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 1.54 by root, Tue Feb 23 14:32:45 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>
79# include <sys/mman.h>
36# include <unistd.h> 80# include <unistd.h>
37# include <utime.h> 81# include <utime.h>
38# include <signal.h> 82# include <signal.h>
39# include <dirent.h> 83# include <dirent.h>
40 84
85/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
86# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
87# define _DIRENT_HAVE_D_TYPE /* sigh */
88# define D_INO(de) (de)->d_fileno
89# define D_NAMLEN(de) (de)->d_namlen
90# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
91# define D_INO(de) (de)->d_ino
92# endif
93
94#ifdef _D_EXACT_NAMLEN
95# undef D_NAMLEN
96# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
97#endif
98
99# ifdef _DIRENT_HAVE_D_TYPE
100# define D_TYPE(de) (de)->d_type
101# endif
102
41# ifndef EIO_STRUCT_DIRENT 103# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 104# define EIO_STRUCT_DIRENT struct dirent
43# endif 105# endif
44 106
45#endif 107#endif
46 108
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif
50
51#if HAVE_SENDFILE 109#if HAVE_SENDFILE
52# if __linux 110# if __linux
53# include <sys/sendfile.h> 111# include <sys/sendfile.h>
54# elif __freebsd 112# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 113# include <sys/socket.h>
56# include <sys/uio.h> 114# include <sys/uio.h>
57# elif __hpux 115# elif __hpux
58# include <sys/socket.h> 116# include <sys/socket.h>
59# elif __solaris /* not yet */ 117# elif __solaris
60# include <sys/sendfile.h> 118# include <sys/sendfile.h>
61# else 119# else
62# error sendfile support requested but not available 120# error sendfile support requested but not available
63# endif 121# endif
64#endif 122#endif
65 123
124#ifndef D_TYPE
125# define D_TYPE(de) 0
126#endif
127#ifndef D_INO
128# define D_INO(de) 0
129#endif
130#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name)
132#endif
133
66/* number of seconds after which an idle threads exit */ 134/* number of seconds after which an idle threads exit */
67#define IDLE_TIMEOUT 10 135#define IDLE_TIMEOUT 10
68 136
69/* used for struct dirent, AIX doesn't provide it */ 137/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 138#ifndef NAME_MAX
71# define NAME_MAX 4096 139# define NAME_MAX 4096
72#endif 140#endif
73 141
142/* used for readlink etc. */
143#ifndef PATH_MAX
144# define PATH_MAX 4096
145#endif
146
74/* buffer size for various temporary buffers */ 147/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 148#define EIO_BUFSIZE 65536
76 149
77#define dBUF \ 150#define dBUF \
78 char *eio_buf; \ 151 char *eio_buf; \
79 X_LOCK (wrklock); \ 152 ETP_WORKER_LOCK (self); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 153 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 154 ETP_WORKER_UNLOCK (self); \
155 errno = ENOMEM; \
82 if (!eio_buf) \ 156 if (!eio_buf) \
83 return -1; 157 return -1;
84 158
85#define EIO_TICKS ((1000000 + 1023) >> 10) 159#define EIO_TICKS ((1000000 + 1023) >> 10)
86 160
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
87static void (*want_poll_cb) (void); 216static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 217static void (*done_poll_cb) (void);
89 218
90static unsigned int max_poll_time = 0; 219static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 220static unsigned int max_poll_reqs; /* reslock */
92 221
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 222static volatile unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 223static volatile unsigned int nready; /* reqlock */
95{ 224static volatile unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 225static volatile unsigned int max_idle = 4;
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
98}
99 226
100static unsigned int started, idle, wanted;
101
102/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT; 227static mutex_t wrklock = X_MUTEX_INIT;
104
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)
120{
121 if (wrk->dirp)
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}
133
134static void worker_free (worker *wrk)
135{
136 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next;
138
139 free (wrk);
140}
141
142static volatile unsigned int nreqs, nready, npending;
143static volatile unsigned int max_idle = 4;
144
145static mutex_t reslock = X_MUTEX_INIT; 228static mutex_t reslock = X_MUTEX_INIT;
146static mutex_t reqlock = X_MUTEX_INIT; 229static mutex_t reqlock = X_MUTEX_INIT;
147static cond_t reqwait = X_COND_INIT; 230static cond_t reqwait = X_COND_INIT;
148 231
232#if !HAVE_PREADWRITE
233/*
234 * make our pread/pwrite emulation safe against themselves, but not against
235 * normal read/write by using a mutex. slows down execution a lot,
236 * but that's your problem, not mine.
237 */
238static mutex_t preadwritelock = X_MUTEX_INIT;
239#endif
240
241typedef struct etp_worker
242{
243 /* locked by wrklock */
244 struct etp_worker *prev, *next;
245
246 thread_t tid;
247
248 /* locked by reslock, reqlock or wrklock */
249 ETP_REQ *req; /* currently processed request */
250
251 ETP_WORKER_COMMON
252} etp_worker;
253
254static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
255
256#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
257#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
258
259/* worker threads management */
260
261static void etp_worker_clear (etp_worker *wrk)
262{
263 ETP_WORKER_CLEAR (wrk);
264}
265
266static void etp_worker_free (etp_worker *wrk)
267{
268 wrk->next->prev = wrk->prev;
269 wrk->prev->next = wrk->next;
270
271 free (wrk);
272}
273
149unsigned int eio_nreqs (void) 274static unsigned int etp_nreqs (void)
150{ 275{
276 int retval;
277 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
278 retval = nreqs;
279 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 280 return retval;
152} 281}
153 282
154unsigned int eio_nready (void) 283static unsigned int etp_nready (void)
155{ 284{
156 unsigned int retval; 285 unsigned int retval;
157 286
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 287 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 288 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 289 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 290
162 return retval; 291 return retval;
163} 292}
164 293
165unsigned int eio_npending (void) 294static unsigned int etp_npending (void)
166{ 295{
167 unsigned int retval; 296 unsigned int retval;
168 297
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 298 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 299 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 300 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 301
173 return retval; 302 return retval;
174} 303}
175 304
176unsigned int eio_nthreads (void) 305static unsigned int etp_nthreads (void)
177{ 306{
178 unsigned int retval; 307 unsigned int retval;
179 308
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 309 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 310 retval = started;
188 * a somewhat faster data structure might be nice, but 317 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 318 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 319 * per shift, the most expensive operation.
191 */ 320 */
192typedef struct { 321typedef struct {
193 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 322 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194 int size; 323 int size;
195} reqq; 324} etp_reqq;
196 325
197static reqq req_queue; 326static etp_reqq req_queue;
198static reqq res_queue; 327static etp_reqq res_queue;
199 328
200static int reqq_push (reqq *q, eio_req *req) 329static int reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 330{
202 int pri = req->pri; 331 int pri = req->pri;
203 req->next = 0; 332 req->next = 0;
204 333
205 if (q->qe[pri]) 334 if (q->qe[pri])
211 q->qe[pri] = q->qs[pri] = req; 340 q->qe[pri] = q->qs[pri] = req;
212 341
213 return q->size++; 342 return q->size++;
214} 343}
215 344
216static eio_req *reqq_shift (reqq *q) 345static ETP_REQ *reqq_shift (etp_reqq *q)
217{ 346{
218 int pri; 347 int pri;
219 348
220 if (!q->size) 349 if (!q->size)
221 return 0; 350 return 0;
222 351
223 --q->size; 352 --q->size;
224 353
225 for (pri = EIO_NUM_PRI; pri--; ) 354 for (pri = ETP_NUM_PRI; pri--; )
226 { 355 {
227 eio_req *req = q->qs[pri]; 356 eio_req *req = q->qs[pri];
228 357
229 if (req) 358 if (req)
230 { 359 {
236 } 365 }
237 366
238 abort (); 367 abort ();
239} 368}
240 369
241static void grp_try_feed (eio_req *grp) 370static void etp_atfork_prepare (void)
242{ 371{
243 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 372 X_LOCK (wrklock);
373 X_LOCK (reqlock);
374 X_LOCK (reslock);
375#if !HAVE_PREADWRITE
376 X_LOCK (preadwritelock);
377#endif
378}
379
380static void etp_atfork_parent (void)
381{
382#if !HAVE_PREADWRITE
383 X_UNLOCK (preadwritelock);
384#endif
385 X_UNLOCK (reslock);
386 X_UNLOCK (reqlock);
387 X_UNLOCK (wrklock);
388}
389
390static void etp_atfork_child (void)
391{
392 ETP_REQ *prv;
393
394 while ((prv = reqq_shift (&req_queue)))
395 ETP_DESTROY (prv);
396
397 while ((prv = reqq_shift (&res_queue)))
398 ETP_DESTROY (prv);
399
400 while (wrk_first.next != &wrk_first)
244 { 401 {
245 int old_len = grp->size; 402 etp_worker *wrk = wrk_first.next;
246 403
247 EIO_FEED (grp); 404 if (wrk->req)
405 ETP_DESTROY (wrk->req);
248 406
249 /* stop if no progress has been made */ 407 etp_worker_clear (wrk);
250 if (old_len == grp->size) 408 etp_worker_free (wrk);
251 {
252 grp->feed = 0;
253 break;
254 }
255 } 409 }
256}
257 410
258static int eio_finish (eio_req *req); 411 started = 0;
412 idle = 0;
413 nreqs = 0;
414 nready = 0;
415 npending = 0;
259 416
260static int grp_dec (eio_req *grp) 417 etp_atfork_parent ();
261{ 418}
262 --grp->size;
263 419
264 /* call feeder, if applicable */ 420static void
265 grp_try_feed (grp); 421etp_once_init (void)
422{
423 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
424}
266 425
267 /* finish, if done */ 426static int
268 if (!grp->size && grp->int1) 427etp_init (void (*want_poll)(void), void (*done_poll)(void))
269 return eio_finish (grp); 428{
270 else 429 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
430
431 pthread_once (&doinit, etp_once_init);
432
433 want_poll_cb = want_poll;
434 done_poll_cb = done_poll;
435
271 return 0; 436 return 0;
272} 437}
273 438
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); 439X_THREAD_PROC (etp_proc);
325 440
326static void start_thread (void) 441static void etp_start_thread (void)
327{ 442{
328 worker *wrk = calloc (1, sizeof (worker)); 443 etp_worker *wrk = calloc (1, sizeof (etp_worker));
329 444
330 /*TODO*/ 445 /*TODO*/
331 assert (("unable to allocate worker thread data", !wrk)); 446 assert (("unable to allocate worker thread data", wrk));
332 447
333 X_LOCK (wrklock); 448 X_LOCK (wrklock);
334 449
335 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 450 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
336 { 451 {
337 wrk->prev = &wrk_first; 452 wrk->prev = &wrk_first;
338 wrk->next = wrk_first.next; 453 wrk->next = wrk_first.next;
339 wrk_first.next->prev = wrk; 454 wrk_first.next->prev = wrk;
340 wrk_first.next = wrk; 455 wrk_first.next = wrk;
344 free (wrk); 459 free (wrk);
345 460
346 X_UNLOCK (wrklock); 461 X_UNLOCK (wrklock);
347} 462}
348 463
349static void maybe_start_thread (void) 464static void etp_maybe_start_thread (void)
350{ 465{
351 if (eio_nthreads () >= wanted) 466 if (expect_true (etp_nthreads () >= wanted))
352 return; 467 return;
353 468
354 /* todo: maybe use idle here, but might be less exact */ 469 /* todo: maybe use idle here, but might be less exact */
355 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 470 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
356 return; 471 return;
357 472
358 start_thread (); 473 etp_start_thread ();
359} 474}
360 475
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) 476static void etp_end_thread (void)
375{ 477{
376 eio_req *req = calloc (1, sizeof (eio_req)); 478 eio_req *req = calloc (1, sizeof (eio_req));
377 479
378 req->type = EIO_QUIT; 480 req->type = -1;
379 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 481 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
380 482
381 X_LOCK (reqlock); 483 X_LOCK (reqlock);
382 reqq_push (&req_queue, req); 484 reqq_push (&req_queue, req);
383 X_COND_SIGNAL (reqwait); 485 X_COND_SIGNAL (reqwait);
384 X_UNLOCK (reqlock); 486 X_UNLOCK (reqlock);
386 X_LOCK (wrklock); 488 X_LOCK (wrklock);
387 --started; 489 --started;
388 X_UNLOCK (wrklock); 490 X_UNLOCK (wrklock);
389} 491}
390 492
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) 493static int etp_poll (void)
428{ 494{
429 int maxreqs = max_poll_reqs; 495 unsigned int maxreqs;
496 unsigned int maxtime;
430 struct timeval tv_start, tv_now; 497 struct timeval tv_start, tv_now;
431 eio_req *req;
432 498
499 X_LOCK (reslock);
500 maxreqs = max_poll_reqs;
501 maxtime = max_poll_time;
502 X_UNLOCK (reslock);
503
433 if (max_poll_time) 504 if (maxtime)
434 gettimeofday (&tv_start, 0); 505 gettimeofday (&tv_start, 0);
435 506
436 for (;;) 507 for (;;)
437 { 508 {
509 ETP_REQ *req;
510
438 maybe_start_thread (); 511 etp_maybe_start_thread ();
439 512
440 X_LOCK (reslock); 513 X_LOCK (reslock);
441 req = reqq_shift (&res_queue); 514 req = reqq_shift (&res_queue);
442 515
443 if (req) 516 if (req)
451 X_UNLOCK (reslock); 524 X_UNLOCK (reslock);
452 525
453 if (!req) 526 if (!req)
454 return 0; 527 return 0;
455 528
529 X_LOCK (reqlock);
456 --nreqs; 530 --nreqs;
531 X_UNLOCK (reqlock);
457 532
458 if (req->type == EIO_GROUP && req->size) 533 if (expect_false (req->type == EIO_GROUP && req->size))
459 { 534 {
460 req->int1 = 1; /* mark request as delayed */ 535 req->int1 = 1; /* mark request as delayed */
461 continue; 536 continue;
462 } 537 }
463 else 538 else
464 { 539 {
465 int res = eio_finish (req); 540 int res = ETP_FINISH (req);
466 if (res) 541 if (expect_false (res))
467 return res; 542 return res;
468 } 543 }
469 544
470 if (maxreqs && !--maxreqs) 545 if (expect_false (maxreqs && !--maxreqs))
471 break; 546 break;
472 547
473 if (max_poll_time) 548 if (maxtime)
474 { 549 {
475 gettimeofday (&tv_now, 0); 550 gettimeofday (&tv_now, 0);
476 551
477 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 552 if (tvdiff (&tv_start, &tv_now) >= maxtime)
478 break; 553 break;
479 } 554 }
480 } 555 }
481 556
482 errno = EAGAIN; 557 errno = EAGAIN;
483 return -1; 558 return -1;
484} 559}
485 560
561static void etp_cancel (ETP_REQ *req)
562{
563 X_LOCK (wrklock);
564 req->flags |= EIO_FLAG_CANCELLED;
565 X_UNLOCK (wrklock);
566
567 eio_grp_cancel (req);
568}
569
570static void etp_submit (ETP_REQ *req)
571{
572 req->pri -= ETP_PRI_MIN;
573
574 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
575 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
576
577 if (expect_false (req->type == EIO_GROUP))
578 {
579 /* I hope this is worth it :/ */
580 X_LOCK (reqlock);
581 ++nreqs;
582 X_UNLOCK (reqlock);
583
584 X_LOCK (reslock);
585
586 ++npending;
587
588 if (!reqq_push (&res_queue, req) && want_poll_cb)
589 want_poll_cb ();
590
591 X_UNLOCK (reslock);
592 }
593 else
594 {
595 X_LOCK (reqlock);
596 ++nreqs;
597 ++nready;
598 reqq_push (&req_queue, req);
599 X_COND_SIGNAL (reqwait);
600 X_UNLOCK (reqlock);
601
602 etp_maybe_start_thread ();
603 }
604}
605
606static void etp_set_max_poll_time (double nseconds)
607{
608 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609 max_poll_time = nseconds * EIO_TICKS;
610 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611}
612
613static void etp_set_max_poll_reqs (unsigned int maxreqs)
614{
615 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
616 max_poll_reqs = maxreqs;
617 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
618}
619
620static void etp_set_max_idle (unsigned int nthreads)
621{
622 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
623 max_idle = nthreads <= 0 ? 1 : nthreads;
624 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
625}
626
627static void etp_set_min_parallel (unsigned int nthreads)
628{
629 if (wanted < nthreads)
630 wanted = nthreads;
631}
632
633static void etp_set_max_parallel (unsigned int nthreads)
634{
635 if (wanted > nthreads)
636 wanted = nthreads;
637
638 while (started > wanted)
639 etp_end_thread ();
640}
641
642/*****************************************************************************/
643
644static void grp_try_feed (eio_req *grp)
645{
646 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
647 {
648 grp->flags &= ~EIO_FLAG_GROUPADD;
649
650 EIO_FEED (grp);
651
652 /* stop if no progress has been made */
653 if (!(grp->flags & EIO_FLAG_GROUPADD))
654 {
655 grp->feed = 0;
656 break;
657 }
658 }
659}
660
661static int grp_dec (eio_req *grp)
662{
663 --grp->size;
664
665 /* call feeder, if applicable */
666 grp_try_feed (grp);
667
668 /* finish, if done */
669 if (!grp->size && grp->int1)
670 return eio_finish (grp);
671 else
672 return 0;
673}
674
675void eio_destroy (eio_req *req)
676{
677 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
678 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
679
680 EIO_DESTROY (req);
681}
682
683static int eio_finish (eio_req *req)
684{
685 int res = EIO_FINISH (req);
686
687 if (req->grp)
688 {
689 int res2;
690 eio_req *grp = req->grp;
691
692 /* unlink request */
693 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
694 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
695
696 if (grp->grp_first == req)
697 grp->grp_first = req->grp_next;
698
699 res2 = grp_dec (grp);
700
701 if (!res && res2)
702 res = res2;
703 }
704
705 eio_destroy (req);
706
707 return res;
708}
709
710void eio_grp_cancel (eio_req *grp)
711{
712 for (grp = grp->grp_first; grp; grp = grp->grp_next)
713 eio_cancel (grp);
714}
715
716void eio_cancel (eio_req *req)
717{
718 etp_cancel (req);
719}
720
721void eio_submit (eio_req *req)
722{
723 etp_submit (req);
724}
725
726unsigned int eio_nreqs (void)
727{
728 return etp_nreqs ();
729}
730
731unsigned int eio_nready (void)
732{
733 return etp_nready ();
734}
735
736unsigned int eio_npending (void)
737{
738 return etp_npending ();
739}
740
741unsigned int eio_nthreads (void)
742{
743 return etp_nthreads ();
744}
745
746void eio_set_max_poll_time (double nseconds)
747{
748 etp_set_max_poll_time (nseconds);
749}
750
751void eio_set_max_poll_reqs (unsigned int maxreqs)
752{
753 etp_set_max_poll_reqs (maxreqs);
754}
755
756void eio_set_max_idle (unsigned int nthreads)
757{
758 etp_set_max_idle (nthreads);
759}
760
761void eio_set_min_parallel (unsigned int nthreads)
762{
763 etp_set_min_parallel (nthreads);
764}
765
766void eio_set_max_parallel (unsigned int nthreads)
767{
768 etp_set_max_parallel (nthreads);
769}
770
771int eio_poll (void)
772{
773 return etp_poll ();
774}
775
486/*****************************************************************************/ 776/*****************************************************************************/
487/* work around various missing functions */ 777/* work around various missing functions */
488 778
489#if !HAVE_PREADWRITE 779#if !HAVE_PREADWRITE
780# undef pread
781# undef pwrite
490# define pread aio_pread 782# define pread eio__pread
491# define pwrite aio_pwrite 783# define pwrite eio__pwrite
492 784
493/* 785static ssize_t
494 * make our pread/pwrite safe against themselves, but not against 786eio__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{ 787{
502 ssize_t res; 788 ssize_t res;
503 off_t ooffset; 789 off_t ooffset;
504 790
505 X_LOCK (preadwritelock); 791 X_LOCK (preadwritelock);
510 X_UNLOCK (preadwritelock); 796 X_UNLOCK (preadwritelock);
511 797
512 return res; 798 return res;
513} 799}
514 800
515static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 801static ssize_t
802eio__pwrite (int fd, void *buf, size_t count, off_t offset)
516{ 803{
517 ssize_t res; 804 ssize_t res;
518 off_t ooffset; 805 off_t ooffset;
519 806
520 X_LOCK (preadwritelock); 807 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR); 808 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET); 809 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count); 810 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET); 811 lseek (fd, ooffset, SEEK_SET);
525 X_UNLOCK (preadwritelock); 812 X_UNLOCK (preadwritelock);
526 813
527 return res; 814 return res;
528} 815}
529#endif 816#endif
530 817
531#ifndef HAVE_FUTIMES 818#ifndef HAVE_FUTIMES
532 819
820# undef utimes
821# undef futimes
533# define utimes(path,times) aio_utimes (path, times) 822# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) aio_futimes (fd, times) 823# define futimes(fd,times) eio__futimes (fd, times)
535 824
825static int
536static int aio_utimes (const char *filename, const struct timeval times[2]) 826eio__utimes (const char *filename, const struct timeval times[2])
537{ 827{
538 if (times) 828 if (times)
539 { 829 {
540 struct utimbuf buf; 830 struct utimbuf buf;
541 831
546 } 836 }
547 else 837 else
548 return utime (filename, 0); 838 return utime (filename, 0);
549} 839}
550 840
551static int aio_futimes (int fd, const struct timeval tv[2]) 841static int eio__futimes (int fd, const struct timeval tv[2])
552{ 842{
553 errno = ENOSYS; 843 errno = ENOSYS;
554 return -1; 844 return -1;
555} 845}
556 846
557#endif 847#endif
558 848
559#if !HAVE_FDATASYNC 849#if !HAVE_FDATASYNC
850# undef fdatasync
560# define fdatasync fsync 851# define fdatasync(fd) fsync (fd)
561#endif 852#endif
853
854/* sync_file_range always needs emulation */
855int
856eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
857{
858#if HAVE_SYNC_FILE_RANGE
859 int res;
860
861 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
862 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
863 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
864 {
865 flags = 0
866 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
867 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
868 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
869 }
870
871 res = sync_file_range (fd, offset, nbytes, flags);
872
873 if (!res || errno != ENOSYS)
874 return res;
875#endif
876
877 /* even though we could play tricks with the flags, it's better to always
878 * call fdatasync, as that matches the expectation of its users best */
879 return fdatasync (fd);
880}
562 881
563#if !HAVE_READAHEAD 882#if !HAVE_READAHEAD
883# undef readahead
564# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 884# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565 885
886static ssize_t
566static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 887eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
567{ 888{
568 size_t todo = count; 889 size_t todo = count;
569 dBUF; 890 dBUF;
570 891
571 while (todo > 0) 892 while (todo > 0)
581 return count; 902 return count;
582} 903}
583 904
584#endif 905#endif
585 906
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 */ 907/* sendfile always needs emulation */
908static ssize_t
617static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 909eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
618{ 910{
619 ssize_t res; 911 ssize_t res;
620 912
621 if (!count) 913 if (!count)
622 return 0; 914 return 0;
623 915
624#if HAVE_SENDFILE 916#if HAVE_SENDFILE
625# if __linux 917# if __linux
626 res = sendfile (ofd, ifd, &offset, count); 918 res = sendfile (ofd, ifd, &offset, count);
627 919
628# elif __freebsd 920# elif __FreeBSD__
629 /* 921 /*
630 * Of course, the freebsd sendfile is a dire hack with no thoughts 922 * Of course, the freebsd sendfile is a dire hack with no thoughts
631 * wasted on making it similar to other I/O functions. 923 * wasted on making it similar to other I/O functions.
632 */ 924 */
633 { 925 {
634 off_t sbytes; 926 off_t sbytes;
635 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 927 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636 928
637 if (res < 0 && sbytes) 929 #if 0 /* according to the manpage, this is correct, but broken behaviour */
638 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 930 /* freebsd' sendfile will return 0 on success */
931 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
932 /* not on e.g. EIO or EPIPE - sounds broken */
933 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
934 res = sbytes;
935 #endif
936
937 /* according to source inspection, this is correct, and useful behaviour */
938 if (sbytes)
939 res = sbytes;
940 }
941
942# elif defined (__APPLE__)
943
944 {
945 off_t sbytes = count;
946 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
947
948 /* according to the manpage, sbytes is always valid */
949 if (sbytes)
639 res = sbytes; 950 res = sbytes;
640 } 951 }
641 952
642# elif __hpux 953# elif __hpux
643 res = sendfile (ofd, ifd, offset, count, 0, 0); 954 res = sendfile (ofd, ifd, offset, count, 0, 0);
657 if (res < 0 && sbytes) 968 if (res < 0 && sbytes)
658 res = sbytes; 969 res = sbytes;
659 } 970 }
660 971
661# endif 972# endif
973
974#elif defined (_WIN32)
975
976 /* does not work, just for documentation of what would need to be done */
977 {
978 HANDLE h = TO_SOCKET (ifd);
979 SetFilePointer (h, offset, 0, FILE_BEGIN);
980 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
981 }
982
662#else 983#else
663 res = -1; 984 res = -1;
664 errno = ENOSYS; 985 errno = ENOSYS;
665#endif 986#endif
666 987
667 if (res < 0 988 if (res < 0
668 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 989 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
990 /* BSDs */
991#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
992 || errno == ENOTSUP
993#endif
994 || errno == EOPNOTSUPP /* BSDs */
669#if __solaris 995#if __solaris
670 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 996 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
671#endif 997#endif
672 ) 998 )
673 ) 999 )
704 } 1030 }
705 1031
706 return res; 1032 return res;
707} 1033}
708 1034
1035static signed char
1036eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1037{
1038 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1039 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1040}
1041
1042#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1043
1044#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1045#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1046
1047static void
1048eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1049{
1050 unsigned char bits [9 + sizeof (ino_t) * 8];
1051 unsigned char *bit = bits;
1052
1053 assert (CHAR_BIT == 8);
1054 assert (sizeof (eio_dirent) * 8 < 256);
1055 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1056 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1057
1058 if (size <= EIO_SORT_FAST)
1059 return;
1060
1061 /* first prepare an array of bits to test in our radix sort */
1062 /* try to take endianness into account, as well as differences in ino_t sizes */
1063 /* inode_bits must contain all inodes ORed together */
1064 /* which is used to skip bits that are 0 everywhere, which is very common */
1065 {
1066 ino_t endianness;
1067 int i, j;
1068
1069 /* we store the byte offset of byte n into byte n of "endianness" */
1070 for (i = 0; i < sizeof (ino_t); ++i)
1071 ((unsigned char *)&endianness)[i] = i;
1072
1073 *bit++ = 0;
1074
1075 for (i = 0; i < sizeof (ino_t); ++i)
1076 {
1077 /* shifting off the byte offsets out of "endianness" */
1078 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1079 endianness >>= 8;
1080
1081 for (j = 0; j < 8; ++j)
1082 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1083 *bit++ = offs + j;
1084 }
1085
1086 for (j = 0; j < 8; ++j)
1087 if (score_bits & (1 << j))
1088 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1089 }
1090
1091 /* now actually do the sorting (a variant of MSD radix sort) */
1092 {
1093 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1094 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1095 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1096 int stk_idx = 0;
1097
1098 base_stk [stk_idx] = dents;
1099 end_stk [stk_idx] = dents + size;
1100 bit_stk [stk_idx] = bit - 1;
1101
1102 do
1103 {
1104 base = base_stk [stk_idx];
1105 end = end_stk [stk_idx];
1106 bit = bit_stk [stk_idx];
1107
1108 for (;;)
1109 {
1110 unsigned char O = *bit >> 3;
1111 unsigned char M = 1 << (*bit & 7);
1112
1113 eio_dirent *a = base;
1114 eio_dirent *b = end;
1115
1116 if (b - a < EIO_SORT_CUTOFF)
1117 break;
1118
1119 /* now bit-partition the array on the bit */
1120 /* this ugly asymmetric loop seems to perform much better than typical */
1121 /* partition algos found in the literature */
1122 do
1123 if (!(((unsigned char *)a)[O] & M))
1124 ++a;
1125 else if (!(((unsigned char *)--b)[O] & M))
1126 {
1127 eio_dirent tmp = *a; *a = *b; *b = tmp;
1128 ++a;
1129 }
1130 while (b > a);
1131
1132 /* next bit, or stop, if no bits left in this path */
1133 if (!*--bit)
1134 break;
1135
1136 base_stk [stk_idx] = a;
1137 end_stk [stk_idx] = end;
1138 bit_stk [stk_idx] = bit;
1139 ++stk_idx;
1140
1141 end = a;
1142 }
1143 }
1144 while (stk_idx--);
1145 }
1146}
1147
1148static void
1149eio_dent_insertion_sort (eio_dirent *dents, int size)
1150{
1151 /* first move the smallest element to the front, to act as a sentinel */
1152 {
1153 int i;
1154 eio_dirent *min = dents;
1155
1156 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1157 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1158 if (EIO_DENT_CMP (dents [i], <, *min))
1159 min = &dents [i];
1160
1161 /* swap elements 0 and j (minimum) */
1162 {
1163 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1164 }
1165 }
1166
1167 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1168 {
1169 eio_dirent *i, *j;
1170
1171 for (i = dents + 1; i < dents + size; ++i)
1172 {
1173 eio_dirent value = *i;
1174
1175 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1176 j [1] = j [0];
1177
1178 j [1] = value;
1179 }
1180 }
1181}
1182
1183static void
1184eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1185{
1186 if (size <= 1)
1187 return; /* our insertion sort relies on size > 0 */
1188
1189 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1190 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1191 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1192
1193 /* use an insertion sort at the end, or for small arrays, */
1194 /* as insertion sort is more efficient for small partitions */
1195 eio_dent_insertion_sort (dents, size);
1196}
1197
709/* read a full directory */ 1198/* read a full directory */
1199static void
710static void scandir_ (eio_req *req, worker *self) 1200eio__scandir (eio_req *req, etp_worker *self)
711{ 1201{
712 DIR *dirp; 1202 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; 1203 EIO_STRUCT_DIRENT *entp;
719 char *name, *names; 1204 char *name, *names;
720 int memlen = 4096; 1205 int namesalloc = 4096;
721 int memofs = 0; 1206 int namesoffs = 0;
1207 int flags = req->int1;
1208 eio_dirent *dents = 0;
1209 int dentalloc = 128;
722 int res = 0; 1210 int dentoffs = 0;
1211 ino_t inode_bits = 0;
1212
1213 req->result = -1;
1214
1215 if (!(flags & EIO_READDIR_DENTS))
1216 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
723 1217
724 X_LOCK (wrklock); 1218 X_LOCK (wrklock);
1219 /* the corresponding closedir is in ETP_WORKER_CLEAR */
725 self->dirp = dirp = opendir (req->ptr1); 1220 self->dirp = dirp = opendir (req->ptr1);
726 self->dbuf = u = malloc (sizeof (*u));
727 req->flags |= EIO_FLAG_PTR2_FREE; 1221 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1222 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
728 req->ptr2 = names = malloc (memlen); 1223 req->ptr2 = names = malloc (namesalloc);
729 X_UNLOCK (wrklock); 1224 X_UNLOCK (wrklock);
730 1225
731 if (dirp && u && names) 1226 if (dirp && names && (!flags || dents))
732 for (;;) 1227 for (;;)
733 { 1228 {
734 errno = 0; 1229 errno = 0;
735 readdir_r (dirp, &u->d, &entp); 1230 entp = readdir (dirp);
736 1231
737 if (!entp) 1232 if (!entp)
1233 {
1234 if (errno)
1235 break;
1236
1237 /* sort etc. */
1238 req->int1 = flags;
1239 req->result = dentoffs;
1240
1241 if (flags & EIO_READDIR_STAT_ORDER)
1242 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
1243 else if (flags & EIO_READDIR_DIRS_FIRST)
1244 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1245 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1246 else
1247 {
1248 /* in this case, all is known, and we just put dirs first and sort them */
1249 eio_dirent *oth = dents + dentoffs;
1250 eio_dirent *dir = dents;
1251
1252 /* now partition dirs to the front, and non-dirs to the back */
1253 /* by walking from both sides and swapping if necessary */
1254 /* also clear score, so it doesn't influence sorting */
1255 while (oth > dir)
1256 {
1257 if (dir->type == EIO_DT_DIR)
1258 ++dir;
1259 else if ((--oth)->type == EIO_DT_DIR)
1260 {
1261 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1262
1263 ++dir;
1264 }
1265 }
1266
1267 /* now sort the dirs only */
1268 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1269 }
1270
738 break; 1271 break;
1272 }
739 1273
1274 /* now add the entry to our list(s) */
740 name = entp->d_name; 1275 name = entp->d_name;
741 1276
1277 /* skip . and .. entries */
742 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1278 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
743 { 1279 {
744 int len = strlen (name) + 1; 1280 int len = D_NAMLEN (entp) + 1;
745 1281
746 res++; 1282 while (expect_false (namesoffs + len > namesalloc))
747
748 while (memofs + len > memlen)
749 { 1283 {
750 memlen *= 2; 1284 namesalloc *= 2;
751 X_LOCK (wrklock); 1285 X_LOCK (wrklock);
752 req->ptr2 = names = realloc (names, memlen); 1286 req->ptr2 = names = realloc (names, namesalloc);
753 X_UNLOCK (wrklock); 1287 X_UNLOCK (wrklock);
754 1288
755 if (!names) 1289 if (!names)
756 break; 1290 break;
757 } 1291 }
758 1292
759 memcpy (names + memofs, name, len); 1293 memcpy (names + namesoffs, name, len);
1294
1295 if (dents)
1296 {
1297 struct eio_dirent *ent;
1298
1299 if (expect_false (dentoffs == dentalloc))
1300 {
1301 dentalloc *= 2;
1302 X_LOCK (wrklock);
1303 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1304 X_UNLOCK (wrklock);
1305
1306 if (!dents)
1307 break;
1308 }
1309
1310 ent = dents + dentoffs;
1311
1312 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1313 ent->namelen = len - 1;
1314 ent->inode = D_INO (entp);
1315
1316 inode_bits |= ent->inode;
1317
1318 switch (D_TYPE (entp))
1319 {
1320 default:
1321 ent->type = EIO_DT_UNKNOWN;
1322 flags |= EIO_READDIR_FOUND_UNKNOWN;
1323 break;
1324
1325 #ifdef DT_FIFO
1326 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1327 #endif
1328 #ifdef DT_CHR
1329 case DT_CHR: ent->type = EIO_DT_CHR; break;
1330 #endif
1331 #ifdef DT_MPC
1332 case DT_MPC: ent->type = EIO_DT_MPC; break;
1333 #endif
1334 #ifdef DT_DIR
1335 case DT_DIR: ent->type = EIO_DT_DIR; break;
1336 #endif
1337 #ifdef DT_NAM
1338 case DT_NAM: ent->type = EIO_DT_NAM; break;
1339 #endif
1340 #ifdef DT_BLK
1341 case DT_BLK: ent->type = EIO_DT_BLK; break;
1342 #endif
1343 #ifdef DT_MPB
1344 case DT_MPB: ent->type = EIO_DT_MPB; break;
1345 #endif
1346 #ifdef DT_REG
1347 case DT_REG: ent->type = EIO_DT_REG; break;
1348 #endif
1349 #ifdef DT_NWK
1350 case DT_NWK: ent->type = EIO_DT_NWK; break;
1351 #endif
1352 #ifdef DT_CMP
1353 case DT_CMP: ent->type = EIO_DT_CMP; break;
1354 #endif
1355 #ifdef DT_LNK
1356 case DT_LNK: ent->type = EIO_DT_LNK; break;
1357 #endif
1358 #ifdef DT_SOCK
1359 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1360 #endif
1361 #ifdef DT_DOOR
1362 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1363 #endif
1364 #ifdef DT_WHT
1365 case DT_WHT: ent->type = EIO_DT_WHT; break;
1366 #endif
1367 }
1368
1369 ent->score = 7;
1370
1371 if (flags & EIO_READDIR_DIRS_FIRST)
1372 {
1373 if (ent->type == EIO_DT_UNKNOWN)
1374 {
1375 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1376 ent->score = 1;
1377 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1378 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1379 }
1380 else if (ent->type == EIO_DT_DIR)
1381 ent->score = 0;
1382 }
1383 }
1384
760 memofs += len; 1385 namesoffs += len;
1386 ++dentoffs;
1387 }
1388
1389 if (EIO_CANCELLED (req))
1390 {
1391 errno = ECANCELED;
1392 break;
761 } 1393 }
762 } 1394 }
1395}
763 1396
764 if (errno) 1397#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
765 res = -1; 1398# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
766 1399#else
767 req->result = res; 1400
1401int
1402eio__msync (void *mem, size_t len, int flags)
1403{
1404 if (EIO_MS_ASYNC != MS_SYNC
1405 || EIO_MS_INVALIDATE != MS_INVALIDATE
1406 || EIO_MS_SYNC != MS_SYNC)
1407 {
1408 flags = 0
1409 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1410 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1411 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1412 }
1413
1414 return msync (mem, len, flags);
1415}
1416
1417#endif
1418
1419int
1420eio__mtouch (void *mem, size_t len, int flags)
1421{
1422 intptr_t addr = (intptr_t)mem;
1423 intptr_t end = addr + len;
1424#ifdef PAGESIZE
1425 const intptr_t page = PAGESIZE;
1426#else
1427 static intptr_t page;
1428
1429 if (!page)
1430 page = sysconf (_SC_PAGESIZE);
1431#endif
1432
1433 /* round down to start of page, although this is probably useless */
1434 addr &= ~(page - 1); /* assume page size is always a power of two */
1435
1436 if (addr < end)
1437 if (flags & EIO_MT_MODIFY) /* modify */
1438 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1439 else
1440 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1441
1442 return 0;
768} 1443}
769 1444
770/*****************************************************************************/ 1445/*****************************************************************************/
771 1446
772#define ALLOC(len) \ 1447#define ALLOC(len) \
782 req->result = -1; \ 1457 req->result = -1; \
783 break; \ 1458 break; \
784 } \ 1459 } \
785 } 1460 }
786 1461
787X_THREAD_PROC (eio_proc) 1462X_THREAD_PROC (etp_proc)
788{ 1463{
789 eio_req *req; 1464 ETP_REQ *req;
790 struct timespec ts; 1465 struct timespec ts;
791 worker *self = (worker *)thr_arg; 1466 etp_worker *self = (etp_worker *)thr_arg;
792 1467
793 /* try to distribute timeouts somewhat randomly */ 1468 /* try to distribute timeouts somewhat randomly */
794 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1469 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
795 1470
796 for (;;) 1471 for (;;)
797 { 1472 {
798 ts.tv_sec = time (0) + IDLE_TIMEOUT;
799
800 X_LOCK (reqlock); 1473 X_LOCK (reqlock);
801 1474
802 for (;;) 1475 for (;;)
803 { 1476 {
804 self->req = req = reqq_shift (&req_queue); 1477 self->req = req = reqq_shift (&req_queue);
806 if (req) 1479 if (req)
807 break; 1480 break;
808 1481
809 ++idle; 1482 ++idle;
810 1483
1484 ts.tv_sec = time (0) + IDLE_TIMEOUT;
811 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1485 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
812 { 1486 {
813 if (idle > max_idle) 1487 if (idle > max_idle)
814 { 1488 {
815 --idle; 1489 --idle;
820 goto quit; 1494 goto quit;
821 } 1495 }
822 1496
823 /* we are allowed to idle, so do so without any timeout */ 1497 /* we are allowed to idle, so do so without any timeout */
824 X_COND_WAIT (reqwait, reqlock); 1498 X_COND_WAIT (reqwait, reqlock);
825 ts.tv_sec = time (0) + IDLE_TIMEOUT;
826 } 1499 }
827 1500
828 --idle; 1501 --idle;
829 } 1502 }
830 1503
831 --nready; 1504 --nready;
832 1505
833 X_UNLOCK (reqlock); 1506 X_UNLOCK (reqlock);
834 1507
835 errno = 0; /* strictly unnecessary */ 1508 if (req->type < 0)
1509 goto quit;
836 1510
837 if (!EIO_CANCELLED (req)) 1511 if (!EIO_CANCELLED (req))
838 switch (req->type) 1512 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 1513
938 X_LOCK (reslock); 1514 X_LOCK (reslock);
939 1515
940 ++npending; 1516 ++npending;
941 1517
942 if (!reqq_push (&res_queue, req) && want_poll_cb) 1518 if (!reqq_push (&res_queue, req) && want_poll_cb)
943 want_poll_cb (); 1519 want_poll_cb ();
944 1520
945 self->req = 0; 1521 self->req = 0;
946 worker_clear (self); 1522 etp_worker_clear (self);
947 1523
948 X_UNLOCK (reslock); 1524 X_UNLOCK (reslock);
949 } 1525 }
950 1526
951quit: 1527quit:
952 X_LOCK (wrklock); 1528 X_LOCK (wrklock);
953 worker_free (self); 1529 etp_worker_free (self);
954 X_UNLOCK (wrklock); 1530 X_UNLOCK (wrklock);
955 1531
956 return 0; 1532 return 0;
957} 1533}
958 1534
959/*****************************************************************************/ 1535/*****************************************************************************/
960 1536
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)) 1537int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1018{ 1538{
1019 want_poll_cb = want_poll; 1539 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} 1540}
1035 1541
1036static void eio_api_destroy (eio_req *req) 1542static void eio_api_destroy (eio_req *req)
1037{ 1543{
1038 free (req); 1544 free (req);
1043 \ 1549 \
1044 req = (eio_req *)calloc (1, sizeof *req); \ 1550 req = (eio_req *)calloc (1, sizeof *req); \
1045 if (!req) \ 1551 if (!req) \
1046 return 0; \ 1552 return 0; \
1047 \ 1553 \
1048 req->type = EIO_ ## rtype; \ 1554 req->type = rtype; \
1049 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 1555 req->pri = pri; \
1050 req->finish = cb; \ 1556 req->finish = cb; \
1557 req->data = data; \
1051 req->destroy = eio_api_destroy; 1558 req->destroy = eio_api_destroy;
1052 1559
1053#define SEND eio_submit (req); return req 1560#define SEND eio_submit (req); return req
1054 1561
1055#define PATH (void)0 1562#define PATH \
1563 req->flags |= EIO_FLAG_PTR1_FREE; \
1564 req->ptr1 = strdup (path); \
1565 if (!req->ptr1) \
1566 { \
1567 eio_api_destroy (req); \
1568 return 0; \
1569 }
1056 1570
1571static void eio_execute (etp_worker *self, eio_req *req)
1572{
1573 errno = 0;
1574
1575 switch (req->type)
1576 {
1577 case EIO_READ: ALLOC (req->size);
1578 req->result = req->offs >= 0
1579 ? pread (req->int1, req->ptr2, req->size, req->offs)
1580 : read (req->int1, req->ptr2, req->size); break;
1581 case EIO_WRITE: req->result = req->offs >= 0
1582 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1583 : write (req->int1, req->ptr2, req->size); break;
1584
1585 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1586 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1587
1588 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1589 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1590 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1591 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1592 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1593 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1594
1595 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1596 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1597 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1598 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1599
1600 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1601 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1602 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1603 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1604 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1605 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1606
1607 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1608 case EIO_CLOSE: req->result = close (req->int1); break;
1609 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1610 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1611 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1612 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1613 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1614 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1615 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1616 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1617
1618 case EIO_READLINK: ALLOC (PATH_MAX);
1619 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1620
1621 case EIO_SYNC: req->result = 0; sync (); break;
1622 case EIO_FSYNC: req->result = fsync (req->int1); break;
1623 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1624 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1625 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1626 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1627
1628 case EIO_READDIR: eio__scandir (req, self); break;
1629
1630 case EIO_BUSY:
1631#ifdef _WIN32
1632 Sleep (req->nv1 * 1000.);
1633#else
1634 {
1635 struct timeval tv;
1636
1637 tv.tv_sec = req->nv1;
1638 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
1639
1640 req->result = select (0, 0, 0, 0, &tv);
1641 }
1642#endif
1643 break;
1644
1645 case EIO_UTIME:
1646 case EIO_FUTIME:
1647 {
1648 struct timeval tv[2];
1649 struct timeval *times;
1650
1651 if (req->nv1 != -1. || req->nv2 != -1.)
1652 {
1653 tv[0].tv_sec = req->nv1;
1654 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1655 tv[1].tv_sec = req->nv2;
1656 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1657
1658 times = tv;
1659 }
1660 else
1661 times = 0;
1662
1663
1664 req->result = req->type == EIO_FUTIME
1665 ? futimes (req->int1, times)
1666 : utimes (req->ptr1, times);
1667 }
1668 break;
1669
1670 case EIO_GROUP:
1671 abort (); /* handled in eio_request */
1672
1673 case EIO_NOP:
1674 req->result = 0;
1675 break;
1676
1677 case EIO_CUSTOM:
1678 ((void (*)(eio_req *))req->feed) (req);
1679 break;
1680
1681 default:
1682 req->result = -1;
1683 break;
1684 }
1685
1686 req->errorno = errno;
1687}
1688
1689#ifndef EIO_NO_WRAPPERS
1690
1691eio_req *eio_nop (int pri, eio_cb cb, void *data)
1692{
1693 REQ (EIO_NOP); SEND;
1694}
1695
1696eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1697{
1698 REQ (EIO_BUSY); req->nv1 = delay; SEND;
1699}
1700
1057eio_req *eio_fsync (int fd, eio_cb cb) 1701eio_req *eio_sync (int pri, eio_cb cb, void *data)
1058{ 1702{
1703 REQ (EIO_SYNC); SEND;
1704}
1705
1706eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1707{
1059 REQ (FSYNC); req->int1 = fd; SEND; 1708 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1060} 1709}
1061 1710
1711eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1712{
1713 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1714}
1715
1716eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1717{
1718 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1719}
1720
1721eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1722{
1723 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1724}
1725
1062eio_req *eio_fdatasync (int fd, eio_cb cb) 1726eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1063{ 1727{
1064 REQ (FDATASYNC); req->int1 = fd; SEND; 1728 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1065} 1729}
1066 1730
1731eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1732{
1733 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1734}
1735
1067eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb) 1736eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1068{ 1737{
1069 REQ (READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 1738 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1070} 1739}
1071 1740
1072eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb) 1741eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1073{ 1742{
1074 REQ (READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 1743 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1075} 1744}
1076 1745
1077eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb) 1746eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1078{ 1747{
1079 REQ (WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 1748 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1080} 1749}
1081 1750
1082eio_req *eio_fstat (int fd, eio_cb cb) 1751eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1083{ 1752{
1084 REQ (FSTAT); req->int1 = fd; SEND; 1753 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1085} 1754}
1086 1755
1756eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1757{
1758 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1759}
1760
1087eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb) 1761eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1088{ 1762{
1089 REQ (FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1763 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1090} 1764}
1091 1765
1092eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb) 1766eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1093{ 1767{
1094 REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 1768 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1095} 1769}
1096 1770
1097eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb) 1771eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1098{ 1772{
1099 REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 1773 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1100} 1774}
1101 1775
1102eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb) 1776eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1103{ 1777{
1104 REQ (FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 1778 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1105} 1779}
1106 1780
1107eio_req *eio_dup2 (int fd, int fd2, eio_cb cb) 1781eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1108{ 1782{
1109 REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND; 1783 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1110} 1784}
1111 1785
1112eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb) 1786eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1113{ 1787{
1114 REQ (SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 1788 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1115} 1789}
1116 1790
1117eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb) 1791eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1118{ 1792{
1119 REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 1793 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1120} 1794}
1121 1795
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) 1796eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1138{ 1797{
1139 REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 1798 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1140} 1799}
1141 1800
1142eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb) 1801eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1143{ 1802{
1144 REQ (TRUNCATE); PATH; req->offs = offset; SEND; 1803 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1145} 1804}
1146 1805
1147eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 1806eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1148{ 1807{
1149 REQ (CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 1808 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1150} 1809}
1151 1810
1152eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb) 1811eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1153{ 1812{
1154 REQ (CHMOD); PATH; req->int2 = (long)mode; SEND; 1813 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1155} 1814}
1156 1815
1157eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb) 1816eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1158{ 1817{
1159 REQ (MKDIR); PATH; req->int2 = (long)mode; SEND; 1818 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1160} 1819}
1161 1820
1162eio_req *eio_unlink (const char *path, eio_cb cb) 1821static eio_req *
1822eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1163{ 1823{
1164 REQ (UNLINK); PATH; SEND;
1165}
1166
1167eio_req *eio_rmdir (const char *path, eio_cb cb)
1168{
1169 REQ (RMDIR); PATH; SEND; 1824 REQ (type); PATH; SEND;
1170} 1825}
1171 1826
1827eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1828{
1829 return eio__1path (EIO_READLINK, path, pri, cb, data);
1830}
1831
1832eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1833{
1834 return eio__1path (EIO_STAT, path, pri, cb, data);
1835}
1836
1837eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1838{
1839 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1840}
1841
1842eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1843{
1844 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1845}
1846
1847eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1848{
1849 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1850}
1851
1172eio_req *eio_readdir (const char *path, eio_cb cb) 1852eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1173{ 1853{
1174 REQ (READDIR); PATH; SEND; 1854 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1175} 1855}
1176 1856
1857eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1858{
1859 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1860}
1861
1177eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb) 1862eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1178{ 1863{
1179 REQ (MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1864 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1180} 1865}
1181 1866
1182eio_req *eio_busy (double delay, eio_cb cb) 1867static eio_req *
1868eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1183{ 1869{
1184 REQ (BUSY); req->nv1 = delay; SEND; 1870 REQ (type); PATH;
1185}
1186 1871
1872 req->flags |= EIO_FLAG_PTR2_FREE;
1873 req->ptr2 = strdup (new_path);
1874 if (!req->ptr2)
1875 {
1876 eio_api_destroy (req);
1877 return 0;
1878 }
1879
1880 SEND;
1881}
1882
1883eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1884{
1885 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
1886}
1887
1888eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1889{
1890 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
1891}
1892
1893eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1894{
1895 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1896}
1897
1898eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1899{
1900 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1901}
1902
1903#endif
1904
1187eio_req *eio_nop (eio_cb cb) 1905eio_req *eio_grp (eio_cb cb, void *data)
1188{ 1906{
1907 const int pri = EIO_PRI_MAX;
1908
1189 REQ (NOP); SEND; 1909 REQ (EIO_GROUP); SEND;
1190} 1910}
1191 1911
1192#undef REQ 1912#undef REQ
1193#undef PATH 1913#undef PATH
1194#undef SEND 1914#undef SEND
1195 1915
1196#if 0 1916/*****************************************************************************/
1197 1917/* 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 1918
1571void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 1919void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1572{ 1920{
1573 grp->int2 = limit; 1921 grp->int2 = limit;
1574 grp->feed = feed; 1922 grp->feed = feed;
1585 1933
1586void eio_grp_add (eio_req *grp, eio_req *req) 1934void eio_grp_add (eio_req *grp, eio_req *req)
1587{ 1935{
1588 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1936 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1589 1937
1938 grp->flags |= EIO_FLAG_GROUPADD;
1939
1590 ++grp->size; 1940 ++grp->size;
1591 req->grp = grp; 1941 req->grp = grp;
1592 1942
1593 req->grp_prev = 0; 1943 req->grp_prev = 0;
1594 req->grp_next = grp->grp_first; 1944 req->grp_next = grp->grp_first;
1597 grp->grp_first->grp_prev = req; 1947 grp->grp_first->grp_prev = req;
1598 1948
1599 grp->grp_first = req; 1949 grp->grp_first = req;
1600} 1950}
1601 1951
1952/*****************************************************************************/
1953/* misc garbage */
1602 1954
1955ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1956{
1957 etp_worker wrk;
1958 ssize_t ret;
1959
1960 wrk.dbuf = 0;
1961
1962 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1963
1964 if (wrk.dbuf)
1965 free (wrk.dbuf);
1966
1967 return ret;
1968}
1969

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines