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 1.77 by root, Tue Jul 5 14:02:15 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 form stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H
67# include <stdint.h>
68#endif
69
15#ifndef EIO_FINISH 70#ifndef EIO_FINISH
16# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 71# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17#endif 72#endif
18 73
19#ifndef EIO_DESTROY 74#ifndef EIO_DESTROY
… …
25#endif 80#endif
26 81
27#ifdef _WIN32 82#ifdef _WIN32
28 83
29 /*doh*/ 84 /*doh*/
30
31#else 85#else
32 86
33# include "config.h"
34# include <sys/time.h> 87# include <sys/time.h>
35# include <sys/select.h> 88# include <sys/select.h>
36# include <unistd.h> 89# include <unistd.h>
37# include <utime.h> 90# include <utime.h>
38# include <signal.h> 91# include <signal.h>
39# include <dirent.h> 92# include <dirent.h>
40 93
94#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
95# include <sys/mman.h>
96#endif
97
98/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
99# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
100# define _DIRENT_HAVE_D_TYPE /* sigh */
101# define D_INO(de) (de)->d_fileno
102# define D_NAMLEN(de) (de)->d_namlen
103# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
104# define D_INO(de) (de)->d_ino
105# endif
106
107#ifdef _D_EXACT_NAMLEN
108# undef D_NAMLEN
109# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
110#endif
111
112# ifdef _DIRENT_HAVE_D_TYPE
113# define D_TYPE(de) (de)->d_type
114# endif
115
41# ifndef EIO_STRUCT_DIRENT 116# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 117# define EIO_STRUCT_DIRENT struct dirent
43# endif 118# endif
44 119
45#endif 120#endif
46 121
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif
50
51#if HAVE_SENDFILE 122#if HAVE_SENDFILE
52# if __linux 123# if __linux
53# include <sys/sendfile.h> 124# include <sys/sendfile.h>
54# elif __freebsd 125# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 126# include <sys/socket.h>
56# include <sys/uio.h> 127# include <sys/uio.h>
57# elif __hpux 128# elif __hpux
58# include <sys/socket.h> 129# include <sys/socket.h>
59# elif __solaris /* not yet */ 130# elif __solaris
60# include <sys/sendfile.h> 131# include <sys/sendfile.h>
61# else 132# else
62# error sendfile support requested but not available 133# error sendfile support requested but not available
63# endif 134# endif
64#endif 135#endif
65 136
66/* number of seconds after which an idle threads exit */ 137#ifndef D_TYPE
67#define IDLE_TIMEOUT 10 138# define D_TYPE(de) 0
139#endif
140#ifndef D_INO
141# define D_INO(de) 0
142#endif
143#ifndef D_NAMLEN
144# define D_NAMLEN(de) strlen ((de)->d_name)
145#endif
68 146
69/* used for struct dirent, AIX doesn't provide it */ 147/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 148#ifndef NAME_MAX
71# define NAME_MAX 4096 149# define NAME_MAX 4096
72#endif 150#endif
73 151
152/* used for readlink etc. */
153#ifndef PATH_MAX
154# define PATH_MAX 4096
155#endif
156
74/* buffer size for various temporary buffers */ 157/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 158#define EIO_BUFSIZE 65536
76 159
77#define dBUF \ 160#define dBUF \
78 char *eio_buf; \ 161 char *eio_buf; \
79 X_LOCK (wrklock); \ 162 ETP_WORKER_LOCK (self); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 163 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 164 ETP_WORKER_UNLOCK (self); \
165 errno = ENOMEM; \
82 if (!eio_buf) \ 166 if (!eio_buf) \
83 return -1; 167 return -1;
84 168
85#define EIO_TICKS ((1000000 + 1023) >> 10) 169#define EIO_TICKS ((1000000 + 1023) >> 10)
86 170
171#define ETP_PRI_MIN EIO_PRI_MIN
172#define ETP_PRI_MAX EIO_PRI_MAX
173
174struct etp_worker;
175
176#define ETP_REQ eio_req
177#define ETP_DESTROY(req) eio_destroy (req)
178static int eio_finish (eio_req *req);
179#define ETP_FINISH(req) eio_finish (req)
180static void eio_execute (struct etp_worker *self, eio_req *req);
181#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
182
183#define ETP_WORKER_CLEAR(req) \
184 if (wrk->dbuf) \
185 { \
186 free (wrk->dbuf); \
187 wrk->dbuf = 0; \
188 } \
189 \
190 if (wrk->dirp) \
191 { \
192 closedir (wrk->dirp); \
193 wrk->dirp = 0; \
194 }
195
196#define ETP_WORKER_COMMON \
197 void *dbuf; \
198 DIR *dirp;
199
200/*****************************************************************************/
201
202#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
203
204/* calculate time difference in ~1/EIO_TICKS of a second */
205ecb_inline int
206tvdiff (struct timeval *tv1, struct timeval *tv2)
207{
208 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
209 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
210}
211
212static unsigned int started, idle, wanted = 4;
213
87static void (*want_poll_cb) (void); 214static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 215static void (*done_poll_cb) (void);
89 216
90static unsigned int max_poll_time = 0; 217static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 218static unsigned int max_poll_reqs; /* reslock */
92 219
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 220static volatile unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 221static volatile unsigned int nready; /* reqlock */
95{ 222static volatile unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 223static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 224static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 225
100static unsigned int started, idle, wanted; 226static xmutex_t wrklock;
227static xmutex_t reslock;
228static xmutex_t reqlock;
229static xcond_t reqwait;
230
231#if !HAVE_PREADWRITE
232/*
233 * make our pread/pwrite emulation safe against themselves, but not against
234 * normal read/write by using a mutex. slows down execution a lot,
235 * but that's your problem, not mine.
236 */
237static xmutex_t preadwritelock = X_MUTEX_INIT;
238#endif
239
240typedef struct etp_worker
241{
242 /* locked by wrklock */
243 struct etp_worker *prev, *next;
244
245 xthread_t tid;
246
247 /* locked by reslock, reqlock or wrklock */
248 ETP_REQ *req; /* currently processed request */
249
250 ETP_WORKER_COMMON
251} etp_worker;
252
253static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
254
255#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
256#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
101 257
102/* worker threads management */ 258/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 259
105typedef struct worker { 260static 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) 261etp_worker_clear (etp_worker *wrk)
120{ 262{
121 if (wrk->dirp) 263 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} 264}
133 265
134static void worker_free (worker *wrk) 266static void ecb_cold
267etp_worker_free (etp_worker *wrk)
135{ 268{
136 wrk->next->prev = wrk->prev; 269 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 270 wrk->prev->next = wrk->next;
138 271
139 free (wrk); 272 free (wrk);
140} 273}
141 274
142static volatile unsigned int nreqs, nready, npending; 275static unsigned int
143static volatile unsigned int max_idle = 4; 276etp_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{ 277{
278 int retval;
279 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
280 retval = nreqs;
281 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 282 return retval;
152} 283}
153 284
154unsigned int eio_nready (void) 285static unsigned int
286etp_nready (void)
155{ 287{
156 unsigned int retval; 288 unsigned int retval;
157 289
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 291 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 293
162 return retval; 294 return retval;
163} 295}
164 296
165unsigned int eio_npending (void) 297static unsigned int
298etp_npending (void)
166{ 299{
167 unsigned int retval; 300 unsigned int retval;
168 301
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 302 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 303 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 304 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 305
173 return retval; 306 return retval;
174} 307}
175 308
176unsigned int eio_nthreads (void) 309static unsigned int
310etp_nthreads (void)
177{ 311{
178 unsigned int retval; 312 unsigned int retval;
179 313
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 314 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 315 retval = started;
… …
188 * a somewhat faster data structure might be nice, but 322 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 323 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 324 * per shift, the most expensive operation.
191 */ 325 */
192typedef struct { 326typedef struct {
193 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 327 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194 int size; 328 int size;
195} reqq; 329} etp_reqq;
196 330
197static reqq req_queue; 331static etp_reqq req_queue;
198static reqq res_queue; 332static etp_reqq res_queue;
199 333
200static int reqq_push (reqq *q, eio_req *req) 334static int ecb_noinline
335reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 336{
202 int pri = req->pri; 337 int pri = req->pri;
203 req->next = 0; 338 req->next = 0;
204 339
205 if (q->qe[pri]) 340 if (q->qe[pri])
… …
211 q->qe[pri] = q->qs[pri] = req; 346 q->qe[pri] = q->qs[pri] = req;
212 347
213 return q->size++; 348 return q->size++;
214} 349}
215 350
216static eio_req *reqq_shift (reqq *q) 351static ETP_REQ * ecb_noinline
352reqq_shift (etp_reqq *q)
217{ 353{
218 int pri; 354 int pri;
219 355
220 if (!q->size) 356 if (!q->size)
221 return 0; 357 return 0;
222 358
223 --q->size; 359 --q->size;
224 360
225 for (pri = EIO_NUM_PRI; pri--; ) 361 for (pri = ETP_NUM_PRI; pri--; )
226 { 362 {
227 eio_req *req = q->qs[pri]; 363 eio_req *req = q->qs[pri];
228 364
229 if (req) 365 if (req)
230 { 366 {
… …
236 } 372 }
237 373
238 abort (); 374 abort ();
239} 375}
240 376
241static void grp_try_feed (eio_req *grp) 377static void ecb_cold
378etp_thread_init (void)
242{ 379{
243 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 380 X_MUTEX_CREATE (wrklock);
381 X_MUTEX_CREATE (reslock);
382 X_MUTEX_CREATE (reqlock);
383 X_COND_CREATE (reqwait);
384}
385
386static void ecb_cold
387etp_atfork_prepare (void)
388{
389 X_LOCK (wrklock);
390 X_LOCK (reqlock);
391 X_LOCK (reslock);
392#if !HAVE_PREADWRITE
393 X_LOCK (preadwritelock);
394#endif
395}
396
397static void ecb_cold
398etp_atfork_parent (void)
399{
400#if !HAVE_PREADWRITE
401 X_UNLOCK (preadwritelock);
402#endif
403 X_UNLOCK (reslock);
404 X_UNLOCK (reqlock);
405 X_UNLOCK (wrklock);
406}
407
408static void ecb_cold
409etp_atfork_child (void)
410{
411 ETP_REQ *prv;
412
413 while ((prv = reqq_shift (&req_queue)))
414 ETP_DESTROY (prv);
415
416 while ((prv = reqq_shift (&res_queue)))
417 ETP_DESTROY (prv);
418
419 while (wrk_first.next != &wrk_first)
244 { 420 {
245 int old_len = grp->size; 421 etp_worker *wrk = wrk_first.next;
246 422
247 EIO_FEED (grp); 423 if (wrk->req)
424 ETP_DESTROY (wrk->req);
248 425
249 /* stop if no progress has been made */ 426 etp_worker_clear (wrk);
250 if (old_len == grp->size) 427 etp_worker_free (wrk);
251 {
252 grp->feed = 0;
253 break;
254 }
255 } 428 }
256}
257 429
258static int eio_finish (eio_req *req); 430 started = 0;
431 idle = 0;
432 nreqs = 0;
433 nready = 0;
434 npending = 0;
259 435
260static int grp_dec (eio_req *grp) 436 etp_thread_init ();
261{ 437}
262 --grp->size;
263 438
264 /* call feeder, if applicable */ 439static void ecb_cold
265 grp_try_feed (grp); 440etp_once_init (void)
441{
442 etp_thread_init ();
443 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
444}
266 445
267 /* finish, if done */ 446static int ecb_cold
268 if (!grp->size && grp->int1) 447etp_init (void (*want_poll)(void), void (*done_poll)(void))
269 return eio_finish (grp); 448{
270 else 449 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
450
451 pthread_once (&doinit, etp_once_init);
452
453 want_poll_cb = want_poll;
454 done_poll_cb = done_poll;
455
271 return 0; 456 return 0;
272} 457}
273 458
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); 459X_THREAD_PROC (etp_proc);
325 460
326static void start_thread (void) 461static void ecb_cold
462etp_start_thread (void)
327{ 463{
328 worker *wrk = calloc (1, sizeof (worker)); 464 etp_worker *wrk = calloc (1, sizeof (etp_worker));
329 465
330 /*TODO*/ 466 /*TODO*/
331 assert (("unable to allocate worker thread data", !wrk)); 467 assert (("unable to allocate worker thread data", wrk));
332 468
333 X_LOCK (wrklock); 469 X_LOCK (wrklock);
334 470
335 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 471 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
336 { 472 {
337 wrk->prev = &wrk_first; 473 wrk->prev = &wrk_first;
338 wrk->next = wrk_first.next; 474 wrk->next = wrk_first.next;
339 wrk_first.next->prev = wrk; 475 wrk_first.next->prev = wrk;
340 wrk_first.next = wrk; 476 wrk_first.next = wrk;
… …
344 free (wrk); 480 free (wrk);
345 481
346 X_UNLOCK (wrklock); 482 X_UNLOCK (wrklock);
347} 483}
348 484
485static void
349static void maybe_start_thread (void) 486etp_maybe_start_thread (void)
350{ 487{
351 if (eio_nthreads () >= wanted) 488 if (ecb_expect_true (etp_nthreads () >= wanted))
352 return; 489 return;
353 490
354 /* todo: maybe use idle here, but might be less exact */ 491 /* todo: maybe use idle here, but might be less exact */
355 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 492 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
356 return; 493 return;
357 494
358 start_thread (); 495 etp_start_thread ();
359} 496}
360 497
361void eio_submit (eio_req *req) 498static void ecb_cold
362{ 499etp_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{ 500{
376 eio_req *req = calloc (1, sizeof (eio_req)); 501 eio_req *req = calloc (1, sizeof (eio_req));
377 502
378 req->type = EIO_QUIT; 503 req->type = -1;
379 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 504 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
380 505
381 X_LOCK (reqlock); 506 X_LOCK (reqlock);
382 reqq_push (&req_queue, req); 507 reqq_push (&req_queue, req);
383 X_COND_SIGNAL (reqwait); 508 X_COND_SIGNAL (reqwait);
384 X_UNLOCK (reqlock); 509 X_UNLOCK (reqlock);
… …
386 X_LOCK (wrklock); 511 X_LOCK (wrklock);
387 --started; 512 --started;
388 X_UNLOCK (wrklock); 513 X_UNLOCK (wrklock);
389} 514}
390 515
391void eio_set_max_poll_time (double nseconds) 516static 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) 517etp_poll (void)
428{ 518{
429 int maxreqs = max_poll_reqs; 519 unsigned int maxreqs;
520 unsigned int maxtime;
430 struct timeval tv_start, tv_now; 521 struct timeval tv_start, tv_now;
431 eio_req *req;
432 522
523 X_LOCK (reslock);
524 maxreqs = max_poll_reqs;
525 maxtime = max_poll_time;
526 X_UNLOCK (reslock);
527
433 if (max_poll_time) 528 if (maxtime)
434 gettimeofday (&tv_start, 0); 529 gettimeofday (&tv_start, 0);
435 530
436 for (;;) 531 for (;;)
437 { 532 {
533 ETP_REQ *req;
534
438 maybe_start_thread (); 535 etp_maybe_start_thread ();
439 536
440 X_LOCK (reslock); 537 X_LOCK (reslock);
441 req = reqq_shift (&res_queue); 538 req = reqq_shift (&res_queue);
442 539
443 if (req) 540 if (req)
… …
451 X_UNLOCK (reslock); 548 X_UNLOCK (reslock);
452 549
453 if (!req) 550 if (!req)
454 return 0; 551 return 0;
455 552
553 X_LOCK (reqlock);
456 --nreqs; 554 --nreqs;
555 X_UNLOCK (reqlock);
457 556
458 if (req->type == EIO_GROUP && req->size) 557 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
459 { 558 {
460 req->int1 = 1; /* mark request as delayed */ 559 req->int1 = 1; /* mark request as delayed */
461 continue; 560 continue;
462 } 561 }
463 else 562 else
464 { 563 {
465 int res = eio_finish (req); 564 int res = ETP_FINISH (req);
466 if (res) 565 if (ecb_expect_false (res))
467 return res; 566 return res;
468 } 567 }
469 568
470 if (maxreqs && !--maxreqs) 569 if (ecb_expect_false (maxreqs && !--maxreqs))
471 break; 570 break;
472 571
473 if (max_poll_time) 572 if (maxtime)
474 { 573 {
475 gettimeofday (&tv_now, 0); 574 gettimeofday (&tv_now, 0);
476 575
477 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 576 if (tvdiff (&tv_start, &tv_now) >= maxtime)
478 break; 577 break;
479 } 578 }
480 } 579 }
481 580
482 errno = EAGAIN; 581 errno = EAGAIN;
483 return -1; 582 return -1;
484} 583}
485 584
585static void
586etp_cancel (ETP_REQ *req)
587{
588 X_LOCK (wrklock);
589 req->flags |= EIO_FLAG_CANCELLED;
590 X_UNLOCK (wrklock);
591
592 eio_grp_cancel (req);
593}
594
595static void
596etp_submit (ETP_REQ *req)
597{
598 req->pri -= ETP_PRI_MIN;
599
600 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
601 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
602
603 if (ecb_expect_false (req->type == EIO_GROUP))
604 {
605 /* I hope this is worth it :/ */
606 X_LOCK (reqlock);
607 ++nreqs;
608 X_UNLOCK (reqlock);
609
610 X_LOCK (reslock);
611
612 ++npending;
613
614 if (!reqq_push (&res_queue, req) && want_poll_cb)
615 want_poll_cb ();
616
617 X_UNLOCK (reslock);
618 }
619 else
620 {
621 X_LOCK (reqlock);
622 ++nreqs;
623 ++nready;
624 reqq_push (&req_queue, req);
625 X_COND_SIGNAL (reqwait);
626 X_UNLOCK (reqlock);
627
628 etp_maybe_start_thread ();
629 }
630}
631
632static void ecb_cold
633etp_set_max_poll_time (double nseconds)
634{
635 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
636 max_poll_time = nseconds * EIO_TICKS;
637 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
638}
639
640static void ecb_cold
641etp_set_max_poll_reqs (unsigned int maxreqs)
642{
643 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
644 max_poll_reqs = maxreqs;
645 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
646}
647
648static void ecb_cold
649etp_set_max_idle (unsigned int nthreads)
650{
651 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
652 max_idle = nthreads;
653 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
654}
655
656static void ecb_cold
657etp_set_idle_timeout (unsigned int seconds)
658{
659 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
660 idle_timeout = seconds;
661 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
662}
663
664static void ecb_cold
665etp_set_min_parallel (unsigned int nthreads)
666{
667 if (wanted < nthreads)
668 wanted = nthreads;
669}
670
671static void ecb_cold
672etp_set_max_parallel (unsigned int nthreads)
673{
674 if (wanted > nthreads)
675 wanted = nthreads;
676
677 while (started > wanted)
678 etp_end_thread ();
679}
680
681/*****************************************************************************/
682
683static void
684grp_try_feed (eio_req *grp)
685{
686 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
687 {
688 grp->flags &= ~EIO_FLAG_GROUPADD;
689
690 EIO_FEED (grp);
691
692 /* stop if no progress has been made */
693 if (!(grp->flags & EIO_FLAG_GROUPADD))
694 {
695 grp->feed = 0;
696 break;
697 }
698 }
699}
700
701static int
702grp_dec (eio_req *grp)
703{
704 --grp->size;
705
706 /* call feeder, if applicable */
707 grp_try_feed (grp);
708
709 /* finish, if done */
710 if (!grp->size && grp->int1)
711 return eio_finish (grp);
712 else
713 return 0;
714}
715
716void
717eio_destroy (eio_req *req)
718{
719 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
720 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
721
722 EIO_DESTROY (req);
723}
724
725static int
726eio_finish (eio_req *req)
727{
728 int res = EIO_FINISH (req);
729
730 if (req->grp)
731 {
732 int res2;
733 eio_req *grp = req->grp;
734
735 /* unlink request */
736 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
737 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
738
739 if (grp->grp_first == req)
740 grp->grp_first = req->grp_next;
741
742 res2 = grp_dec (grp);
743
744 if (!res && res2)
745 res = res2;
746 }
747
748 eio_destroy (req);
749
750 return res;
751}
752
753void
754eio_grp_cancel (eio_req *grp)
755{
756 for (grp = grp->grp_first; grp; grp = grp->grp_next)
757 eio_cancel (grp);
758}
759
760void
761eio_cancel (eio_req *req)
762{
763 etp_cancel (req);
764}
765
766void
767eio_submit (eio_req *req)
768{
769 etp_submit (req);
770}
771
772unsigned int
773eio_nreqs (void)
774{
775 return etp_nreqs ();
776}
777
778unsigned int
779eio_nready (void)
780{
781 return etp_nready ();
782}
783
784unsigned int
785eio_npending (void)
786{
787 return etp_npending ();
788}
789
790unsigned int ecb_cold
791eio_nthreads (void)
792{
793 return etp_nthreads ();
794}
795
796void ecb_cold
797eio_set_max_poll_time (double nseconds)
798{
799 etp_set_max_poll_time (nseconds);
800}
801
802void ecb_cold
803eio_set_max_poll_reqs (unsigned int maxreqs)
804{
805 etp_set_max_poll_reqs (maxreqs);
806}
807
808void ecb_cold
809eio_set_max_idle (unsigned int nthreads)
810{
811 etp_set_max_idle (nthreads);
812}
813
814void ecb_cold
815eio_set_idle_timeout (unsigned int seconds)
816{
817 etp_set_idle_timeout (seconds);
818}
819
820void ecb_cold
821eio_set_min_parallel (unsigned int nthreads)
822{
823 etp_set_min_parallel (nthreads);
824}
825
826void ecb_cold
827eio_set_max_parallel (unsigned int nthreads)
828{
829 etp_set_max_parallel (nthreads);
830}
831
832int eio_poll (void)
833{
834 return etp_poll ();
835}
836
486/*****************************************************************************/ 837/*****************************************************************************/
487/* work around various missing functions */ 838/* work around various missing functions */
488 839
489#if !HAVE_PREADWRITE 840#if !HAVE_PREADWRITE
841# undef pread
842# undef pwrite
490# define pread aio_pread 843# define pread eio__pread
491# define pwrite aio_pwrite 844# define pwrite eio__pwrite
492 845
493/* 846static ssize_t
494 * make our pread/pwrite safe against themselves, but not against 847eio__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{ 848{
502 ssize_t res; 849 ssize_t res;
503 off_t ooffset; 850 off_t ooffset;
504 851
505 X_LOCK (preadwritelock); 852 X_LOCK (preadwritelock);
… …
510 X_UNLOCK (preadwritelock); 857 X_UNLOCK (preadwritelock);
511 858
512 return res; 859 return res;
513} 860}
514 861
515static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 862static ssize_t
863eio__pwrite (int fd, void *buf, size_t count, off_t offset)
516{ 864{
517 ssize_t res; 865 ssize_t res;
518 off_t ooffset; 866 off_t ooffset;
519 867
520 X_LOCK (preadwritelock); 868 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR); 869 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET); 870 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count); 871 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET); 872 lseek (fd, ooffset, SEEK_SET);
525 X_UNLOCK (preadwritelock); 873 X_UNLOCK (preadwritelock);
526 874
527 return res; 875 return res;
528} 876}
529#endif 877#endif
530 878
531#ifndef HAVE_FUTIMES 879#ifndef HAVE_UTIMES
532 880
881# undef utimes
533# define utimes(path,times) aio_utimes (path, times) 882# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) aio_futimes (fd, times)
535 883
884static int
536static int aio_utimes (const char *filename, const struct timeval times[2]) 885eio__utimes (const char *filename, const struct timeval times[2])
537{ 886{
538 if (times) 887 if (times)
539 { 888 {
540 struct utimbuf buf; 889 struct utimbuf buf;
541 890
… …
546 } 895 }
547 else 896 else
548 return utime (filename, 0); 897 return utime (filename, 0);
549} 898}
550 899
900#endif
901
902#ifndef HAVE_FUTIMES
903
904# undef futimes
905# define futimes(fd,times) eio__futimes (fd, times)
906
907static int
551static int aio_futimes (int fd, const struct timeval tv[2]) 908eio__futimes (int fd, const struct timeval tv[2])
552{ 909{
553 errno = ENOSYS; 910 errno = ENOSYS;
554 return -1; 911 return -1;
555} 912}
556 913
557#endif 914#endif
558 915
559#if !HAVE_FDATASYNC 916#if !HAVE_FDATASYNC
917# undef fdatasync
560# define fdatasync fsync 918# define fdatasync(fd) fsync (fd)
561#endif 919#endif
920
921/* sync_file_range always needs emulation */
922static int
923eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
924{
925#if HAVE_SYNC_FILE_RANGE
926 int res;
927
928 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
929 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
930 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
931 {
932 flags = 0
933 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
934 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
935 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
936 }
937
938 res = sync_file_range (fd, offset, nbytes, flags);
939
940 if (!res || errno != ENOSYS)
941 return res;
942#endif
943
944 /* even though we could play tricks with the flags, it's better to always
945 * call fdatasync, as that matches the expectation of its users best */
946 return fdatasync (fd);
947}
562 948
563#if !HAVE_READAHEAD 949#if !HAVE_READAHEAD
950# undef readahead
564# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 951# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565 952
953static ssize_t
566static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 954eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
567{ 955{
568 size_t todo = count; 956 size_t todo = count;
569 dBUF; 957 dBUF;
570 958
571 while (todo > 0) 959 while (todo > 0)
… …
581 return count; 969 return count;
582} 970}
583 971
584#endif 972#endif
585 973
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 */ 974/* sendfile always needs emulation */
975static ssize_t
617static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 976eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
618{ 977{
978 ssize_t written = 0;
619 ssize_t res; 979 ssize_t res;
620 980
621 if (!count) 981 if (!count)
622 return 0; 982 return 0;
623 983
984 for (;;)
985 {
624#if HAVE_SENDFILE 986#if HAVE_SENDFILE
625# if __linux 987# if __linux
988 off_t soffset = offset;
626 res = sendfile (ofd, ifd, &offset, count); 989 res = sendfile (ofd, ifd, &soffset, count);
627 990
628# elif __freebsd 991# elif __FreeBSD__
629 /* 992 /*
630 * Of course, the freebsd sendfile is a dire hack with no thoughts 993 * Of course, the freebsd sendfile is a dire hack with no thoughts
631 * wasted on making it similar to other I/O functions. 994 * wasted on making it similar to other I/O functions.
632 */ 995 */
633 {
634 off_t sbytes; 996 off_t sbytes;
635 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 997 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636 998
637 if (res < 0 && sbytes) 999 #if 0 /* according to the manpage, this is correct, but broken behaviour */
638 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1000 /* freebsd' sendfile will return 0 on success */
1001 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1002 /* not on e.g. EIO or EPIPE - sounds broken */
1003 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
639 res = sbytes; 1004 res = sbytes;
640 } 1005 #endif
1006
1007 /* according to source inspection, this is correct, and useful behaviour */
1008 if (sbytes)
1009 res = sbytes;
1010
1011# elif defined (__APPLE__) && 0 /* broken, as everything on os x */
1012 off_t sbytes = count;
1013 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1014
1015 /* according to the manpage, sbytes is always valid */
1016 if (sbytes)
1017 res = sbytes;
641 1018
642# elif __hpux 1019# elif __hpux
643 res = sendfile (ofd, ifd, offset, count, 0, 0); 1020 res = sendfile (ofd, ifd, offset, count, 0, 0);
644 1021
645# elif __solaris 1022# elif __solaris
646 {
647 struct sendfilevec vec; 1023 struct sendfilevec vec;
648 size_t sbytes; 1024 size_t sbytes;
649 1025
650 vec.sfv_fd = ifd; 1026 vec.sfv_fd = ifd;
651 vec.sfv_flag = 0; 1027 vec.sfv_flag = 0;
652 vec.sfv_off = offset; 1028 vec.sfv_off = offset;
653 vec.sfv_len = count; 1029 vec.sfv_len = count;
654 1030
655 res = sendfilev (ofd, &vec, 1, &sbytes); 1031 res = sendfilev (ofd, &vec, 1, &sbytes);
656 1032
657 if (res < 0 && sbytes) 1033 if (res < 0 && sbytes)
658 res = sbytes; 1034 res = sbytes;
659 }
660 1035
661# endif 1036# endif
1037
1038#elif defined (_WIN32)
1039 /* does not work, just for documentation of what would need to be done */
1040 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1041 /* libeio guarantees that the file offset does not change, and windows */
1042 /* has no way to get an independent handle to the same file description */
1043 HANDLE h = TO_SOCKET (ifd);
1044 SetFilePointer (h, offset, 0, FILE_BEGIN);
1045 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1046
662#else 1047#else
663 res = -1; 1048 res = -1;
664 errno = ENOSYS; 1049 errno = ENOSYS;
665#endif 1050#endif
666 1051
1052 /* we assume sendfile can copy at least 128mb in one go */
1053 if (res <= 128 * 1024 * 1024)
1054 {
1055 if (res > 0)
1056 written += res;
1057
1058 if (written)
1059 return written;
1060
1061 break;
1062 }
1063 else
1064 {
1065 /* if we requested more, then probably the kernel was lazy */
1066 written += res;
1067 offset += res;
1068 count -= res;
1069
1070 if (!count)
1071 return written;
1072 }
1073 }
1074
667 if (res < 0 1075 if (res < 0
668 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1076 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1077 /* BSDs */
1078#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1079 || errno == ENOTSUP
1080#endif
1081 || errno == EOPNOTSUPP /* BSDs */
669#if __solaris 1082#if __solaris
670 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1083 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
671#endif 1084#endif
672 ) 1085 )
673 ) 1086 )
… …
704 } 1117 }
705 1118
706 return res; 1119 return res;
707} 1120}
708 1121
1122#ifdef PAGESIZE
1123# define eio_pagesize() PAGESIZE
1124#else
1125static intptr_t
1126eio_pagesize (void)
1127{
1128 static intptr_t page;
1129
1130 if (!page)
1131 page = sysconf (_SC_PAGESIZE);
1132
1133 return page;
1134}
1135#endif
1136
1137static void
1138eio_page_align (void **addr, size_t *length)
1139{
1140 intptr_t mask = eio_pagesize () - 1;
1141
1142 /* round down addr */
1143 intptr_t adj = mask & (intptr_t)*addr;
1144
1145 *addr = (void *)((intptr_t)*addr - adj);
1146 *length += adj;
1147
1148 /* round up length */
1149 *length = (*length + mask) & ~mask;
1150}
1151
1152#if !_POSIX_MEMLOCK
1153# define eio__mlockall(a) ((errno = ENOSYS), -1)
1154#else
1155
1156static int
1157eio__mlockall (int flags)
1158{
1159 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1160 extern int mallopt (int, int);
1161 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1162 #endif
1163
1164 if (EIO_MCL_CURRENT != MCL_CURRENT
1165 || EIO_MCL_FUTURE != MCL_FUTURE)
1166 {
1167 flags = 0
1168 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1169 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1170 }
1171
1172 return mlockall (flags);
1173}
1174#endif
1175
1176#if !_POSIX_MEMLOCK_RANGE
1177# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1178#else
1179
1180static int
1181eio__mlock (void *addr, size_t length)
1182{
1183 eio_page_align (&addr, &length);
1184
1185 return mlock (addr, length);
1186}
1187
1188#endif
1189
1190#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1191# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1192#else
1193
1194static int
1195eio__msync (void *mem, size_t len, int flags)
1196{
1197 eio_page_align (&mem, &len);
1198
1199 if (EIO_MS_ASYNC != MS_SYNC
1200 || EIO_MS_INVALIDATE != MS_INVALIDATE
1201 || EIO_MS_SYNC != MS_SYNC)
1202 {
1203 flags = 0
1204 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1205 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1206 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1207 }
1208
1209 return msync (mem, len, flags);
1210}
1211
1212#endif
1213
1214static int
1215eio__mtouch (eio_req *req)
1216{
1217 void *mem = req->ptr2;
1218 size_t len = req->size;
1219 int flags = req->int1;
1220
1221 eio_page_align (&mem, &len);
1222
1223 {
1224 intptr_t addr = (intptr_t)mem;
1225 intptr_t end = addr + len;
1226 intptr_t page = eio_pagesize ();
1227
1228 if (addr < end)
1229 if (flags & EIO_MT_MODIFY) /* modify */
1230 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1231 else
1232 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1233 }
1234
1235 return 0;
1236}
1237
1238/*****************************************************************************/
1239/* requests implemented outside eio_execute, because they are so large */
1240
1241static void
1242eio__realpath (eio_req *req, etp_worker *self)
1243{
1244 char *rel = req->ptr1;
1245 char *res;
1246 char *tmp1, *tmp2;
1247#if SYMLOOP_MAX > 32
1248 int links = SYMLOOP_MAX;
1249#else
1250 int links = 32;
1251#endif
1252
1253 req->result = -1;
1254
1255 errno = EINVAL;
1256 if (!rel)
1257 return;
1258
1259 errno = ENOENT;
1260 if (!*rel)
1261 return;
1262
1263 if (!req->ptr2)
1264 {
1265 X_LOCK (wrklock);
1266 req->flags |= EIO_FLAG_PTR2_FREE;
1267 X_UNLOCK (wrklock);
1268 req->ptr2 = malloc (PATH_MAX * 3);
1269
1270 errno = ENOMEM;
1271 if (!req->ptr2)
1272 return;
1273 }
1274
1275 res = req->ptr2;
1276 tmp1 = res + PATH_MAX;
1277 tmp2 = tmp1 + PATH_MAX;
1278
1279 if (*rel != '/')
1280 {
1281 if (!getcwd (res, PATH_MAX))
1282 return;
1283
1284 if (res [1]) /* only use if not / */
1285 res += strlen (res);
1286 }
1287
1288 while (*rel)
1289 {
1290 ssize_t len, linklen;
1291 char *beg = rel;
1292
1293 while (*rel && *rel != '/')
1294 ++rel;
1295
1296 len = rel - beg;
1297
1298 if (!len) /* skip slashes */
1299 {
1300 ++rel;
1301 continue;
1302 }
1303
1304 if (beg [0] == '.')
1305 {
1306 if (len == 1)
1307 continue; /* . - nop */
1308
1309 if (beg [1] == '.' && len == 2)
1310 {
1311 /* .. - back up one component, if possible */
1312
1313 while (res != req->ptr2)
1314 if (*--res == '/')
1315 break;
1316
1317 continue;
1318 }
1319 }
1320
1321 errno = ENAMETOOLONG;
1322 if (res + 1 + len + 1 >= tmp1)
1323 return;
1324
1325 /* copy one component */
1326 *res = '/';
1327 memcpy (res + 1, beg, len);
1328
1329 /* zero-terminate, for readlink */
1330 res [len + 1] = 0;
1331
1332 /* now check if it's a symlink */
1333 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1334
1335 if (linklen < 0)
1336 {
1337 if (errno != EINVAL)
1338 return;
1339
1340 /* it's a normal directory. hopefully */
1341 res += len + 1;
1342 }
1343 else
1344 {
1345 /* yay, it was a symlink - build new path in tmp2 */
1346 int rellen = strlen (rel);
1347
1348 errno = ENAMETOOLONG;
1349 if (linklen + 1 + rellen >= PATH_MAX)
1350 return;
1351
1352 if (*tmp1 == '/')
1353 res = req->ptr2; /* symlink resolves to an absolute path */
1354
1355 /* we need to be careful, as rel might point into tmp2 already */
1356 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1357 tmp2 [linklen] = '/';
1358 memcpy (tmp2, tmp1, linklen);
1359
1360 rel = tmp2;
1361 }
1362 }
1363
1364 /* special case for the lone root path */
1365 if (res == req->ptr2)
1366 *res++ = '/';
1367
1368 req->result = res - (char *)req->ptr2;
1369 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1370}
1371
1372static signed char
1373eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1374{
1375 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1376 : a->inode < b->inode ? -1
1377 : a->inode > b->inode ? 1
1378 : 0;
1379}
1380
1381#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1382
1383#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1384#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1385
1386static void
1387eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1388{
1389 unsigned char bits [9 + sizeof (ino_t) * 8];
1390 unsigned char *bit = bits;
1391
1392 assert (CHAR_BIT == 8);
1393 assert (sizeof (eio_dirent) * 8 < 256);
1394 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1395 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1396
1397 if (size <= EIO_SORT_FAST)
1398 return;
1399
1400 /* first prepare an array of bits to test in our radix sort */
1401 /* try to take endianness into account, as well as differences in ino_t sizes */
1402 /* inode_bits must contain all inodes ORed together */
1403 /* which is used to skip bits that are 0 everywhere, which is very common */
1404 {
1405 ino_t endianness;
1406 int i, j;
1407
1408 /* we store the byte offset of byte n into byte n of "endianness" */
1409 for (i = 0; i < sizeof (ino_t); ++i)
1410 ((unsigned char *)&endianness)[i] = i;
1411
1412 *bit++ = 0;
1413
1414 for (i = 0; i < sizeof (ino_t); ++i)
1415 {
1416 /* shifting off the byte offsets out of "endianness" */
1417 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1418 endianness >>= 8;
1419
1420 for (j = 0; j < 8; ++j)
1421 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1422 *bit++ = offs + j;
1423 }
1424
1425 for (j = 0; j < 8; ++j)
1426 if (score_bits & (1 << j))
1427 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1428 }
1429
1430 /* now actually do the sorting (a variant of MSD radix sort) */
1431 {
1432 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1433 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1434 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1435 int stk_idx = 0;
1436
1437 base_stk [stk_idx] = dents;
1438 end_stk [stk_idx] = dents + size;
1439 bit_stk [stk_idx] = bit - 1;
1440
1441 do
1442 {
1443 base = base_stk [stk_idx];
1444 end = end_stk [stk_idx];
1445 bit = bit_stk [stk_idx];
1446
1447 for (;;)
1448 {
1449 unsigned char O = *bit >> 3;
1450 unsigned char M = 1 << (*bit & 7);
1451
1452 eio_dirent *a = base;
1453 eio_dirent *b = end;
1454
1455 if (b - a < EIO_SORT_CUTOFF)
1456 break;
1457
1458 /* now bit-partition the array on the bit */
1459 /* this ugly asymmetric loop seems to perform much better than typical */
1460 /* partition algos found in the literature */
1461 do
1462 if (!(((unsigned char *)a)[O] & M))
1463 ++a;
1464 else if (!(((unsigned char *)--b)[O] & M))
1465 {
1466 eio_dirent tmp = *a; *a = *b; *b = tmp;
1467 ++a;
1468 }
1469 while (b > a);
1470
1471 /* next bit, or stop, if no bits left in this path */
1472 if (!*--bit)
1473 break;
1474
1475 base_stk [stk_idx] = a;
1476 end_stk [stk_idx] = end;
1477 bit_stk [stk_idx] = bit;
1478 ++stk_idx;
1479
1480 end = a;
1481 }
1482 }
1483 while (stk_idx--);
1484 }
1485}
1486
1487static void
1488eio_dent_insertion_sort (eio_dirent *dents, int size)
1489{
1490 /* first move the smallest element to the front, to act as a sentinel */
1491 {
1492 int i;
1493 eio_dirent *min = dents;
1494
1495 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1496 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1497 if (EIO_DENT_CMP (dents [i], <, *min))
1498 min = &dents [i];
1499
1500 /* swap elements 0 and j (minimum) */
1501 {
1502 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1503 }
1504 }
1505
1506 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1507 {
1508 eio_dirent *i, *j;
1509
1510 for (i = dents + 1; i < dents + size; ++i)
1511 {
1512 eio_dirent value = *i;
1513
1514 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1515 j [1] = j [0];
1516
1517 j [1] = value;
1518 }
1519 }
1520}
1521
1522static void
1523eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1524{
1525 if (size <= 1)
1526 return; /* our insertion sort relies on size > 0 */
1527
1528 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1529 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1530 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1531
1532 /* use an insertion sort at the end, or for small arrays, */
1533 /* as insertion sort is more efficient for small partitions */
1534 eio_dent_insertion_sort (dents, size);
1535}
1536
709/* read a full directory */ 1537/* read a full directory */
1538static void
710static void scandir_ (eio_req *req, worker *self) 1539eio__scandir (eio_req *req, etp_worker *self)
711{ 1540{
712 DIR *dirp; 1541 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; 1542 EIO_STRUCT_DIRENT *entp;
719 char *name, *names; 1543 char *name, *names;
720 int memlen = 4096; 1544 int namesalloc = 4096;
721 int memofs = 0; 1545 int namesoffs = 0;
1546 int flags = req->int1;
1547 eio_dirent *dents = 0;
1548 int dentalloc = 128;
722 int res = 0; 1549 int dentoffs = 0;
1550 ino_t inode_bits = 0;
1551
1552 req->result = -1;
1553
1554 if (!(flags & EIO_READDIR_DENTS))
1555 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
723 1556
724 X_LOCK (wrklock); 1557 X_LOCK (wrklock);
1558 /* the corresponding closedir is in ETP_WORKER_CLEAR */
725 self->dirp = dirp = opendir (req->ptr1); 1559 self->dirp = dirp = opendir (req->ptr1);
726 self->dbuf = u = malloc (sizeof (*u)); 1560
727 req->flags |= EIO_FLAG_PTR2_FREE; 1561 if (req->flags & EIO_FLAG_PTR1_FREE)
1562 free (req->ptr1);
1563
1564 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1565 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
728 req->ptr2 = names = malloc (memlen); 1566 req->ptr2 = names = malloc (namesalloc);
729 X_UNLOCK (wrklock); 1567 X_UNLOCK (wrklock);
730 1568
731 if (dirp && u && names) 1569 if (dirp && names && (!flags || dents))
732 for (;;) 1570 for (;;)
733 { 1571 {
734 errno = 0; 1572 errno = 0;
735 readdir_r (dirp, &u->d, &entp); 1573 entp = readdir (dirp);
736 1574
737 if (!entp) 1575 if (!entp)
1576 {
1577 if (errno)
1578 break;
1579
1580 /* sort etc. */
1581 req->int1 = flags;
1582 req->result = dentoffs;
1583
1584 if (flags & EIO_READDIR_STAT_ORDER)
1585 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1586 else if (flags & EIO_READDIR_DIRS_FIRST)
1587 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1588 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1589 else
1590 {
1591 /* in this case, all is known, and we just put dirs first and sort them */
1592 eio_dirent *oth = dents + dentoffs;
1593 eio_dirent *dir = dents;
1594
1595 /* now partition dirs to the front, and non-dirs to the back */
1596 /* by walking from both sides and swapping if necessary */
1597 while (oth > dir)
1598 {
1599 if (dir->type == EIO_DT_DIR)
1600 ++dir;
1601 else if ((--oth)->type == EIO_DT_DIR)
1602 {
1603 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1604
1605 ++dir;
1606 }
1607 }
1608
1609 /* now sort the dirs only (dirs all have the same score) */
1610 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1611 }
1612
738 break; 1613 break;
1614 }
739 1615
1616 /* now add the entry to our list(s) */
740 name = entp->d_name; 1617 name = entp->d_name;
741 1618
1619 /* skip . and .. entries */
742 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1620 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
743 { 1621 {
744 int len = strlen (name) + 1; 1622 int len = D_NAMLEN (entp) + 1;
745 1623
746 res++; 1624 while (ecb_expect_false (namesoffs + len > namesalloc))
747
748 while (memofs + len > memlen)
749 { 1625 {
750 memlen *= 2; 1626 namesalloc *= 2;
751 X_LOCK (wrklock); 1627 X_LOCK (wrklock);
752 req->ptr2 = names = realloc (names, memlen); 1628 req->ptr2 = names = realloc (names, namesalloc);
753 X_UNLOCK (wrklock); 1629 X_UNLOCK (wrklock);
754 1630
755 if (!names) 1631 if (!names)
756 break; 1632 break;
757 } 1633 }
758 1634
759 memcpy (names + memofs, name, len); 1635 memcpy (names + namesoffs, name, len);
1636
1637 if (dents)
1638 {
1639 struct eio_dirent *ent;
1640
1641 if (ecb_expect_false (dentoffs == dentalloc))
1642 {
1643 dentalloc *= 2;
1644 X_LOCK (wrklock);
1645 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1646 X_UNLOCK (wrklock);
1647
1648 if (!dents)
1649 break;
1650 }
1651
1652 ent = dents + dentoffs;
1653
1654 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1655 ent->namelen = len - 1;
1656 ent->inode = D_INO (entp);
1657
1658 inode_bits |= ent->inode;
1659
1660 switch (D_TYPE (entp))
1661 {
1662 default:
1663 ent->type = EIO_DT_UNKNOWN;
1664 flags |= EIO_READDIR_FOUND_UNKNOWN;
1665 break;
1666
1667 #ifdef DT_FIFO
1668 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1669 #endif
1670 #ifdef DT_CHR
1671 case DT_CHR: ent->type = EIO_DT_CHR; break;
1672 #endif
1673 #ifdef DT_MPC
1674 case DT_MPC: ent->type = EIO_DT_MPC; break;
1675 #endif
1676 #ifdef DT_DIR
1677 case DT_DIR: ent->type = EIO_DT_DIR; break;
1678 #endif
1679 #ifdef DT_NAM
1680 case DT_NAM: ent->type = EIO_DT_NAM; break;
1681 #endif
1682 #ifdef DT_BLK
1683 case DT_BLK: ent->type = EIO_DT_BLK; break;
1684 #endif
1685 #ifdef DT_MPB
1686 case DT_MPB: ent->type = EIO_DT_MPB; break;
1687 #endif
1688 #ifdef DT_REG
1689 case DT_REG: ent->type = EIO_DT_REG; break;
1690 #endif
1691 #ifdef DT_NWK
1692 case DT_NWK: ent->type = EIO_DT_NWK; break;
1693 #endif
1694 #ifdef DT_CMP
1695 case DT_CMP: ent->type = EIO_DT_CMP; break;
1696 #endif
1697 #ifdef DT_LNK
1698 case DT_LNK: ent->type = EIO_DT_LNK; break;
1699 #endif
1700 #ifdef DT_SOCK
1701 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1702 #endif
1703 #ifdef DT_DOOR
1704 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1705 #endif
1706 #ifdef DT_WHT
1707 case DT_WHT: ent->type = EIO_DT_WHT; break;
1708 #endif
1709 }
1710
1711 ent->score = 7;
1712
1713 if (flags & EIO_READDIR_DIRS_FIRST)
1714 {
1715 if (ent->type == EIO_DT_UNKNOWN)
1716 {
1717 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1718 ent->score = 1;
1719 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1720 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1721 }
1722 else if (ent->type == EIO_DT_DIR)
1723 ent->score = 0;
1724 }
1725 }
1726
760 memofs += len; 1727 namesoffs += len;
1728 ++dentoffs;
1729 }
1730
1731 if (EIO_CANCELLED (req))
1732 {
1733 errno = ECANCELED;
1734 break;
761 } 1735 }
762 } 1736 }
763
764 if (errno)
765 res = -1;
766
767 req->result = res;
768} 1737}
769 1738
770/*****************************************************************************/ 1739/*****************************************************************************/
771 1740
772#define ALLOC(len) \ 1741#define ALLOC(len) \
… …
782 req->result = -1; \ 1751 req->result = -1; \
783 break; \ 1752 break; \
784 } \ 1753 } \
785 } 1754 }
786 1755
787X_THREAD_PROC (eio_proc) 1756X_THREAD_PROC (etp_proc)
788{ 1757{
789 eio_req *req; 1758 ETP_REQ *req;
790 struct timespec ts; 1759 struct timespec ts;
791 worker *self = (worker *)thr_arg; 1760 etp_worker *self = (etp_worker *)thr_arg;
792 1761
793 /* try to distribute timeouts somewhat randomly */ 1762 /* try to distribute timeouts somewhat randomly */
794 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1763 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
795 1764
796 for (;;) 1765 for (;;)
797 { 1766 {
798 ts.tv_sec = time (0) + IDLE_TIMEOUT;
799
800 X_LOCK (reqlock); 1767 X_LOCK (reqlock);
801 1768
802 for (;;) 1769 for (;;)
803 { 1770 {
804 self->req = req = reqq_shift (&req_queue); 1771 self->req = req = reqq_shift (&req_queue);
… …
806 if (req) 1773 if (req)
807 break; 1774 break;
808 1775
809 ++idle; 1776 ++idle;
810 1777
1778 ts.tv_sec = time (0) + idle_timeout;
811 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1779 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
812 { 1780 {
813 if (idle > max_idle) 1781 if (idle > max_idle)
814 { 1782 {
815 --idle; 1783 --idle;
… …
820 goto quit; 1788 goto quit;
821 } 1789 }
822 1790
823 /* we are allowed to idle, so do so without any timeout */ 1791 /* we are allowed to idle, so do so without any timeout */
824 X_COND_WAIT (reqwait, reqlock); 1792 X_COND_WAIT (reqwait, reqlock);
825 ts.tv_sec = time (0) + IDLE_TIMEOUT;
826 } 1793 }
827 1794
828 --idle; 1795 --idle;
829 } 1796 }
830 1797
831 --nready; 1798 --nready;
832 1799
833 X_UNLOCK (reqlock); 1800 X_UNLOCK (reqlock);
834 1801
835 errno = 0; /* strictly unnecessary */ 1802 if (req->type < 0)
1803 goto quit;
836 1804
837 if (!EIO_CANCELLED (req)) 1805 if (!EIO_CANCELLED (req))
838 switch (req->type) 1806 ETP_EXECUTE (self, req);
839 {
840 case EIO_READ: ALLOC (req->size);
841 req->result = req->offs >= 0
842 ? pread (req->int1, req->ptr2, req->size, req->offs)
843 : read (req->int1, req->ptr2, req->size); break;
844 case EIO_WRITE: req->result = req->offs >= 0
845 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
846 : write (req->int1, req->ptr2, req->size); break;
847
848 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
849 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
850
851 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
852 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
853 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
854 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
855 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
856 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
857
858 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
859 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
860 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
861 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
862 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
863 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
864
865 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
866 case EIO_CLOSE: req->result = close (req->int1); break;
867 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
868 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
869 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
870 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
871 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
872 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
873 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
874 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
875
876 case EIO_READLINK: ALLOC (NAME_MAX);
877 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
878
879 case EIO_SYNC: req->result = 0; sync (); break;
880 case EIO_FSYNC: req->result = fsync (req->int1); break;
881 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
882
883 case EIO_READDIR: scandir_ (req, self); break;
884
885 case EIO_BUSY:
886#ifdef _WIN32
887 Sleep (req->nv1 * 1000.);
888#else
889 {
890 struct timeval tv;
891
892 tv.tv_sec = req->nv1;
893 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
894
895 req->result = select (0, 0, 0, 0, &tv);
896 }
897#endif
898 break;
899
900 case EIO_UTIME:
901 case EIO_FUTIME:
902 {
903 struct timeval tv[2];
904 struct timeval *times;
905
906 if (req->nv1 != -1. || req->nv2 != -1.)
907 {
908 tv[0].tv_sec = req->nv1;
909 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
910 tv[1].tv_sec = req->nv2;
911 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
912
913 times = tv;
914 }
915 else
916 times = 0;
917
918
919 req->result = req->type == EIO_FUTIME
920 ? futimes (req->int1, times)
921 : utimes (req->ptr1, times);
922 }
923
924 case EIO_GROUP:
925 case EIO_NOP:
926 break;
927
928 case EIO_QUIT:
929 goto quit;
930
931 default:
932 req->result = -1;
933 break;
934 }
935
936 req->errorno = errno;
937 1807
938 X_LOCK (reslock); 1808 X_LOCK (reslock);
939 1809
940 ++npending; 1810 ++npending;
941 1811
942 if (!reqq_push (&res_queue, req) && want_poll_cb) 1812 if (!reqq_push (&res_queue, req) && want_poll_cb)
943 want_poll_cb (); 1813 want_poll_cb ();
944 1814
945 self->req = 0; 1815 self->req = 0;
946 worker_clear (self); 1816 etp_worker_clear (self);
947 1817
948 X_UNLOCK (reslock); 1818 X_UNLOCK (reslock);
949 } 1819 }
950 1820
951quit: 1821quit:
952 X_LOCK (wrklock); 1822 X_LOCK (wrklock);
953 worker_free (self); 1823 etp_worker_free (self);
954 X_UNLOCK (wrklock); 1824 X_UNLOCK (wrklock);
955 1825
956 return 0; 1826 return 0;
957} 1827}
958 1828
959/*****************************************************************************/ 1829/*****************************************************************************/
960 1830
961static void eio_atfork_prepare (void) 1831int 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)) 1832eio_init (void (*want_poll)(void), void (*done_poll)(void))
1018{ 1833{
1019 want_poll_cb = want_poll; 1834 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} 1835}
1035 1836
1837ecb_inline void
1036static void eio_api_destroy (eio_req *req) 1838eio_api_destroy (eio_req *req)
1037{ 1839{
1038 free (req); 1840 free (req);
1039} 1841}
1040 1842
1041#define REQ(rtype) \ 1843#define REQ(rtype) \
… …
1043 \ 1845 \
1044 req = (eio_req *)calloc (1, sizeof *req); \ 1846 req = (eio_req *)calloc (1, sizeof *req); \
1045 if (!req) \ 1847 if (!req) \
1046 return 0; \ 1848 return 0; \
1047 \ 1849 \
1048 req->type = EIO_ ## rtype; \ 1850 req->type = rtype; \
1049 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 1851 req->pri = pri; \
1050 req->finish = cb; \ 1852 req->finish = cb; \
1853 req->data = data; \
1051 req->destroy = eio_api_destroy; 1854 req->destroy = eio_api_destroy;
1052 1855
1053#define SEND eio_submit (req); return req 1856#define SEND eio_submit (req); return req
1054 1857
1055#define PATH (void)0 1858#define PATH \
1859 req->flags |= EIO_FLAG_PTR1_FREE; \
1860 req->ptr1 = strdup (path); \
1861 if (!req->ptr1) \
1862 { \
1863 eio_api_destroy (req); \
1864 return 0; \
1865 }
1056 1866
1867static void
1868eio_execute (etp_worker *self, eio_req *req)
1869{
1870 switch (req->type)
1871 {
1872 case EIO_READ: ALLOC (req->size);
1873 req->result = req->offs >= 0
1874 ? pread (req->int1, req->ptr2, req->size, req->offs)
1875 : read (req->int1, req->ptr2, req->size); break;
1876 case EIO_WRITE: req->result = req->offs >= 0
1877 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1878 : write (req->int1, req->ptr2, req->size); break;
1879
1880 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1881 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1882
1883 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1884 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1885 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1886 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1887 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1888 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1889
1890 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1891 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1892 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1893 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1894
1895 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1896 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1897 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1898 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1899 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1900 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1901
1902 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1903 case EIO_CLOSE: req->result = close (req->int1); break;
1904 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1905 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1906 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1907 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1908 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1909 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1910 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1911 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1912
1913 case EIO_REALPATH: eio__realpath (req, self); break;
1914
1915 case EIO_READLINK: ALLOC (PATH_MAX);
1916 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1917
1918 case EIO_SYNC: req->result = 0; sync (); break;
1919 case EIO_FSYNC: req->result = fsync (req->int1); break;
1920 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1921 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1922 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1923 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1924 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1925 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1926
1927 case EIO_READDIR: eio__scandir (req, self); break;
1928
1929 case EIO_BUSY:
1930#ifdef _WIN32
1931 Sleep (req->nv1 * 1e3);
1932#else
1933 {
1934 struct timeval tv;
1935
1936 tv.tv_sec = req->nv1;
1937 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1938
1939 req->result = select (0, 0, 0, 0, &tv);
1940 }
1941#endif
1942 break;
1943
1944 case EIO_UTIME:
1945 case EIO_FUTIME:
1946 {
1947 struct timeval tv[2];
1948 struct timeval *times;
1949
1950 if (req->nv1 != -1. || req->nv2 != -1.)
1951 {
1952 tv[0].tv_sec = req->nv1;
1953 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1954 tv[1].tv_sec = req->nv2;
1955 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1956
1957 times = tv;
1958 }
1959 else
1960 times = 0;
1961
1962 req->result = req->type == EIO_FUTIME
1963 ? futimes (req->int1, times)
1964 : utimes (req->ptr1, times);
1965 }
1966 break;
1967
1968 case EIO_GROUP:
1969 abort (); /* handled in eio_request */
1970
1971 case EIO_NOP:
1972 req->result = 0;
1973 break;
1974
1975 case EIO_CUSTOM:
1976 req->feed (req);
1977 break;
1978
1979 default:
1980 errno = ENOSYS;
1981 req->result = -1;
1982 break;
1983 }
1984
1985 req->errorno = errno;
1986}
1987
1988#ifndef EIO_NO_WRAPPERS
1989
1990eio_req *eio_nop (int pri, eio_cb cb, void *data)
1991{
1992 REQ (EIO_NOP); SEND;
1993}
1994
1995eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1996{
1997 REQ (EIO_BUSY); req->nv1 = delay; SEND;
1998}
1999
1057eio_req *eio_fsync (int fd, eio_cb cb) 2000eio_req *eio_sync (int pri, eio_cb cb, void *data)
1058{ 2001{
2002 REQ (EIO_SYNC); SEND;
2003}
2004
2005eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2006{
1059 REQ (FSYNC); req->int1 = fd; SEND; 2007 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1060} 2008}
1061 2009
2010eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2011{
2012 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2013}
2014
2015eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2016{
2017 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2018}
2019
2020eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2021{
2022 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2023}
2024
2025eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2026{
2027 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2028}
2029
2030eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2031{
2032 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2033}
2034
1062eio_req *eio_fdatasync (int fd, eio_cb cb) 2035eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1063{ 2036{
1064 REQ (FDATASYNC); req->int1 = fd; SEND; 2037 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1065} 2038}
1066 2039
2040eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2041{
2042 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2043}
2044
1067eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb) 2045eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1068{ 2046{
1069 REQ (READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2047 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1070} 2048}
1071 2049
1072eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2050eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1073{ 2051{
1074 REQ (READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2052 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1075} 2053}
1076 2054
1077eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb) 2055eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1078{ 2056{
1079 REQ (WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2057 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1080} 2058}
1081 2059
1082eio_req *eio_fstat (int fd, eio_cb cb) 2060eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1083{ 2061{
1084 REQ (FSTAT); req->int1 = fd; SEND; 2062 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1085} 2063}
1086 2064
2065eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2066{
2067 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2068}
2069
1087eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb) 2070eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1088{ 2071{
1089 REQ (FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2072 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1090} 2073}
1091 2074
1092eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb) 2075eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1093{ 2076{
1094 REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 2077 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1095} 2078}
1096 2079
1097eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb) 2080eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1098{ 2081{
1099 REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2082 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1100} 2083}
1101 2084
1102eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb) 2085eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1103{ 2086{
1104 REQ (FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2087 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1105} 2088}
1106 2089
1107eio_req *eio_dup2 (int fd, int fd2, eio_cb cb) 2090eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1108{ 2091{
1109 REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND; 2092 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1110} 2093}
1111 2094
1112eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb) 2095eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1113{ 2096{
1114 REQ (SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 2097 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1115} 2098}
1116 2099
1117eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb) 2100eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1118{ 2101{
1119 REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 2102 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1120} 2103}
1121 2104
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) 2105eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1138{ 2106{
1139 REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 2107 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1140} 2108}
1141 2109
1142eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb) 2110eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1143{ 2111{
1144 REQ (TRUNCATE); PATH; req->offs = offset; SEND; 2112 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1145} 2113}
1146 2114
1147eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 2115eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1148{ 2116{
1149 REQ (CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2117 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1150} 2118}
1151 2119
1152eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb) 2120eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1153{ 2121{
1154 REQ (CHMOD); PATH; req->int2 = (long)mode; SEND; 2122 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1155} 2123}
1156 2124
1157eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb) 2125eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1158{ 2126{
1159 REQ (MKDIR); PATH; req->int2 = (long)mode; SEND; 2127 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1160} 2128}
1161 2129
1162eio_req *eio_unlink (const char *path, eio_cb cb) 2130static eio_req *
2131eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1163{ 2132{
1164 REQ (UNLINK); PATH; SEND;
1165}
1166
1167eio_req *eio_rmdir (const char *path, eio_cb cb)
1168{
1169 REQ (RMDIR); PATH; SEND; 2133 REQ (type); PATH; SEND;
1170} 2134}
1171 2135
2136eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2137{
2138 return eio__1path (EIO_READLINK, path, pri, cb, data);
2139}
2140
2141eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2142{
2143 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2144}
2145
2146eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2147{
2148 return eio__1path (EIO_STAT, path, pri, cb, data);
2149}
2150
2151eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2152{
2153 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2154}
2155
2156eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2157{
2158 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2159}
2160
2161eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2162{
2163 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2164}
2165
1172eio_req *eio_readdir (const char *path, eio_cb cb) 2166eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1173{ 2167{
1174 REQ (READDIR); PATH; SEND; 2168 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1175} 2169}
1176 2170
2171eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2172{
2173 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2174}
2175
1177eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb) 2176eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1178{ 2177{
1179 REQ (MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2178 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1180} 2179}
1181 2180
1182eio_req *eio_busy (double delay, eio_cb cb) 2181static eio_req *
2182eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1183{ 2183{
1184 REQ (BUSY); req->nv1 = delay; SEND; 2184 REQ (type); PATH;
1185}
1186 2185
2186 req->flags |= EIO_FLAG_PTR2_FREE;
2187 req->ptr2 = strdup (new_path);
2188 if (!req->ptr2)
2189 {
2190 eio_api_destroy (req);
2191 return 0;
2192 }
2193
2194 SEND;
2195}
2196
2197eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2198{
2199 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2200}
2201
2202eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2203{
2204 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2205}
2206
2207eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2208{
2209 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2210}
2211
2212eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2213{
2214 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2215}
2216
2217#endif
2218
1187eio_req *eio_nop (eio_cb cb) 2219eio_req *eio_grp (eio_cb cb, void *data)
1188{ 2220{
2221 const int pri = EIO_PRI_MAX;
2222
1189 REQ (NOP); SEND; 2223 REQ (EIO_GROUP); SEND;
1190} 2224}
1191 2225
1192#undef REQ 2226#undef REQ
1193#undef PATH 2227#undef PATH
1194#undef SEND 2228#undef SEND
1195 2229
1196#if 0 2230/*****************************************************************************/
2231/* grp functions */
1197 2232
1198void 2233void
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) 2234eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1572{ 2235{
1573 grp->int2 = limit; 2236 grp->int2 = limit;
1574 grp->feed = feed; 2237 grp->feed = feed;
1575 2238
1576 grp_try_feed (grp); 2239 grp_try_feed (grp);
1577} 2240}
1578 2241
2242void
1579void eio_grp_limit (eio_req *grp, int limit) 2243eio_grp_limit (eio_req *grp, int limit)
1580{ 2244{
1581 grp->int2 = limit; 2245 grp->int2 = limit;
1582 2246
1583 grp_try_feed (grp); 2247 grp_try_feed (grp);
1584} 2248}
1585 2249
2250void
1586void eio_grp_add (eio_req *grp, eio_req *req) 2251eio_grp_add (eio_req *grp, eio_req *req)
1587{ 2252{
1588 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2253 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2254
2255 grp->flags |= EIO_FLAG_GROUPADD;
1589 2256
1590 ++grp->size; 2257 ++grp->size;
1591 req->grp = grp; 2258 req->grp = grp;
1592 2259
1593 req->grp_prev = 0; 2260 req->grp_prev = 0;
… …
1597 grp->grp_first->grp_prev = req; 2264 grp->grp_first->grp_prev = req;
1598 2265
1599 grp->grp_first = req; 2266 grp->grp_first = req;
1600} 2267}
1601 2268
2269/*****************************************************************************/
2270/* misc garbage */
1602 2271
2272ssize_t
2273eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2274{
2275 etp_worker wrk;
2276 ssize_t ret;
2277
2278 wrk.dbuf = 0;
2279
2280 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2281
2282 if (wrk.dbuf)
2283 free (wrk.dbuf);
2284
2285 return ret;
2286}
2287

Diff Legend

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