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

Comparing libeio/eio.c (file contents):
Revision 1.2 by root, Sat May 10 23:33:40 2008 UTC vs.
Revision 1.79 by root, Tue Jul 5 16:57:41 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
6#include "EXTERN.h"
7#include "perl.h"
8#include "XSUB.h"
9
10#include <stddef.h> 53#include <stddef.h>
11#include <stdlib.h> 54#include <stdlib.h>
55#include <string.h>
12#include <errno.h> 56#include <errno.h>
13#include <sys/types.h> 57#include <sys/types.h>
14#include <sys/stat.h> 58#include <sys/stat.h>
59#include <sys/statvfs.h>
15#include <limits.h> 60#include <limits.h>
16#include <fcntl.h> 61#include <fcntl.h>
17#include <sched.h> 62#include <assert.h>
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
18 69
19#ifndef EIO_FINISH 70#ifndef EIO_FINISH
20# 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
21#endif 72#endif
22 73
… …
29#endif 80#endif
30 81
31#ifdef _WIN32 82#ifdef _WIN32
32 83
33 /*doh*/ 84 /*doh*/
34
35#else 85#else
36 86
37# include "config.h"
38# include <sys/time.h> 87# include <sys/time.h>
39# include <sys/select.h> 88# include <sys/select.h>
40# include <unistd.h> 89# include <unistd.h>
41# include <utime.h> 90# include <utime.h>
42# include <signal.h> 91# include <signal.h>
92# include <dirent.h>
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
43 115
44# ifndef EIO_STRUCT_DIRENT 116# ifndef EIO_STRUCT_DIRENT
45# define EIO_STRUCT_DIRENT struct dirent 117# define EIO_STRUCT_DIRENT struct dirent
46# endif 118# endif
47 119
48#endif 120#endif
49 121
50# ifndef EIO_STRUCT_STAT
51# define EIO_STRUCT_STAT struct stat
52# endif
53
54#if HAVE_SENDFILE 122#if HAVE_SENDFILE
55# if __linux 123# if __linux
56# include <sys/sendfile.h> 124# include <sys/sendfile.h>
57# elif __freebsd 125# elif __FreeBSD__ || defined __APPLE__
58# include <sys/socket.h> 126# include <sys/socket.h>
59# include <sys/uio.h> 127# include <sys/uio.h>
60# elif __hpux 128# elif __hpux
61# include <sys/socket.h> 129# include <sys/socket.h>
62# elif __solaris /* not yet */ 130# elif __solaris
63# include <sys/sendfile.h> 131# include <sys/sendfile.h>
64# else 132# else
65# error sendfile support requested but not available 133# error sendfile support requested but not available
66# endif 134# endif
67#endif 135#endif
68 136
69/* number of seconds after which an idle threads exit */ 137#ifndef D_TYPE
70#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
71 146
72/* used for struct dirent, AIX doesn't provide it */ 147/* used for struct dirent, AIX doesn't provide it */
73#ifndef NAME_MAX 148#ifndef NAME_MAX
74# define NAME_MAX 4096 149# define NAME_MAX 4096
75#endif 150#endif
76 151
152/* used for readlink etc. */
153#ifndef PATH_MAX
154# define PATH_MAX 4096
155#endif
156
77/* buffer size for various temporary buffers */ 157/* buffer size for various temporary buffers */
78#define EIO_BUFSIZE 65536 158#define EIO_BUFSIZE 65536
79 159
80#define dBUF \ 160#define dBUF \
81 char *eio_buf; \ 161 char *eio_buf; \
82 X_LOCK (wrklock); \ 162 ETP_WORKER_LOCK (self); \
83 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 163 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
84 X_UNLOCK (wrklock); \ 164 ETP_WORKER_UNLOCK (self); \
165 errno = ENOMEM; \
85 if (!eio_buf) \ 166 if (!eio_buf) \
86 return -1; 167 return -1;
87 168
88#define EIO_TICKS ((1000000 + 1023) >> 10) 169#define EIO_TICKS ((1000000 + 1023) >> 10)
89 170
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
90static void (*want_poll_cb) (void); 214static void (*want_poll_cb) (void);
91static void (*done_poll_cb) (void); 215static void (*done_poll_cb) (void);
92 216
93static unsigned int max_poll_time = 0; 217static unsigned int max_poll_time; /* reslock */
94static unsigned int max_poll_reqs = 0; 218static unsigned int max_poll_reqs; /* reslock */
95 219
96/* calculcate time difference in ~1/EIO_TICKS of a second */ 220static volatile unsigned int nreqs; /* reqlock */
97static int tvdiff (struct timeval *tv1, struct timeval *tv2) 221static volatile unsigned int nready; /* reqlock */
98{ 222static volatile unsigned int npending; /* reqlock */
99 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 223static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
100 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 224static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
101}
102 225
103static 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)
104 257
105/* worker threads management */ 258/* worker threads management */
106static mutex_t wrklock = X_MUTEX_INIT;
107 259
108typedef struct worker { 260static void ecb_cold
109 /* locked by wrklock */
110 struct worker *prev, *next;
111
112 thread_t tid;
113
114 /* locked by reslock, reqlock or wrklock */
115 eio_req *req; /* currently processed request */
116 void *dbuf;
117 DIR *dirp;
118} worker;
119
120static worker wrk_first = { &wrk_first, &wrk_first, 0 };
121
122static void worker_clear (worker *wrk) 261etp_worker_clear (etp_worker *wrk)
123{ 262{
124 if (wrk->dirp) 263 ETP_WORKER_CLEAR (wrk);
125 {
126 closedir (wrk->dirp);
127 wrk->dirp = 0;
128 }
129
130 if (wrk->dbuf)
131 {
132 free (wrk->dbuf);
133 wrk->dbuf = 0;
134 }
135} 264}
136 265
137static void worker_free (worker *wrk) 266static void ecb_cold
267etp_worker_free (etp_worker *wrk)
138{ 268{
139 wrk->next->prev = wrk->prev; 269 wrk->next->prev = wrk->prev;
140 wrk->prev->next = wrk->next; 270 wrk->prev->next = wrk->next;
141 271
142 free (wrk); 272 free (wrk);
143} 273}
144 274
145static volatile unsigned int nreqs, nready, npending; 275static unsigned int
146static volatile unsigned int max_idle = 4; 276etp_nreqs (void)
147
148static mutex_t reslock = X_MUTEX_INIT;
149static mutex_t reqlock = X_MUTEX_INIT;
150static cond_t reqwait = X_COND_INIT;
151
152unsigned int eio_nreqs (void)
153{ 277{
278 int retval;
279 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
280 retval = nreqs;
281 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 282 return retval;
155} 283}
156 284
157unsigned int eio_nready (void) 285static unsigned int
286etp_nready (void)
158{ 287{
159 unsigned int retval; 288 unsigned int retval;
160 289
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 291 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 293
165 return retval; 294 return retval;
166} 295}
167 296
168unsigned int eio_npending (void) 297static unsigned int
298etp_npending (void)
169{ 299{
170 unsigned int retval; 300 unsigned int retval;
171 301
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 302 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 303 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 304 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 305
176 return retval; 306 return retval;
177} 307}
178 308
179unsigned int eio_nthreads (void) 309static unsigned int
310etp_nthreads (void)
180{ 311{
181 unsigned int retval; 312 unsigned int retval;
182 313
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 314 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 315 retval = started;
… …
191 * a somewhat faster data structure might be nice, but 322 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 323 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 324 * per shift, the most expensive operation.
194 */ 325 */
195typedef struct { 326typedef struct {
196 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 */
197 int size; 328 int size;
198} reqq; 329} etp_reqq;
199 330
200static reqq req_queue; 331static etp_reqq req_queue;
201static reqq res_queue; 332static etp_reqq res_queue;
202 333
203static int reqq_push (reqq *q, eio_req *req) 334static int ecb_noinline
335reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 336{
205 int pri = req->pri; 337 int pri = req->pri;
206 req->next = 0; 338 req->next = 0;
207 339
208 if (q->qe[pri]) 340 if (q->qe[pri])
… …
214 q->qe[pri] = q->qs[pri] = req; 346 q->qe[pri] = q->qs[pri] = req;
215 347
216 return q->size++; 348 return q->size++;
217} 349}
218 350
219static eio_req *reqq_shift (reqq *q) 351static ETP_REQ * ecb_noinline
352reqq_shift (etp_reqq *q)
220{ 353{
221 int pri; 354 int pri;
222 355
223 if (!q->size) 356 if (!q->size)
224 return 0; 357 return 0;
225 358
226 --q->size; 359 --q->size;
227 360
228 for (pri = EIO_NUM_PRI; pri--; ) 361 for (pri = ETP_NUM_PRI; pri--; )
229 { 362 {
230 eio_req *req = q->qs[pri]; 363 eio_req *req = q->qs[pri];
231 364
232 if (req) 365 if (req)
233 { 366 {
… …
239 } 372 }
240 373
241 abort (); 374 abort ();
242} 375}
243 376
244static void grp_try_feed (eio_req *grp) 377static void ecb_cold
378etp_thread_init (void)
245{ 379{
246 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)
247 { 420 {
248 int old_len = grp->size; 421 etp_worker *wrk = wrk_first.next;
249 422
250 EIO_FEED (grp); 423 if (wrk->req)
424 ETP_DESTROY (wrk->req);
251 425
252 /* stop if no progress has been made */ 426 etp_worker_clear (wrk);
253 if (old_len == grp->size) 427 etp_worker_free (wrk);
254 {
255 grp->feed = 0;
256 break;
257 }
258 } 428 }
259}
260 429
261static int eio_finish (eio_req *req); 430 started = 0;
431 idle = 0;
432 nreqs = 0;
433 nready = 0;
434 npending = 0;
262 435
263static int grp_dec (eio_req *grp) 436 etp_thread_init ();
264{ 437}
265 --grp->size;
266 438
267 /* call feeder, if applicable */ 439static void ecb_cold
268 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}
269 445
270 /* finish, if done */ 446static int ecb_cold
271 if (!grp->size && grp->int1) 447etp_init (void (*want_poll)(void), void (*done_poll)(void))
272 return eio_finish (grp); 448{
273 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
274 return 0; 456 return 0;
275} 457}
276 458
277void eio_destroy (eio_req *req)
278{
279 if ((req)->flags & EIO_FLAG_PTR2_FREE)
280 free (req->ptr2);
281
282 EIO_DESTROY (req);
283}
284
285static int eio_finish (eio_req *req)
286{
287 int res = EIO_FINISH (req);
288
289 if (req->grp)
290 {
291 int res2;
292 eio_req *grp = req->grp;
293
294 /* unlink request */
295 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
296 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
297
298 if (grp->grp_first == req)
299 grp->grp_first = req->grp_next;
300
301 res2 = grp_dec (grp);
302
303 if (!res && res2)
304 res = res2;
305 }
306
307 eio_destroy (req);
308
309 return res;
310}
311
312void eio_grp_cancel (eio_req *grp)
313{
314 for (grp = grp->grp_first; grp; grp = grp->grp_next)
315 eio_cancel (grp);
316}
317
318void eio_cancel (eio_req *req)
319{
320 req->flags |= EIO_FLAG_CANCELLED;
321
322 eio_grp_cancel (req);
323}
324
325X_THREAD_PROC (eio_proc); 459X_THREAD_PROC (etp_proc);
326 460
327static void start_thread (void) 461static void ecb_cold
462etp_start_thread (void)
328{ 463{
329 worker *wrk = calloc (1, sizeof (worker)); 464 etp_worker *wrk = calloc (1, sizeof (etp_worker));
330 465
331 if (!wrk) 466 /*TODO*/
332 croak ("unable to allocate worker thread data"); 467 assert (("unable to allocate worker thread data", wrk));
333 468
334 X_LOCK (wrklock); 469 X_LOCK (wrklock);
335 470
336 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 471 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
337 { 472 {
338 wrk->prev = &wrk_first; 473 wrk->prev = &wrk_first;
339 wrk->next = wrk_first.next; 474 wrk->next = wrk_first.next;
340 wrk_first.next->prev = wrk; 475 wrk_first.next->prev = wrk;
341 wrk_first.next = wrk; 476 wrk_first.next = wrk;
… …
345 free (wrk); 480 free (wrk);
346 481
347 X_UNLOCK (wrklock); 482 X_UNLOCK (wrklock);
348} 483}
349 484
485static void
350static void maybe_start_thread (void) 486etp_maybe_start_thread (void)
351{ 487{
352 if (eio_nthreads () >= wanted) 488 if (ecb_expect_true (etp_nthreads () >= wanted))
353 return; 489 return;
354 490
355 /* todo: maybe use idle here, but might be less exact */ 491 /* todo: maybe use idle here, but might be less exact */
356 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 ()))
357 return; 493 return;
358 494
359 start_thread (); 495 etp_start_thread ();
360} 496}
361 497
362void eio_submit (eio_req *req) 498static void ecb_cold
363{ 499etp_end_thread (void)
364 ++nreqs;
365
366 X_LOCK (reqlock);
367 ++nready;
368 reqq_push (&req_queue, req);
369 X_COND_SIGNAL (reqwait);
370 X_UNLOCK (reqlock);
371
372 maybe_start_thread ();
373}
374
375static void end_thread (void)
376{ 500{
377 eio_req *req = calloc (1, sizeof (eio_req)); 501 eio_req *req = calloc (1, sizeof (eio_req));
378 502
379 req->type = EIO_QUIT; 503 req->type = -1;
380 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 504 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381 505
382 X_LOCK (reqlock); 506 X_LOCK (reqlock);
383 reqq_push (&req_queue, req); 507 reqq_push (&req_queue, req);
384 X_COND_SIGNAL (reqwait); 508 X_COND_SIGNAL (reqwait);
385 X_UNLOCK (reqlock); 509 X_UNLOCK (reqlock);
… …
387 X_LOCK (wrklock); 511 X_LOCK (wrklock);
388 --started; 512 --started;
389 X_UNLOCK (wrklock); 513 X_UNLOCK (wrklock);
390} 514}
391 515
392void eio_set_max_poll_time (double nseconds) 516static int
393{
394 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
395 max_poll_time = nseconds;
396 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397}
398
399void eio_set_max_poll_reqs (unsigned int maxreqs)
400{
401 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
402 max_poll_reqs = maxreqs;
403 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404}
405
406void eio_set_max_idle (unsigned int nthreads)
407{
408 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
409 max_idle = nthreads <= 0 ? 1 : nthreads;
410 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411}
412
413void eio_set_min_parallel (unsigned int nthreads)
414{
415 if (wanted < nthreads)
416 wanted = nthreads;
417}
418
419void eio_set_max_parallel (unsigned int nthreads)
420{
421 if (wanted > nthreads)
422 wanted = nthreads;
423
424 while (started > wanted)
425 end_thread ();
426}
427
428int eio_poll (void) 517etp_poll (void)
429{ 518{
430 int maxreqs = max_poll_reqs; 519 unsigned int maxreqs;
520 unsigned int maxtime;
431 struct timeval tv_start, tv_now; 521 struct timeval tv_start, tv_now;
432 eio_req *req;
433 522
523 X_LOCK (reslock);
524 maxreqs = max_poll_reqs;
525 maxtime = max_poll_time;
526 X_UNLOCK (reslock);
527
434 if (max_poll_time) 528 if (maxtime)
435 gettimeofday (&tv_start, 0); 529 gettimeofday (&tv_start, 0);
436 530
437 for (;;) 531 for (;;)
438 { 532 {
533 ETP_REQ *req;
534
439 maybe_start_thread (); 535 etp_maybe_start_thread ();
440 536
441 X_LOCK (reslock); 537 X_LOCK (reslock);
442 req = reqq_shift (&res_queue); 538 req = reqq_shift (&res_queue);
443 539
444 if (req) 540 if (req)
… …
452 X_UNLOCK (reslock); 548 X_UNLOCK (reslock);
453 549
454 if (!req) 550 if (!req)
455 return 0; 551 return 0;
456 552
553 X_LOCK (reqlock);
457 --nreqs; 554 --nreqs;
555 X_UNLOCK (reqlock);
458 556
459 if (req->type == EIO_GROUP && req->size) 557 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
460 { 558 {
461 req->int1 = 1; /* mark request as delayed */ 559 req->int1 = 1; /* mark request as delayed */
462 continue; 560 continue;
463 } 561 }
464 else 562 else
465 { 563 {
466 int res = eio_finish (req); 564 int res = ETP_FINISH (req);
467 if (res) 565 if (ecb_expect_false (res))
468 return res; 566 return res;
469 } 567 }
470 568
471 if (maxreqs && !--maxreqs) 569 if (ecb_expect_false (maxreqs && !--maxreqs))
472 break; 570 break;
473 571
474 if (max_poll_time) 572 if (maxtime)
475 { 573 {
476 gettimeofday (&tv_now, 0); 574 gettimeofday (&tv_now, 0);
477 575
478 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 576 if (tvdiff (&tv_start, &tv_now) >= maxtime)
479 break; 577 break;
480 } 578 }
481 } 579 }
482 580
483 errno = EAGAIN; 581 errno = EAGAIN;
484 return -1; 582 return -1;
485} 583}
486 584
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)
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
487/*****************************************************************************/ 837/*****************************************************************************/
488/* work around various missing functions */ 838/* work around various missing functions */
489 839
490#if !HAVE_PREADWRITE 840#if !HAVE_PREADWRITE
841# undef pread
842# undef pwrite
491# define pread aio_pread 843# define pread eio__pread
492# define pwrite aio_pwrite 844# define pwrite eio__pwrite
493 845
494/* 846static ssize_t
495 * make our pread/pwrite safe against themselves, but not against 847eio__pread (int fd, void *buf, size_t count, off_t offset)
496 * normal read/write by using a mutex. slows down execution a lot,
497 * but that's your problem, not mine.
498 */
499static mutex_t preadwritelock = X_MUTEX_INIT;
500
501static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
502{ 848{
503 ssize_t res; 849 ssize_t res;
504 off_t ooffset; 850 off_t ooffset;
505 851
506 X_LOCK (preadwritelock); 852 X_LOCK (preadwritelock);
… …
511 X_UNLOCK (preadwritelock); 857 X_UNLOCK (preadwritelock);
512 858
513 return res; 859 return res;
514} 860}
515 861
516static 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)
517{ 864{
518 ssize_t res; 865 ssize_t res;
519 off_t ooffset; 866 off_t ooffset;
520 867
521 X_LOCK (preadwritelock); 868 X_LOCK (preadwritelock);
522 ooffset = lseek (fd, 0, SEEK_CUR); 869 ooffset = lseek (fd, 0, SEEK_CUR);
523 lseek (fd, offset, SEEK_SET); 870 lseek (fd, offset, SEEK_SET);
524 res = write (fd, buf, count); 871 res = write (fd, buf, count);
525 lseek (fd, offset, SEEK_SET); 872 lseek (fd, ooffset, SEEK_SET);
526 X_UNLOCK (preadwritelock); 873 X_UNLOCK (preadwritelock);
527 874
528 return res; 875 return res;
529} 876}
530#endif 877#endif
531 878
532#ifndef HAVE_FUTIMES 879#ifndef HAVE_UTIMES
533 880
881# undef utimes
534# define utimes(path,times) aio_utimes (path, times) 882# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times)
536 883
884static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 885eio__utimes (const char *filename, const struct timeval times[2])
538{ 886{
539 if (times) 887 if (times)
540 { 888 {
541 struct utimbuf buf; 889 struct utimbuf buf;
542 890
… …
547 } 895 }
548 else 896 else
549 return utime (filename, 0); 897 return utime (filename, 0);
550} 898}
551 899
900#endif
901
902#ifndef HAVE_FUTIMES
903
904# undef futimes
905# define futimes(fd,times) eio__futimes (fd, times)
906
907static int
552static int aio_futimes (int fd, const struct timeval tv[2]) 908eio__futimes (int fd, const struct timeval tv[2])
553{ 909{
554 errno = ENOSYS; 910 errno = ENOSYS;
555 return -1; 911 return -1;
556} 912}
557 913
558#endif 914#endif
559 915
560#if !HAVE_FDATASYNC 916#if !HAVE_FDATASYNC
917# undef fdatasync
561# define fdatasync fsync 918# define fdatasync(fd) fsync (fd)
562#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}
563 948
564#if !HAVE_READAHEAD 949#if !HAVE_READAHEAD
950# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 951# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 952
953static ssize_t
567static 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)
568{ 955{
569 size_t todo = count; 956 size_t todo = count;
570 dBUF; 957 dBUF;
571 958
572 while (todo > 0) 959 while (todo > 0)
573 { 960 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 961 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 962
576 pread (fd, aio_buf, len, offset); 963 pread (fd, eio_buf, len, offset);
577 offset += len; 964 offset += len;
578 todo -= len; 965 todo -= len;
579 } 966 }
580 967
581 errno = 0; 968 errno = 0;
582 return count; 969 return count;
583} 970}
584 971
585#endif 972#endif
586 973
587#if !HAVE_READDIR_R
588# define readdir_r aio_readdir_r
589
590static mutex_t readdirlock = X_MUTEX_INIT;
591
592static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
593{
594 X_DIRENT *e;
595 int errorno;
596
597 X_LOCK (readdirlock);
598
599 e = readdir (dirp);
600 errorno = errno;
601
602 if (e)
603 {
604 *res = ent;
605 strcpy (ent->d_name, e->d_name);
606 }
607 else
608 *res = 0;
609
610 X_UNLOCK (readdirlock);
611
612 errno = errorno;
613 return e ? 0 : -1;
614}
615#endif
616
617/* sendfile always needs emulation */ 974/* sendfile always needs emulation */
975static ssize_t
618static 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)
619{ 977{
978 ssize_t written = 0;
620 ssize_t res; 979 ssize_t res;
621 980
622 if (!count) 981 if (!count)
623 return 0; 982 return 0;
624 983
984 for (;;)
985 {
625#if HAVE_SENDFILE 986#if HAVE_SENDFILE
626# if __linux 987# if __linux
988 off_t soffset = offset;
627 res = sendfile (ofd, ifd, &offset, count); 989 res = sendfile (ofd, ifd, &soffset, count);
628 990
629# elif __freebsd 991# elif __FreeBSD__
630 /* 992 /*
631 * 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
632 * wasted on making it similar to other I/O functions. 994 * wasted on making it similar to other I/O functions.
633 */ 995 */
634 {
635 off_t sbytes; 996 off_t sbytes;
636 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 997 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637 998
638 if (res < 0 && sbytes) 999 #if 0 /* according to the manpage, this is correct, but broken behaviour */
639 /* 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)
640 res = sbytes; 1004 res = sbytes;
641 } 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;
642 1018
643# elif __hpux 1019# elif __hpux
644 res = sendfile (ofd, ifd, offset, count, 0, 0); 1020 res = sendfile (ofd, ifd, offset, count, 0, 0);
645 1021
646# elif __solaris 1022# elif __solaris
647 {
648 struct sendfilevec vec; 1023 struct sendfilevec vec;
649 size_t sbytes; 1024 size_t sbytes;
650 1025
651 vec.sfv_fd = ifd; 1026 vec.sfv_fd = ifd;
652 vec.sfv_flag = 0; 1027 vec.sfv_flag = 0;
653 vec.sfv_off = offset; 1028 vec.sfv_off = offset;
654 vec.sfv_len = count; 1029 vec.sfv_len = count;
655 1030
656 res = sendfilev (ofd, &vec, 1, &sbytes); 1031 res = sendfilev (ofd, &vec, 1, &sbytes);
657 1032
658 if (res < 0 && sbytes) 1033 if (res < 0 && sbytes)
659 res = sbytes; 1034 res = sbytes;
660 }
661 1035
662# 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
663#else 1047#else
664 res = -1; 1048 res = -1;
665 errno = ENOSYS; 1049 errno = ENOSYS;
666#endif 1050#endif
667 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
668 if (res < 0 1075 if (res < 0
669 && (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 */
670#if __solaris 1082#if __solaris
671 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1083 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
672#endif 1084#endif
673 ) 1085 )
674 ) 1086 )
… …
705 } 1117 }
706 1118
707 return res; 1119 return res;
708} 1120}
709 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 symlinks = SYMLOOP_MAX;
1249#else
1250 int symlinks = 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 0 /* disabled, the musl way to do things is just too racy */
1280#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1281 /* on linux we may be able to ask the kernel */
1282 {
1283 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1284
1285 if (fd >= 0)
1286 {
1287 sprintf (tmp1, "/proc/self/fd/%d", fd);
1288 req->result = readlink (tmp1, res, PATH_MAX);
1289 close (fd);
1290
1291 /* here we should probably stat the open file and the disk file, to make sure they still match */
1292
1293 if (req->result > 0)
1294 goto done;
1295 }
1296 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1297 return;
1298 }
1299#endif
1300#endif
1301
1302 if (*rel != '/')
1303 {
1304 if (!getcwd (res, PATH_MAX))
1305 return;
1306
1307 if (res [1]) /* only use if not / */
1308 res += strlen (res);
1309 }
1310
1311 while (*rel)
1312 {
1313 ssize_t len, linklen;
1314 char *beg = rel;
1315
1316 while (*rel && *rel != '/')
1317 ++rel;
1318
1319 len = rel - beg;
1320
1321 if (!len) /* skip slashes */
1322 {
1323 ++rel;
1324 continue;
1325 }
1326
1327 if (beg [0] == '.')
1328 {
1329 if (len == 1)
1330 continue; /* . - nop */
1331
1332 if (beg [1] == '.' && len == 2)
1333 {
1334 /* .. - back up one component, if possible */
1335
1336 while (res != req->ptr2)
1337 if (*--res == '/')
1338 break;
1339
1340 continue;
1341 }
1342 }
1343
1344 errno = ENAMETOOLONG;
1345 if (res + 1 + len + 1 >= tmp1)
1346 return;
1347
1348 /* copy one component */
1349 *res = '/';
1350 memcpy (res + 1, beg, len);
1351
1352 /* zero-terminate, for readlink */
1353 res [len + 1] = 0;
1354
1355 /* now check if it's a symlink */
1356 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1357
1358 if (linklen < 0)
1359 {
1360 if (errno != EINVAL)
1361 return;
1362
1363 /* it's a normal directory. hopefully */
1364 res += len + 1;
1365 }
1366 else
1367 {
1368 /* yay, it was a symlink - build new path in tmp2 */
1369 int rellen = strlen (rel);
1370
1371 errno = ENAMETOOLONG;
1372 if (linklen + 1 + rellen >= PATH_MAX)
1373 return;
1374
1375 errno = ELOOP;
1376 if (!--symlinks)
1377 return;
1378
1379 if (*tmp1 == '/')
1380 res = req->ptr2; /* symlink resolves to an absolute path */
1381
1382 /* we need to be careful, as rel might point into tmp2 already */
1383 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1384 tmp2 [linklen] = '/';
1385 memcpy (tmp2, tmp1, linklen);
1386
1387 rel = tmp2;
1388 }
1389 }
1390
1391 /* special case for the lone root path */
1392 if (res == req->ptr2)
1393 *res++ = '/';
1394
1395 req->result = res - (char *)req->ptr2;
1396
1397done:
1398 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1399}
1400
1401static signed char
1402eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1403{
1404 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1405 : a->inode < b->inode ? -1
1406 : a->inode > b->inode ? 1
1407 : 0;
1408}
1409
1410#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1411
1412#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1413#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1414
1415static void
1416eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1417{
1418 unsigned char bits [9 + sizeof (ino_t) * 8];
1419 unsigned char *bit = bits;
1420
1421 assert (CHAR_BIT == 8);
1422 assert (sizeof (eio_dirent) * 8 < 256);
1423 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1424 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1425
1426 if (size <= EIO_SORT_FAST)
1427 return;
1428
1429 /* first prepare an array of bits to test in our radix sort */
1430 /* try to take endianness into account, as well as differences in ino_t sizes */
1431 /* inode_bits must contain all inodes ORed together */
1432 /* which is used to skip bits that are 0 everywhere, which is very common */
1433 {
1434 ino_t endianness;
1435 int i, j;
1436
1437 /* we store the byte offset of byte n into byte n of "endianness" */
1438 for (i = 0; i < sizeof (ino_t); ++i)
1439 ((unsigned char *)&endianness)[i] = i;
1440
1441 *bit++ = 0;
1442
1443 for (i = 0; i < sizeof (ino_t); ++i)
1444 {
1445 /* shifting off the byte offsets out of "endianness" */
1446 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1447 endianness >>= 8;
1448
1449 for (j = 0; j < 8; ++j)
1450 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1451 *bit++ = offs + j;
1452 }
1453
1454 for (j = 0; j < 8; ++j)
1455 if (score_bits & (1 << j))
1456 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1457 }
1458
1459 /* now actually do the sorting (a variant of MSD radix sort) */
1460 {
1461 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1462 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1463 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1464 int stk_idx = 0;
1465
1466 base_stk [stk_idx] = dents;
1467 end_stk [stk_idx] = dents + size;
1468 bit_stk [stk_idx] = bit - 1;
1469
1470 do
1471 {
1472 base = base_stk [stk_idx];
1473 end = end_stk [stk_idx];
1474 bit = bit_stk [stk_idx];
1475
1476 for (;;)
1477 {
1478 unsigned char O = *bit >> 3;
1479 unsigned char M = 1 << (*bit & 7);
1480
1481 eio_dirent *a = base;
1482 eio_dirent *b = end;
1483
1484 if (b - a < EIO_SORT_CUTOFF)
1485 break;
1486
1487 /* now bit-partition the array on the bit */
1488 /* this ugly asymmetric loop seems to perform much better than typical */
1489 /* partition algos found in the literature */
1490 do
1491 if (!(((unsigned char *)a)[O] & M))
1492 ++a;
1493 else if (!(((unsigned char *)--b)[O] & M))
1494 {
1495 eio_dirent tmp = *a; *a = *b; *b = tmp;
1496 ++a;
1497 }
1498 while (b > a);
1499
1500 /* next bit, or stop, if no bits left in this path */
1501 if (!*--bit)
1502 break;
1503
1504 base_stk [stk_idx] = a;
1505 end_stk [stk_idx] = end;
1506 bit_stk [stk_idx] = bit;
1507 ++stk_idx;
1508
1509 end = a;
1510 }
1511 }
1512 while (stk_idx--);
1513 }
1514}
1515
1516static void
1517eio_dent_insertion_sort (eio_dirent *dents, int size)
1518{
1519 /* first move the smallest element to the front, to act as a sentinel */
1520 {
1521 int i;
1522 eio_dirent *min = dents;
1523
1524 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1525 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1526 if (EIO_DENT_CMP (dents [i], <, *min))
1527 min = &dents [i];
1528
1529 /* swap elements 0 and j (minimum) */
1530 {
1531 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1532 }
1533 }
1534
1535 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1536 {
1537 eio_dirent *i, *j;
1538
1539 for (i = dents + 1; i < dents + size; ++i)
1540 {
1541 eio_dirent value = *i;
1542
1543 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1544 j [1] = j [0];
1545
1546 j [1] = value;
1547 }
1548 }
1549}
1550
1551static void
1552eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1553{
1554 if (size <= 1)
1555 return; /* our insertion sort relies on size > 0 */
1556
1557 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1558 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1559 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1560
1561 /* use an insertion sort at the end, or for small arrays, */
1562 /* as insertion sort is more efficient for small partitions */
1563 eio_dent_insertion_sort (dents, size);
1564}
1565
710/* read a full directory */ 1566/* read a full directory */
1567static void
711static void scandir_ (eio_req *req, worker *self) 1568eio__scandir (eio_req *req, etp_worker *self)
712{ 1569{
713 DIR *dirp; 1570 DIR *dirp;
714 union
715 {
716 EIO_STRUCT_DIRENT d;
717 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
718 } *u;
719 EIO_STRUCT_DIRENT *entp; 1571 EIO_STRUCT_DIRENT *entp;
720 char *name, *names; 1572 char *name, *names;
721 int memlen = 4096; 1573 int namesalloc = 4096;
722 int memofs = 0; 1574 int namesoffs = 0;
1575 int flags = req->int1;
1576 eio_dirent *dents = 0;
1577 int dentalloc = 128;
723 int res = 0; 1578 int dentoffs = 0;
1579 ino_t inode_bits = 0;
1580
1581 req->result = -1;
1582
1583 if (!(flags & EIO_READDIR_DENTS))
1584 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
724 1585
725 X_LOCK (wrklock); 1586 X_LOCK (wrklock);
1587 /* the corresponding closedir is in ETP_WORKER_CLEAR */
726 self->dirp = dirp = opendir (req->ptr1); 1588 self->dirp = dirp = opendir (req->ptr1);
727 self->dbuf = u = malloc (sizeof (*u)); 1589
728 req->flags |= EIO_FLAG_PTR2_FREE; 1590 if (req->flags & EIO_FLAG_PTR1_FREE)
1591 free (req->ptr1);
1592
1593 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1594 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
729 req->ptr2 = names = malloc (memlen); 1595 req->ptr2 = names = malloc (namesalloc);
730 X_UNLOCK (wrklock); 1596 X_UNLOCK (wrklock);
731 1597
732 if (dirp && u && names) 1598 if (dirp && names && (!flags || dents))
733 for (;;) 1599 for (;;)
734 { 1600 {
735 errno = 0; 1601 errno = 0;
736 readdir_r (dirp, &u->d, &entp); 1602 entp = readdir (dirp);
737 1603
738 if (!entp) 1604 if (!entp)
1605 {
1606 if (errno)
1607 break;
1608
1609 /* sort etc. */
1610 req->int1 = flags;
1611 req->result = dentoffs;
1612
1613 if (flags & EIO_READDIR_STAT_ORDER)
1614 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1615 else if (flags & EIO_READDIR_DIRS_FIRST)
1616 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1617 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1618 else
1619 {
1620 /* in this case, all is known, and we just put dirs first and sort them */
1621 eio_dirent *oth = dents + dentoffs;
1622 eio_dirent *dir = dents;
1623
1624 /* now partition dirs to the front, and non-dirs to the back */
1625 /* by walking from both sides and swapping if necessary */
1626 while (oth > dir)
1627 {
1628 if (dir->type == EIO_DT_DIR)
1629 ++dir;
1630 else if ((--oth)->type == EIO_DT_DIR)
1631 {
1632 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1633
1634 ++dir;
1635 }
1636 }
1637
1638 /* now sort the dirs only (dirs all have the same score) */
1639 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1640 }
1641
739 break; 1642 break;
1643 }
740 1644
1645 /* now add the entry to our list(s) */
741 name = entp->d_name; 1646 name = entp->d_name;
742 1647
1648 /* skip . and .. entries */
743 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1649 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
744 { 1650 {
745 int len = strlen (name) + 1; 1651 int len = D_NAMLEN (entp) + 1;
746 1652
747 res++; 1653 while (ecb_expect_false (namesoffs + len > namesalloc))
748
749 while (memofs + len > memlen)
750 { 1654 {
751 memlen *= 2; 1655 namesalloc *= 2;
752 X_LOCK (wrklock); 1656 X_LOCK (wrklock);
753 req->ptr2 = names = realloc (names, memlen); 1657 req->ptr2 = names = realloc (names, namesalloc);
754 X_UNLOCK (wrklock); 1658 X_UNLOCK (wrklock);
755 1659
756 if (!names) 1660 if (!names)
757 break; 1661 break;
758 } 1662 }
759 1663
760 memcpy (names + memofs, name, len); 1664 memcpy (names + namesoffs, name, len);
1665
1666 if (dents)
1667 {
1668 struct eio_dirent *ent;
1669
1670 if (ecb_expect_false (dentoffs == dentalloc))
1671 {
1672 dentalloc *= 2;
1673 X_LOCK (wrklock);
1674 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1675 X_UNLOCK (wrklock);
1676
1677 if (!dents)
1678 break;
1679 }
1680
1681 ent = dents + dentoffs;
1682
1683 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1684 ent->namelen = len - 1;
1685 ent->inode = D_INO (entp);
1686
1687 inode_bits |= ent->inode;
1688
1689 switch (D_TYPE (entp))
1690 {
1691 default:
1692 ent->type = EIO_DT_UNKNOWN;
1693 flags |= EIO_READDIR_FOUND_UNKNOWN;
1694 break;
1695
1696 #ifdef DT_FIFO
1697 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1698 #endif
1699 #ifdef DT_CHR
1700 case DT_CHR: ent->type = EIO_DT_CHR; break;
1701 #endif
1702 #ifdef DT_MPC
1703 case DT_MPC: ent->type = EIO_DT_MPC; break;
1704 #endif
1705 #ifdef DT_DIR
1706 case DT_DIR: ent->type = EIO_DT_DIR; break;
1707 #endif
1708 #ifdef DT_NAM
1709 case DT_NAM: ent->type = EIO_DT_NAM; break;
1710 #endif
1711 #ifdef DT_BLK
1712 case DT_BLK: ent->type = EIO_DT_BLK; break;
1713 #endif
1714 #ifdef DT_MPB
1715 case DT_MPB: ent->type = EIO_DT_MPB; break;
1716 #endif
1717 #ifdef DT_REG
1718 case DT_REG: ent->type = EIO_DT_REG; break;
1719 #endif
1720 #ifdef DT_NWK
1721 case DT_NWK: ent->type = EIO_DT_NWK; break;
1722 #endif
1723 #ifdef DT_CMP
1724 case DT_CMP: ent->type = EIO_DT_CMP; break;
1725 #endif
1726 #ifdef DT_LNK
1727 case DT_LNK: ent->type = EIO_DT_LNK; break;
1728 #endif
1729 #ifdef DT_SOCK
1730 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1731 #endif
1732 #ifdef DT_DOOR
1733 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1734 #endif
1735 #ifdef DT_WHT
1736 case DT_WHT: ent->type = EIO_DT_WHT; break;
1737 #endif
1738 }
1739
1740 ent->score = 7;
1741
1742 if (flags & EIO_READDIR_DIRS_FIRST)
1743 {
1744 if (ent->type == EIO_DT_UNKNOWN)
1745 {
1746 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1747 ent->score = 1;
1748 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1749 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1750 }
1751 else if (ent->type == EIO_DT_DIR)
1752 ent->score = 0;
1753 }
1754 }
1755
761 memofs += len; 1756 namesoffs += len;
1757 ++dentoffs;
1758 }
1759
1760 if (EIO_CANCELLED (req))
1761 {
1762 errno = ECANCELED;
1763 break;
762 } 1764 }
763 } 1765 }
764
765 if (errno)
766 res = -1;
767
768 req->result = res;
769} 1766}
770 1767
771/*****************************************************************************/ 1768/*****************************************************************************/
772 1769
1770#define ALLOC(len) \
1771 if (!req->ptr2) \
1772 { \
1773 X_LOCK (wrklock); \
1774 req->flags |= EIO_FLAG_PTR2_FREE; \
1775 X_UNLOCK (wrklock); \
1776 req->ptr2 = malloc (len); \
1777 if (!req->ptr2) \
1778 { \
1779 errno = ENOMEM; \
1780 req->result = -1; \
1781 break; \
1782 } \
1783 }
1784
773X_THREAD_PROC (eio_proc) 1785X_THREAD_PROC (etp_proc)
774{ 1786{
775 eio_req *req; 1787 ETP_REQ *req;
776 struct timespec ts; 1788 struct timespec ts;
777 worker *self = (worker *)thr_arg; 1789 etp_worker *self = (etp_worker *)thr_arg;
778 1790
779 /* try to distribute timeouts somewhat randomly */ 1791 /* try to distribute timeouts somewhat randomly */
780 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1792 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781 1793
782 for (;;) 1794 for (;;)
783 { 1795 {
784 ts.tv_sec = time (0) + IDLE_TIMEOUT;
785
786 X_LOCK (reqlock); 1796 X_LOCK (reqlock);
787 1797
788 for (;;) 1798 for (;;)
789 { 1799 {
790 self->req = req = reqq_shift (&req_queue); 1800 self->req = req = reqq_shift (&req_queue);
… …
792 if (req) 1802 if (req)
793 break; 1803 break;
794 1804
795 ++idle; 1805 ++idle;
796 1806
1807 ts.tv_sec = time (0) + idle_timeout;
797 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1808 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
798 { 1809 {
799 if (idle > max_idle) 1810 if (idle > max_idle)
800 { 1811 {
801 --idle; 1812 --idle;
… …
806 goto quit; 1817 goto quit;
807 } 1818 }
808 1819
809 /* we are allowed to idle, so do so without any timeout */ 1820 /* we are allowed to idle, so do so without any timeout */
810 X_COND_WAIT (reqwait, reqlock); 1821 X_COND_WAIT (reqwait, reqlock);
811 ts.tv_sec = time (0) + IDLE_TIMEOUT;
812 } 1822 }
813 1823
814 --idle; 1824 --idle;
815 } 1825 }
816 1826
817 --nready; 1827 --nready;
818 1828
819 X_UNLOCK (reqlock); 1829 X_UNLOCK (reqlock);
820 1830
821 errno = 0; /* strictly unnecessary */ 1831 if (req->type < 0)
1832 goto quit;
822 1833
823 if (!EIO_CANCELLED (req)) 1834 if (!EIO_CANCELLED (req))
824 switch (req->type) 1835 ETP_EXECUTE (self, req);
825 {
826 case EIO_READ: req->result = req->offs >= 0
827 ? pread (req->int1, req->ptr2, req->size, req->offs)
828 : read (req->int1, req->ptr2, req->size); break;
829 case EIO_WRITE: req->result = req->offs >= 0
830 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
831 : write (req->int1, req->ptr2, req->size); break;
832
833 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
834 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
835
836 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
837 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
838 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
839
840 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
841 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
842 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
843 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
844 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
845 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
846
847 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
848 case EIO_CLOSE: req->result = close (req->int1); break;
849 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
850 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
851 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
852 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
853 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
854 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
855 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
856 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
857 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
858
859 case EIO_SYNC: req->result = 0; sync (); break;
860 case EIO_FSYNC: req->result = fsync (req->int1); break;
861 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
862
863 case EIO_READDIR: scandir_ (req, self); break;
864
865 case EIO_BUSY:
866#ifdef _WIN32
867 Sleep (req->nv1 * 1000.);
868#else
869 {
870 struct timeval tv;
871
872 tv.tv_sec = req->nv1;
873 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
874
875 req->result = select (0, 0, 0, 0, &tv);
876 }
877#endif
878 break;
879
880 case EIO_UTIME:
881 case EIO_FUTIME:
882 {
883 struct timeval tv[2];
884 struct timeval *times;
885
886 if (req->nv1 != -1. || req->nv2 != -1.)
887 {
888 tv[0].tv_sec = req->nv1;
889 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
890 tv[1].tv_sec = req->nv2;
891 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
892
893 times = tv;
894 }
895 else
896 times = 0;
897
898
899 req->result = req->type == EIO_FUTIME
900 ? futimes (req->int1, times)
901 : utimes (req->ptr1, times);
902 }
903
904 case EIO_GROUP:
905 case EIO_NOP:
906 break;
907
908 case EIO_QUIT:
909 goto quit;
910
911 default:
912 req->result = -1;
913 break;
914 }
915
916 req->errorno = errno;
917 1836
918 X_LOCK (reslock); 1837 X_LOCK (reslock);
919 1838
920 ++npending; 1839 ++npending;
921 1840
922 if (!reqq_push (&res_queue, req) && want_poll_cb) 1841 if (!reqq_push (&res_queue, req) && want_poll_cb)
923 want_poll_cb (); 1842 want_poll_cb ();
924 1843
925 self->req = 0; 1844 self->req = 0;
926 worker_clear (self); 1845 etp_worker_clear (self);
927 1846
928 X_UNLOCK (reslock); 1847 X_UNLOCK (reslock);
929 } 1848 }
930 1849
931quit: 1850quit:
932 X_LOCK (wrklock); 1851 X_LOCK (wrklock);
933 worker_free (self); 1852 etp_worker_free (self);
934 X_UNLOCK (wrklock); 1853 X_UNLOCK (wrklock);
935 1854
936 return 0; 1855 return 0;
937} 1856}
938 1857
939/*****************************************************************************/ 1858/*****************************************************************************/
940 1859
941static void eio_atfork_prepare (void) 1860int ecb_cold
1861eio_init (void (*want_poll)(void), void (*done_poll)(void))
942{ 1862{
943 X_LOCK (wrklock); 1863 return etp_init (want_poll, done_poll);
944 X_LOCK (reqlock);
945 X_LOCK (reslock);
946#if !HAVE_PREADWRITE
947 X_LOCK (preadwritelock);
948#endif
949#if !HAVE_READDIR_R
950 X_LOCK (readdirlock);
951#endif
952} 1864}
953 1865
954static void eio_atfork_parent (void) 1866ecb_inline void
1867eio_api_destroy (eio_req *req)
955{ 1868{
956#if !HAVE_READDIR_R 1869 free (req);
957 X_UNLOCK (readdirlock);
958#endif
959#if !HAVE_PREADWRITE
960 X_UNLOCK (preadwritelock);
961#endif
962 X_UNLOCK (reslock);
963 X_UNLOCK (reqlock);
964 X_UNLOCK (wrklock);
965} 1870}
966 1871
967static void eio_atfork_child (void) 1872#define REQ(rtype) \
968{ 1873 eio_req *req; \
969 eio_req *prv; 1874 \
1875 req = (eio_req *)calloc (1, sizeof *req); \
1876 if (!req) \
1877 return 0; \
1878 \
1879 req->type = rtype; \
1880 req->pri = pri; \
1881 req->finish = cb; \
1882 req->data = data; \
1883 req->destroy = eio_api_destroy;
970 1884
971 while (prv = reqq_shift (&req_queue)) 1885#define SEND eio_submit (req); return req
972 eio_destroy (prv);
973 1886
974 while (prv = reqq_shift (&res_queue)) 1887#define PATH \
975 eio_destroy (prv); 1888 req->flags |= EIO_FLAG_PTR1_FREE; \
1889 req->ptr1 = strdup (path); \
1890 if (!req->ptr1) \
1891 { \
1892 eio_api_destroy (req); \
1893 return 0; \
1894 }
976 1895
977 while (wrk_first.next != &wrk_first) 1896static void
1897eio_execute (etp_worker *self, eio_req *req)
1898{
1899 switch (req->type)
978 { 1900 {
979 worker *wrk = wrk_first.next; 1901 case EIO_READ: ALLOC (req->size);
1902 req->result = req->offs >= 0
1903 ? pread (req->int1, req->ptr2, req->size, req->offs)
1904 : read (req->int1, req->ptr2, req->size); break;
1905 case EIO_WRITE: req->result = req->offs >= 0
1906 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1907 : write (req->int1, req->ptr2, req->size); break;
980 1908
981 if (wrk->req) 1909 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
982 eio_destroy (wrk->req); 1910 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
983 1911
984 worker_clear (wrk); 1912 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
985 worker_free (wrk); 1913 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
986 } 1914 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1915 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1916 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1917 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
987 1918
988 started = 0; 1919 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
989 idle = 0; 1920 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
990 nreqs = 0; 1921 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
991 nready = 0; 1922 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
992 npending = 0;
993 1923
994 eio_atfork_parent (); 1924 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
995} 1925 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1926 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1927 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1928 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1929 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
996 1930
997int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1931 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
998{ 1932 case EIO_CLOSE: req->result = close (req->int1); break;
999 want_poll_cb = want_poll; 1933 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1000 done_poll_cb = done_poll; 1934 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1935 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1936 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1937 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1938 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1939 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1940 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1001 1941
1942 case EIO_REALPATH: eio__realpath (req, self); break;
1943
1944 case EIO_READLINK: ALLOC (PATH_MAX);
1945 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1946
1947 case EIO_SYNC: req->result = 0; sync (); break;
1948 case EIO_FSYNC: req->result = fsync (req->int1); break;
1949 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1950 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1951 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1952 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1953 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1954 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1955
1956 case EIO_READDIR: eio__scandir (req, self); break;
1957
1958 case EIO_BUSY:
1002#ifdef _WIN32 1959#ifdef _WIN32
1003 X_MUTEX_CHECK (wrklock); 1960 Sleep (req->nv1 * 1e3);
1004 X_MUTEX_CHECK (reslock); 1961#else
1005 X_MUTEX_CHECK (reqlock); 1962 {
1006 X_MUTEX_CHECK (reqwait); 1963 struct timeval tv;
1007 X_MUTEX_CHECK (preadwritelock);
1008 X_MUTEX_CHECK (readdirlock);
1009 1964
1010 X_COND_CHECK (reqwait); 1965 tv.tv_sec = req->nv1;
1011#endif 1966 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1012 1967
1013 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child); 1968 req->result = select (0, 0, 0, 0, &tv);
1014} 1969 }
1970#endif
1971 break;
1015 1972
1016#if 0 1973 case EIO_UTIME:
1974 case EIO_FUTIME:
1975 {
1976 struct timeval tv[2];
1977 struct timeval *times;
1017 1978
1979 if (req->nv1 != -1. || req->nv2 != -1.)
1980 {
1981 tv[0].tv_sec = req->nv1;
1982 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1983 tv[1].tv_sec = req->nv2;
1984 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1985
1986 times = tv;
1987 }
1988 else
1989 times = 0;
1990
1991 req->result = req->type == EIO_FUTIME
1992 ? futimes (req->int1, times)
1993 : utimes (req->ptr1, times);
1994 }
1995 break;
1996
1997 case EIO_GROUP:
1998 abort (); /* handled in eio_request */
1999
2000 case EIO_NOP:
2001 req->result = 0;
2002 break;
2003
2004 case EIO_CUSTOM:
2005 req->feed (req);
2006 break;
2007
2008 default:
2009 errno = ENOSYS;
2010 req->result = -1;
2011 break;
2012 }
2013
2014 req->errorno = errno;
2015}
2016
2017#ifndef EIO_NO_WRAPPERS
2018
2019eio_req *eio_nop (int pri, eio_cb cb, void *data)
2020{
2021 REQ (EIO_NOP); SEND;
2022}
2023
2024eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
2025{
2026 REQ (EIO_BUSY); req->nv1 = delay; SEND;
2027}
2028
1018eio_req *eio_fsync (int fd, eio_cb cb); 2029eio_req *eio_sync (int pri, eio_cb cb, void *data)
2030{
2031 REQ (EIO_SYNC); SEND;
2032}
2033
2034eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
2035{
2036 REQ (EIO_FSYNC); req->int1 = fd; SEND;
2037}
2038
2039eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2040{
2041 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2042}
2043
2044eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2045{
2046 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2047}
2048
2049eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2050{
2051 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2052}
2053
2054eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2055{
2056 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2057}
2058
2059eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2060{
2061 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2062}
2063
1019eio_req *eio_fdatasync (int fd, eio_cb cb); 2064eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1020eio_req *eio_dupclose (int fd, eio_cb cb); 2065{
2066 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2067}
2068
2069eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2070{
2071 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2072}
2073
1021eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb); 2074eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1022eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2075{
1023eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb); 2076 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1024eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2077}
2078
2079eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2080{
2081 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2082}
2083
2084eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2085{
2086 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2087}
2088
2089eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2090{
2091 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2092}
2093
2094eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2095{
2096 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2097}
2098
1025eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb); 2099eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
2100{
2101 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
2102}
2103
1026eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb); 2104eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
2105{
2106 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
2107}
2108
1027eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb); 2109eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2110{
2111 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2112}
2113
1028eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb); 2114eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
2115{
2116 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2117}
2118
1029eio_req *eio_dup2 (int fd, int fd2, eio_cb cb); 2119eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2120{
2121 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
2122}
2123
1030eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb); 2124eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
2125{
2126 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
2127}
2128
1031eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb); 2129eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1032eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2130{
1033eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2131 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1034eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */ 2132}
2133
1035eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb); 2134eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
2135{
2136 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
2137}
2138
1036eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb); 2139eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1037eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb); 2140{
1038eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb); 2141 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2142}
2143
1039eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb); 2144eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
2145{
2146 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2147}
2148
2149eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2150{
2151 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
2152}
2153
2154eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2155{
2156 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
2157}
2158
2159static eio_req *
2160eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
2161{
2162 REQ (type); PATH; SEND;
2163}
2164
2165eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2166{
2167 return eio__1path (EIO_READLINK, path, pri, cb, data);
2168}
2169
2170eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2171{
2172 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2173}
2174
2175eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2176{
2177 return eio__1path (EIO_STAT, path, pri, cb, data);
2178}
2179
2180eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
2181{
2182 return eio__1path (EIO_LSTAT, path, pri, cb, data);
2183}
2184
2185eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2186{
2187 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2188}
2189
1040eio_req *eio_unlink (const char *path, eio_cb cb); 2190eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
2191{
2192 return eio__1path (EIO_UNLINK, path, pri, cb, data);
2193}
2194
1041eio_req *eio_rmdir (const char *path, eio_cb cb); 2195eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1042eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */ 2196{
2197 return eio__1path (EIO_RMDIR, path, pri, cb, data);
2198}
2199
2200eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
2201{
2202 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
2203}
2204
1043eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb); 2205eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1044eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */ 2206{
1045eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */ 2207 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2208}
2209
2210static eio_req *
2211eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2212{
2213 REQ (type); PATH;
2214
2215 req->flags |= EIO_FLAG_PTR2_FREE;
2216 req->ptr2 = strdup (new_path);
2217 if (!req->ptr2)
2218 {
2219 eio_api_destroy (req);
2220 return 0;
2221 }
2222
2223 SEND;
2224}
2225
2226eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2227{
2228 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
2229}
2230
2231eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2232{
2233 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
2234}
2235
2236eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2237{
2238 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2239}
2240
2241eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2242{
2243 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2244}
2245
2246#endif
2247
2248eio_req *eio_grp (eio_cb cb, void *data)
2249{
2250 const int pri = EIO_PRI_MAX;
2251
2252 REQ (EIO_GROUP); SEND;
2253}
2254
2255#undef REQ
2256#undef PATH
2257#undef SEND
2258
2259/*****************************************************************************/
2260/* grp functions */
2261
1046void 2262void
1047aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1048 PROTOTYPE: $$$;$
1049 PPCODE:
1050{
1051 dREQ;
1052
1053 req->type = EIO_OPEN;
1054 req->sv1 = newSVsv (pathname);
1055 req->ptr1 = SvPVbyte_nolen (req->sv1);
1056 req->int1 = flags;
1057 req->int2 = mode;
1058
1059 EIO_SEND;
1060}
1061
1062void
1063aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1064 PROTOTYPE: $;$
1065 ALIAS:
1066 aio_fsync = EIO_FSYNC
1067 aio_fdatasync = EIO_FDATASYNC
1068 PPCODE:
1069{
1070 dREQ;
1071
1072 req->type = ix;
1073 req->sv1 = newSVsv (fh);
1074 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1075
1076 EIO_SEND (req);
1077}
1078
1079void
1080aio_close (SV *fh, SV *callback=&PL_sv_undef)
1081 PROTOTYPE: $;$
1082 PPCODE:
1083{
1084 dREQ;
1085
1086 req->type = EIO_CLOSE;
1087 req->sv1 = newSVsv (fh);
1088 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1089
1090 EIO_SEND (req);
1091}
1092
1093void
1094aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1095 ALIAS:
1096 aio_read = EIO_READ
1097 aio_write = EIO_WRITE
1098 PROTOTYPE: $$$$$;$
1099 PPCODE:
1100{
1101 STRLEN svlen;
1102 char *svptr = SvPVbyte (data, svlen);
1103 UV len = SvUV (length);
1104
1105 SvUPGRADE (data, SVt_PV);
1106 SvPOK_on (data);
1107
1108 if (dataoffset < 0)
1109 dataoffset += svlen;
1110
1111 if (dataoffset < 0 || dataoffset > svlen)
1112 croak ("dataoffset outside of data scalar");
1113
1114 if (ix == EIO_WRITE)
1115 {
1116 /* write: check length and adjust. */
1117 if (!SvOK (length) || len + dataoffset > svlen)
1118 len = svlen - dataoffset;
1119 }
1120 else
1121 {
1122 /* read: grow scalar as necessary */
1123 svptr = SvGROW (data, len + dataoffset + 1);
1124 }
1125
1126 if (len < 0)
1127 croak ("length must not be negative");
1128
1129 {
1130 dREQ;
1131
1132 req->type = ix;
1133 req->sv1 = newSVsv (fh);
1134 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1135 : IoOFP (sv_2io (fh)));
1136 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1137 req->size = len;
1138 req->sv2 = SvREFCNT_inc (data);
1139 req->ptr2 = (char *)svptr + dataoffset;
1140 req->stroffset = dataoffset;
1141
1142 if (!SvREADONLY (data))
1143 {
1144 SvREADONLY_on (data);
1145 req->flags |= FLAG_SV2_RO_OFF;
1146 }
1147
1148 EIO_SEND;
1149 }
1150}
1151
1152void
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{
1157 SV *data;
1158 dREQ;
1159
1160 data = newSV (NAME_MAX);
1161 SvPOK_on (data);
1162
1163 req->type = EIO_READLINK;
1164 req->sv1 = newSVsv (path);
1165 req->ptr1 = SvPVbyte_nolen (req->sv1);
1166 req->sv2 = data;
1167 req->ptr2 = SvPVbyte_nolen (data);
1168
1169 EIO_SEND;
1170}
1171
1172void
1173aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174 PROTOTYPE: $$$$;$
1175 PPCODE:
1176{
1177 dREQ;
1178
1179 req->type = EIO_SENDFILE;
1180 req->sv1 = newSVsv (out_fh);
1181 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182 req->sv2 = newSVsv (in_fh);
1183 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184 req->offs = SvVAL64 (in_offset);
1185 req->size = length;
1186
1187 EIO_SEND;
1188}
1189
1190void
1191aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192 PROTOTYPE: $$$;$
1193 PPCODE:
1194{
1195 dREQ;
1196
1197 req->type = EIO_READAHEAD;
1198 req->sv1 = newSVsv (fh);
1199 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1200 req->offs = SvVAL64 (offset);
1201 req->size = length;
1202
1203 EIO_SEND;
1204}
1205
1206void
1207aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1208 ALIAS:
1209 aio_stat = EIO_STAT
1210 aio_lstat = EIO_LSTAT
1211 PPCODE:
1212{
1213 dREQ;
1214
1215 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1216 if (!req->ptr2)
1217 {
1218 req_destroy (req);
1219 croak ("out of memory during aio_stat statdata allocation");
1220 }
1221
1222 req->flags |= FLAG_PTR2_FREE;
1223 req->sv1 = newSVsv (fh_or_path);
1224
1225 if (SvPOK (fh_or_path))
1226 {
1227 req->type = ix;
1228 req->ptr1 = SvPVbyte_nolen (req->sv1);
1229 }
1230 else
1231 {
1232 req->type = EIO_FSTAT;
1233 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1234 }
1235
1236 EIO_SEND;
1237}
1238
1239void
1240aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1241 PPCODE:
1242{
1243 dREQ;
1244
1245 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1246 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1247 req->sv1 = newSVsv (fh_or_path);
1248
1249 if (SvPOK (fh_or_path))
1250 {
1251 req->type = EIO_UTIME;
1252 req->ptr1 = SvPVbyte_nolen (req->sv1);
1253 }
1254 else
1255 {
1256 req->type = EIO_FUTIME;
1257 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1258 }
1259
1260 EIO_SEND;
1261}
1262
1263void
1264aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1265 PPCODE:
1266{
1267 dREQ;
1268
1269 req->sv1 = newSVsv (fh_or_path);
1270 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1271
1272 if (SvPOK (fh_or_path))
1273 {
1274 req->type = EIO_TRUNCATE;
1275 req->ptr1 = SvPVbyte_nolen (req->sv1);
1276 }
1277 else
1278 {
1279 req->type = EIO_FTRUNCATE;
1280 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1281 }
1282
1283 EIO_SEND;
1284}
1285
1286void
1287aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1288 ALIAS:
1289 aio_chmod = EIO_CHMOD
1290 aio_fchmod = EIO_FCHMOD
1291 aio_mkdir = EIO_MKDIR
1292 PPCODE:
1293{
1294 dREQ;
1295
1296 req->type = type;
1297 req->int2 = mode;
1298 req->sv1 = newSVsv (fh_or_path);
1299
1300 if (ix == EIO_FCHMOD)
1301 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1302 else
1303 req->ptr1 = SvPVbyte_nolen (req->sv1);
1304
1305 EIO_SEND;
1306}
1307
1308void
1309aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1310 PPCODE:
1311{
1312 dREQ;
1313
1314 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1315 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1316 req->sv1 = newSVsv (fh_or_path);
1317
1318 if (SvPOK (fh_or_path))
1319 {
1320 req->type = EIO_CHOWN;
1321 req->ptr1 = SvPVbyte_nolen (req->sv1);
1322 }
1323 else
1324 {
1325 req->type = EIO_FCHOWN;
1326 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1327 }
1328
1329 EIO_SEND;
1330}
1331
1332void
1333aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1334 ALIAS:
1335 aio_unlink = EIO_UNLINK
1336 aio_rmdir = EIO_RMDIR
1337 aio_readdir = EIO_READDIR
1338 PPCODE:
1339{
1340 dREQ;
1341
1342 req->type = ix;
1343 req->sv1 = newSVsv (pathname);
1344 req->ptr1 = SvPVbyte_nolen (req->sv1);
1345
1346 EIO_SEND;
1347}
1348
1349void
1350aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1351 ALIAS:
1352 aio_link = EIO_LINK
1353 aio_symlink = EIO_SYMLINK
1354 aio_rename = EIO_RENAME
1355 PPCODE:
1356{
1357 dREQ;
1358
1359 req->type = ix;
1360 req->sv1 = newSVsv (oldpath);
1361 req->ptr1 = SvPVbyte_nolen (req->sv1);
1362 req->sv2 = newSVsv (newpath);
1363 req->ptr2 = SvPVbyte_nolen (req->sv2);
1364
1365 EIO_SEND;
1366}
1367
1368void
1369aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1370 PPCODE:
1371{
1372 dREQ;
1373
1374 req->type = EIO_MKNOD;
1375 req->sv1 = newSVsv (pathname);
1376 req->ptr1 = SvPVbyte_nolen (req->sv1);
1377 req->int2 = (mode_t)mode;
1378 req->offs = dev;
1379
1380 EIO_SEND;
1381}
1382
1383void
1384aio_busy (double delay, SV *callback=&PL_sv_undef)
1385 PPCODE:
1386{
1387 dREQ;
1388
1389 req->type = EIO_BUSY;
1390 req->nv1 = delay < 0. ? 0. : delay;
1391
1392 EIO_SEND;
1393}
1394
1395void
1396aio_group (SV *callback=&PL_sv_undef)
1397 PROTOTYPE: ;$
1398 PPCODE:
1399{
1400 dREQ;
1401
1402 req->type = EIO_GROUP;
1403
1404 req_send (req);
1405 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1406}
1407
1408void
1409aio_nop (SV *callback=&PL_sv_undef)
1410 ALIAS:
1411 aio_nop = EIO_NOP
1412 aio_sync = EIO_SYNC
1413 PPCODE:
1414{
1415 dREQ;
1416
1417#endif
1418
1419void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2263eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 2264{
1421 grp->int2 = limit; 2265 grp->int2 = limit;
1422 grp->feed = feed; 2266 grp->feed = feed;
1423 2267
1424 grp_try_feed (grp); 2268 grp_try_feed (grp);
1425} 2269}
1426 2270
2271void
1427void eio_grp_limit (eio_req *grp, int limit) 2272eio_grp_limit (eio_req *grp, int limit)
1428{ 2273{
1429 grp->int2 = limit; 2274 grp->int2 = limit;
1430 2275
1431 grp_try_feed (grp); 2276 grp_try_feed (grp);
1432} 2277}
1433 2278
2279void
1434void eio_grp_add (eio_req *grp, eio_req *req) 2280eio_grp_add (eio_req *grp, eio_req *req)
1435{ 2281{
1436 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2282 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2283
2284 grp->flags |= EIO_FLAG_GROUPADD;
1437 2285
1438 ++grp->size; 2286 ++grp->size;
1439 req->grp = grp; 2287 req->grp = grp;
1440 2288
1441 req->grp_prev = 0; 2289 req->grp_prev = 0;
… …
1445 grp->grp_first->grp_prev = req; 2293 grp->grp_first->grp_prev = req;
1446 2294
1447 grp->grp_first = req; 2295 grp->grp_first = req;
1448} 2296}
1449 2297
2298/*****************************************************************************/
2299/* misc garbage */
1450 2300
2301ssize_t
2302eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2303{
2304 etp_worker wrk;
2305 ssize_t ret;
2306
2307 wrk.dbuf = 0;
2308
2309 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2310
2311 if (wrk.dbuf)
2312 free (wrk.dbuf);
2313
2314 return ret;
2315}
2316

Diff Legend

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