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 1.57 by root, Mon Sep 13 01:59:05 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>
40# include <unistd.h> 79# include <unistd.h>
41# include <utime.h> 80# include <utime.h>
42# include <signal.h> 81# include <signal.h>
82# include <dirent.h>
83
84#if _POSIX_MEMLOCK || _POSIX_MAPPED_FILES
85# include <sys/mman.h>
86#endif
87
88/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
89# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
90# define _DIRENT_HAVE_D_TYPE /* sigh */
91# define D_INO(de) (de)->d_fileno
92# define D_NAMLEN(de) (de)->d_namlen
93# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
94# define D_INO(de) (de)->d_ino
95# endif
96
97#ifdef _D_EXACT_NAMLEN
98# undef D_NAMLEN
99# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
100#endif
101
102# ifdef _DIRENT_HAVE_D_TYPE
103# define D_TYPE(de) (de)->d_type
104# endif
43 105
44# ifndef EIO_STRUCT_DIRENT 106# ifndef EIO_STRUCT_DIRENT
45# define EIO_STRUCT_DIRENT struct dirent 107# define EIO_STRUCT_DIRENT struct dirent
46# endif 108# endif
47 109
48#endif 110#endif
49 111
50# ifndef EIO_STRUCT_STAT
51# define EIO_STRUCT_STAT struct stat
52# endif
53
54#if HAVE_SENDFILE 112#if HAVE_SENDFILE
55# if __linux 113# if __linux
56# include <sys/sendfile.h> 114# include <sys/sendfile.h>
57# elif __freebsd 115# elif __FreeBSD__ || defined __APPLE__
58# include <sys/socket.h> 116# include <sys/socket.h>
59# include <sys/uio.h> 117# include <sys/uio.h>
60# elif __hpux 118# elif __hpux
61# include <sys/socket.h> 119# include <sys/socket.h>
62# elif __solaris /* not yet */ 120# elif __solaris
63# include <sys/sendfile.h> 121# include <sys/sendfile.h>
64# else 122# else
65# error sendfile support requested but not available 123# error sendfile support requested but not available
66# endif 124# endif
67#endif 125#endif
68 126
127#ifndef D_TYPE
128# define D_TYPE(de) 0
129#endif
130#ifndef D_INO
131# define D_INO(de) 0
132#endif
133#ifndef D_NAMLEN
134# define D_NAMLEN(de) strlen ((de)->d_name)
135#endif
136
69/* number of seconds after which an idle threads exit */ 137/* number of seconds after which an idle threads exit */
70#define IDLE_TIMEOUT 10 138#define IDLE_TIMEOUT 10
71 139
72/* used for struct dirent, AIX doesn't provide it */ 140/* used for struct dirent, AIX doesn't provide it */
73#ifndef NAME_MAX 141#ifndef NAME_MAX
74# define NAME_MAX 4096 142# define NAME_MAX 4096
75#endif 143#endif
76 144
145/* used for readlink etc. */
146#ifndef PATH_MAX
147# define PATH_MAX 4096
148#endif
149
77/* buffer size for various temporary buffers */ 150/* buffer size for various temporary buffers */
78#define EIO_BUFSIZE 65536 151#define EIO_BUFSIZE 65536
79 152
80#define dBUF \ 153#define dBUF \
81 char *eio_buf; \ 154 char *eio_buf; \
82 X_LOCK (wrklock); \ 155 ETP_WORKER_LOCK (self); \
83 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 156 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
84 X_UNLOCK (wrklock); \ 157 ETP_WORKER_UNLOCK (self); \
158 errno = ENOMEM; \
85 if (!eio_buf) \ 159 if (!eio_buf) \
86 return -1; 160 return -1;
87 161
88#define EIO_TICKS ((1000000 + 1023) >> 10) 162#define EIO_TICKS ((1000000 + 1023) >> 10)
89 163
164/*****************************************************************************/
165
166#if __GNUC__ >= 3
167# define expect(expr,value) __builtin_expect ((expr),(value))
168#else
169# define expect(expr,value) (expr)
170#endif
171
172#define expect_false(expr) expect ((expr) != 0, 0)
173#define expect_true(expr) expect ((expr) != 0, 1)
174
175/*****************************************************************************/
176
177#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX
179
180struct etp_worker;
181
182#define ETP_REQ eio_req
183#define ETP_DESTROY(req) eio_destroy (req)
184static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
188
189#define ETP_WORKER_CLEAR(req) \
190 if (wrk->dbuf) \
191 { \
192 free (wrk->dbuf); \
193 wrk->dbuf = 0; \
194 } \
195 \
196 if (wrk->dirp) \
197 { \
198 closedir (wrk->dirp); \
199 wrk->dirp = 0; \
200 }
201
202#define ETP_WORKER_COMMON \
203 void *dbuf; \
204 DIR *dirp;
205
206/*****************************************************************************/
207
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209
210/* calculate time difference in ~1/EIO_TICKS of a second */
211static int tvdiff (struct timeval *tv1, struct timeval *tv2)
212{
213 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
214 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
215}
216
217static unsigned int started, idle, wanted = 4;
218
90static void (*want_poll_cb) (void); 219static void (*want_poll_cb) (void);
91static void (*done_poll_cb) (void); 220static void (*done_poll_cb) (void);
92 221
93static unsigned int max_poll_time = 0; 222static unsigned int max_poll_time; /* reslock */
94static unsigned int max_poll_reqs = 0; 223static unsigned int max_poll_reqs; /* reslock */
95 224
96/* calculcate time difference in ~1/EIO_TICKS of a second */ 225static volatile unsigned int nreqs; /* reqlock */
97static int tvdiff (struct timeval *tv1, struct timeval *tv2) 226static volatile unsigned int nready; /* reqlock */
98{ 227static volatile unsigned int npending; /* reqlock */
99 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 228static volatile unsigned int max_idle = 4;
100 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
101}
102 229
103static unsigned int started, idle, wanted; 230static xmutex_t wrklock = X_MUTEX_INIT;
231static xmutex_t reslock = X_MUTEX_INIT;
232static xmutex_t reqlock = X_MUTEX_INIT;
233static xcond_t reqwait = X_COND_INIT;
234
235#if !HAVE_PREADWRITE
236/*
237 * make our pread/pwrite emulation safe against themselves, but not against
238 * normal read/write by using a mutex. slows down execution a lot,
239 * but that's your problem, not mine.
240 */
241static xmutex_t preadwritelock = X_MUTEX_INIT;
242#endif
243
244typedef struct etp_worker
245{
246 /* locked by wrklock */
247 struct etp_worker *prev, *next;
248
249 xthread_t tid;
250
251 /* locked by reslock, reqlock or wrklock */
252 ETP_REQ *req; /* currently processed request */
253
254 ETP_WORKER_COMMON
255} etp_worker;
256
257static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
258
259#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
260#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
104 261
105/* worker threads management */ 262/* worker threads management */
106static mutex_t wrklock = X_MUTEX_INIT;
107 263
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) 264static void etp_worker_clear (etp_worker *wrk)
123{ 265{
124 if (wrk->dirp) 266 ETP_WORKER_CLEAR (wrk);
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} 267}
136 268
137static void worker_free (worker *wrk) 269static void etp_worker_free (etp_worker *wrk)
138{ 270{
139 wrk->next->prev = wrk->prev; 271 wrk->next->prev = wrk->prev;
140 wrk->prev->next = wrk->next; 272 wrk->prev->next = wrk->next;
141 273
142 free (wrk); 274 free (wrk);
143} 275}
144 276
145static volatile unsigned int nreqs, nready, npending;
146static volatile unsigned int max_idle = 4;
147
148static mutex_t reslock = X_MUTEX_INIT;
149static mutex_t reqlock = X_MUTEX_INIT;
150static cond_t reqwait = X_COND_INIT;
151
152unsigned int eio_nreqs (void) 277static unsigned int etp_nreqs (void)
153{ 278{
279 int retval;
280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281 retval = nreqs;
282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 283 return retval;
155} 284}
156 285
157unsigned int eio_nready (void) 286static unsigned int etp_nready (void)
158{ 287{
159 unsigned int retval; 288 unsigned int retval;
160 289
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 291 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 293
165 return retval; 294 return retval;
166} 295}
167 296
168unsigned int eio_npending (void) 297static unsigned int etp_npending (void)
169{ 298{
170 unsigned int retval; 299 unsigned int retval;
171 300
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 301 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 302 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 303 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 304
176 return retval; 305 return retval;
177} 306}
178 307
179unsigned int eio_nthreads (void) 308static unsigned int etp_nthreads (void)
180{ 309{
181 unsigned int retval; 310 unsigned int retval;
182 311
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 312 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 313 retval = started;
… …
191 * a somewhat faster data structure might be nice, but 320 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 321 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 322 * per shift, the most expensive operation.
194 */ 323 */
195typedef struct { 324typedef struct {
196 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 325 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197 int size; 326 int size;
198} reqq; 327} etp_reqq;
199 328
200static reqq req_queue; 329static etp_reqq req_queue;
201static reqq res_queue; 330static etp_reqq res_queue;
202 331
203static int reqq_push (reqq *q, eio_req *req) 332static int reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 333{
205 int pri = req->pri; 334 int pri = req->pri;
206 req->next = 0; 335 req->next = 0;
207 336
208 if (q->qe[pri]) 337 if (q->qe[pri])
… …
214 q->qe[pri] = q->qs[pri] = req; 343 q->qe[pri] = q->qs[pri] = req;
215 344
216 return q->size++; 345 return q->size++;
217} 346}
218 347
219static eio_req *reqq_shift (reqq *q) 348static ETP_REQ *reqq_shift (etp_reqq *q)
220{ 349{
221 int pri; 350 int pri;
222 351
223 if (!q->size) 352 if (!q->size)
224 return 0; 353 return 0;
225 354
226 --q->size; 355 --q->size;
227 356
228 for (pri = EIO_NUM_PRI; pri--; ) 357 for (pri = ETP_NUM_PRI; pri--; )
229 { 358 {
230 eio_req *req = q->qs[pri]; 359 eio_req *req = q->qs[pri];
231 360
232 if (req) 361 if (req)
233 { 362 {
… …
239 } 368 }
240 369
241 abort (); 370 abort ();
242} 371}
243 372
244static void grp_try_feed (eio_req *grp) 373static void etp_atfork_prepare (void)
245{ 374{
246 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 375 X_LOCK (wrklock);
376 X_LOCK (reqlock);
377 X_LOCK (reslock);
378#if !HAVE_PREADWRITE
379 X_LOCK (preadwritelock);
380#endif
381}
382
383static void etp_atfork_parent (void)
384{
385#if !HAVE_PREADWRITE
386 X_UNLOCK (preadwritelock);
387#endif
388 X_UNLOCK (reslock);
389 X_UNLOCK (reqlock);
390 X_UNLOCK (wrklock);
391}
392
393static void etp_atfork_child (void)
394{
395 ETP_REQ *prv;
396
397 while ((prv = reqq_shift (&req_queue)))
398 ETP_DESTROY (prv);
399
400 while ((prv = reqq_shift (&res_queue)))
401 ETP_DESTROY (prv);
402
403 while (wrk_first.next != &wrk_first)
247 { 404 {
248 int old_len = grp->size; 405 etp_worker *wrk = wrk_first.next;
249 406
250 EIO_FEED (grp); 407 if (wrk->req)
408 ETP_DESTROY (wrk->req);
251 409
252 /* stop if no progress has been made */ 410 etp_worker_clear (wrk);
253 if (old_len == grp->size) 411 etp_worker_free (wrk);
254 {
255 grp->feed = 0;
256 break;
257 }
258 } 412 }
259}
260 413
261static int eio_finish (eio_req *req); 414 started = 0;
415 idle = 0;
416 nreqs = 0;
417 nready = 0;
418 npending = 0;
262 419
263static int grp_dec (eio_req *grp) 420 etp_atfork_parent ();
264{ 421}
265 --grp->size;
266 422
267 /* call feeder, if applicable */ 423static void
268 grp_try_feed (grp); 424etp_once_init (void)
425{
426 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
427}
269 428
270 /* finish, if done */ 429static int
271 if (!grp->size && grp->int1) 430etp_init (void (*want_poll)(void), void (*done_poll)(void))
272 return eio_finish (grp); 431{
273 else 432 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
433
434 pthread_once (&doinit, etp_once_init);
435
436 want_poll_cb = want_poll;
437 done_poll_cb = done_poll;
438
274 return 0; 439 return 0;
275} 440}
276 441
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); 442X_THREAD_PROC (etp_proc);
326 443
327static void start_thread (void) 444static void etp_start_thread (void)
328{ 445{
329 worker *wrk = calloc (1, sizeof (worker)); 446 etp_worker *wrk = calloc (1, sizeof (etp_worker));
330 447
331 if (!wrk) 448 /*TODO*/
332 croak ("unable to allocate worker thread data"); 449 assert (("unable to allocate worker thread data", wrk));
333 450
334 X_LOCK (wrklock); 451 X_LOCK (wrklock);
335 452
336 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 453 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
337 { 454 {
338 wrk->prev = &wrk_first; 455 wrk->prev = &wrk_first;
339 wrk->next = wrk_first.next; 456 wrk->next = wrk_first.next;
340 wrk_first.next->prev = wrk; 457 wrk_first.next->prev = wrk;
341 wrk_first.next = wrk; 458 wrk_first.next = wrk;
… …
345 free (wrk); 462 free (wrk);
346 463
347 X_UNLOCK (wrklock); 464 X_UNLOCK (wrklock);
348} 465}
349 466
350static void maybe_start_thread (void) 467static void etp_maybe_start_thread (void)
351{ 468{
352 if (eio_nthreads () >= wanted) 469 if (expect_true (etp_nthreads () >= wanted))
353 return; 470 return;
354 471
355 /* todo: maybe use idle here, but might be less exact */ 472 /* todo: maybe use idle here, but might be less exact */
356 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 473 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
357 return; 474 return;
358 475
359 start_thread (); 476 etp_start_thread ();
360} 477}
361 478
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) 479static void etp_end_thread (void)
376{ 480{
377 eio_req *req = calloc (1, sizeof (eio_req)); 481 eio_req *req = calloc (1, sizeof (eio_req));
378 482
379 req->type = EIO_QUIT; 483 req->type = -1;
380 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 484 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381 485
382 X_LOCK (reqlock); 486 X_LOCK (reqlock);
383 reqq_push (&req_queue, req); 487 reqq_push (&req_queue, req);
384 X_COND_SIGNAL (reqwait); 488 X_COND_SIGNAL (reqwait);
385 X_UNLOCK (reqlock); 489 X_UNLOCK (reqlock);
… …
387 X_LOCK (wrklock); 491 X_LOCK (wrklock);
388 --started; 492 --started;
389 X_UNLOCK (wrklock); 493 X_UNLOCK (wrklock);
390} 494}
391 495
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) 496static int etp_poll (void)
429{ 497{
430 int maxreqs = max_poll_reqs; 498 unsigned int maxreqs;
499 unsigned int maxtime;
431 struct timeval tv_start, tv_now; 500 struct timeval tv_start, tv_now;
432 eio_req *req;
433 501
502 X_LOCK (reslock);
503 maxreqs = max_poll_reqs;
504 maxtime = max_poll_time;
505 X_UNLOCK (reslock);
506
434 if (max_poll_time) 507 if (maxtime)
435 gettimeofday (&tv_start, 0); 508 gettimeofday (&tv_start, 0);
436 509
437 for (;;) 510 for (;;)
438 { 511 {
512 ETP_REQ *req;
513
439 maybe_start_thread (); 514 etp_maybe_start_thread ();
440 515
441 X_LOCK (reslock); 516 X_LOCK (reslock);
442 req = reqq_shift (&res_queue); 517 req = reqq_shift (&res_queue);
443 518
444 if (req) 519 if (req)
… …
452 X_UNLOCK (reslock); 527 X_UNLOCK (reslock);
453 528
454 if (!req) 529 if (!req)
455 return 0; 530 return 0;
456 531
532 X_LOCK (reqlock);
457 --nreqs; 533 --nreqs;
534 X_UNLOCK (reqlock);
458 535
459 if (req->type == EIO_GROUP && req->size) 536 if (expect_false (req->type == EIO_GROUP && req->size))
460 { 537 {
461 req->int1 = 1; /* mark request as delayed */ 538 req->int1 = 1; /* mark request as delayed */
462 continue; 539 continue;
463 } 540 }
464 else 541 else
465 { 542 {
466 int res = eio_finish (req); 543 int res = ETP_FINISH (req);
467 if (res) 544 if (expect_false (res))
468 return res; 545 return res;
469 } 546 }
470 547
471 if (maxreqs && !--maxreqs) 548 if (expect_false (maxreqs && !--maxreqs))
472 break; 549 break;
473 550
474 if (max_poll_time) 551 if (maxtime)
475 { 552 {
476 gettimeofday (&tv_now, 0); 553 gettimeofday (&tv_now, 0);
477 554
478 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 555 if (tvdiff (&tv_start, &tv_now) >= maxtime)
479 break; 556 break;
480 } 557 }
481 } 558 }
482 559
483 errno = EAGAIN; 560 errno = EAGAIN;
484 return -1; 561 return -1;
485} 562}
486 563
564static void etp_cancel (ETP_REQ *req)
565{
566 X_LOCK (wrklock);
567 req->flags |= EIO_FLAG_CANCELLED;
568 X_UNLOCK (wrklock);
569
570 eio_grp_cancel (req);
571}
572
573static void etp_submit (ETP_REQ *req)
574{
575 req->pri -= ETP_PRI_MIN;
576
577 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
578 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
579
580 if (expect_false (req->type == EIO_GROUP))
581 {
582 /* I hope this is worth it :/ */
583 X_LOCK (reqlock);
584 ++nreqs;
585 X_UNLOCK (reqlock);
586
587 X_LOCK (reslock);
588
589 ++npending;
590
591 if (!reqq_push (&res_queue, req) && want_poll_cb)
592 want_poll_cb ();
593
594 X_UNLOCK (reslock);
595 }
596 else
597 {
598 X_LOCK (reqlock);
599 ++nreqs;
600 ++nready;
601 reqq_push (&req_queue, req);
602 X_COND_SIGNAL (reqwait);
603 X_UNLOCK (reqlock);
604
605 etp_maybe_start_thread ();
606 }
607}
608
609static void etp_set_max_poll_time (double nseconds)
610{
611 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
612 max_poll_time = nseconds * EIO_TICKS;
613 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
614}
615
616static void etp_set_max_poll_reqs (unsigned int maxreqs)
617{
618 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
619 max_poll_reqs = maxreqs;
620 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
621}
622
623static void etp_set_max_idle (unsigned int nthreads)
624{
625 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
626 max_idle = nthreads <= 0 ? 1 : nthreads;
627 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
628}
629
630static void etp_set_min_parallel (unsigned int nthreads)
631{
632 if (wanted < nthreads)
633 wanted = nthreads;
634}
635
636static void etp_set_max_parallel (unsigned int nthreads)
637{
638 if (wanted > nthreads)
639 wanted = nthreads;
640
641 while (started > wanted)
642 etp_end_thread ();
643}
644
645/*****************************************************************************/
646
647static void grp_try_feed (eio_req *grp)
648{
649 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
650 {
651 grp->flags &= ~EIO_FLAG_GROUPADD;
652
653 EIO_FEED (grp);
654
655 /* stop if no progress has been made */
656 if (!(grp->flags & EIO_FLAG_GROUPADD))
657 {
658 grp->feed = 0;
659 break;
660 }
661 }
662}
663
664static int grp_dec (eio_req *grp)
665{
666 --grp->size;
667
668 /* call feeder, if applicable */
669 grp_try_feed (grp);
670
671 /* finish, if done */
672 if (!grp->size && grp->int1)
673 return eio_finish (grp);
674 else
675 return 0;
676}
677
678void eio_destroy (eio_req *req)
679{
680 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
681 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
682
683 EIO_DESTROY (req);
684}
685
686static int eio_finish (eio_req *req)
687{
688 int res = EIO_FINISH (req);
689
690 if (req->grp)
691 {
692 int res2;
693 eio_req *grp = req->grp;
694
695 /* unlink request */
696 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
