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 1.63 by root, Thu May 26 03:51:55 2011 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL.
38 */
39
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_MEMLOCK_RANGE || _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
69/* number of seconds after which an idle threads exit */ 127#ifndef D_TYPE
70#define IDLE_TIMEOUT 10 128# 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
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
74# define NAME_MAX 4096 139# define NAME_MAX 4096
75#endif 140#endif
76 141
142/* used for readlink etc. */
143#ifndef PATH_MAX
144# define PATH_MAX 4096
145#endif
146
77/* buffer size for various temporary buffers */ 147/* buffer size for various temporary buffers */
78#define EIO_BUFSIZE 65536 148#define EIO_BUFSIZE 65536
79 149
80#define dBUF \ 150#define dBUF \
81 char *eio_buf; \ 151 char *eio_buf; \
82 X_LOCK (wrklock); \ 152 ETP_WORKER_LOCK (self); \
83 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 153 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
84 X_UNLOCK (wrklock); \ 154 ETP_WORKER_UNLOCK (self); \
155 errno = ENOMEM; \
85 if (!eio_buf) \ 156 if (!eio_buf) \
86 return -1; 157 return -1;
87 158
88#define EIO_TICKS ((1000000 + 1023) >> 10) 159#define EIO_TICKS ((1000000 + 1023) >> 10)
89 160
161/*****************************************************************************/
162
163#if __GNUC__ >= 3
164# define expect(expr,value) __builtin_expect ((expr),(value))
165#else
166# define expect(expr,value) (expr)
167#endif
168
169#define expect_false(expr) expect ((expr) != 0, 0)
170#define expect_true(expr) expect ((expr) != 0, 1)
171
172/*****************************************************************************/
173
174#define ETP_PRI_MIN EIO_PRI_MIN
175#define ETP_PRI_MAX EIO_PRI_MAX
176
177struct etp_worker;
178
179#define ETP_REQ eio_req
180#define ETP_DESTROY(req) eio_destroy (req)
181static int eio_finish (eio_req *req);
182#define ETP_FINISH(req) eio_finish (req)
183static void eio_execute (struct etp_worker *self, eio_req *req);
184#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
185
186#define ETP_WORKER_CLEAR(req) \
187 if (wrk->dbuf) \
188 { \
189 free (wrk->dbuf); \
190 wrk->dbuf = 0; \
191 } \
192 \
193 if (wrk->dirp) \
194 { \
195 closedir (wrk->dirp); \
196 wrk->dirp = 0; \
197 }
198
199#define ETP_WORKER_COMMON \
200 void *dbuf; \
201 DIR *dirp;
202
203/*****************************************************************************/
204
205#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
206
207/* calculate time difference in ~1/EIO_TICKS of a second */
208static int tvdiff (struct timeval *tv1, struct timeval *tv2)
209{
210 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
211 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
212}
213
214static unsigned int started, idle, wanted = 4;
215
90static void (*want_poll_cb) (void); 216static void (*want_poll_cb) (void);
91static void (*done_poll_cb) (void); 217static void (*done_poll_cb) (void);
92 218
93static unsigned int max_poll_time = 0; 219static unsigned int max_poll_time; /* reslock */
94static unsigned int max_poll_reqs = 0; 220static unsigned int max_poll_reqs; /* reslock */
95 221
96/* calculcate time difference in ~1/EIO_TICKS of a second */ 222static volatile unsigned int nreqs; /* reqlock */
97static int tvdiff (struct timeval *tv1, struct timeval *tv2) 223static volatile unsigned int nready; /* reqlock */
98{ 224static volatile unsigned int npending; /* reqlock */
99 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 225static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
100 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 226static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
101}
102 227
103static unsigned int started, idle, wanted; 228static xmutex_t wrklock;
229static xmutex_t reslock;
230static xmutex_t reqlock;
231static xcond_t reqwait;
232
233#if !HAVE_PREADWRITE
234/*
235 * make our pread/pwrite emulation safe against themselves, but not against
236 * normal read/write by using a mutex. slows down execution a lot,
237 * but that's your problem, not mine.
238 */
239static xmutex_t preadwritelock = X_MUTEX_INIT;
240#endif
241
242typedef struct etp_worker
243{
244 /* locked by wrklock */
245 struct etp_worker *prev, *next;
246
247 xthread_t tid;
248
249 /* locked by reslock, reqlock or wrklock */
250 ETP_REQ *req; /* currently processed request */
251
252 ETP_WORKER_COMMON
253} etp_worker;
254
255static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
256
257#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
258#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
104 259
105/* worker threads management */ 260/* worker threads management */
106static mutex_t wrklock = X_MUTEX_INIT;
107 261
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) 262static void etp_worker_clear (etp_worker *wrk)
123{ 263{
124 if (wrk->dirp) 264 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} 265}
136 266
137static void worker_free (worker *wrk) 267static void etp_worker_free (etp_worker *wrk)
138{ 268{
139 wrk->next->prev = wrk->prev; 269 wrk->next->prev = wrk->prev;
140 wrk->prev->next = wrk->next; 270 wrk->prev->next = wrk->next;
141 271
142 free (wrk); 272 free (wrk);
143} 273}
144 274
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) 275static unsigned int etp_nreqs (void)
153{ 276{
277 int retval;
278 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
279 retval = nreqs;
280 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 281 return retval;
155} 282}
156 283
157unsigned int eio_nready (void) 284static unsigned int etp_nready (void)
158{ 285{
159 unsigned int retval; 286 unsigned int retval;
160 287
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 288 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 289 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 290 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 291
165 return retval; 292 return retval;
166} 293}
167 294
168unsigned int eio_npending (void) 295static unsigned int etp_npending (void)
169{ 296{
170 unsigned int retval; 297 unsigned int retval;
171 298
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 299 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 300 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 301 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 302
176 return retval; 303 return retval;
177} 304}
178 305
179unsigned int eio_nthreads (void) 306static unsigned int etp_nthreads (void)
180{ 307{
181 unsigned int retval; 308 unsigned int retval;
182 309
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 310 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 311 retval = started;
… …
191 * a somewhat faster data structure might be nice, but 318 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 319 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 320 * per shift, the most expensive operation.
194 */ 321 */
195typedef struct { 322typedef struct {
196 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 323 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197 int size; 324 int size;
198} reqq; 325} etp_reqq;
199 326
200static reqq req_queue; 327static etp_reqq req_queue;
201static reqq res_queue; 328static etp_reqq res_queue;
202 329
203static int reqq_push (reqq *q, eio_req *req) 330static int reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 331{
205 int pri = req->pri; 332 int pri = req->pri;
206 req->next = 0; 333 req->next = 0;
207 334
208 if (q->qe[pri]) 335 if (q->qe[pri])
… …
214 q->qe[pri] = q->qs[pri] = req; 341 q->qe[pri] = q->qs[pri] = req;
215 342
216 return q->size++; 343 return q->size++;
217} 344}
218 345
219static eio_req *reqq_shift (reqq *q) 346static ETP_REQ *reqq_shift (etp_reqq *q)
220{ 347{
221 int pri; 348 int pri;
222 349
223 if (!q->size) 350 if (!q->size)
224 return 0; 351 return 0;
225 352
226 --q->size; 353 --q->size;
227 354
228 for (pri = EIO_NUM_PRI; pri--; ) 355 for (pri = ETP_NUM_PRI; pri--; )
229 { 356 {
230 eio_req *req = q->qs[pri]; 357 eio_req *req = q->qs[pri];
231 358
232 if (req) 359 if (req)
233 { 360 {
… …
239 } 366 }
240 367
241 abort (); 368 abort ();
242} 369}
243 370
244static void grp_try_feed (eio_req *grp) 371static void etp_thread_init (void)
245{ 372{
246 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 373 X_MUTEX_CREATE (wrklock);
374 X_MUTEX_CREATE (reslock);
375 X_MUTEX_CREATE (reqlock);
376 X_COND_CREATE (reqwait);
377}
378
379static void etp_atfork_prepare (void)
380{
381 X_LOCK (wrklock);
382 X_LOCK (reqlock);
383 X_LOCK (reslock);
384#if !HAVE_PREADWRITE
385 X_LOCK (preadwritelock);
386#endif
387}
388
389static void etp_atfork_parent (void)
390{
391#if !HAVE_PREADWRITE
392 X_UNLOCK (preadwritelock);
393#endif
394 X_UNLOCK (reslock);
395 X_UNLOCK (reqlock);
396 X_UNLOCK (wrklock);
397}
398
399static void etp_atfork_child (void)
400{
401 ETP_REQ *prv;
402
403 while ((prv = reqq_shift (&req_queue)))
404 ETP_DESTROY (prv);
405
406 while ((prv = reqq_shift (&res_queue)))
407 ETP_DESTROY (prv);
408
409 while (wrk_first.next != &wrk_first)
247 { 410 {
248 int old_len = grp->size; 411 etp_worker *wrk = wrk_first.next;
249 412
250 EIO_FEED (grp); 413 if (wrk->req)
414 ETP_DESTROY (wrk->req);
251 415
252 /* stop if no progress has been made */ 416 etp_worker_clear (wrk);
253 if (old_len == grp->size) 417 etp_worker_free (wrk);
254 {
255 grp->feed = 0;
256 break;
257 }
258 } 418 }
259}
260 419
261static int eio_finish (eio_req *req); 420 started = 0;
421 idle = 0;
422 nreqs = 0;
423 nready = 0;
424 npending = 0;
262 425
263static int grp_dec (eio_req *grp) 426 etp_thread_init ();
264{ 427}
265 --grp->size;
266 428
267 /* call feeder, if applicable */ 429static void
268 grp_try_feed (grp); 430etp_once_init (void)
431{
432 etp_thread_init ();
433 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
434}
269 435
270 /* finish, if done */ 436static int
271 if (!grp->size && grp->int1) 437etp_init (void (*want_poll)(void), void (*done_poll)(void))
272 return eio_finish (grp); 438{
273 else 439 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
440
441 pthread_once (&doinit, etp_once_init);
442
443 want_poll_cb = want_poll;
444 done_poll_cb = done_poll;
445
274 return 0; 446 return 0;
275} 447}
276 448
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); 449X_THREAD_PROC (etp_proc);
326 450
327static void start_thread (void) 451static void etp_start_thread (void)
328{ 452{
329 worker *wrk = calloc (1, sizeof (worker)); 453 etp_worker *wrk = calloc (1, sizeof (etp_worker));
330 454
331 if (!wrk) 455 /*TODO*/
332 croak ("unable to allocate worker thread data"); 456 assert (("unable to allocate worker thread data", wrk));
333 457
334 X_LOCK (wrklock); 458 X_LOCK (wrklock);
335 459
336 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 460 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
337 { 461 {
338 wrk->prev = &wrk_first; 462 wrk->prev = &wrk_first;
339 wrk->next = wrk_first.next; 463 wrk->next = wrk_first.next;
340 wrk_first.next->prev = wrk; 464 wrk_first.next->prev = wrk;
341 wrk_first.next = wrk; 465 wrk_first.next = wrk;
… …
345 free (wrk); 469 free (wrk);
346 470
347 X_UNLOCK (wrklock); 471 X_UNLOCK (wrklock);
348} 472}
349 473
350static void maybe_start_thread (void) 474static void etp_maybe_start_thread (void)
351{ 475{
352 if (eio_nthreads () >= wanted) 476 if (expect_true (etp_nthreads () >= wanted))
353 return; 477 return;
354 478
355 /* todo: maybe use idle here, but might be less exact */ 479 /* todo: maybe use idle here, but might be less exact */
356 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 480 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
357 return; 481 return;
358 482
359 start_thread (); 483 etp_start_thread ();
360} 484}
361 485
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) 486static void etp_end_thread (void)
376{ 487{
377 eio_req *req = calloc (1, sizeof (eio_req)); 488 eio_req *req = calloc (1, sizeof (eio_req));
378 489
379 req->type = EIO_QUIT; 490 req->type = -1;
380 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 491 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381 492
382 X_LOCK (reqlock); 493 X_LOCK (reqlock);
383 reqq_push (&req_queue, req); 494 reqq_push (&req_queue, req);
384 X_COND_SIGNAL (reqwait); 495 X_COND_SIGNAL (reqwait);
385 X_UNLOCK (reqlock); 496 X_UNLOCK (reqlock);
… …
387 X_LOCK (wrklock); 498 X_LOCK (wrklock);
388 --started; 499 --started;
389 X_UNLOCK (wrklock); 500 X_UNLOCK (wrklock);
390} 501}
391 502
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) 503static int etp_poll (void)
429{ 504{
430 int maxreqs = max_poll_reqs; 505 unsigned int maxreqs;
506 unsigned int maxtime;
431 struct timeval tv_start, tv_now; 507 struct timeval tv_start, tv_now;
432 eio_req *req;
433 508
509 X_LOCK (reslock);
510 maxreqs = max_poll_reqs;
511 maxtime = max_poll_time;
512 X_UNLOCK (reslock);
513
434 if (max_poll_time) 514 if (maxtime)
435 gettimeofday (&tv_start, 0); 515 gettimeofday (&tv_start, 0);
436 516
437 for (;;) 517 for (;;)
438 { 518 {
519 ETP_REQ *req;
520
439 maybe_start_thread (); 521 etp_maybe_start_thread ();
440 522
441 X_LOCK (reslock); 523 X_LOCK (reslock);
442 req = reqq_shift (&res_queue); 524 req = reqq_shift (&res_queue);
443 525
444 if (req) 526 if (req)
… …
452 X_UNLOCK (reslock); 534 X_UNLOCK (reslock);
453 535
454 if (!req) 536 if (!req)
455 return 0; 537 return 0;
456 538
539 X_LOCK (reqlock);
457 --nreqs; 540 --nreqs;
541 X_UNLOCK (reqlock);
458 542
459 if (req->type == EIO_GROUP && req->size) 543 if (expect_false (req->type == EIO_GROUP && req->size))
460 { 544 {
461 req->int1 = 1; /* mark request as delayed */ 545 req->int1 = 1; /* mark request as delayed */
462 continue; 546 continue;
463 } 547 }
464 else 548 else
465 { 549 {
466 int res = eio_finish (req); 550 int res = ETP_FINISH (req);
467 if (res) 551 if (expect_false (res))
468 return res; 552 return res;
469 } 553 }
470 554
471 if (maxreqs && !--maxreqs) 555 if (expect_false (maxreqs && !--maxreqs))
472 break; 556 break;
473 557
474 if (max_poll_time) 558 if (maxtime)
475 { 559 {
476 gettimeofday (&tv_now, 0); 560 gettimeofday (&tv_now, 0);
477 561
478 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 562 if (tvdiff (&tv_start, &tv_now) >= maxtime)
479 break; 563 break;
480 } 564 }
481 } 565 }
482 566
483 errno = EAGAIN; 567 errno = EAGAIN;
484 return -1; 568 return -1;
485} 569}
486 570
571static void etp_cancel (ETP_REQ *req)
572{
573 X_LOCK (wrklock);
574 req->flags |= EIO_FLAG_CANCELLED;
575 X_UNLOCK (wrklock);
576
577 eio_grp_cancel (req);
578}
579
580static void etp_submit (ETP_REQ *req)
581{
582 req->pri -= ETP_PRI_MIN;
583
584 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
585 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
586
587 if (expect_false (req->type == EIO_GROUP))
588 {
589 /* I hope this is worth it :/ */
590 X_LOCK (reqlock);
591 ++nreqs;
592 X_UNLOCK (reqlock);
593
594 X_LOCK (reslock);
595
596 ++npending;
597
598 if (!reqq_push (&res_queue, req) && want_poll_cb)
599 want_poll_cb ();
600
601 X_UNLOCK (reslock);
602 }
603 else
604 {
605 X_LOCK (reqlock);
606 ++nreqs;
607 ++nready;
608 reqq_push (&req_queue, req);
609 X_COND_SIGNAL (reqwait);
610 X_UNLOCK (reqlock);
611
612 etp_maybe_start_thread ();
613 }
614}
615
616static void etp_set_max_poll_time (double nseconds)
617{
618 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
619 max_poll_time = nseconds * EIO_TICKS;
620 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
621}
622
623static void etp_set_max_poll_reqs (unsigned int maxreqs)
624{
625 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
626 max_poll_reqs = maxreqs;
627 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
628}
629
630static void etp_set_max_idle (unsigned int nthreads)
631{
632 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
633 max_idle = nthreads;
634 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
635}
636
637static void etp_set_idle_timeout (unsigned int seconds)
638{
639 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
640 idle_timeout = seconds;
641 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
642}
643
644static void etp_set_min_parallel (unsigned int nthreads)
645{
646 if (wanted < nthreads)
647 wanted = nthreads;
648}
649
650static void etp_set_max_parallel (unsigned int nthreads)
651{
652 if (wanted > nthreads)
653 wanted = nthreads;
654
655 while (started > wanted)
656 etp_end_thread ();
657}
658
659/*****************************************************************************/
660
661static void grp_try_feed (eio_req *grp)
662{
663 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
664 {
665 grp->flags &= ~EIO_FLAG_GROUPADD;
666
667 EIO_FEED (grp);
668
669 /* stop if no progress has been made */
670 if (!(grp->flags & EIO_FLAG_GROUPADD))
671 {
672 grp->feed = 0;
673 break;
674 }
675 }
676}
677
678static int grp_dec (eio_req *grp)
679{
680 --grp->size;
681
682 /* call feeder, if applicable */
683 grp_try_feed (grp);
684
685 /* finish, if done */
686 if (!grp->size && grp->int1)
687 return eio_finish (grp);
688 else
689 return 0;
690}
691
692void eio_destroy (eio_req *req)
693{
694 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
695 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
696
697 EIO_DESTROY (req);
698}
699
700static int eio_finish (eio_req *req)
701{
702 int res = EIO_FINISH (req);
703
704 if (req->grp)
705 {
706 int res2;
707 eio_req *grp = req->grp;
708
709 /* unlink request */
710 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
711 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
712
713 if (grp->grp_first == req)
714 grp->grp_first = req->grp_next;
715
716 res2 = grp_dec (grp);
717
718 if (!res && res2)
719 res = res2;
720 }
721
722 eio_destroy (req);
723
724 return res;
725}
726
727void eio_grp_cancel (eio_req *grp)
728{
729 for (grp = grp->grp_first; grp; grp = grp->grp_next)
730 eio_cancel (grp);
731}
732
733void eio_cancel (eio_req *req)
734{
735 etp_cancel (req);
736}
737
738void eio_submit (eio_req *req)
739{
740 etp_submit (req);
741}
742
743unsigned int eio_nreqs (void)
744{
745 return etp_nreqs ();
746}
747
748unsigned int eio_nready (void)
749{
750 return etp_nready ();
751}
752
753unsigned int eio_npending (void)
754{
755 return etp_npending ();
756}
757
758unsigned int eio_nthreads (void)
759{
760 return etp_nthreads ();
761}
762
763void eio_set_max_poll_time (double nseconds)
764{
765 etp_set_max_poll_time (nseconds);
766}
767
768void eio_set_max_poll_reqs (unsigned int maxreqs)
769{
770 etp_set_max_poll_reqs (maxreqs);
771}
772
773void eio_set_max_idle (unsigned int nthreads)
774{
775 etp_set_max_idle (nthreads);
776}
777
778void eio_set_idle_timeout (unsigned int seconds)
779{
780 etp_set_idle_timeout (seconds);
781}
782
783void eio_set_min_parallel (unsigned int nthreads)
784{
785 etp_set_min_parallel (nthreads);
786}
787
788void eio_set_max_parallel (unsigned int nthreads)
789{
790 etp_set_max_parallel (nthreads);
791}
792
793int eio_poll (void)
794{
795 return etp_poll ();
796}
797
487/*****************************************************************************/ 798/*****************************************************************************/
488/* work around various missing functions */ 799/* work around various missing functions */
489 800
490#if !HAVE_PREADWRITE 801#if !HAVE_PREADWRITE
802# undef pread
803# undef pwrite
491# define pread aio_pread 804# define pread eio__pread
492# define pwrite aio_pwrite 805# define pwrite eio__pwrite
493 806
494/* 807static ssize_t
495 * make our pread/pwrite safe against themselves, but not against 808eio__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{ 809{
503 ssize_t res; 810 ssize_t res;
504 off_t ooffset; 811 off_t ooffset;
505 812
506 X_LOCK (preadwritelock); 813 X_LOCK (preadwritelock);
… …
511 X_UNLOCK (preadwritelock); 818 X_UNLOCK (preadwritelock);
512 819
513 return res; 820 return res;
514} 821}
515 822
516static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 823static ssize_t
824eio__pwrite (int fd, void *buf, size_t count, off_t offset)
517{ 825{
518 ssize_t res; 826 ssize_t res;
519 off_t ooffset; 827 off_t ooffset;
520 828
521 X_LOCK (preadwritelock); 829 X_LOCK (preadwritelock);
522 ooffset = lseek (fd, 0, SEEK_CUR); 830 ooffset = lseek (fd, 0, SEEK_CUR);
523 lseek (fd, offset, SEEK_SET); 831 lseek (fd, offset, SEEK_SET);
524 res = write (fd, buf, count); 832 res = write (fd, buf, count);
525 lseek (fd, offset, SEEK_SET); 833 lseek (fd, ooffset, SEEK_SET);
526 X_UNLOCK (preadwritelock); 834 X_UNLOCK (preadwritelock);
527 835
528 return res; 836 return res;
529} 837}
530#endif 838#endif
531 839
532#ifndef HAVE_FUTIMES 840#ifndef HAVE_UTIMES
533 841
842# undef utimes
534# define utimes(path,times) aio_utimes (path, times) 843# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times)
536 844
845static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 846eio__utimes (const char *filename, const struct timeval times[2])
538{ 847{
539 if (times) 848 if (times)
540 { 849 {
541 struct utimbuf buf; 850 struct utimbuf buf;
542 851
… …
547 } 856 }
548 else 857 else
549 return utime (filename, 0); 858 return utime (filename, 0);
550} 859}
551 860
861#endif
862
863#ifndef HAVE_FUTIMES
864
865# undef futimes
866# define futimes(fd,times) eio__futimes (fd, times)
867
552static int aio_futimes (int fd, const struct timeval tv[2]) 868static int eio__futimes (int fd, const struct timeval tv[2])
553{ 869{
554 errno = ENOSYS; 870 errno = ENOSYS;
555 return -1; 871 return -1;
556} 872}
557 873
558#endif 874#endif
559 875
560#if !HAVE_FDATASYNC 876#if !HAVE_FDATASYNC
877# undef fdatasync
561# define fdatasync fsync 878# define fdatasync(fd) fsync (fd)
562#endif 879#endif
880
881/* sync_file_range always needs emulation */
882int
883eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
884{
885#if HAVE_SYNC_FILE_RANGE
886 int res;
887
888 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
889 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
890 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
891 {
892 flags = 0
893 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
894 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
895 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
896 }
897
898 res = sync_file_range (fd, offset, nbytes, flags);
899
900 if (!res || errno != ENOSYS)
901 return res;
902#endif
903
904 /* even though we could play tricks with the flags, it's better to always
905 * call fdatasync, as that matches the expectation of its users best */
906 return fdatasync (fd);
907}
563 908
564#if !HAVE_READAHEAD 909#if !HAVE_READAHEAD
910# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 911# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 912
913static ssize_t
567static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 914eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568{ 915{
569 size_t todo = count; 916 size_t todo = count;
570 dBUF; 917 dBUF;
571 918
572 while (todo > 0) 919 while (todo > 0)
573 { 920 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 921 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 922
576 pread (fd, aio_buf, len, offset); 923 pread (fd, eio_buf, len, offset);
577 offset += len; 924 offset += len;
578 todo -= len; 925 todo -= len;
579 } 926 }
580 927
581 errno = 0; 928 errno = 0;
582 return count; 929 return count;
583} 930}
584 931
585#endif 932#endif
586 933
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 */ 934/* sendfile always needs emulation */
935static ssize_t
618static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 936eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
619{ 937{
620 ssize_t res; 938 ssize_t res;
621 939
622 if (!count) 940 if (!count)
623 return 0; 941 return 0;
624 942
625#if HAVE_SENDFILE 943#if HAVE_SENDFILE
626# if __linux 944# if __linux
627 res = sendfile (ofd, ifd, &offset, count); 945 res = sendfile (ofd, ifd, &offset, count);
628 946
629# elif __freebsd 947# elif __FreeBSD__
630 /* 948 /*
631 * Of course, the freebsd sendfile is a dire hack with no thoughts 949 * Of course, the freebsd sendfile is a dire hack with no thoughts
632 * wasted on making it similar to other I/O functions. 950 * wasted on making it similar to other I/O functions.
633 */ 951 */
634 { 952 {
635 off_t sbytes; 953 off_t sbytes;
636 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 954 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637 955
638 if (res < 0 && sbytes) 956 #if 0 /* according to the manpage, this is correct, but broken behaviour */
639 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 957 /* freebsd' sendfile will return 0 on success */
958 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
959 /* not on e.g. EIO or EPIPE - sounds broken */
960 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
961 res = sbytes;
962 #endif
963
964 /* according to source inspection, this is correct, and useful behaviour */
965 if (sbytes)
966 res = sbytes;
967 }
968
969# elif defined (__APPLE__)
970
971 {
972 off_t sbytes = count;
973 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
974
975 /* according to the manpage, sbytes is always valid */
976 if (sbytes)
640 res = sbytes; 977 res = sbytes;
641 } 978 }
642 979
643# elif __hpux 980# elif __hpux
644 res = sendfile (ofd, ifd, offset, count, 0, 0); 981 res = sendfile (ofd, ifd, offset, count, 0, 0);
… …
658 if (res < 0 && sbytes) 995 if (res < 0 && sbytes)
659 res = sbytes; 996 res = sbytes;
660 } 997 }
661 998
662# endif 999# endif
1000
1001#elif defined (_WIN32)
1002
1003 /* does not work, just for documentation of what would need to be done */
1004 {
1005 HANDLE h = TO_SOCKET (ifd);
1006 SetFilePointer (h, offset, 0, FILE_BEGIN);
1007 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1008 }
1009
663#else 1010#else
664 res = -1; 1011 res = -1;
665 errno = ENOSYS; 1012 errno = ENOSYS;
666#endif 1013#endif
667 1014
668 if (res < 0 1015 if (res < 0
669 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1016 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1017 /* BSDs */
1018#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1019 || errno == ENOTSUP
1020#endif
1021 || errno == EOPNOTSUPP /* BSDs */
670#if __solaris 1022#if __solaris
671 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1023 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
672#endif 1024#endif
673 ) 1025 )
674 ) 1026 )
… …
705 } 1057 }
706 1058
707 return res; 1059 return res;
708} 1060}
709 1061
1062static signed char
1063eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1064{
1065 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1066 : a->inode < b->inode ? -1
1067 : a->inode > b->inode ? 1
1068 : 0;
1069}
1070
1071#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1072
1073#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1074#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1075
1076static void
1077eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1078{
1079 unsigned char bits [9 + sizeof (ino_t) * 8];
1080 unsigned char *bit = bits;
1081
1082 assert (CHAR_BIT == 8);
1083 assert (sizeof (eio_dirent) * 8 < 256);
1084 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1085 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1086
1087 if (size <= EIO_SORT_FAST)
1088 return;
1089
1090 /* first prepare an array of bits to test in our radix sort */
1091 /* try to take endianness into account, as well as differences in ino_t sizes */
1092 /* inode_bits must contain all inodes ORed together */
1093 /* which is used to skip bits that are 0 everywhere, which is very common */
1094 {
1095 ino_t endianness;
1096 int i, j;
1097
1098 /* we store the byte offset of byte n into byte n of "endianness" */
1099 for (i = 0; i < sizeof (ino_t); ++i)
1100 ((unsigned char *)&endianness)[i] = i;
1101
1102 *bit++ = 0;
1103
1104 for (i = 0; i < sizeof (ino_t); ++i)
1105 {
1106 /* shifting off the byte offsets out of "endianness" */
1107 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1108 endianness >>= 8;
1109
1110 for (j = 0; j < 8; ++j)
1111 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1112 *bit++ = offs + j;
1113 }
1114
1115 for (j = 0; j < 8; ++j)
1116 if (score_bits & (1 << j))
1117 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1118 }
1119
1120 /* now actually do the sorting (a variant of MSD radix sort) */
1121 {
1122 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1123 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1124 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1125 int stk_idx = 0;
1126
1127 base_stk [stk_idx] = dents;
1128 end_stk [stk_idx] = dents + size;
1129 bit_stk [stk_idx] = bit - 1;
1130
1131 do
1132 {
1133 base = base_stk [stk_idx];
1134 end = end_stk [stk_idx];
1135 bit = bit_stk [stk_idx];
1136
1137 for (;;)
1138 {
1139 unsigned char O = *bit >> 3;
1140 unsigned char M = 1 << (*bit & 7);
1141
1142 eio_dirent *a = base;
1143 eio_dirent *b = end;
1144
1145 if (b - a < EIO_SORT_CUTOFF)
1146 break;
1147
1148 /* now bit-partition the array on the bit */
1149 /* this ugly asymmetric loop seems to perform much better than typical */
1150 /* partition algos found in the literature */
1151 do
1152 if (!(((unsigned char *)a)[O] & M))
1153 ++a;
1154 else if (!(((unsigned char *)--b)[O] & M))
1155 {
1156 eio_dirent tmp = *a; *a = *b; *b = tmp;
1157 ++a;
1158 }
1159 while (b > a);
1160
1161 /* next bit, or stop, if no bits left in this path */
1162 if (!*--bit)
1163 break;
1164
1165 base_stk [stk_idx] = a;
1166 end_stk [stk_idx] = end;
1167 bit_stk [stk_idx] = bit;
1168 ++stk_idx;
1169
1170 end = a;
1171 }
1172 }
1173 while (stk_idx--);
1174 }
1175}
1176
1177static void
1178eio_dent_insertion_sort (eio_dirent *dents, int size)
1179{
1180 /* first move the smallest element to the front, to act as a sentinel */
1181 {
1182 int i;
1183 eio_dirent *min = dents;
1184
1185 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1186 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1187 if (EIO_DENT_CMP (dents [i], <, *min))
1188 min = &dents [i];
1189
1190 /* swap elements 0 and j (minimum) */
1191 {
1192 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1193 }
1194 }
1195
1196 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1197 {
1198 eio_dirent *i, *j;
1199
1200 for (i = dents + 1; i < dents + size; ++i)
1201 {
1202 eio_dirent value = *i;
1203
1204 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1205 j [1] = j [0];
1206
1207 j [1] = value;
1208 }
1209 }
1210}
1211
1212static void
1213eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1214{
1215 if (size <= 1)
1216 return; /* our insertion sort relies on size > 0 */
1217
1218 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1219 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1220 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1221
1222 /* use an insertion sort at the end, or for small arrays, */
1223 /* as insertion sort is more efficient for small partitions */
1224 eio_dent_insertion_sort (dents, size);
1225}
1226
710/* read a full directory */ 1227/* read a full directory */
1228static void
711static void scandir_ (eio_req *req, worker *self) 1229eio__scandir (eio_req *req, etp_worker *self)
712{ 1230{
713 DIR *dirp; 1231 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; 1232 EIO_STRUCT_DIRENT *entp;
720 char *name, *names; 1233 char *name, *names;
721 int memlen = 4096; 1234 int namesalloc = 4096;
722 int memofs = 0; 1235 int namesoffs = 0;
1236 int flags = req->int1;
1237 eio_dirent *dents = 0;
1238 int dentalloc = 128;
723 int res = 0; 1239 int dentoffs = 0;
1240 ino_t inode_bits = 0;
1241
1242 req->result = -1;
1243
1244 if (!(flags & EIO_READDIR_DENTS))
1245 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
724 1246
725 X_LOCK (wrklock); 1247 X_LOCK (wrklock);
1248 /* the corresponding closedir is in ETP_WORKER_CLEAR */
726 self->dirp = dirp = opendir (req->ptr1); 1249 self->dirp = dirp = opendir (req->ptr1);
727 self->dbuf = u = malloc (sizeof (*u)); 1250
728 req->flags |= EIO_FLAG_PTR2_FREE; 1251 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1252 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
729 req->ptr2 = names = malloc (memlen); 1253 req->ptr2 = names = malloc (namesalloc);
730 X_UNLOCK (wrklock); 1254 X_UNLOCK (wrklock);
731 1255
732 if (dirp && u && names) 1256 if (dirp && names && (!flags || dents))
733 for (;;) 1257 for (;;)
734 { 1258 {
735 errno = 0; 1259 errno = 0;
736 readdir_r (dirp, &u->d, &entp); 1260 entp = readdir (dirp);
737 1261
738 if (!entp) 1262 if (!entp)
1263 {
1264 if (errno)
1265 break;
1266
1267 /* sort etc. */
1268 req->int1 = flags;
1269 req->result = dentoffs;
1270
1271 if (flags & EIO_READDIR_STAT_ORDER)
1272 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1273 else if (flags & EIO_READDIR_DIRS_FIRST)
1274 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1275 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1276 else
1277 {
1278 /* in this case, all is known, and we just put dirs first and sort them */
1279 eio_dirent *oth = dents + dentoffs;
1280 eio_dirent *dir = dents;
1281
1282 /* now partition dirs to the front, and non-dirs to the back */
1283 /* by walking from both sides and swapping if necessary */
1284 /* also clear score, so it doesn't influence sorting */
1285 while (oth > dir)
1286 {
1287 if (dir->type == EIO_DT_DIR)
1288 ++dir;
1289 else if ((--oth)->type == EIO_DT_DIR)
1290 {
1291 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1292
1293 ++dir;
1294 }
1295 }
1296
1297 /* now sort the dirs only */
1298 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1299 }
1300
739 break; 1301 break;
1302 }
740 1303
1304 /* now add the entry to our list(s) */
741 name = entp->d_name; 1305 name = entp->d_name;
742 1306
1307 /* skip . and .. entries */
743 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1308 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
744 { 1309 {
745 int len = strlen (name) + 1; 1310 int len = D_NAMLEN (entp) + 1;
746 1311
747 res++; 1312 while (expect_false (namesoffs + len > namesalloc))
748
749 while (memofs + len > memlen)
750 { 1313 {
751 memlen *= 2; 1314 namesalloc *= 2;
752 X_LOCK (wrklock); 1315 X_LOCK (wrklock);
753 req->ptr2 = names = realloc (names, memlen); 1316 req->ptr2 = names = realloc (names, namesalloc);
754 X_UNLOCK (wrklock); 1317 X_UNLOCK (wrklock);
755 1318
756 if (!names) 1319 if (!names)
757 break; 1320 break;
758 } 1321 }
759 1322
760 memcpy (names + memofs, name, len); 1323 memcpy (names + namesoffs, name, len);
1324
1325 if (dents)
1326 {
1327 struct eio_dirent *ent;
1328
1329 if (expect_false (dentoffs == dentalloc))
1330 {
1331 dentalloc *= 2;
1332 X_LOCK (wrklock);
1333 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1334 X_UNLOCK (wrklock);
1335
1336 if (!dents)
1337 break;
1338 }
1339
1340 ent = dents + dentoffs;
1341
1342 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1343 ent->namelen = len - 1;
1344 ent->inode = D_INO (entp);
1345
1346 inode_bits |= ent->inode;
1347
1348 switch (D_TYPE (entp))
1349 {
1350 default:
1351 ent->type = EIO_DT_UNKNOWN;
1352 flags |= EIO_READDIR_FOUND_UNKNOWN;
1353 break;
1354
1355 #ifdef DT_FIFO
1356 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1357 #endif
1358 #ifdef DT_CHR
1359 case DT_CHR: ent->type = EIO_DT_CHR; break;
1360 #endif
1361 #ifdef DT_MPC
1362 case DT_MPC: ent->type = EIO_DT_MPC; break;
1363 #endif
1364 #ifdef DT_DIR
1365 case DT_DIR: ent->type = EIO_DT_DIR; break;
1366 #endif
1367 #ifdef DT_NAM
1368 case DT_NAM: ent->type = EIO_DT_NAM; break;
1369 #endif
1370 #ifdef DT_BLK
1371 case DT_BLK: ent->type = EIO_DT_BLK; break;
1372 #endif
1373 #ifdef DT_MPB
1374 case DT_MPB: ent->type = EIO_DT_MPB; break;
1375 #endif
1376 #ifdef DT_REG
1377 case DT_REG: ent->type = EIO_DT_REG; break;
1378 #endif
1379 #ifdef DT_NWK
1380 case DT_NWK: ent->type = EIO_DT_NWK; break;
1381 #endif
1382 #ifdef DT_CMP
1383 case DT_CMP: ent->type = EIO_DT_CMP; break;
1384 #endif
1385 #ifdef DT_LNK
1386 case DT_LNK: ent->type = EIO_DT_LNK; break;
1387 #endif
1388 #ifdef DT_SOCK
1389 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1390 #endif
1391 #ifdef DT_DOOR
1392 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1393 #endif
1394 #ifdef DT_WHT
1395 case DT_WHT: ent->type = EIO_DT_WHT; break;
1396 #endif
1397 }
1398
1399 ent->score = 7;
1400
1401 if (flags & EIO_READDIR_DIRS_FIRST)
1402 {
1403 if (ent->type == EIO_DT_UNKNOWN)
1404 {
1405 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1406 ent->score = 1;
1407 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1408 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1409 }
1410 else if (ent->type == EIO_DT_DIR)
1411 ent->score = 0;
1412 }
1413 }
1414
761 memofs += len; 1415 namesoffs += len;
1416 ++dentoffs;
1417 }
1418
1419 if (EIO_CANCELLED (req))
1420 {
1421 errno = ECANCELED;
1422 break;
762 } 1423 }
763 } 1424 }
1425}
764 1426
765 if (errno) 1427#ifdef PAGESIZE
766 res = -1; 1428# define eio_pagesize() PAGESIZE
1429#else
1430static intptr_t
1431eio_pagesize (void)
1432{
1433 static intptr_t page;
1434
1435 if (!page)
1436 page = sysconf (_SC_PAGESIZE);
1437
1438 return page;
1439}
1440#endif
1441
1442static void
1443eio_page_align (void **addr, size_t *length)
1444{
1445 intptr_t mask = eio_pagesize () - 1;
1446
1447 /* round down addr */
1448 intptr_t adj = mask & (intptr_t)*addr;
1449
1450 *addr = (void *)((intptr_t)*addr - adj);
1451 *length += adj;
1452
1453 /* round up length */
1454 *length = (*length + mask) & ~mask;
1455}
1456
1457#if !_POSIX_MEMLOCK
1458# define eio__mlockall(a) ((errno = ENOSYS), -1)
1459#else
1460
1461static int
1462eio__mlockall (int flags)
1463{
1464 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1465 extern int mallopt (int, int);
1466 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1467 #endif
1468
1469 if (EIO_MCL_CURRENT != MCL_CURRENT
1470 || EIO_MCL_FUTURE != MCL_FUTURE)
1471 {
1472 flags = 0
1473 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1474 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1475 }
1476
1477 return mlockall (flags);
1478}
1479#endif
1480
1481#if !_POSIX_MEMLOCK_RANGE
1482# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1483#else
1484
1485static int
1486eio__mlock (void *addr, size_t length)
1487{
1488 eio_page_align (&addr, &length);
1489
1490 return mlock (addr, length);
1491}
1492
1493#endif
1494
1495#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1496# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1497#else
1498
1499int
1500eio__msync (void *mem, size_t len, int flags)
1501{
1502 eio_page_align (&mem, &len);
1503
1504 if (EIO_MS_ASYNC != MS_SYNC
1505 || EIO_MS_INVALIDATE != MS_INVALIDATE
1506 || EIO_MS_SYNC != MS_SYNC)
1507 {
1508 flags = 0
1509 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1510 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1511 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1512 }
1513
1514 return msync (mem, len, flags);
1515}
1516
1517#endif
1518
1519int
1520eio__mtouch (void *mem, size_t len, int flags)
1521{
1522 eio_page_align (&mem, &len);
1523
767 1524 {
768 req->result = res; 1525 intptr_t addr = (intptr_t)mem;
1526 intptr_t end = addr + len;
1527 intptr_t page = eio_pagesize ();
1528
1529 if (addr < end)
1530 if (flags & EIO_MT_MODIFY) /* modify */
1531 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1532 else
1533 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1534 }
1535
1536 return 0;
769} 1537}
770 1538
771/*****************************************************************************/ 1539/*****************************************************************************/
772 1540
1541#define ALLOC(len) \
1542 if (!req->ptr2) \
1543 { \
1544 X_LOCK (wrklock); \
1545 req->flags |= EIO_FLAG_PTR2_FREE; \
1546 X_UNLOCK (wrklock); \
1547 req->ptr2 = malloc (len); \
1548 if (!req->ptr2) \
1549 { \
1550 errno = ENOMEM; \
1551 req->result = -1; \
1552 break; \
1553 } \
1554 }
1555
773X_THREAD_PROC (eio_proc) 1556X_THREAD_PROC (etp_proc)
774{ 1557{
775 eio_req *req; 1558 ETP_REQ *req;
776 struct timespec ts; 1559 struct timespec ts;
777 worker *self = (worker *)thr_arg; 1560 etp_worker *self = (etp_worker *)thr_arg;
778 1561
779 /* try to distribute timeouts somewhat randomly */ 1562 /* try to distribute timeouts somewhat randomly */
780 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1563 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781 1564
782 for (;;) 1565 for (;;)
783 { 1566 {
784 ts.tv_sec = time (0) + IDLE_TIMEOUT;
785
786 X_LOCK (reqlock); 1567 X_LOCK (reqlock);
787 1568
788 for (;;) 1569 for (;;)
789 { 1570 {
790 self->req = req = reqq_shift (&req_queue); 1571 self->req = req = reqq_shift (&req_queue);
… …
792 if (req) 1573 if (req)
793 break; 1574 break;
794 1575
795 ++idle; 1576 ++idle;
796 1577
1578 ts.tv_sec = time (0) + idle_timeout;
797 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1579 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
798 { 1580 {
799 if (idle > max_idle) 1581 if (idle > max_idle)
800 { 1582 {
801 --idle; 1583 --idle;
… …
806 goto quit; 1588 goto quit;
807 } 1589 }
808 1590
809 /* we are allowed to idle, so do so without any timeout */ 1591 /* we are allowed to idle, so do so without any timeout */
810 X_COND_WAIT (reqwait, reqlock); 1592 X_COND_WAIT (reqwait, reqlock);
811 ts.tv_sec = time (0) + IDLE_TIMEOUT;
812 } 1593 }
813 1594
814 --idle; 1595 --idle;
815 } 1596 }
816 1597
817 --nready; 1598 --nready;
818 1599
819 X_UNLOCK (reqlock); 1600 X_UNLOCK (reqlock);
820 1601
821 errno = 0; /* strictly unnecessary */ 1602 if (req->type < 0)
1603 goto quit;
822 1604
823 if (!EIO_CANCELLED (req)) 1605 if (!EIO_CANCELLED (req))
824 switch (req->type) 1606 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 1607
918 X_LOCK (reslock); 1608 X_LOCK (reslock);
919 1609
920 ++npending; 1610 ++npending;
921 1611
922 if (!reqq_push (&res_queue, req) && want_poll_cb) 1612 if (!reqq_push (&res_queue, req) && want_poll_cb)
923 want_poll_cb (); 1613 want_poll_cb ();
924 1614
925 self->req = 0; 1615 self->req = 0;
926 worker_clear (self); 1616 etp_worker_clear (self);
927 1617
928 X_UNLOCK (reslock); 1618 X_UNLOCK (reslock);
929 } 1619 }
930 1620
931quit: 1621quit:
932 X_LOCK (wrklock); 1622 X_LOCK (wrklock);
933 worker_free (self); 1623 etp_worker_free (self);
934 X_UNLOCK (wrklock); 1624 X_UNLOCK (wrklock);
935 1625
936 return 0; 1626 return 0;
937} 1627}
938 1628
939/*****************************************************************************/ 1629/*****************************************************************************/
940 1630
941static void eio_atfork_prepare (void) 1631int eio_init (void (*want_poll)(void), void (*done_poll)(void))
942{ 1632{
943 X_LOCK (wrklock); 1633 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} 1634}
953 1635
954static void eio_atfork_parent (void) 1636static void eio_api_destroy (eio_req *req)
955{ 1637{
956#if !HAVE_READDIR_R 1638 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} 1639}
966 1640
967static void eio_atfork_child (void) 1641#define REQ(rtype) \
968{ 1642 eio_req *req; \
969 eio_req *prv; 1643 \
1644 req = (eio_req *)calloc (1, sizeof *req); \
1645 if (!req) \
1646 return 0; \
1647 \
1648 req->type = rtype; \
1649 req->pri = pri; \
1650 req->finish = cb; \
1651 req->data = data; \
1652 req->destroy = eio_api_destroy;
970 1653
971 while (prv = reqq_shift (&req_queue)) 1654#define SEND eio_submit (req); return req
972 eio_destroy (prv);
973 1655
974 while (prv = reqq_shift (&res_queue)) 1656#define PATH \
975 eio_destroy (prv); 1657 req->flags |= EIO_FLAG_PTR1_FREE; \
1658 req->ptr1 = strdup (path); \
1659 if (!req->ptr1) \
1660 { \
1661 eio_api_destroy (req); \
1662 return 0; \
1663 }
976 1664
977 while (wrk_first.next != &wrk_first) 1665static void eio_execute (etp_worker *self, eio_req *req)
1666{
1667 switch (req->type)
978 { 1668 {
979 worker *wrk = wrk_first.next; 1669 case EIO_READ: ALLOC (req->size);
1670 req->result = req->offs >= 0
1671 ? pread (req->int1, req->ptr2, req->size, req->offs)
1672 : read (req->int1, req->ptr2, req->size); break;
1673 case EIO_WRITE: req->result = req->offs >= 0
1674 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1675 : write (req->int1, req->ptr2, req->size); break;
980 1676
981 if (wrk->req) 1677 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
982 eio_destroy (wrk->req); 1678 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
983 1679
984 worker_clear (wrk); 1680 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
985 worker_free (wrk); 1681 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
986 } 1682 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1683 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1684 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1685 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
987 1686
988 started = 0; 1687 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
989 idle = 0; 1688 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
990 nreqs = 0; 1689 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
991 nready = 0; 1690 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
992 npending = 0;
993 1691
994 eio_atfork_parent (); 1692 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
995} 1693 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1694 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1695 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1696 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1697 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
996 1698
997int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1699 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
998{ 1700 case EIO_CLOSE: req->result = close (req->int1); break;
999 want_poll_cb = want_poll; 1701 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1000 done_poll_cb = done_poll; 1702 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1703 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1704 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1705 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1706 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1707 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1708 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1001 1709
1710 case EIO_READLINK: ALLOC (PATH_MAX);
1711 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1712
1713 case EIO_SYNC: req->result = 0; sync (); break;
1714 case EIO_FSYNC: req->result = fsync (req->int1); break;
1715 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1716 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1717 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1718 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1719 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1720 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1721
1722 case EIO_READDIR: eio__scandir (req, self); break;
1723
1724 case EIO_BUSY:
1002#ifdef _WIN32 1725#ifdef _WIN32
1003 X_MUTEX_CHECK (wrklock); 1726 Sleep (req->nv1 * 1e3);
1004 X_MUTEX_CHECK (reslock); 1727#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 { 1728 {
1130 dREQ; 1729 struct timeval tv;
1131 1730
1132 req->type = ix; 1731 tv.tv_sec = req->nv1;
1133 req->sv1 = newSVsv (fh); 1732 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 1733
1142 if (!SvREADONLY (data)) 1734 req->result = select (0, 0, 0, 0, &tv);
1735 }
1736#endif
1737 break;
1738
1739 case EIO_UTIME:
1740 case EIO_FUTIME:
1741 {
1742 struct timeval tv[2];
1743 struct timeval *times;
1744
1745 if (req->nv1 != -1. || req->nv2 != -1.)
1143 { 1746 {
1144 SvREADONLY_on (data); 1747 tv[0].tv_sec = req->nv1;
1145 req->flags |= FLAG_SV2_RO_OFF; 1748 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1749 tv[1].tv_sec = req->nv2;
1750 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1751
1752 times = tv;
1146 } 1753 }
1754 else
1755 times = 0;
1147 1756
1148 EIO_SEND; 1757 req->result = req->type == EIO_FUTIME
1758 ? futimes (req->int1, times)
1759 : utimes (req->ptr1, times);
1149 } 1760 }
1150} 1761 break;
1151 1762
1152void 1763 case EIO_GROUP:
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef) 1764 abort (); /* handled in eio_request */
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{
1157 SV *data;
1158 dREQ;
1159 1765
1160 data = newSV (NAME_MAX); 1766 case EIO_NOP:
1161 SvPOK_on (data); 1767 req->result = 0;
1768 break;
1162 1769
1163 req->type = EIO_READLINK; 1770 case EIO_CUSTOM:
1164 req->sv1 = newSVsv (path); 1771 ((void (*)(eio_req *))req->feed) (req);
1165 req->ptr1 = SvPVbyte_nolen (req->sv1); 1772 break;
1166 req->sv2 = data;
1167 req->ptr2 = SvPVbyte_nolen (data);
1168 1773
1169 EIO_SEND; 1774 default:
1170} 1775 errno = ENOSYS;
1776 req->result = -1;
1777 break;
1778 }
1171 1779
1172void 1780 req->errorno = errno;
1173aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174 PROTOTYPE: $$$$;$
1175 PPCODE:
1176{
1177 dREQ;
1178
1179 req->type = EIO_SENDFILE;
1180 req->sv1 = newSVsv (out_fh);
1181 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182 req->sv2 = newSVsv (in_fh);
1183 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184 req->offs = SvVAL64 (in_offset);
1185 req->size = length;
1186
1187 EIO_SEND;
1188} 1781}
1189 1782
1190void 1783#ifndef EIO_NO_WRAPPERS
1191aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192 PROTOTYPE: $$$;$
1193 PPCODE:
1194{
1195 dREQ;
1196 1784
1197 req->type = EIO_READAHEAD; 1785eio_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{ 1786{
1213 dREQ; 1787 REQ (EIO_NOP); SEND;
1788}
1214 1789
1215 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT)); 1790eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1216 if (!req->ptr2) 1791{
1217 { 1792 REQ (EIO_BUSY); req->nv1 = delay; SEND;
1218 req_destroy (req); 1793}
1219 croak ("out of memory during aio_stat statdata allocation");
1220 }
1221 1794
1795eio_req *eio_sync (int pri, eio_cb cb, void *data)
1796{
1797 REQ (EIO_SYNC); SEND;
1798}
1799
1800eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1801{
1802 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1803}
1804
1805eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1806{
1807 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1808}
1809
1810eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1811{
1812 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1813}
1814
1815eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1816{
1817 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1818}
1819
1820eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1821{
1822 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1823}
1824
1825eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1826{
1827 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1828}
1829
1830eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1831{
1832 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1833}
1834
1835eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1836{
1837 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1838}
1839
1840eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1841{
1842 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1843}
1844
1845eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1846{
1847 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1848}
1849
1850eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1851{
1852 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1853}
1854
1855eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1856{
1857 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1858}
1859
1860eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1861{
1862 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1863}
1864
1865eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1866{
1867 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1868}
1869
1870eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1871{
1872 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1873}
1874
1875eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1876{
1877 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1878}
1879
1880eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1881{
1882 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1883}
1884
1885eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1886{
1887 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1888}
1889
1890eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1891{
1892 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1893}
1894
1895eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1896{
1897 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1898}
1899
1900eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1901{
1902 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1903}
1904
1905eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1906{
1907 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1908}
1909
1910eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1911{
1912 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1913}
1914
1915eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1916{
1917 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1918}
1919
1920eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1921{
1922 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1923}
1924
1925static eio_req *
1926eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1927{
1928 REQ (type); PATH; SEND;
1929}
1930
1931eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1932{
1933 return eio__1path (EIO_READLINK, path, pri, cb, data);
1934}
1935
1936eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1937{
1938 return eio__1path (EIO_STAT, path, pri, cb, data);
1939}
1940
1941eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1942{
1943 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1944}
1945
1946eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1947{
1948 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1949}
1950
1951eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1952{
1953 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1954}
1955
1956eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1957{
1958 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1959}
1960
1961eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1962{
1963 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1964}
1965
1966eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1967{
1968 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1969}
1970
1971static eio_req *
1972eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1973{
1974 REQ (type); PATH;
1975
1222 req->flags |= FLAG_PTR2_FREE; 1976 req->flags |= EIO_FLAG_PTR2_FREE;
1223 req->sv1 = newSVsv (fh_or_path); 1977 req->ptr2 = strdup (new_path);
1978 if (!req->ptr2)
1979 {
1980 eio_api_destroy (req);
1981 return 0;
1982 }
1224 1983
1225 if (SvPOK (fh_or_path)) 1984 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} 1985}
1238 1986
1239void 1987eio_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{ 1988{
1243 dREQ; 1989 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} 1990}
1262 1991
1263void 1992eio_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{ 1993{
1267 dREQ; 1994 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} 1995}
1285 1996
1286void 1997eio_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{ 1998{
1294 dREQ; 1999 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} 2000}
1307 2001
1308void 2002eio_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{ 2003{
1312 dREQ; 2004 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} 2005}
1331 2006
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 2007#endif
2008
2009eio_req *eio_grp (eio_cb cb, void *data)
2010{
2011 const int pri = EIO_PRI_MAX;
2012
2013 REQ (EIO_GROUP); SEND;
2014}
2015
2016#undef REQ
2017#undef PATH
2018#undef SEND
2019
2020/*****************************************************************************/
2021/* grp functions */
1418 2022
1419void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2023void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 2024{
1421 grp->int2 = limit; 2025 grp->int2 = limit;
1422 grp->feed = feed; 2026 grp->feed = feed;
… …
1433 2037
1434void eio_grp_add (eio_req *grp, eio_req *req) 2038void eio_grp_add (eio_req *grp, eio_req *req)
1435{ 2039{
1436 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2040 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1437 2041
2042 grp->flags |= EIO_FLAG_GROUPADD;
2043
1438 ++grp->size; 2044 ++grp->size;
1439 req->grp = grp; 2045 req->grp = grp;
1440 2046
1441 req->grp_prev = 0; 2047 req->grp_prev = 0;
1442 req->grp_next = grp->grp_first; 2048 req->grp_next = grp->grp_first;
… …
1445 grp->grp_first->grp_prev = req; 2051 grp->grp_first->grp_prev = req;
1446 2052
1447 grp->grp_first = req; 2053 grp->grp_first = req;
1448} 2054}
1449 2055
2056/*****************************************************************************/
2057/* misc garbage */
1450 2058
2059ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2060{
2061 etp_worker wrk;
2062 ssize_t ret;
2063
2064 wrk.dbuf = 0;
2065
2066 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2067
2068 if (wrk.dbuf)
2069 free (wrk.dbuf);
2070
2071 return ret;
2072}
2073

Diff Legend

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