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

Comparing libeio/eio.c (file contents):
Revision 1.4 by root, Sun May 11 00:10:15 2008 UTC vs.
Revision 1.76 by root, Tue Jul 5 09:24:12 2011 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL.
38 */
39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
1#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
2#include "xthread.h" 50#include "xthread.h"
3 51
4#include <errno.h> 52#include <errno.h>
5#include <stddef.h> 53#include <stddef.h>
6#include <stdlib.h> 54#include <stdlib.h>
7#include <string.h> 55#include <string.h>
8#include <errno.h> 56#include <errno.h>
9#include <sys/types.h> 57#include <sys/types.h>
10#include <sys/stat.h> 58#include <sys/stat.h>
59#include <sys/statvfs.h>
11#include <limits.h> 60#include <limits.h>
12#include <fcntl.h> 61#include <fcntl.h>
13#include <sched.h> 62#include <assert.h>
63
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65/* intptr_t only comes form stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H
67# include <stdint.h>
68#endif
14 69
15#ifndef EIO_FINISH 70#ifndef EIO_FINISH
16# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 71# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17#endif 72#endif
18 73
… …
25#endif 80#endif
26 81
27#ifdef _WIN32 82#ifdef _WIN32
28 83
29 /*doh*/ 84 /*doh*/
30
31#else 85#else
32 86
33# include "config.h"
34# include <sys/time.h> 87# include <sys/time.h>
35# include <sys/select.h> 88# include <sys/select.h>
36# include <unistd.h> 89# include <unistd.h>
37# include <utime.h> 90# include <utime.h>
38# include <signal.h> 91# include <signal.h>
39# include <dirent.h> 92# include <dirent.h>
40 93
94#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
95# include <sys/mman.h>
96#endif
97
98/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
99# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
100# define _DIRENT_HAVE_D_TYPE /* sigh */
101# define D_INO(de) (de)->d_fileno
102# define D_NAMLEN(de) (de)->d_namlen
103# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
104# define D_INO(de) (de)->d_ino
105# endif
106
107#ifdef _D_EXACT_NAMLEN
108# undef D_NAMLEN
109# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
110#endif
111
112# ifdef _DIRENT_HAVE_D_TYPE
113# define D_TYPE(de) (de)->d_type
114# endif
115
41# ifndef EIO_STRUCT_DIRENT 116# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 117# define EIO_STRUCT_DIRENT struct dirent
43# endif 118# endif
44 119
45#endif 120#endif
46 121
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif
50
51#if HAVE_SENDFILE 122#if HAVE_SENDFILE
52# if __linux 123# if __linux
53# include <sys/sendfile.h> 124# include <sys/sendfile.h>
54# elif __freebsd 125# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 126# include <sys/socket.h>
56# include <sys/uio.h> 127# include <sys/uio.h>
57# elif __hpux 128# elif __hpux
58# include <sys/socket.h> 129# include <sys/socket.h>
59# elif __solaris /* not yet */ 130# elif __solaris
60# include <sys/sendfile.h> 131# include <sys/sendfile.h>
61# else 132# else
62# error sendfile support requested but not available 133# error sendfile support requested but not available
63# endif 134# endif
64#endif 135#endif
65 136
66/* number of seconds after which an idle threads exit */ 137#ifndef D_TYPE
67#define IDLE_TIMEOUT 10 138# define D_TYPE(de) 0
139#endif
140#ifndef D_INO
141# define D_INO(de) 0
142#endif
143#ifndef D_NAMLEN
144# define D_NAMLEN(de) strlen ((de)->d_name)
145#endif
68 146
69/* used for struct dirent, AIX doesn't provide it */ 147/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 148#ifndef NAME_MAX
71# define NAME_MAX 4096 149# define NAME_MAX 4096
72#endif 150#endif
73 151
152/* used for readlink etc. */
153#ifndef PATH_MAX
154# define PATH_MAX 4096
155#endif
156
74/* buffer size for various temporary buffers */ 157/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 158#define EIO_BUFSIZE 65536
76 159
77#define dBUF \ 160#define dBUF \
78 char *eio_buf; \ 161 char *eio_buf; \
79 X_LOCK (wrklock); \ 162 ETP_WORKER_LOCK (self); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 163 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 164 ETP_WORKER_UNLOCK (self); \
165 errno = ENOMEM; \
82 if (!eio_buf) \ 166 if (!eio_buf) \
83 return -1; 167 return -1;
84 168
85#define EIO_TICKS ((1000000 + 1023) >> 10) 169#define EIO_TICKS ((1000000 + 1023) >> 10)
86 170
171#define ETP_PRI_MIN EIO_PRI_MIN
172#define ETP_PRI_MAX EIO_PRI_MAX
173
174struct etp_worker;
175
176#define ETP_REQ eio_req
177#define ETP_DESTROY(req) eio_destroy (req)
178static int eio_finish (eio_req *req);
179#define ETP_FINISH(req) eio_finish (req)
180static void eio_execute (struct etp_worker *self, eio_req *req);
181#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
182
183#define ETP_WORKER_CLEAR(req) \
184 if (wrk->dbuf) \
185 { \
186 free (wrk->dbuf); \
187 wrk->dbuf = 0; \
188 } \
189 \
190 if (wrk->dirp) \
191 { \
192 closedir (wrk->dirp); \
193 wrk->dirp = 0; \
194 }
195
196#define ETP_WORKER_COMMON \
197 void *dbuf; \
198 DIR *dirp;
199
200/*****************************************************************************/
201
202#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
203
204/* calculate time difference in ~1/EIO_TICKS of a second */
205ecb_inline int
206tvdiff (struct timeval *tv1, struct timeval *tv2)
207{
208 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
209 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
210}
211
212static unsigned int started, idle, wanted = 4;
213
87static void (*want_poll_cb) (void); 214static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 215static void (*done_poll_cb) (void);
89 216
90static unsigned int max_poll_time = 0; 217static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 218static unsigned int max_poll_reqs; /* reslock */
92 219
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 220static volatile unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 221static volatile unsigned int nready; /* reqlock */
95{ 222static volatile unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 223static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 224static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 225
100static unsigned int started, idle, wanted; 226static xmutex_t wrklock;
227static xmutex_t reslock;
228static xmutex_t reqlock;
229static xcond_t reqwait;
230
231#if !HAVE_PREADWRITE
232/*
233 * make our pread/pwrite emulation safe against themselves, but not against
234 * normal read/write by using a mutex. slows down execution a lot,
235 * but that's your problem, not mine.
236 */
237static xmutex_t preadwritelock = X_MUTEX_INIT;
238#endif
239
240typedef struct etp_worker
241{
242 /* locked by wrklock */
243 struct etp_worker *prev, *next;
244
245 xthread_t tid;
246
247 /* locked by reslock, reqlock or wrklock */
248 ETP_REQ *req; /* currently processed request */
249
250 ETP_WORKER_COMMON
251} etp_worker;
252
253static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
254
255#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
256#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
101 257
102/* worker threads management */ 258/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 259
105typedef struct worker { 260static void ecb_cold
106 /* locked by wrklock */
107 struct worker *prev, *next;
108
109 thread_t tid;
110
111 /* locked by reslock, reqlock or wrklock */
112 eio_req *req; /* currently processed request */
113 void *dbuf;
114 DIR *dirp;
115} worker;
116
117static worker wrk_first = { &wrk_first, &wrk_first, 0 };
118
119static void worker_clear (worker *wrk) 261etp_worker_clear (etp_worker *wrk)
120{ 262{
121 if (wrk->dirp) 263 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} 264}
133 265
134static void worker_free (worker *wrk) 266static void ecb_cold
267etp_worker_free (etp_worker *wrk)
135{ 268{
136 wrk->next->prev = wrk->prev; 269 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 270 wrk->prev->next = wrk->next;
138 271
139 free (wrk); 272 free (wrk);
140} 273}
141 274
142static volatile unsigned int nreqs, nready, npending; 275static unsigned int
143static volatile unsigned int max_idle = 4; 276etp_nreqs (void)
144
145static mutex_t reslock = X_MUTEX_INIT;
146static mutex_t reqlock = X_MUTEX_INIT;
147static cond_t reqwait = X_COND_INIT;
148
149unsigned int eio_nreqs (void)
150{ 277{
278 int retval;
279 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
280 retval = nreqs;
281 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 282 return retval;
152} 283}
153 284
154unsigned int eio_nready (void) 285static unsigned int
286etp_nready (void)
155{ 287{
156 unsigned int retval; 288 unsigned int retval;
157 289
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 291 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 293
162 return retval; 294 return retval;
163} 295}
164 296
165unsigned int eio_npending (void) 297static unsigned int
298etp_npending (void)
166{ 299{
167 unsigned int retval; 300 unsigned int retval;
168 301
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 302 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 303 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 304 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 305
173 return retval; 306 return retval;
174} 307}
175 308
176unsigned int eio_nthreads (void) 309static unsigned int
310etp_nthreads (void)
177{ 311{
178 unsigned int retval; 312 unsigned int retval;
179 313
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 314 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 315 retval = started;
… …
188 * a somewhat faster data structure might be nice, but 322 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 323 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 324 * per shift, the most expensive operation.