697 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
698
699 if (grp->grp_first == req)
700 grp->grp_first = req->grp_next;
701
702 res2 = grp_dec (grp);
703
704 if (!res && res2)
705 res = res2;
706 }
707
708 eio_destroy (req);
709
710 return res;
711}
712
713void eio_grp_cancel (eio_req *grp)
714{
715 for (grp = grp->grp_first; grp; grp = grp->grp_next)
716 eio_cancel (grp);
717}
718
719void eio_cancel (eio_req *req)
720{
721 etp_cancel (req);
722}
723
724void eio_submit (eio_req *req)
725{
726 etp_submit (req);
727}
728
729unsigned int eio_nreqs (void)
730{
731 return etp_nreqs ();
732}
733
734unsigned int eio_nready (void)
735{
736 return etp_nready ();
737}
738
739unsigned int eio_npending (void)
740{
741 return etp_npending ();
742}
743
744unsigned int eio_nthreads (void)
745{
746 return etp_nthreads ();
747}
748
749void eio_set_max_poll_time (double nseconds)
750{
751 etp_set_max_poll_time (nseconds);
752}
753
754void eio_set_max_poll_reqs (unsigned int maxreqs)
755{
756 etp_set_max_poll_reqs (maxreqs);
757}
758
759void eio_set_max_idle (unsigned int nthreads)
760{
761 etp_set_max_idle (nthreads);
762}
763
764void eio_set_min_parallel (unsigned int nthreads)
765{
766 etp_set_min_parallel (nthreads);
767}
768
769void eio_set_max_parallel (unsigned int nthreads)
770{
771 etp_set_max_parallel (nthreads);
772}
773
774int eio_poll (void)
775{
776 return etp_poll ();
777}
778
487/*****************************************************************************/ 779/*****************************************************************************/
488/* work around various missing functions */ 780/* work around various missing functions */
489 781
490#if !HAVE_PREADWRITE 782#if !HAVE_PREADWRITE
783# undef pread
784# undef pwrite
491# define pread aio_pread 785# define pread eio__pread
492# define pwrite aio_pwrite 786# define pwrite eio__pwrite
493 787
494/* 788static ssize_t
495 * make our pread/pwrite safe against themselves, but not against 789eio__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{ 790{
503 ssize_t res; 791 ssize_t res;
504 off_t ooffset; 792 off_t ooffset;
505 793
506 X_LOCK (preadwritelock); 794 X_LOCK (preadwritelock);
… …
511 X_UNLOCK (preadwritelock); 799 X_UNLOCK (preadwritelock);
512 800
513 return res; 801 return res;
514} 802}
515 803
516static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 804static ssize_t
805eio__pwrite (int fd, void *buf, size_t count, off_t offset)
517{ 806{
518 ssize_t res; 807 ssize_t res;
519 off_t ooffset; 808 off_t ooffset;
520 809
521 X_LOCK (preadwritelock); 810 X_LOCK (preadwritelock);
522 ooffset = lseek (fd, 0, SEEK_CUR); 811 ooffset = lseek (fd, 0, SEEK_CUR);
523 lseek (fd, offset, SEEK_SET); 812 lseek (fd, offset, SEEK_SET);
524 res = write (fd, buf, count); 813 res = write (fd, buf, count);
525 lseek (fd, offset, SEEK_SET); 814 lseek (fd, ooffset, SEEK_SET);
526 X_UNLOCK (preadwritelock); 815 X_UNLOCK (preadwritelock);
527 816
528 return res; 817 return res;
529} 818}
530#endif 819#endif
531 820
532#ifndef HAVE_FUTIMES 821#ifndef HAVE_FUTIMES
533 822
823# undef utimes
824# undef futimes
534# define utimes(path,times) aio_utimes (path, times) 825# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times) 826# define futimes(fd,times) eio__futimes (fd, times)
536 827
828static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 829eio__utimes (const char *filename, const struct timeval times[2])
538{ 830{
539 if (times) 831 if (times)
540 { 832 {
541 struct utimbuf buf; 833 struct utimbuf buf;
542 834
… …
547 } 839 }
548 else 840 else
549 return utime (filename, 0); 841 return utime (filename, 0);
550} 842}
551 843
552static int aio_futimes (int fd, const struct timeval tv[2]) 844static int eio__futimes (int fd, const struct timeval tv[2])
553{ 845{
554 errno = ENOSYS; 846 errno = ENOSYS;
555 return -1; 847 return -1;
556} 848}
557 849
558#endif 850#endif
559 851
560#if !HAVE_FDATASYNC 852#if !HAVE_FDATASYNC
853# undef fdatasync
561# define fdatasync fsync 854# define fdatasync(fd) fsync (fd)
562#endif 855#endif
856
857/* sync_file_range always needs emulation */
858int
859eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
860{
861#if HAVE_SYNC_FILE_RANGE
862 int res;
863
864 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
865 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
866 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
867 {
868 flags = 0
869 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
870 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
871 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
872 }
873
874 res = sync_file_range (fd, offset, nbytes, flags);
875
876 if (!res || errno != ENOSYS)
877 return res;
878#endif
879
880 /* even though we could play tricks with the flags, it's better to always
881 * call fdatasync, as that matches the expectation of its users best */
882 return fdatasync (fd);
883}
563 884
564#if !HAVE_READAHEAD 885#if !HAVE_READAHEAD
886# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 887# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 888
889static ssize_t
567static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 890eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568{ 891{
569 size_t todo = count; 892 size_t todo = count;
570 dBUF; 893 dBUF;
571 894
572 while (todo > 0) 895 while (todo > 0)
573 { 896 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 897 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 898
576 pread (fd, aio_buf, len, offset); 899 pread (fd, eio_buf, len, offset);
577 offset += len; 900 offset += len;
578 todo -= len; 901 todo -= len;
579 } 902 }
580 903
581 errno = 0; 904 errno = 0;
582 return count; 905 return count;
583} 906}
584 907
585#endif 908#endif
586 909
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 */ 910/* sendfile always needs emulation */
911static ssize_t
618static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 912eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
619{ 913{
620 ssize_t res; 914 ssize_t res;
621 915
622 if (!count) 916 if (!count)
623 return 0; 917 return 0;
624 918
625#if HAVE_SENDFILE 919#if HAVE_SENDFILE
626# if __linux 920# if __linux
627 res = sendfile (ofd, ifd, &offset, count); 921 res = sendfile (ofd, ifd, &offset, count);
628 922
629# elif __freebsd 923# elif __FreeBSD__
630 /* 924 /*
631 * Of course, the freebsd sendfile is a dire hack with no thoughts 925 * Of course, the freebsd sendfile is a dire hack with no thoughts
632 * wasted on making it similar to other I/O functions. 926 * wasted on making it similar to other I/O functions.
633 */ 927 */
634 { 928 {
635 off_t sbytes; 929 off_t sbytes;
636 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 930 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637 931
638 if (res < 0 && sbytes) 932 #if 0 /* according to the manpage, this is correct, but broken behaviour */
639 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 933 /* freebsd' sendfile will return 0 on success */
934 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
935 /* not on e.g. EIO or EPIPE - sounds broken */
936 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
937 res = sbytes;
938 #endif
939
940 /* according to source inspection, this is correct, and useful behaviour */
941 if (sbytes)
942 res = sbytes;
943 }
944
945# elif defined (__APPLE__)
946
947 {
948 off_t sbytes = count;
949 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
950
951 /* according to the manpage, sbytes is always valid */
952 if (sbytes)
640 res = sbytes; 953 res = sbytes;
641 } 954 }
642 955
643# elif __hpux 956# elif __hpux
644 res = sendfile (ofd, ifd, offset, count, 0, 0); 957 res = sendfile (ofd, ifd, offset, count, 0, 0);
… …
658 if (res < 0 && sbytes) 971 if (res < 0 && sbytes)
659 res = sbytes; 972 res = sbytes;
660 } 973 }
661 974
662# endif 975# endif
976
977#elif defined (_WIN32)
978
979 /* does not work, just for documentation of what would need to be done */
980 {
981 HANDLE h = TO_SOCKET (ifd);
982 SetFilePointer (h, offset, 0, FILE_BEGIN);
983 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
984 }
985
663#else 986#else
664 res = -1; 987 res = -1;
665 errno = ENOSYS; 988 errno = ENOSYS;
666#endif 989#endif
667 990
668 if (res < 0 991 if (res < 0
669 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 992 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
993 /* BSDs */
994#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
995 || errno == ENOTSUP
996#endif
997 || errno == EOPNOTSUPP /* BSDs */
670#if __solaris 998#if __solaris
671 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 999 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
672#endif 1000#endif
673 ) 1001 )
674 ) 1002 )
… …
705 } 1033 }
706 1034
707 return res; 1035 return res;
708} 1036}
709 1037
1038static signed char
1039eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1040{
1041 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1042 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1043}
1044
1045#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1046
1047#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1048#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1049
1050static void
1051eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1052{
1053 unsigned char bits [9 + sizeof (ino_t) * 8];
1054 unsigned char *bit = bits;
1055
1056 assert (CHAR_BIT == 8);
1057 assert (sizeof (eio_dirent) * 8 < 256);
1058 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1059 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1060
1061 if (size <= EIO_SORT_FAST)
1062 return;
1063
1064 /* first prepare an array of bits to test in our radix sort */
1065 /* try to take endianness into account, as well as differences in ino_t sizes */
1066 /* inode_bits must contain all inodes ORed together */
1067 /* which is used to skip bits that are 0 everywhere, which is very common */
1068 {
1069 ino_t endianness;
1070 int i, j;
1071
1072 /* we store the byte offset of byte n into byte n of "endianness" */
1073 for (i = 0; i < sizeof (ino_t); ++i)
1074 ((unsigned char *)&endianness)[i] = i;
1075
1076 *bit++ = 0;
1077
1078 for (i = 0; i < sizeof (ino_t); ++i)
1079 {
1080 /* shifting off the byte offsets out of "endianness" */
1081 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1082 endianness >>= 8;
1083
1084 for (j = 0; j < 8; ++j)
1085 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1086 *bit++ = offs + j;
1087 }
1088
1089 for (j = 0; j < 8; ++j)
1090 if (score_bits & (1 << j))
1091 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1092 }
1093
1094 /* now actually do the sorting (a variant of MSD radix sort) */
1095 {
1096 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1097 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1098 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1099 int stk_idx = 0;
1100
1101 base_stk [stk_idx] = dents;
1102 end_stk [stk_idx] = dents + size;
1103 bit_stk [stk_idx] = bit - 1;
1104
1105 do
1106 {
1107 base = base_stk [stk_idx];
1108 end = end_stk [stk_idx];
1109 bit = bit_stk [stk_idx];
1110
1111 for (;;)
1112 {
1113 unsigned char O = *bit >> 3;
1114 unsigned char M = 1 << (*bit & 7);
1115
1116 eio_dirent *a = base;
1117 eio_dirent *b = end;
1118
1119 if (b - a < EIO_SORT_CUTOFF)
1120 break;
1121
1122 /* now bit-partition the array on the bit */
1123 /* this ugly asymmetric loop seems to perform much better than typical */
1124 /* partition algos found in the literature */
1125 do
1126 if (!(((unsigned char *)a)[O] & M))
1127 ++a;
1128 else if (!(((unsigned char *)--b)[O] & M))
1129 {
1130 eio_dirent tmp = *a; *a = *b; *b = tmp;
1131 ++a;
1132 }
1133 while (b > a);
1134
1135 /* next bit, or stop, if no bits left in this path */
1136 if (!*--bit)
1137 break;
1138
1139 base_stk [stk_idx] = a;
1140 end_stk [stk_idx] = end;
1141 bit_stk [stk_idx] = bit;
1142 ++stk_idx;
1143
1144 end = a;
1145 }
1146 }
1147 while (stk_idx--);
1148 }
1149}
1150
1151static void
1152eio_dent_insertion_sort (eio_dirent *dents, int size)
1153{
1154 /* first move the smallest element to the front, to act as a sentinel */
1155 {
1156 int i;
1157 eio_dirent *min = dents;
1158
1159 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1160 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1161 if (EIO_DENT_CMP (dents [i], <, *min))
1162 min = &dents [i];
1163
1164 /* swap elements 0 and j (minimum) */
1165 {
1166 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1167 }
1168 }
1169
1170 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1171 {
1172 eio_dirent *i, *j;
1173
1174 for (i = dents + 1; i < dents + size; ++i)
1175 {
1176 eio_dirent value = *i;
1177
1178 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1179 j [1] = j [0];
1180
1181 j [1] = value;
1182 }
1183 }
1184}
1185
1186static void
1187eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1188{
1189 if (size <= 1)
1190 return; /* our insertion sort relies on size > 0 */
1191
1192 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1193 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1194 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1195
1196 /* use an insertion sort at the end, or for small arrays, */
1197 /* as insertion sort is more efficient for small partitions */
1198 eio_dent_insertion_sort (dents, size);
1199}
1200
710/* read a full directory */ 1201/* read a full directory */
1202static void
711static void scandir_ (eio_req *req, worker *self) 1203eio__scandir (eio_req *req, etp_worker *self)
712{ 1204{
713 DIR *dirp; 1205 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; 1206 EIO_STRUCT_DIRENT *entp;
720 char *name, *names; 1207 char *name, *names;
721 int memlen = 4096; 1208 int namesalloc = 4096;
722 int memofs = 0; 1209 int namesoffs = 0;
1210 int flags = req->int1;
1211 eio_dirent *dents = 0;
1212 int dentalloc = 128;
723 int res = 0; 1213 int dentoffs = 0;
1214 ino_t inode_bits = 0;
1215
1216 req->result = -1;
1217
1218 if (!(flags & EIO_READDIR_DENTS))
1219 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
724 1220
725 X_LOCK (wrklock); 1221 X_LOCK (wrklock);
1222 /* the corresponding closedir is in ETP_WORKER_CLEAR */
726 self->dirp = dirp = opendir (req->ptr1); 1223 self->dirp = dirp = opendir (req->ptr1);
727 self->dbuf = u = malloc (sizeof (*u));
728 req->flags |= EIO_FLAG_PTR2_FREE; 1224 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1225 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
729 req->ptr2 = names = malloc (memlen); 1226 req->ptr2 = names = malloc (namesalloc);
730 X_UNLOCK (wrklock); 1227 X_UNLOCK (wrklock);
731 1228
732 if (dirp && u && names) 1229 if (dirp && names && (!flags || dents))
733 for (;;) 1230 for (;;)
734 { 1231 {
735 errno = 0; 1232 errno = 0;
736 readdir_r (dirp, &u->d, &entp); 1233 entp = readdir (dirp);
737 1234
738 if (!entp) 1235 if (!entp)
1236 {
1237 if (errno)
1238 break;
1239
1240 /* sort etc. */
1241 req->int1 = flags;
1242 req->result = dentoffs;
1243
1244 if (flags & EIO_READDIR_STAT_ORDER)
1245 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
1246 else if (flags & EIO_READDIR_DIRS_FIRST)
1247 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1248 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1249 else
1250 {
1251 /* in this case, all is known, and we just put dirs first and sort them */
1252 eio_dirent *oth = dents + dentoffs;
1253 eio_dirent *dir = dents;
1254
1255 /* now partition dirs to the front, and non-dirs to the back */
1256 /* by walking from both sides and swapping if necessary */
1257 /* also clear score, so it doesn't influence sorting */
1258 while (oth > dir)
1259 {
1260 if (dir->type == EIO_DT_DIR)
1261 ++dir;
1262 else if ((--oth)->type == EIO_DT_DIR)
1263 {
1264 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1265
1266 ++dir;
1267 }
1268 }
1269
1270 /* now sort the dirs only */
1271 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1272 }
1273
739 break; 1274 break;
1275 }
740 1276
1277 /* now add the entry to our list(s) */
741 name = entp->d_name; 1278 name = entp->d_name;
742 1279
1280 /* skip . and .. entries */
743 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1281 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
744 { 1282 {
745 int len = strlen (name) + 1; 1283 int len = D_NAMLEN (entp) + 1;
746 1284
747 res++; 1285 while (expect_false (namesoffs + len > namesalloc))
748
749 while (memofs + len > memlen)
750 { 1286 {
751 memlen *= 2; 1287 namesalloc *= 2;
752 X_LOCK (wrklock); 1288 X_LOCK (wrklock);
753 req->ptr2 = names = realloc (names, memlen); 1289 req->ptr2 = names = realloc (names, namesalloc);
754 X_UNLOCK (wrklock); 1290 X_UNLOCK (wrklock);
755 1291
756 if (!names) 1292 if (!names)
757 break; 1293 break;
758 } 1294 }
759 1295
760 memcpy (names + memofs, name, len); 1296 memcpy (names + namesoffs, name, len);
1297
1298 if (dents)
1299 {
1300 struct eio_dirent *ent;
1301
1302 if (expect_false (dentoffs == dentalloc))
1303 {
1304 dentalloc *= 2;
1305 X_LOCK (wrklock);
1306 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1307 X_UNLOCK (wrklock);
1308
1309 if (!dents)
1310 break;
1311 }
1312
1313 ent = dents + dentoffs;
1314
1315 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1316 ent->namelen = len - 1;
1317 ent->inode = D_INO (entp);
1318
1319 inode_bits |= ent->inode;
1320
1321 switch (D_TYPE (entp))
1322 {
1323 default:
1324 ent->type = EIO_DT_UNKNOWN;
1325 flags |= EIO_READDIR_FOUND_UNKNOWN;
1326 break;
1327
1328 #ifdef DT_FIFO
1329 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1330 #endif
1331 #ifdef DT_CHR
1332 case DT_CHR: ent->type = EIO_DT_CHR; break;
1333 #endif
1334 #ifdef DT_MPC
1335 case DT_MPC: ent->type = EIO_DT_MPC; break;
1336 #endif
1337 #ifdef DT_DIR
1338 case DT_DIR: ent->type = EIO_DT_DIR; break;
1339 #endif
1340 #ifdef DT_NAM
1341 case DT_NAM: ent->type = EIO_DT_NAM; break;
1342 #endif
1343 #ifdef DT_BLK
1344 case DT_BLK: ent->type = EIO_DT_BLK; break;
1345 #endif
1346 #ifdef DT_MPB
1347 case DT_MPB: ent->type = EIO_DT_MPB; break;
1348 #endif
1349 #ifdef DT_REG
1350 case DT_REG: ent->type = EIO_DT_REG; break;
1351 #endif
1352 #ifdef DT_NWK
1353 case DT_NWK: ent->type = EIO_DT_NWK; break;
1354 #endif
1355 #ifdef DT_CMP
1356 case DT_CMP: ent->type = EIO_DT_CMP; break;
1357 #endif
1358 #ifdef DT_LNK
1359 case DT_LNK: ent->type = EIO_DT_LNK; break;
1360 #endif
1361 #ifdef DT_SOCK
1362 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1363 #endif
1364 #ifdef DT_DOOR
1365 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1366 #endif
1367 #ifdef DT_WHT
1368 case DT_WHT: ent->type = EIO_DT_WHT; break;
1369 #endif
1370 }
1371
1372 ent->score = 7;
1373
1374 if (flags & EIO_READDIR_DIRS_FIRST)
1375 {
1376 if (ent->type == EIO_DT_UNKNOWN)
1377 {
1378 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1379 ent->score = 1;
1380 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1381 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1382 }
1383 else if (ent->type == EIO_DT_DIR)
1384 ent->score = 0;
1385 }
1386 }
1387
761 memofs += len; 1388 namesoffs += len;
1389 ++dentoffs;
1390 }
1391
1392 if (EIO_CANCELLED (req))
1393 {
1394 errno = ECANCELED;
1395 break;
762 } 1396 }
763 } 1397 }
1398}
764 1399
765 if (errno) 1400#ifdef PAGESIZE
766 res = -1; 1401# define eio_pagesize() PAGESIZE
1402#else
1403static intptr_t
1404eio_pagesize (void)
1405{
1406 static intptr_t page;
1407
1408 if (!page)
1409 page = sysconf (_SC_PAGESIZE);
1410
1411 return page;
1412}
1413#endif
1414
1415static void
1416eio_page_align (void **addr, size_t *length)
1417{
1418 intptr_t mask = eio_pagesize () - 1;
1419
1420 /* round down addr */
1421 intptr_t adj = mask & (intptr_t)*addr;
1422
1423 *addr = (void *)((intptr_t)*addr - adj);
1424 *length += adj;
1425
1426 /* round up length */
1427 *length = (*length + mask) & ~mask;
1428}
1429
1430#if !_POSIX_MEMLOCK
1431# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1432# define eio__mlockall(a) ((errno = ENOSYS), -1)
1433#else
1434
1435static int
1436eio__mlock (void *addr, size_t length)
1437{
1438 eio_page_align (&addr, &length);
1439
1440 return mlock (addr, length);
1441}
1442
1443static int
1444eio__mlockall (int flags)
1445{
1446 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1447 extern int mallopt (int, int);
1448 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1449 #endif
1450
1451 if (EIO_MCL_CURRENT != MCL_CURRENT
1452 || EIO_MCL_FUTURE != MCL_FUTURE)
1453 {
1454 flags = 0
1455 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1456 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1457 }
1458
1459 return mlockall (flags);
1460}
1461#endif
1462
1463#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1464# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1465#else
1466
1467int
1468eio__msync (void *mem, size_t len, int flags)
1469{
1470 eio_page_align (&mem, &len);
1471
1472 if (EIO_MS_ASYNC != MS_SYNC
1473 || EIO_MS_INVALIDATE != MS_INVALIDATE
1474 || EIO_MS_SYNC != MS_SYNC)
1475 {
1476 flags = 0
1477 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1478 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1479 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1480 }
1481
1482 return msync (mem, len, flags);
1483}
1484
1485#endif
1486
1487int
1488eio__mtouch (void *mem, size_t len, int flags)
1489{
1490 eio_page_align (&mem, &len);
1491
767 1492 {
768 req->result = res; 1493 intptr_t addr = (intptr_t)mem;
1494 intptr_t end = addr + len;
1495 intptr_t page = eio_pagesize ();
1496
1497 if (addr < end)
1498 if (flags & EIO_MT_MODIFY) /* modify */
1499 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1500 else
1501 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1502 }
1503
1504 return 0;
769} 1505}
770 1506
771/*****************************************************************************/ 1507/*****************************************************************************/
772 1508
1509#define ALLOC(len) \
1510 if (!req->ptr2) \
1511 { \
1512 X_LOCK (wrklock); \
1513 req->flags |= EIO_FLAG_PTR2_FREE; \
1514 X_UNLOCK (wrklock); \
1515 req->ptr2 = malloc (len); \
1516 if (!req->ptr2) \
1517 { \
1518 errno = ENOMEM; \
1519 req->result = -1; \
1520 break; \
1521 } \
1522 }
1523
773X_THREAD_PROC (eio_proc) 1524X_THREAD_PROC (etp_proc)
774{ 1525{
775 eio_req *req; 1526 ETP_REQ *req;
776 struct timespec ts; 1527 struct timespec ts;
777 worker *self = (worker *)thr_arg; 1528 etp_worker *self = (etp_worker *)thr_arg;
778 1529
779 /* try to distribute timeouts somewhat randomly */ 1530 /* try to distribute timeouts somewhat randomly */
780 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1531 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781 1532
782 for (;;) 1533 for (;;)
783 { 1534 {
784 ts.tv_sec = time (0) + IDLE_TIMEOUT;
785
786 X_LOCK (reqlock); 1535 X_LOCK (reqlock);
787 1536
788 for (;;) 1537 for (;;)
789 { 1538 {
790 self->req = req = reqq_shift (&req_queue); 1539 self->req = req = reqq_shift (&req_queue);
… …
792 if (req) 1541 if (req)
793 break; 1542 break;
794 1543
795 ++idle; 1544 ++idle;
796 1545
1546 ts.tv_sec = time (0) + IDLE_TIMEOUT;
797 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1547 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
798 { 1548 {
799 if (idle > max_idle) 1549 if (idle > max_idle)
800 { 1550 {
801 --idle; 1551 --idle;
… …
806 goto quit; 1556 goto quit;
807 } 1557 }
808 1558
809 /* we are allowed to idle, so do so without any timeout */ 1559 /* we are allowed to idle, so do so without any timeout */
810 X_COND_WAIT (reqwait, reqlock); 1560 X_COND_WAIT (reqwait, reqlock);
811 ts.tv_sec = time (0) + IDLE_TIMEOUT;
812 } 1561 }
813 1562
814 --idle; 1563 --idle;
815 } 1564 }
816 1565
817 --nready; 1566 --nready;
818 1567
819 X_UNLOCK (reqlock); 1568 X_UNLOCK (reqlock);
820 1569
821 errno = 0; /* strictly unnecessary */ 1570 if (req->type < 0)
1571 goto quit;
822 1572
823 if (!EIO_CANCELLED (req)) 1573 if (!EIO_CANCELLED (req))
824 switch (req->type) 1574 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 1575
918 X_LOCK (reslock); 1576 X_LOCK (reslock);
919 1577
920 ++npending; 1578 ++npending;
921 1579
922 if (!reqq_push (&res_queue, req) && want_poll_cb) 1580 if (!reqq_push (&res_queue, req) && want_poll_cb)
923 want_poll_cb (); 1581 want_poll_cb ();
924 1582
925 self->req = 0; 1583 self->req = 0;
926 worker_clear (self); 1584 etp_worker_clear (self);
927 1585
928 X_UNLOCK (reslock); 1586 X_UNLOCK (reslock);
929 } 1587 }
930 1588
931quit: 1589quit:
932 X_LOCK (wrklock); 1590 X_LOCK (wrklock);
933 worker_free (self); 1591 etp_worker_free (self);
934 X_UNLOCK (wrklock); 1592 X_UNLOCK (wrklock);
935 1593
936 return 0; 1594 return 0;
937} 1595}
938 1596
939/*****************************************************************************/ 1597/*****************************************************************************/
940 1598
941static void eio_atfork_prepare (void) 1599int eio_init (void (*want_poll)(void), void (*done_poll)(void))
942{ 1600{
943 X_LOCK (wrklock); 1601 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} 1602}
953 1603
954static void eio_atfork_parent (void) 1604static void eio_api_destroy (eio_req *req)
955{ 1605{
956#if !HAVE_READDIR_R 1606 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} 1607}
966 1608
967static void eio_atfork_child (void) 1609#define REQ(rtype) \
968{ 1610 eio_req *req; \
969 eio_req *prv; 1611 \
1612 req = (eio_req *)calloc (1, sizeof *req); \
1613 if (!req) \
1614 return 0; \
1615 \
1616 req->type = rtype; \
1617 req->pri = pri; \
1618 req->finish = cb; \
1619 req->data = data; \
1620 req->destroy = eio_api_destroy;
970 1621
971 while (prv = reqq_shift (&req_queue)) 1622#define SEND eio_submit (req); return req
972 eio_destroy (prv);
973 1623
974 while (prv = reqq_shift (&res_queue)) 1624#define PATH \
975 eio_destroy (prv); 1625 req->flags |= EIO_FLAG_PTR1_FREE; \
1626 req->ptr1 = strdup (path); \
1627 if (!req->ptr1) \
1628 { \
1629 eio_api_destroy (req); \
1630 return 0; \
1631 }
976 1632
977 while (wrk_first.next != &wrk_first) 1633static void eio_execute (etp_worker *self, eio_req *req)
1634{
1635 errno = 0;
1636
1637 switch (req->type)
978 { 1638 {
979 worker *wrk = wrk_first.next; 1639 case EIO_READ: ALLOC (req->size);
1640 req->result = req->offs >= 0
1641 ? pread (req->int1, req->ptr2, req->size, req->offs)
1642 : read (req->int1, req->ptr2, req->size); break;
1643 case EIO_WRITE: req->result = req->offs >= 0
1644 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1645 : write (req->int1, req->ptr2, req->size); break;
980 1646
981 if (wrk->req) 1647 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
982 eio_destroy (wrk->req); 1648 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
983 1649
984 worker_clear (wrk); 1650 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
985 worker_free (wrk); 1651 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
986 } 1652 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1653 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1654 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1655 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
987 1656
988 started = 0; 1657 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
989 idle = 0; 1658 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
990 nreqs = 0; 1659 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
991 nready = 0; 1660 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
992 npending = 0;
993 1661
994 eio_atfork_parent (); 1662 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
995} 1663 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1664 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1665 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1666 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1667 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
996 1668
997int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1669 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
998{ 1670 case EIO_CLOSE: req->result = close (req->int1); break;
999 want_poll_cb = want_poll; 1671 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1000 done_poll_cb = done_poll; 1672 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1673 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1674 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1675 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1676 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1677 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1678 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1001 1679
1680 case EIO_READLINK: ALLOC (PATH_MAX);
1681 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1682
1683 case EIO_SYNC: req->result = 0; sync (); break;
1684 case EIO_FSYNC: req->result = fsync (req->int1); break;
1685 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1686 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1687 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1688 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1689 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1690 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1691
1692 case EIO_READDIR: eio__scandir (req, self); break;
1693
1694 case EIO_BUSY:
1002#ifdef _WIN32 1695#ifdef _WIN32
1003 X_MUTEX_CHECK (wrklock); 1696 Sleep (req->nv1 * 1e3);
1004 X_MUTEX_CHECK (reslock); 1697#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 { 1698 {
1130 dREQ; 1699 struct timeval tv;
1131 1700
1132 req->type = ix; 1701 tv.tv_sec = req->nv1;
1133 req->sv1 = newSVsv (fh); 1702 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
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 1703
1142 if (!SvREADONLY (data)) 1704 req->result = select (0, 0, 0, 0, &tv);
1705 }
1706#endif
1707 break;
1708
1709 case EIO_UTIME:
1710 case EIO_FUTIME:
1711 {
1712 struct timeval tv[2];
1713 struct timeval *times;
1714
1715 if (req->nv1 != -1. || req->nv2 != -1.)
1143 { 1716 {
1144 SvREADONLY_on (data); 1717 tv[0].tv_sec = req->nv1;
1145 req->flags |= FLAG_SV2_RO_OFF; 1718 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1719 tv[1].tv_sec = req->nv2;
1720 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1721
1722 times = tv;
1146 } 1723 }
1724 else
1725 times = 0;
1147 1726
1148 EIO_SEND; 1727
1728 req->result = req->type == EIO_FUTIME
1729 ? futimes (req->int1, times)
1730 : utimes (req->ptr1, times);
1149 } 1731 }
1150} 1732 break;
1151 1733
1152void 1734 case EIO_GROUP:
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef) 1735 abort (); /* handled in eio_request */
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{
1157 SV *data;
1158 dREQ;
1159 1736
1160 data = newSV (NAME_MAX); 1737 case EIO_NOP:
1161 SvPOK_on (data); 1738 req->result = 0;
1739 break;
1162 1740
1163 req->type = EIO_READLINK; 1741 case EIO_CUSTOM:
1164 req->sv1 = newSVsv (path); 1742 ((void (*)(eio_req *))req->feed) (req);
1165 req->ptr1 = SvPVbyte_nolen (req->sv1); 1743 break;
1166 req->sv2 = data;
1167 req->ptr2 = SvPVbyte_nolen (data);
1168 1744
1169 EIO_SEND; 1745 default:
1170} 1746 req->result = -1;
1747 break;
1748 }
1171 1749
1172void 1750 req->errorno = errno;
1173aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174 PROTOTYPE: $$$$;$
1175 PPCODE:
1176{
1177 dREQ;
1178
1179 req->type = EIO_SENDFILE;
1180 req->sv1 = newSVsv (out_fh);
1181 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182 req->sv2 = newSVsv (in_fh);
1183 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184 req->offs = SvVAL64 (in_offset);
1185 req->size = length;
1186
1187 EIO_SEND;
1188} 1751}
1189 1752
1190void 1753#ifndef EIO_NO_WRAPPERS
1191aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192 PROTOTYPE: $$$;$
1193 PPCODE:
1194{
1195 dREQ;
1196 1754
1197 req->type = EIO_READAHEAD; 1755eio_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{ 1756{
1213 dREQ; 1757 REQ (EIO_NOP); SEND;
1758}
1214 1759
1215 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT)); 1760eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1216 if (!req->ptr2) 1761{
1217 { 1762 REQ (EIO_BUSY); req->nv1 = delay; SEND;
1218 req_destroy (req); 1763}
1219 croak ("out of memory during aio_stat statdata allocation");
1220 }
1221 1764
1765eio_req *eio_sync (int pri, eio_cb cb, void *data)
1766{
1767 REQ (EIO_SYNC); SEND;
1768}
1769
1770eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1771{
1772 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1773}
1774
1775eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1776{
1777 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1778}
1779
1780eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1781{
1782 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1783}
1784
1785eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1786{
1787 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1788}
1789
1790eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1791{
1792 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1793}
1794
1795eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1796{
1797 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1798}
1799
1800eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1801{
1802 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1803}
1804
1805eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1806{
1807 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1808}
1809
1810eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1811{
1812 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1813}
1814
1815eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1816{
1817 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1818}
1819
1820eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1821{
1822 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1823}
1824
1825eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1826{
1827 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1828}
1829
1830eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1831{
1832 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1833}
1834
1835eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1836{
1837 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1838}
1839
1840eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1841{
1842 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1843}
1844
1845eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1846{
1847 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1848}
1849
1850eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1851{
1852 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1853}
1854
1855eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1856{
1857 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1858}
1859
1860eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1861{
1862 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1863}
1864
1865eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1866{
1867 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1868}
1869
1870eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1871{
1872 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1873}
1874
1875eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1876{
1877 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1878}
1879
1880eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1881{
1882 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1883}
1884
1885eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1886{
1887 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1888}
1889
1890eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1891{
1892 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1893}
1894
1895static eio_req *
1896eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1897{
1898 REQ (type); PATH; SEND;
1899}
1900
1901eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1902{
1903 return eio__1path (EIO_READLINK, path, pri, cb, data);
1904}
1905
1906eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1907{
1908 return eio__1path (EIO_STAT, path, pri, cb, data);
1909}
1910
1911eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1912{
1913 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1914}
1915
1916eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1917{
1918 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1919}
1920
1921eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1922{
1923 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1924}
1925
1926eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1927{
1928 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1929}
1930
1931eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1932{
1933 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1934}
1935
1936eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1937{
1938 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1939}
1940
1941static eio_req *
1942eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1943{
1944 REQ (type); PATH;
1945
1222 req->flags |= FLAG_PTR2_FREE; 1946 req->flags |= EIO_FLAG_PTR2_FREE;
1223 req->sv1 = newSVsv (fh_or_path); 1947 req->ptr2 = strdup (new_path);
1948 if (!req->ptr2)
1949 {
1950 eio_api_destroy (req);
1951 return 0;
1952 }
1224 1953
1225 if (SvPOK (fh_or_path)) 1954 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} 1955}
1238 1956
1239void 1957eio_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{ 1958{
1243 dREQ; 1959 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} 1960}
1262 1961
1263void 1962eio_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{ 1963{
1267 dREQ; 1964 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} 1965}
1285 1966
1286void 1967eio_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{ 1968{
1294 dREQ; 1969 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} 1970}
1307 1971
1308void 1972eio_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{ 1973{
1312 dREQ; 1974 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} 1975}
1331 1976
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 1977#endif
1978
1979eio_req *eio_grp (eio_cb cb, void *data)
1980{
1981 const int pri = EIO_PRI_MAX;
1982
1983 REQ (EIO_GROUP); SEND;
1984}
1985
1986#undef REQ
1987#undef PATH
1988#undef SEND
1989
1990/*****************************************************************************/
1991/* grp functions */
1418 1992
1419void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 1993void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 1994{
1421 grp->int2 = limit; 1995 grp->int2 = limit;
1422 grp->feed = feed; 1996 grp->feed = feed;
… …
1433 2007
1434void eio_grp_add (eio_req *grp, eio_req *req) 2008void eio_grp_add (eio_req *grp, eio_req *req)
1435{ 2009{
1436 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2010 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1437 2011
2012 grp->flags |= EIO_FLAG_GROUPADD;
2013
1438 ++grp->size; 2014 ++grp->size;
1439 req->grp = grp; 2015 req->grp = grp;
1440 2016
1441 req->grp_prev = 0; 2017 req->grp_prev = 0;
1442 req->grp_next = grp->grp_first; 2018 req->grp_next = grp->grp_first;
… …
1445 grp->grp_first->grp_prev = req; 2021 grp->grp_first->grp_prev = req;
1446 2022
1447 grp->grp_first = req; 2023 grp->grp_first = req;
1448} 2024}
1449 2025
2026/*****************************************************************************/
2027/* misc garbage */
1450 2028
2029ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2030{
2031 etp_worker wrk;
2032 ssize_t ret;
2033
2034 wrk.dbuf = 0;
2035
2036 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2037
2038 if (wrk.dbuf)
2039 free (wrk.dbuf);
2040
2041 return ret;
2042}
2043

Diff Legend

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