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

Comparing libeio/eio.c (file contents):
Revision 1.8 by root, Sun May 11 20:08:37 2008 UTC vs.
Revision 1.59 by root, Wed Dec 1 07:30:39 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#include <stddef.h> 48#include <stddef.h>
6#include <stdlib.h> 49#include <stdlib.h>
7#include <string.h> 50#include <string.h>
8#include <errno.h> 51#include <errno.h>
9#include <sys/types.h> 52#include <sys/types.h>
10#include <sys/stat.h> 53#include <sys/stat.h>
54#include <sys/statvfs.h>
11#include <limits.h> 55#include <limits.h>
12#include <fcntl.h> 56#include <fcntl.h>
13#include <assert.h> 57#include <assert.h>
14 58
15#ifndef EIO_FINISH 59#ifndef EIO_FINISH
… …
25#endif 69#endif
26 70
27#ifdef _WIN32 71#ifdef _WIN32
28 72
29 /*doh*/ 73 /*doh*/
30
31#else 74#else
32 75
33# include "config.h" 76# include "config.h"
34# include <sys/time.h> 77# include <sys/time.h>
35# include <sys/select.h> 78# include <sys/select.h>
36# include <unistd.h> 79# include <unistd.h>
37# include <utime.h> 80# include <utime.h>
38# include <signal.h> 81# include <signal.h>
39# include <dirent.h> 82# include <dirent.h>
40 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
105
41# ifndef EIO_STRUCT_DIRENT 106# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 107# define EIO_STRUCT_DIRENT struct dirent
43# endif 108# endif
44 109
45#endif 110#endif
46 111
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif
50
51#if HAVE_SENDFILE 112#if HAVE_SENDFILE
52# if __linux 113# if __linux
53# include <sys/sendfile.h> 114# include <sys/sendfile.h>
54# elif __freebsd 115# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 116# include <sys/socket.h>
56# include <sys/uio.h> 117# include <sys/uio.h>
57# elif __hpux 118# elif __hpux
58# include <sys/socket.h> 119# include <sys/socket.h>
59# elif __solaris /* not yet */ 120# elif __solaris
60# include <sys/sendfile.h> 121# include <sys/sendfile.h>
61# else 122# else
62# error sendfile support requested but not available 123# error sendfile support requested but not available
63# endif 124# endif
64#endif 125#endif
65 126
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
66/* number of seconds after which an idle threads exit */ 137/* number of seconds after which an idle threads exit */
67#define IDLE_TIMEOUT 10 138#define IDLE_TIMEOUT 10
68 139
69/* used for struct dirent, AIX doesn't provide it */ 140/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 141#ifndef NAME_MAX
71# define NAME_MAX 4096 142# define NAME_MAX 4096
72#endif 143#endif
73 144
145/* used for readlink etc. */
146#ifndef PATH_MAX
147# define PATH_MAX 4096
148#endif
149
74/* buffer size for various temporary buffers */ 150/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 151#define EIO_BUFSIZE 65536
76 152
77#define dBUF \ 153#define dBUF \
78 char *eio_buf; \ 154 char *eio_buf; \
79 X_LOCK (wrklock); \ 155 ETP_WORKER_LOCK (self); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 156 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 157 ETP_WORKER_UNLOCK (self); \
158 errno = ENOMEM; \
82 if (!eio_buf) \ 159 if (!eio_buf) \
83 return -1; 160 return -1;
84 161
85#define EIO_TICKS ((1000000 + 1023) >> 10) 162#define EIO_TICKS ((1000000 + 1023) >> 10)
86 163
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
87static void (*want_poll_cb) (void); 219static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 220static void (*done_poll_cb) (void);
89 221
90static unsigned int max_poll_time = 0; 222static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 223static unsigned int max_poll_reqs; /* reslock */
92 224
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 225static volatile unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 226static volatile unsigned int nready; /* reqlock */
95{ 227static volatile unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 228static volatile unsigned int max_idle = 4;
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
98}
99 229
100static 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)
101 261
102/* worker threads management */ 262/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 263
105typedef struct worker {
106 /* locked by wrklock */
107 struct worker *prev, *next;
108
109 thread_t tid;
110
111 /* locked by reslock, reqlock or wrklock */
112 eio_req *req; /* currently processed request */
113 void *dbuf;
114 DIR *dirp;
115} worker;
116
117static worker wrk_first = { &wrk_first, &wrk_first, 0 };
118
119static void worker_clear (worker *wrk) 264static void etp_worker_clear (etp_worker *wrk)
120{ 265{
121 if (wrk->dirp) 266 ETP_WORKER_CLEAR (wrk);
122 {
123 closedir (wrk->dirp);
124 wrk->dirp = 0;
125 }
126
127 if (wrk->dbuf)
128 {
129 free (wrk->dbuf);
130 wrk->dbuf = 0;
131 }
132} 267}
133 268
134static void worker_free (worker *wrk) 269static void etp_worker_free (etp_worker *wrk)
135{ 270{
136 wrk->next->prev = wrk->prev; 271 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 272 wrk->prev->next = wrk->next;
138 273
139 free (wrk); 274 free (wrk);
140} 275}
141 276
142static volatile unsigned int nreqs, nready, npending;
143static volatile unsigned int max_idle = 4;
144
145static mutex_t reslock = X_MUTEX_INIT;
146static mutex_t reqlock = X_MUTEX_INIT;
147static cond_t reqwait = X_COND_INIT;
148
149unsigned int eio_nreqs (void) 277static unsigned int etp_nreqs (void)
150{ 278{
279 int retval;
280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281 retval = nreqs;
282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 283 return retval;
152} 284}
153 285
154unsigned int eio_nready (void) 286static unsigned int etp_nready (void)
155{ 287{
156 unsigned int retval; 288 unsigned int retval;
157 289
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 291 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 293
162 return retval; 294 return retval;
163} 295}
164 296
165unsigned int eio_npending (void) 297static unsigned int etp_npending (void)
166{ 298{
167 unsigned int retval; 299 unsigned int retval;
168 300
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 301 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 302 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 303 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 304
173 return retval; 305 return retval;
174} 306}
175 307
176unsigned int eio_nthreads (void) 308static unsigned int etp_nthreads (void)
177{ 309{
178 unsigned int retval; 310 unsigned int retval;
179 311
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 312 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 313 retval = started;
… …
188 * a somewhat faster data structure might be nice, but 320 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 321 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 322 * per shift, the most expensive operation.
191 */ 323 */
192typedef struct { 324typedef struct {
193 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 */
194 int size; 326 int size;
195} reqq; 327} etp_reqq;
196 328
197static reqq req_queue; 329static etp_reqq req_queue;
198static reqq res_queue; 330static etp_reqq res_queue;
199 331
200static int reqq_push (reqq *q, eio_req *req) 332static int reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 333{
202 int pri = req->pri; 334 int pri = req->pri;
203 req->next = 0; 335 req->next = 0;
204 336
205 if (q->qe[pri]) 337 if (q->qe[pri])
… …
211 q->qe[pri] = q->qs[pri] = req; 343 q->qe[pri] = q->qs[pri] = req;
212 344
213 return q->size++; 345 return q->size++;
214} 346}
215 347
216static eio_req *reqq_shift (reqq *q) 348static ETP_REQ *reqq_shift (etp_reqq *q)
217{ 349{
218 int pri; 350 int pri;
219 351
220 if (!q->size) 352 if (!q->size)
221 return 0; 353 return 0;
222 354
223 --q->size; 355 --q->size;
224 356
225 for (pri = EIO_NUM_PRI; pri--; ) 357 for (pri = ETP_NUM_PRI; pri--; )
226 { 358 {
227 eio_req *req = q->qs[pri]; 359 eio_req *req = q->qs[pri];
228 360
229 if (req) 361 if (req)
230 { 362 {
… …
236 } 368 }
237 369
238 abort (); 370 abort ();
239} 371}
240 372
241static void grp_try_feed (eio_req *grp) 373static void etp_atfork_prepare (void)
242{ 374{
243 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)
244 { 404 {
245 int old_len = grp->size; 405 etp_worker *wrk = wrk_first.next;
246 406
247 EIO_FEED (grp); 407 if (wrk->req)
408 ETP_DESTROY (wrk->req);
248 409
249 /* stop if no progress has been made */ 410 etp_worker_clear (wrk);
250 if (old_len == grp->size) 411 etp_worker_free (wrk);
251 {
252 grp->feed = 0;
253 break;
254 }
255 } 412 }
256}
257 413
258static int eio_finish (eio_req *req); 414 started = 0;
415 idle = 0;
416 nreqs = 0;
417 nready = 0;
418 npending = 0;
259 419
260static int grp_dec (eio_req *grp) 420 etp_atfork_parent ();
261{ 421}
262 --grp->size;
263 422
264 /* call feeder, if applicable */ 423static void
265 grp_try_feed (grp); 424etp_once_init (void)
425{
426 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
427}
266 428
267 /* finish, if done */ 429static int
268 if (!grp->size && grp->int1) 430etp_init (void (*want_poll)(void), void (*done_poll)(void))
269 return eio_finish (grp); 431{
270 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
271 return 0; 439 return 0;
272} 440}
273 441
274void eio_destroy (eio_req *req)
275{
276 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
277 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
278
279 EIO_DESTROY (req);
280}
281
282static int eio_finish (eio_req *req)
283{
284 int res = EIO_FINISH (req);
285
286 if (req->grp)
287 {
288 int res2;
289 eio_req *grp = req->grp;
290
291 /* unlink request */
292 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
293 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
294
295 if (grp->grp_first == req)
296 grp->grp_first = req->grp_next;
297
298 res2 = grp_dec (grp);
299
300 if (!res && res2)
301 res = res2;
302 }
303
304 eio_destroy (req);
305
306 return res;
307}
308
309void eio_grp_cancel (eio_req *grp)
310{
311 for (grp = grp->grp_first; grp; grp = grp->grp_next)
312 eio_cancel (grp);
313}
314
315void eio_cancel (eio_req *req)
316{
317 X_LOCK (wrklock);
318 req->flags |= EIO_FLAG_CANCELLED;
319 X_UNLOCK (wrklock);
320
321 eio_grp_cancel (req);
322}
323
324X_THREAD_PROC (eio_proc); 442X_THREAD_PROC (etp_proc);
325 443
326static void start_thread (void) 444static void etp_start_thread (void)
327{ 445{
328 worker *wrk = calloc (1, sizeof (worker)); 446 etp_worker *wrk = calloc (1, sizeof (etp_worker));
329 447
330 /*TODO*/ 448 /*TODO*/
331 assert (("unable to allocate worker thread data", !wrk)); 449 assert (("unable to allocate worker thread data", wrk));
332 450
333 X_LOCK (wrklock); 451 X_LOCK (wrklock);
334 452
335 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 453 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
336 { 454 {
337 wrk->prev = &wrk_first; 455 wrk->prev = &wrk_first;
338 wrk->next = wrk_first.next; 456 wrk->next = wrk_first.next;
339 wrk_first.next->prev = wrk; 457 wrk_first.next->prev = wrk;
340 wrk_first.next = wrk; 458 wrk_first.next = wrk;
… …
344 free (wrk); 462 free (wrk);
345 463
346 X_UNLOCK (wrklock); 464 X_UNLOCK (wrklock);
347} 465}
348 466
349static void maybe_start_thread (void) 467static void etp_maybe_start_thread (void)
350{ 468{
351 if (eio_nthreads () >= wanted) 469 if (expect_true (etp_nthreads () >= wanted))
352 return; 470 return;
353 471
354 /* todo: maybe use idle here, but might be less exact */ 472 /* todo: maybe use idle here, but might be less exact */
355 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 ()))
356 return; 474 return;
357 475
358 start_thread (); 476 etp_start_thread ();
359} 477}
360 478
361void eio_submit (eio_req *req)
362{
363 ++nreqs;
364
365 X_LOCK (reqlock);
366 ++nready;
367 reqq_push (&req_queue, req);
368 X_COND_SIGNAL (reqwait);
369 X_UNLOCK (reqlock);
370
371 maybe_start_thread ();
372}
373
374static void end_thread (void) 479static void etp_end_thread (void)
375{ 480{
376 eio_req *req = calloc (1, sizeof (eio_req)); 481 eio_req *req = calloc (1, sizeof (eio_req));
377 482
378 req->type = EIO_QUIT; 483 req->type = -1;
379 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 484 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
380 485
381 X_LOCK (reqlock); 486 X_LOCK (reqlock);
382 reqq_push (&req_queue, req); 487 reqq_push (&req_queue, req);
383 X_COND_SIGNAL (reqwait); 488 X_COND_SIGNAL (reqwait);
384 X_UNLOCK (reqlock); 489 X_UNLOCK (reqlock);
… …
386 X_LOCK (wrklock); 491 X_LOCK (wrklock);
387 --started; 492 --started;
388 X_UNLOCK (wrklock); 493 X_UNLOCK (wrklock);
389} 494}
390 495
391void eio_set_max_poll_time (double nseconds)
392{
393 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
394 max_poll_time = nseconds;
395 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
396}
397
398void eio_set_max_poll_reqs (unsigned int maxreqs)
399{
400 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
401 max_poll_reqs = maxreqs;
402 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
403}
404
405void eio_set_max_idle (unsigned int nthreads)
406{
407 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
408 max_idle = nthreads <= 0 ? 1 : nthreads;
409 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
410}
411
412void eio_set_min_parallel (unsigned int nthreads)
413{
414 if (wanted < nthreads)
415 wanted = nthreads;
416}
417
418void eio_set_max_parallel (unsigned int nthreads)
419{
420 if (wanted > nthreads)
421 wanted = nthreads;
422
423 while (started > wanted)
424 end_thread ();
425}
426
427int eio_poll (void) 496static int etp_poll (void)
428{ 497{
429 int maxreqs = max_poll_reqs; 498 unsigned int maxreqs;
499 unsigned int maxtime;
430 struct timeval tv_start, tv_now; 500 struct timeval tv_start, tv_now;
431 eio_req *req;
432 501
502 X_LOCK (reslock);
503 maxreqs = max_poll_reqs;
504 maxtime = max_poll_time;
505 X_UNLOCK (reslock);
506
433 if (max_poll_time) 507 if (maxtime)
434 gettimeofday (&tv_start, 0); 508 gettimeofday (&tv_start, 0);
435 509
436 for (;;) 510 for (;;)
437 { 511 {
512 ETP_REQ *req;
513
438 maybe_start_thread (); 514 etp_maybe_start_thread ();
439 515
440 X_LOCK (reslock); 516 X_LOCK (reslock);
441 req = reqq_shift (&res_queue); 517 req = reqq_shift (&res_queue);
442 518
443 if (req) 519 if (req)
… …
451 X_UNLOCK (reslock); 527 X_UNLOCK (reslock);
452 528
453 if (!req) 529 if (!req)
454 return 0; 530 return 0;
455 531
532 X_LOCK (reqlock);
456 --nreqs; 533 --nreqs;
534 X_UNLOCK (reqlock);
457 535
458 if (req->type == EIO_GROUP && req->size) 536 if (expect_false (req->type == EIO_GROUP && req->size))
459 { 537 {
460 req->int1 = 1; /* mark request as delayed */ 538 req->int1 = 1; /* mark request as delayed */
461 continue; 539 continue;
462 } 540 }
463 else 541 else
464 { 542 {
465 int res = eio_finish (req); 543 int res = ETP_FINISH (req);
466 if (res) 544 if (expect_false (res))
467 return res; 545 return res;
468 } 546 }
469 547
470 if (maxreqs && !--maxreqs) 548 if (expect_false (maxreqs && !--maxreqs))
471 break; 549 break;
472 550
473 if (max_poll_time) 551 if (maxtime)
474 { 552 {
475 gettimeofday (&tv_now, 0); 553 gettimeofday (&tv_now, 0);
476 554
477 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 555 if (tvdiff (&tv_start, &tv_now) >= maxtime)
478 break; 556 break;
479 } 557 }
480 } 558 }
481 559
482 errno = EAGAIN; 560 errno = EAGAIN;
483 return -1; 561 return -1;
484} 562}
485 563
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
486/*****************************************************************************/ 779/*****************************************************************************/
487/* work around various missing functions */ 780/* work around various missing functions */
488 781
489#if !HAVE_PREADWRITE 782#if !HAVE_PREADWRITE
783# undef pread
784# undef pwrite
490# define pread aio_pread 785# define pread eio__pread
491# define pwrite aio_pwrite 786# define pwrite eio__pwrite
492 787
493/* 788static ssize_t
494 * make our pread/pwrite safe against themselves, but not against 789eio__pread (int fd, void *buf, size_t count, off_t offset)
495 * normal read/write by using a mutex. slows down execution a lot,
496 * but that's your problem, not mine.
497 */
498static mutex_t preadwritelock = X_MUTEX_INIT;
499
500static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
501{ 790{
502 ssize_t res; 791 ssize_t res;
503 off_t ooffset; 792 off_t ooffset;
504 793
505 X_LOCK (preadwritelock); 794 X_LOCK (preadwritelock);
… …
510 X_UNLOCK (preadwritelock); 799 X_UNLOCK (preadwritelock);
511 800
512 return res; 801 return res;
513} 802}
514 803
515static 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)
516{ 806{
517 ssize_t res; 807 ssize_t res;
518 off_t ooffset; 808 off_t ooffset;
519 809
520 X_LOCK (preadwritelock); 810 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR); 811 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET); 812 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count); 813 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET); 814 lseek (fd, ooffset, SEEK_SET);
525 X_UNLOCK (preadwritelock); 815 X_UNLOCK (preadwritelock);
526 816
527 return res; 817 return res;
528} 818}
529#endif 819#endif
530 820
531#ifndef HAVE_FUTIMES 821#ifndef HAVE_UTIMES
532 822
823# undef utimes
533# define utimes(path,times) aio_utimes (path, times) 824# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) aio_futimes (fd, times)
535 825
826static int
536static int aio_utimes (const char *filename, const struct timeval times[2]) 827eio__utimes (const char *filename, const struct timeval times[2])
537{ 828{
538 if (times) 829 if (times)
539 { 830 {
540 struct utimbuf buf; 831 struct utimbuf buf;
541 832
… …
546 } 837 }
547 else 838 else
548 return utime (filename, 0); 839 return utime (filename, 0);
549} 840}
550 841
842#endif
843
844#ifndef HAVE_FUTIMES
845
846# undef futimes
847# define futimes(fd,times) eio__futimes (fd, times)
848
551static int aio_futimes (int fd, const struct timeval tv[2]) 849static int eio__futimes (int fd, const struct timeval tv[2])
552{ 850{
553 errno = ENOSYS; 851 errno = ENOSYS;
554 return -1; 852 return -1;
555} 853}
556 854
557#endif 855#endif
558 856
559#if !HAVE_FDATASYNC 857#if !HAVE_FDATASYNC
858# undef fdatasync
560# define fdatasync fsync 859# define fdatasync(fd) fsync (fd)
561#endif 860#endif
861
862/* sync_file_range always needs emulation */
863int
864eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
865{
866#if HAVE_SYNC_FILE_RANGE
867 int res;
868
869 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
870 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
871 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
872 {
873 flags = 0
874 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
875 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
876 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
877 }
878
879 res = sync_file_range (fd, offset, nbytes, flags);
880
881 if (!res || errno != ENOSYS)
882 return res;
883#endif
884
885 /* even though we could play tricks with the flags, it's better to always
886 * call fdatasync, as that matches the expectation of its users best */
887 return fdatasync (fd);
888}
562 889
563#if !HAVE_READAHEAD 890#if !HAVE_READAHEAD
891# undef readahead
564# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 892# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565 893
894static ssize_t
566static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 895eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
567{ 896{
568 size_t todo = count; 897 size_t todo = count;
569 dBUF; 898 dBUF;
570 899
571 while (todo > 0) 900 while (todo > 0)
… …
581 return count; 910 return count;
582} 911}
583 912
584#endif 913#endif
585 914
586#if !HAVE_READDIR_R
587# define readdir_r aio_readdir_r
588
589static mutex_t readdirlock = X_MUTEX_INIT;
590
591static int readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
592{
593 EIO_STRUCT_DIRENT *e;
594 int errorno;
595
596 X_LOCK (readdirlock);
597
598 e = readdir (dirp);
599 errorno = errno;
600
601 if (e)
602 {
603 *res = ent;
604 strcpy (ent->d_name, e->d_name);
605 }
606 else
607 *res = 0;
608
609 X_UNLOCK (readdirlock);
610
611 errno = errorno;
612 return e ? 0 : -1;
613}
614#endif
615
616/* sendfile always needs emulation */ 915/* sendfile always needs emulation */
916static ssize_t
617static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 917eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
618{ 918{
619 ssize_t res; 919 ssize_t res;
620 920
621 if (!count) 921 if (!count)
622 return 0; 922 return 0;
623 923
624#if HAVE_SENDFILE 924#if HAVE_SENDFILE
625# if __linux 925# if __linux
626 res = sendfile (ofd, ifd, &offset, count); 926 res = sendfile (ofd, ifd, &offset, count);
627 927
628# elif __freebsd 928# elif __FreeBSD__
629 /* 929 /*
630 * Of course, the freebsd sendfile is a dire hack with no thoughts 930 * Of course, the freebsd sendfile is a dire hack with no thoughts
631 * wasted on making it similar to other I/O functions. 931 * wasted on making it similar to other I/O functions.
632 */ 932 */
633 { 933 {
634 off_t sbytes; 934 off_t sbytes;
635 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 935 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636 936
637 if (res < 0 && sbytes) 937 #if 0 /* according to the manpage, this is correct, but broken behaviour */
638 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 938 /* freebsd' sendfile will return 0 on success */
939 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
940 /* not on e.g. EIO or EPIPE - sounds broken */
941 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
942 res = sbytes;
943 #endif
944
945 /* according to source inspection, this is correct, and useful behaviour */
946 if (sbytes)
947 res = sbytes;
948 }
949
950# elif defined (__APPLE__)
951
952 {
953 off_t sbytes = count;
954 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
955
956 /* according to the manpage, sbytes is always valid */
957 if (sbytes)
639 res = sbytes; 958 res = sbytes;
640 } 959 }
641 960
642# elif __hpux 961# elif __hpux
643 res = sendfile (ofd, ifd, offset, count, 0, 0); 962 res = sendfile (ofd, ifd, offset, count, 0, 0);
… …
657 if (res < 0 && sbytes) 976 if (res < 0 && sbytes)
658 res = sbytes; 977 res = sbytes;
659 } 978 }
660 979
661# endif 980# endif
981
982#elif defined (_WIN32)
983
984 /* does not work, just for documentation of what would need to be done */
985 {
986 HANDLE h = TO_SOCKET (ifd);
987 SetFilePointer (h, offset, 0, FILE_BEGIN);
988 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
989 }
990
662#else 991#else
663 res = -1; 992 res = -1;
664 errno = ENOSYS; 993 errno = ENOSYS;
665#endif 994#endif
666 995
667 if (res < 0 996 if (res < 0
668 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 997 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
998 /* BSDs */
999#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1000 || errno == ENOTSUP
1001#endif
1002 || errno == EOPNOTSUPP /* BSDs */
669#if __solaris 1003#if __solaris
670 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1004 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
671#endif 1005#endif
672 ) 1006 )
673 ) 1007 )
… …
704 } 1038 }
705 1039
706 return res; 1040 return res;
707} 1041}
708 1042
1043static signed char
1044eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1045{
1046 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1047 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1048}
1049
1050#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1051
1052#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1053#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1054
1055static void
1056eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1057{
1058 unsigned char bits [9 + sizeof (ino_t) * 8];
1059 unsigned char *bit = bits;
1060
1061 assert (CHAR_BIT == 8);
1062 assert (sizeof (eio_dirent) * 8 < 256);
1063 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1064 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1065
1066 if (size <= EIO_SORT_FAST)
1067 return;
1068
1069 /* first prepare an array of bits to test in our radix sort */
1070 /* try to take endianness into account, as well as differences in ino_t sizes */
1071 /* inode_bits must contain all inodes ORed together */
1072 /* which is used to skip bits that are 0 everywhere, which is very common */
1073 {
1074 ino_t endianness;
1075 int i, j;
1076
1077 /* we store the byte offset of byte n into byte n of "endianness" */
1078 for (i = 0; i < sizeof (ino_t); ++i)
1079 ((unsigned char *)&endianness)[i] = i;
1080
1081 *bit++ = 0;
1082
1083 for (i = 0; i < sizeof (ino_t); ++i)
1084 {
1085 /* shifting off the byte offsets out of "endianness" */
1086 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1087 endianness >>= 8;
1088
1089 for (j = 0; j < 8; ++j)
1090 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1091 *bit++ = offs + j;
1092 }
1093
1094 for (j = 0; j < 8; ++j)
1095 if (score_bits & (1 << j))
1096 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1097 }
1098
1099 /* now actually do the sorting (a variant of MSD radix sort) */
1100 {
1101 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1102 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1103 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1104 int stk_idx = 0;
1105
1106 base_stk [stk_idx] = dents;
1107 end_stk [stk_idx] = dents + size;
1108 bit_stk [stk_idx] = bit - 1;
1109
1110 do
1111 {
1112 base = base_stk [stk_idx];
1113 end = end_stk [stk_idx];
1114 bit = bit_stk [stk_idx];
1115
1116 for (;;)
1117 {
1118 unsigned char O = *bit >> 3;
1119 unsigned char M = 1 << (*bit & 7);
1120
1121 eio_dirent *a = base;
1122 eio_dirent *b = end;
1123
1124 if (b - a < EIO_SORT_CUTOFF)
1125 break;
1126
1127 /* now bit-partition the array on the bit */
1128 /* this ugly asymmetric loop seems to perform much better than typical */
1129 /* partition algos found in the literature */
1130 do
1131 if (!(((unsigned char *)a)[O] & M))
1132 ++a;
1133 else if (!(((unsigned char *)--b)[O] & M))
1134 {
1135 eio_dirent tmp = *a; *a = *b; *b = tmp;
1136 ++a;
1137 }
1138 while (b > a);
1139
1140 /* next bit, or stop, if no bits left in this path */
1141 if (!*--bit)
1142 break;
1143
1144 base_stk [stk_idx] = a;
1145 end_stk [stk_idx] = end;
1146 bit_stk [stk_idx] = bit;
1147 ++stk_idx;
1148
1149 end = a;
1150 }
1151 }
1152 while (stk_idx--);
1153 }
1154}
1155
1156static void
1157eio_dent_insertion_sort (eio_dirent *dents, int size)
1158{
1159 /* first move the smallest element to the front, to act as a sentinel */
1160 {
1161 int i;
1162 eio_dirent *min = dents;
1163
1164 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1165 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1166 if (EIO_DENT_CMP (dents [i], <, *min))
1167 min = &dents [i];
1168
1169 /* swap elements 0 and j (minimum) */
1170 {
1171 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1172 }
1173 }
1174
1175 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1176 {
1177 eio_dirent *i, *j;
1178
1179 for (i = dents + 1; i < dents + size; ++i)
1180 {
1181 eio_dirent value = *i;
1182
1183 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1184 j [1] = j [0];
1185
1186 j [1] = value;
1187 }
1188 }
1189}
1190
1191static void
1192eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1193{
1194 if (size <= 1)
1195 return; /* our insertion sort relies on size > 0 */
1196
1197 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1198 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1199 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1200
1201 /* use an insertion sort at the end, or for small arrays, */
1202 /* as insertion sort is more efficient for small partitions */
1203 eio_dent_insertion_sort (dents, size);
1204}
1205
709/* read a full directory */ 1206/* read a full directory */
1207static void
710static void scandir_ (eio_req *req, worker *self) 1208eio__scandir (eio_req *req, etp_worker *self)
711{ 1209{
712 DIR *dirp; 1210 DIR *dirp;
713 union
714 {
715 EIO_STRUCT_DIRENT d;
716 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
717 } *u;
718 EIO_STRUCT_DIRENT *entp; 1211 EIO_STRUCT_DIRENT *entp;
719 char *name, *names; 1212 char *name, *names;
720 int memlen = 4096; 1213 int namesalloc = 4096;
721 int memofs = 0; 1214 int namesoffs = 0;
1215 int flags = req->int1;
1216 eio_dirent *dents = 0;
1217 int dentalloc = 128;
722 int res = 0; 1218 int dentoffs = 0;
1219 ino_t inode_bits = 0;
1220
1221 req->result = -1;
1222
1223 if (!(flags & EIO_READDIR_DENTS))
1224 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
723 1225
724 X_LOCK (wrklock); 1226 X_LOCK (wrklock);
1227 /* the corresponding closedir is in ETP_WORKER_CLEAR */
725 self->dirp = dirp = opendir (req->ptr1); 1228 self->dirp = dirp = opendir (req->ptr1);
726 self->dbuf = u = malloc (sizeof (*u));
727 req->flags |= EIO_FLAG_PTR2_FREE; 1229 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1230 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
728 req->ptr2 = names = malloc (memlen); 1231 req->ptr2 = names = malloc (namesalloc);
729 X_UNLOCK (wrklock); 1232 X_UNLOCK (wrklock);
730 1233
731 if (dirp && u && names) 1234 if (dirp && names && (!flags || dents))
732 for (;;) 1235 for (;;)
733 { 1236 {
734 errno = 0; 1237 errno = 0;
735 readdir_r (dirp, &u->d, &entp); 1238 entp = readdir (dirp);
736 1239
737 if (!entp) 1240 if (!entp)
1241 {
1242 if (errno)
1243 break;
1244
1245 /* sort etc. */
1246 req->int1 = flags;
1247 req->result = dentoffs;
1248
1249 if (flags & EIO_READDIR_STAT_ORDER)
1250 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
1251 else if (flags & EIO_READDIR_DIRS_FIRST)
1252 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1253 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1254 else
1255 {
1256 /* in this case, all is known, and we just put dirs first and sort them */
1257 eio_dirent *oth = dents + dentoffs;
1258 eio_dirent *dir = dents;
1259
1260 /* now partition dirs to the front, and non-dirs to the back */
1261 /* by walking from both sides and swapping if necessary */
1262 /* also clear score, so it doesn't influence sorting */
1263 while (oth > dir)
1264 {
1265 if (dir->type == EIO_DT_DIR)
1266 ++dir;
1267 else if ((--oth)->type == EIO_DT_DIR)
1268 {
1269 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1270
1271 ++dir;
1272 }
1273 }
1274
1275 /* now sort the dirs only */
1276 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1277 }
1278
738 break; 1279 break;
1280 }
739 1281
1282 /* now add the entry to our list(s) */
740 name = entp->d_name; 1283 name = entp->d_name;
741 1284
1285 /* skip . and .. entries */
742 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1286 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
743 { 1287 {
744 int len = strlen (name) + 1; 1288 int len = D_NAMLEN (entp) + 1;
745 1289
746 res++; 1290 while (expect_false (namesoffs + len > namesalloc))
747
748 while (memofs + len > memlen)
749 { 1291 {
750 memlen *= 2; 1292 namesalloc *= 2;
751 X_LOCK (wrklock); 1293 X_LOCK (wrklock);
752 req->ptr2 = names = realloc (names, memlen); 1294 req->ptr2 = names = realloc (names, namesalloc);
753 X_UNLOCK (wrklock); 1295 X_UNLOCK (wrklock);
754 1296
755 if (!names) 1297 if (!names)
756 break; 1298 break;
757 } 1299 }
758 1300
759 memcpy (names + memofs, name, len); 1301 memcpy (names + namesoffs, name, len);
1302
1303 if (dents)
1304 {
1305 struct eio_dirent *ent;
1306
1307 if (expect_false (dentoffs == dentalloc))
1308 {
1309 dentalloc *= 2;
1310 X_LOCK (wrklock);
1311 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1312 X_UNLOCK (wrklock);
1313
1314 if (!dents)
1315 break;
1316 }
1317
1318 ent = dents + dentoffs;
1319
1320 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1321 ent->namelen = len - 1;
1322 ent->inode = D_INO (entp);
1323
1324 inode_bits |= ent->inode;
1325
1326 switch (D_TYPE (entp))
1327 {
1328 default:
1329 ent->type = EIO_DT_UNKNOWN;
1330 flags |= EIO_READDIR_FOUND_UNKNOWN;
1331 break;
1332
1333 #ifdef DT_FIFO
1334 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1335 #endif
1336 #ifdef DT_CHR
1337 case DT_CHR: ent->type = EIO_DT_CHR; break;
1338 #endif
1339 #ifdef DT_MPC
1340 case DT_MPC: ent->type = EIO_DT_MPC; break;
1341 #endif
1342 #ifdef DT_DIR
1343 case DT_DIR: ent->type = EIO_DT_DIR; break;
1344 #endif
1345 #ifdef DT_NAM
1346 case DT_NAM: ent->type = EIO_DT_NAM; break;
1347 #endif
1348 #ifdef DT_BLK
1349 case DT_BLK: ent->type = EIO_DT_BLK; break;
1350 #endif
1351 #ifdef DT_MPB
1352 case DT_MPB: ent->type = EIO_DT_MPB; break;
1353 #endif
1354 #ifdef DT_REG
1355 case DT_REG: ent->type = EIO_DT_REG; break;
1356 #endif
1357 #ifdef DT_NWK
1358 case DT_NWK: ent->type = EIO_DT_NWK; break;
1359 #endif
1360 #ifdef DT_CMP
1361 case DT_CMP: ent->type = EIO_DT_CMP; break;
1362 #endif
1363 #ifdef DT_LNK
1364 case DT_LNK: ent->type = EIO_DT_LNK; break;
1365 #endif
1366 #ifdef DT_SOCK
1367 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1368 #endif
1369 #ifdef DT_DOOR
1370 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1371 #endif
1372 #ifdef DT_WHT
1373 case DT_WHT: ent->type = EIO_DT_WHT; break;
1374 #endif
1375 }
1376
1377 ent->score = 7;
1378
1379 if (flags & EIO_READDIR_DIRS_FIRST)
1380 {
1381 if (ent->type == EIO_DT_UNKNOWN)
1382 {
1383 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1384 ent->score = 1;
1385 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1386 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1387 }
1388 else if (ent->type == EIO_DT_DIR)
1389 ent->score = 0;
1390 }
1391 }
1392
760 memofs += len; 1393 namesoffs += len;
1394 ++dentoffs;
1395 }
1396
1397 if (EIO_CANCELLED (req))
1398 {
1399 errno = ECANCELED;
1400 break;
761 } 1401 }
762 } 1402 }
1403}
763 1404
764 if (errno) 1405#ifdef PAGESIZE
765 res = -1; 1406# define eio_pagesize() PAGESIZE
1407#else
1408static intptr_t
1409eio_pagesize (void)
1410{
1411 static intptr_t page;
1412
1413 if (!page)
1414 page = sysconf (_SC_PAGESIZE);
1415
1416 return page;
1417}
1418#endif
1419
1420static void
1421eio_page_align (void **addr, size_t *length)
1422{
1423 intptr_t mask = eio_pagesize () - 1;
1424
1425 /* round down addr */
1426 intptr_t adj = mask & (intptr_t)*addr;
1427
1428 *addr = (void *)((intptr_t)*addr - adj);
1429 *length += adj;
1430
1431 /* round up length */
1432 *length = (*length + mask) & ~mask;
1433}
1434
1435#if !_POSIX_MEMLOCK
1436# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1437# define eio__mlockall(a) ((errno = ENOSYS), -1)
1438#else
1439
1440static int
1441eio__mlock (void *addr, size_t length)
1442{
1443 eio_page_align (&addr, &length);
1444
1445 return mlock (addr, length);
1446}
1447
1448static int
1449eio__mlockall (int flags)
1450{
1451 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1452 extern int mallopt (int, int);
1453 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1454 #endif
1455
1456 if (EIO_MCL_CURRENT != MCL_CURRENT
1457 || EIO_MCL_FUTURE != MCL_FUTURE)
1458 {
1459 flags = 0
1460 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1461 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1462 }
1463
1464 return mlockall (flags);
1465}
1466#endif
1467
1468#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1469# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1470#else
1471
1472int
1473eio__msync (void *mem, size_t len, int flags)
1474{
1475 eio_page_align (&mem, &len);
1476
1477 if (EIO_MS_ASYNC != MS_SYNC
1478 || EIO_MS_INVALIDATE != MS_INVALIDATE
1479 || EIO_MS_SYNC != MS_SYNC)
1480 {
1481 flags = 0
1482 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1483 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1484 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1485 }
1486
1487 return msync (mem, len, flags);
1488}
1489
1490#endif
1491
1492int
1493eio__mtouch (void *mem, size_t len, int flags)
1494{
1495 eio_page_align (&mem, &len);
1496
766 1497 {
767 req->result = res; 1498 intptr_t addr = (intptr_t)mem;
1499 intptr_t end = addr + len;
1500 intptr_t page = eio_pagesize ();
1501
1502 if (addr < end)
1503 if (flags & EIO_MT_MODIFY) /* modify */
1504 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1505 else
1506 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1507 }
1508
1509 return 0;
768} 1510}
769 1511
770/*****************************************************************************/ 1512/*****************************************************************************/
771 1513
772#define ALLOC(len) \ 1514#define ALLOC(len) \
… …
782 req->result = -1; \ 1524 req->result = -1; \
783 break; \ 1525 break; \
784 } \ 1526 } \
785 } 1527 }
786 1528
787X_THREAD_PROC (eio_proc) 1529X_THREAD_PROC (etp_proc)
788{ 1530{
789 eio_req *req; 1531 ETP_REQ *req;
790 struct timespec ts; 1532 struct timespec ts;
791 worker *self = (worker *)thr_arg; 1533 etp_worker *self = (etp_worker *)thr_arg;
792 1534
793 /* try to distribute timeouts somewhat randomly */ 1535 /* try to distribute timeouts somewhat randomly */
794 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1536 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
795 1537
796 for (;;) 1538 for (;;)
797 { 1539 {
798 ts.tv_sec = time (0) + IDLE_TIMEOUT;
799
800 X_LOCK (reqlock); 1540 X_LOCK (reqlock);
801 1541
802 for (;;) 1542 for (;;)
803 { 1543 {
804 self->req = req = reqq_shift (&req_queue); 1544 self->req = req = reqq_shift (&req_queue);
… …
806 if (req) 1546 if (req)
807 break; 1547 break;
808 1548
809 ++idle; 1549 ++idle;
810 1550
1551 ts.tv_sec = time (0) + IDLE_TIMEOUT;
811 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1552 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
812 { 1553 {
813 if (idle > max_idle) 1554 if (idle > max_idle)
814 { 1555 {
815 --idle; 1556 --idle;
… …
820 goto quit; 1561 goto quit;
821 } 1562 }
822 1563
823 /* we are allowed to idle, so do so without any timeout */ 1564 /* we are allowed to idle, so do so without any timeout */
824 X_COND_WAIT (reqwait, reqlock); 1565 X_COND_WAIT (reqwait, reqlock);
825 ts.tv_sec = time (0) + IDLE_TIMEOUT;
826 } 1566 }
827 1567
828 --idle; 1568 --idle;
829 } 1569 }
830 1570
831 --nready; 1571 --nready;
832 1572
833 X_UNLOCK (reqlock); 1573 X_UNLOCK (reqlock);
834 1574
835 errno = 0; /* strictly unnecessary */ 1575 if (req->type < 0)
1576 goto quit;
836 1577
837 if (!EIO_CANCELLED (req)) 1578 if (!EIO_CANCELLED (req))
838 switch (req->type) 1579 ETP_EXECUTE (self, req);
839 {
840 case EIO_READ: ALLOC (req->size);
841 req->result = req->offs >= 0
842 ? pread (req->int1, req->ptr2, req->size, req->offs)
843 : read (req->int1, req->ptr2, req->size); break;
844 case EIO_WRITE: req->result = req->offs >= 0
845 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
846 : write (req->int1, req->ptr2, req->size); break;
847
848 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
849 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
850
851 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
852 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
853 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
854 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
855 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
856 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
857
858 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
859 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
860 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
861 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
862 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
863 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
864
865 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
866 case EIO_CLOSE: req->result = close (req->int1); break;
867 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
868 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
869 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
870 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
871 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
872 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
873 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
874 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
875
876 case EIO_READLINK: ALLOC (NAME_MAX);
877 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
878
879 case EIO_SYNC: req->result = 0; sync (); break;
880 case EIO_FSYNC: req->result = fsync (req->int1); break;
881 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
882
883 case EIO_READDIR: scandir_ (req, self); break;
884
885 case EIO_BUSY:
886#ifdef _WIN32
887 Sleep (req->nv1 * 1000.);
888#else
889 {
890 struct timeval tv;
891
892 tv.tv_sec = req->nv1;
893 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
894
895 req->result = select (0, 0, 0, 0, &tv);
896 }
897#endif
898 break;
899
900 case EIO_UTIME:
901 case EIO_FUTIME:
902 {
903 struct timeval tv[2];
904 struct timeval *times;
905
906 if (req->nv1 != -1. || req->nv2 != -1.)
907 {
908 tv[0].tv_sec = req->nv1;
909 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
910 tv[1].tv_sec = req->nv2;
911 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
912
913 times = tv;
914 }
915 else
916 times = 0;
917
918
919 req->result = req->type == EIO_FUTIME
920 ? futimes (req->int1, times)
921 : utimes (req->ptr1, times);
922 }
923
924 case EIO_GROUP:
925 case EIO_NOP:
926 break;
927
928 case EIO_QUIT:
929 goto quit;
930
931 default:
932 req->result = -1;
933 break;
934 }
935
936 req->errorno = errno;
937 1580
938 X_LOCK (reslock); 1581 X_LOCK (reslock);
939 1582
940 ++npending; 1583 ++npending;
941 1584
942 if (!reqq_push (&res_queue, req) && want_poll_cb) 1585 if (!reqq_push (&res_queue, req) && want_poll_cb)
943 want_poll_cb (); 1586 want_poll_cb ();
944 1587
945 self->req = 0; 1588 self->req = 0;
946 worker_clear (self); 1589 etp_worker_clear (self);
947 1590
948 X_UNLOCK (reslock); 1591 X_UNLOCK (reslock);
949 } 1592 }
950 1593
951quit: 1594quit:
952 X_LOCK (wrklock); 1595 X_LOCK (wrklock);
953 worker_free (self); 1596 etp_worker_free (self);
954 X_UNLOCK (wrklock); 1597 X_UNLOCK (wrklock);
955 1598
956 return 0; 1599 return 0;
957} 1600}
958 1601
959/*****************************************************************************/ 1602/*****************************************************************************/
960 1603
961static void eio_atfork_prepare (void)
962{
963 X_LOCK (wrklock);
964 X_LOCK (reqlock);
965 X_LOCK (reslock);
966#if !HAVE_PREADWRITE
967 X_LOCK (preadwritelock);
968#endif
969#if !HAVE_READDIR_R
970 X_LOCK (readdirlock);
971#endif
972}
973
974static void eio_atfork_parent (void)
975{
976#if !HAVE_READDIR_R
977 X_UNLOCK (readdirlock);
978#endif
979#if !HAVE_PREADWRITE
980 X_UNLOCK (preadwritelock);
981#endif
982 X_UNLOCK (reslock);
983 X_UNLOCK (reqlock);
984 X_UNLOCK (wrklock);
985}
986
987static void eio_atfork_child (void)
988{
989 eio_req *prv;
990
991 while (prv = reqq_shift (&req_queue))
992 eio_destroy (prv);
993
994 while (prv = reqq_shift (&res_queue))
995 eio_destroy (prv);
996
997 while (wrk_first.next != &wrk_first)
998 {
999 worker *wrk = wrk_first.next;
1000
1001 if (wrk->req)
1002 eio_destroy (wrk->req);
1003
1004 worker_clear (wrk);
1005 worker_free (wrk);
1006 }
1007
1008 started = 0;
1009 idle = 0;
1010 nreqs = 0;
1011 nready = 0;
1012 npending = 0;
1013
1014 eio_atfork_parent ();
1015}
1016
1017int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1604int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1018{ 1605{
1019 want_poll_cb = want_poll; 1606 return etp_init (want_poll, done_poll);
1020 done_poll_cb = done_poll;
1021
1022#ifdef _WIN32
1023 X_MUTEX_CHECK (wrklock);
1024 X_MUTEX_CHECK (reslock);
1025 X_MUTEX_CHECK (reqlock);
1026 X_MUTEX_CHECK (reqwait);
1027 X_MUTEX_CHECK (preadwritelock);
1028 X_MUTEX_CHECK (readdirlock);
1029
1030 X_COND_CHECK (reqwait);
1031#endif
1032
1033 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1034} 1607}
1035 1608
1036static void eio_api_destroy (eio_req *req) 1609static void eio_api_destroy (eio_req *req)
1037{ 1610{
1038 free (req); 1611 free (req);
… …
1043 \ 1616 \
1044 req = (eio_req *)calloc (1, sizeof *req); \ 1617 req = (eio_req *)calloc (1, sizeof *req); \
1045 if (!req) \ 1618 if (!req) \
1046 return 0; \ 1619 return 0; \
1047 \ 1620 \
1048 req->type = EIO_ ## rtype; \ 1621 req->type = rtype; \
1049 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 1622 req->pri = pri; \
1050 req->finish = cb; \ 1623 req->finish = cb; \
1624 req->data = data; \
1051 req->destroy = eio_api_destroy; 1625 req->destroy = eio_api_destroy;
1052 1626
1053#define SEND eio_submit (req); return req 1627#define SEND eio_submit (req); return req
1054 1628
1055#define PATH (void)0 1629#define PATH \
1630 req->flags |= EIO_FLAG_PTR1_FREE; \
1631 req->ptr1 = strdup (path); \
1632 if (!req->ptr1) \
1633 { \
1634 eio_api_destroy (req); \
1635 return 0; \
1636 }
1056 1637
1638static void eio_execute (etp_worker *self, eio_req *req)
1639{
1640 switch (req->type)
1641 {
1642 case EIO_READ: ALLOC (req->size);
1643 req->result = req->offs >= 0
1644 ? pread (req->int1, req->ptr2, req->size, req->offs)
1645 : read (req->int1, req->ptr2, req->size); break;
1646 case EIO_WRITE: req->result = req->offs >= 0
1647 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1648 : write (req->int1, req->ptr2, req->size); break;
1649
1650 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1651 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1652
1653 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1654 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1655 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1656 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1657 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1658 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1659
1660 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1661 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1662 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1663 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1664
1665 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1666 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1667 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1668 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1669 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1670 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1671
1672 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1673 case EIO_CLOSE: req->result = close (req->int1); break;
1674 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1675 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1676 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1677 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1678 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1679 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1680 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1681 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1682
1683 case EIO_READLINK: ALLOC (PATH_MAX);
1684 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1685
1686 case EIO_SYNC: req->result = 0; sync (); break;
1687 case EIO_FSYNC: req->result = fsync (req->int1); break;
1688 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1689 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1690 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1691 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1692 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1693 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1694
1695 case EIO_READDIR: eio__scandir (req, self); break;
1696
1697 case EIO_BUSY:
1698#ifdef _WIN32
1699 Sleep (req->nv1 * 1e3);
1700#else
1701 {
1702 struct timeval tv;
1703
1704 tv.tv_sec = req->nv1;
1705 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1706
1707 req->result = select (0, 0, 0, 0, &tv);
1708 }
1709#endif
1710 break;
1711
1712 case EIO_UTIME:
1713 case EIO_FUTIME:
1714 {
1715 struct timeval tv[2];
1716 struct timeval *times;
1717
1718 if (req->nv1 != -1. || req->nv2 != -1.)
1719 {
1720 tv[0].tv_sec = req->nv1;
1721 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1722 tv[1].tv_sec = req->nv2;
1723 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1724
1725 times = tv;
1726 }
1727 else
1728 times = 0;
1729
1730 req->result = req->type == EIO_FUTIME
1731 ? futimes (req->int1, times)
1732 : utimes (req->ptr1, times);
1733 }
1734 break;
1735
1736 case EIO_GROUP:
1737 abort (); /* handled in eio_request */
1738
1739 case EIO_NOP:
1740 req->result = 0;
1741 break;
1742
1743 case EIO_CUSTOM:
1744 ((void (*)(eio_req *))req->feed) (req);
1745 break;
1746
1747 default:
1748 errno = ENOSYS;
1749 req->result = -1;
1750 break;
1751 }
1752
1753 req->errorno = errno;
1754}
1755
1756#ifndef EIO_NO_WRAPPERS
1757
1758eio_req *eio_nop (int pri, eio_cb cb, void *data)
1759{
1760 REQ (EIO_NOP); SEND;
1761}
1762
1763eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1764{
1765 REQ (EIO_BUSY); req->nv1 = delay; SEND;
1766}
1767
1057eio_req *eio_fsync (int fd, eio_cb cb) 1768eio_req *eio_sync (int pri, eio_cb cb, void *data)
1058{ 1769{
1770 REQ (EIO_SYNC); SEND;
1771}
1772
1773eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1774{
1059 REQ (FSYNC); req->int1 = fd; SEND; 1775 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1060} 1776}
1061 1777
1778eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1779{
1780 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1781}
1782
1783eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1784{
1785 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1786}
1787
1788eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1789{
1790 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1791}
1792
1793eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1794{
1795 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1796}
1797
1798eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1799{
1800 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1801}
1802
1062eio_req *eio_fdatasync (int fd, eio_cb cb) 1803eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1063{ 1804{
1064 REQ (FDATASYNC); req->int1 = fd; SEND; 1805 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1065} 1806}
1066 1807
1808eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1809{
1810 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1811}
1812
1067eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb) 1813eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1068{ 1814{
1069 REQ (READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 1815 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1070} 1816}
1071 1817
1072eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb) 1818eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1073{ 1819{
1074 REQ (READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 1820 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1075} 1821}
1076 1822
1077eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb) 1823eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1078{ 1824{
1079 REQ (WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 1825 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1080} 1826}
1081 1827
1082eio_req *eio_fstat (int fd, eio_cb cb) 1828eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1083{ 1829{
1084 REQ (FSTAT); req->int1 = fd; SEND; 1830 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1085} 1831}
1086 1832
1833eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1834{
1835 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1836}
1837
1087eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb) 1838eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1088{ 1839{
1089 REQ (FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1840 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1090} 1841}
1091 1842
1092eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb) 1843eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1093{ 1844{
1094 REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 1845 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1095} 1846}
1096 1847
1097eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb) 1848eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1098{ 1849{
1099 REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 1850 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1100} 1851}
1101 1852
1102eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb) 1853eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1103{ 1854{
1104 REQ (FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 1855 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1105} 1856}
1106 1857
1107eio_req *eio_dup2 (int fd, int fd2, eio_cb cb) 1858eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1108{ 1859{
1109 REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND; 1860 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1110} 1861}
1111 1862
1112eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb) 1863eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1113{ 1864{
1114 REQ (SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 1865 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1115} 1866}
1116 1867
1117eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb) 1868eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1118{ 1869{
1119 REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 1870 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1120} 1871}
1121 1872
1122eio_req *eio_readlink (const char *path, eio_cb cb)
1123{
1124 REQ (READLINK); PATH; SEND;
1125}
1126
1127eio_req *eio_stat (const char *path, eio_cb cb)
1128{
1129 REQ (STAT); PATH; SEND;
1130}
1131
1132eio_req *eio_lstat (const char *path, eio_cb cb)
1133{
1134 REQ (LSTAT); PATH; SEND;
1135}
1136
1137eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb) 1873eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1138{ 1874{
1139 REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 1875 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1140} 1876}
1141 1877
1142eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb) 1878eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1143{ 1879{
1144 REQ (TRUNCATE); PATH; req->offs = offset; SEND; 1880 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1145} 1881}
1146 1882
1147eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 1883eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1148{ 1884{
1149 REQ (CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 1885 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1150} 1886}
1151 1887
1152eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb) 1888eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1153{ 1889{
1154 REQ (CHMOD); PATH; req->int2 = (long)mode; SEND; 1890 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1155} 1891}
1156 1892
1157eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb) 1893eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1158{ 1894{
1159 REQ (MKDIR); PATH; req->int2 = (long)mode; SEND; 1895 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1160} 1896}
1161 1897
1162eio_req *eio_unlink (const char *path, eio_cb cb) 1898static eio_req *
1899eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1163{ 1900{
1164 REQ (UNLINK); PATH; SEND;
1165}
1166
1167eio_req *eio_rmdir (const char *path, eio_cb cb)
1168{
1169 REQ (RMDIR); PATH; SEND; 1901 REQ (type); PATH; SEND;
1170} 1902}
1171 1903
1904eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1905{
1906 return eio__1path (EIO_READLINK, path, pri, cb, data);
1907}
1908
1909eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1910{
1911 return eio__1path (EIO_STAT, path, pri, cb, data);
1912}
1913
1914eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1915{
1916 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1917}
1918
1919eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1920{
1921 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1922}
1923
1924eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1925{
1926 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1927}
1928
1172eio_req *eio_readdir (const char *path, eio_cb cb) 1929eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1173{ 1930{
1174 REQ (READDIR); PATH; SEND; 1931 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1175} 1932}
1176 1933
1934eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1935{
1936 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1937}
1938
1177eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb) 1939eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1178{ 1940{
1179 REQ (MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1941 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1180} 1942}
1181 1943
1182eio_req *eio_busy (double delay, eio_cb cb) 1944static eio_req *
1945eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1183{ 1946{
1184 REQ (BUSY); req->nv1 = delay; SEND; 1947 REQ (type); PATH;
1185}
1186 1948
1949 req->flags |= EIO_FLAG_PTR2_FREE;
1950 req->ptr2 = strdup (new_path);
1951 if (!req->ptr2)
1952 {
1953 eio_api_destroy (req);
1954 return 0;
1955 }
1956
1957 SEND;
1958}
1959
1960eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1961{
1962 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
1963}
1964
1965eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1966{
1967 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
1968}
1969
1970eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1971{
1972 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1973}
1974
1975eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1976{
1977 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1978}
1979
1980#endif
1981
1187eio_req *eio_nop (eio_cb cb) 1982eio_req *eio_grp (eio_cb cb, void *data)
1188{ 1983{
1984 const int pri = EIO_PRI_MAX;
1985
1189 REQ (NOP); SEND; 1986 REQ (EIO_GROUP); SEND;
1190} 1987}
1191 1988
1192#undef REQ 1989#undef REQ
1193#undef PATH 1990#undef PATH
1194#undef SEND 1991#undef SEND
1195 1992
1196#if 0 1993/*****************************************************************************/
1197 1994/* grp functions */
1198void
1199aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1200 PROTOTYPE: $$$;$
1201 PPCODE:
1202{
1203 dREQ;
1204
1205 req->type = EIO_OPEN;
1206 req->sv1 = newSVsv (pathname);
1207 req->ptr1 = SvPVbyte_nolen (req->sv1);
1208 req->int1 = flags;
1209 req->int2 = mode;
1210
1211 EIO_SEND;
1212}
1213
1214void
1215aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1216 PROTOTYPE: $;$
1217 ALIAS:
1218 aio_fsync = EIO_FSYNC
1219 aio_fdatasync = EIO_FDATASYNC
1220 PPCODE:
1221{
1222 dREQ;
1223
1224 req->type = ix;
1225 req->sv1 = newSVsv (fh);
1226 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1227
1228 EIO_SEND (req);
1229}
1230
1231void
1232aio_close (SV *fh, SV *callback=&PL_sv_undef)
1233 PROTOTYPE: $;$
1234 PPCODE:
1235{
1236 dREQ;
1237
1238 req->type = EIO_CLOSE;
1239 req->sv1 = newSVsv (fh);
1240 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1241
1242 EIO_SEND (req);
1243}
1244
1245void
1246aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1247 ALIAS:
1248 aio_read = EIO_READ
1249 aio_write = EIO_WRITE
1250 PROTOTYPE: $$$$$;$
1251 PPCODE:
1252{
1253 STRLEN svlen;
1254 char *svptr = SvPVbyte (data, svlen);
1255 UV len = SvUV (length);
1256
1257 SvUPGRADE (data, SVt_PV);
1258 SvPOK_on (data);
1259
1260 if (dataoffset < 0)
1261 dataoffset += svlen;
1262
1263 if (dataoffset < 0 || dataoffset > svlen)
1264 croak ("dataoffset outside of data scalar");
1265
1266 if (ix == EIO_WRITE)
1267 {
1268 /* write: check length and adjust. */
1269 if (!SvOK (length) || len + dataoffset > svlen)
1270 len = svlen - dataoffset;
1271 }
1272 else
1273 {
1274 /* read: grow scalar as necessary */
1275 svptr = SvGROW (data, len + dataoffset + 1);
1276 }
1277
1278 if (len < 0)
1279 croak ("length must not be negative");
1280
1281 {
1282 dREQ;
1283
1284 req->type = ix;
1285 req->sv1 = newSVsv (fh);
1286 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1287 : IoOFP (sv_2io (fh)));
1288 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1289 req->size = len;
1290 req->sv2 = SvREFCNT_inc (data);
1291 req->ptr2 = (char *)svptr + dataoffset;
1292 req->stroffset = dataoffset;
1293
1294 if (!SvREADONLY (data))
1295 {
1296 SvREADONLY_on (data);
1297 req->flags |= FLAG_SV2_RO_OFF;
1298 }
1299
1300 EIO_SEND;
1301 }
1302}
1303
1304void
1305aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1306 PROTOTYPE: $$;$
1307 PPCODE:
1308{
1309 SV *data;
1310 dREQ;
1311
1312 data = newSV (NAME_MAX);
1313 SvPOK_on (data);
1314
1315 req->type = EIO_READLINK;
1316 req->sv1 = newSVsv (path);
1317 req->ptr1 = SvPVbyte_nolen (req->sv1);
1318 req->sv2 = data;
1319 req->ptr2 = SvPVbyte_nolen (data);
1320
1321 EIO_SEND;
1322}
1323
1324void
1325aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1326 PROTOTYPE: $$$$;$
1327 PPCODE:
1328{
1329 dREQ;
1330
1331 req->type = EIO_SENDFILE;
1332 req->sv1 = newSVsv (out_fh);
1333 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1334 req->sv2 = newSVsv (in_fh);
1335 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1336 req->offs = SvVAL64 (in_offset);
1337 req->size = length;
1338
1339 EIO_SEND;
1340}
1341
1342void
1343aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1344 PROTOTYPE: $$$;$
1345 PPCODE:
1346{
1347 dREQ;
1348
1349 req->type = EIO_READAHEAD;
1350 req->sv1 = newSVsv (fh);
1351 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1352 req->offs = SvVAL64 (offset);
1353 req->size = length;
1354
1355 EIO_SEND;
1356}
1357
1358void
1359aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1360 ALIAS:
1361 aio_stat = EIO_STAT
1362 aio_lstat = EIO_LSTAT
1363 PPCODE:
1364{
1365 dREQ;
1366
1367 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1368 if (!req->ptr2)
1369 {
1370 req_destroy (req);
1371 croak ("out of memory during aio_stat statdata allocation");
1372 }
1373
1374 req->flags |= FLAG_PTR2_FREE;
1375 req->sv1 = newSVsv (fh_or_path);
1376
1377 if (SvPOK (fh_or_path))
1378 {
1379 req->type = ix;
1380 req->ptr1 = SvPVbyte_nolen (req->sv1);
1381 }
1382 else
1383 {
1384 req->type = EIO_FSTAT;
1385 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1386 }
1387
1388 EIO_SEND;
1389}
1390
1391void
1392aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1393 PPCODE:
1394{
1395 dREQ;
1396
1397 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1398 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1399 req->sv1 = newSVsv (fh_or_path);
1400
1401 if (SvPOK (fh_or_path))
1402 {
1403 req->type = EIO_UTIME;
1404 req->ptr1 = SvPVbyte_nolen (req->sv1);
1405 }
1406 else
1407 {
1408 req->type = EIO_FUTIME;
1409 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1410 }
1411
1412 EIO_SEND;
1413}
1414
1415void
1416aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1417 PPCODE:
1418{
1419 dREQ;
1420
1421 req->sv1 = newSVsv (fh_or_path);
1422 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1423
1424 if (SvPOK (fh_or_path))
1425 {
1426 req->type = EIO_TRUNCATE;
1427 req->ptr1 = SvPVbyte_nolen (req->sv1);
1428 }
1429 else
1430 {
1431 req->type = EIO_FTRUNCATE;
1432 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1433 }
1434
1435 EIO_SEND;
1436}
1437
1438void
1439aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1440 ALIAS:
1441 aio_chmod = EIO_CHMOD
1442 aio_fchmod = EIO_FCHMOD
1443 aio_mkdir = EIO_MKDIR
1444 PPCODE:
1445{
1446 dREQ;
1447
1448 req->type = type;
1449 req->int2 = mode;
1450 req->sv1 = newSVsv (fh_or_path);
1451
1452 if (ix == EIO_FCHMOD)
1453 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1454 else
1455 req->ptr1 = SvPVbyte_nolen (req->sv1);
1456
1457 EIO_SEND;
1458}
1459
1460void
1461aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1462 PPCODE:
1463{
1464 dREQ;
1465
1466 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1467 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1468 req->sv1 = newSVsv (fh_or_path);
1469
1470 if (SvPOK (fh_or_path))
1471 {
1472 req->type = EIO_CHOWN;
1473 req->ptr1 = SvPVbyte_nolen (req->sv1);
1474 }
1475 else
1476 {
1477 req->type = EIO_FCHOWN;
1478 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1479 }
1480
1481 EIO_SEND;
1482}
1483
1484void
1485aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1486 ALIAS:
1487 aio_unlink = EIO_UNLINK
1488 aio_rmdir = EIO_RMDIR
1489 aio_readdir = EIO_READDIR
1490 PPCODE:
1491{
1492 dREQ;
1493
1494 req->type = ix;
1495 req->sv1 = newSVsv (pathname);
1496 req->ptr1 = SvPVbyte_nolen (req->sv1);
1497
1498 EIO_SEND;
1499}
1500
1501void
1502aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1503 ALIAS:
1504 aio_link = EIO_LINK
1505 aio_symlink = EIO_SYMLINK
1506 aio_rename = EIO_RENAME
1507 PPCODE:
1508{
1509 dREQ;
1510
1511 req->type = ix;
1512 req->sv1 = newSVsv (oldpath);
1513 req->ptr1 = SvPVbyte_nolen (req->sv1);
1514 req->sv2 = newSVsv (newpath);
1515 req->ptr2 = SvPVbyte_nolen (req->sv2);
1516
1517 EIO_SEND;
1518}
1519
1520void
1521aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1522 PPCODE:
1523{
1524 dREQ;
1525
1526 req->type = EIO_MKNOD;
1527 req->sv1 = newSVsv (pathname);
1528 req->ptr1 = SvPVbyte_nolen (req->sv1);
1529 req->int2 = (mode_t)mode;
1530 req->offs = dev;
1531
1532 EIO_SEND;
1533}
1534
1535void
1536aio_busy (double delay, SV *callback=&PL_sv_undef)
1537 PPCODE:
1538{
1539 dREQ;
1540
1541 req->type = EIO_BUSY;
1542 req->nv1 = delay < 0. ? 0. : delay;
1543
1544 EIO_SEND;
1545}
1546
1547void
1548aio_group (SV *callback=&PL_sv_undef)
1549 PROTOTYPE: ;$
1550 PPCODE:
1551{
1552 dREQ;
1553
1554 req->type = EIO_GROUP;
1555
1556 req_send (req);
1557 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1558}
1559
1560void
1561aio_nop (SV *callback=&PL_sv_undef)
1562 ALIAS:
1563 aio_nop = EIO_NOP
1564 aio_sync = EIO_SYNC
1565 PPCODE:
1566{
1567 dREQ;
1568
1569#endif
1570 1995
1571void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 1996void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1572{ 1997{
1573 grp->int2 = limit; 1998 grp->int2 = limit;
1574 grp->feed = feed; 1999 grp->feed = feed;
… …
1585 2010
1586void eio_grp_add (eio_req *grp, eio_req *req) 2011void eio_grp_add (eio_req *grp, eio_req *req)
1587{ 2012{
1588 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2013 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1589 2014
2015 grp->flags |= EIO_FLAG_GROUPADD;
2016
1590 ++grp->size; 2017 ++grp->size;
1591 req->grp = grp; 2018 req->grp = grp;
1592 2019
1593 req->grp_prev = 0; 2020 req->grp_prev = 0;
1594 req->grp_next = grp->grp_first; 2021 req->grp_next = grp->grp_first;
… …
1597 grp->grp_first->grp_prev = req; 2024 grp->grp_first->grp_prev = req;
1598 2025
1599 grp->grp_first = req; 2026 grp->grp_first = req;
1600} 2027}
1601 2028
2029/*****************************************************************************/
2030/* misc garbage */
1602 2031
2032ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2033{
2034 etp_worker wrk;
2035 ssize_t ret;
2036
2037 wrk.dbuf = 0;
2038
2039 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2040
2041 if (wrk.dbuf)
2042 free (wrk.dbuf);
2043
2044 return ret;
2045}
2046

Diff Legend

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