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 1.82 by root, Thu Jul 7 15:44:44 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 req->cancelled = 1;
595
596 eio_grp_cancel (req);
597}
598
599static void
600etp_submit (ETP_REQ *req)
601{
602 req->pri -= ETP_PRI_MIN;
603
604 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
605 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
606
607 if (ecb_expect_false (req->type == EIO_GROUP))
608 {
609 /* I hope this is worth it :/ */
610 X_LOCK (reqlock);
611 ++nreqs;
612 X_UNLOCK (reqlock);
613
614 X_LOCK (reslock);
615
616 ++npending;
617
618 if (!reqq_push (&res_queue, req) && want_poll_cb)
619 want_poll_cb ();
620
621 X_UNLOCK (reslock);
622 }
623 else
624 {
625 X_LOCK (reqlock);
626 ++nreqs;
627 ++nready;
628 reqq_push (&req_queue, req);
629 X_COND_SIGNAL (reqwait);
630 X_UNLOCK (reqlock);
631
632 etp_maybe_start_thread ();
633 }
634}
635
636static void ecb_cold
637etp_set_max_poll_time (double nseconds)
638{
639 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
640 max_poll_time = nseconds * EIO_TICKS;
641 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
642}
643
644static void ecb_cold
645etp_set_max_poll_reqs (unsigned int maxreqs)
646{
647 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
648 max_poll_reqs = maxreqs;
649 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
650}
651
652static void ecb_cold
653etp_set_max_idle (unsigned int nthreads)
654{
655 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
656 max_idle = nthreads;
657 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
658}
659
660static void ecb_cold
661etp_set_idle_timeout (unsigned int seconds)
662{
663 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
664 idle_timeout = seconds;
665 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
666}
667
668static void ecb_cold
669etp_set_min_parallel (unsigned int nthreads)
670{
671 if (wanted < nthreads)
672 wanted = nthreads;
673}
674
675static void ecb_cold
676etp_set_max_parallel (unsigned int nthreads)
677{
678 if (wanted > nthreads)
679 wanted = nthreads;
680
681 while (started > wanted)
682 etp_end_thread ();
683}
684
685/*****************************************************************************/
686
687static void
688grp_try_feed (eio_req *grp)
689{
690 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
691 {
692 grp->flags &= ~EIO_FLAG_GROUPADD;
693
694 EIO_FEED (grp);
695
696 /* stop if no progress has been made */
697 if (!(grp->flags & EIO_FLAG_GROUPADD))
698 {
699 grp->feed = 0;
700 break;
701 }
702 }
703}
704
705static int
706grp_dec (eio_req *grp)
707{
708 --grp->size;
709
710 /* call feeder, if applicable */
711 grp_try_feed (grp);
712
713 /* finish, if done */
714 if (!grp->size && grp->int1)
715 return eio_finish (grp);
716 else
717 return 0;
718}
719
720static void
721eio_destroy (eio_req *req)
722{
723 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
724 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
725
726 EIO_DESTROY (req);
727}
728
729static int
730eio_finish (eio_req *req)
731{
732 int res = EIO_FINISH (req);
733
734 if (req->grp)
735 {
736 int res2;
737 eio_req *grp = req->grp;
738
739 /* unlink request */
740 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
741 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
742
743 if (grp->grp_first == req)
744 grp->grp_first = req->grp_next;
745
746 res2 = grp_dec (grp);
747
748 if (!res)
749 res = res2;
750 }
751
752 eio_destroy (req);
753
754 return res;
755}
756
757void
758eio_grp_cancel (eio_req *grp)
759{
760 for (grp = grp->grp_first; grp; grp = grp->grp_next)
761 eio_cancel (grp);
762}
763
764void
765eio_cancel (eio_req *req)
766{
767 etp_cancel (req);
768}
769
770void
771eio_submit (eio_req *req)
772{
773 etp_submit (req);
774}
775
776unsigned int
777eio_nreqs (void)
778{
779 return etp_nreqs ();
780}
781
782unsigned int
783eio_nready (void)
784{
785 return etp_nready ();
786}
787
788unsigned int
789eio_npending (void)
790{
791 return etp_npending ();
792}
793
794unsigned int ecb_cold
795eio_nthreads (void)
796{
797 return etp_nthreads ();
798}
799
800void ecb_cold
801eio_set_max_poll_time (double nseconds)
802{
803 etp_set_max_poll_time (nseconds);
804}
805
806void ecb_cold
807eio_set_max_poll_reqs (unsigned int maxreqs)
808{
809 etp_set_max_poll_reqs (maxreqs);
810}
811
812void ecb_cold
813eio_set_max_idle (unsigned int nthreads)
814{
815 etp_set_max_idle (nthreads);
816}
817
818void ecb_cold
819eio_set_idle_timeout (unsigned int seconds)
820{
821 etp_set_idle_timeout (seconds);
822}
823
824void ecb_cold
825eio_set_min_parallel (unsigned int nthreads)
826{
827 etp_set_min_parallel (nthreads);
828}
829
830void ecb_cold
831eio_set_max_parallel (unsigned int nthreads)
832{
833 etp_set_max_parallel (nthreads);
834}
835
836int eio_poll (void)
837{
838 return etp_poll ();
839}
840
486/*****************************************************************************/ 841/*****************************************************************************/
487/* work around various missing functions */ 842/* work around various missing functions */
488 843
489#if !HAVE_PREADWRITE 844#if !HAVE_PREADWRITE
845# undef pread
846# undef pwrite
490# define pread aio_pread 847# define pread eio__pread
491# define pwrite aio_pwrite 848# define pwrite eio__pwrite
492 849
493/* 850static ssize_t
494 * make our pread/pwrite safe against themselves, but not against 851eio__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{ 852{
502 ssize_t res; 853 ssize_t res;
503 off_t ooffset; 854 off_t ooffset;
504 855
505 X_LOCK (preadwritelock); 856 X_LOCK (preadwritelock);
510 X_UNLOCK (preadwritelock); 861 X_UNLOCK (preadwritelock);
511 862
512 return res; 863 return res;
513} 864}
514 865
515static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 866static ssize_t
867eio__pwrite (int fd, void *buf, size_t count, off_t offset)
516{ 868{
517 ssize_t res; 869 ssize_t res;
518 off_t ooffset; 870 off_t ooffset;
519 871
520 X_LOCK (preadwritelock); 872 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR); 873 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET); 874 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count); 875 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET); 876 lseek (fd, ooffset, SEEK_SET);
525 X_UNLOCK (preadwritelock); 877 X_UNLOCK (preadwritelock);
526 878
527 return res; 879 return res;
528} 880}
529#endif 881#endif
530 882
531#ifndef HAVE_FUTIMES 883#ifndef HAVE_UTIMES
532 884
885# undef utimes
533# define utimes(path,times) aio_utimes (path, times) 886# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) aio_futimes (fd, times)
535 887
888static int
536static int aio_utimes (const char *filename, const struct timeval times[2]) 889eio__utimes (const char *filename, const struct timeval times[2])
537{ 890{
538 if (times) 891 if (times)
539 { 892 {
540 struct utimbuf buf; 893 struct utimbuf buf;
541 894
546 } 899 }
547 else 900 else
548 return utime (filename, 0); 901 return utime (filename, 0);
549} 902}
550 903
904#endif
905
906#ifndef HAVE_FUTIMES
907
908# undef futimes
909# define futimes(fd,times) eio__futimes (fd, times)
910
911static int
551static int aio_futimes (int fd, const struct timeval tv[2]) 912eio__futimes (int fd, const struct timeval tv[2])
552{ 913{
553 errno = ENOSYS; 914 errno = ENOSYS;
554 return -1; 915 return -1;
555} 916}
556 917
557#endif 918#endif
558 919
559#if !HAVE_FDATASYNC 920#if !HAVE_FDATASYNC
921# undef fdatasync
560# define fdatasync fsync 922# define fdatasync(fd) fsync (fd)
561#endif 923#endif
924
925/* sync_file_range always needs emulation */
926static int
927eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
928{
929#if HAVE_SYNC_FILE_RANGE
930 int res;
931
932 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
933 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
934 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
935 {
936 flags = 0
937 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
938 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
939 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
940 }
941
942 res = sync_file_range (fd, offset, nbytes, flags);
943
944 if (!res || errno != ENOSYS)
945 return res;
946#endif
947
948 /* even though we could play tricks with the flags, it's better to always
949 * call fdatasync, as that matches the expectation of its users best */
950 return fdatasync (fd);
951}
562 952
563#if !HAVE_READAHEAD 953#if !HAVE_READAHEAD
954# undef readahead
564# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 955# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565 956
957static ssize_t
566static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 958eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
567{ 959{
568 size_t todo = count; 960 size_t todo = count;
569 dBUF; 961 dBUF;
570 962
571 while (todo > 0) 963 while (todo > 0)
581 return count; 973 return count;
582} 974}
583 975
584#endif 976#endif
585 977
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 */ 978/* sendfile always needs emulation */
979static ssize_t
617static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 980eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
618{ 981{
982 ssize_t written = 0;
619 ssize_t res; 983 ssize_t res;
620 984
621 if (!count) 985 if (!count)
622 return 0; 986 return 0;
623 987
988 for (;;)
989 {
990#ifdef __APPLE__
991# undef HAVE_SENDFILE /* broken, as everything on os x */
992#endif
624#if HAVE_SENDFILE 993#if HAVE_SENDFILE
625# if __linux 994# if __linux
995 off_t soffset = offset;
626 res = sendfile (ofd, ifd, &offset, count); 996 res = sendfile (ofd, ifd, &soffset, count);
627 997
628# elif __freebsd 998# elif __FreeBSD__
629 /* 999 /*
630 * Of course, the freebsd sendfile is a dire hack with no thoughts 1000 * Of course, the freebsd sendfile is a dire hack with no thoughts
631 * wasted on making it similar to other I/O functions. 1001 * wasted on making it similar to other I/O functions.
632 */ 1002 */
633 {
634 off_t sbytes; 1003 off_t sbytes;
635 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1004 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636 1005
637 if (res < 0 && sbytes) 1006 #if 0 /* according to the manpage, this is correct, but broken behaviour */
638 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1007 /* freebsd' sendfile will return 0 on success */
1008 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1009 /* not on e.g. EIO or EPIPE - sounds broken */
1010 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
639 res = sbytes; 1011 res = sbytes;
640 } 1012 #endif
1013
1014 /* according to source inspection, this is correct, and useful behaviour */
1015 if (sbytes)
1016 res = sbytes;
1017
1018# elif defined (__APPLE__)
1019 off_t sbytes = count;
1020 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1021
1022 /* according to the manpage, sbytes is always valid */
1023 if (sbytes)
1024 res = sbytes;
641 1025
642# elif __hpux 1026# elif __hpux
643 res = sendfile (ofd, ifd, offset, count, 0, 0); 1027 res = sendfile (ofd, ifd, offset, count, 0, 0);
644 1028
645# elif __solaris 1029# elif __solaris
646 {
647 struct sendfilevec vec; 1030 struct sendfilevec vec;
648 size_t sbytes; 1031 size_t sbytes;
649 1032
650 vec.sfv_fd = ifd; 1033 vec.sfv_fd = ifd;
651 vec.sfv_flag = 0; 1034 vec.sfv_flag = 0;
652 vec.sfv_off = offset; 1035 vec.sfv_off = offset;
653 vec.sfv_len = count; 1036 vec.sfv_len = count;
654 1037
655 res = sendfilev (ofd, &vec, 1, &sbytes); 1038 res = sendfilev (ofd, &vec, 1, &sbytes);
656 1039
657 if (res < 0 && sbytes) 1040 if (res < 0 && sbytes)
658 res = sbytes; 1041 res = sbytes;
659 }
660 1042
661# endif 1043# endif
1044
1045#elif defined (_WIN32)
1046 /* does not work, just for documentation of what would need to be done */
1047 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1048 /* libeio guarantees that the file offset does not change, and windows */
1049 /* has no way to get an independent handle to the same file description */
1050 HANDLE h = TO_SOCKET (ifd);
1051 SetFilePointer (h, offset, 0, FILE_BEGIN);
1052 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1053
662#else 1054#else
663 res = -1; 1055 res = -1;
664 errno = ENOSYS; 1056 errno = ENOSYS;
665#endif 1057#endif
666 1058
1059 /* we assume sendfile can copy at least 128mb in one go */
1060 if (res <= 128 * 1024 * 1024)
1061 {
1062 if (res > 0)
1063 written += res;
1064
1065 if (written)
1066 return written;
1067
1068 break;
1069 }
1070 else
1071 {
1072 /* if we requested more, then probably the kernel was lazy */
1073 written += res;
1074 offset += res;
1075 count -= res;
1076
1077 if (!count)
1078 return written;
1079 }
1080 }
1081
667 if (res < 0 1082 if (res < 0
668 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1083 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1084 /* BSDs */
1085#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1086 || errno == ENOTSUP
1087#endif
1088 || errno == EOPNOTSUPP /* BSDs */
669#if __solaris 1089#if __solaris
670 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1090 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
671#endif 1091#endif
672 ) 1092 )
673 ) 1093 )
704 } 1124 }
705 1125
706 return res; 1126 return res;
707} 1127}
708 1128
1129#ifdef PAGESIZE
1130# define eio_pagesize() PAGESIZE
1131#else
1132static intptr_t
1133eio_pagesize (void)
1134{
1135 static intptr_t page;
1136
1137 if (!page)
1138 page = sysconf (_SC_PAGESIZE);
1139
1140 return page;
1141}
1142#endif
1143
1144static void
1145eio_page_align (void **addr, size_t *length)
1146{
1147 intptr_t mask = eio_pagesize () - 1;
1148
1149 /* round down addr */
1150 intptr_t adj = mask & (intptr_t)*addr;
1151
1152 *addr = (void *)((intptr_t)*addr - adj);
1153 *length += adj;
1154
1155 /* round up length */
1156 *length = (*length + mask) & ~mask;
1157}
1158
1159#if !_POSIX_MEMLOCK
1160# define eio__mlockall(a) ((errno = ENOSYS), -1)
1161#else
1162
1163static int
1164eio__mlockall (int flags)
1165{
1166 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1167 extern int mallopt (int, int);
1168 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1169 #endif
1170
1171 if (EIO_MCL_CURRENT != MCL_CURRENT
1172 || EIO_MCL_FUTURE != MCL_FUTURE)
1173 {
1174 flags = 0
1175 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1176 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1177 }
1178
1179 return mlockall (flags);
1180}
1181#endif
1182
1183#if !_POSIX_MEMLOCK_RANGE
1184# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1185#else
1186
1187static int
1188eio__mlock (void *addr, size_t length)
1189{
1190 eio_page_align (&addr, &length);
1191
1192 return mlock (addr, length);
1193}
1194
1195#endif
1196
1197#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1198# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1199#else
1200
1201static int
1202eio__msync (void *mem, size_t len, int flags)
1203{
1204 eio_page_align (&mem, &len);
1205
1206 if (EIO_MS_ASYNC != MS_SYNC
1207 || EIO_MS_INVALIDATE != MS_INVALIDATE
1208 || EIO_MS_SYNC != MS_SYNC)
1209 {
1210 flags = 0
1211 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1212 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1213 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1214 }
1215
1216 return msync (mem, len, flags);
1217}
1218
1219#endif
1220
1221static int
1222eio__mtouch (eio_req *req)
1223{
1224 void *mem = req->ptr2;
1225 size_t len = req->size;
1226 int flags = req->int1;
1227
1228 eio_page_align (&mem, &len);
1229
1230 {
1231 intptr_t addr = (intptr_t)mem;
1232 intptr_t end = addr + len;
1233 intptr_t page = eio_pagesize ();
1234
1235 if (addr < end)
1236 if (flags & EIO_MT_MODIFY) /* modify */
1237 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1238 else
1239 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1240 }
1241
1242 return 0;
1243}
1244
1245/*****************************************************************************/
1246/* requests implemented outside eio_execute, because they are so large */
1247
1248static void
1249eio__realpath (eio_req *req, etp_worker *self)
1250{
1251 char *rel = req->ptr1;
1252 char *res;
1253 char *tmp1, *tmp2;
1254#if SYMLOOP_MAX > 32
1255 int symlinks = SYMLOOP_MAX;
1256#else
1257 int symlinks = 32;
1258#endif
1259
1260 req->result = -1;
1261
1262 errno = EINVAL;
1263 if (!rel)
1264 return;
1265
1266 errno = ENOENT;
1267 if (!*rel)
1268 return;
1269
1270 if (!req->ptr2)
1271 {
1272 X_LOCK (wrklock);
1273 req->flags |= EIO_FLAG_PTR2_FREE;
1274 X_UNLOCK (wrklock);
1275 req->ptr2 = malloc (PATH_MAX * 3);
1276
1277 errno = ENOMEM;
1278 if (!req->ptr2)
1279 return;
1280 }
1281
1282 res = req->ptr2;
1283 tmp1 = res + PATH_MAX;
1284 tmp2 = tmp1 + PATH_MAX;
1285
1286#if 0 /* disabled, the musl way to do things is just too racy */
1287#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1288 /* on linux we may be able to ask the kernel */
1289 {
1290 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1291
1292 if (fd >= 0)
1293 {
1294 sprintf (tmp1, "/proc/self/fd/%d", fd);
1295 req->result = readlink (tmp1, res, PATH_MAX);
1296 close (fd);
1297
1298 /* here we should probably stat the open file and the disk file, to make sure they still match */
1299
1300 if (req->result > 0)
1301 goto done;
1302 }
1303 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1304 return;
1305 }
1306#endif
1307#endif
1308
1309 if (*rel != '/')
1310 {
1311 if (!getcwd (res, PATH_MAX))
1312 return;
1313
1314 if (res [1]) /* only use if not / */
1315 res += strlen (res);
1316 }
1317
1318 while (*rel)
1319 {
1320 ssize_t len, linklen;
1321 char *beg = rel;
1322
1323 while (*rel && *rel != '/')
1324 ++rel;
1325
1326 len = rel - beg;
1327
1328 if (!len) /* skip slashes */
1329 {
1330 ++rel;
1331 continue;
1332 }
1333
1334 if (beg [0] == '.')
1335 {
1336 if (len == 1)
1337 continue; /* . - nop */
1338
1339 if (beg [1] == '.' && len == 2)
1340 {
1341 /* .. - back up one component, if possible */
1342
1343 while (res != req->ptr2)
1344 if (*--res == '/')
1345 break;
1346
1347 continue;
1348 }
1349 }
1350
1351 errno = ENAMETOOLONG;
1352 if (res + 1 + len + 1 >= tmp1)
1353 return;
1354
1355 /* copy one component */
1356 *res = '/';
1357 memcpy (res + 1, beg, len);
1358
1359 /* zero-terminate, for readlink */
1360 res [len + 1] = 0;
1361
1362 /* now check if it's a symlink */
1363 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1364
1365 if (linklen < 0)
1366 {
1367 if (errno != EINVAL)
1368 return;
1369
1370 /* it's a normal directory. hopefully */
1371 res += len + 1;
1372 }
1373 else
1374 {
1375 /* yay, it was a symlink - build new path in tmp2 */
1376 int rellen = strlen (rel);
1377
1378 errno = ENAMETOOLONG;
1379 if (linklen + 1 + rellen >= PATH_MAX)
1380 return;
1381
1382 errno = ELOOP;
1383 if (!--symlinks)
1384 return;
1385
1386 if (*tmp1 == '/')
1387 res = req->ptr2; /* symlink resolves to an absolute path */
1388
1389 /* we need to be careful, as rel might point into tmp2 already */
1390 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1391 tmp2 [linklen] = '/';
1392 memcpy (tmp2, tmp1, linklen);
1393
1394 rel = tmp2;
1395 }
1396 }
1397
1398 /* special case for the lone root path */
1399 if (res == req->ptr2)
1400 *res++ = '/';
1401
1402 req->result = res - (char *)req->ptr2;
1403
1404done:
1405 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1406}
1407
1408static signed char
1409eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1410{
1411 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1412 : a->inode < b->inode ? -1
1413 : a->inode > b->inode ? 1
1414 : 0;
1415}
1416
1417#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1418
1419#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1420#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1421
1422static void
1423eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1424{
1425 unsigned char bits [9 + sizeof (ino_t) * 8];
1426 unsigned char *bit = bits;
1427
1428 assert (CHAR_BIT == 8);
1429 assert (sizeof (eio_dirent) * 8 < 256);
1430 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1431 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1432
1433 if (size <= EIO_SORT_FAST)
1434 return;
1435
1436 /* first prepare an array of bits to test in our radix sort */
1437 /* try to take endianness into account, as well as differences in ino_t sizes */
1438 /* inode_bits must contain all inodes ORed together */
1439 /* which is used to skip bits that are 0 everywhere, which is very common */
1440 {
1441 ino_t endianness;
1442 int i, j;
1443
1444 /* we store the byte offset of byte n into byte n of "endianness" */
1445 for (i = 0; i < sizeof (ino_t); ++i)
1446 ((unsigned char *)&endianness)[i] = i;
1447
1448 *bit++ = 0;
1449
1450 for (i = 0; i < sizeof (ino_t); ++i)
1451 {
1452 /* shifting off the byte offsets out of "endianness" */
1453 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1454 endianness >>= 8;
1455
1456 for (j = 0; j < 8; ++j)
1457 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1458 *bit++ = offs + j;
1459 }
1460
1461 for (j = 0; j < 8; ++j)
1462 if (score_bits & (1 << j))
1463 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1464 }
1465
1466 /* now actually do the sorting (a variant of MSD radix sort) */
1467 {
1468 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1469 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1470 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1471 int stk_idx = 0;
1472
1473 base_stk [stk_idx] = dents;
1474 end_stk [stk_idx] = dents + size;
1475 bit_stk [stk_idx] = bit - 1;
1476
1477 do
1478 {
1479 base = base_stk [stk_idx];
1480 end = end_stk [stk_idx];
1481 bit = bit_stk [stk_idx];
1482
1483 for (;;)
1484 {
1485 unsigned char O = *bit >> 3;
1486 unsigned char M = 1 << (*bit & 7);
1487
1488 eio_dirent *a = base;
1489 eio_dirent *b = end;
1490
1491 if (b - a < EIO_SORT_CUTOFF)
1492 break;
1493
1494 /* now bit-partition the array on the bit */
1495 /* this ugly asymmetric loop seems to perform much better than typical */
1496 /* partition algos found in the literature */
1497 do
1498 if (!(((unsigned char *)a)[O] & M))
1499 ++a;
1500 else if (!(((unsigned char *)--b)[O] & M))
1501 {
1502 eio_dirent tmp = *a; *a = *b; *b = tmp;
1503 ++a;
1504 }
1505 while (b > a);
1506
1507 /* next bit, or stop, if no bits left in this path */
1508 if (!*--bit)
1509 break;
1510
1511 base_stk [stk_idx] = a;
1512 end_stk [stk_idx] = end;
1513 bit_stk [stk_idx] = bit;
1514 ++stk_idx;
1515
1516 end = a;
1517 }
1518 }
1519 while (stk_idx--);
1520 }
1521}
1522
1523static void
1524eio_dent_insertion_sort (eio_dirent *dents, int size)
1525{
1526 /* first move the smallest element to the front, to act as a sentinel */
1527 {
1528 int i;
1529 eio_dirent *min = dents;
1530
1531 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1532 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1533 if (EIO_DENT_CMP (dents [i], <, *min))
1534 min = &dents [i];
1535
1536 /* swap elements 0 and j (minimum) */
1537 {
1538 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1539 }
1540 }
1541
1542 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1543 {
1544 eio_dirent *i, *j;
1545
1546 for (i = dents + 1; i < dents + size; ++i)
1547 {
1548 eio_dirent value = *i;
1549
1550 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1551 j [1] = j [0];
1552
1553 j [1] = value;
1554 }
1555 }
1556}
1557
1558static void
1559eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1560{
1561 if (size <= 1)
1562 return; /* our insertion sort relies on size > 0 */
1563
1564 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1565 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1566 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1567
1568 /* use an insertion sort at the end, or for small arrays, */
1569 /* as insertion sort is more efficient for small partitions */
1570 eio_dent_insertion_sort (dents, size);
1571}
1572
709/* read a full directory */ 1573/* read a full directory */
1574static void
710static void scandir_ (eio_req *req, worker *self) 1575eio__scandir (eio_req *req, etp_worker *self)
711{ 1576{
712 DIR *dirp; 1577 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; 1578 EIO_STRUCT_DIRENT *entp;
719 char *name, *names; 1579 char *name, *names;
720 int memlen = 4096; 1580 int namesalloc = 4096;
721 int memofs = 0; 1581 int namesoffs = 0;
1582 int flags = req->int1;
1583 eio_dirent *dents = 0;
1584 int dentalloc = 128;
722 int res = 0; 1585 int dentoffs = 0;
1586 ino_t inode_bits = 0;
1587
1588 req->result = -1;
1589
1590 if (!(flags & EIO_READDIR_DENTS))
1591 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
723 1592
724 X_LOCK (wrklock); 1593 X_LOCK (wrklock);
1594 /* the corresponding closedir is in ETP_WORKER_CLEAR */
725 self->dirp = dirp = opendir (req->ptr1); 1595 self->dirp = dirp = opendir (req->ptr1);
726 self->dbuf = u = malloc (sizeof (*u)); 1596
727 req->flags |= EIO_FLAG_PTR2_FREE; 1597 if (req->flags & EIO_FLAG_PTR1_FREE)
1598 free (req->ptr1);
1599
1600 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1601 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
728 req->ptr2 = names = malloc (memlen); 1602 req->ptr2 = names = malloc (namesalloc);
729 X_UNLOCK (wrklock); 1603 X_UNLOCK (wrklock);
730 1604
731 if (dirp && u && names) 1605 if (dirp && names && (!flags || dents))
732 for (;;) 1606 for (;;)
733 { 1607 {
734 errno = 0; 1608 errno = 0;
735 readdir_r (dirp, &u->d, &entp); 1609 entp = readdir (dirp);
736 1610
737 if (!entp) 1611 if (!entp)
1612 {
1613 if (errno)
1614 break;
1615
1616 /* sort etc. */
1617 req->int1 = flags;
1618 req->result = dentoffs;
1619
1620 if (flags & EIO_READDIR_STAT_ORDER)
1621 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1622 else if (flags & EIO_READDIR_DIRS_FIRST)
1623 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1624 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1625 else
1626 {
1627 /* in this case, all is known, and we just put dirs first and sort them */
1628 eio_dirent *oth = dents + dentoffs;
1629 eio_dirent *dir = dents;
1630
1631 /* now partition dirs to the front, and non-dirs to the back */
1632 /* by walking from both sides and swapping if necessary */
1633 while (oth > dir)
1634 {
1635 if (dir->type == EIO_DT_DIR)
1636 ++dir;
1637 else if ((--oth)->type == EIO_DT_DIR)
1638 {
1639 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1640
1641 ++dir;
1642 }
1643 }
1644
1645 /* now sort the dirs only (dirs all have the same score) */
1646 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1647 }
1648
738 break; 1649 break;
1650 }
739 1651
1652 /* now add the entry to our list(s) */
740 name = entp->d_name; 1653 name = entp->d_name;
741 1654
1655 /* skip . and .. entries */
742 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1656 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
743 { 1657 {
744 int len = strlen (name) + 1; 1658 int len = D_NAMLEN (entp) + 1;
745 1659
746 res++; 1660 while (ecb_expect_false (namesoffs + len > namesalloc))
747
748 while (memofs + len > memlen)
749 { 1661 {
750 memlen *= 2; 1662 namesalloc *= 2;
751 X_LOCK (wrklock); 1663 X_LOCK (wrklock);
752 req->ptr2 = names = realloc (names, memlen); 1664 req->ptr2 = names = realloc (names, namesalloc);
753 X_UNLOCK (wrklock); 1665 X_UNLOCK (wrklock);
754 1666
755 if (!names) 1667 if (!names)
756 break; 1668 break;
757 } 1669 }
758 1670
759 memcpy (names + memofs, name, len); 1671 memcpy (names + namesoffs, name, len);
1672
1673 if (dents)
1674 {
1675 struct eio_dirent *ent;
1676
1677 if (ecb_expect_false (dentoffs == dentalloc))
1678 {
1679 dentalloc *= 2;
1680 X_LOCK (wrklock);
1681 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1682 X_UNLOCK (wrklock);
1683
1684 if (!dents)
1685 break;
1686 }
1687
1688 ent = dents + dentoffs;
1689
1690 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1691 ent->namelen = len - 1;
1692 ent->inode = D_INO (entp);
1693
1694 inode_bits |= ent->inode;
1695
1696 switch (D_TYPE (entp))
1697 {
1698 default:
1699 ent->type = EIO_DT_UNKNOWN;
1700 flags |= EIO_READDIR_FOUND_UNKNOWN;
1701 break;
1702
1703 #ifdef DT_FIFO
1704 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1705 #endif
1706 #ifdef DT_CHR
1707 case DT_CHR: ent->type = EIO_DT_CHR; break;
1708 #endif
1709 #ifdef DT_MPC
1710 case DT_MPC: ent->type = EIO_DT_MPC; break;
1711 #endif
1712 #ifdef DT_DIR
1713 case DT_DIR: ent->type = EIO_DT_DIR; break;
1714 #endif
1715 #ifdef DT_NAM
1716 case DT_NAM: ent->type = EIO_DT_NAM; break;
1717 #endif
1718 #ifdef DT_BLK
1719 case DT_BLK: ent->type = EIO_DT_BLK; break;
1720 #endif
1721 #ifdef DT_MPB
1722 case DT_MPB: ent->type = EIO_DT_MPB; break;
1723 #endif
1724 #ifdef DT_REG
1725 case DT_REG: ent->type = EIO_DT_REG; break;
1726 #endif
1727 #ifdef DT_NWK
1728 case DT_NWK: ent->type = EIO_DT_NWK; break;
1729 #endif
1730 #ifdef DT_CMP
1731 case DT_CMP: ent->type = EIO_DT_CMP; break;
1732 #endif
1733 #ifdef DT_LNK
1734 case DT_LNK: ent->type = EIO_DT_LNK; break;
1735 #endif
1736 #ifdef DT_SOCK
1737 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1738 #endif
1739 #ifdef DT_DOOR
1740 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1741 #endif
1742 #ifdef DT_WHT
1743 case DT_WHT: ent->type = EIO_DT_WHT; break;
1744 #endif
1745 }
1746
1747 ent->score = 7;
1748
1749 if (flags & EIO_READDIR_DIRS_FIRST)
1750 {
1751 if (ent->type == EIO_DT_UNKNOWN)
1752 {
1753 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1754 ent->score = 1;
1755 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1756 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1757 }
1758 else if (ent->type == EIO_DT_DIR)
1759 ent->score = 0;
1760 }
1761 }
1762
760 memofs += len; 1763 namesoffs += len;
1764 ++dentoffs;
1765 }
1766
1767 if (EIO_CANCELLED (req))
1768 {
1769 errno = ECANCELED;
1770 break;
761 } 1771 }
762 } 1772 }
763
764 if (errno)
765 res = -1;
766
767 req->result = res;
768} 1773}
769 1774
770/*****************************************************************************/ 1775/*****************************************************************************/
771 1776
772#define ALLOC(len) \ 1777#define ALLOC(len) \
782 req->result = -1; \ 1787 req->result = -1; \
783 break; \ 1788 break; \
784 } \ 1789 } \
785 } 1790 }
786 1791
787X_THREAD_PROC (eio_proc) 1792X_THREAD_PROC (etp_proc)
788{ 1793{
789 eio_req *req; 1794 ETP_REQ *req;
790 struct timespec ts; 1795 struct timespec ts;
791 worker *self = (worker *)thr_arg; 1796 etp_worker *self = (etp_worker *)thr_arg;
792 1797
793 /* try to distribute timeouts somewhat randomly */ 1798 /* try to distribute timeouts somewhat randomly */
794 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1799 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
795 1800
796 for (;;) 1801 for (;;)
797 { 1802 {
798 ts.tv_sec = time (0) + IDLE_TIMEOUT;
799
800 X_LOCK (reqlock); 1803 X_LOCK (reqlock);
801 1804
802 for (;;) 1805 for (;;)
803 { 1806 {
804 self->req = req = reqq_shift (&req_queue); 1807 self->req = req = reqq_shift (&req_queue);
806 if (req) 1809 if (req)
807 break; 1810 break;
808 1811
809 ++idle; 1812 ++idle;
810 1813
1814 ts.tv_sec = time (0) + idle_timeout;
811 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1815 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
812 { 1816 {
813 if (idle > max_idle) 1817 if (idle > max_idle)
814 { 1818 {
815 --idle; 1819 --idle;
820 goto quit; 1824 goto quit;
821 } 1825 }
822 1826
823 /* we are allowed to idle, so do so without any timeout */ 1827 /* we are allowed to idle, so do so without any timeout */
824 X_COND_WAIT (reqwait, reqlock); 1828 X_COND_WAIT (reqwait, reqlock);
825 ts.tv_sec = time (0) + IDLE_TIMEOUT;
826 } 1829 }
827 1830
828 --idle; 1831 --idle;
829 } 1832 }
830 1833
831 --nready; 1834 --nready;
832 1835
833 X_UNLOCK (reqlock); 1836 X_UNLOCK (reqlock);
834 1837
835 errno = 0; /* strictly unnecessary */
836
837 if (!EIO_CANCELLED (req))
838 switch (req->type) 1838 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; 1839 goto quit;
930 1840
931 default: 1841 ETP_EXECUTE (self, req);
932 req->result = -1;
933 break;
934 }
935
936 req->errorno = errno;
937 1842
938 X_LOCK (reslock); 1843 X_LOCK (reslock);
939 1844
940 ++npending; 1845 ++npending;
941 1846
942 if (!reqq_push (&res_queue, req) && want_poll_cb) 1847 if (!reqq_push (&res_queue, req) && want_poll_cb)
943 want_poll_cb (); 1848 want_poll_cb ();
944 1849
945 self->req = 0; 1850 self->req = 0;
946 worker_clear (self); 1851 etp_worker_clear (self);
947 1852
948 X_UNLOCK (reslock); 1853 X_UNLOCK (reslock);
949 } 1854 }
950 1855
951quit: 1856quit:
952 X_LOCK (wrklock); 1857 X_LOCK (wrklock);
953 worker_free (self); 1858 etp_worker_free (self);
954 X_UNLOCK (wrklock); 1859 X_UNLOCK (wrklock);
955 1860
956 return 0; 1861 return 0;
957} 1862}
958 1863
959/*****************************************************************************/ 1864/*****************************************************************************/
960 1865
961static void eio_atfork_prepare (void) 1866int 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)) 1867eio_init (void (*want_poll)(void), void (*done_poll)(void))
1018{ 1868{
1019 want_poll_cb = want_poll; 1869 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} 1870}
1035 1871
1872ecb_inline void
1036static void eio_api_destroy (eio_req *req) 1873eio_api_destroy (eio_req *req)
1037{ 1874{
1038 free (req); 1875 free (req);
1039} 1876}
1040 1877
1041#define REQ(rtype) \ 1878#define REQ(rtype) \
1043 \ 1880 \
1044 req = (eio_req *)calloc (1, sizeof *req); \ 1881 req = (eio_req *)calloc (1, sizeof *req); \
1045 if (!req) \ 1882 if (!req) \
1046 return 0; \ 1883 return 0; \
1047 \ 1884 \
1048 req->type = EIO_ ## rtype; \ 1885 req->type = rtype; \
1049 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 1886 req->pri = pri; \
1050 req->finish = cb; \ 1887 req->finish = cb; \
1888 req->data = data; \
1051 req->destroy = eio_api_destroy; 1889 req->destroy = eio_api_destroy;
1052 1890
1053#define SEND eio_submit (req); return req 1891#define SEND eio_submit (req); return req
1054 1892
1055#define PATH (void)0 1893#define PATH \
1894 req->flags |= EIO_FLAG_PTR1_FREE; \
1895 req->ptr1 = strdup (path); \
1896 if (!req->ptr1) \
1897 { \
1898 eio_api_destroy (req); \
1899 return 0; \
1900 }
1056 1901
1902static void
1903eio_execute (etp_worker *self, eio_req *req)
1904{
1905 if (ecb_expect_false (EIO_CANCELLED (req)))
1906 {
1907 req->result = -1;
1908 req->errorno = ECANCELED;
1909 return;
1910 }
1911
1912 switch (req->type)
1913 {
1914 case EIO_READ: ALLOC (req->size);
1915 req->result = req->offs >= 0
1916 ? pread (req->int1, req->ptr2, req->size, req->offs)
1917 : read (req->int1, req->ptr2, req->size); break;
1918 case EIO_WRITE: req->result = req->offs >= 0
1919 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1920 : write (req->int1, req->ptr2, req->size); break;
1921
1922 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1923 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1924
1925 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1926 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1927 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1928 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1929 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1930 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1931
1932 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1933 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1934 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1935 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1936
1937 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1938 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1939 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1940 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1941 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1942 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1943
1944 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1945 case EIO_CLOSE: req->result = close (req->int1); break;
1946 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1947 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1948 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1949 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1950 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1951 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1952 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1953 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1954
1955 case EIO_REALPATH: eio__realpath (req, self); break;
1956
1957 case EIO_READLINK: ALLOC (PATH_MAX);
1958 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1959
1960 case EIO_SYNC: req->result = 0; sync (); break;
1961 case EIO_FSYNC: req->result = fsync (req->int1); break;
1962 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1963 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1964 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1965 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1966 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1967 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1968
1969 case EIO_READDIR: eio__scandir (req, self); break;
1970
1971 case EIO_BUSY:
1972#ifdef _WIN32
1973 Sleep (req->nv1 * 1e3);
1974#else
1975 {
1976 struct timeval tv;
1977
1978 tv.tv_sec = req->nv1;
1979 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1980
1981 req->result = select (0, 0, 0, 0, &tv);
1982 }
1983#endif
1984 break;
1985
1986 case EIO_UTIME:
1987 case EIO_FUTIME:
1988 {
1989 struct timeval tv[2];
1990 struct timeval *times;
1991
1992 if (req->nv1 != -1. || req->nv2 != -1.)
1993 {
1994 tv[0].tv_sec = req->nv1;
1995 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1996 tv[1].tv_sec = req->nv2;
1997 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1998
1999 times = tv;
2000 }
2001 else
2002 times = 0;
2003
2004 req->result = req->type == EIO_FUTIME
2005 ? futimes (req->int1, times)
2006 : utimes (req->ptr1, times);
2007 }
2008 break;
2009
2010 case EIO_GROUP:
2011 abort (); /* handled in eio_request */
2012
2013 case EIO_NOP:
2014 req->result = 0;
2015 break;
2016
2017 case EIO_CUSTOM:
2018 req->feed (req);
2019 break;
2020
2021 default:
2022 errno = ENOSYS;
2023 req->result = -1;
2024 break;
2025 }
2026
2027 req->errorno = errno;
2028}
2029
2030#ifndef EIO_NO_WRAPPERS
2031
2032eio_req *eio_nop (int pri, eio_cb cb, void *data)
2033{
2034 REQ (EIO_NOP); SEND;
2035}
2036
2037eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2038{
2039 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2040}
2041
1057eio_req *eio_fsync (int fd, eio_cb cb) 2042eio_req *eio_sync (int pri, eio_cb cb, void *data)
1058{ 2043{
2044 REQ (EIO_SYNC); SEND;
2045}
2046
2047eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2048{
1059 REQ (FSYNC); req->int1 = fd; SEND; 2049 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1060} 2050}
1061 2051
2052eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2053{
2054 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2055}
2056
2057eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2058{
2059 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2060}
2061
2062eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2063{
2064 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2065}
2066
2067eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2068{
2069 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2070}
2071
2072eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2073{
2074 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2075}
2076
1062eio_req *eio_fdatasync (int fd, eio_cb cb) 2077eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1063{ 2078{
1064 REQ (FDATASYNC); req->int1 = fd; SEND; 2079 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1065} 2080}
1066 2081
2082eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2083{
2084 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2085}
2086
1067eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb) 2087eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1068{ 2088{
1069 REQ (READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2089 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1070} 2090}
1071 2091
1072eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2092eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1073{ 2093{
1074 REQ (READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2094 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1075} 2095}
1076 2096
1077eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2097eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1078{ 2098{
1079 REQ (WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2099 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1080} 2100}
1081 2101
1082eio_req *eio_fstat (int fd, eio_cb cb) 2102eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1083{ 2103{
1084 REQ (FSTAT); req->int1 = fd; SEND; 2104 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1085} 2105}
1086 2106
2107eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2108{
2109 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2110}
2111
1087eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb) 2112eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1088{ 2113{
1089 REQ (FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2114 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1090} 2115}
1091 2116
1092eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb) 2117eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1093{ 2118{
1094 REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 2119 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1095} 2120}
1096 2121
1097eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb) 2122eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1098{ 2123{
1099 REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2124 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1100} 2125}
1101 2126
1102eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb) 2127eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1103{ 2128{
1104 REQ (FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2129 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1105} 2130}
1106 2131
1107eio_req *eio_dup2 (int fd, int fd2, eio_cb cb) 2132eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1108{ 2133{
1109 REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND; 2134 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1110} 2135}
1111 2136
1112eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb) 2137eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1113{ 2138{
1114 REQ (SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 2139 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1115} 2140}
1116 2141
1117eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb) 2142eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1118{ 2143{
1119 REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 2144 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1120} 2145}
1121 2146
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) 2147eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1138{ 2148{
1139 REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 2149 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1140} 2150}
1141 2151
1142eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb) 2152eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1143{ 2153{
1144 REQ (TRUNCATE); PATH; req->offs = offset; SEND; 2154 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1145} 2155}
1146 2156
1147eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 2157eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1148{ 2158{
1149 REQ (CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2159 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1150} 2160}
1151 2161
1152eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb) 2162eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1153{ 2163{
1154 REQ (CHMOD); PATH; req->int2 = (long)mode; SEND; 2164 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1155} 2165}
1156 2166
1157eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb) 2167eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1158{ 2168{
1159 REQ (MKDIR); PATH; req->int2 = (long)mode; SEND; 2169 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1160} 2170}
1161 2171
1162eio_req *eio_unlink (const char *path, eio_cb cb) 2172static eio_req *
2173eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1163{ 2174{
1164 REQ (UNLINK); PATH; SEND;
1165}
1166
1167eio_req *eio_rmdir (const char *path, eio_cb cb)
1168{
1169 REQ (RMDIR); PATH; SEND; 2175 REQ (type); PATH; SEND;
1170} 2176}
1171 2177
2178eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2179{
2180 return eio__1path (EIO_READLINK, path, pri, cb, data);
2181}
2182
2183eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2184{
2185 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2186}
2187
2188eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2189{
2190 return eio__1path (EIO_STAT, path, pri, cb, data);
2191}
2192
2193eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2194{
2195 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2196}
2197
2198eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2199{
2200 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2201}
2202
2203eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2204{
2205 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2206}
2207
1172eio_req *eio_readdir (const char *path, eio_cb cb) 2208eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1173{ 2209{
1174 REQ (READDIR); PATH; SEND; 2210 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1175} 2211}
1176 2212
2213eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2214{
2215 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2216}
2217
1177eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb) 2218eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1178{ 2219{
1179 REQ (MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2220 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1180} 2221}
1181 2222
1182eio_req *eio_busy (double delay, eio_cb cb) 2223static eio_req *
2224eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1183{ 2225{
1184 REQ (BUSY); req->nv1 = delay; SEND; 2226 REQ (type); PATH;
1185}
1186 2227
2228 req->flags |= EIO_FLAG_PTR2_FREE;
2229 req->ptr2 = strdup (new_path);
2230 if (!req->ptr2)
2231 {
2232 eio_api_destroy (req);
2233 return 0;
2234 }
2235
2236 SEND;
2237}
2238
2239eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2240{
2241 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2242}
2243
2244eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2245{
2246 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2247}
2248
2249eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2250{
2251 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2252}
2253
2254eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2255{
2256 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2257}
2258
2259#endif
2260
1187eio_req *eio_nop (eio_cb cb) 2261eio_req *eio_grp (eio_cb cb, void *data)
1188{ 2262{
2263 const int pri = EIO_PRI_MAX;
2264
1189 REQ (NOP); SEND; 2265 REQ (EIO_GROUP); SEND;
1190} 2266}
1191 2267
1192#undef REQ 2268#undef REQ
1193#undef PATH 2269#undef PATH
1194#undef SEND 2270#undef SEND
1195 2271
1196#if 0 2272/*****************************************************************************/
2273/* grp functions */
1197 2274
1198void 2275void
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) 2276eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1572{ 2277{
1573 grp->int2 = limit; 2278 grp->int2 = limit;
1574 grp->feed = feed; 2279 grp->feed = feed;
1575 2280
1576 grp_try_feed (grp); 2281 grp_try_feed (grp);
1577} 2282}
1578 2283
2284void
1579void eio_grp_limit (eio_req *grp, int limit) 2285eio_grp_limit (eio_req *grp, int limit)
1580{ 2286{
1581 grp->int2 = limit; 2287 grp->int2 = limit;
1582 2288
1583 grp_try_feed (grp); 2289 grp_try_feed (grp);
1584} 2290}
1585 2291
2292void
1586void eio_grp_add (eio_req *grp, eio_req *req) 2293eio_grp_add (eio_req *grp, eio_req *req)
1587{ 2294{
1588 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2295 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2296
2297 grp->flags |= EIO_FLAG_GROUPADD;
1589 2298
1590 ++grp->size; 2299 ++grp->size;
1591 req->grp = grp; 2300 req->grp = grp;
1592 2301
1593 req->grp_prev = 0; 2302 req->grp_prev = 0;
1597 grp->grp_first->grp_prev = req; 2306 grp->grp_first->grp_prev = req;
1598 2307
1599 grp->grp_first = req; 2308 grp->grp_first = req;
1600} 2309}
1601 2310
2311/*****************************************************************************/
2312/* misc garbage */
1602 2313
2314ssize_t
2315eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2316{
2317 etp_worker wrk;
2318 ssize_t ret;
2319
2320 wrk.dbuf = 0;
2321
2322 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2323
2324 if (wrk.dbuf)
2325 free (wrk.dbuf);
2326
2327 return ret;
2328}
2329

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines