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

Comparing libeio/eio.c (file contents):
Revision 1.2 by root, Sat May 10 23:33:40 2008 UTC vs.
Revision 1.53 by root, Sat Jan 9 10:03:09 2010 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010 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
1#include "eio.h" 40#include "eio.h"
41
42#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif
2#include "xthread.h" 45#include "xthread.h"
3 46
4#include <errno.h> 47#include <errno.h>
5
6#include "EXTERN.h"
7#include "perl.h"
8#include "XSUB.h"
9
10#include <stddef.h> 48#include <stddef.h>
11#include <stdlib.h> 49#include <stdlib.h>
50#include <string.h>
12#include <errno.h> 51#include <errno.h>
13#include <sys/types.h> 52#include <sys/types.h>
14#include <sys/stat.h> 53#include <sys/stat.h>
54#include <sys/statvfs.h>
15#include <limits.h> 55#include <limits.h>
16#include <fcntl.h> 56#include <fcntl.h>
17#include <sched.h> 57#include <assert.h>
18 58
19#ifndef EIO_FINISH 59#ifndef EIO_FINISH
20# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 60# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
21#endif 61#endif
22 62
29#endif 69#endif
30 70
31#ifdef _WIN32 71#ifdef _WIN32
32 72
33 /*doh*/ 73 /*doh*/
34
35#else 74#else
36 75
37# include "config.h" 76# include "config.h"
38# include <sys/time.h> 77# include <sys/time.h>
39# include <sys/select.h> 78# include <sys/select.h>
79# include <sys/mman.h>
40# include <unistd.h> 80# include <unistd.h>
41# include <utime.h> 81# include <utime.h>
42# include <signal.h> 82# include <signal.h>
83# include <dirent.h>
84
85/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
86# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
87# define _DIRENT_HAVE_D_TYPE /* sigh */
88# define D_INO(de) (de)->d_fileno
89# define D_NAMLEN(de) (de)->d_namlen
90# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
91# define D_INO(de) (de)->d_ino
92# endif
93
94#ifdef _D_EXACT_NAMLEN
95# undef D_NAMLEN
96# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
97#endif
98
99# ifdef _DIRENT_HAVE_D_TYPE
100# define D_TYPE(de) (de)->d_type
101# endif
43 102
44# ifndef EIO_STRUCT_DIRENT 103# ifndef EIO_STRUCT_DIRENT
45# define EIO_STRUCT_DIRENT struct dirent 104# define EIO_STRUCT_DIRENT struct dirent
46# endif 105# endif
47 106
48#endif 107#endif
49 108
50# ifndef EIO_STRUCT_STAT
51# define EIO_STRUCT_STAT struct stat
52# endif
53
54#if HAVE_SENDFILE 109#if HAVE_SENDFILE
55# if __linux 110# if __linux
56# include <sys/sendfile.h> 111# include <sys/sendfile.h>
57# elif __freebsd 112# elif __FreeBSD__ || defined __APPLE__
58# include <sys/socket.h> 113# include <sys/socket.h>
59# include <sys/uio.h> 114# include <sys/uio.h>
60# elif __hpux 115# elif __hpux
61# include <sys/socket.h> 116# include <sys/socket.h>
62# elif __solaris /* not yet */ 117# elif __solaris
63# include <sys/sendfile.h> 118# include <sys/sendfile.h>
64# else 119# else
65# error sendfile support requested but not available 120# error sendfile support requested but not available
66# endif 121# endif
67#endif 122#endif
68 123
124#ifndef D_TYPE
125# define D_TYPE(de) 0
126#endif
127#ifndef D_INO
128# define D_INO(de) 0
129#endif
130#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name)
132#endif
133
69/* number of seconds after which an idle threads exit */ 134/* number of seconds after which an idle threads exit */
70#define IDLE_TIMEOUT 10 135#define IDLE_TIMEOUT 10
71 136
72/* used for struct dirent, AIX doesn't provide it */ 137/* used for struct dirent, AIX doesn't provide it */
73#ifndef NAME_MAX 138#ifndef NAME_MAX
77/* buffer size for various temporary buffers */ 142/* buffer size for various temporary buffers */
78#define EIO_BUFSIZE 65536 143#define EIO_BUFSIZE 65536
79 144
80#define dBUF \ 145#define dBUF \
81 char *eio_buf; \ 146 char *eio_buf; \
82 X_LOCK (wrklock); \ 147 ETP_WORKER_LOCK (self); \
83 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 148 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
84 X_UNLOCK (wrklock); \ 149 ETP_WORKER_UNLOCK (self); \
150 errno = ENOMEM; \
85 if (!eio_buf) \ 151 if (!eio_buf) \
86 return -1; 152 return -1;
87 153
88#define EIO_TICKS ((1000000 + 1023) >> 10) 154#define EIO_TICKS ((1000000 + 1023) >> 10)
89 155
156/*****************************************************************************/
157
158#if __GNUC__ >= 3
159# define expect(expr,value) __builtin_expect ((expr),(value))
160#else
161# define expect(expr,value) (expr)
162#endif
163
164#define expect_false(expr) expect ((expr) != 0, 0)
165#define expect_true(expr) expect ((expr) != 0, 1)
166
167/*****************************************************************************/
168
169#define ETP_PRI_MIN EIO_PRI_MIN
170#define ETP_PRI_MAX EIO_PRI_MAX
171
172struct etp_worker;
173
174#define ETP_REQ eio_req
175#define ETP_DESTROY(req) eio_destroy (req)
176static int eio_finish (eio_req *req);
177#define ETP_FINISH(req) eio_finish (req)
178static void eio_execute (struct etp_worker *self, eio_req *req);
179#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
180
181#define ETP_WORKER_CLEAR(req) \
182 if (wrk->dbuf) \
183 { \
184 free (wrk->dbuf); \
185 wrk->dbuf = 0; \
186 } \
187 \
188 if (wrk->dirp) \
189 { \
190 closedir (wrk->dirp); \
191 wrk->dirp = 0; \
192 }
193
194#define ETP_WORKER_COMMON \
195 void *dbuf; \
196 DIR *dirp;
197
198/*****************************************************************************/
199
200#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
201
202/* calculate time difference in ~1/EIO_TICKS of a second */
203static int tvdiff (struct timeval *tv1, struct timeval *tv2)
204{
205 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
206 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
207}
208
209static unsigned int started, idle, wanted = 4;
210
90static void (*want_poll_cb) (void); 211static void (*want_poll_cb) (void);
91static void (*done_poll_cb) (void); 212static void (*done_poll_cb) (void);
92 213
93static unsigned int max_poll_time = 0; 214static unsigned int max_poll_time; /* reslock */
94static unsigned int max_poll_reqs = 0; 215static unsigned int max_poll_reqs; /* reslock */
95 216
96/* calculcate time difference in ~1/EIO_TICKS of a second */ 217static volatile unsigned int nreqs; /* reqlock */
97static int tvdiff (struct timeval *tv1, struct timeval *tv2) 218static volatile unsigned int nready; /* reqlock */
98{ 219static volatile unsigned int npending; /* reqlock */
99 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 220static volatile unsigned int max_idle = 4;
100 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
101}
102 221
103static unsigned int started, idle, wanted;
104
105/* worker threads management */
106static mutex_t wrklock = X_MUTEX_INIT; 222static mutex_t wrklock = X_MUTEX_INIT;
107
108typedef struct worker {
109 /* locked by wrklock */
110 struct worker *prev, *next;
111
112 thread_t tid;
113
114 /* locked by reslock, reqlock or wrklock */
115 eio_req *req; /* currently processed request */
116 void *dbuf;
117 DIR *dirp;
118} worker;
119
120static worker wrk_first = { &wrk_first, &wrk_first, 0 };
121
122static void worker_clear (worker *wrk)
123{
124 if (wrk->dirp)
125 {
126 closedir (wrk->dirp);
127 wrk->dirp = 0;
128 }
129
130 if (wrk->dbuf)
131 {
132 free (wrk->dbuf);
133 wrk->dbuf = 0;
134 }
135}
136
137static void worker_free (worker *wrk)
138{
139 wrk->next->prev = wrk->prev;
140 wrk->prev->next = wrk->next;
141
142 free (wrk);
143}
144
145static volatile unsigned int nreqs, nready, npending;
146static volatile unsigned int max_idle = 4;
147
148static mutex_t reslock = X_MUTEX_INIT; 223static mutex_t reslock = X_MUTEX_INIT;
149static mutex_t reqlock = X_MUTEX_INIT; 224static mutex_t reqlock = X_MUTEX_INIT;
150static cond_t reqwait = X_COND_INIT; 225static cond_t reqwait = X_COND_INIT;
151 226
227#if !HAVE_PREADWRITE
228/*
229 * make our pread/pwrite emulation safe against themselves, but not against
230 * normal read/write by using a mutex. slows down execution a lot,
231 * but that's your problem, not mine.
232 */
233static mutex_t preadwritelock = X_MUTEX_INIT;
234#endif
235
236typedef struct etp_worker
237{
238 /* locked by wrklock */
239 struct etp_worker *prev, *next;
240
241 thread_t tid;
242
243 /* locked by reslock, reqlock or wrklock */
244 ETP_REQ *req; /* currently processed request */
245
246 ETP_WORKER_COMMON
247} etp_worker;
248
249static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
250
251#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
252#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
253
254/* worker threads management */
255
256static void etp_worker_clear (etp_worker *wrk)
257{
258 ETP_WORKER_CLEAR (wrk);
259}
260
261static void etp_worker_free (etp_worker *wrk)
262{
263 wrk->next->prev = wrk->prev;
264 wrk->prev->next = wrk->next;
265
266 free (wrk);
267}
268
152unsigned int eio_nreqs (void) 269static unsigned int etp_nreqs (void)
153{ 270{
271 int retval;
272 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
273 retval = nreqs;
274 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 275 return retval;
155} 276}
156 277
157unsigned int eio_nready (void) 278static unsigned int etp_nready (void)
158{ 279{
159 unsigned int retval; 280 unsigned int retval;
160 281
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 282 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 283 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 284 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 285
165 return retval; 286 return retval;
166} 287}
167 288
168unsigned int eio_npending (void) 289static unsigned int etp_npending (void)
169{ 290{
170 unsigned int retval; 291 unsigned int retval;
171 292
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 293 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 294 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 295 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 296
176 return retval; 297 return retval;
177} 298}
178 299
179unsigned int eio_nthreads (void) 300static unsigned int etp_nthreads (void)
180{ 301{
181 unsigned int retval; 302 unsigned int retval;
182 303
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 304 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 305 retval = started;
191 * a somewhat faster data structure might be nice, but 312 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 313 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 314 * per shift, the most expensive operation.
194 */ 315 */
195typedef struct { 316typedef struct {
196 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 317 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197 int size; 318 int size;
198} reqq; 319} etp_reqq;
199 320
200static reqq req_queue; 321static etp_reqq req_queue;
201static reqq res_queue; 322static etp_reqq res_queue;
202 323
203static int reqq_push (reqq *q, eio_req *req) 324static int reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 325{
205 int pri = req->pri; 326 int pri = req->pri;
206 req->next = 0; 327 req->next = 0;
207 328
208 if (q->qe[pri]) 329 if (q->qe[pri])
214 q->qe[pri] = q->qs[pri] = req; 335 q->qe[pri] = q->qs[pri] = req;
215 336
216 return q->size++; 337 return q->size++;
217} 338}
218 339
219static eio_req *reqq_shift (reqq *q) 340static ETP_REQ *reqq_shift (etp_reqq *q)
220{ 341{
221 int pri; 342 int pri;
222 343
223 if (!q->size) 344 if (!q->size)
224 return 0; 345 return 0;
225 346
226 --q->size; 347 --q->size;
227 348
228 for (pri = EIO_NUM_PRI; pri--; ) 349 for (pri = ETP_NUM_PRI; pri--; )
229 { 350 {
230 eio_req *req = q->qs[pri]; 351 eio_req *req = q->qs[pri];
231 352
232 if (req) 353 if (req)
233 { 354 {
239 } 360 }
240 361
241 abort (); 362 abort ();
242} 363}
243 364
244static void grp_try_feed (eio_req *grp) 365static void etp_atfork_prepare (void)
245{ 366{
246 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 367 X_LOCK (wrklock);
368 X_LOCK (reqlock);
369 X_LOCK (reslock);
370#if !HAVE_PREADWRITE
371 X_LOCK (preadwritelock);
372#endif
373}
374
375static void etp_atfork_parent (void)
376{
377#if !HAVE_PREADWRITE
378 X_UNLOCK (preadwritelock);
379#endif
380 X_UNLOCK (reslock);
381 X_UNLOCK (reqlock);
382 X_UNLOCK (wrklock);
383}
384
385static void etp_atfork_child (void)
386{
387 ETP_REQ *prv;
388
389 while ((prv = reqq_shift (&req_queue)))
390 ETP_DESTROY (prv);
391
392 while ((prv = reqq_shift (&res_queue)))
393 ETP_DESTROY (prv);
394
395 while (wrk_first.next != &wrk_first)
247 { 396 {
248 int old_len = grp->size; 397 etp_worker *wrk = wrk_first.next;
249 398
250 EIO_FEED (grp); 399 if (wrk->req)
400 ETP_DESTROY (wrk->req);
251 401
252 /* stop if no progress has been made */ 402 etp_worker_clear (wrk);
253 if (old_len == grp->size) 403 etp_worker_free (wrk);
254 {
255 grp->feed = 0;
256 break;
257 }
258 } 404 }
259}
260 405
261static int eio_finish (eio_req *req); 406 started = 0;
407 idle = 0;
408 nreqs = 0;
409 nready = 0;
410 npending = 0;
262 411
263static int grp_dec (eio_req *grp) 412 etp_atfork_parent ();
264{ 413}
265 --grp->size;
266 414
267 /* call feeder, if applicable */ 415static void
268 grp_try_feed (grp); 416etp_once_init (void)
417{
418 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
419}
269 420
270 /* finish, if done */ 421static int
271 if (!grp->size && grp->int1) 422etp_init (void (*want_poll)(void), void (*done_poll)(void))
272 return eio_finish (grp); 423{
273 else 424 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
425
426 pthread_once (&doinit, etp_once_init);
427
428 want_poll_cb = want_poll;
429 done_poll_cb = done_poll;
430
274 return 0; 431 return 0;
275} 432}
276 433
277void eio_destroy (eio_req *req)
278{
279 if ((req)->flags & EIO_FLAG_PTR2_FREE)
280 free (req->ptr2);
281
282 EIO_DESTROY (req);
283}
284
285static int eio_finish (eio_req *req)
286{
287 int res = EIO_FINISH (req);
288
289 if (req->grp)
290 {
291 int res2;
292 eio_req *grp = req->grp;
293
294 /* unlink request */
295 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
296 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
297
298 if (grp->grp_first == req)
299 grp->grp_first = req->grp_next;
300
301 res2 = grp_dec (grp);
302
303 if (!res && res2)
304 res = res2;
305 }
306
307 eio_destroy (req);
308
309 return res;
310}
311
312void eio_grp_cancel (eio_req *grp)
313{
314 for (grp = grp->grp_first; grp; grp = grp->grp_next)
315 eio_cancel (grp);
316}
317
318void eio_cancel (eio_req *req)
319{
320 req->flags |= EIO_FLAG_CANCELLED;
321
322 eio_grp_cancel (req);
323}
324
325X_THREAD_PROC (eio_proc); 434X_THREAD_PROC (etp_proc);
326 435
327static void start_thread (void) 436static void etp_start_thread (void)
328{ 437{
329 worker *wrk = calloc (1, sizeof (worker)); 438 etp_worker *wrk = calloc (1, sizeof (etp_worker));
330 439
331 if (!wrk) 440 /*TODO*/
332 croak ("unable to allocate worker thread data"); 441 assert (("unable to allocate worker thread data", wrk));
333 442
334 X_LOCK (wrklock); 443 X_LOCK (wrklock);
335 444
336 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 445 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
337 { 446 {
338 wrk->prev = &wrk_first; 447 wrk->prev = &wrk_first;
339 wrk->next = wrk_first.next; 448 wrk->next = wrk_first.next;
340 wrk_first.next->prev = wrk; 449 wrk_first.next->prev = wrk;
341 wrk_first.next = wrk; 450 wrk_first.next = wrk;
345 free (wrk); 454 free (wrk);
346 455
347 X_UNLOCK (wrklock); 456 X_UNLOCK (wrklock);
348} 457}
349 458
350static void maybe_start_thread (void) 459static void etp_maybe_start_thread (void)
351{ 460{
352 if (eio_nthreads () >= wanted) 461 if (expect_true (etp_nthreads () >= wanted))
353 return; 462 return;
354 463
355 /* todo: maybe use idle here, but might be less exact */ 464 /* todo: maybe use idle here, but might be less exact */
356 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 465 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
357 return; 466 return;
358 467
359 start_thread (); 468 etp_start_thread ();
360} 469}
361 470
362void eio_submit (eio_req *req)
363{
364 ++nreqs;
365
366 X_LOCK (reqlock);
367 ++nready;
368 reqq_push (&req_queue, req);
369 X_COND_SIGNAL (reqwait);
370 X_UNLOCK (reqlock);
371
372 maybe_start_thread ();
373}
374
375static void end_thread (void) 471static void etp_end_thread (void)
376{ 472{
377 eio_req *req = calloc (1, sizeof (eio_req)); 473 eio_req *req = calloc (1, sizeof (eio_req));
378 474
379 req->type = EIO_QUIT; 475 req->type = -1;
380 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 476 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381 477
382 X_LOCK (reqlock); 478 X_LOCK (reqlock);
383 reqq_push (&req_queue, req); 479 reqq_push (&req_queue, req);
384 X_COND_SIGNAL (reqwait); 480 X_COND_SIGNAL (reqwait);
385 X_UNLOCK (reqlock); 481 X_UNLOCK (reqlock);
387 X_LOCK (wrklock); 483 X_LOCK (wrklock);
388 --started; 484 --started;
389 X_UNLOCK (wrklock); 485 X_UNLOCK (wrklock);
390} 486}
391 487
392void eio_set_max_poll_time (double nseconds)
393{
394 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
395 max_poll_time = nseconds;
396 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397}
398
399void eio_set_max_poll_reqs (unsigned int maxreqs)
400{
401 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
402 max_poll_reqs = maxreqs;
403 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404}
405
406void eio_set_max_idle (unsigned int nthreads)
407{
408 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
409 max_idle = nthreads <= 0 ? 1 : nthreads;
410 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411}
412
413void eio_set_min_parallel (unsigned int nthreads)
414{
415 if (wanted < nthreads)
416 wanted = nthreads;
417}
418
419void eio_set_max_parallel (unsigned int nthreads)
420{
421 if (wanted > nthreads)
422 wanted = nthreads;
423
424 while (started > wanted)
425 end_thread ();
426}
427
428int eio_poll (void) 488static int etp_poll (void)
429{ 489{
430 int maxreqs = max_poll_reqs; 490 unsigned int maxreqs;
491 unsigned int maxtime;
431 struct timeval tv_start, tv_now; 492 struct timeval tv_start, tv_now;
432 eio_req *req;
433 493
494 X_LOCK (reslock);
495 maxreqs = max_poll_reqs;
496 maxtime = max_poll_time;
497 X_UNLOCK (reslock);
498
434 if (max_poll_time) 499 if (maxtime)
435 gettimeofday (&tv_start, 0); 500 gettimeofday (&tv_start, 0);
436 501
437 for (;;) 502 for (;;)
438 { 503 {
504 ETP_REQ *req;
505
439 maybe_start_thread (); 506 etp_maybe_start_thread ();
440 507
441 X_LOCK (reslock); 508 X_LOCK (reslock);
442 req = reqq_shift (&res_queue); 509 req = reqq_shift (&res_queue);
443 510
444 if (req) 511 if (req)
452 X_UNLOCK (reslock); 519 X_UNLOCK (reslock);
453 520
454 if (!req) 521 if (!req)
455 return 0; 522 return 0;
456 523
524 X_LOCK (reqlock);
457 --nreqs; 525 --nreqs;
526 X_UNLOCK (reqlock);
458 527
459 if (req->type == EIO_GROUP && req->size) 528 if (expect_false (req->type == EIO_GROUP && req->size))
460 { 529 {
461 req->int1 = 1; /* mark request as delayed */ 530 req->int1 = 1; /* mark request as delayed */
462 continue; 531 continue;
463 } 532 }
464 else 533 else
465 { 534 {
466 int res = eio_finish (req); 535 int res = ETP_FINISH (req);
467 if (res) 536 if (expect_false (res))
468 return res; 537 return res;
469 } 538 }
470 539
471 if (maxreqs && !--maxreqs) 540 if (expect_false (maxreqs && !--maxreqs))
472 break; 541 break;
473 542
474 if (max_poll_time) 543 if (maxtime)
475 { 544 {
476 gettimeofday (&tv_now, 0); 545 gettimeofday (&tv_now, 0);
477 546
478 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 547 if (tvdiff (&tv_start, &tv_now) >= maxtime)
479 break; 548 break;
480 } 549 }
481 } 550 }
482 551
483 errno = EAGAIN; 552 errno = EAGAIN;
484 return -1; 553 return -1;
485} 554}
486 555
556static void etp_cancel (ETP_REQ *req)
557{
558 X_LOCK (wrklock);
559 req->flags |= EIO_FLAG_CANCELLED;
560 X_UNLOCK (wrklock);
561
562 eio_grp_cancel (req);
563}
564
565static void etp_submit (ETP_REQ *req)
566{
567 req->pri -= ETP_PRI_MIN;
568
569 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
570 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
571
572 if (expect_false (req->type == EIO_GROUP))
573 {
574 /* I hope this is worth it :/ */
575 X_LOCK (reqlock);
576 ++nreqs;
577 X_UNLOCK (reqlock);
578
579 X_LOCK (reslock);
580
581 ++npending;
582
583 if (!reqq_push (&res_queue, req) && want_poll_cb)
584 want_poll_cb ();
585
586 X_UNLOCK (reslock);
587 }
588 else
589 {
590 X_LOCK (reqlock);
591 ++nreqs;
592 ++nready;
593 reqq_push (&req_queue, req);
594 X_COND_SIGNAL (reqwait);
595 X_UNLOCK (reqlock);
596
597 etp_maybe_start_thread ();
598 }
599}
600
601static void etp_set_max_poll_time (double nseconds)
602{
603 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
604 max_poll_time = nseconds * EIO_TICKS;
605 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
606}
607
608static void etp_set_max_poll_reqs (unsigned int maxreqs)
609{
610 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
611 max_poll_reqs = maxreqs;
612 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
613}
614
615static void etp_set_max_idle (unsigned int nthreads)
616{
617 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
618 max_idle = nthreads <= 0 ? 1 : nthreads;
619 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
620}
621
622static void etp_set_min_parallel (unsigned int nthreads)
623{
624 if (wanted < nthreads)
625 wanted = nthreads;
626}
627
628static void etp_set_max_parallel (unsigned int nthreads)
629{
630 if (wanted > nthreads)
631 wanted = nthreads;
632
633 while (started > wanted)
634 etp_end_thread ();
635}
636
637/*****************************************************************************/
638
639static void grp_try_feed (eio_req *grp)
640{
641 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
642 {
643 grp->flags &= ~EIO_FLAG_GROUPADD;
644
645 EIO_FEED (grp);
646
647 /* stop if no progress has been made */
648 if (!(grp->flags & EIO_FLAG_GROUPADD))
649 {
650 grp->feed = 0;
651 break;
652 }
653 }
654}
655
656static int grp_dec (eio_req *grp)
657{
658 --grp->size;
659
660 /* call feeder, if applicable */
661 grp_try_feed (grp);
662
663 /* finish, if done */
664 if (!grp->size && grp->int1)
665 return eio_finish (grp);
666 else
667 return 0;
668}
669
670void eio_destroy (eio_req *req)
671{
672 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
673 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
674
675 EIO_DESTROY (req);
676}
677
678static int eio_finish (eio_req *req)
679{
680 int res = EIO_FINISH (req);
681
682 if (req->grp)
683 {
684 int res2;
685 eio_req *grp = req->grp;
686
687 /* unlink request */
688 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
689 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
690
691 if (grp->grp_first == req)
692 grp->grp_first = req->grp_next;
693
694 res2 = grp_dec (grp);
695
696 if (!res && res2)
697 res = res2;
698 }
699
700 eio_destroy (req);
701
702 return res;
703}
704
705void eio_grp_cancel (eio_req *grp)
706{
707 for (grp = grp->grp_first; grp; grp = grp->grp_next)
708 eio_cancel (grp);
709}
710
711void eio_cancel (eio_req *req)
712{
713 etp_cancel (req);
714}
715
716void eio_submit (eio_req *req)
717{
718 etp_submit (req);
719}
720
721unsigned int eio_nreqs (void)
722{
723 return etp_nreqs ();
724}
725
726unsigned int eio_nready (void)
727{
728 return etp_nready ();
729}
730
731unsigned int eio_npending (void)
732{
733 return etp_npending ();
734}
735
736unsigned int eio_nthreads (void)
737{
738 return etp_nthreads ();
739}
740
741void eio_set_max_poll_time (double nseconds)
742{
743 etp_set_max_poll_time (nseconds);
744}
745
746void eio_set_max_poll_reqs (unsigned int maxreqs)
747{
748 etp_set_max_poll_reqs (maxreqs);
749}
750
751void eio_set_max_idle (unsigned int nthreads)
752{
753 etp_set_max_idle (nthreads);
754}
755
756void eio_set_min_parallel (unsigned int nthreads)
757{
758 etp_set_min_parallel (nthreads);
759}
760
761void eio_set_max_parallel (unsigned int nthreads)
762{
763 etp_set_max_parallel (nthreads);
764}
765
766int eio_poll (void)
767{
768 return etp_poll ();
769}
770
487/*****************************************************************************/ 771/*****************************************************************************/
488/* work around various missing functions */ 772/* work around various missing functions */
489 773
490#if !HAVE_PREADWRITE 774#if !HAVE_PREADWRITE
775# undef pread
776# undef pwrite
491# define pread aio_pread 777# define pread eio__pread
492# define pwrite aio_pwrite 778# define pwrite eio__pwrite
493 779
494/* 780static ssize_t
495 * make our pread/pwrite safe against themselves, but not against 781eio__pread (int fd, void *buf, size_t count, off_t offset)
496 * normal read/write by using a mutex. slows down execution a lot,
497 * but that's your problem, not mine.
498 */
499static mutex_t preadwritelock = X_MUTEX_INIT;
500
501static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
502{ 782{
503 ssize_t res; 783 ssize_t res;
504 off_t ooffset; 784 off_t ooffset;
505 785
506 X_LOCK (preadwritelock); 786 X_LOCK (preadwritelock);
511 X_UNLOCK (preadwritelock); 791 X_UNLOCK (preadwritelock);
512 792
513 return res; 793 return res;
514} 794}
515 795
516static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 796static ssize_t
797eio__pwrite (int fd, void *buf, size_t count, off_t offset)
517{ 798{
518 ssize_t res; 799 ssize_t res;
519 off_t ooffset; 800 off_t ooffset;
520 801
521 X_LOCK (preadwritelock); 802 X_LOCK (preadwritelock);
522 ooffset = lseek (fd, 0, SEEK_CUR); 803 ooffset = lseek (fd, 0, SEEK_CUR);
523 lseek (fd, offset, SEEK_SET); 804 lseek (fd, offset, SEEK_SET);
524 res = write (fd, buf, count); 805 res = write (fd, buf, count);
525 lseek (fd, offset, SEEK_SET); 806 lseek (fd, ooffset, SEEK_SET);
526 X_UNLOCK (preadwritelock); 807 X_UNLOCK (preadwritelock);
527 808
528 return res; 809 return res;
529} 810}
530#endif 811#endif
531 812
532#ifndef HAVE_FUTIMES 813#ifndef HAVE_FUTIMES
533 814
815# undef utimes
816# undef futimes
534# define utimes(path,times) aio_utimes (path, times) 817# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times) 818# define futimes(fd,times) eio__futimes (fd, times)
536 819
820static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 821eio__utimes (const char *filename, const struct timeval times[2])
538{ 822{
539 if (times) 823 if (times)
540 { 824 {
541 struct utimbuf buf; 825 struct utimbuf buf;
542 826
547 } 831 }
548 else 832 else
549 return utime (filename, 0); 833 return utime (filename, 0);
550} 834}
551 835
552static int aio_futimes (int fd, const struct timeval tv[2]) 836static int eio__futimes (int fd, const struct timeval tv[2])
553{ 837{
554 errno = ENOSYS; 838 errno = ENOSYS;
555 return -1; 839 return -1;
556} 840}
557 841
558#endif 842#endif
559 843
560#if !HAVE_FDATASYNC 844#if !HAVE_FDATASYNC
845# undef fdatasync
561# define fdatasync fsync 846# define fdatasync(fd) fsync (fd)
562#endif 847#endif
848
849/* sync_file_range always needs emulation */
850int
851eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
852{
853#if HAVE_SYNC_FILE_RANGE
854 int res;
855
856 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
857 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
858 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
859 {
860 flags = 0
861 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
862 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
863 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
864 }
865
866 res = sync_file_range (fd, offset, nbytes, flags);
867
868 if (!res || errno != ENOSYS)
869 return res;
870#endif
871
872 /* even though we could play tricks with the flags, it's better to always
873 * call fdatasync, as that matches the expectation of its users best */
874 return fdatasync (fd);
875}
563 876
564#if !HAVE_READAHEAD 877#if !HAVE_READAHEAD
878# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 879# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 880
881static ssize_t
567static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 882eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568{ 883{
569 size_t todo = count; 884 size_t todo = count;
570 dBUF; 885 dBUF;
571 886
572 while (todo > 0) 887 while (todo > 0)
573 { 888 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 889 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 890
576 pread (fd, aio_buf, len, offset); 891 pread (fd, eio_buf, len, offset);
577 offset += len; 892 offset += len;
578 todo -= len; 893 todo -= len;
579 } 894 }
580 895
581 errno = 0; 896 errno = 0;
582 return count; 897 return count;
583} 898}
584 899
585#endif 900#endif
586 901
587#if !HAVE_READDIR_R
588# define readdir_r aio_readdir_r
589
590static mutex_t readdirlock = X_MUTEX_INIT;
591
592static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
593{
594 X_DIRENT *e;
595 int errorno;
596
597 X_LOCK (readdirlock);
598
599 e = readdir (dirp);
600 errorno = errno;
601
602 if (e)
603 {
604 *res = ent;
605 strcpy (ent->d_name, e->d_name);
606 }
607 else
608 *res = 0;
609
610 X_UNLOCK (readdirlock);
611
612 errno = errorno;
613 return e ? 0 : -1;
614}
615#endif
616
617/* sendfile always needs emulation */ 902/* sendfile always needs emulation */
903static ssize_t
618static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 904eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
619{ 905{
620 ssize_t res; 906 ssize_t res;
621 907
622 if (!count) 908 if (!count)
623 return 0; 909 return 0;
624 910
625#if HAVE_SENDFILE 911#if HAVE_SENDFILE
626# if __linux 912# if __linux
627 res = sendfile (ofd, ifd, &offset, count); 913 res = sendfile (ofd, ifd, &offset, count);
628 914
629# elif __freebsd 915# elif __FreeBSD__
630 /* 916 /*
631 * Of course, the freebsd sendfile is a dire hack with no thoughts 917 * Of course, the freebsd sendfile is a dire hack with no thoughts
632 * wasted on making it similar to other I/O functions. 918 * wasted on making it similar to other I/O functions.
633 */ 919 */
634 { 920 {
635 off_t sbytes; 921 off_t sbytes;
636 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 922 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637 923
638 if (res < 0 && sbytes) 924 #if 0 /* according to the manpage, this is correct, but broken behaviour */
639 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 925 /* freebsd' sendfile will return 0 on success */
926 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
927 /* not on e.g. EIO or EPIPE - sounds broken */
928 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
929 res = sbytes;
930 #endif
931
932 /* according to source inspection, this is correct, and useful behaviour */
933 if (sbytes)
934 res = sbytes;
935 }
936
937# elif defined (__APPLE__)
938
939 {
940 off_t sbytes = count;
941 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
942
943 /* according to the manpage, sbytes is always valid */
944 if (sbytes)
640 res = sbytes; 945 res = sbytes;
641 } 946 }
642 947
643# elif __hpux 948# elif __hpux
644 res = sendfile (ofd, ifd, offset, count, 0, 0); 949 res = sendfile (ofd, ifd, offset, count, 0, 0);
658 if (res < 0 && sbytes) 963 if (res < 0 && sbytes)
659 res = sbytes; 964 res = sbytes;
660 } 965 }
661 966
662# endif 967# endif
968
969#elif defined (_WIN32)
970
971 /* does not work, just for documentation of what would need to be done */
972 {
973 HANDLE h = TO_SOCKET (ifd);
974 SetFilePointer (h, offset, 0, FILE_BEGIN);
975 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
976 }
977
663#else 978#else
664 res = -1; 979 res = -1;
665 errno = ENOSYS; 980 errno = ENOSYS;
666#endif 981#endif
667 982
668 if (res < 0 983 if (res < 0
669 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 984 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
985 /* BSDs */
986#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
987 || errno == ENOTSUP
988#endif
989 || errno == EOPNOTSUPP /* BSDs */
670#if __solaris 990#if __solaris
671 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 991 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
672#endif 992#endif
673 ) 993 )
674 ) 994 )
705 } 1025 }
706 1026
707 return res; 1027 return res;
708} 1028}
709 1029
1030static signed char
1031eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1032{
1033 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1034 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1035}
1036
1037#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1038
1039#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1040#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1041
1042static void
1043eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1044{
1045 unsigned char bits [9 + sizeof (ino_t) * 8];
1046 unsigned char *bit = bits;
1047
1048 assert (CHAR_BIT == 8);
1049 assert (sizeof (eio_dirent) * 8 < 256);
1050 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1051 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1052
1053 if (size <= EIO_SORT_FAST)
1054 return;
1055
1056 /* first prepare an array of bits to test in our radix sort */
1057 /* try to take endianness into account, as well as differences in ino_t sizes */
1058 /* inode_bits must contain all inodes ORed together */
1059 /* which is used to skip bits that are 0 everywhere, which is very common */
1060 {
1061 ino_t endianness;
1062 int i, j;
1063
1064 /* we store the byte offset of byte n into byte n of "endianness" */
1065 for (i = 0; i < sizeof (ino_t); ++i)
1066 ((unsigned char *)&endianness)[i] = i;
1067
1068 *bit++ = 0;
1069
1070 for (i = 0; i < sizeof (ino_t); ++i)
1071 {
1072 /* shifting off the byte offsets out of "endianness" */
1073 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1074 endianness >>= 8;
1075
1076 for (j = 0; j < 8; ++j)
1077 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1078 *bit++ = offs + j;
1079 }
1080
1081 for (j = 0; j < 8; ++j)
1082 if (score_bits & (1 << j))
1083 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1084 }
1085
1086 /* now actually do the sorting (a variant of MSD radix sort) */
1087 {
1088 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1089 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1090 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1091 int stk_idx = 0;
1092
1093 base_stk [stk_idx] = dents;
1094 end_stk [stk_idx] = dents + size;
1095 bit_stk [stk_idx] = bit - 1;
1096
1097 do
1098 {
1099 base = base_stk [stk_idx];
1100 end = end_stk [stk_idx];
1101 bit = bit_stk [stk_idx];
1102
1103 for (;;)
1104 {
1105 unsigned char O = *bit >> 3;
1106 unsigned char M = 1 << (*bit & 7);
1107
1108 eio_dirent *a = base;
1109 eio_dirent *b = end;
1110
1111 if (b - a < EIO_SORT_CUTOFF)
1112 break;
1113
1114 /* now bit-partition the array on the bit */
1115 /* this ugly asymmetric loop seems to perform much better than typical */
1116 /* partition algos found in the literature */
1117 do
1118 if (!(((unsigned char *)a)[O] & M))
1119 ++a;
1120 else if (!(((unsigned char *)--b)[O] & M))
1121 {
1122 eio_dirent tmp = *a; *a = *b; *b = tmp;
1123 ++a;
1124 }
1125 while (b > a);
1126
1127 /* next bit, or stop, if no bits left in this path */
1128 if (!*--bit)
1129 break;
1130
1131 base_stk [stk_idx] = a;
1132 end_stk [stk_idx] = end;
1133 bit_stk [stk_idx] = bit;
1134 ++stk_idx;
1135
1136 end = a;
1137 }
1138 }
1139 while (stk_idx--);
1140 }
1141}
1142
1143static void
1144eio_dent_insertion_sort (eio_dirent *dents, int size)
1145{
1146 /* first move the smallest element to the front, to act as a sentinel */
1147 {
1148 int i;
1149 eio_dirent *min = dents;
1150
1151 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1152 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1153 if (EIO_DENT_CMP (dents [i], <, *min))
1154 min = &dents [i];
1155
1156 /* swap elements 0 and j (minimum) */
1157 {
1158 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1159 }
1160 }
1161
1162 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1163 {
1164 eio_dirent *i, *j;
1165
1166 for (i = dents + 1; i < dents + size; ++i)
1167 {
1168 eio_dirent value = *i;
1169
1170 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1171 j [1] = j [0];
1172
1173 j [1] = value;
1174 }
1175 }
1176}
1177
1178static void
1179eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1180{
1181 if (size <= 1)
1182 return; /* our insertion sort relies on size > 0 */
1183
1184 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1185 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1186 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1187
1188 /* use an insertion sort at the end, or for small arrays, */
1189 /* as insertion sort is more efficient for small partitions */
1190 eio_dent_insertion_sort (dents, size);
1191}
1192
710/* read a full directory */ 1193/* read a full directory */
1194static void
711static void scandir_ (eio_req *req, worker *self) 1195eio__scandir (eio_req *req, etp_worker *self)
712{ 1196{
713 DIR *dirp; 1197 DIR *dirp;
714 union
715 {
716 EIO_STRUCT_DIRENT d;
717 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
718 } *u;
719 EIO_STRUCT_DIRENT *entp; 1198 EIO_STRUCT_DIRENT *entp;
720 char *name, *names; 1199 char *name, *names;
721 int memlen = 4096; 1200 int namesalloc = 4096;
722 int memofs = 0; 1201 int namesoffs = 0;
1202 int flags = req->int1;
1203 eio_dirent *dents = 0;
1204 int dentalloc = 128;
723 int res = 0; 1205 int dentoffs = 0;
1206 ino_t inode_bits = 0;
1207
1208 req->result = -1;
1209
1210 if (!(flags & EIO_READDIR_DENTS))
1211 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
724 1212
725 X_LOCK (wrklock); 1213 X_LOCK (wrklock);
1214 /* the corresponding closedir is in ETP_WORKER_CLEAR */
726 self->dirp = dirp = opendir (req->ptr1); 1215 self->dirp = dirp = opendir (req->ptr1);
727 self->dbuf = u = malloc (sizeof (*u));
728 req->flags |= EIO_FLAG_PTR2_FREE; 1216 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1217 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
729 req->ptr2 = names = malloc (memlen); 1218 req->ptr2 = names = malloc (namesalloc);
730 X_UNLOCK (wrklock); 1219 X_UNLOCK (wrklock);
731 1220
732 if (dirp && u && names) 1221 if (dirp && names && (!flags || dents))
733 for (;;) 1222 for (;;)
734 { 1223 {
735 errno = 0; 1224 errno = 0;
736 readdir_r (dirp, &u->d, &entp); 1225 entp = readdir (dirp);
737 1226
738 if (!entp) 1227 if (!entp)
1228 {
1229 if (errno)
1230 break;
1231
1232 /* sort etc. */
1233 req->int1 = flags;
1234 req->result = dentoffs;
1235
1236 if (flags & EIO_READDIR_STAT_ORDER)
1237 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
1238 else if (flags & EIO_READDIR_DIRS_FIRST)
1239 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1240 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1241 else
1242 {
1243 /* in this case, all is known, and we just put dirs first and sort them */
1244 eio_dirent *oth = dents + dentoffs;
1245 eio_dirent *dir = dents;
1246
1247 /* now partition dirs to the front, and non-dirs to the back */
1248 /* by walking from both sides and swapping if necessary */
1249 /* also clear score, so it doesn't influence sorting */
1250 while (oth > dir)
1251 {
1252 if (dir->type == EIO_DT_DIR)
1253 ++dir;
1254 else if ((--oth)->type == EIO_DT_DIR)
1255 {
1256 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1257
1258 ++dir;
1259 }
1260 }
1261
1262 /* now sort the dirs only */
1263 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1264 }
1265
739 break; 1266 break;
1267 }
740 1268
1269 /* now add the entry to our list(s) */
741 name = entp->d_name; 1270 name = entp->d_name;
742 1271
1272 /* skip . and .. entries */
743 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1273 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
744 { 1274 {
745 int len = strlen (name) + 1; 1275 int len = D_NAMLEN (entp) + 1;
746 1276
747 res++; 1277 while (expect_false (namesoffs + len > namesalloc))
748
749 while (memofs + len > memlen)
750 { 1278 {
751 memlen *= 2; 1279 namesalloc *= 2;
752 X_LOCK (wrklock); 1280 X_LOCK (wrklock);
753 req->ptr2 = names = realloc (names, memlen); 1281 req->ptr2 = names = realloc (names, namesalloc);
754 X_UNLOCK (wrklock); 1282 X_UNLOCK (wrklock);
755 1283
756 if (!names) 1284 if (!names)
757 break; 1285 break;
758 } 1286 }
759 1287
760 memcpy (names + memofs, name, len); 1288 memcpy (names + namesoffs, name, len);
1289
1290 if (dents)
1291 {
1292 struct eio_dirent *ent;
1293
1294 if (expect_false (dentoffs == dentalloc))
1295 {
1296 dentalloc *= 2;
1297 X_LOCK (wrklock);
1298 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1299 X_UNLOCK (wrklock);
1300
1301 if (!dents)
1302 break;
1303 }
1304
1305 ent = dents + dentoffs;
1306
1307 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1308 ent->namelen = len - 1;
1309 ent->inode = D_INO (entp);
1310
1311 inode_bits |= ent->inode;
1312
1313 switch (D_TYPE (entp))
1314 {
1315 default:
1316 ent->type = EIO_DT_UNKNOWN;
1317 flags |= EIO_READDIR_FOUND_UNKNOWN;
1318 break;
1319
1320 #ifdef DT_FIFO
1321 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1322 #endif
1323 #ifdef DT_CHR
1324 case DT_CHR: ent->type = EIO_DT_CHR; break;
1325 #endif
1326 #ifdef DT_MPC
1327 case DT_MPC: ent->type = EIO_DT_MPC; break;
1328 #endif
1329 #ifdef DT_DIR
1330 case DT_DIR: ent->type = EIO_DT_DIR; break;
1331 #endif
1332 #ifdef DT_NAM
1333 case DT_NAM: ent->type = EIO_DT_NAM; break;
1334 #endif
1335 #ifdef DT_BLK
1336 case DT_BLK: ent->type = EIO_DT_BLK; break;
1337 #endif
1338 #ifdef DT_MPB
1339 case DT_MPB: ent->type = EIO_DT_MPB; break;
1340 #endif
1341 #ifdef DT_REG
1342 case DT_REG: ent->type = EIO_DT_REG; break;
1343 #endif
1344 #ifdef DT_NWK
1345 case DT_NWK: ent->type = EIO_DT_NWK; break;
1346 #endif
1347 #ifdef DT_CMP
1348 case DT_CMP: ent->type = EIO_DT_CMP; break;
1349 #endif
1350 #ifdef DT_LNK
1351 case DT_LNK: ent->type = EIO_DT_LNK; break;
1352 #endif
1353 #ifdef DT_SOCK
1354 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1355 #endif
1356 #ifdef DT_DOOR
1357 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1358 #endif
1359 #ifdef DT_WHT
1360 case DT_WHT: ent->type = EIO_DT_WHT; break;
1361 #endif
1362 }
1363
1364 ent->score = 7;
1365
1366 if (flags & EIO_READDIR_DIRS_FIRST)
1367 {
1368 if (ent->type == EIO_DT_UNKNOWN)
1369 {
1370 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1371 ent->score = 1;
1372 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1373 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1374 }
1375 else if (ent->type == EIO_DT_DIR)
1376 ent->score = 0;
1377 }
1378 }
1379
761 memofs += len; 1380 namesoffs += len;
1381 ++dentoffs;
1382 }
1383
1384 if (EIO_CANCELLED (req))
1385 {
1386 errno = ECANCELED;
1387 break;
762 } 1388 }
763 } 1389 }
1390}
764 1391
765 if (errno) 1392#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
766 res = -1; 1393# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
767 1394#else
768 req->result = res; 1395
1396int
1397eio__msync (void *mem, size_t len, int flags)
1398{
1399 if (EIO_MS_ASYNC != MS_SYNC
1400 || EIO_MS_INVALIDATE != MS_INVALIDATE
1401 || EIO_MS_SYNC != MS_SYNC)
1402 {
1403 flags = 0
1404 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1405 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1406 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1407 }
1408
1409 return msync (mem, len, flags);
1410}
1411
1412#endif
1413
1414int
1415eio__mtouch (void *mem, size_t len, int flags)
1416{
1417 intptr_t addr = (intptr_t)mem;
1418 intptr_t end = addr + len;
1419#ifdef PAGESIZE
1420 const intptr_t page = PAGESIZE;
1421#else
1422 static intptr_t page;
1423
1424 if (!page)
1425 page = sysconf (_SC_PAGESIZE);
1426#endif
1427
1428 /* round down to start of page, although this is probably useless */
1429 addr &= ~(page - 1); /* assume page size is always a power of two */
1430
1431 if (addr < end)
1432 if (flags & EIO_MT_MODIFY) /* modify */
1433 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1434 else
1435 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1436
1437 return 0;
769} 1438}
770 1439
771/*****************************************************************************/ 1440/*****************************************************************************/
772 1441
1442#define ALLOC(len) \
1443 if (!req->ptr2) \
1444 { \
1445 X_LOCK (wrklock); \
1446 req->flags |= EIO_FLAG_PTR2_FREE; \
1447 X_UNLOCK (wrklock); \
1448 req->ptr2 = malloc (len); \
1449 if (!req->ptr2) \
1450 { \
1451 errno = ENOMEM; \
1452 req->result = -1; \
1453 break; \
1454 } \
1455 }
1456
773X_THREAD_PROC (eio_proc) 1457X_THREAD_PROC (etp_proc)
774{ 1458{
775 eio_req *req; 1459 ETP_REQ *req;
776 struct timespec ts; 1460 struct timespec ts;
777 worker *self = (worker *)thr_arg; 1461 etp_worker *self = (etp_worker *)thr_arg;
778 1462
779 /* try to distribute timeouts somewhat randomly */ 1463 /* try to distribute timeouts somewhat randomly */
780 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1464 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781 1465
782 for (;;) 1466 for (;;)
783 { 1467 {
784 ts.tv_sec = time (0) + IDLE_TIMEOUT;
785
786 X_LOCK (reqlock); 1468 X_LOCK (reqlock);
787 1469
788 for (;;) 1470 for (;;)
789 { 1471 {
790 self->req = req = reqq_shift (&req_queue); 1472 self->req = req = reqq_shift (&req_queue);
792 if (req) 1474 if (req)
793 break; 1475 break;
794 1476
795 ++idle; 1477 ++idle;
796 1478
1479 ts.tv_sec = time (0) + IDLE_TIMEOUT;
797 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1480 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
798 { 1481 {
799 if (idle > max_idle) 1482 if (idle > max_idle)
800 { 1483 {
801 --idle; 1484 --idle;
806 goto quit; 1489 goto quit;
807 } 1490 }
808 1491
809 /* we are allowed to idle, so do so without any timeout */ 1492 /* we are allowed to idle, so do so without any timeout */
810 X_COND_WAIT (reqwait, reqlock); 1493 X_COND_WAIT (reqwait, reqlock);
811 ts.tv_sec = time (0) + IDLE_TIMEOUT;
812 } 1494 }
813 1495
814 --idle; 1496 --idle;
815 } 1497 }
816 1498
817 --nready; 1499 --nready;
818 1500
819 X_UNLOCK (reqlock); 1501 X_UNLOCK (reqlock);
820 1502
821 errno = 0; /* strictly unnecessary */ 1503 if (req->type < 0)
1504 goto quit;
822 1505
823 if (!EIO_CANCELLED (req)) 1506 if (!EIO_CANCELLED (req))
824 switch (req->type) 1507 ETP_EXECUTE (self, req);
825 {
826 case EIO_READ: req->result = req->offs >= 0
827 ? pread (req->int1, req->ptr2, req->size, req->offs)
828 : read (req->int1, req->ptr2, req->size); break;
829 case EIO_WRITE: req->result = req->offs >= 0
830 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
831 : write (req->int1, req->ptr2, req->size); break;
832
833 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
834 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
835
836 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
837 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
838 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
839
840 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
841 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
842 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
843 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
844 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
845 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
846
847 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
848 case EIO_CLOSE: req->result = close (req->int1); break;
849 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
850 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
851 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
852 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
853 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
854 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
855 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
856 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
857 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
858
859 case EIO_SYNC: req->result = 0; sync (); break;
860 case EIO_FSYNC: req->result = fsync (req->int1); break;
861 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
862
863 case EIO_READDIR: scandir_ (req, self); break;
864
865 case EIO_BUSY:
866#ifdef _WIN32
867 Sleep (req->nv1 * 1000.);
868#else
869 {
870 struct timeval tv;
871
872 tv.tv_sec = req->nv1;
873 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
874
875 req->result = select (0, 0, 0, 0, &tv);
876 }
877#endif
878 break;
879
880 case EIO_UTIME:
881 case EIO_FUTIME:
882 {
883 struct timeval tv[2];
884 struct timeval *times;
885
886 if (req->nv1 != -1. || req->nv2 != -1.)
887 {
888 tv[0].tv_sec = req->nv1;
889 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
890 tv[1].tv_sec = req->nv2;
891 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
892
893 times = tv;
894 }
895 else
896 times = 0;
897
898
899 req->result = req->type == EIO_FUTIME
900 ? futimes (req->int1, times)
901 : utimes (req->ptr1, times);
902 }
903
904 case EIO_GROUP:
905 case EIO_NOP:
906 break;
907
908 case EIO_QUIT:
909 goto quit;
910
911 default:
912 req->result = -1;
913 break;
914 }
915
916 req->errorno = errno;
917 1508
918 X_LOCK (reslock); 1509 X_LOCK (reslock);
919 1510
920 ++npending; 1511 ++npending;
921 1512
922 if (!reqq_push (&res_queue, req) && want_poll_cb) 1513 if (!reqq_push (&res_queue, req) && want_poll_cb)
923 want_poll_cb (); 1514 want_poll_cb ();
924 1515
925 self->req = 0; 1516 self->req = 0;
926 worker_clear (self); 1517 etp_worker_clear (self);
927 1518
928 X_UNLOCK (reslock); 1519 X_UNLOCK (reslock);
929 } 1520 }
930 1521
931quit: 1522quit:
932 X_LOCK (wrklock); 1523 X_LOCK (wrklock);
933 worker_free (self); 1524 etp_worker_free (self);
934 X_UNLOCK (wrklock); 1525 X_UNLOCK (wrklock);
935 1526
936 return 0; 1527 return 0;
937} 1528}
938 1529
939/*****************************************************************************/ 1530/*****************************************************************************/
940 1531
941static void eio_atfork_prepare (void) 1532int eio_init (void (*want_poll)(void), void (*done_poll)(void))
942{ 1533{
943 X_LOCK (wrklock); 1534 return etp_init (want_poll, done_poll);
944 X_LOCK (reqlock);
945 X_LOCK (reslock);
946#if !HAVE_PREADWRITE
947 X_LOCK (preadwritelock);
948#endif
949#if !HAVE_READDIR_R
950 X_LOCK (readdirlock);
951#endif
952} 1535}
953 1536
954static void eio_atfork_parent (void) 1537static void eio_api_destroy (eio_req *req)
955{ 1538{
956#if !HAVE_READDIR_R 1539 free (req);
957 X_UNLOCK (readdirlock);
958#endif
959#if !HAVE_PREADWRITE
960 X_UNLOCK (preadwritelock);
961#endif
962 X_UNLOCK (reslock);
963 X_UNLOCK (reqlock);
964 X_UNLOCK (wrklock);
965} 1540}
966 1541
967static void eio_atfork_child (void) 1542#define REQ(rtype) \
968{ 1543 eio_req *req; \
969 eio_req *prv; 1544 \
1545 req = (eio_req *)calloc (1, sizeof *req); \
1546 if (!req) \
1547 return 0; \
1548 \
1549 req->type = rtype; \
1550 req->pri = pri; \
1551 req->finish = cb; \
1552 req->data = data; \
1553 req->destroy = eio_api_destroy;
970 1554
971 while (prv = reqq_shift (&req_queue)) 1555#define SEND eio_submit (req); return req
972 eio_destroy (prv);
973 1556
974 while (prv = reqq_shift (&res_queue)) 1557#define PATH \
975 eio_destroy (prv); 1558 req->flags |= EIO_FLAG_PTR1_FREE; \
1559 req->ptr1 = strdup (path); \
1560 if (!req->ptr1) \
1561 { \
1562 eio_api_destroy (req); \
1563 return 0; \
1564 }
976 1565
977 while (wrk_first.next != &wrk_first) 1566static void eio_execute (etp_worker *self, eio_req *req)
1567{
1568 errno = 0;
1569
1570 switch (req->type)
978 { 1571 {
979 worker *wrk = wrk_first.next; 1572 case EIO_READ: ALLOC (req->size);
1573 req->result = req->offs >= 0
1574 ? pread (req->int1, req->ptr2, req->size, req->offs)
1575 : read (req->int1, req->ptr2, req->size); break;
1576 case EIO_WRITE: req->result = req->offs >= 0
1577 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1578 : write (req->int1, req->ptr2, req->size); break;
980 1579
981 if (wrk->req) 1580 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
982 eio_destroy (wrk->req); 1581 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
983 1582
984 worker_clear (wrk); 1583 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
985 worker_free (wrk); 1584 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
986 } 1585 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1586 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1587 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1588 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
987 1589
988 started = 0; 1590 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
989 idle = 0; 1591 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
990 nreqs = 0; 1592 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
991 nready = 0; 1593 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
992 npending = 0;
993 1594
994 eio_atfork_parent (); 1595 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
995} 1596 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1597 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1598 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1599 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1600 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
996 1601
997int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1602 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
998{ 1603 case EIO_CLOSE: req->result = close (req->int1); break;
999 want_poll_cb = want_poll; 1604 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1000 done_poll_cb = done_poll; 1605 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1606 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1607 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1608 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1609 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1610 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1611 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1001 1612
1613 case EIO_READLINK: ALLOC (NAME_MAX);
1614 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1615
1616 case EIO_SYNC: req->result = 0; sync (); break;
1617 case EIO_FSYNC: req->result = fsync (req->int1); break;
1618 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1619 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1620 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1621 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1622
1623 case EIO_READDIR: eio__scandir (req, self); break;
1624
1625 case EIO_BUSY:
1002#ifdef _WIN32 1626#ifdef _WIN32
1003 X_MUTEX_CHECK (wrklock); 1627 Sleep (req->nv1 * 1000.);
1004 X_MUTEX_CHECK (reslock); 1628#else
1005 X_MUTEX_CHECK (reqlock);
1006 X_MUTEX_CHECK (reqwait);
1007 X_MUTEX_CHECK (preadwritelock);
1008 X_MUTEX_CHECK (readdirlock);
1009
1010 X_COND_CHECK (reqwait);
1011#endif
1012
1013 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1014}
1015
1016#if 0
1017
1018eio_req *eio_fsync (int fd, eio_cb cb);
1019eio_req *eio_fdatasync (int fd, eio_cb cb);
1020eio_req *eio_dupclose (int fd, eio_cb cb);
1021eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb);
1022eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1023eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1024eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1025eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb);
1026eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb);
1027eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb);
1028eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb);
1029eio_req *eio_dup2 (int fd, int fd2, eio_cb cb);
1030eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb);
1031eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb);
1032eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1033eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1034eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1035eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb);
1036eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb);
1037eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb);
1038eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb);
1039eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb);
1040eio_req *eio_unlink (const char *path, eio_cb cb);
1041eio_req *eio_rmdir (const char *path, eio_cb cb);
1042eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1043eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb);
1044eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */
1045eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */
1046void
1047aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1048 PROTOTYPE: $$$;$
1049 PPCODE:
1050{
1051 dREQ;
1052
1053 req->type = EIO_OPEN;
1054 req->sv1 = newSVsv (pathname);
1055 req->ptr1 = SvPVbyte_nolen (req->sv1);
1056 req->int1 = flags;
1057 req->int2 = mode;
1058
1059 EIO_SEND;
1060}
1061
1062void
1063aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1064 PROTOTYPE: $;$
1065 ALIAS:
1066 aio_fsync = EIO_FSYNC
1067 aio_fdatasync = EIO_FDATASYNC
1068 PPCODE:
1069{
1070 dREQ;
1071
1072 req->type = ix;
1073 req->sv1 = newSVsv (fh);
1074 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1075
1076 EIO_SEND (req);
1077}
1078
1079void
1080aio_close (SV *fh, SV *callback=&PL_sv_undef)
1081 PROTOTYPE: $;$
1082 PPCODE:
1083{
1084 dREQ;
1085
1086 req->type = EIO_CLOSE;
1087 req->sv1 = newSVsv (fh);
1088 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1089
1090 EIO_SEND (req);
1091}
1092
1093void
1094aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1095 ALIAS:
1096 aio_read = EIO_READ
1097 aio_write = EIO_WRITE
1098 PROTOTYPE: $$$$$;$
1099 PPCODE:
1100{
1101 STRLEN svlen;
1102 char *svptr = SvPVbyte (data, svlen);
1103 UV len = SvUV (length);
1104
1105 SvUPGRADE (data, SVt_PV);
1106 SvPOK_on (data);
1107
1108 if (dataoffset < 0)
1109 dataoffset += svlen;
1110
1111 if (dataoffset < 0 || dataoffset > svlen)
1112 croak ("dataoffset outside of data scalar");
1113
1114 if (ix == EIO_WRITE)
1115 {
1116 /* write: check length and adjust. */
1117 if (!SvOK (length) || len + dataoffset > svlen)
1118 len = svlen - dataoffset;
1119 }
1120 else
1121 {
1122 /* read: grow scalar as necessary */
1123 svptr = SvGROW (data, len + dataoffset + 1);
1124 }
1125
1126 if (len < 0)
1127 croak ("length must not be negative");
1128
1129 { 1629 {
1130 dREQ; 1630 struct timeval tv;
1131 1631
1132 req->type = ix; 1632 tv.tv_sec = req->nv1;
1133 req->sv1 = newSVsv (fh); 1633 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
1134 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1135 : IoOFP (sv_2io (fh)));
1136 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1137 req->size = len;
1138 req->sv2 = SvREFCNT_inc (data);
1139 req->ptr2 = (char *)svptr + dataoffset;
1140 req->stroffset = dataoffset;
1141 1634
1142 if (!SvREADONLY (data)) 1635 req->result = select (0, 0, 0, 0, &tv);
1636 }
1637#endif
1638 break;
1639
1640 case EIO_UTIME:
1641 case EIO_FUTIME:
1642 {
1643 struct timeval tv[2];
1644 struct timeval *times;
1645
1646 if (req->nv1 != -1. || req->nv2 != -1.)
1143 { 1647 {
1144 SvREADONLY_on (data); 1648 tv[0].tv_sec = req->nv1;
1145 req->flags |= FLAG_SV2_RO_OFF; 1649 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1650 tv[1].tv_sec = req->nv2;
1651 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1652
1653 times = tv;
1146 } 1654 }
1655 else
1656 times = 0;
1147 1657
1148 EIO_SEND; 1658
1659 req->result = req->type == EIO_FUTIME
1660 ? futimes (req->int1, times)
1661 : utimes (req->ptr1, times);
1149 } 1662 }
1150} 1663 break;
1151 1664
1152void 1665 case EIO_GROUP:
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef) 1666 abort (); /* handled in eio_request */
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{
1157 SV *data;
1158 dREQ;
1159 1667
1160 data = newSV (NAME_MAX); 1668 case EIO_NOP:
1161 SvPOK_on (data); 1669 req->result = 0;
1670 break;
1162 1671
1163 req->type = EIO_READLINK; 1672 case EIO_CUSTOM:
1164 req->sv1 = newSVsv (path); 1673 ((void (*)(eio_req *))req->feed) (req);
1165 req->ptr1 = SvPVbyte_nolen (req->sv1); 1674 break;
1166 req->sv2 = data;
1167 req->ptr2 = SvPVbyte_nolen (data);
1168 1675
1169 EIO_SEND; 1676 default:
1170} 1677 req->result = -1;
1678 break;
1679 }
1171 1680
1172void 1681 req->errorno = errno;
1173aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174 PROTOTYPE: $$$$;$
1175 PPCODE:
1176{
1177 dREQ;
1178
1179 req->type = EIO_SENDFILE;
1180 req->sv1 = newSVsv (out_fh);
1181 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182 req->sv2 = newSVsv (in_fh);
1183 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184 req->offs = SvVAL64 (in_offset);
1185 req->size = length;
1186
1187 EIO_SEND;
1188} 1682}
1189 1683
1190void 1684#ifndef EIO_NO_WRAPPERS
1191aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192 PROTOTYPE: $$$;$
1193 PPCODE:
1194{
1195 dREQ;
1196 1685
1197 req->type = EIO_READAHEAD; 1686eio_req *eio_nop (int pri, eio_cb cb, void *data)
1198 req->sv1 = newSVsv (fh);
1199 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1200 req->offs = SvVAL64 (offset);
1201 req->size = length;
1202
1203 EIO_SEND;
1204}
1205
1206void
1207aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1208 ALIAS:
1209 aio_stat = EIO_STAT
1210 aio_lstat = EIO_LSTAT
1211 PPCODE:
1212{ 1687{
1213 dREQ; 1688 REQ (EIO_NOP); SEND;
1689}
1214 1690
1215 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT)); 1691eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1216 if (!req->ptr2) 1692{
1217 { 1693 REQ (EIO_BUSY); req->nv1 = delay; SEND;
1218 req_destroy (req); 1694}
1219 croak ("out of memory during aio_stat statdata allocation");
1220 }
1221 1695
1696eio_req *eio_sync (int pri, eio_cb cb, void *data)
1697{
1698 REQ (EIO_SYNC); SEND;
1699}
1700
1701eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1702{
1703 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1704}
1705
1706eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1707{
1708 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1709}
1710
1711eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1712{
1713 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1714}
1715
1716eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1717{
1718 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1719}
1720
1721eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1722{
1723 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1724}
1725
1726eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1727{
1728 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1729}
1730
1731eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1732{
1733 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1734}
1735
1736eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1737{
1738 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1739}
1740
1741eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1742{
1743 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1744}
1745
1746eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1747{
1748 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1749}
1750
1751eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1752{
1753 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1754}
1755
1756eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1757{
1758 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1759}
1760
1761eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1762{
1763 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1764}
1765
1766eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1767{
1768 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1769}
1770
1771eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1772{
1773 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1774}
1775
1776eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1777{
1778 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1779}
1780
1781eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1782{
1783 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1784}
1785
1786eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1787{
1788 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1789}
1790
1791eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1792{
1793 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1794}
1795
1796eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1797{
1798 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1799}
1800
1801eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1802{
1803 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1804}
1805
1806eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1807{
1808 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1809}
1810
1811eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1812{
1813 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1814}
1815
1816static eio_req *
1817eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1818{
1819 REQ (type); PATH; SEND;
1820}
1821
1822eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1823{
1824 return eio__1path (EIO_READLINK, path, pri, cb, data);
1825}
1826
1827eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1828{
1829 return eio__1path (EIO_STAT, path, pri, cb, data);
1830}
1831
1832eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1833{
1834 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1835}
1836
1837eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1838{
1839 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1840}
1841
1842eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1843{
1844 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1845}
1846
1847eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1848{
1849 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1850}
1851
1852eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1853{
1854 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1855}
1856
1857eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1858{
1859 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1860}
1861
1862static eio_req *
1863eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1864{
1865 REQ (type); PATH;
1866
1222 req->flags |= FLAG_PTR2_FREE; 1867 req->flags |= EIO_FLAG_PTR2_FREE;
1223 req->sv1 = newSVsv (fh_or_path); 1868 req->ptr2 = strdup (new_path);
1869 if (!req->ptr2)
1870 {
1871 eio_api_destroy (req);
1872 return 0;
1873 }
1224 1874
1225 if (SvPOK (fh_or_path)) 1875 SEND;
1226 {
1227 req->type = ix;
1228 req->ptr1 = SvPVbyte_nolen (req->sv1);
1229 }
1230 else
1231 {
1232 req->type = EIO_FSTAT;
1233 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1234 }
1235
1236 EIO_SEND;
1237} 1876}
1238 1877
1239void 1878eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1240aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1241 PPCODE:
1242{ 1879{
1243 dREQ; 1880 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
1244
1245 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1246 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1247 req->sv1 = newSVsv (fh_or_path);
1248
1249 if (SvPOK (fh_or_path))
1250 {
1251 req->type = EIO_UTIME;
1252 req->ptr1 = SvPVbyte_nolen (req->sv1);
1253 }
1254 else
1255 {
1256 req->type = EIO_FUTIME;
1257 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1258 }
1259
1260 EIO_SEND;
1261} 1881}
1262 1882
1263void 1883eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1264aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1265 PPCODE:
1266{ 1884{
1267 dREQ; 1885 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
1268
1269 req->sv1 = newSVsv (fh_or_path);
1270 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1271
1272 if (SvPOK (fh_or_path))
1273 {
1274 req->type = EIO_TRUNCATE;
1275 req->ptr1 = SvPVbyte_nolen (req->sv1);
1276 }
1277 else
1278 {
1279 req->type = EIO_FTRUNCATE;
1280 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1281 }
1282
1283 EIO_SEND;
1284} 1886}
1285 1887
1286void 1888eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1287aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1288 ALIAS:
1289 aio_chmod = EIO_CHMOD
1290 aio_fchmod = EIO_FCHMOD
1291 aio_mkdir = EIO_MKDIR
1292 PPCODE:
1293{ 1889{
1294 dREQ; 1890 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1295
1296 req->type = type;
1297 req->int2 = mode;
1298 req->sv1 = newSVsv (fh_or_path);
1299
1300 if (ix == EIO_FCHMOD)
1301 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1302 else
1303 req->ptr1 = SvPVbyte_nolen (req->sv1);
1304
1305 EIO_SEND;
1306} 1891}
1307 1892
1308void 1893eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1309aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1310 PPCODE:
1311{ 1894{
1312 dREQ; 1895 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1313
1314 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1315 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1316 req->sv1 = newSVsv (fh_or_path);
1317
1318 if (SvPOK (fh_or_path))
1319 {
1320 req->type = EIO_CHOWN;
1321 req->ptr1 = SvPVbyte_nolen (req->sv1);
1322 }
1323 else
1324 {
1325 req->type = EIO_FCHOWN;
1326 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1327 }
1328
1329 EIO_SEND;
1330} 1896}
1331 1897
1332void
1333aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1334 ALIAS:
1335 aio_unlink = EIO_UNLINK
1336 aio_rmdir = EIO_RMDIR
1337 aio_readdir = EIO_READDIR
1338 PPCODE:
1339{
1340 dREQ;
1341
1342 req->type = ix;
1343 req->sv1 = newSVsv (pathname);
1344 req->ptr1 = SvPVbyte_nolen (req->sv1);
1345
1346 EIO_SEND;
1347}
1348
1349void
1350aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1351 ALIAS:
1352 aio_link = EIO_LINK
1353 aio_symlink = EIO_SYMLINK
1354 aio_rename = EIO_RENAME
1355 PPCODE:
1356{
1357 dREQ;
1358
1359 req->type = ix;
1360 req->sv1 = newSVsv (oldpath);
1361 req->ptr1 = SvPVbyte_nolen (req->sv1);
1362 req->sv2 = newSVsv (newpath);
1363 req->ptr2 = SvPVbyte_nolen (req->sv2);
1364
1365 EIO_SEND;
1366}
1367
1368void
1369aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1370 PPCODE:
1371{
1372 dREQ;
1373
1374 req->type = EIO_MKNOD;
1375 req->sv1 = newSVsv (pathname);
1376 req->ptr1 = SvPVbyte_nolen (req->sv1);
1377 req->int2 = (mode_t)mode;
1378 req->offs = dev;
1379
1380 EIO_SEND;
1381}
1382
1383void
1384aio_busy (double delay, SV *callback=&PL_sv_undef)
1385 PPCODE:
1386{
1387 dREQ;
1388
1389 req->type = EIO_BUSY;
1390 req->nv1 = delay < 0. ? 0. : delay;
1391
1392 EIO_SEND;
1393}
1394
1395void
1396aio_group (SV *callback=&PL_sv_undef)
1397 PROTOTYPE: ;$
1398 PPCODE:
1399{
1400 dREQ;
1401
1402 req->type = EIO_GROUP;
1403
1404 req_send (req);
1405 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1406}
1407
1408void
1409aio_nop (SV *callback=&PL_sv_undef)
1410 ALIAS:
1411 aio_nop = EIO_NOP
1412 aio_sync = EIO_SYNC
1413 PPCODE:
1414{
1415 dREQ;
1416
1417#endif 1898#endif
1899
1900eio_req *eio_grp (eio_cb cb, void *data)
1901{
1902 const int pri = EIO_PRI_MAX;
1903
1904 REQ (EIO_GROUP); SEND;
1905}
1906
1907#undef REQ
1908#undef PATH
1909#undef SEND
1910
1911/*****************************************************************************/
1912/* grp functions */
1418 1913
1419void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 1914void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 1915{
1421 grp->int2 = limit; 1916 grp->int2 = limit;
1422 grp->feed = feed; 1917 grp->feed = feed;
1433 1928
1434void eio_grp_add (eio_req *grp, eio_req *req) 1929void eio_grp_add (eio_req *grp, eio_req *req)
1435{ 1930{
1436 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1931 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1437 1932
1933 grp->flags |= EIO_FLAG_GROUPADD;
1934
1438 ++grp->size; 1935 ++grp->size;
1439 req->grp = grp; 1936 req->grp = grp;
1440 1937
1441 req->grp_prev = 0; 1938 req->grp_prev = 0;
1442 req->grp_next = grp->grp_first; 1939 req->grp_next = grp->grp_first;
1445 grp->grp_first->grp_prev = req; 1942 grp->grp_first->grp_prev = req;
1446 1943
1447 grp->grp_first = req; 1944 grp->grp_first = req;
1448} 1945}
1449 1946
1947/*****************************************************************************/
1948/* misc garbage */
1450 1949
1950ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1951{
1952 etp_worker wrk;
1953 ssize_t ret;
1954
1955 wrk.dbuf = 0;
1956
1957 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1958
1959 if (wrk.dbuf)
1960 free (wrk.dbuf);
1961
1962 return ret;
1963}
1964

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines