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 1.80 by root, Tue Jul 5 18:59:28 2011 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL.
38 */
39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
1#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
2#include "xthread.h" 50#include "xthread.h"
3 51
4#include <errno.h> 52#include <errno.h>
5#include <stddef.h> 53#include <stddef.h>
6#include <stdlib.h> 54#include <stdlib.h>
7#include <string.h> 55#include <string.h>
8#include <errno.h> 56#include <errno.h>
9#include <sys/types.h> 57#include <sys/types.h>
10#include <sys/stat.h> 58#include <sys/stat.h>
59#include <sys/statvfs.h>
11#include <limits.h> 60#include <limits.h>
12#include <fcntl.h> 61#include <fcntl.h>
13#include <assert.h> 62#include <assert.h>
14 63
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H
67# include <stdint.h>
68#endif
69
70#ifndef ECANCELED
71# define ECANCELED EDOM
72#endif
73
74static void eio_destroy (eio_req *req);
75
15#ifndef EIO_FINISH 76#ifndef EIO_FINISH
16# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 77# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17#endif 78#endif
18 79
19#ifndef EIO_DESTROY 80#ifndef EIO_DESTROY
25#endif 86#endif
26 87
27#ifdef _WIN32 88#ifdef _WIN32
28 89
29 /*doh*/ 90 /*doh*/
30
31#else 91#else
32 92
33# include "config.h"
34# include <sys/time.h> 93# include <sys/time.h>
35# include <sys/select.h> 94# include <sys/select.h>
36# include <unistd.h> 95# include <unistd.h>
37# include <utime.h> 96# include <utime.h>
38# include <signal.h> 97# include <signal.h>
39# include <dirent.h> 98# include <dirent.h>
40 99
100#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
101# include <sys/mman.h>
102#endif
103
104/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
105# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
106# define _DIRENT_HAVE_D_TYPE /* sigh */
107# define D_INO(de) (de)->d_fileno
108# define D_NAMLEN(de) (de)->d_namlen
109# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
110# define D_INO(de) (de)->d_ino
111# endif
112
113#ifdef _D_EXACT_NAMLEN
114# undef D_NAMLEN
115# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
116#endif
117
118# ifdef _DIRENT_HAVE_D_TYPE
119# define D_TYPE(de) (de)->d_type
120# endif
121
41# ifndef EIO_STRUCT_DIRENT 122# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 123# define EIO_STRUCT_DIRENT struct dirent
43# endif 124# endif
44 125
45#endif 126#endif
46 127
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif
50
51#if HAVE_SENDFILE 128#if HAVE_SENDFILE
52# if __linux 129# if __linux
53# include <sys/sendfile.h> 130# include <sys/sendfile.h>
54# elif __freebsd 131# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 132# include <sys/socket.h>
56# include <sys/uio.h> 133# include <sys/uio.h>
57# elif __hpux 134# elif __hpux
58# include <sys/socket.h> 135# include <sys/socket.h>
59# elif __solaris /* not yet */ 136# elif __solaris
60# include <sys/sendfile.h> 137# include <sys/sendfile.h>
61# else 138# else
62# error sendfile support requested but not available 139# error sendfile support requested but not available
63# endif 140# endif
64#endif 141#endif
65 142
66/* number of seconds after which an idle threads exit */ 143#ifndef D_TYPE
67#define IDLE_TIMEOUT 10 144# define D_TYPE(de) 0
145#endif
146#ifndef D_INO
147# define D_INO(de) 0
148#endif
149#ifndef D_NAMLEN
150# define D_NAMLEN(de) strlen ((de)->d_name)
151#endif
68 152
69/* used for struct dirent, AIX doesn't provide it */ 153/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 154#ifndef NAME_MAX
71# define NAME_MAX 4096 155# define NAME_MAX 4096
72#endif 156#endif
73 157
158/* used for readlink etc. */
159#ifndef PATH_MAX
160# define PATH_MAX 4096
161#endif
162
74/* buffer size for various temporary buffers */ 163/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 164#define EIO_BUFSIZE 65536
76 165
77#define dBUF \ 166#define dBUF \
78 char *eio_buf; \ 167 char *eio_buf; \
79 X_LOCK (wrklock); \ 168 ETP_WORKER_LOCK (self); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 169 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 170 ETP_WORKER_UNLOCK (self); \
171 errno = ENOMEM; \
82 if (!eio_buf) \ 172 if (!eio_buf) \
83 return -1; 173 return -1;
84 174
85#define EIO_TICKS ((1000000 + 1023) >> 10) 175#define EIO_TICKS ((1000000 + 1023) >> 10)
86 176
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 */
211ecb_inline int
212tvdiff (struct timeval *tv1, struct timeval *tv2)
213{
214 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
215 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
216}
217
218static unsigned int started, idle, wanted = 4;
219
87static void (*want_poll_cb) (void); 220static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 221static void (*done_poll_cb) (void);
89 222
90static unsigned int max_poll_time = 0; 223static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 224static unsigned int max_poll_reqs; /* reslock */
92 225
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 226static volatile unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 227static volatile unsigned int nready; /* reqlock */
95{ 228static volatile unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 229static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 230static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 231
100static unsigned int started, idle, wanted; 232static xmutex_t wrklock;
233static xmutex_t reslock;
234static xmutex_t reqlock;
235static xcond_t reqwait;
236
237#if !HAVE_PREADWRITE
238/*
239 * make our pread/pwrite emulation safe against themselves, but not against
240 * normal read/write by using a mutex. slows down execution a lot,
241 * but that's your problem, not mine.
242 */
243static xmutex_t preadwritelock = X_MUTEX_INIT;
244#endif
245
246typedef struct etp_worker
247{
248 /* locked by wrklock */
249 struct etp_worker *prev, *next;
250
251 xthread_t tid;
252
253 /* locked by reslock, reqlock or wrklock */
254 ETP_REQ *req; /* currently processed request */
255
256 ETP_WORKER_COMMON
257} etp_worker;
258
259static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
260
261#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
262#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
101 263
102/* worker threads management */ 264/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 265
105typedef struct worker { 266static void ecb_cold
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) 267etp_worker_clear (etp_worker *wrk)
120{ 268{
121 if (wrk->dirp) 269 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} 270}
133 271
134static void worker_free (worker *wrk) 272static void ecb_cold
273etp_worker_free (etp_worker *wrk)
135{ 274{
136 wrk->next->prev = wrk->prev; 275 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 276 wrk->prev->next = wrk->next;
138 277
139 free (wrk); 278 free (wrk);
140} 279}
141 280
142static volatile unsigned int nreqs, nready, npending; 281static unsigned int
143static volatile unsigned int max_idle = 4; 282etp_nreqs (void)
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)
150{ 283{
284 int retval;
285 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
286 retval = nreqs;
287 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 288 return retval;
152} 289}
153 290
154unsigned int eio_nready (void) 291static unsigned int
292etp_nready (void)
155{ 293{
156 unsigned int retval; 294 unsigned int retval;
157 295
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 296 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 297 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 298 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 299
162 return retval; 300 return retval;
163} 301}
164 302
165unsigned int eio_npending (void) 303static unsigned int
304etp_npending (void)
166{ 305{
167 unsigned int retval; 306 unsigned int retval;
168 307
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 308 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 309 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 310 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 311
173 return retval; 312 return retval;
174} 313}
175 314
176unsigned int eio_nthreads (void) 315static unsigned int
316etp_nthreads (void)
177{ 317{
178 unsigned int retval; 318 unsigned int retval;
179 319
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 320 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 321 retval = started;
188 * a somewhat faster data structure might be nice, but 328 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 329 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 330 * per shift, the most expensive operation.
191 */ 331 */
192typedef struct { 332typedef struct {
193 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 333 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194 int size; 334 int size;
195} reqq; 335} etp_reqq;
196 336
197static reqq req_queue; 337static etp_reqq req_queue;
198static reqq res_queue; 338static etp_reqq res_queue;
199 339
200static int reqq_push (reqq *q, eio_req *req) 340static int ecb_noinline
341reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 342{
202 int pri = req->pri; 343 int pri = req->pri;
203 req->next = 0; 344 req->next = 0;
204 345
205 if (q->qe[pri]) 346 if (q->qe[pri])
211 q->qe[pri] = q->qs[pri] = req; 352 q->qe[pri] = q->qs[pri] = req;
212 353
213 return q->size++; 354 return q->size++;
214} 355}
215 356
216static eio_req *reqq_shift (reqq *q) 357static ETP_REQ * ecb_noinline
358reqq_shift (etp_reqq *q)
217{ 359{
218 int pri; 360 int pri;
219 361
220 if (!q->size) 362 if (!q->size)
221 return 0; 363 return 0;
222 364
223 --q->size; 365 --q->size;
224 366
225 for (pri = EIO_NUM_PRI; pri--; ) 367 for (pri = ETP_NUM_PRI; pri--; )
226 { 368 {
227 eio_req *req = q->qs[pri]; 369 eio_req *req = q->qs[pri];
228 370
229 if (req) 371 if (req)
230 { 372 {
236 } 378 }
237 379
238 abort (); 380 abort ();
239} 381}
240 382
241static void grp_try_feed (eio_req *grp) 383static void ecb_cold
384etp_thread_init (void)
242{ 385{
243 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 386 X_MUTEX_CREATE (wrklock);
387 X_MUTEX_CREATE (reslock);
388 X_MUTEX_CREATE (reqlock);
389 X_COND_CREATE (reqwait);
390}
391
392static void ecb_cold
393etp_atfork_prepare (void)
394{
395 X_LOCK (wrklock);
396 X_LOCK (reqlock);
397 X_LOCK (reslock);
398#if !HAVE_PREADWRITE
399 X_LOCK (preadwritelock);
400#endif
401}
402
403static void ecb_cold
404etp_atfork_parent (void)
405{
406#if !HAVE_PREADWRITE
407 X_UNLOCK (preadwritelock);
408#endif
409 X_UNLOCK (reslock);
410 X_UNLOCK (reqlock);
411 X_UNLOCK (wrklock);
412}
413
414static void ecb_cold
415etp_atfork_child (void)
416{
417 ETP_REQ *prv;
418
419 while ((prv = reqq_shift (&req_queue)))
420 ETP_DESTROY (prv);
421
422 while ((prv = reqq_shift (&res_queue)))
423 ETP_DESTROY (prv);
424
425 while (wrk_first.next != &wrk_first)
244 { 426 {
245 int old_len = grp->size; 427 etp_worker *wrk = wrk_first.next;
246 428
247 EIO_FEED (grp); 429 if (wrk->req)
430 ETP_DESTROY (wrk->req);
248 431
249 /* stop if no progress has been made */ 432 etp_worker_clear (wrk);
250 if (old_len == grp->size) 433 etp_worker_free (wrk);
251 {
252 grp->feed = 0;
253 break;
254 }
255 } 434 }
256}
257 435
258static int eio_finish (eio_req *req); 436 started = 0;
437 idle = 0;
438 nreqs = 0;
439 nready = 0;
440 npending = 0;
259 441
260static int grp_dec (eio_req *grp) 442 etp_thread_init ();
261{ 443}
262 --grp->size;
263 444
264 /* call feeder, if applicable */ 445static void ecb_cold
265 grp_try_feed (grp); 446etp_once_init (void)
447{
448 etp_thread_init ();
449 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
450}
266 451
267 /* finish, if done */ 452static int ecb_cold
268 if (!grp->size && grp->int1) 453etp_init (void (*want_poll)(void), void (*done_poll)(void))
269 return eio_finish (grp); 454{
270 else 455 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
456
457 pthread_once (&doinit, etp_once_init);
458
459 want_poll_cb = want_poll;
460 done_poll_cb = done_poll;
461
271 return 0; 462 return 0;
272} 463}
273 464
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); 465X_THREAD_PROC (etp_proc);
325 466
326static void start_thread (void) 467static void ecb_cold
468etp_start_thread (void)
327{ 469{
328 worker *wrk = calloc (1, sizeof (worker)); 470 etp_worker *wrk = calloc (1, sizeof (etp_worker));
329 471
330 /*TODO*/ 472 /*TODO*/
331 assert (("unable to allocate worker thread data", !wrk)); 473 assert (("unable to allocate worker thread data", wrk));
332 474
333 X_LOCK (wrklock); 475 X_LOCK (wrklock);
334 476
335 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 477 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
336 { 478 {
337 wrk->prev = &wrk_first; 479 wrk->prev = &wrk_first;
338 wrk->next = wrk_first.next; 480 wrk->next = wrk_first.next;
339 wrk_first.next->prev = wrk; 481 wrk_first.next->prev = wrk;
340 wrk_first.next = wrk; 482 wrk_first.next = wrk;
344 free (wrk); 486 free (wrk);
345 487
346 X_UNLOCK (wrklock); 488 X_UNLOCK (wrklock);
347} 489}
348 490
491static void
349static void maybe_start_thread (void) 492etp_maybe_start_thread (void)
350{ 493{
351 if (eio_nthreads () >= wanted) 494 if (ecb_expect_true (etp_nthreads () >= wanted))
352 return; 495 return;
353 496
354 /* todo: maybe use idle here, but might be less exact */ 497 /* todo: maybe use idle here, but might be less exact */
355 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 498 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
356 return; 499 return;
357 500
358 start_thread (); 501 etp_start_thread ();
359} 502}
360 503
361void eio_submit (eio_req *req) 504static void ecb_cold
362{ 505etp_end_thread (void)
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)
375{ 506{
376 eio_req *req = calloc (1, sizeof (eio_req)); 507 eio_req *req = calloc (1, sizeof (eio_req));
377 508
378 req->type = EIO_QUIT; 509 req->type = -1;
379 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 510 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
380 511
381 X_LOCK (reqlock); 512 X_LOCK (reqlock);
382 reqq_push (&req_queue, req); 513 reqq_push (&req_queue, req);
383 X_COND_SIGNAL (reqwait); 514 X_COND_SIGNAL (reqwait);
384 X_UNLOCK (reqlock); 515 X_UNLOCK (reqlock);
386 X_LOCK (wrklock); 517 X_LOCK (wrklock);
387 --started; 518 --started;
388 X_UNLOCK (wrklock); 519 X_UNLOCK (wrklock);
389} 520}
390 521
391void eio_set_max_poll_time (double nseconds) 522static int
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) 523etp_poll (void)
428{ 524{
429 int maxreqs = max_poll_reqs; 525 unsigned int maxreqs;
526 unsigned int maxtime;
430 struct timeval tv_start, tv_now; 527 struct timeval tv_start, tv_now;
431 eio_req *req;
432 528
529 X_LOCK (reslock);
530 maxreqs = max_poll_reqs;
531 maxtime = max_poll_time;
532 X_UNLOCK (reslock);
533
433 if (max_poll_time) 534 if (maxtime)
434 gettimeofday (&tv_start, 0); 535 gettimeofday (&tv_start, 0);
435 536
436 for (;;) 537 for (;;)
437 { 538 {
539 ETP_REQ *req;
540
438 maybe_start_thread (); 541 etp_maybe_start_thread ();
439 542
440 X_LOCK (reslock); 543 X_LOCK (reslock);
441 req = reqq_shift (&res_queue); 544 req = reqq_shift (&res_queue);
442 545
443 if (req) 546 if (req)
451 X_UNLOCK (reslock); 554 X_UNLOCK (reslock);
452 555
453 if (!req) 556 if (!req)
454 return 0; 557 return 0;
455 558
559 X_LOCK (reqlock);
456 --nreqs; 560 --nreqs;
561 X_UNLOCK (reqlock);
457 562
458 if (req->type == EIO_GROUP && req->size) 563 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
459 { 564 {
460 req->int1 = 1; /* mark request as delayed */ 565 req->int1 = 1; /* mark request as delayed */
461 continue; 566 continue;
462 } 567 }
463 else 568 else
464 { 569 {
465 int res = eio_finish (req); 570 int res = ETP_FINISH (req);
466 if (res) 571 if (ecb_expect_false (res))
467 return res; 572 return res;
468 } 573 }
469 574
470 if (maxreqs && !--maxreqs) 575 if (ecb_expect_false (maxreqs && !--maxreqs))
471 break; 576 break;
472 577
473 if (max_poll_time) 578 if (maxtime)
474 { 579 {
475 gettimeofday (&tv_now, 0); 580 gettimeofday (&tv_now, 0);
476 581
477 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 582 if (tvdiff (&tv_start, &tv_now) >= maxtime)
478 break; 583 break;
479 } 584 }
480 } 585 }
481 586
482 errno = EAGAIN; 587 errno = EAGAIN;
483 return -1; 588 return -1;
484} 589}
485 590
591static void
592etp_cancel (ETP_REQ *req)
593{
594 X_LOCK (wrklock);
595 req->flags |= EIO_FLAG_CANCELLED;
596 X_UNLOCK (wrklock);
597
598 eio_grp_cancel (req);
599}
600
601static void
602etp_submit (ETP_REQ *req)
603{
604 req->pri -= ETP_PRI_MIN;
605
606 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
607 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
608
609 if (ecb_expect_false (req->type == EIO_GROUP))
610 {
611 /* I hope this is worth it :/ */
612 X_LOCK (reqlock);
613 ++nreqs;
614 X_UNLOCK (reqlock);
615
616 X_LOCK (reslock);
617
618 ++npending;
619
620 if (!reqq_push (&res_queue, req) && want_poll_cb)
621 want_poll_cb ();
622
623 X_UNLOCK (reslock);
624 }
625 else
626 {
627 X_LOCK (reqlock);
628 ++nreqs;
629 ++nready;
630 reqq_push (&req_queue, req);
631 X_COND_SIGNAL (reqwait);
632 X_UNLOCK (reqlock);
633
634 etp_maybe_start_thread ();
635 }
636}
637
638static void ecb_cold
639etp_set_max_poll_time (double nseconds)
640{
641 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
642 max_poll_time = nseconds * EIO_TICKS;
643 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
644}
645
646static void ecb_cold
647etp_set_max_poll_reqs (unsigned int maxreqs)
648{
649 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
650 max_poll_reqs = maxreqs;
651 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
652}
653
654static void ecb_cold
655etp_set_max_idle (unsigned int nthreads)
656{
657 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
658 max_idle = nthreads;
659 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
660}
661
662static void ecb_cold
663etp_set_idle_timeout (unsigned int seconds)
664{
665 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
666 idle_timeout = seconds;
667 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
668}
669
670static void ecb_cold
671etp_set_min_parallel (unsigned int nthreads)
672{
673 if (wanted < nthreads)
674 wanted = nthreads;
675}
676
677static void ecb_cold
678etp_set_max_parallel (unsigned int nthreads)
679{
680 if (wanted > nthreads)
681 wanted = nthreads;
682
683 while (started > wanted)
684 etp_end_thread ();
685}
686
687/*****************************************************************************/
688
689static void
690grp_try_feed (eio_req *grp)
691{
692 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
693 {
694 grp->flags &= ~EIO_FLAG_GROUPADD;
695
696 EIO_FEED (grp);
697
698 /* stop if no progress has been made */
699 if (!(grp->flags & EIO_FLAG_GROUPADD))
700 {
701 grp->feed = 0;
702 break;
703 }
704 }
705}
706
707static int
708grp_dec (eio_req *grp)
709{
710 --grp->size;
711
712 /* call feeder, if applicable */
713 grp_try_feed (grp);
714
715 /* finish, if done */
716 if (!grp->size && grp->int1)
717 return eio_finish (grp);
718 else
719 return 0;
720}
721
722static void
723eio_destroy (eio_req *req)
724{
725 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
726 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
727
728 EIO_DESTROY (req);
729}
730
731static int
732eio_finish (eio_req *req)
733{
734 int res = EIO_FINISH (req);
735
736 if (req->grp)
737 {
738 int res2;
739 eio_req *grp = req->grp;
740
741 /* unlink request */
742 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
743 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
744
745 if (grp->grp_first == req)
746 grp->grp_first = req->grp_next;
747
748 res2 = grp_dec (grp);
749
750 if (!res)
751 res = res2;
752 }
753
754 eio_destroy (req);
755
756 return res;
757}
758
759void
760eio_grp_cancel (eio_req *grp)
761{
762 for (grp = grp->grp_first; grp; grp = grp->grp_next)
763 eio_cancel (grp);
764}
765
766void
767eio_cancel (eio_req *req)
768{
769 etp_cancel (req);
770}
771
772void
773eio_submit (eio_req *req)
774{
775 etp_submit (req);
776}
777
778unsigned int
779eio_nreqs (void)
780{
781 return etp_nreqs ();
782}
783
784unsigned int
785eio_nready (void)
786{
787 return etp_nready ();
788}
789
790unsigned int
791eio_npending (void)
792{
793 return etp_npending ();
794}
795
796unsigned int ecb_cold
797eio_nthreads (void)
798{
799 return etp_nthreads ();
800}
801
802void ecb_cold
803eio_set_max_poll_time (double nseconds)
804{
805 etp_set_max_poll_time (nseconds);
806}
807
808void ecb_cold
809eio_set_max_poll_reqs (unsigned int maxreqs)
810{
811 etp_set_max_poll_reqs (maxreqs);
812}
813
814void ecb_cold
815eio_set_max_idle (unsigned int nthreads)
816{
817 etp_set_max_idle (nthreads);
818}
819
820void ecb_cold
821eio_set_idle_timeout (unsigned int seconds)
822{
823 etp_set_idle_timeout (seconds);
824}
825
826void ecb_cold
827eio_set_min_parallel (unsigned int nthreads)
828{
829 etp_set_min_parallel (nthreads);
830}
831
832void ecb_cold
833eio_set_max_parallel (unsigned int nthreads)
834{
835 etp_set_max_parallel (nthreads);
836}
837
838int eio_poll (void)
839{
840 return etp_poll ();
841}
842
486/*****************************************************************************/ 843/*****************************************************************************/
487/* work around various missing functions */ 844/* work around various missing functions */
488 845
489#if !HAVE_PREADWRITE 846#if !HAVE_PREADWRITE
847# undef pread
848# undef pwrite
490# define pread aio_pread 849# define pread eio__pread
491# define pwrite aio_pwrite 850# define pwrite eio__pwrite
492 851
493/* 852static ssize_t
494 * make our pread/pwrite safe against themselves, but not against 853eio__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{ 854{
502 ssize_t res; 855 ssize_t res;
503 off_t ooffset; 856 off_t ooffset;
504 857
505 X_LOCK (preadwritelock); 858 X_LOCK (preadwritelock);
510 X_UNLOCK (preadwritelock); 863 X_UNLOCK (preadwritelock);
511 864
512 return res; 865 return res;
513} 866}
514 867
515static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 868static ssize_t
869eio__pwrite (int fd, void *buf, size_t count, off_t offset)
516{ 870{
517 ssize_t res; 871 ssize_t res;
518 off_t ooffset; 872 off_t ooffset;
519 873
520 X_LOCK (preadwritelock); 874 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR); 875 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET); 876 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count); 877 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET); 878 lseek (fd, ooffset, SEEK_SET);
525 X_UNLOCK (preadwritelock); 879 X_UNLOCK (preadwritelock);
526 880
527 return res; 881 return res;
528} 882}
529#endif 883#endif
530 884
531#ifndef HAVE_FUTIMES 885#ifndef HAVE_UTIMES
532 886
887# undef utimes
533# define utimes(path,times) aio_utimes (path, times) 888# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) aio_futimes (fd, times)
535 889
890static int
536static int aio_utimes (const char *filename, const struct timeval times[2]) 891eio__utimes (const char *filename, const struct timeval times[2])
537{ 892{
538 if (times) 893 if (times)
539 { 894 {
540 struct utimbuf buf; 895 struct utimbuf buf;
541 896
546 } 901 }
547 else 902 else
548 return utime (filename, 0); 903 return utime (filename, 0);
549} 904}
550 905
906#endif
907
908#ifndef HAVE_FUTIMES
909
910# undef futimes
911# define futimes(fd,times) eio__futimes (fd, times)
912
913static int
551static int aio_futimes (int fd, const struct timeval tv[2]) 914eio__futimes (int fd, const struct timeval tv[2])
552{ 915{
553 errno = ENOSYS; 916 errno = ENOSYS;
554 return -1; 917 return -1;
555} 918}
556 919
557#endif 920#endif
558 921
559#if !HAVE_FDATASYNC 922#if !HAVE_FDATASYNC
923# undef fdatasync
560# define fdatasync fsync 924# define fdatasync(fd) fsync (fd)
561#endif 925#endif
926
927/* sync_file_range always needs emulation */
928static int
929eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
930{
931#if HAVE_SYNC_FILE_RANGE
932 int res;
933
934 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
935 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
936 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
937 {
938 flags = 0
939 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
940 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
941 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
942 }
943
944 res = sync_file_range (fd, offset, nbytes, flags);
945
946 if (!res || errno != ENOSYS)
947 return res;
948#endif
949
950 /* even though we could play tricks with the flags, it's better to always
951 * call fdatasync, as that matches the expectation of its users best */
952 return fdatasync (fd);
953}
562 954
563#if !HAVE_READAHEAD 955#if !HAVE_READAHEAD
956# undef readahead
564# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 957# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565 958
959static ssize_t
566static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 960eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
567{ 961{
568 size_t todo = count; 962 size_t todo = count;
569 dBUF; 963 dBUF;
570 964
571 while (todo > 0) 965 while (todo > 0)
581 return count; 975 return count;
582} 976}
583 977
584#endif 978#endif
585 979
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 */ 980/* sendfile always needs emulation */
981static ssize_t
617static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 982eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
618{ 983{
984 ssize_t written = 0;
619 ssize_t res; 985 ssize_t res;
620 986
621 if (!count) 987 if (!count)
622 return 0; 988 return 0;
623 989
990 for (;;)
991 {
624#if HAVE_SENDFILE 992#if HAVE_SENDFILE
625# if __linux 993# if __linux
994 off_t soffset = offset;
626 res = sendfile (ofd, ifd, &offset, count); 995 res = sendfile (ofd, ifd, &soffset, count);
627 996
628# elif __freebsd 997# elif __FreeBSD__
629 /* 998 /*
630 * Of course, the freebsd sendfile is a dire hack with no thoughts 999 * Of course, the freebsd sendfile is a dire hack with no thoughts
631 * wasted on making it similar to other I/O functions. 1000 * wasted on making it similar to other I/O functions.
632 */ 1001 */
633 {
634 off_t sbytes; 1002 off_t sbytes;
635 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1003 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636 1004
637 if (res < 0 && sbytes) 1005 #if 0 /* according to the manpage, this is correct, but broken behaviour */
638 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1006 /* freebsd' sendfile will return 0 on success */
1007 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1008 /* not on e.g. EIO or EPIPE - sounds broken */
1009 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
639 res = sbytes; 1010 res = sbytes;
640 } 1011 #endif
1012
1013 /* according to source inspection, this is correct, and useful behaviour */
1014 if (sbytes)
1015 res = sbytes;
1016
1017# elif defined (__APPLE__) && 0 /* broken, as everything on os x */
1018 off_t sbytes = count;
1019 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1020
1021 /* according to the manpage, sbytes is always valid */
1022 if (sbytes)
1023 res = sbytes;
641 1024
642# elif __hpux 1025# elif __hpux
643 res = sendfile (ofd, ifd, offset, count, 0, 0); 1026 res = sendfile (ofd, ifd, offset, count, 0, 0);
644 1027
645# elif __solaris 1028# elif __solaris
646 {
647 struct sendfilevec vec; 1029 struct sendfilevec vec;
648 size_t sbytes; 1030 size_t sbytes;
649 1031
650 vec.sfv_fd = ifd; 1032 vec.sfv_fd = ifd;
651 vec.sfv_flag = 0; 1033 vec.sfv_flag = 0;
652 vec.sfv_off = offset; 1034 vec.sfv_off = offset;
653 vec.sfv_len = count; 1035 vec.sfv_len = count;
654 1036
655 res = sendfilev (ofd, &vec, 1, &sbytes); 1037 res = sendfilev (ofd, &vec, 1, &sbytes);
656 1038
657 if (res < 0 && sbytes) 1039 if (res < 0 && sbytes)
658 res = sbytes; 1040 res = sbytes;
659 }
660 1041
661# endif 1042# endif
1043
1044#elif defined (_WIN32)
1045 /* does not work, just for documentation of what would need to be done */
1046 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1047 /* libeio guarantees that the file offset does not change, and windows */
1048 /* has no way to get an independent handle to the same file description */
1049 HANDLE h = TO_SOCKET (ifd);
1050 SetFilePointer (h, offset, 0, FILE_BEGIN);
1051 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1052
662#else 1053#else
663 res = -1; 1054 res = -1;
664 errno = ENOSYS; 1055 errno = ENOSYS;
665#endif 1056#endif
666 1057
1058 /* we assume sendfile can copy at least 128mb in one go */
1059 if (res <= 128 * 1024 * 1024)
1060 {
1061 if (res > 0)
1062 written += res;
1063
1064 if (written)
1065 return written;
1066
1067 break;
1068 }
1069 else
1070 {
1071 /* if we requested more, then probably the kernel was lazy */
1072 written += res;
1073 offset += res;
1074 count -= res;
1075
1076 if (!count)
1077 return written;
1078 }
1079 }
1080
667 if (res < 0 1081 if (res < 0
668 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1082 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1083 /* BSDs */
1084#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1085 || errno == ENOTSUP
1086#endif
1087 || errno == EOPNOTSUPP /* BSDs */
669#if __solaris 1088#if __solaris
670 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1089 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
671#endif 1090#endif
672 ) 1091 )
673 ) 1092 )
704 } 1123 }
705 1124
706 return res; 1125 return res;
707} 1126}
708 1127
1128#ifdef PAGESIZE
1129# define eio_pagesize() PAGESIZE
1130#else
1131static intptr_t
1132eio_pagesize (void)
1133{
1134 static intptr_t page;
1135
1136 if (!page)
1137 page = sysconf (_SC_PAGESIZE);
1138
1139 return page;
1140}
1141#endif
1142
1143static void
1144eio_page_align (void **addr, size_t *length)
1145{
1146 intptr_t mask = eio_pagesize () - 1;
1147
1148 /* round down addr */
1149 intptr_t adj = mask & (intptr_t)*addr;
1150
1151 *addr = (void *)((intptr_t)*addr - adj);
1152 *length += adj;
1153
1154 /* round up length */
1155 *length = (*length + mask) & ~mask;
1156}
1157
1158#if !_POSIX_MEMLOCK
1159# define eio__mlockall(a) ((errno = ENOSYS), -1)
1160#else
1161
1162static int
1163eio__mlockall (int flags)
1164{
1165 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1166 extern int mallopt (int, int);
1167 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1168 #endif
1169
1170 if (EIO_MCL_CURRENT != MCL_CURRENT
1171 || EIO_MCL_FUTURE != MCL_FUTURE)
1172 {
1173 flags = 0
1174 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1175 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1176 }
1177
1178 return mlockall (flags);
1179}
1180#endif
1181
1182#if !_POSIX_MEMLOCK_RANGE
1183# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1184#else
1185
1186static int
1187eio__mlock (void *addr, size_t length)
1188{
1189 eio_page_align (&addr, &length);
1190
1191 return mlock (addr, length);
1192}
1193
1194#endif
1195
1196#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1197# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1198#else
1199
1200static int
1201eio__msync (void *mem, size_t len, int flags)
1202{
1203 eio_page_align (&mem, &len);
1204
1205 if (EIO_MS_ASYNC != MS_SYNC
1206 || EIO_MS_INVALIDATE != MS_INVALIDATE
1207 || EIO_MS_SYNC != MS_SYNC)
1208 {
1209 flags = 0
1210 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1211 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1212 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1213 }
1214
1215 return msync (mem, len, flags);
1216}
1217
1218#endif
1219
1220static int
1221eio__mtouch (eio_req *req)
1222{
1223 void *mem = req->ptr2;
1224 size_t len = req->size;
1225 int flags = req->int1;
1226
1227 eio_page_align (&mem, &len);
1228
1229 {
1230 intptr_t addr = (intptr_t)mem;
1231 intptr_t end = addr + len;
1232 intptr_t page = eio_pagesize ();
1233
1234 if (addr < end)
1235 if (flags & EIO_MT_MODIFY) /* modify */
1236 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1237 else
1238 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1239 }
1240
1241 return 0;
1242}
1243
1244/*****************************************************************************/
1245/* requests implemented outside eio_execute, because they are so large */
1246
1247static void
1248eio__realpath (eio_req *req, etp_worker *self)
1249{
1250 char *rel = req->ptr1;
1251 char *res;
1252 char *tmp1, *tmp2;
1253#if SYMLOOP_MAX > 32
1254 int symlinks = SYMLOOP_MAX;
1255#else
1256 int symlinks = 32;
1257#endif
1258
1259 req->result = -1;
1260
1261 errno = EINVAL;
1262 if (!rel)
1263 return;
1264
1265 errno = ENOENT;
1266 if (!*rel)
1267 return;
1268
1269 if (!req->ptr2)
1270 {
1271 X_LOCK (wrklock);
1272 req->flags |= EIO_FLAG_PTR2_FREE;
1273 X_UNLOCK (wrklock);
1274 req->ptr2 = malloc (PATH_MAX * 3);
1275
1276 errno = ENOMEM;
1277 if (!req->ptr2)
1278 return;
1279 }
1280
1281 res = req->ptr2;
1282 tmp1 = res + PATH_MAX;
1283 tmp2 = tmp1 + PATH_MAX;
1284
1285#if 0 /* disabled, the musl way to do things is just too racy */
1286#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1287 /* on linux we may be able to ask the kernel */
1288 {
1289 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1290
1291 if (fd >= 0)
1292 {
1293 sprintf (tmp1, "/proc/self/fd/%d", fd);
1294 req->result = readlink (tmp1, res, PATH_MAX);
1295 close (fd);
1296
1297 /* here we should probably stat the open file and the disk file, to make sure they still match */
1298
1299 if (req->result > 0)
1300 goto done;
1301 }
1302 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1303 return;
1304 }
1305#endif
1306#endif
1307
1308 if (*rel != '/')
1309 {
1310 if (!getcwd (res, PATH_MAX))
1311 return;
1312
1313 if (res [1]) /* only use if not / */
1314 res += strlen (res);
1315 }
1316
1317 while (*rel)
1318 {
1319 ssize_t len, linklen;
1320 char *beg = rel;
1321
1322 while (*rel && *rel != '/')
1323 ++rel;
1324
1325 len = rel - beg;
1326
1327 if (!len) /* skip slashes */
1328 {
1329 ++rel;
1330 continue;
1331 }
1332
1333 if (beg [0] == '.')
1334 {
1335 if (len == 1)
1336 continue; /* . - nop */
1337
1338 if (beg [1] == '.' && len == 2)
1339 {
1340 /* .. - back up one component, if possible */
1341
1342 while (res != req->ptr2)
1343 if (*--res == '/')
1344 break;
1345
1346 continue;
1347 }
1348 }
1349
1350 errno = ENAMETOOLONG;
1351 if (res + 1 + len + 1 >= tmp1)
1352 return;
1353
1354 /* copy one component */
1355 *res = '/';
1356 memcpy (res + 1, beg, len);
1357
1358 /* zero-terminate, for readlink */
1359 res [len + 1] = 0;
1360
1361 /* now check if it's a symlink */
1362 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1363
1364 if (linklen < 0)
1365 {
1366 if (errno != EINVAL)
1367 return;
1368
1369 /* it's a normal directory. hopefully */
1370 res += len + 1;
1371 }
1372 else
1373 {
1374 /* yay, it was a symlink - build new path in tmp2 */
1375 int rellen = strlen (rel);
1376
1377 errno = ENAMETOOLONG;
1378 if (linklen + 1 + rellen >= PATH_MAX)
1379 return;
1380
1381 errno = ELOOP;
1382 if (!--symlinks)
1383 return;
1384
1385 if (*tmp1 == '/')
1386 res = req->ptr2; /* symlink resolves to an absolute path */
1387
1388 /* we need to be careful, as rel might point into tmp2 already */
1389 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1390 tmp2 [linklen] = '/';
1391 memcpy (tmp2, tmp1, linklen);
1392
1393 rel = tmp2;
1394 }
1395 }
1396
1397 /* special case for the lone root path */
1398 if (res == req->ptr2)
1399 *res++ = '/';
1400
1401 req->result = res - (char *)req->ptr2;
1402
1403done:
1404 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1405}
1406
1407static signed char
1408eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1409{
1410 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1411 : a->inode < b->inode ? -1
1412 : a->inode > b->inode ? 1
1413 : 0;
1414}
1415
1416#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1417
1418#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1419#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1420
1421static void
1422eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1423{
1424 unsigned char bits [9 + sizeof (ino_t) * 8];
1425 unsigned char *bit = bits;
1426
1427 assert (CHAR_BIT == 8);
1428 assert (sizeof (eio_dirent) * 8 < 256);
1429 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1430 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1431
1432 if (size <= EIO_SORT_FAST)
1433 return;
1434
1435 /* first prepare an array of bits to test in our radix sort */
1436 /* try to take endianness into account, as well as differences in ino_t sizes */
1437 /* inode_bits must contain all inodes ORed together */
1438 /* which is used to skip bits that are 0 everywhere, which is very common */
1439 {
1440 ino_t endianness;
1441 int i, j;
1442
1443 /* we store the byte offset of byte n into byte n of "endianness" */
1444 for (i = 0; i < sizeof (ino_t); ++i)
1445 ((unsigned char *)&endianness)[i] = i;
1446
1447 *bit++ = 0;
1448
1449 for (i = 0; i < sizeof (ino_t); ++i)
1450 {
1451 /* shifting off the byte offsets out of "endianness" */
1452 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1453 endianness >>= 8;
1454
1455 for (j = 0; j < 8; ++j)
1456 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1457 *bit++ = offs + j;
1458 }
1459
1460 for (j = 0; j < 8; ++j)
1461 if (score_bits & (1 << j))
1462 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1463 }
1464
1465 /* now actually do the sorting (a variant of MSD radix sort) */
1466 {
1467 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1468 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1469 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1470 int stk_idx = 0;
1471
1472 base_stk [stk_idx] = dents;
1473 end_stk [stk_idx] = dents + size;
1474 bit_stk [stk_idx] = bit - 1;
1475
1476 do
1477 {
1478 base = base_stk [stk_idx];
1479 end = end_stk [stk_idx];
1480 bit = bit_stk [stk_idx];
1481
1482 for (;;)
1483 {
1484 unsigned char O = *bit >> 3;
1485 unsigned char M = 1 << (*bit & 7);
1486
1487 eio_dirent *a = base;
1488 eio_dirent *b = end;
1489
1490 if (b - a < EIO_SORT_CUTOFF)
1491 break;
1492
1493 /* now bit-partition the array on the bit */
1494 /* this ugly asymmetric loop seems to perform much better than typical */
1495 /* partition algos found in the literature */
1496 do
1497 if (!(((unsigned char *)a)[O] & M))
1498 ++a;
1499 else if (!(((unsigned char *)--b)[O] & M))
1500 {
1501 eio_dirent tmp = *a; *a = *b; *b = tmp;
1502 ++a;
1503 }
1504 while (b > a);
1505
1506 /* next bit, or stop, if no bits left in this path */
1507 if (!*--bit)
1508 break;
1509
1510 base_stk [stk_idx] = a;
1511 end_stk [stk_idx] = end;
1512 bit_stk [stk_idx] = bit;
1513 ++stk_idx;
1514
1515 end = a;
1516 }
1517 }
1518 while (stk_idx--);
1519 }
1520}
1521
1522static void
1523eio_dent_insertion_sort (eio_dirent *dents, int size)
1524{
1525 /* first move the smallest element to the front, to act as a sentinel */
1526 {
1527 int i;
1528 eio_dirent *min = dents;
1529
1530 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1531 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1532 if (EIO_DENT_CMP (dents [i], <, *min))
1533 min = &dents [i];
1534
1535 /* swap elements 0 and j (minimum) */
1536 {
1537 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1538 }
1539 }
1540
1541 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1542 {
1543 eio_dirent *i, *j;
1544
1545 for (i = dents + 1; i < dents + size; ++i)
1546 {
1547 eio_dirent value = *i;
1548
1549 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1550 j [1] = j [0];
1551
1552 j [1] = value;
1553 }
1554 }
1555}
1556
1557static void
1558eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1559{
1560 if (size <= 1)
1561 return; /* our insertion sort relies on size > 0 */
1562
1563 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1564 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1565 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1566
1567 /* use an insertion sort at the end, or for small arrays, */
1568 /* as insertion sort is more efficient for small partitions */
1569 eio_dent_insertion_sort (dents, size);
1570}
1571
709/* read a full directory */ 1572/* read a full directory */
1573static void
710static void scandir_ (eio_req *req, worker *self) 1574eio__scandir (eio_req *req, etp_worker *self)
711{ 1575{
712 DIR *dirp; 1576 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; 1577 EIO_STRUCT_DIRENT *entp;
719 char *name, *names; 1578 char *name, *names;
720 int memlen = 4096; 1579 int namesalloc = 4096;
721 int memofs = 0; 1580 int namesoffs = 0;
1581 int flags = req->int1;
1582 eio_dirent *dents = 0;
1583 int dentalloc = 128;
722 int res = 0; 1584 int dentoffs = 0;
1585 ino_t inode_bits = 0;
1586
1587 req->result = -1;
1588
1589 if (!(flags & EIO_READDIR_DENTS))
1590 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
723 1591
724 X_LOCK (wrklock); 1592 X_LOCK (wrklock);
1593 /* the corresponding closedir is in ETP_WORKER_CLEAR */
725 self->dirp = dirp = opendir (req->ptr1); 1594 self->dirp = dirp = opendir (req->ptr1);
726 self->dbuf = u = malloc (sizeof (*u)); 1595
727 req->flags |= EIO_FLAG_PTR2_FREE; 1596 if (req->flags & EIO_FLAG_PTR1_FREE)
1597 free (req->ptr1);
1598
1599 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1600 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
728 req->ptr2 = names = malloc (memlen); 1601 req->ptr2 = names = malloc (namesalloc);
729 X_UNLOCK (wrklock); 1602 X_UNLOCK (wrklock);
730 1603
731 if (dirp && u && names) 1604 if (dirp && names && (!flags || dents))
732 for (;;) 1605 for (;;)
733 { 1606 {
734 errno = 0; 1607 errno = 0;
735 readdir_r (dirp, &u->d, &entp); 1608 entp = readdir (dirp);
736 1609
737 if (!entp) 1610 if (!entp)
1611 {
1612 if (errno)
1613 break;
1614
1615 /* sort etc. */
1616 req->int1 = flags;
1617 req->result = dentoffs;
1618
1619 if (flags & EIO_READDIR_STAT_ORDER)
1620 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1621 else if (flags & EIO_READDIR_DIRS_FIRST)
1622 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1623 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1624 else
1625 {
1626 /* in this case, all is known, and we just put dirs first and sort them */
1627 eio_dirent *oth = dents + dentoffs;
1628 eio_dirent *dir = dents;
1629
1630 /* now partition dirs to the front, and non-dirs to the back */
1631 /* by walking from both sides and swapping if necessary */
1632 while (oth > dir)
1633 {
1634 if (dir->type == EIO_DT_DIR)
1635 ++dir;
1636 else if ((--oth)->type == EIO_DT_DIR)
1637 {
1638 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1639
1640 ++dir;
1641 }
1642 }
1643
1644 /* now sort the dirs only (dirs all have the same score) */
1645 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1646 }
1647
738 break; 1648 break;
1649 }
739 1650
1651 /* now add the entry to our list(s) */
740 name = entp->d_name; 1652 name = entp->d_name;
741 1653
1654 /* skip . and .. entries */
742 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1655 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
743 { 1656 {
744 int len = strlen (name) + 1; 1657 int len = D_NAMLEN (entp) + 1;
745 1658
746 res++; 1659 while (ecb_expect_false (namesoffs + len > namesalloc))
747
748 while (memofs + len > memlen)
749 { 1660 {
750 memlen *= 2; 1661 namesalloc *= 2;
751 X_LOCK (wrklock); 1662 X_LOCK (wrklock);
752 req->ptr2 = names = realloc (names, memlen); 1663 req->ptr2 = names = realloc (names, namesalloc);
753 X_UNLOCK (wrklock); 1664 X_UNLOCK (wrklock);
754 1665
755 if (!names) 1666 if (!names)
756 break; 1667 break;
757 } 1668 }
758 1669
759 memcpy (names + memofs, name, len); 1670 memcpy (names + namesoffs, name, len);
1671
1672 if (dents)
1673 {
1674 struct eio_dirent *ent;
1675
1676 if (ecb_expect_false (dentoffs == dentalloc))
1677 {
1678 dentalloc *= 2;
1679 X_LOCK (wrklock);
1680 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1681 X_UNLOCK (wrklock);
1682
1683 if (!dents)
1684 break;
1685 }
1686
1687 ent = dents + dentoffs;
1688
1689 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1690 ent->namelen = len - 1;
1691 ent->inode = D_INO (entp);
1692
1693 inode_bits |= ent->inode;
1694
1695 switch (D_TYPE (entp))
1696 {
1697 default:
1698 ent->type = EIO_DT_UNKNOWN;
1699 flags |= EIO_READDIR_FOUND_UNKNOWN;
1700 break;
1701
1702 #ifdef DT_FIFO
1703 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1704 #endif
1705 #ifdef DT_CHR
1706 case DT_CHR: ent->type = EIO_DT_CHR; break;
1707 #endif
1708 #ifdef DT_MPC
1709 case DT_MPC: ent->type = EIO_DT_MPC; break;
1710 #endif
1711 #ifdef DT_DIR
1712 case DT_DIR: ent->type = EIO_DT_DIR; break;
1713 #endif
1714 #ifdef DT_NAM
1715 case DT_NAM: ent->type = EIO_DT_NAM; break;
1716 #endif
1717 #ifdef DT_BLK
1718 case DT_BLK: ent->type = EIO_DT_BLK; break;
1719 #endif
1720 #ifdef DT_MPB
1721 case DT_MPB: ent->type = EIO_DT_MPB; break;
1722 #endif
1723 #ifdef DT_REG
1724 case DT_REG: ent->type = EIO_DT_REG; break;
1725 #endif
1726 #ifdef DT_NWK
1727 case DT_NWK: ent->type = EIO_DT_NWK; break;
1728 #endif
1729 #ifdef DT_CMP
1730 case DT_CMP: ent->type = EIO_DT_CMP; break;
1731 #endif
1732 #ifdef DT_LNK
1733 case DT_LNK: ent->type = EIO_DT_LNK; break;
1734 #endif
1735 #ifdef DT_SOCK
1736 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1737 #endif
1738 #ifdef DT_DOOR
1739 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1740 #endif
1741 #ifdef DT_WHT
1742 case DT_WHT: ent->type = EIO_DT_WHT; break;
1743 #endif
1744 }
1745
1746 ent->score = 7;
1747
1748 if (flags & EIO_READDIR_DIRS_FIRST)
1749 {
1750 if (ent->type == EIO_DT_UNKNOWN)
1751 {
1752 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1753 ent->score = 1;
1754 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1755 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1756 }
1757 else if (ent->type == EIO_DT_DIR)
1758 ent->score = 0;
1759 }
1760 }
1761
760 memofs += len; 1762 namesoffs += len;
1763 ++dentoffs;
1764 }
1765
1766 if (EIO_CANCELLED (req))
1767 {
1768 errno = ECANCELED;
1769 break;
761 } 1770 }
762 } 1771 }
763
764 if (errno)
765 res = -1;
766
767 req->result = res;
768} 1772}
769 1773
770/*****************************************************************************/ 1774/*****************************************************************************/
771 1775
772#define ALLOC(len) \ 1776#define ALLOC(len) \
782 req->result = -1; \ 1786 req->result = -1; \
783 break; \ 1787 break; \
784 } \ 1788 } \
785 } 1789 }
786 1790
787X_THREAD_PROC (eio_proc) 1791X_THREAD_PROC (etp_proc)
788{ 1792{
789 eio_req *req; 1793 ETP_REQ *req;
790 struct timespec ts; 1794 struct timespec ts;
791 worker *self = (worker *)thr_arg; 1795 etp_worker *self = (etp_worker *)thr_arg;
792 1796
793 /* try to distribute timeouts somewhat randomly */ 1797 /* try to distribute timeouts somewhat randomly */
794 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1798 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
795 1799
796 for (;;) 1800 for (;;)
797 { 1801 {
798 ts.tv_sec = time (0) + IDLE_TIMEOUT;
799
800 X_LOCK (reqlock); 1802 X_LOCK (reqlock);
801 1803
802 for (;;) 1804 for (;;)
803 { 1805 {
804 self->req = req = reqq_shift (&req_queue); 1806 self->req = req = reqq_shift (&req_queue);
806 if (req) 1808 if (req)
807 break; 1809 break;
808 1810
809 ++idle; 1811 ++idle;
810 1812
1813 ts.tv_sec = time (0) + idle_timeout;
811 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1814 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
812 { 1815 {
813 if (idle > max_idle) 1816 if (idle > max_idle)
814 { 1817 {
815 --idle; 1818 --idle;
820 goto quit; 1823 goto quit;
821 } 1824 }
822 1825
823 /* we are allowed to idle, so do so without any timeout */ 1826 /* we are allowed to idle, so do so without any timeout */
824 X_COND_WAIT (reqwait, reqlock); 1827 X_COND_WAIT (reqwait, reqlock);
825 ts.tv_sec = time (0) + IDLE_TIMEOUT;
826 } 1828 }
827 1829
828 --idle; 1830 --idle;
829 } 1831 }
830 1832
831 --nready; 1833 --nready;
832 1834
833 X_UNLOCK (reqlock); 1835 X_UNLOCK (reqlock);
834 1836
835 errno = 0; /* strictly unnecessary */
836
837 if (!EIO_CANCELLED (req))
838 switch (req->type) 1837 if (req->type < 0)
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; 1838 goto quit;
930 1839
931 default: 1840 ETP_EXECUTE (self, req);
932 req->result = -1;
933 break;
934 }
935
936 req->errorno = errno;
937 1841
938 X_LOCK (reslock); 1842 X_LOCK (reslock);
939 1843
940 ++npending; 1844 ++npending;
941 1845
942 if (!reqq_push (&res_queue, req) && want_poll_cb) 1846 if (!reqq_push (&res_queue, req) && want_poll_cb)
943 want_poll_cb (); 1847 want_poll_cb ();
944 1848
945 self->req = 0; 1849 self->req = 0;
946 worker_clear (self); 1850 etp_worker_clear (self);
947 1851
948 X_UNLOCK (reslock); 1852 X_UNLOCK (reslock);
949 } 1853 }
950 1854
951quit: 1855quit:
952 X_LOCK (wrklock); 1856 X_LOCK (wrklock);
953 worker_free (self); 1857 etp_worker_free (self);
954 X_UNLOCK (wrklock); 1858 X_UNLOCK (wrklock);
955 1859
956 return 0; 1860 return 0;
957} 1861}
958 1862
959/*****************************************************************************/ 1863/*****************************************************************************/
960 1864
961static void eio_atfork_prepare (void) 1865int ecb_cold
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)) 1866eio_init (void (*want_poll)(void), void (*done_poll)(void))
1018{ 1867{
1019 want_poll_cb = want_poll; 1868 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} 1869}
1035 1870
1871ecb_inline void
1036static void eio_api_destroy (eio_req *req) 1872eio_api_destroy (eio_req *req)
1037{ 1873{
1038 free (req); 1874 free (req);
1039} 1875}
1040 1876
1041#define REQ(rtype) \ 1877#define REQ(rtype) \
1043 \ 1879 \
1044 req = (eio_req *)calloc (1, sizeof *req); \ 1880 req = (eio_req *)calloc (1, sizeof *req); \
1045 if (!req) \ 1881 if (!req) \
1046 return 0; \ 1882 return 0; \
1047 \ 1883 \
1048 req->type = EIO_ ## rtype; \ 1884 req->type = rtype; \
1049 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 1885 req->pri = pri; \
1050 req->finish = cb; \ 1886 req->finish = cb; \
1887 req->data = data; \
1051 req->destroy = eio_api_destroy; 1888 req->destroy = eio_api_destroy;
1052 1889
1053#define SEND eio_submit (req); return req 1890#define SEND eio_submit (req); return req
1054 1891
1055#define PATH (void)0 1892#define PATH \
1893 req->flags |= EIO_FLAG_PTR1_FREE; \
1894 req->ptr1 = strdup (path); \
1895 if (!req->ptr1) \
1896 { \
1897 eio_api_destroy (req); \
1898 return 0; \
1899 }
1056 1900
1901static void
1902eio_execute (etp_worker *self, eio_req *req)
1903{
1904 if (ecb_expect_false (EIO_CANCELLED (req)))
1905 {
1906 req->result = -1;
1907 req->errorno = ECANCELED;
1908 return;
1909 }
1910
1911 switch (req->type)
1912 {
1913 case EIO_READ: ALLOC (req->size);
1914 req->result = req->offs >= 0
1915 ? pread (req->int1, req->ptr2, req->size, req->offs)
1916 : read (req->int1, req->ptr2, req->size); break;
1917 case EIO_WRITE: req->result = req->offs >= 0
1918 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1919 : write (req->int1, req->ptr2, req->size); break;
1920
1921 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1922 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1923
1924 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1925 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1926 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1927 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1928 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1929 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1930
1931 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1932 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1933 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1934 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1935
1936 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1937 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1938 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1939 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1940 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1941 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1942
1943 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1944 case EIO_CLOSE: req->result = close (req->int1); break;
1945 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1946 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1947 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1948 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1949 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1950 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1951 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1952 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1953
1954 case EIO_REALPATH: eio__realpath (req, self); break;
1955
1956 case EIO_READLINK: ALLOC (PATH_MAX);
1957 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1958
1959 case EIO_SYNC: req->result = 0; sync (); break;
1960 case EIO_FSYNC: req->result = fsync (req->int1); break;
1961 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1962 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1963 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1964 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1965 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1966 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1967
1968 case EIO_READDIR: eio__scandir (req, self); break;
1969
1970 case EIO_BUSY:
1971#ifdef _WIN32
1972 Sleep (req->nv1 * 1e3);
1973#else
1974 {
1975 struct timeval tv;
1976
1977 tv.tv_sec = req->nv1;
1978 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1979
1980 req->result = select (0, 0, 0, 0, &tv);
1981 }
1982#endif
1983 break;
1984
1985 case EIO_UTIME:
1986 case EIO_FUTIME:
1987 {
1988 struct timeval tv[2];
1989 struct timeval *times;
1990
1991 if (req->nv1 != -1. || req->nv2 != -1.)
1992 {
1993 tv[0].tv_sec = req->nv1;
1994 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1995 tv[1].tv_sec = req->nv2;
1996 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1997
1998 times = tv;
1999 }
2000 else
2001 times = 0;
2002
2003 req->result = req->type == EIO_FUTIME
2004 ? futimes (req->int1, times)
2005 : utimes (req->ptr1, times);
2006 }
2007 break;
2008
2009 case EIO_GROUP:
2010 abort (); /* handled in eio_request */
2011
2012 case EIO_NOP:
2013 req->result = 0;
2014 break;
2015
2016 case EIO_CUSTOM:
2017 req->feed (req);
2018 break;
2019
2020 default:
2021 errno = ENOSYS;
2022 req->result = -1;
2023 break;
2024 }
2025
2026 req->errorno = errno;
2027}
2028
2029#ifndef EIO_NO_WRAPPERS
2030
2031eio_req *eio_nop (int pri, eio_cb cb, void *data)
2032{
2033 REQ (EIO_NOP); SEND;
2034}
2035
2036eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2037{
2038 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2039}
2040
1057eio_req *eio_fsync (int fd, eio_cb cb) 2041eio_req *eio_sync (int pri, eio_cb cb, void *data)
1058{ 2042{
2043 REQ (EIO_SYNC); SEND;
2044}
2045
2046eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2047{
1059 REQ (FSYNC); req->int1 = fd; SEND; 2048 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1060} 2049}
1061 2050
2051eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2052{
2053 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2054}
2055
2056eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2057{
2058 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2059}
2060
2061eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2062{
2063 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2064}
2065
2066eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2067{
2068 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2069}
2070
2071eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2072{
2073 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2074}
2075
1062eio_req *eio_fdatasync (int fd, eio_cb cb) 2076eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1063{ 2077{
1064 REQ (FDATASYNC); req->int1 = fd; SEND; 2078 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1065} 2079}
1066 2080
2081eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2082{
2083 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2084}
2085
1067eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb) 2086eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1068{ 2087{
1069 REQ (READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2088 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1070} 2089}
1071 2090
1072eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2091eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1073{ 2092{
1074 REQ (READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2093 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1075} 2094}
1076 2095
1077eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2096eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1078{ 2097{
1079 REQ (WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2098 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1080} 2099}
1081 2100
1082eio_req *eio_fstat (int fd, eio_cb cb) 2101eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1083{ 2102{
1084 REQ (FSTAT); req->int1 = fd; SEND; 2103 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1085} 2104}
1086 2105
2106eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2107{
2108 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2109}
2110
1087eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb) 2111eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1088{ 2112{
1089 REQ (FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2113 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1090} 2114}
1091 2115
1092eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb) 2116eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1093{ 2117{
1094 REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 2118 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1095} 2119}
1096 2120
1097eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb) 2121eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1098{ 2122{
1099 REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2123 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1100} 2124}
1101 2125
1102eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb) 2126eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1103{ 2127{
1104 REQ (FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2128 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1105} 2129}
1106 2130
1107eio_req *eio_dup2 (int fd, int fd2, eio_cb cb) 2131eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1108{ 2132{
1109 REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND; 2133 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1110} 2134}
1111 2135
1112eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb) 2136eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1113{ 2137{
1114 REQ (SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 2138 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1115} 2139}
1116 2140
1117eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb) 2141eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1118{ 2142{
1119 REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 2143 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1120} 2144}
1121 2145
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) 2146eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1138{ 2147{
1139 REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 2148 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1140} 2149}
1141 2150
1142eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb) 2151eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1143{ 2152{
1144 REQ (TRUNCATE); PATH; req->offs = offset; SEND; 2153 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1145} 2154}
1146 2155
1147eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 2156eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1148{ 2157{
1149 REQ (CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2158 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1150} 2159}
1151 2160
1152eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb) 2161eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1153{ 2162{
1154 REQ (CHMOD); PATH; req->int2 = (long)mode; SEND; 2163 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1155} 2164}
1156 2165
1157eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb) 2166eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1158{ 2167{
1159 REQ (MKDIR); PATH; req->int2 = (long)mode; SEND; 2168 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1160} 2169}
1161 2170
1162eio_req *eio_unlink (const char *path, eio_cb cb) 2171static eio_req *
2172eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1163{ 2173{
1164 REQ (UNLINK); PATH; SEND;
1165}
1166
1167eio_req *eio_rmdir (const char *path, eio_cb cb)
1168{
1169 REQ (RMDIR); PATH; SEND; 2174 REQ (type); PATH; SEND;
1170} 2175}
1171 2176
2177eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2178{
2179 return eio__1path (EIO_READLINK, path, pri, cb, data);
2180}
2181
2182eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2183{
2184 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2185}
2186
2187eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2188{
2189 return eio__1path (EIO_STAT, path, pri, cb, data);
2190}
2191
2192eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2193{
2194 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2195}
2196
2197eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2198{
2199 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2200}
2201
2202eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2203{
2204 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2205}
2206
1172eio_req *eio_readdir (const char *path, eio_cb cb) 2207eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1173{ 2208{
1174 REQ (READDIR); PATH; SEND; 2209 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1175} 2210}
1176 2211
2212eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2213{
2214 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2215}
2216
1177eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb) 2217eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1178{ 2218{
1179 REQ (MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2219 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1180} 2220}
1181 2221
1182eio_req *eio_busy (double delay, eio_cb cb) 2222static eio_req *
2223eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1183{ 2224{
1184 REQ (BUSY); req->nv1 = delay; SEND; 2225 REQ (type); PATH;
1185}
1186 2226
2227 req->flags |= EIO_FLAG_PTR2_FREE;
2228 req->ptr2 = strdup (new_path);
2229 if (!req->ptr2)
2230 {
2231 eio_api_destroy (req);
2232 return 0;
2233 }
2234
2235 SEND;
2236}
2237
2238eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2239{
2240 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2241}
2242
2243eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2244{
2245 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2246}
2247
2248eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2249{
2250 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2251}
2252
2253eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2254{
2255 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2256}
2257
2258#endif
2259
1187eio_req *eio_nop (eio_cb cb) 2260eio_req *eio_grp (eio_cb cb, void *data)
1188{ 2261{
2262 const int pri = EIO_PRI_MAX;
2263
1189 REQ (NOP); SEND; 2264 REQ (EIO_GROUP); SEND;
1190} 2265}
1191 2266
1192#undef REQ 2267#undef REQ
1193#undef PATH 2268#undef PATH
1194#undef SEND 2269#undef SEND
1195 2270
1196#if 0 2271/*****************************************************************************/
2272/* grp functions */
1197 2273
1198void 2274void
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
1571void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2275eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1572{ 2276{
1573 grp->int2 = limit; 2277 grp->int2 = limit;
1574 grp->feed = feed; 2278 grp->feed = feed;
1575 2279
1576 grp_try_feed (grp); 2280 grp_try_feed (grp);
1577} 2281}
1578 2282
2283void
1579void eio_grp_limit (eio_req *grp, int limit) 2284eio_grp_limit (eio_req *grp, int limit)
1580{ 2285{
1581 grp->int2 = limit; 2286 grp->int2 = limit;
1582 2287
1583 grp_try_feed (grp); 2288 grp_try_feed (grp);
1584} 2289}
1585 2290
2291void
1586void eio_grp_add (eio_req *grp, eio_req *req) 2292eio_grp_add (eio_req *grp, eio_req *req)
1587{ 2293{
1588 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2294 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2295
2296 grp->flags |= EIO_FLAG_GROUPADD;
1589 2297
1590 ++grp->size; 2298 ++grp->size;
1591 req->grp = grp; 2299 req->grp = grp;
1592 2300
1593 req->grp_prev = 0; 2301 req->grp_prev = 0;
1597 grp->grp_first->grp_prev = req; 2305 grp->grp_first->grp_prev = req;
1598 2306
1599 grp->grp_first = req; 2307 grp->grp_first = req;
1600} 2308}
1601 2309
2310/*****************************************************************************/
2311/* misc garbage */
1602 2312
2313ssize_t
2314eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2315{
2316 etp_worker wrk;
2317 ssize_t ret;
2318
2319 wrk.dbuf = 0;
2320
2321 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2322
2323 if (wrk.dbuf)
2324 free (wrk.dbuf);
2325
2326 return ret;
2327}
2328

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines