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

Comparing libeio/eio.c (file contents):
Revision 1.6 by root, Sun May 11 19:11:05 2008 UTC vs.
Revision 1.65 by root, Mon May 30 12:56:50 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
46#ifdef EIO_STACKSIZE
47# define XTHREAD_STACKSIZE EIO_STACKSIZE
48#endif
2#include "xthread.h" 49#include "xthread.h"
3 50
4#include <errno.h> 51#include <errno.h>
5#include <stddef.h> 52#include <stddef.h>
6#include <stdlib.h> 53#include <stdlib.h>
7#include <string.h> 54#include <string.h>
8#include <errno.h> 55#include <errno.h>
9#include <sys/types.h> 56#include <sys/types.h>
10#include <sys/stat.h> 57#include <sys/stat.h>
58#include <sys/statvfs.h>
11#include <limits.h> 59#include <limits.h>
12#include <fcntl.h> 60#include <fcntl.h>
13#include <sched.h> 61#include <assert.h>
14 62
15#ifndef EIO_FINISH 63#ifndef EIO_FINISH
16# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 64# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17#endif 65#endif
18 66
… …
25#endif 73#endif
26 74
27#ifdef _WIN32 75#ifdef _WIN32
28 76
29 /*doh*/ 77 /*doh*/
30
31#else 78#else
32 79
33# include "config.h"
34# include <sys/time.h> 80# include <sys/time.h>
35# include <sys/select.h> 81# include <sys/select.h>
36# include <unistd.h> 82# include <unistd.h>
37# include <utime.h> 83# include <utime.h>
38# include <signal.h> 84# include <signal.h>
39# include <dirent.h> 85# include <dirent.h>
40 86
87#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
88# include <sys/mman.h>
89#endif
90
91/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
92# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
93# define _DIRENT_HAVE_D_TYPE /* sigh */
94# define D_INO(de) (de)->d_fileno
95# define D_NAMLEN(de) (de)->d_namlen
96# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
97# define D_INO(de) (de)->d_ino
98# endif
99
100#ifdef _D_EXACT_NAMLEN
101# undef D_NAMLEN
102# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
103#endif
104
105# ifdef _DIRENT_HAVE_D_TYPE
106# define D_TYPE(de) (de)->d_type
107# endif
108
41# ifndef EIO_STRUCT_DIRENT 109# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 110# define EIO_STRUCT_DIRENT struct dirent
43# endif 111# endif
44 112
45#endif 113#endif
46 114
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif
50
51#if HAVE_SENDFILE 115#if HAVE_SENDFILE
52# if __linux 116# if __linux
53# include <sys/sendfile.h> 117# include <sys/sendfile.h>
54# elif __freebsd 118# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 119# include <sys/socket.h>
56# include <sys/uio.h> 120# include <sys/uio.h>
57# elif __hpux 121# elif __hpux
58# include <sys/socket.h> 122# include <sys/socket.h>
59# elif __solaris /* not yet */ 123# elif __solaris
60# include <sys/sendfile.h> 124# include <sys/sendfile.h>
61# else 125# else
62# error sendfile support requested but not available 126# error sendfile support requested but not available
63# endif 127# endif
64#endif 128#endif
65 129
66/* number of seconds after which an idle threads exit */ 130#ifndef D_TYPE
67#define IDLE_TIMEOUT 10 131# define D_TYPE(de) 0
132#endif
133#ifndef D_INO
134# define D_INO(de) 0
135#endif
136#ifndef D_NAMLEN
137# define D_NAMLEN(de) strlen ((de)->d_name)
138#endif
68 139
69/* used for struct dirent, AIX doesn't provide it */ 140/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 141#ifndef NAME_MAX
71# define NAME_MAX 4096 142# define NAME_MAX 4096
72#endif 143#endif
73 144
145/* used for readlink etc. */
146#ifndef PATH_MAX
147# define PATH_MAX 4096
148#endif
149
74/* buffer size for various temporary buffers */ 150/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 151#define EIO_BUFSIZE 65536
76 152
77#define dBUF \ 153#define dBUF \
78 char *eio_buf; \ 154 char *eio_buf; \
79 X_LOCK (wrklock); \ 155 ETP_WORKER_LOCK (self); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 156 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 157 ETP_WORKER_UNLOCK (self); \
158 errno = ENOMEM; \
82 if (!eio_buf) \ 159 if (!eio_buf) \
83 return -1; 160 return -1;
84 161
85#define EIO_TICKS ((1000000 + 1023) >> 10) 162#define EIO_TICKS ((1000000 + 1023) >> 10)
86 163
164/*****************************************************************************/
165
166#if __GNUC__ >= 3
167# define expect(expr,value) __builtin_expect ((expr),(value))
168#else
169# define expect(expr,value) (expr)
170#endif
171
172#define expect_false(expr) expect ((expr) != 0, 0)
173#define expect_true(expr) expect ((expr) != 0, 1)
174
175/*****************************************************************************/
176
177#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX
179
180struct etp_worker;
181
182#define ETP_REQ eio_req
183#define ETP_DESTROY(req) eio_destroy (req)
184static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
188
189#define ETP_WORKER_CLEAR(req) \
190 if (wrk->dbuf) \
191 { \
192 free (wrk->dbuf); \
193 wrk->dbuf = 0; \
194 } \
195 \
196 if (wrk->dirp) \
197 { \
198 closedir (wrk->dirp); \
199 wrk->dirp = 0; \
200 }
201
202#define ETP_WORKER_COMMON \
203 void *dbuf; \
204 DIR *dirp;
205
206/*****************************************************************************/
207
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209
210/* calculate time difference in ~1/EIO_TICKS of a second */
211static int tvdiff (struct timeval *tv1, struct timeval *tv2)
212{
213 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
214 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
215}
216
217static unsigned int started, idle, wanted = 4;
218
87static void (*want_poll_cb) (void); 219static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 220static void (*done_poll_cb) (void);
89 221
90static unsigned int max_poll_time = 0; 222static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 223static unsigned int max_poll_reqs; /* reslock */
92 224
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 225static volatile unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 226static volatile unsigned int nready; /* reqlock */
95{ 227static volatile unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 228static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 229static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 230
100static unsigned int started, idle, wanted; 231static xmutex_t wrklock;
232static xmutex_t reslock;
233static xmutex_t reqlock;
234static xcond_t reqwait;
235
236#if !HAVE_PREADWRITE
237/*
238 * make our pread/pwrite emulation safe against themselves, but not against
239 * normal read/write by using a mutex. slows down execution a lot,
240 * but that's your problem, not mine.
241 */
242static xmutex_t preadwritelock = X_MUTEX_INIT;
243#endif
244
245typedef struct etp_worker
246{
247 /* locked by wrklock */
248 struct etp_worker *prev, *next;
249
250 xthread_t tid;
251
252 /* locked by reslock, reqlock or wrklock */
253 ETP_REQ *req; /* currently processed request */
254
255 ETP_WORKER_COMMON
256} etp_worker;
257
258static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
259
260#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
261#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
101 262
102/* worker threads management */ 263/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 264
105typedef struct worker {
106 /* locked by wrklock */
107 struct worker *prev, *next;
108
109 thread_t tid;
110
111 /* locked by reslock, reqlock or wrklock */
112 eio_req *req; /* currently processed request */
113 void *dbuf;
114 DIR *dirp;
115} worker;
116
117static worker wrk_first = { &wrk_first, &wrk_first, 0 };
118
119static void worker_clear (worker *wrk) 265static void etp_worker_clear (etp_worker *wrk)
120{ 266{
121 if (wrk->dirp) 267 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} 268}
133 269
134static void worker_free (worker *wrk) 270static void etp_worker_free (etp_worker *wrk)
135{ 271{
136 wrk->next->prev = wrk->prev; 272 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 273 wrk->prev->next = wrk->next;
138 274
139 free (wrk); 275 free (wrk);
140} 276}
141 277
142static volatile unsigned int nreqs, nready, npending;
143static volatile unsigned int max_idle = 4;
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) 278static unsigned int etp_nreqs (void)
150{ 279{
280 int retval;
281 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
282 retval = nreqs;
283 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 284 return retval;
152} 285}
153 286
154unsigned int eio_nready (void) 287static unsigned int etp_nready (void)
155{ 288{
156 unsigned int retval; 289 unsigned int retval;
157 290
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 291 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 292 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 293 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 294
162 return retval; 295 return retval;
163} 296}
164 297
165unsigned int eio_npending (void) 298static unsigned int etp_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 etp_nthreads (void)
177{ 310{
178 unsigned int retval; 311 unsigned int retval;
179 312
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 313 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 314 retval = started;
… …
188 * a somewhat faster data structure might be nice, but 321 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 322 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 323 * per shift, the most expensive operation.
