627 959
628# elif __freebsd 960# elif __FreeBSD__
629 /* 961 /*
630 * 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
631 * wasted on making it similar to other I/O functions. 963 * wasted on making it similar to other I/O functions.
632 */ 964 */
633 {
634 off_t sbytes; 965 off_t sbytes;
635 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 966 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636 967
637 if (res < 0 && sbytes) 968 #if 0 /* according to the manpage, this is correct, but broken behaviour */
638 /* 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)
639 res = sbytes; 973 res = sbytes;
640 } 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;
641 987
642# elif __hpux 988# elif __hpux
643 res = sendfile (ofd, ifd, offset, count, 0, 0); 989 res = sendfile (ofd, ifd, offset, count, 0, 0);
644 990
645# elif __solaris 991# elif __solaris
646 {
647 struct sendfilevec vec; 992 struct sendfilevec vec;
648 size_t sbytes; 993 size_t sbytes;
649 994
650 vec.sfv_fd = ifd; 995 vec.sfv_fd = ifd;
651 vec.sfv_flag = 0; 996 vec.sfv_flag = 0;
652 vec.sfv_off = offset; 997 vec.sfv_off = offset;
653 vec.sfv_len = count; 998 vec.sfv_len = count;
654 999
655 res = sendfilev (ofd, &vec, 1, &sbytes); 1000 res = sendfilev (ofd, &vec, 1, &sbytes);
656 1001
657 if (res < 0 && sbytes) 1002 if (res < 0 && sbytes)
658 res = sbytes; 1003 res = sbytes;
659 }
660 1004
661# 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
662#else 1013#else
663 res = -1; 1014 res = -1;
664 errno = ENOSYS; 1015 errno = ENOSYS;
665#endif 1016#endif
666 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
667 if (res < 0 1041 if (res < 0
668 && (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 */
669#if __solaris 1048#if __solaris
670 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1049 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
671#endif 1050#endif
672 ) 1051 )
673 ) 1052 )
… …
704 } 1083 }
705 1084
706 return res; 1085 return res;
707} 1086}
708 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
709/* read a full directory */ 1253/* read a full directory */
1254static void
710static void scandir_ (eio_req *req, worker *self) 1255eio__scandir (eio_req *req, etp_worker *self)
711{ 1256{
712 DIR *dirp; 1257 DIR *dirp;
713 union
714 {
715 EIO_STRUCT_DIRENT d;
716 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
717 } *u;
718 EIO_STRUCT_DIRENT *entp; 1258 EIO_STRUCT_DIRENT *entp;
719 char *name, *names; 1259 char *name, *names;
720 int memlen = 4096; 1260 int namesalloc = 4096;
721 int memofs = 0; 1261 int namesoffs = 0;
1262 int flags = req->int1;
1263 eio_dirent *dents = 0;
1264 int dentalloc = 128;
722 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);
723 1272
724 X_LOCK (wrklock); 1273 X_LOCK (wrklock);
1274 /* the corresponding closedir is in ETP_WORKER_CLEAR */
725 self->dirp = dirp = opendir (req->ptr1); 1275 self->dirp = dirp = opendir (req->ptr1);
726 self->dbuf = u = malloc (sizeof (*u)); 1276
727 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;
728 req->ptr2 = names = malloc (memlen); 1279 req->ptr2 = names = malloc (namesalloc);
729 X_UNLOCK (wrklock); 1280 X_UNLOCK (wrklock);
730 1281
731 if (dirp && u && names) 1282 if (dirp && names && (!flags || dents))
732 for (;;) 1283 for (;;)
733 { 1284 {
734 errno = 0; 1285 errno = 0;
735 readdir_r (dirp, &u->d, &entp); 1286 entp = readdir (dirp);
736 1287
737 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
738 break; 1326 break;
1327 }
739 1328
1329 /* now add the entry to our list(s) */
740 name = entp->d_name; 1330 name = entp->d_name;
741 1331
1332 /* skip . and .. entries */
742 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1333 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
743 { 1334 {
744 int len = strlen (name) + 1; 1335 int len = D_NAMLEN (entp) + 1;
745 1336
746 res++; 1337 while (expect_false (namesoffs + len > namesalloc))
747
748 while (memofs + len > memlen)
749 { 1338 {
750 memlen *= 2; 1339 namesalloc *= 2;
751 X_LOCK (wrklock); 1340 X_LOCK (wrklock);
752 req->ptr2 = names = realloc (names, memlen); 1341 req->ptr2 = names = realloc (names, namesalloc);
753 X_UNLOCK (wrklock); 1342 X_UNLOCK (wrklock);
754 1343
755 if (!names) 1344 if (!names)
756 break; 1345 break;
757 } 1346 }
758 1347
759 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
760 memofs += len; 1440 namesoffs += len;
1441 ++dentoffs;
1442 }
1443
1444 if (EIO_CANCELLED (req))
1445 {
1446 errno = ECANCELED;
1447 break;
761 } 1448 }
762 } 1449 }
1450}
763 1451
764 if (errno) 1452#ifdef PAGESIZE
765 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
766 1553 {
767 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;
768} 1566}
769 1567
770/*****************************************************************************/ 1568/*****************************************************************************/
771 1569
772#define ALLOC(len) \ 1570#define ALLOC(len) \
… …
782 req->result = -1; \ 1580 req->result = -1; \
783 break; \ 1581 break; \
784 } \ 1582 } \
785 } 1583 }
786 1584
787X_THREAD_PROC (eio_proc) 1585X_THREAD_PROC (etp_proc)
788{ 1586{
789 eio_req *req; 1587 ETP_REQ *req;
790 struct timespec ts; 1588 struct timespec ts;
791 worker *self = (worker *)thr_arg; 1589 etp_worker *self = (etp_worker *)thr_arg;
792 1590
793 /* try to distribute timeouts somewhat randomly */ 1591 /* try to distribute timeouts somewhat randomly */
794 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1592 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
795 1593
796 for (;;) 1594 for (;;)
797 { 1595 {
798 ts.tv_sec = time (0) + IDLE_TIMEOUT;
799
800 X_LOCK (reqlock); 1596 X_LOCK (reqlock);
801 1597
802 for (;;) 1598 for (;;)
803 { 1599 {
804 self->req = req = reqq_shift (&req_queue); 1600 self->req = req = reqq_shift (&req_queue);
… …
806 if (req) 1602 if (req)
807 break; 1603 break;
808 1604
809 ++idle; 1605 ++idle;
810 1606
1607 ts.tv_sec = time (0) + idle_timeout;
811 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1608 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
812 { 1609 {
813 if (idle > max_idle) 1610 if (idle > max_idle)
814 { 1611 {
815 --idle; 1612 --idle;
… …
820 goto quit; 1617 goto quit;
821 } 1618 }
822 1619
823 /* we are allowed to idle, so do so without any timeout */ 1620 /* we are allowed to idle, so do so without any timeout */
824 X_COND_WAIT (reqwait, reqlock); 1621 X_COND_WAIT (reqwait, reqlock);
825 ts.tv_sec = time (0) + IDLE_TIMEOUT;
826 } 1622 }
827 1623
828 --idle; 1624 --idle;
829 } 1625 }
830 1626
831 --nready; 1627 --nready;
832 1628
833 X_UNLOCK (reqlock); 1629 X_UNLOCK (reqlock);
834 1630
835 errno = 0; /* strictly unnecessary */ 1631 if (req->type < 0)
1632 goto quit;
836 1633
837 if (!EIO_CANCELLED (req)) 1634 if (!EIO_CANCELLED (req))
838 switch (req->type) 1635 ETP_EXECUTE (self, req);
839 {
840 case EIO_READ: ALLOC (req->size);
841 req->result = req->offs >= 0
842 ? pread (req->int1, req->ptr2, req->size, req->offs)
843 : read (req->int1, req->ptr2, req->size); break;
844 case EIO_WRITE: req->result = req->offs >= 0
845 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
846 : write (req->int1, req->ptr2, req->size); break;
847
848 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
849 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
850
851 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
852 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
853 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
854 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
855 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
856 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
857
858 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
859 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
860 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
861 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
862 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
863 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
864
865 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
866 case EIO_CLOSE: req->result = close (req->int1); break;
867 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
868 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
869 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
870 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
871 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
872 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
873 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
874 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
875
876 case EIO_READLINK: ALLOC (NAME_MAX);
877 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
878
879 case EIO_SYNC: req->result = 0; sync (); break;
880 case EIO_FSYNC: req->result = fsync (req->int1); break;
881 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
882
883 case EIO_READDIR: scandir_ (req, self); break;
884
885 case EIO_BUSY:
886#ifdef _WIN32
887 Sleep (req->nv1 * 1000.);
888#else
889 {
890 struct timeval tv;
891
892 tv.tv_sec = req->nv1;
893 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
894
895 req->result = select (0, 0, 0, 0, &tv);
896 }
897#endif
898 break;
899
900 case EIO_UTIME:
901 case EIO_FUTIME:
902 {
903 struct timeval tv[2];
904 struct timeval *times;
905
906 if (req->nv1 != -1. || req->nv2 != -1.)
907 {
908 tv[0].tv_sec = req->nv1;
909 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
910 tv[1].tv_sec = req->nv2;
911 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
912
913 times = tv;
914 }
915 else
916 times = 0;
917
918
919 req->result = req->type == EIO_FUTIME
920 ? futimes (req->int1, times)
921 : utimes (req->ptr1, times);
922 }
923
924 case EIO_GROUP:
925 case EIO_NOP:
926 break;
927
928 case EIO_QUIT:
929 goto quit;
930
931 default:
932 req->result = -1;
933 break;
934 }
935
936 req->errorno = errno;
937 1636
938 X_LOCK (reslock); 1637 X_LOCK (reslock);
939 1638
940 ++npending; 1639 ++npending;
941 1640
942 if (!reqq_push (&res_queue, req) && want_poll_cb) 1641 if (!reqq_push (&res_queue, req) && want_poll_cb)
943 want_poll_cb (); 1642 want_poll_cb ();
944 1643
945 self->req = 0; 1644 self->req = 0;
946 worker_clear (self); 1645 etp_worker_clear (self);
947 1646
948 X_UNLOCK (reslock); 1647 X_UNLOCK (reslock);
949 } 1648 }
950 1649
951quit: 1650quit:
952 X_LOCK (wrklock); 1651 X_LOCK (wrklock);
953 worker_free (self); 1652 etp_worker_free (self);
954 X_UNLOCK (wrklock); 1653 X_UNLOCK (wrklock);
955 1654
956 return 0; 1655 return 0;
957} 1656}
958 1657
959/*****************************************************************************/ 1658/*****************************************************************************/
960 1659
961static void eio_atfork_prepare (void)
962{
963 X_LOCK (wrklock);
964 X_LOCK (reqlock);
965 X_LOCK (reslock);
966#if !HAVE_PREADWRITE
967 X_LOCK (preadwritelock);
968#endif
969#if !HAVE_READDIR_R
970 X_LOCK (readdirlock);
971#endif
972}
973
974static void eio_atfork_parent (void)
975{
976#if !HAVE_READDIR_R
977 X_UNLOCK (readdirlock);
978#endif
979#if !HAVE_PREADWRITE
980 X_UNLOCK (preadwritelock);
981#endif
982 X_UNLOCK (reslock);
983 X_UNLOCK (reqlock);
984 X_UNLOCK (wrklock);
985}
986
987static void eio_atfork_child (void)
988{
989 eio_req *prv;
990
991 while (prv = reqq_shift (&req_queue))
992 eio_destroy (prv);
993
994 while (prv = reqq_shift (&res_queue))
995 eio_destroy (prv);
996
997 while (wrk_first.next != &wrk_first)
998 {
999 worker *wrk = wrk_first.next;
1000
1001 if (wrk->req)
1002 eio_destroy (wrk->req);
1003
1004 worker_clear (wrk);
1005 worker_free (wrk);
1006 }
1007
1008 started = 0;
1009 idle = 0;
1010 nreqs = 0;
1011 nready = 0;
1012 npending = 0;
1013
1014 eio_atfork_parent ();
1015}
1016
1017int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1660int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1018{ 1661{
1019 want_poll_cb = want_poll; 1662 return etp_init (want_poll, done_poll);
1020 done_poll_cb = done_poll;
1021
1022#ifdef _WIN32
1023 X_MUTEX_CHECK (wrklock);
1024 X_MUTEX_CHECK (reslock);
1025 X_MUTEX_CHECK (reqlock);
1026 X_MUTEX_CHECK (reqwait);
1027 X_MUTEX_CHECK (preadwritelock);
1028 X_MUTEX_CHECK (readdirlock);
1029
1030 X_COND_CHECK (reqwait);
1031#endif
1032
1033 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1034} 1663}
1035 1664
1036static void eio_api_destroy (eio_req *req) 1665static void eio_api_destroy (eio_req *req)
1037{ 1666{
1038 free (req); 1667 free (req);
… …
1043 \ 1672 \
1044 req = (eio_req *)calloc (1, sizeof *req); \ 1673 req = (eio_req *)calloc (1, sizeof *req); \
1045 if (!req) \ 1674 if (!req) \
1046 return 0; \ 1675 return 0; \
1047 \ 1676 \
1048 req->type = EIO_ ## rtype; \ 1677 req->type = rtype; \
1049 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 1678 req->pri = pri; \
1050 req->finish = cb; \ 1679 req->finish = cb; \
1680 req->data = data; \
1051 req->destroy = eio_api_destroy; 1681 req->destroy = eio_api_destroy;
1052 1682
1053#define SEND eio_submit (req); return req 1683#define SEND eio_submit (req); return req
1054 1684
1055#define PATH (void)0 1685#define PATH \
1686 req->flags |= EIO_FLAG_PTR1_FREE; \
1687 req->ptr1 = strdup (path); \
1688 if (!req->ptr1) \
1689 { \
1690 eio_api_destroy (req); \
1691 return 0; \
1692 }
1056 1693
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)
1815{
1816 REQ (EIO_NOP); SEND;
1817}
1818
1819eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1820{
1821 REQ (EIO_BUSY); req->nv1 = delay; SEND;
1822}
1823
1057eio_req *eio_fsync (int fd, eio_cb cb) 1824eio_req *eio_sync (int pri, eio_cb cb, void *data)
1058{ 1825{
1826 REQ (EIO_SYNC); SEND;
1827}
1828
1829eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1830{
1059 REQ (FSYNC); req->int1 = fd; SEND; 1831 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1060} 1832}
1061 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
1062eio_req *eio_fdatasync (int fd, eio_cb cb) 1859eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1063{ 1860{
1064 REQ (FDATASYNC); req->int1 = fd; SEND; 1861 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1065} 1862}
1066 1863
1864eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1865{
1866 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1867}
1868
1067eio_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)
1068{ 1870{
1069 REQ (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;
1070} 1872}
1071 1873
1072eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb) 1874eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1073{ 1875{
1074 REQ (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;
1075} 1877}
1076 1878
1077eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb) 1879eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1078{ 1880{
1079 REQ (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;
1080} 1882}
1081 1883
1082eio_req *eio_fstat (int fd, eio_cb cb) 1884eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1083{ 1885{
1084 REQ (FSTAT); req->int1 = fd; SEND; 1886 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1085} 1887}
1086 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
1087eio_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)
1088{ 1895{
1089 REQ (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;
1090} 1897}
1091 1898
1092eio_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)
1093{ 1900{
1094 REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 1901 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1095} 1902}
1096 1903
1097eio_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)
1098{ 1905{
1099 REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 1906 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1100} 1907}
1101 1908
1102eio_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)
1103{ 1910{
1104 REQ (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;
1105} 1912}
1106 1913
1107eio_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)
1108{ 1915{
1109 REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND; 1916 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1110} 1917}
1111 1918
1112eio_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)
1113{ 1920{
1114 REQ (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;
1115} 1922}
1116 1923
1117eio_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)
1118{ 1925{
1119 REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 1926 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1120} 1927}
1121 1928
1122eio_req *eio_readlink (const char *path, eio_cb cb)
1123{
1124 REQ (READLINK); PATH; SEND;
1125}
1126
1127eio_req *eio_stat (const char *path, eio_cb cb)
1128{
1129 REQ (STAT); PATH; SEND;
1130}
1131
1132eio_req *eio_lstat (const char *path, eio_cb cb)
1133{
1134 REQ (LSTAT); PATH; SEND;
1135}
1136
1137eio_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)
1138{ 1930{
1139 REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 1931 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1140} 1932}
1141 1933
1142eio_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)
1143{ 1935{
1144 REQ (TRUNCATE); PATH; req->offs = offset; SEND; 1936 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1145} 1937}
1146 1938
1147eio_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)
1148{ 1940{
1149 REQ (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;
1150} 1942}
1151 1943
1152eio_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)
1153{ 1945{
1154 REQ (CHMOD); PATH; req->int2 = (long)mode; SEND; 1946 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1155} 1947}
1156 1948
1157eio_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)
1158{ 1950{
1159 REQ (MKDIR); PATH; req->int2 = (long)mode; SEND; 1951 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1160} 1952}
1161 1953
1162eio_req *eio_unlink (const char *path, eio_cb cb) 1954static eio_req *
1955eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1163{ 1956{
1164 REQ (UNLINK); PATH; SEND;
1165}
1166
1167eio_req *eio_rmdir (const char *path, eio_cb cb)
1168{
1169 REQ (RMDIR); PATH; SEND; 1957 REQ (type); PATH; SEND;
1170} 1958}
1171 1959
1960eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1961{
1962 return eio__1path (EIO_READLINK, path, pri, cb, data);
1963}
1964
1965eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1966{
1967 return eio__1path (EIO_STAT, path, pri, cb, data);
1968}
1969
1970eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1971{
1972 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1973}
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
1980eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1981{
1982 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1983}
1984
1172eio_req *eio_readdir (const char *path, eio_cb cb) 1985eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1173{ 1986{
1174 REQ (READDIR); PATH; SEND; 1987 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1175} 1988}
1176 1989
1990eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1991{
1992 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1993}
1994
1177eio_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)
1178{ 1996{
1179 REQ (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;
1180} 1998}
1181 1999
1182eio_req *eio_busy (double delay, eio_cb cb) 2000static eio_req *
2001eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1183{ 2002{
1184 REQ (BUSY); req->nv1 = delay; SEND; 2003 REQ (type); PATH;
1185}
1186 2004
2005 req->flags |= EIO_FLAG_PTR2_FREE;
2006 req->ptr2 = strdup (new_path);
2007 if (!req->ptr2)
2008 {
2009 eio_api_destroy (req);
2010 return 0;
2011 }
2012
2013 SEND;
2014}
2015
2016eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2017{
2018 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2019}
2020
2021eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2022{
2023 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2024}
2025
2026eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2027{
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
1187eio_req *eio_nop (eio_cb cb) 2038eio_req *eio_grp (eio_cb cb, void *data)
1188{ 2039{
2040 const int pri = EIO_PRI_MAX;
2041
1189 REQ (NOP); SEND; 2042 REQ (EIO_GROUP); SEND;
1190} 2043}
1191 2044
1192#undef REQ 2045#undef REQ
1193#undef PATH 2046#undef PATH
1194#undef SEND 2047#undef SEND
1195 2048
1196#if 0 2049/*****************************************************************************/
1197 2050/* grp functions */
1198void
1199aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1200 PROTOTYPE: $$$;$
1201 PPCODE:
1202{
1203 dREQ;
1204
1205 req->type = EIO_OPEN;
1206 req->sv1 = newSVsv (pathname);
1207 req->ptr1 = SvPVbyte_nolen (req->sv1);
1208 req->int1 = flags;
1209 req->int2 = mode;
1210
1211 EIO_SEND;
1212}
1213
1214void
1215aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1216 PROTOTYPE: $;$
1217 ALIAS:
1218 aio_fsync = EIO_FSYNC
1219 aio_fdatasync = EIO_FDATASYNC
1220 PPCODE:
1221{
1222 dREQ;
1223
1224 req->type = ix;
1225 req->sv1 = newSVsv (fh);
1226 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1227
1228 EIO_SEND (req);
1229}
1230
1231void
1232aio_close (SV *fh, SV *callback=&PL_sv_undef)
1233 PROTOTYPE: $;$
1234 PPCODE:
1235{
1236 dREQ;
1237
1238 req->type = EIO_CLOSE;
1239 req->sv1 = newSVsv (fh);
1240 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1241
1242 EIO_SEND (req);
1243}
1244
1245void
1246aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1247 ALIAS:
1248 aio_read = EIO_READ
1249 aio_write = EIO_WRITE
1250 PROTOTYPE: $$$$$;$
1251 PPCODE:
1252{
1253 STRLEN svlen;
1254 char *svptr = SvPVbyte (data, svlen);
1255 UV len = SvUV (length);
1256
1257 SvUPGRADE (data, SVt_PV);
1258 SvPOK_on (data);
1259
1260 if (dataoffset < 0)
1261 dataoffset += svlen;
1262
1263 if (dataoffset < 0 || dataoffset > svlen)
1264 croak ("dataoffset outside of data scalar");
1265
1266 if (ix == EIO_WRITE)
1267 {
1268 /* write: check length and adjust. */
1269 if (!SvOK (length) || len + dataoffset > svlen)
1270 len = svlen - dataoffset;
1271 }
1272 else
1273 {
1274 /* read: grow scalar as necessary */
1275 svptr = SvGROW (data, len + dataoffset + 1);
1276 }
1277
1278 if (len < 0)
1279 croak ("length must not be negative");
1280
1281 {
1282 dREQ;
1283
1284 req->type = ix;
1285 req->sv1 = newSVsv (fh);
1286 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1287 : IoOFP (sv_2io (fh)));
1288 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1289 req->size = len;
1290 req->sv2 = SvREFCNT_inc (data);
1291 req->ptr2 = (char *)svptr + dataoffset;
1292 req->stroffset = dataoffset;
1293
1294 if (!SvREADONLY (data))
1295 {
1296 SvREADONLY_on (data);
1297 req->flags |= FLAG_SV2_RO_OFF;
1298 }
1299
1300 EIO_SEND;
1301 }
1302}
1303
1304void
1305aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1306 PROTOTYPE: $$;$
1307 PPCODE:
1308{
1309 SV *data;
1310 dREQ;
1311
1312 data = newSV (NAME_MAX);
1313 SvPOK_on (data);
1314
1315 req->type = EIO_READLINK;
1316 req->sv1 = newSVsv (path);
1317 req->ptr1 = SvPVbyte_nolen (req->sv1);
1318 req->sv2 = data;
1319 req->ptr2 = SvPVbyte_nolen (data);
1320
1321 EIO_SEND;
1322}
1323
1324void
1325aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1326 PROTOTYPE: $$$$;$
1327 PPCODE:
1328{
1329 dREQ;
1330
1331 req->type = EIO_SENDFILE;
1332 req->sv1 = newSVsv (out_fh);
1333 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1334 req->sv2 = newSVsv (in_fh);
1335 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1336 req->offs = SvVAL64 (in_offset);
1337 req->size = length;
1338
1339 EIO_SEND;
1340}
1341
1342void
1343aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1344 PROTOTYPE: $$$;$
1345 PPCODE:
1346{
1347 dREQ;
1348
1349 req->type = EIO_READAHEAD;
1350 req->sv1 = newSVsv (fh);
1351 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1352 req->offs = SvVAL64 (offset);
1353 req->size = length;
1354
1355 EIO_SEND;
1356}
1357
1358void
1359aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1360 ALIAS:
1361 aio_stat = EIO_STAT
1362 aio_lstat = EIO_LSTAT
1363 PPCODE:
1364{
1365 dREQ;
1366
1367 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1368 if (!req->ptr2)
1369 {
1370 req_destroy (req);
1371 croak ("out of memory during aio_stat statdata allocation");
1372 }
1373
1374 req->flags |= FLAG_PTR2_FREE;
1375 req->sv1 = newSVsv (fh_or_path);
1376
1377 if (SvPOK (fh_or_path))
1378 {
1379 req->type = ix;
1380 req->ptr1 = SvPVbyte_nolen (req->sv1);
1381 }
1382 else
1383 {
1384 req->type = EIO_FSTAT;
1385 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1386 }
1387
1388 EIO_SEND;
1389}
1390
1391void
1392aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1393 PPCODE:
1394{
1395 dREQ;
1396
1397 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1398 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1399 req->sv1 = newSVsv (fh_or_path);
1400
1401 if (SvPOK (fh_or_path))
1402 {
1403 req->type = EIO_UTIME;
1404 req->ptr1 = SvPVbyte_nolen (req->sv1);
1405 }
1406 else
1407 {
1408 req->type = EIO_FUTIME;
1409 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1410 }
1411
1412 EIO_SEND;
1413}
1414
1415void
1416aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1417 PPCODE:
1418{
1419 dREQ;
1420
1421 req->sv1 = newSVsv (fh_or_path);
1422 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1423
1424 if (SvPOK (fh_or_path))
1425 {
1426 req->type = EIO_TRUNCATE;
1427 req->ptr1 = SvPVbyte_nolen (req->sv1);
1428 }
1429 else
1430 {
1431 req->type = EIO_FTRUNCATE;
1432 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1433 }
1434
1435 EIO_SEND;
1436}
1437
1438void
1439aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1440 ALIAS:
1441 aio_chmod = EIO_CHMOD
1442 aio_fchmod = EIO_FCHMOD
1443 aio_mkdir = EIO_MKDIR
1444 PPCODE:
1445{
1446 dREQ;
1447
1448 req->type = type;
1449 req->int2 = mode;
1450 req->sv1 = newSVsv (fh_or_path);
1451
1452 if (ix == EIO_FCHMOD)
1453 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1454 else
1455 req->ptr1 = SvPVbyte_nolen (req->sv1);
1456
1457 EIO_SEND;
1458}
1459
1460void
1461aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1462 PPCODE:
1463{
1464 dREQ;
1465
1466 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1467 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1468 req->sv1 = newSVsv (fh_or_path);
1469
1470 if (SvPOK (fh_or_path))
1471 {
1472 req->type = EIO_CHOWN;
1473 req->ptr1 = SvPVbyte_nolen (req->sv1);
1474 }
1475 else
1476 {
1477 req->type = EIO_FCHOWN;
1478 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1479 }
1480
1481 EIO_SEND;
1482}
1483
1484void
1485aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1486 ALIAS:
1487 aio_unlink = EIO_UNLINK
1488 aio_rmdir = EIO_RMDIR
1489 aio_readdir = EIO_READDIR
1490 PPCODE:
1491{
1492 dREQ;
1493
1494 req->type = ix;
1495 req->sv1 = newSVsv (pathname);
1496 req->ptr1 = SvPVbyte_nolen (req->sv1);
1497
1498 EIO_SEND;
1499}
1500
1501void
1502aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1503 ALIAS:
1504 aio_link = EIO_LINK
1505 aio_symlink = EIO_SYMLINK
1506 aio_rename = EIO_RENAME
1507 PPCODE:
1508{
1509 dREQ;
1510
1511 req->type = ix;
1512 req->sv1 = newSVsv (oldpath);
1513 req->ptr1 = SvPVbyte_nolen (req->sv1);
1514 req->sv2 = newSVsv (newpath);
1515 req->ptr2 = SvPVbyte_nolen (req->sv2);
1516
1517 EIO_SEND;
1518}
1519
1520void
1521aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1522 PPCODE:
1523{
1524 dREQ;
1525
1526 req->type = EIO_MKNOD;
1527 req->sv1 = newSVsv (pathname);
1528 req->ptr1 = SvPVbyte_nolen (req->sv1);
1529 req->int2 = (mode_t)mode;
1530 req->offs = dev;
1531
1532 EIO_SEND;
1533}
1534
1535void
1536aio_busy (double delay, SV *callback=&PL_sv_undef)
1537 PPCODE:
1538{
1539 dREQ;
1540
1541 req->type = EIO_BUSY;
1542 req->nv1 = delay < 0. ? 0. : delay;
1543
1544 EIO_SEND;
1545}
1546
1547void
1548aio_group (SV *callback=&PL_sv_undef)
1549 PROTOTYPE: ;$
1550 PPCODE:
1551{
1552 dREQ;
1553
1554 req->type = EIO_GROUP;
1555
1556 req_send (req);
1557 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1558}
1559
1560void
1561aio_nop (SV *callback=&PL_sv_undef)
1562 ALIAS:
1563 aio_nop = EIO_NOP
1564 aio_sync = EIO_SYNC
1565 PPCODE:
1566{
1567 dREQ;
1568
1569#endif
1570 2051
1571void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2052void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1572{ 2053{
1573 grp->int2 = limit; 2054 grp->int2 = limit;
1574 grp->feed = feed; 2055 grp->feed = feed;
… …
1585 2066
1586void eio_grp_add (eio_req *grp, eio_req *req) 2067void eio_grp_add (eio_req *grp, eio_req *req)
1587{ 2068{
1588 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));
1589 2070
2071 grp->flags |= EIO_FLAG_GROUPADD;
2072
1590 ++grp->size; 2073 ++grp->size;
1591 req->grp = grp; 2074 req->grp = grp;
1592 2075
1593 req->grp_prev = 0; 2076 req->grp_prev = 0;
1594 req->grp_next = grp->grp_first; 2077 req->grp_next = grp->grp_first;
… …
1597 grp->grp_first->grp_prev = req; 2080 grp->grp_first->grp_prev = req;
1598 2081
1599 grp->grp_first = req; 2082 grp->grp_first = req;
1600} 2083}
1601 2084
2085/*****************************************************************************/
2086/* misc garbage */
1602 2087
2088ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2089{
2090 etp_worker wrk;
2091 ssize_t ret;
2092
2093 wrk.dbuf = 0;
2094
2095 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2096
2097 if (wrk.dbuf)
2098 free (wrk.dbuf);
2099
2100 return ret;
2101}
2102

Diff Legend

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