191 */ 325 */
192typedef struct { 326typedef struct {
193 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 327 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194 int size; 328 int size;
195} reqq; 329} etp_reqq;
196 330
197static reqq req_queue; 331static etp_reqq req_queue;
198static reqq res_queue; 332static etp_reqq res_queue;
199 333
200static int reqq_push (reqq *q, eio_req *req) 334static int ecb_noinline
335reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 336{
202 int pri = req->pri; 337 int pri = req->pri;
203 req->next = 0; 338 req->next = 0;
204 339
205 if (q->qe[pri]) 340 if (q->qe[pri])
… …
211 q->qe[pri] = q->qs[pri] = req; 346 q->qe[pri] = q->qs[pri] = req;
212 347
213 return q->size++; 348 return q->size++;
214} 349}
215 350
216static eio_req *reqq_shift (reqq *q) 351static ETP_REQ * ecb_noinline
352reqq_shift (etp_reqq *q)
217{ 353{
218 int pri; 354 int pri;
219 355
220 if (!q->size) 356 if (!q->size)
221 return 0; 357 return 0;
222 358
223 --q->size; 359 --q->size;
224 360
225 for (pri = EIO_NUM_PRI; pri--; ) 361 for (pri = ETP_NUM_PRI; pri--; )
226 { 362 {
227 eio_req *req = q->qs[pri]; 363 eio_req *req = q->qs[pri];
228 364
229 if (req) 365 if (req)
230 { 366 {
… …
236 } 372 }
237 373
238 abort (); 374 abort ();
239} 375}
240 376
241static void grp_try_feed (eio_req *grp) 377static void ecb_cold
378etp_thread_init (void)
242{ 379{
243 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 380 X_MUTEX_CREATE (wrklock);
381 X_MUTEX_CREATE (reslock);
382 X_MUTEX_CREATE (reqlock);
383 X_COND_CREATE (reqwait);
384}
385
386static void ecb_cold
387etp_atfork_prepare (void)
388{
389 X_LOCK (wrklock);
390 X_LOCK (reqlock);
391 X_LOCK (reslock);
392#if !HAVE_PREADWRITE
393 X_LOCK (preadwritelock);
394#endif
395}
396
397static void ecb_cold
398etp_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 ecb_cold
409etp_atfork_child (void)
410{
411 ETP_REQ *prv;
412
413 while ((prv = reqq_shift (&req_queue)))
414 ETP_DESTROY (prv);
415
416 while ((prv = reqq_shift (&res_queue)))
417 ETP_DESTROY (prv);
418
419 while (wrk_first.next != &wrk_first)
244 { 420 {
245 int old_len = grp->size; 421 etp_worker *wrk = wrk_first.next;
246 422
247 EIO_FEED (grp); 423 if (wrk->req)
424 ETP_DESTROY (wrk->req);
248 425
249 /* stop if no progress has been made */ 426 etp_worker_clear (wrk);
250 if (old_len == grp->size) 427 etp_worker_free (wrk);
251 {
252 grp->feed = 0;
253 break;
254 }
255 } 428 }
256}
257 429
258static int eio_finish (eio_req *req); 430 started = 0;
431 idle = 0;
432 nreqs = 0;
433 nready = 0;
434 npending = 0;
259 435
260static int grp_dec (eio_req *grp) 436 etp_thread_init ();
261{ 437}
262 --grp->size;
263 438
264 /* call feeder, if applicable */ 439static void ecb_cold
265 grp_try_feed (grp); 440etp_once_init (void)
441{
442 etp_thread_init ();
443 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
444}
266 445
267 /* finish, if done */ 446static int ecb_cold
268 if (!grp->size && grp->int1) 447etp_init (void (*want_poll)(void), void (*done_poll)(void))
269 return eio_finish (grp); 448{
270 else 449 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
450
451 pthread_once (&doinit, etp_once_init);
452
453 want_poll_cb = want_poll;
454 done_poll_cb = done_poll;
455
271 return 0; 456 return 0;
272} 457}
273 458
274void eio_destroy (eio_req *req)
275{
276 if ((req)->flags & EIO_FLAG_PTR2_FREE)
277 free (req->ptr2);
278
279 EIO_DESTROY (req);
280}
281
282static int eio_finish (eio_req *req)
283{
284 int res = EIO_FINISH (req);
285
286 if (req->grp)
287 {
288 int res2;
289 eio_req *grp = req->grp;
290
291 /* unlink request */
292 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
293 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
294
295 if (grp->grp_first == req)
296 grp->grp_first = req->grp_next;
297
298 res2 = grp_dec (grp);
299
300 if (!res && res2)
301 res = res2;
302 }
303
304 eio_destroy (req);
305
306 return res;
307}
308
309void eio_grp_cancel (eio_req *grp)
310{
311 for (grp = grp->grp_first; grp; grp = grp->grp_next)
312 eio_cancel (grp);
313}
314
315void eio_cancel (eio_req *req)
316{
317 req->flags |= EIO_FLAG_CANCELLED;
318
319 eio_grp_cancel (req);
320}
321
322X_THREAD_PROC (eio_proc); 459X_THREAD_PROC (etp_proc);
323 460
324static void start_thread (void) 461static void ecb_cold
462etp_start_thread (void)
325{ 463{
326 worker *wrk = calloc (1, sizeof (worker)); 464 etp_worker *wrk = calloc (1, sizeof (etp_worker));
327 465
328 if (!wrk) 466 /*TODO*/
329 croak ("unable to allocate worker thread data"); 467 assert (("unable to allocate worker thread data", wrk));
330 468
331 X_LOCK (wrklock); 469 X_LOCK (wrklock);
332 470
333 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 471 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334 { 472 {
335 wrk->prev = &wrk_first; 473 wrk->prev = &wrk_first;
336 wrk->next = wrk_first.next; 474 wrk->next = wrk_first.next;
337 wrk_first.next->prev = wrk; 475 wrk_first.next->prev = wrk;
338 wrk_first.next = wrk; 476 wrk_first.next = wrk;
… …
342 free (wrk); 480 free (wrk);
343 481
344 X_UNLOCK (wrklock); 482 X_UNLOCK (wrklock);
345} 483}
346 484
485static void
347static void maybe_start_thread (void) 486etp_maybe_start_thread (void)
348{ 487{
349 if (eio_nthreads () >= wanted) 488 if (ecb_expect_true (etp_nthreads () >= wanted))
350 return; 489 return;
351 490
352 /* todo: maybe use idle here, but might be less exact */ 491 /* todo: maybe use idle here, but might be less exact */
353 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 492 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354 return; 493 return;
355 494
356 start_thread (); 495 etp_start_thread ();
357} 496}
358 497
359void eio_submit (eio_req *req) 498static void ecb_cold
360{ 499etp_end_thread (void)
361 ++nreqs;
362
363 X_LOCK (reqlock);
364 ++nready;
365 reqq_push (&req_queue, req);
366 X_COND_SIGNAL (reqwait);
367 X_UNLOCK (reqlock);
368
369 maybe_start_thread ();
370}
371
372static void end_thread (void)
373{ 500{
374 eio_req *req = calloc (1, sizeof (eio_req)); 501 eio_req *req = calloc (1, sizeof (eio_req));
375 502
376 req->type = EIO_QUIT; 503 req->type = -1;
377 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 504 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
378 505
379 X_LOCK (reqlock); 506 X_LOCK (reqlock);
380 reqq_push (&req_queue, req); 507 reqq_push (&req_queue, req);
381 X_COND_SIGNAL (reqwait); 508 X_COND_SIGNAL (reqwait);
382 X_UNLOCK (reqlock); 509 X_UNLOCK (reqlock);
… …
384 X_LOCK (wrklock); 511 X_LOCK (wrklock);
385 --started; 512 --started;
386 X_UNLOCK (wrklock); 513 X_UNLOCK (wrklock);
387} 514}
388 515
389void eio_set_max_poll_time (double nseconds) 516static int
390{
391 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
392 max_poll_time = nseconds;
393 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
394}
395
396void eio_set_max_poll_reqs (unsigned int maxreqs)
397{
398 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
399 max_poll_reqs = maxreqs;
400 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
401}
402
403void eio_set_max_idle (unsigned int nthreads)
404{
405 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
406 max_idle = nthreads <= 0 ? 1 : nthreads;
407 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
408}
409
410void eio_set_min_parallel (unsigned int nthreads)
411{
412 if (wanted < nthreads)
413 wanted = nthreads;
414}
415
416void eio_set_max_parallel (unsigned int nthreads)
417{
418 if (wanted > nthreads)
419 wanted = nthreads;
420
421 while (started > wanted)
422 end_thread ();
423}
424
425int eio_poll (void) 517etp_poll (void)
426{ 518{
427 int maxreqs = max_poll_reqs; 519 unsigned int maxreqs;
520 unsigned int maxtime;
428 struct timeval tv_start, tv_now; 521 struct timeval tv_start, tv_now;
429 eio_req *req;
430 522
523 X_LOCK (reslock);
524 maxreqs = max_poll_reqs;
525 maxtime = max_poll_time;
526 X_UNLOCK (reslock);
527
431 if (max_poll_time) 528 if (maxtime)
432 gettimeofday (&tv_start, 0); 529 gettimeofday (&tv_start, 0);
433 530
434 for (;;) 531 for (;;)
435 { 532 {
533 ETP_REQ *req;
534
436 maybe_start_thread (); 535 etp_maybe_start_thread ();
437 536
438 X_LOCK (reslock); 537 X_LOCK (reslock);
439 req = reqq_shift (&res_queue); 538 req = reqq_shift (&res_queue);
440 539
441 if (req) 540 if (req)
… …
449 X_UNLOCK (reslock); 548 X_UNLOCK (reslock);
450 549
451 if (!req) 550 if (!req)
452 return 0; 551 return 0;
453 552
553 X_LOCK (reqlock);
454 --nreqs; 554 --nreqs;
555 X_UNLOCK (reqlock);
455 556
456 if (req->type == EIO_GROUP && req->size) 557 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
457 { 558 {
458 req->int1 = 1; /* mark request as delayed */ 559 req->int1 = 1; /* mark request as delayed */
459 continue; 560 continue;
460 } 561 }
461 else 562 else
462 { 563 {
463 int res = eio_finish (req); 564 int res = ETP_FINISH (req);
464 if (res) 565 if (ecb_expect_false (res))
465 return res; 566 return res;
466 } 567 }
467 568
468 if (maxreqs && !--maxreqs) 569 if (ecb_expect_false (maxreqs && !--maxreqs))
469 break; 570 break;
470 571
471 if (max_poll_time) 572 if (maxtime)
472 { 573 {
473 gettimeofday (&tv_now, 0); 574 gettimeofday (&tv_now, 0);
474 575
475 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 576 if (tvdiff (&tv_start, &tv_now) >= maxtime)
476 break; 577 break;
477 } 578 }
478 } 579 }
479 580
480 errno = EAGAIN; 581 errno = EAGAIN;
481 return -1; 582 return -1;
482} 583}
483 584
585static void
586etp_cancel (ETP_REQ *req)
587{
588 X_LOCK (wrklock);
589 req->flags |= EIO_FLAG_CANCELLED;
590 X_UNLOCK (wrklock);
591
592 eio_grp_cancel (req);
593}
594
595static void
596etp_submit (ETP_REQ *req)
597{
598 req->pri -= ETP_PRI_MIN;
599
600 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
601 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
602
603 if (ecb_expect_false (req->type == EIO_GROUP))
604 {
605 /* I hope this is worth it :/ */
606 X_LOCK (reqlock);
607 ++nreqs;
608 X_UNLOCK (reqlock);
609
610 X_LOCK (reslock);
611
612 ++npending;
613
614 if (!reqq_push (&res_queue, req) && want_poll_cb)
615 want_poll_cb ();
616
617 X_UNLOCK (reslock);
618 }
619 else
620 {
621 X_LOCK (reqlock);
622 ++nreqs;
623 ++nready;
624 reqq_push (&req_queue, req);
625 X_COND_SIGNAL (reqwait);
626 X_UNLOCK (reqlock);
627
628 etp_maybe_start_thread ();
629 }
630}
631
632static void ecb_cold
633etp_set_max_poll_time (double nseconds)
634{
635 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
636 max_poll_time = nseconds * EIO_TICKS;
637 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
638}
639
640static void ecb_cold
641etp_set_max_poll_reqs (unsigned int maxreqs)
642{
643 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
644 max_poll_reqs = maxreqs;
645 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
646}
647
648static void ecb_cold
649etp_set_max_idle (unsigned int nthreads)
650{
651 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
652 max_idle = nthreads;
653 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
654}
655
656static void ecb_cold
657etp_set_idle_timeout (unsigned int seconds)
658{
659 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
660 idle_timeout = seconds;
661 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
662}
663
664static void ecb_cold
665etp_set_min_parallel (unsigned int nthreads)
666{
667 if (wanted < nthreads)
668 wanted = nthreads;
669}
670
671static void ecb_cold
672etp_set_max_parallel (unsigned int nthreads)
673{
674 if (wanted > nthreads)
675 wanted = nthreads;
676
677 while (started > wanted)
678 etp_end_thread ();
679}
680
681/*****************************************************************************/
682
683static void
684grp_try_feed (eio_req *grp)
685{
686 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
687 {
688 grp->flags &= ~EIO_FLAG_GROUPADD;
689
690 EIO_FEED (grp);
691
692 /* stop if no progress has been made */
693 if (!(grp->flags & EIO_FLAG_GROUPADD))
694 {
695 grp->feed = 0;
696 break;
697 }
698 }
699}
700
701static int
702grp_dec (eio_req *grp)
703{
704 --grp->size;
705
706 /* call feeder, if applicable */
707 grp_try_feed (grp);
708
709 /* finish, if done */
710 if (!grp->size && grp->int1)
711 return eio_finish (grp);
712 else
713 return 0;
714}
715
716void
717eio_destroy (eio_req *req)
718{
719 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
720 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
721
722 EIO_DESTROY (req);
723}
724
725static int
726eio_finish (eio_req *req)
727{
728 int res = EIO_FINISH (req);
729
730 if (req->grp)
731 {
732 int res2;
733 eio_req *grp = req->grp;
734
735 /* unlink request */
736 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
737 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
738
739 if (grp->grp_first == req)
740 grp->grp_first = req->grp_next;
741
742 res2 = grp_dec (grp);
743
744 if (!res && res2)
745 res = res2;
746 }
747
748 eio_destroy (req);
749
750 return res;
751}
752
753void
754eio_grp_cancel (eio_req *grp)
755{
756 for (grp = grp->grp_first; grp; grp = grp->grp_next)
757 eio_cancel (grp);
758}
759
760void
761eio_cancel (eio_req *req)
762{
763 etp_cancel (req);
764}
765
766void
767eio_submit (eio_req *req)
768{
769 etp_submit (req);
770}
771
772unsigned int
773eio_nreqs (void)
774{
775 return etp_nreqs ();
776}
777
778unsigned int
779eio_nready (void)
780{
781 return etp_nready ();
782}
783
784unsigned int
785eio_npending (void)
786{
787 return etp_npending ();
788}
789
790unsigned int ecb_cold
791eio_nthreads (void)
792{
793 return etp_nthreads ();
794}
795
796void ecb_cold
797eio_set_max_poll_time (double nseconds)
798{
799 etp_set_max_poll_time (nseconds);
800}
801
802void ecb_cold
803eio_set_max_poll_reqs (unsigned int maxreqs)
804{
805 etp_set_max_poll_reqs (maxreqs);
806}
807
808void ecb_cold
809eio_set_max_idle (unsigned int nthreads)
810{
811 etp_set_max_idle (nthreads);
812}
813
814void ecb_cold
815eio_set_idle_timeout (unsigned int seconds)
816{
817 etp_set_idle_timeout (seconds);
818}
819
820void ecb_cold
821eio_set_min_parallel (unsigned int nthreads)
822{
823 etp_set_min_parallel (nthreads);
824}
825
826void ecb_cold
827eio_set_max_parallel (unsigned int nthreads)
828{
829 etp_set_max_parallel (nthreads);
830}
831
832int eio_poll (void)
833{
834 return etp_poll ();
835}
836
484/*****************************************************************************/ 837/*****************************************************************************/
485/* work around various missing functions */ 838/* work around various missing functions */
486 839
840#if _POSIX_VERSION < 200809L
841# define realpath(path,resolved_path) (errno = ENOSYS, 0)
842#endif
843
487#if !HAVE_PREADWRITE 844#if !HAVE_PREADWRITE
845# undef pread
846# undef pwrite
488# define pread aio_pread 847# define pread eio__pread
489# define pwrite aio_pwrite 848# define pwrite eio__pwrite
490 849
491/* 850static ssize_t
492 * make our pread/pwrite safe against themselves, but not against 851eio__pread (int fd, void *buf, size_t count, off_t offset)
493 * normal read/write by using a mutex. slows down execution a lot,
494 * but that's your problem, not mine.
495 */
496static mutex_t preadwritelock = X_MUTEX_INIT;
497
498static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
499{ 852{
500 ssize_t res; 853 ssize_t res;
501 off_t ooffset; 854 off_t ooffset;
502 855
503 X_LOCK (preadwritelock); 856 X_LOCK (preadwritelock);
… …
508 X_UNLOCK (preadwritelock); 861 X_UNLOCK (preadwritelock);
509 862
510 return res; 863 return res;
511} 864}
512 865
513static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 866static ssize_t
867eio__pwrite (int fd, void *buf, size_t count, off_t offset)
514{ 868{
515 ssize_t res; 869 ssize_t res;
516 off_t ooffset; 870 off_t ooffset;
517 871
518 X_LOCK (preadwritelock); 872 X_LOCK (preadwritelock);
519 ooffset = lseek (fd, 0, SEEK_CUR); 873 ooffset = lseek (fd, 0, SEEK_CUR);
520 lseek (fd, offset, SEEK_SET); 874 lseek (fd, offset, SEEK_SET);
521 res = write (fd, buf, count); 875 res = write (fd, buf, count);
522 lseek (fd, offset, SEEK_SET); 876 lseek (fd, ooffset, SEEK_SET);
523 X_UNLOCK (preadwritelock); 877 X_UNLOCK (preadwritelock);
524 878
525 return res; 879 return res;
526} 880}
527#endif 881#endif
528 882
529#ifndef HAVE_FUTIMES 883#ifndef HAVE_UTIMES
530 884
885# undef utimes
531# define utimes(path,times) aio_utimes (path, times) 886# define utimes(path,times) eio__utimes (path, times)
532# define futimes(fd,times) aio_futimes (fd, times)
533 887
888static int
534static int aio_utimes (const char *filename, const struct timeval times[2]) 889eio__utimes (const char *filename, const struct timeval times[2])
535{ 890{
536 if (times) 891 if (times)
537 { 892 {
538 struct utimbuf buf; 893 struct utimbuf buf;
539 894
… …
544 } 899 }
545 else 900 else
546 return utime (filename, 0); 901 return utime (filename, 0);
547} 902}
548 903
904#endif
905
906#ifndef HAVE_FUTIMES
907
908# undef futimes
909# define futimes(fd,times) eio__futimes (fd, times)
910
911static int
549static int aio_futimes (int fd, const struct timeval tv[2]) 912eio__futimes (int fd, const struct timeval tv[2])
550{ 913{
551 errno = ENOSYS; 914 errno = ENOSYS;
552 return -1; 915 return -1;
553} 916}
554 917
555#endif 918#endif
556 919
557#if !HAVE_FDATASYNC 920#if !HAVE_FDATASYNC
921# undef fdatasync
558# define fdatasync fsync 922# define fdatasync(fd) fsync (fd)
559#endif 923#endif
924
925/* sync_file_range always needs emulation */
926static int
927eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
928{
929#if HAVE_SYNC_FILE_RANGE
930 int res;
931
932 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
933 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
934 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
935 {
936 flags = 0
937 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
938 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
939 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
940 }
941
942 res = sync_file_range (fd, offset, nbytes, flags);
943
944 if (!res || errno != ENOSYS)
945 return res;
946#endif
947
948 /* even though we could play tricks with the flags, it's better to always
949 * call fdatasync, as that matches the expectation of its users best */
950 return fdatasync (fd);
951}
560 952
561#if !HAVE_READAHEAD 953#if !HAVE_READAHEAD
954# undef readahead
562# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 955# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
563 956
957static ssize_t
564static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 958eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
565{ 959{
566 size_t todo = count; 960 size_t todo = count;
567 dBUF; 961 dBUF;
568 962
569 while (todo > 0) 963 while (todo > 0)
… …
579 return count; 973 return count;
580} 974}
581 975
582#endif 976#endif
583 977
584#if !HAVE_READDIR_R
585# define readdir_r aio_readdir_r
586
587static mutex_t readdirlock = X_MUTEX_INIT;
588
589static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
590{
591 X_DIRENT *e;
592 int errorno;
593
594 X_LOCK (readdirlock);
595
596 e = readdir (dirp);
597 errorno = errno;
598
599 if (e)
600 {
601 *res = ent;
602 strcpy (ent->d_name, e->d_name);
603 }
604 else
605 *res = 0;
606
607 X_UNLOCK (readdirlock);
608
609 errno = errorno;
610 return e ? 0 : -1;
611}
612#endif
613
614/* sendfile always needs emulation */ 978/* sendfile always needs emulation */
979static ssize_t
615static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 980eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
616{ 981{
982 ssize_t written = 0;
617 ssize_t res; 983 ssize_t res;
618 984
619 if (!count) 985 if (!count)
620 return 0; 986 return 0;
621 987
988 for (;;)
989 {
622#if HAVE_SENDFILE 990#if HAVE_SENDFILE
623# if __linux 991# if __linux
992 off_t soffset = offset;
624 res = sendfile (ofd, ifd, &offset, count); 993 res = sendfile (ofd, ifd, &soffset, count);
625 994
626# elif __freebsd 995# elif __FreeBSD__
627 /* 996 /*
628 * Of course, the freebsd sendfile is a dire hack with no thoughts 997 * Of course, the freebsd sendfile is a dire hack with no thoughts
629 * wasted on making it similar to other I/O functions. 998 * wasted on making it similar to other I/O functions.
630 */ 999 */
631 {
632 off_t sbytes; 1000 off_t sbytes;
633 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1001 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
634 1002
635 if (res < 0 && sbytes) 1003 #if 0 /* according to the manpage, this is correct, but broken behaviour */
636 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1004 /* freebsd' sendfile will return 0 on success */
1005 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1006 /* not on e.g. EIO or EPIPE - sounds broken */
1007 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
637 res = sbytes; 1008 res = sbytes;
638 } 1009 #endif
1010
1011 /* according to source inspection, this is correct, and useful behaviour */
1012 if (sbytes)
1013 res = sbytes;
1014
1015# elif defined (__APPLE__) && 0 /* broken, as everything on os x */
1016 off_t sbytes = count;
1017 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1018
1019 /* according to the manpage, sbytes is always valid */
1020 if (sbytes)
1021 res = sbytes;
639 1022
640# elif __hpux 1023# elif __hpux
641 res = sendfile (ofd, ifd, offset, count, 0, 0); 1024 res = sendfile (ofd, ifd, offset, count, 0, 0);
642 1025
643# elif __solaris 1026# elif __solaris
644 {
645 struct sendfilevec vec; 1027 struct sendfilevec vec;
646 size_t sbytes; 1028 size_t sbytes;
647 1029
648 vec.sfv_fd = ifd; 1030 vec.sfv_fd = ifd;
649 vec.sfv_flag = 0; 1031 vec.sfv_flag = 0;
650 vec.sfv_off = offset; 1032 vec.sfv_off = offset;
651 vec.sfv_len = count; 1033 vec.sfv_len = count;
652 1034
653 res = sendfilev (ofd, &vec, 1, &sbytes); 1035 res = sendfilev (ofd, &vec, 1, &sbytes);
654 1036
655 if (res < 0 && sbytes) 1037 if (res < 0 && sbytes)
656 res = sbytes; 1038 res = sbytes;
657 }
658 1039
659# endif 1040# endif
1041
1042#elif defined (_WIN32)
1043 /* does not work, just for documentation of what would need to be done */
1044 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1045 /* libeio guarantees that the file offset does not change, and windows */
1046 /* has no way to get an independent handle to the same file description */
1047 HANDLE h = TO_SOCKET (ifd);
1048 SetFilePointer (h, offset, 0, FILE_BEGIN);
1049 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1050
660#else 1051#else
661 res = -1; 1052 res = -1;
662 errno = ENOSYS; 1053 errno = ENOSYS;
663#endif 1054#endif
664 1055
1056 /* we assume sendfile can copy at least 128mb in one go */
1057 if (res <= 128 * 1024 * 1024)
1058 {
1059 if (res > 0)
1060 written += res;
1061
1062 if (written)
1063 return written;
1064
1065 break;
1066 }
1067 else
1068 {
1069 /* if we requested more, then probably the kernel was lazy */
1070 written += res;
1071 offset += res;
1072 count -= res;
1073
1074 if (!count)
1075 return written;
1076 }
1077 }
1078
665 if (res < 0 1079 if (res < 0
666 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1080 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1081 /* BSDs */
1082#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1083 || errno == ENOTSUP
1084#endif
1085 || errno == EOPNOTSUPP /* BSDs */
667#if __solaris 1086#if __solaris
668 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1087 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
669#endif 1088#endif
670 ) 1089 )
671 ) 1090 )
… …
702 } 1121 }
703 1122
704 return res; 1123 return res;
705} 1124}
706 1125
1126static signed char
1127eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1128{
1129 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1130 : a->inode < b->inode ? -1
1131 : a->inode > b->inode ? 1
1132 : 0;
1133}
1134
1135#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1136
1137#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1138#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1139
1140static void
1141eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1142{
1143 unsigned char bits [9 + sizeof (ino_t) * 8];
1144 unsigned char *bit = bits;
1145
1146 assert (CHAR_BIT == 8);
1147 assert (sizeof (eio_dirent) * 8 < 256);
1148 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1149 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1150
1151 if (size <= EIO_SORT_FAST)
1152 return;
1153
1154 /* first prepare an array of bits to test in our radix sort */
1155 /* try to take endianness into account, as well as differences in ino_t sizes */
1156 /* inode_bits must contain all inodes ORed together */
1157 /* which is used to skip bits that are 0 everywhere, which is very common */
1158 {
1159 ino_t endianness;
1160 int i, j;
1161
1162 /* we store the byte offset of byte n into byte n of "endianness" */
1163 for (i = 0; i < sizeof (ino_t); ++i)
1164 ((unsigned char *)&endianness)[i] = i;
1165
1166 *bit++ = 0;
1167
1168 for (i = 0; i < sizeof (ino_t); ++i)
1169 {
1170 /* shifting off the byte offsets out of "endianness" */
1171 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1172 endianness >>= 8;
1173
1174 for (j = 0; j < 8; ++j)
1175 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1176 *bit++ = offs + j;
1177 }
1178
1179 for (j = 0; j < 8; ++j)
1180 if (score_bits & (1 << j))
1181 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1182 }
1183
1184 /* now actually do the sorting (a variant of MSD radix sort) */
1185 {
1186 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1187 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1188 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1189 int stk_idx = 0;
1190
1191 base_stk [stk_idx] = dents;
1192 end_stk [stk_idx] = dents + size;
1193 bit_stk [stk_idx] = bit - 1;
1194
1195 do
1196 {
1197 base = base_stk [stk_idx];
1198 end = end_stk [stk_idx];
1199 bit = bit_stk [stk_idx];
1200
1201 for (;;)
1202 {
1203 unsigned char O = *bit >> 3;
1204 unsigned char M = 1 << (*bit & 7);
1205
1206 eio_dirent *a = base;
1207 eio_dirent *b = end;
1208
1209 if (b - a < EIO_SORT_CUTOFF)
1210 break;
1211
1212 /* now bit-partition the array on the bit */
1213 /* this ugly asymmetric loop seems to perform much better than typical */
1214 /* partition algos found in the literature */
1215 do
1216 if (!(((unsigned char *)a)[O] & M))
1217 ++a;
1218 else if (!(((unsigned char *)--b)[O] & M))
1219 {
1220 eio_dirent tmp = *a; *a = *b; *b = tmp;
1221 ++a;
1222 }
1223 while (b > a);
1224
1225 /* next bit, or stop, if no bits left in this path */
1226 if (!*--bit)
1227 break;
1228
1229 base_stk [stk_idx] = a;
1230 end_stk [stk_idx] = end;
1231 bit_stk [stk_idx] = bit;
1232 ++stk_idx;
1233
1234 end = a;
1235 }
1236 }
1237 while (stk_idx--);
1238 }
1239}
1240
1241static void
1242eio_dent_insertion_sort (eio_dirent *dents, int size)
1243{
1244 /* first move the smallest element to the front, to act as a sentinel */
1245 {
1246 int i;
1247 eio_dirent *min = dents;
1248
1249 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1250 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1251 if (EIO_DENT_CMP (dents [i], <, *min))
1252 min = &dents [i];
1253
1254 /* swap elements 0 and j (minimum) */
1255 {
1256 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1257 }
1258 }
1259
1260 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1261 {
1262 eio_dirent *i, *j;
1263
1264 for (i = dents + 1; i < dents + size; ++i)
1265 {
1266 eio_dirent value = *i;
1267
1268 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1269 j [1] = j [0];
1270
1271 j [1] = value;
1272 }
1273 }
1274}
1275
1276static void
1277eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1278{
1279 if (size <= 1)
1280 return; /* our insertion sort relies on size > 0 */
1281
1282 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1283 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1284 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1285
1286 /* use an insertion sort at the end, or for small arrays, */
1287 /* as insertion sort is more efficient for small partitions */
1288 eio_dent_insertion_sort (dents, size);
1289}
1290
707/* read a full directory */ 1291/* read a full directory */
1292static void
708static void scandir_ (eio_req *req, worker *self) 1293eio__scandir (eio_req *req, etp_worker *self)
709{ 1294{
710 DIR *dirp; 1295 DIR *dirp;
711 union
712 {
713 EIO_STRUCT_DIRENT d;
714 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
715 } *u;
716 EIO_STRUCT_DIRENT *entp; 1296 EIO_STRUCT_DIRENT *entp;
717 char *name, *names; 1297 char *name, *names;
718 int memlen = 4096; 1298 int namesalloc = 4096;
719 int memofs = 0; 1299 int namesoffs = 0;
1300 int flags = req->int1;
1301 eio_dirent *dents = 0;
1302 int dentalloc = 128;
720 int res = 0; 1303 int dentoffs = 0;
1304 ino_t inode_bits = 0;
1305
1306 req->result = -1;
1307
1308 if (!(flags & EIO_READDIR_DENTS))
1309 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
721 1310
722 X_LOCK (wrklock); 1311 X_LOCK (wrklock);
1312 /* the corresponding closedir is in ETP_WORKER_CLEAR */
723 self->dirp = dirp = opendir (req->ptr1); 1313 self->dirp = dirp = opendir (req->ptr1);
724 self->dbuf = u = malloc (sizeof (*u)); 1314
725 req->flags |= EIO_FLAG_PTR2_FREE; 1315 if (req->flags & EIO_FLAG_PTR1_FREE)
1316 free (req->ptr1);
1317
1318 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1319 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
726 req->ptr2 = names = malloc (memlen); 1320 req->ptr2 = names = malloc (namesalloc);
727 X_UNLOCK (wrklock); 1321 X_UNLOCK (wrklock);
728 1322
729 if (dirp && u && names) 1323 if (dirp && names && (!flags || dents))
730 for (;;) 1324 for (;;)
731 { 1325 {
732 errno = 0; 1326 errno = 0;
733 readdir_r (dirp, &u->d, &entp); 1327 entp = readdir (dirp);
734 1328
735 if (!entp) 1329 if (!entp)
1330 {
1331 if (errno)
1332 break;
1333
1334 /* sort etc. */
1335 req->int1 = flags;
1336 req->result = dentoffs;
1337
1338 if (flags & EIO_READDIR_STAT_ORDER)
1339 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1340 else if (flags & EIO_READDIR_DIRS_FIRST)
1341 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1342 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1343 else
1344 {
1345 /* in this case, all is known, and we just put dirs first and sort them */
1346 eio_dirent *oth = dents + dentoffs;
1347 eio_dirent *dir = dents;
1348
1349 /* now partition dirs to the front, and non-dirs to the back */
1350 /* by walking from both sides and swapping if necessary */
1351 while (oth > dir)
1352 {
1353 if (dir->type == EIO_DT_DIR)
1354 ++dir;
1355 else if ((--oth)->type == EIO_DT_DIR)
1356 {
1357 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1358
1359 ++dir;
1360 }
1361 }
1362
1363 /* now sort the dirs only (dirs all have the same score) */
1364 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1365 }
1366
736 break; 1367 break;
1368 }
737 1369
1370 /* now add the entry to our list(s) */
738 name = entp->d_name; 1371 name = entp->d_name;
739 1372
1373 /* skip . and .. entries */
740 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1374 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
741 { 1375 {
742 int len = strlen (name) + 1; 1376 int len = D_NAMLEN (entp) + 1;
743 1377
744 res++; 1378 while (ecb_expect_false (namesoffs + len > namesalloc))
745
746 while (memofs + len > memlen)
747 { 1379 {
748 memlen *= 2; 1380 namesalloc *= 2;
749 X_LOCK (wrklock); 1381 X_LOCK (wrklock);
750 req->ptr2 = names = realloc (names, memlen); 1382 req->ptr2 = names = realloc (names, namesalloc);
751 X_UNLOCK (wrklock); 1383 X_UNLOCK (wrklock);
752 1384
753 if (!names) 1385 if (!names)
754 break; 1386 break;
755 } 1387 }
756 1388
757 memcpy (names + memofs, name, len); 1389 memcpy (names + namesoffs, name, len);
1390
1391 if (dents)
1392 {
1393 struct eio_dirent *ent;
1394
1395 if (ecb_expect_false (dentoffs == dentalloc))
1396 {
1397 dentalloc *= 2;
1398 X_LOCK (wrklock);
1399 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1400 X_UNLOCK (wrklock);
1401
1402 if (!dents)
1403 break;
1404 }
1405
1406 ent = dents + dentoffs;
1407
1408 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1409 ent->namelen = len - 1;
1410 ent->inode = D_INO (entp);
1411
1412 inode_bits |= ent->inode;
1413
1414 switch (D_TYPE (entp))
1415 {
1416 default:
1417 ent->type = EIO_DT_UNKNOWN;
1418 flags |= EIO_READDIR_FOUND_UNKNOWN;
1419 break;
1420
1421 #ifdef DT_FIFO
1422 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1423 #endif
1424 #ifdef DT_CHR
1425 case DT_CHR: ent->type = EIO_DT_CHR; break;
1426 #endif
1427 #ifdef DT_MPC
1428 case DT_MPC: ent->type = EIO_DT_MPC; break;
1429 #endif
1430 #ifdef DT_DIR
1431 case DT_DIR: ent->type = EIO_DT_DIR; break;
1432 #endif
1433 #ifdef DT_NAM
1434 case DT_NAM: ent->type = EIO_DT_NAM; break;
1435 #endif
1436 #ifdef DT_BLK
1437 case DT_BLK: ent->type = EIO_DT_BLK; break;
1438 #endif
1439 #ifdef DT_MPB
1440 case DT_MPB: ent->type = EIO_DT_MPB; break;
1441 #endif
1442 #ifdef DT_REG
1443 case DT_REG: ent->type = EIO_DT_REG; break;
1444 #endif
1445 #ifdef DT_NWK
1446 case DT_NWK: ent->type = EIO_DT_NWK; break;
1447 #endif
1448 #ifdef DT_CMP
1449 case DT_CMP: ent->type = EIO_DT_CMP; break;
1450 #endif
1451 #ifdef DT_LNK
1452 case DT_LNK: ent->type = EIO_DT_LNK; break;
1453 #endif
1454 #ifdef DT_SOCK
1455 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1456 #endif
1457 #ifdef DT_DOOR
1458 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1459 #endif
1460 #ifdef DT_WHT
1461 case DT_WHT: ent->type = EIO_DT_WHT; break;
1462 #endif
1463 }
1464
1465 ent->score = 7;
1466
1467 if (flags & EIO_READDIR_DIRS_FIRST)
1468 {
1469 if (ent->type == EIO_DT_UNKNOWN)
1470 {
1471 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1472 ent->score = 1;
1473 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1474 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1475 }
1476 else if (ent->type == EIO_DT_DIR)
1477 ent->score = 0;
1478 }
1479 }
1480
758 memofs += len; 1481 namesoffs += len;
1482 ++dentoffs;
1483 }
1484
1485 if (EIO_CANCELLED (req))
1486 {
1487 errno = ECANCELED;
1488 break;
759 } 1489 }
760 } 1490 }
1491}
761 1492
762 if (errno) 1493#ifdef PAGESIZE
763 res = -1; 1494# define eio_pagesize() PAGESIZE
1495#else
1496static intptr_t
1497eio_pagesize (void)
1498{
1499 static intptr_t page;
1500
1501 if (!page)
1502 page = sysconf (_SC_PAGESIZE);
1503
1504 return page;
1505}
1506#endif
1507
1508static void
1509eio_page_align (void **addr, size_t *length)
1510{
1511 intptr_t mask = eio_pagesize () - 1;
1512
1513 /* round down addr */
1514 intptr_t adj = mask & (intptr_t)*addr;
1515
1516 *addr = (void *)((intptr_t)*addr - adj);
1517 *length += adj;
1518
1519 /* round up length */
1520 *length = (*length + mask) & ~mask;
1521}
1522
1523#if !_POSIX_MEMLOCK
1524# define eio__mlockall(a) ((errno = ENOSYS), -1)
1525#else
1526
1527static int
1528eio__mlockall (int flags)
1529{
1530 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1531 extern int mallopt (int, int);
1532 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1533 #endif
1534
1535 if (EIO_MCL_CURRENT != MCL_CURRENT
1536 || EIO_MCL_FUTURE != MCL_FUTURE)
1537 {
1538 flags = 0
1539 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1540 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1541 }
1542
1543 return mlockall (flags);
1544}
1545#endif
1546
1547#if !_POSIX_MEMLOCK_RANGE
1548# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1549#else
1550
1551static int
1552eio__mlock (void *addr, size_t length)
1553{
1554 eio_page_align (&addr, &length);
1555
1556 return mlock (addr, length);
1557}
1558
1559#endif
1560
1561#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1562# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1563#else
1564
1565static int
1566eio__msync (void *mem, size_t len, int flags)
1567{
1568 eio_page_align (&mem, &len);
1569
1570 if (EIO_MS_ASYNC != MS_SYNC
1571 || EIO_MS_INVALIDATE != MS_INVALIDATE
1572 || EIO_MS_SYNC != MS_SYNC)
1573 {
1574 flags = 0
1575 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1576 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1577 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1578 }
1579
1580 return msync (mem, len, flags);
1581}
1582
1583#endif
1584
1585static int
1586eio__mtouch (eio_req *req)
1587{
1588 void *mem = req->ptr2;
1589 size_t len = req->size;
1590 int flags = req->int1;
1591
1592 eio_page_align (&mem, &len);
1593
764 1594 {
765 req->result = res; 1595 intptr_t addr = (intptr_t)mem;
1596 intptr_t end = addr + len;
1597 intptr_t page = eio_pagesize ();
1598
1599 if (addr < end)
1600 if (flags & EIO_MT_MODIFY) /* modify */
1601 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1602 else
1603 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1604 }
1605
1606 return 0;
766} 1607}
767 1608
768/*****************************************************************************/ 1609/*****************************************************************************/
769 1610
1611#define ALLOC(len) \
1612 if (!req->ptr2) \
1613 { \
1614 X_LOCK (wrklock); \
1615 req->flags |= EIO_FLAG_PTR2_FREE; \
1616 X_UNLOCK (wrklock); \
1617 req->ptr2 = malloc (len); \
1618 if (!req->ptr2) \
1619 { \
1620 errno = ENOMEM; \
1621 req->result = -1; \
1622 break; \
1623 } \
1624 }
1625
770X_THREAD_PROC (eio_proc) 1626X_THREAD_PROC (etp_proc)
771{ 1627{
772 eio_req *req; 1628 ETP_REQ *req;
773 struct timespec ts; 1629 struct timespec ts;
774 worker *self = (worker *)thr_arg; 1630 etp_worker *self = (etp_worker *)thr_arg;
775 1631
776 /* try to distribute timeouts somewhat randomly */ 1632 /* try to distribute timeouts somewhat randomly */
777 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1633 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
778 1634
779 for (;;) 1635 for (;;)
780 { 1636 {
781 ts.tv_sec = time (0) + IDLE_TIMEOUT;
782
783 X_LOCK (reqlock); 1637 X_LOCK (reqlock);
784 1638
785 for (;;) 1639 for (;;)
786 { 1640 {
787 self->req = req = reqq_shift (&req_queue); 1641 self->req = req = reqq_shift (&req_queue);
… …
789 if (req) 1643 if (req)
790 break; 1644 break;
791 1645
792 ++idle; 1646 ++idle;
793 1647
1648 ts.tv_sec = time (0) + idle_timeout;
794 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1649 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
795 { 1650 {
796 if (idle > max_idle) 1651 if (idle > max_idle)
797 { 1652 {
798 --idle; 1653 --idle;
… …
803 goto quit; 1658 goto quit;
804 } 1659 }
805 1660
806 /* we are allowed to idle, so do so without any timeout */ 1661 /* we are allowed to idle, so do so without any timeout */
807 X_COND_WAIT (reqwait, reqlock); 1662 X_COND_WAIT (reqwait, reqlock);
808 ts.tv_sec = time (0) + IDLE_TIMEOUT;
809 } 1663 }
810 1664
811 --idle; 1665 --idle;
812 } 1666 }
813 1667
814 --nready; 1668 --nready;
815 1669
816 X_UNLOCK (reqlock); 1670 X_UNLOCK (reqlock);
817 1671
818 errno = 0; /* strictly unnecessary */ 1672 if (req->type < 0)
1673 goto quit;
819 1674
820 if (!EIO_CANCELLED (req)) 1675 if (!EIO_CANCELLED (req))
821 switch (req->type) 1676 ETP_EXECUTE (self, req);
822 {
823 case EIO_READ: req->result = req->offs >= 0
824 ? pread (req->int1, req->ptr2, req->size, req->offs)
825 : read (req->int1, req->ptr2, req->size); break;
826 case EIO_WRITE: req->result = req->offs >= 0
827 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
828 : write (req->int1, req->ptr2, req->size); break;
829
830 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
831 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
832
833 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
834 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
835 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
836
837 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
838 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
839 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
840 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
841 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
842 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
843
844 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
845 case EIO_CLOSE: req->result = close (req->int1); break;
846 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
847 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
848 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
849 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
850 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
851 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
852 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
853 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
854 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
855
856 case EIO_SYNC: req->result = 0; sync (); break;
857 case EIO_FSYNC: req->result = fsync (req->int1); break;
858 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
859
860 case EIO_READDIR: scandir_ (req, self); break;
861
862 case EIO_BUSY:
863#ifdef _WIN32
864 Sleep (req->nv1 * 1000.);
865#else
866 {
867 struct timeval tv;
868
869 tv.tv_sec = req->nv1;
870 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
871
872 req->result = select (0, 0, 0, 0, &tv);
873 }
874#endif
875 break;
876
877 case EIO_UTIME:
878 case EIO_FUTIME:
879 {
880 struct timeval tv[2];
881 struct timeval *times;
882
883 if (req->nv1 != -1. || req->nv2 != -1.)
884 {
885 tv[0].tv_sec = req->nv1;
886 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
887 tv[1].tv_sec = req->nv2;
888 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
889
890 times = tv;
891 }
892 else
893 times = 0;
894
895
896 req->result = req->type == EIO_FUTIME
897 ? futimes (req->int1, times)
898 : utimes (req->ptr1, times);
899 }
900
901 case EIO_GROUP:
902 case EIO_NOP:
903 break;
904
905 case EIO_QUIT:
906 goto quit;
907
908 default:
909 req->result = -1;
910 break;
911 }
912
913 req->errorno = errno;
914 1677
915 X_LOCK (reslock); 1678 X_LOCK (reslock);
916 1679
917 ++npending; 1680 ++npending;
918 1681
919 if (!reqq_push (&res_queue, req) && want_poll_cb) 1682 if (!reqq_push (&res_queue, req) && want_poll_cb)
920 want_poll_cb (); 1683 want_poll_cb ();
921 1684
922 self->req = 0; 1685 self->req = 0;
923 worker_clear (self); 1686 etp_worker_clear (self);
924 1687
925 X_UNLOCK (reslock); 1688 X_UNLOCK (reslock);
926 } 1689 }
927 1690
928quit: 1691quit:
929 X_LOCK (wrklock); 1692 X_LOCK (wrklock);
930 worker_free (self); 1693 etp_worker_free (self);
931 X_UNLOCK (wrklock); 1694 X_UNLOCK (wrklock);
932 1695
933 return 0; 1696 return 0;
934} 1697}
935 1698
936/*****************************************************************************/ 1699/*****************************************************************************/
937 1700
938static void eio_atfork_prepare (void) 1701int ecb_cold
1702eio_init (void (*want_poll)(void), void (*done_poll)(void))
939{ 1703{
940 X_LOCK (wrklock); 1704 return etp_init (want_poll, done_poll);
941 X_LOCK (reqlock);
942 X_LOCK (reslock);
943#if !HAVE_PREADWRITE
944 X_LOCK (preadwritelock);
945#endif
946#if !HAVE_READDIR_R
947 X_LOCK (readdirlock);
948#endif
949} 1705}
950 1706
951static void eio_atfork_parent (void) 1707ecb_inline void
1708eio_api_destroy (eio_req *req)
952{ 1709{
953#if !HAVE_READDIR_R 1710 free (req);
954 X_UNLOCK (readdirlock);
955#endif
956#if !HAVE_PREADWRITE
957 X_UNLOCK (preadwritelock);
958#endif
959 X_UNLOCK (reslock);
960 X_UNLOCK (reqlock);
961 X_UNLOCK (wrklock);
962} 1711}
963 1712
964static void eio_atfork_child (void) 1713#define REQ(rtype) \
965{ 1714 eio_req *req; \
966 eio_req *prv; 1715 \
1716 req = (eio_req *)calloc (1, sizeof *req); \
1717 if (!req) \
1718 return 0; \
1719 \
1720 req->type = rtype; \
1721 req->pri = pri; \
1722 req->finish = cb; \
1723 req->data = data; \
1724 req->destroy = eio_api_destroy;
967 1725
968 while (prv = reqq_shift (&req_queue)) 1726#define SEND eio_submit (req); return req
969 eio_destroy (prv);
970 1727
971 while (prv = reqq_shift (&res_queue)) 1728#define PATH \
972 eio_destroy (prv); 1729 req->flags |= EIO_FLAG_PTR1_FREE; \
1730 req->ptr1 = strdup (path); \
1731 if (!req->ptr1) \
1732 { \
1733 eio_api_destroy (req); \
1734 return 0; \
1735 }
973 1736
974 while (wrk_first.next != &wrk_first) 1737static void
1738eio_execute (etp_worker *self, eio_req *req)
1739{
1740 switch (req->type)
975 { 1741 {
976 worker *wrk = wrk_first.next; 1742 case EIO_READ: ALLOC (req->size);
1743 req->result = req->offs >= 0
1744 ? pread (req->int1, req->ptr2, req->size, req->offs)
1745 : read (req->int1, req->ptr2, req->size); break;
1746 case EIO_WRITE: req->result = req->offs >= 0
1747 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1748 : write (req->int1, req->ptr2, req->size); break;
977 1749
978 if (wrk->req) 1750 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
979 eio_destroy (wrk->req); 1751 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
980 1752
981 worker_clear (wrk); 1753 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
982 worker_free (wrk); 1754 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
983 } 1755 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1756 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1757 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1758 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
984 1759
985 started = 0; 1760 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
986 idle = 0; 1761 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
987 nreqs = 0; 1762 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
988 nready = 0; 1763 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
989 npending = 0;
990 1764
991 eio_atfork_parent (); 1765 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
992} 1766 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1767 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1768 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1769 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1770 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
993 1771
994int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1772 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
995{ 1773 case EIO_CLOSE: req->result = close (req->int1); break;
996 want_poll_cb = want_poll; 1774 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
997 done_poll_cb = done_poll; 1775 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1776 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1777 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1778 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1779 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1780 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1781 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
998 1782
1783 case EIO_REALPATH: req->flags |= EIO_FLAG_PTR2_FREE;
1784 req->ptr2 = realpath (req->ptr1, 0);
1785 req->result = req->ptr2 ? strlen (req->ptr2) : -1;
1786 break;
1787
1788 case EIO_READLINK: ALLOC (PATH_MAX);
1789 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1790
1791 case EIO_SYNC: req->result = 0; sync (); break;
1792 case EIO_FSYNC: req->result = fsync (req->int1); break;
1793 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1794 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1795 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1796 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1797 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1798 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1799
1800 case EIO_READDIR: eio__scandir (req, self); break;
1801
1802 case EIO_BUSY:
999#ifdef _WIN32 1803#ifdef _WIN32
1000 X_MUTEX_CHECK (wrklock); 1804 Sleep (req->nv1 * 1e3);
1001 X_MUTEX_CHECK (reslock); 1805#else
1002 X_MUTEX_CHECK (reqlock); 1806 {
1003 X_MUTEX_CHECK (reqwait); 1807 struct timeval tv;
1004 X_MUTEX_CHECK (preadwritelock);
1005 X_MUTEX_CHECK (readdirlock);
1006 1808
1007 X_COND_CHECK (reqwait); 1809 tv.tv_sec = req->nv1;
1008#endif 1810 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1009 1811
1010 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child); 1812 req->result = select (0, 0, 0, 0, &tv);
1011} 1813 }
1814#endif
1815 break;
1012 1816
1013#if 0 1817 case EIO_UTIME:
1818 case EIO_FUTIME:
1819 {
1820 struct timeval tv[2];
1821 struct timeval *times;
1014 1822
1823 if (req->nv1 != -1. || req->nv2 != -1.)
1824 {
1825 tv[0].tv_sec = req->nv1;
1826 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1827 tv[1].tv_sec = req->nv2;
1828 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1829
1830 times = tv;
1831 }
1832 else
1833 times = 0;
1834
1835 req->result = req->type == EIO_FUTIME
1836 ? futimes (req->int1, times)
1837 : utimes (req->ptr1, times);
1838 }
1839 break;
1840
1841 case EIO_GROUP:
1842 abort (); /* handled in eio_request */
1843
1844 case EIO_NOP:
1845 req->result = 0;
1846 break;
1847
1848 case EIO_CUSTOM:
1849 req->feed (req);
1850 break;
1851
1852 default:
1853 errno = ENOSYS;
1854 req->result = -1;
1855 break;
1856 }
1857
1858 req->errorno = errno;
1859}
1860
1861#ifndef EIO_NO_WRAPPERS
1862
1863eio_req *eio_nop (int pri, eio_cb cb, void *data)
1864{
1865 REQ (EIO_NOP); SEND;
1866}
1867
1868eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1869{
1870 REQ (EIO_BUSY); req->nv1 = delay; SEND;
1871}
1872
1015eio_req *eio_fsync (int fd, eio_cb cb); 1873eio_req *eio_sync (int pri, eio_cb cb, void *data)
1874{
1875 REQ (EIO_SYNC); SEND;
1876}
1877
1878eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1879{
1880 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1881}
1882
1883eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1884{
1885 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1886}
1887
1888eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1889{
1890 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1891}
1892
1893eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1894{
1895 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1896}
1897
1898eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1899{
1900 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1901}
1902
1903eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1904{
1905 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1906}
1907
1016eio_req *eio_fdatasync (int fd, eio_cb cb); 1908eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1017eio_req *eio_dupclose (int fd, eio_cb cb); 1909{
1910 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1911}
1912
1913eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1914{
1915 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1916}
1917
1018eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb); 1918eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1019eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb); 1919{
1020eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb); 1920 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1021eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 1921}
1922
1923eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1924{
1925 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1926}
1927
1928eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1929{
1930 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1931}
1932
1933eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1934{
1935 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1936}
1937
1938eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1939{
1940 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1941}
1942
1022eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb); 1943eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1944{
1945 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1946}
1947
1023eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb); 1948eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1949{
1950 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1951}
1952
1024eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb); 1953eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1954{
1955 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1956}
1957
1025eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb); 1958eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1959{
1960 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1961}
1962
1026eio_req *eio_dup2 (int fd, int fd2, eio_cb cb); 1963eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1964{
1965 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1966}
1967
1027eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb); 1968eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1969{
1970 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1971}
1972
1028eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb); 1973eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1029eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 1974{
1030eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 1975 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1031eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 1976}
1977
1032eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb); 1978eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1979{
1980 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1981}
1982
1033eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb); 1983eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1034eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb); 1984{
1035eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb); 1985 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1986}
1987
1036eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb); 1988eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1989{
1990 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1991}
1992
1993eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1994{
1995 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1996}
1997
1998eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1999{
2000 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2001}
2002
2003static eio_req *
2004eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2005{
2006 REQ (type); PATH; SEND;
2007}
2008
2009eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2010{
2011 return eio__1path (EIO_READLINK, path, pri, cb, data);
2012}
2013
2014eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2015{
2016 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2017}
2018
2019eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2020{
2021 return eio__1path (EIO_STAT, path, pri, cb, data);
2022}
2023
2024eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2025{
2026 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2027}
2028
2029eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2030{
2031 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2032}
2033
1037eio_req *eio_unlink (const char *path, eio_cb cb); 2034eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2035{
2036 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2037}
2038
1038eio_req *eio_rmdir (const char *path, eio_cb cb); 2039eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1039eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2040{
2041 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2042}
2043
2044eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2045{
2046 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2047}
2048
1040eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb); 2049eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1041eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */ 2050{
1042eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */ 2051 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2052}
2053
2054static eio_req *
2055eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2056{
2057 REQ (type); PATH;
2058
2059 req->flags |= EIO_FLAG_PTR2_FREE;
2060 req->ptr2 = strdup (new_path);
2061 if (!req->ptr2)
2062 {
2063 eio_api_destroy (req);
2064 return 0;
2065 }
2066
2067 SEND;
2068}
2069
2070eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2071{
2072 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2073}
2074
2075eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2076{
2077 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2078}
2079
2080eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2081{
2082 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2083}
2084
2085eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2086{
2087 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2088}
2089
2090#endif
2091
2092eio_req *eio_grp (eio_cb cb, void *data)
2093{
2094 const int pri = EIO_PRI_MAX;
2095
2096 REQ (EIO_GROUP); SEND;
2097}
2098
2099#undef REQ
2100#undef PATH
2101#undef SEND
2102
2103/*****************************************************************************/
2104/* grp functions */
2105
1043void 2106void
1044aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1045 PROTOTYPE: $$$;$
1046 PPCODE:
1047{
1048 dREQ;
1049
1050 req->type = EIO_OPEN;
1051 req->sv1 = newSVsv (pathname);
1052 req->ptr1 = SvPVbyte_nolen (req->sv1);
1053 req->int1 = flags;
1054 req->int2 = mode;
1055
1056 EIO_SEND;
1057}
1058
1059void
1060aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1061 PROTOTYPE: $;$
1062 ALIAS:
1063 aio_fsync = EIO_FSYNC
1064 aio_fdatasync = EIO_FDATASYNC
1065 PPCODE:
1066{
1067 dREQ;
1068
1069 req->type = ix;
1070 req->sv1 = newSVsv (fh);
1071 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1072
1073 EIO_SEND (req);
1074}
1075
1076void
1077aio_close (SV *fh, SV *callback=&PL_sv_undef)
1078 PROTOTYPE: $;$
1079 PPCODE:
1080{
1081 dREQ;
1082
1083 req->type = EIO_CLOSE;
1084 req->sv1 = newSVsv (fh);
1085 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1086
1087 EIO_SEND (req);
1088}
1089
1090void
1091aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1092 ALIAS:
1093 aio_read = EIO_READ
1094 aio_write = EIO_WRITE
1095 PROTOTYPE: $$$$$;$
1096 PPCODE:
1097{
1098 STRLEN svlen;
1099 char *svptr = SvPVbyte (data, svlen);
1100 UV len = SvUV (length);
1101
1102 SvUPGRADE (data, SVt_PV);
1103 SvPOK_on (data);
1104
1105 if (dataoffset < 0)
1106 dataoffset += svlen;
1107
1108 if (dataoffset < 0 || dataoffset > svlen)
1109 croak ("dataoffset outside of data scalar");
1110
1111 if (ix == EIO_WRITE)
1112 {
1113 /* write: check length and adjust. */
1114 if (!SvOK (length) || len + dataoffset > svlen)
1115 len = svlen - dataoffset;
1116 }
1117 else
1118 {
1119 /* read: grow scalar as necessary */
1120 svptr = SvGROW (data, len + dataoffset + 1);
1121 }
1122
1123 if (len < 0)
1124 croak ("length must not be negative");
1125
1126 {
1127 dREQ;
1128
1129 req->type = ix;
1130 req->sv1 = newSVsv (fh);
1131 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1132 : IoOFP (sv_2io (fh)));
1133 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1134 req->size = len;
1135 req->sv2 = SvREFCNT_inc (data);
1136 req->ptr2 = (char *)svptr + dataoffset;
1137 req->stroffset = dataoffset;
1138
1139 if (!SvREADONLY (data))
1140 {
1141 SvREADONLY_on (data);
1142 req->flags |= FLAG_SV2_RO_OFF;
1143 }
1144
1145 EIO_SEND;
1146 }
1147}
1148
1149void
1150aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1151 PROTOTYPE: $$;$
1152 PPCODE:
1153{
1154 SV *data;
1155 dREQ;
1156
1157 data = newSV (NAME_MAX);
1158 SvPOK_on (data);
1159
1160 req->type = EIO_READLINK;
1161 req->sv1 = newSVsv (path);
1162 req->ptr1 = SvPVbyte_nolen (req->sv1);
1163 req->sv2 = data;
1164 req->ptr2 = SvPVbyte_nolen (data);
1165
1166 EIO_SEND;
1167}
1168
1169void
1170aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1171 PROTOTYPE: $$$$;$
1172 PPCODE:
1173{
1174 dREQ;
1175
1176 req->type = EIO_SENDFILE;
1177 req->sv1 = newSVsv (out_fh);
1178 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1179 req->sv2 = newSVsv (in_fh);
1180 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1181 req->offs = SvVAL64 (in_offset);
1182 req->size = length;
1183
1184 EIO_SEND;
1185}
1186
1187void
1188aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1189 PROTOTYPE: $$$;$
1190 PPCODE:
1191{
1192 dREQ;
1193
1194 req->type = EIO_READAHEAD;
1195 req->sv1 = newSVsv (fh);
1196 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1197 req->offs = SvVAL64 (offset);
1198 req->size = length;
1199
1200 EIO_SEND;
1201}
1202
1203void
1204aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1205 ALIAS:
1206 aio_stat = EIO_STAT
1207 aio_lstat = EIO_LSTAT
1208 PPCODE:
1209{
1210 dREQ;
1211
1212 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1213 if (!req->ptr2)
1214 {
1215 req_destroy (req);
1216 croak ("out of memory during aio_stat statdata allocation");
1217 }
1218
1219 req->flags |= FLAG_PTR2_FREE;
1220 req->sv1 = newSVsv (fh_or_path);
1221
1222 if (SvPOK (fh_or_path))
1223 {
1224 req->type = ix;
1225 req->ptr1 = SvPVbyte_nolen (req->sv1);
1226 }
1227 else
1228 {
1229 req->type = EIO_FSTAT;
1230 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1231 }
1232
1233 EIO_SEND;
1234}
1235
1236void
1237aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1238 PPCODE:
1239{
1240 dREQ;
1241
1242 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1243 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1244 req->sv1 = newSVsv (fh_or_path);
1245
1246 if (SvPOK (fh_or_path))
1247 {
1248 req->type = EIO_UTIME;
1249 req->ptr1 = SvPVbyte_nolen (req->sv1);
1250 }
1251 else
1252 {
1253 req->type = EIO_FUTIME;
1254 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1255 }
1256
1257 EIO_SEND;
1258}
1259
1260void
1261aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1262 PPCODE:
1263{
1264 dREQ;
1265
1266 req->sv1 = newSVsv (fh_or_path);
1267 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1268
1269 if (SvPOK (fh_or_path))
1270 {
1271 req->type = EIO_TRUNCATE;
1272 req->ptr1 = SvPVbyte_nolen (req->sv1);
1273 }
1274 else
1275 {
1276 req->type = EIO_FTRUNCATE;
1277 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1278 }
1279
1280 EIO_SEND;
1281}
1282
1283void
1284aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1285 ALIAS:
1286 aio_chmod = EIO_CHMOD
1287 aio_fchmod = EIO_FCHMOD
1288 aio_mkdir = EIO_MKDIR
1289 PPCODE:
1290{
1291 dREQ;
1292
1293 req->type = type;
1294 req->int2 = mode;
1295 req->sv1 = newSVsv (fh_or_path);
1296
1297 if (ix == EIO_FCHMOD)
1298 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1299 else
1300 req->ptr1 = SvPVbyte_nolen (req->sv1);
1301
1302 EIO_SEND;
1303}
1304
1305void
1306aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1307 PPCODE:
1308{
1309 dREQ;
1310
1311 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1312 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1313 req->sv1 = newSVsv (fh_or_path);
1314
1315 if (SvPOK (fh_or_path))
1316 {
1317 req->type = EIO_CHOWN;
1318 req->ptr1 = SvPVbyte_nolen (req->sv1);
1319 }
1320 else
1321 {
1322 req->type = EIO_FCHOWN;
1323 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1324 }
1325
1326 EIO_SEND;
1327}
1328
1329void
1330aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1331 ALIAS:
1332 aio_unlink = EIO_UNLINK
1333 aio_rmdir = EIO_RMDIR
1334 aio_readdir = EIO_READDIR
1335 PPCODE:
1336{
1337 dREQ;
1338
1339 req->type = ix;
1340 req->sv1 = newSVsv (pathname);
1341 req->ptr1 = SvPVbyte_nolen (req->sv1);
1342
1343 EIO_SEND;
1344}
1345
1346void
1347aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1348 ALIAS:
1349 aio_link = EIO_LINK
1350 aio_symlink = EIO_SYMLINK
1351 aio_rename = EIO_RENAME
1352 PPCODE:
1353{
1354 dREQ;
1355
1356 req->type = ix;
1357 req->sv1 = newSVsv (oldpath);
1358 req->ptr1 = SvPVbyte_nolen (req->sv1);
1359 req->sv2 = newSVsv (newpath);
1360 req->ptr2 = SvPVbyte_nolen (req->sv2);
1361
1362 EIO_SEND;
1363}
1364
1365void
1366aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1367 PPCODE:
1368{
1369 dREQ;
1370
1371 req->type = EIO_MKNOD;
1372 req->sv1 = newSVsv (pathname);
1373 req->ptr1 = SvPVbyte_nolen (req->sv1);
1374 req->int2 = (mode_t)mode;
1375 req->offs = dev;
1376
1377 EIO_SEND;
1378}
1379
1380void
1381aio_busy (double delay, SV *callback=&PL_sv_undef)
1382 PPCODE:
1383{
1384 dREQ;
1385
1386 req->type = EIO_BUSY;
1387 req->nv1 = delay < 0. ? 0. : delay;
1388
1389 EIO_SEND;
1390}
1391
1392void
1393aio_group (SV *callback=&PL_sv_undef)
1394 PROTOTYPE: ;$
1395 PPCODE:
1396{
1397 dREQ;
1398
1399 req->type = EIO_GROUP;
1400
1401 req_send (req);
1402 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1403}
1404
1405void
1406aio_nop (SV *callback=&PL_sv_undef)
1407 ALIAS:
1408 aio_nop = EIO_NOP
1409 aio_sync = EIO_SYNC
1410 PPCODE:
1411{
1412 dREQ;
1413
1414#endif
1415
1416void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2107eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1417{ 2108{
1418 grp->int2 = limit; 2109 grp->int2 = limit;
1419 grp->feed = feed; 2110 grp->feed = feed;
1420 2111
1421 grp_try_feed (grp); 2112 grp_try_feed (grp);
1422} 2113}
1423 2114
2115void
1424void eio_grp_limit (eio_req *grp, int limit) 2116eio_grp_limit (eio_req *grp, int limit)
1425{ 2117{
1426 grp->int2 = limit; 2118 grp->int2 = limit;
1427 2119
1428 grp_try_feed (grp); 2120 grp_try_feed (grp);
1429} 2121}
1430 2122
2123void
1431void eio_grp_add (eio_req *grp, eio_req *req) 2124eio_grp_add (eio_req *grp, eio_req *req)
1432{ 2125{
1433 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2126 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2127
2128 grp->flags |= EIO_FLAG_GROUPADD;
1434 2129
1435 ++grp->size; 2130 ++grp->size;
1436 req->grp = grp; 2131 req->grp = grp;
1437 2132
1438 req->grp_prev = 0; 2133 req->grp_prev = 0;
… …
1442 grp->grp_first->grp_prev = req; 2137 grp->grp_first->grp_prev = req;
1443 2138
1444 grp->grp_first = req; 2139 grp->grp_first = req;
1445} 2140}
1446 2141
2142/*****************************************************************************/
2143/* misc garbage */
1447 2144
2145ssize_t
2146eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2147{
2148 etp_worker wrk;
2149 ssize_t ret;
2150
2151 wrk.dbuf = 0;
2152
2153 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2154
2155 if (wrk.dbuf)
2156 free (wrk.dbuf);
2157
2158 return ret;
2159}
2160

Diff Legend

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