191 */ 324 */
192typedef struct { 325typedef struct {
193 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 326 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194 int size; 327 int size;
195} reqq; 328} etp_reqq;
196 329
197static reqq req_queue; 330static etp_reqq req_queue;
198static reqq res_queue; 331static etp_reqq res_queue;
199 332
200static int reqq_push (reqq *q, eio_req *req) 333static int reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 334{
202 int pri = req->pri; 335 int pri = req->pri;
203 req->next = 0; 336 req->next = 0;
204 337
205 if (q->qe[pri]) 338 if (q->qe[pri])
… …
211 q->qe[pri] = q->qs[pri] = req; 344 q->qe[pri] = q->qs[pri] = req;
212 345
213 return q->size++; 346 return q->size++;
214} 347}
215 348
216static eio_req *reqq_shift (reqq *q) 349static ETP_REQ *reqq_shift (etp_reqq *q)
217{ 350{
218 int pri; 351 int pri;
219 352
220 if (!q->size) 353 if (!q->size)
221 return 0; 354 return 0;
222 355
223 --q->size; 356 --q->size;
224 357
225 for (pri = EIO_NUM_PRI; pri--; ) 358 for (pri = ETP_NUM_PRI; pri--; )
226 { 359 {
227 eio_req *req = q->qs[pri]; 360 eio_req *req = q->qs[pri];
228 361
229 if (req) 362 if (req)
230 { 363 {
… …
236 } 369 }
237 370
238 abort (); 371 abort ();
239} 372}
240 373
241static void grp_try_feed (eio_req *grp) 374static void etp_thread_init (void)
242{ 375{
243 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 376 X_MUTEX_CREATE (wrklock);
377 X_MUTEX_CREATE (reslock);
378 X_MUTEX_CREATE (reqlock);
379 X_COND_CREATE (reqwait);
380}
381
382static void etp_atfork_prepare (void)
383{
384 X_LOCK (wrklock);
385 X_LOCK (reqlock);
386 X_LOCK (reslock);
387#if !HAVE_PREADWRITE
388 X_LOCK (preadwritelock);
389#endif
390}
391
392static void etp_atfork_parent (void)
393{
394#if !HAVE_PREADWRITE
395 X_UNLOCK (preadwritelock);
396#endif
397 X_UNLOCK (reslock);
398 X_UNLOCK (reqlock);
399 X_UNLOCK (wrklock);
400}
401
402static void etp_atfork_child (void)
403{
404 ETP_REQ *prv;
405
406 while ((prv = reqq_shift (&req_queue)))
407 ETP_DESTROY (prv);
408
409 while ((prv = reqq_shift (&res_queue)))
410 ETP_DESTROY (prv);
411
412 while (wrk_first.next != &wrk_first)
244 { 413 {
245 int old_len = grp->size; 414 etp_worker *wrk = wrk_first.next;
246 415
247 EIO_FEED (grp); 416 if (wrk->req)
417 ETP_DESTROY (wrk->req);
248 418
249 /* stop if no progress has been made */ 419 etp_worker_clear (wrk);
250 if (old_len == grp->size) 420 etp_worker_free (wrk);
251 {
252 grp->feed = 0;
253 break;
254 }
255 } 421 }
256}
257 422
258static int eio_finish (eio_req *req); 423 started = 0;
424 idle = 0;
425 nreqs = 0;
426 nready = 0;
427 npending = 0;
259 428
260static int grp_dec (eio_req *grp) 429 etp_thread_init ();
261{ 430}
262 --grp->size;
263 431
264 /* call feeder, if applicable */ 432static void
265 grp_try_feed (grp); 433etp_once_init (void)
434{
435 etp_thread_init ();
436 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
437}
266 438
267 /* finish, if done */ 439static int
268 if (!grp->size && grp->int1) 440etp_init (void (*want_poll)(void), void (*done_poll)(void))
269 return eio_finish (grp); 441{
270 else 442 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
443
444 pthread_once (&doinit, etp_once_init);
445
446 want_poll_cb = want_poll;
447 done_poll_cb = done_poll;
448
271 return 0; 449 return 0;
272} 450}
273 451
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); 452X_THREAD_PROC (etp_proc);
325 453
326static void start_thread (void) 454static void etp_start_thread (void)
327{ 455{
328 worker *wrk = calloc (1, sizeof (worker)); 456 etp_worker *wrk = calloc (1, sizeof (etp_worker));
329 457
330 if (!wrk) 458 /*TODO*/
331 croak ("unable to allocate worker thread data"); 459 assert (("unable to allocate worker thread data", wrk));
332 460
333 X_LOCK (wrklock); 461 X_LOCK (wrklock);
334 462
335 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 463 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
336 { 464 {
337 wrk->prev = &wrk_first; 465 wrk->prev = &wrk_first;
338 wrk->next = wrk_first.next; 466 wrk->next = wrk_first.next;
339 wrk_first.next->prev = wrk; 467 wrk_first.next->prev = wrk;
340 wrk_first.next = wrk; 468 wrk_first.next = wrk;
… …
344 free (wrk); 472 free (wrk);
345 473
346 X_UNLOCK (wrklock); 474 X_UNLOCK (wrklock);
347} 475}
348 476
349static void maybe_start_thread (void) 477static void etp_maybe_start_thread (void)
350{ 478{
351 if (eio_nthreads () >= wanted) 479 if (expect_true (etp_nthreads () >= wanted))
352 return; 480 return;
353 481
354 /* todo: maybe use idle here, but might be less exact */ 482 /* todo: maybe use idle here, but might be less exact */
355 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 483 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
356 return; 484 return;
357 485
358 start_thread (); 486 etp_start_thread ();
359} 487}
360 488
361void eio_submit (eio_req *req)
362{
363 ++nreqs;
364
365 X_LOCK (reqlock);
366 ++nready;
367 reqq_push (&req_queue, req);
368 X_COND_SIGNAL (reqwait);
369 X_UNLOCK (reqlock);
370
371 maybe_start_thread ();
372}
373
374static void end_thread (void) 489static void etp_end_thread (void)
375{ 490{
376 eio_req *req = calloc (1, sizeof (eio_req)); 491 eio_req *req = calloc (1, sizeof (eio_req));
377 492
378 req->type = EIO_QUIT; 493 req->type = -1;
379 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 494 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
380 495
381 X_LOCK (reqlock); 496 X_LOCK (reqlock);
382 reqq_push (&req_queue, req); 497 reqq_push (&req_queue, req);
383 X_COND_SIGNAL (reqwait); 498 X_COND_SIGNAL (reqwait);
384 X_UNLOCK (reqlock); 499 X_UNLOCK (reqlock);
… …
386 X_LOCK (wrklock); 501 X_LOCK (wrklock);
387 --started; 502 --started;
388 X_UNLOCK (wrklock); 503 X_UNLOCK (wrklock);
389} 504}
390 505
391void eio_set_max_poll_time (double nseconds)
392{
393 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
394 max_poll_time = nseconds;
395 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
396}
397
398void eio_set_max_poll_reqs (unsigned int maxreqs)
399{
400 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
401 max_poll_reqs = maxreqs;
402 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
403}
404
405void eio_set_max_idle (unsigned int nthreads)
406{
407 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
408 max_idle = nthreads <= 0 ? 1 : nthreads;
409 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
410}
411
412void eio_set_min_parallel (unsigned int nthreads)
413{
414 if (wanted < nthreads)
415 wanted = nthreads;
416}
417
418void eio_set_max_parallel (unsigned int nthreads)
419{
420 if (wanted > nthreads)
421 wanted = nthreads;
422
423 while (started > wanted)
424 end_thread ();
425}
426
427int eio_poll (void) 506static int etp_poll (void)
428{ 507{
429 int maxreqs = max_poll_reqs; 508 unsigned int maxreqs;
509 unsigned int maxtime;
430 struct timeval tv_start, tv_now; 510 struct timeval tv_start, tv_now;
431 eio_req *req;
432 511
512 X_LOCK (reslock);
513 maxreqs = max_poll_reqs;
514 maxtime = max_poll_time;
515 X_UNLOCK (reslock);
516
433 if (max_poll_time) 517 if (maxtime)
434 gettimeofday (&tv_start, 0); 518 gettimeofday (&tv_start, 0);
435 519
436 for (;;) 520 for (;;)
437 { 521 {
522 ETP_REQ *req;
523
438 maybe_start_thread (); 524 etp_maybe_start_thread ();
439 525
440 X_LOCK (reslock); 526 X_LOCK (reslock);
441 req = reqq_shift (&res_queue); 527 req = reqq_shift (&res_queue);
442 528
443 if (req) 529 if (req)
… …
451 X_UNLOCK (reslock); 537 X_UNLOCK (reslock);
452 538
453 if (!req) 539 if (!req)
454 return 0; 540 return 0;
455 541
542 X_LOCK (reqlock);
456 --nreqs; 543 --nreqs;
544 X_UNLOCK (reqlock);
457 545
458 if (req->type == EIO_GROUP && req->size) 546 if (expect_false (req->type == EIO_GROUP && req->size))
459 { 547 {
460 req->int1 = 1; /* mark request as delayed */ 548 req->int1 = 1; /* mark request as delayed */
461 continue; 549 continue;
462 } 550 }
463 else 551 else
464 { 552 {
465 int res = eio_finish (req); 553 int res = ETP_FINISH (req);
466 if (res) 554 if (expect_false (res))
467 return res; 555 return res;
468 } 556 }
469 557
470 if (maxreqs && !--maxreqs) 558 if (expect_false (maxreqs && !--maxreqs))
471 break; 559 break;
472 560
473 if (max_poll_time) 561 if (maxtime)
474 { 562 {
475 gettimeofday (&tv_now, 0); 563 gettimeofday (&tv_now, 0);
476 564
477 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 565 if (tvdiff (&tv_start, &tv_now) >= maxtime)
478 break; 566 break;
479 } 567 }
480 } 568 }
481 569
482 errno = EAGAIN; 570 errno = EAGAIN;
483 return -1; 571 return -1;
484} 572}
485 573
574static void etp_cancel (ETP_REQ *req)
575{
576 X_LOCK (wrklock);
577 req->flags |= EIO_FLAG_CANCELLED;
578 X_UNLOCK (wrklock);
579
580 eio_grp_cancel (req);
581}
582
583static void etp_submit (ETP_REQ *req)
584{
585 req->pri -= ETP_PRI_MIN;
586
587 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
588 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
589
590 if (expect_false (req->type == EIO_GROUP))
591 {
592 /* I hope this is worth it :/ */
593 X_LOCK (reqlock);
594 ++nreqs;
595 X_UNLOCK (reqlock);
596
597 X_LOCK (reslock);
598
599 ++npending;
600
601 if (!reqq_push (&res_queue, req) && want_poll_cb)
602 want_poll_cb ();
603
604 X_UNLOCK (reslock);
605 }
606 else
607 {
608 X_LOCK (reqlock);
609 ++nreqs;
610 ++nready;
611 reqq_push (&req_queue, req);
612 X_COND_SIGNAL (reqwait);
613 X_UNLOCK (reqlock);
614
615 etp_maybe_start_thread ();
616 }
617}
618
619static void etp_set_max_poll_time (double nseconds)
620{
621 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
622 max_poll_time = nseconds * EIO_TICKS;
623 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
624}
625
626static void etp_set_max_poll_reqs (unsigned int maxreqs)
627{
628 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
629 max_poll_reqs = maxreqs;
630 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
631}
632
633static void etp_set_max_idle (unsigned int nthreads)
634{
635 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
636 max_idle = nthreads;
637 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
638}
639
640static void etp_set_idle_timeout (unsigned int seconds)
641{
642 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
643 idle_timeout = seconds;
644 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
645}
646
647static void etp_set_min_parallel (unsigned int nthreads)
648{
649 if (wanted < nthreads)
650 wanted = nthreads;
651}
652
653static void etp_set_max_parallel (unsigned int nthreads)
654{
655 if (wanted > nthreads)
656 wanted = nthreads;
657
658 while (started > wanted)
659 etp_end_thread ();
660}
661
662/*****************************************************************************/
663
664static void grp_try_feed (eio_req *grp)
665{
666 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
667 {
668 grp->flags &= ~EIO_FLAG_GROUPADD;
669
670 EIO_FEED (grp);
671
672 /* stop if no progress has been made */
673 if (!(grp->flags & EIO_FLAG_GROUPADD))
674 {
675 grp->feed = 0;
676 break;
677 }
678 }
679}
680
681static int grp_dec (eio_req *grp)
682{
683 --grp->size;
684
685 /* call feeder, if applicable */
686 grp_try_feed (grp);
687
688 /* finish, if done */
689 if (!grp->size && grp->int1)
690 return eio_finish (grp);
691 else
692 return 0;
693}
694
695void eio_destroy (eio_req *req)
696{
697 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
698 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
699
700 EIO_DESTROY (req);
701}
702
703static int eio_finish (eio_req *req)
704{
705 int res = EIO_FINISH (req);
706
707 if (req->grp)
708 {
709 int res2;
710 eio_req *grp = req->grp;
711
712 /* unlink request */
713 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
714 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
715
716 if (grp->grp_first == req)
717 grp->grp_first = req->grp_next;
718
719 res2 = grp_dec (grp);
720
721 if (!res && res2)
722 res = res2;
723 }
724
725 eio_destroy (req);
726
727 return res;
728}
729
730void eio_grp_cancel (eio_req *grp)
731{
732 for (grp = grp->grp_first; grp; grp = grp->grp_next)
733 eio_cancel (grp);
734}
735
736void eio_cancel (eio_req *req)
737{
738 etp_cancel (req);
739}
740
741void eio_submit (eio_req *req)
742{
743 etp_submit (req);
744}
745
746unsigned int eio_nreqs (void)
747{
748 return etp_nreqs ();
749}
750
751unsigned int eio_nready (void)
752{
753 return etp_nready ();
754}
755
756unsigned int eio_npending (void)
757{
758 return etp_npending ();
759}
760
761unsigned int eio_nthreads (void)
762{
763 return etp_nthreads ();
764}
765
766void eio_set_max_poll_time (double nseconds)
767{
768 etp_set_max_poll_time (nseconds);
769}
770
771void eio_set_max_poll_reqs (unsigned int maxreqs)
772{
773 etp_set_max_poll_reqs (maxreqs);
774}
775
776void eio_set_max_idle (unsigned int nthreads)
777{
778 etp_set_max_idle (nthreads);
779}
780
781void eio_set_idle_timeout (unsigned int seconds)
782{
783 etp_set_idle_timeout (seconds);
784}
785
786void eio_set_min_parallel (unsigned int nthreads)
787{
788 etp_set_min_parallel (nthreads);
789}
790
791void eio_set_max_parallel (unsigned int nthreads)
792{
793 etp_set_max_parallel (nthreads);
794}
795
796int eio_poll (void)
797{
798 return etp_poll ();
799}
800
486/*****************************************************************************/ 801/*****************************************************************************/
487/* work around various missing functions */ 802/* work around various missing functions */
488 803
489#if !HAVE_PREADWRITE 804#if !HAVE_PREADWRITE
805# undef pread
806# undef pwrite
490# define pread aio_pread 807# define pread eio__pread
491# define pwrite aio_pwrite 808# define pwrite eio__pwrite
492 809
493/* 810static ssize_t
494 * make our pread/pwrite safe against themselves, but not against 811eio__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{ 812{
502 ssize_t res; 813 ssize_t res;
503 off_t ooffset; 814 off_t ooffset;
504 815
505 X_LOCK (preadwritelock); 816 X_LOCK (preadwritelock);
… …
510 X_UNLOCK (preadwritelock); 821 X_UNLOCK (preadwritelock);
511 822
512 return res; 823 return res;
513} 824}
514 825
515static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 826static ssize_t
827eio__pwrite (int fd, void *buf, size_t count, off_t offset)
516{ 828{
517 ssize_t res; 829 ssize_t res;
518 off_t ooffset; 830 off_t ooffset;
519 831
520 X_LOCK (preadwritelock); 832 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR); 833 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET); 834 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count); 835 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET); 836 lseek (fd, ooffset, SEEK_SET);
525 X_UNLOCK (preadwritelock); 837 X_UNLOCK (preadwritelock);
526 838
527 return res; 839 return res;
528} 840}
529#endif 841#endif
530 842
531#ifndef HAVE_FUTIMES 843#ifndef HAVE_UTIMES
532 844
845# undef utimes
533# define utimes(path,times) aio_utimes (path, times) 846# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) aio_futimes (fd, times)
535 847
848static int
536static int aio_utimes (const char *filename, const struct timeval times[2]) 849eio__utimes (const char *filename, const struct timeval times[2])
537{ 850{
538 if (times) 851 if (times)
539 { 852 {
540 struct utimbuf buf; 853 struct utimbuf buf;
541 854
… …
546 } 859 }
547 else 860 else
548 return utime (filename, 0); 861 return utime (filename, 0);
549} 862}
550 863
864#endif
865
866#ifndef HAVE_FUTIMES
867
868# undef futimes
869# define futimes(fd,times) eio__futimes (fd, times)
870
551static int aio_futimes (int fd, const struct timeval tv[2]) 871static int eio__futimes (int fd, const struct timeval tv[2])
552{ 872{
553 errno = ENOSYS; 873 errno = ENOSYS;
554 return -1; 874 return -1;
555} 875}
556 876
557#endif 877#endif
558 878
559#if !HAVE_FDATASYNC 879#if !HAVE_FDATASYNC
880# undef fdatasync
560# define fdatasync fsync 881# define fdatasync(fd) fsync (fd)
561#endif 882#endif
883
884/* sync_file_range always needs emulation */
885int
886eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
887{
888#if HAVE_SYNC_FILE_RANGE
889 int res;
890
891 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
892 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
893 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
894 {
895 flags = 0
896 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
897 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
898 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
899 }
900
901 res = sync_file_range (fd, offset, nbytes, flags);
902
903 if (!res || errno != ENOSYS)
904 return res;
905#endif
906
907 /* even though we could play tricks with the flags, it's better to always
908 * call fdatasync, as that matches the expectation of its users best */
909 return fdatasync (fd);
910}
562 911
563#if !HAVE_READAHEAD 912#if !HAVE_READAHEAD
913# undef readahead
564# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 914# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565 915
916static ssize_t
566static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 917eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
567{ 918{
568 size_t todo = count; 919 size_t todo = count;
569 dBUF; 920 dBUF;
570 921
571 while (todo > 0) 922 while (todo > 0)
… …
581 return count; 932 return count;
582} 933}
583 934
584#endif 935#endif
585 936
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 */ 937/* sendfile always needs emulation */
938static ssize_t
617static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 939eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
618{ 940{
619 ssize_t res; 941 ssize_t res;
620 942
621 if (!count) 943 if (!count)
622 return 0; 944 return 0;
623 945
624#if HAVE_SENDFILE 946#if HAVE_SENDFILE
625# if __linux 947# if __linux
626 res = sendfile (ofd, ifd, &offset, count); 948 res = sendfile (ofd, ifd, &offset, count);
627 949
628# elif __freebsd 950# elif __FreeBSD__
629 /* 951 /*
630 * Of course, the freebsd sendfile is a dire hack with no thoughts 952 * Of course, the freebsd sendfile is a dire hack with no thoughts
631 * wasted on making it similar to other I/O functions. 953 * wasted on making it similar to other I/O functions.
632 */ 954 */
633 { 955 {
634 off_t sbytes; 956 off_t sbytes;
635 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 957 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636 958
637 if (res < 0 && sbytes) 959 #if 0 /* according to the manpage, this is correct, but broken behaviour */
638 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 960 /* freebsd' sendfile will return 0 on success */
961 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
962 /* not on e.g. EIO or EPIPE - sounds broken */
963 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
964 res = sbytes;
965 #endif
966
967 /* according to source inspection, this is correct, and useful behaviour */
968 if (sbytes)
969 res = sbytes;
970 }
971
972# elif defined (__APPLE__)
973
974 {
975 off_t sbytes = count;
976 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
977
978 /* according to the manpage, sbytes is always valid */
979 if (sbytes)
639 res = sbytes; 980 res = sbytes;
640 } 981 }
641 982
642# elif __hpux 983# elif __hpux
643 res = sendfile (ofd, ifd, offset, count, 0, 0); 984 res = sendfile (ofd, ifd, offset, count, 0, 0);
… …
657 if (res < 0 && sbytes) 998 if (res < 0 && sbytes)
658 res = sbytes; 999 res = sbytes;
659 } 1000 }
660 1001
661# endif 1002# endif
1003
1004#elif defined (_WIN32)
1005
1006 /* does not work, just for documentation of what would need to be done */
1007 {
1008 HANDLE h = TO_SOCKET (ifd);
1009 SetFilePointer (h, offset, 0, FILE_BEGIN);
1010 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1011 }
1012
662#else 1013#else
663 res = -1; 1014 res = -1;
664 errno = ENOSYS; 1015 errno = ENOSYS;
665#endif 1016#endif
666 1017
667 if (res < 0 1018 if (res < 0
668 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1019 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1020 /* BSDs */
1021#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1022 || errno == ENOTSUP
1023#endif
1024 || errno == EOPNOTSUPP /* BSDs */
669#if __solaris 1025#if __solaris
670 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1026 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
671#endif 1027#endif
672 ) 1028 )
673 ) 1029 )
… …
704 } 1060 }
705 1061
706 return res; 1062 return res;
707} 1063}
708 1064
1065static signed char
1066eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1067{
1068 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1069 : a->inode < b->inode ? -1
1070 : a->inode > b->inode ? 1
1071 : 0;
1072}
1073
1074#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1075
1076#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1077#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1078
1079static void
1080eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1081{
1082 unsigned char bits [9 + sizeof (ino_t) * 8];
1083 unsigned char *bit = bits;
1084
1085 assert (CHAR_BIT == 8);
1086 assert (sizeof (eio_dirent) * 8 < 256);
1087 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1088 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1089
1090 if (size <= EIO_SORT_FAST)
1091 return;
1092
1093 /* first prepare an array of bits to test in our radix sort */
1094 /* try to take endianness into account, as well as differences in ino_t sizes */
1095 /* inode_bits must contain all inodes ORed together */
1096 /* which is used to skip bits that are 0 everywhere, which is very common */
1097 {
1098 ino_t endianness;
1099 int i, j;
1100
1101 /* we store the byte offset of byte n into byte n of "endianness" */
1102 for (i = 0; i < sizeof (ino_t); ++i)
1103 ((unsigned char *)&endianness)[i] = i;
1104
1105 *bit++ = 0;
1106
1107 for (i = 0; i < sizeof (ino_t); ++i)
1108 {
1109 /* shifting off the byte offsets out of "endianness" */
1110 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1111 endianness >>= 8;
1112
1113 for (j = 0; j < 8; ++j)
1114 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1115 *bit++ = offs + j;
1116 }
1117
1118 for (j = 0; j < 8; ++j)
1119 if (score_bits & (1 << j))
1120 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1121 }
1122
1123 /* now actually do the sorting (a variant of MSD radix sort) */
1124 {
1125 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1126 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1127 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1128 int stk_idx = 0;
1129
1130 base_stk [stk_idx] = dents;
1131 end_stk [stk_idx] = dents + size;
1132 bit_stk [stk_idx] = bit - 1;
1133
1134 do
1135 {
1136 base = base_stk [stk_idx];
1137 end = end_stk [stk_idx];
1138 bit = bit_stk [stk_idx];
1139
1140 for (;;)
1141 {
1142 unsigned char O = *bit >> 3;
1143 unsigned char M = 1 << (*bit & 7);
1144
1145 eio_dirent *a = base;
1146 eio_dirent *b = end;
1147
1148 if (b - a < EIO_SORT_CUTOFF)
1149 break;
1150
1151 /* now bit-partition the array on the bit */
1152 /* this ugly asymmetric loop seems to perform much better than typical */
1153 /* partition algos found in the literature */
1154 do
1155 if (!(((unsigned char *)a)[O] & M))
1156 ++a;
1157 else if (!(((unsigned char *)--b)[O] & M))
1158 {
1159 eio_dirent tmp = *a; *a = *b; *b = tmp;
1160 ++a;
1161 }
1162 while (b > a);
1163
1164 /* next bit, or stop, if no bits left in this path */
1165 if (!*--bit)
1166 break;
1167
1168 base_stk [stk_idx] = a;
1169 end_stk [stk_idx] = end;
1170 bit_stk [stk_idx] = bit;
1171 ++stk_idx;
1172
1173 end = a;
1174 }
1175 }
1176 while (stk_idx--);
1177 }
1178}
1179
1180static void
1181eio_dent_insertion_sort (eio_dirent *dents, int size)
1182{
1183 /* first move the smallest element to the front, to act as a sentinel */
1184 {
1185 int i;
1186 eio_dirent *min = dents;
1187
1188 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1189 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1190 if (EIO_DENT_CMP (dents [i], <, *min))
1191 min = &dents [i];
1192
1193 /* swap elements 0 and j (minimum) */
1194 {
1195 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1196 }
1197 }
1198
1199 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1200 {
1201 eio_dirent *i, *j;
1202
1203 for (i = dents + 1; i < dents + size; ++i)
1204 {
1205 eio_dirent value = *i;
1206
1207 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1208 j [1] = j [0];
1209
1210 j [1] = value;
1211 }
1212 }
1213}
1214
1215static void
1216eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1217{
1218 if (size <= 1)
1219 return; /* our insertion sort relies on size > 0 */
1220
1221 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1222 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1223 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1224
1225 /* use an insertion sort at the end, or for small arrays, */
1226 /* as insertion sort is more efficient for small partitions */
1227 eio_dent_insertion_sort (dents, size);
1228}
1229
709/* read a full directory */ 1230/* read a full directory */
1231static void
710static void scandir_ (eio_req *req, worker *self) 1232eio__scandir (eio_req *req, etp_worker *self)
711{ 1233{
712 DIR *dirp; 1234 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; 1235 EIO_STRUCT_DIRENT *entp;
719 char *name, *names; 1236 char *name, *names;
720 int memlen = 4096; 1237 int namesalloc = 4096;
721 int memofs = 0; 1238 int namesoffs = 0;
1239 int flags = req->int1;
1240 eio_dirent *dents = 0;
1241 int dentalloc = 128;
722 int res = 0; 1242 int dentoffs = 0;
1243 ino_t inode_bits = 0;
1244
1245 req->result = -1;
1246
1247 if (!(flags & EIO_READDIR_DENTS))
1248 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
723 1249
724 X_LOCK (wrklock); 1250 X_LOCK (wrklock);
1251 /* the corresponding closedir is in ETP_WORKER_CLEAR */
725 self->dirp = dirp = opendir (req->ptr1); 1252 self->dirp = dirp = opendir (req->ptr1);
726 self->dbuf = u = malloc (sizeof (*u)); 1253
727 req->flags |= EIO_FLAG_PTR2_FREE; 1254 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1255 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
728 req->ptr2 = names = malloc (memlen); 1256 req->ptr2 = names = malloc (namesalloc);
729 X_UNLOCK (wrklock); 1257 X_UNLOCK (wrklock);
730 1258
731 if (dirp && u && names) 1259 if (dirp && names && (!flags || dents))
732 for (;;) 1260 for (;;)
733 { 1261 {
734 errno = 0; 1262 errno = 0;
735 readdir_r (dirp, &u->d, &entp); 1263 entp = readdir (dirp);
736 1264
737 if (!entp) 1265 if (!entp)
1266 {
1267 if (errno)
1268 break;
1269
1270 /* sort etc. */
1271 req->int1 = flags;
1272 req->result = dentoffs;
1273
1274 if (flags & EIO_READDIR_STAT_ORDER)
1275 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1276 else if (flags & EIO_READDIR_DIRS_FIRST)
1277 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1278 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1279 else
1280 {
1281 /* in this case, all is known, and we just put dirs first and sort them */
1282 eio_dirent *oth = dents + dentoffs;
1283 eio_dirent *dir = dents;
1284
1285 /* now partition dirs to the front, and non-dirs to the back */
1286 /* by walking from both sides and swapping if necessary */
1287 while (oth > dir)
1288 {
1289 if (dir->type == EIO_DT_DIR)
1290 ++dir;
1291 else if ((--oth)->type == EIO_DT_DIR)
1292 {
1293 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1294
1295 ++dir;
1296 }
1297 }
1298
1299 /* now sort the dirs only (dirs all have the same score) */
1300 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1301 }
1302
738 break; 1303 break;
1304 }
739 1305
1306 /* now add the entry to our list(s) */
740 name = entp->d_name; 1307 name = entp->d_name;
741 1308
1309 /* skip . and .. entries */
742 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1310 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
743 { 1311 {
744 int len = strlen (name) + 1; 1312 int len = D_NAMLEN (entp) + 1;
745 1313
746 res++; 1314 while (expect_false (namesoffs + len > namesalloc))
747
748 while (memofs + len > memlen)
749 { 1315 {
750 memlen *= 2; 1316 namesalloc *= 2;
751 X_LOCK (wrklock); 1317 X_LOCK (wrklock);
752 req->ptr2 = names = realloc (names, memlen); 1318 req->ptr2 = names = realloc (names, namesalloc);
753 X_UNLOCK (wrklock); 1319 X_UNLOCK (wrklock);
754 1320
755 if (!names) 1321 if (!names)
756 break; 1322 break;
757 } 1323 }
758 1324
759 memcpy (names + memofs, name, len); 1325 memcpy (names + namesoffs, name, len);
1326
1327 if (dents)
1328 {
1329 struct eio_dirent *ent;
1330
1331 if (expect_false (dentoffs == dentalloc))
1332 {
1333 dentalloc *= 2;
1334 X_LOCK (wrklock);
1335 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1336 X_UNLOCK (wrklock);
1337
1338 if (!dents)
1339 break;
1340 }
1341
1342 ent = dents + dentoffs;
1343
1344 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1345 ent->namelen = len - 1;
1346 ent->inode = D_INO (entp);
1347
1348 inode_bits |= ent->inode;
1349
1350 switch (D_TYPE (entp))
1351 {
1352 default:
1353 ent->type = EIO_DT_UNKNOWN;
1354 flags |= EIO_READDIR_FOUND_UNKNOWN;
1355 break;
1356
1357 #ifdef DT_FIFO
1358 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1359 #endif
1360 #ifdef DT_CHR
1361 case DT_CHR: ent->type = EIO_DT_CHR; break;
1362 #endif
1363 #ifdef DT_MPC
1364 case DT_MPC: ent->type = EIO_DT_MPC; break;
1365 #endif
1366 #ifdef DT_DIR
1367 case DT_DIR: ent->type = EIO_DT_DIR; break;
1368 #endif
1369 #ifdef DT_NAM
1370 case DT_NAM: ent->type = EIO_DT_NAM; break;
1371 #endif
1372 #ifdef DT_BLK
1373 case DT_BLK: ent->type = EIO_DT_BLK; break;
1374 #endif
1375 #ifdef DT_MPB
1376 case DT_MPB: ent->type = EIO_DT_MPB; break;
1377 #endif
1378 #ifdef DT_REG
1379 case DT_REG: ent->type = EIO_DT_REG; break;
1380 #endif
1381 #ifdef DT_NWK
1382 case DT_NWK: ent->type = EIO_DT_NWK; break;
1383 #endif
1384 #ifdef DT_CMP
1385 case DT_CMP: ent->type = EIO_DT_CMP; break;
1386 #endif
1387 #ifdef DT_LNK
1388 case DT_LNK: ent->type = EIO_DT_LNK; break;
1389 #endif
1390 #ifdef DT_SOCK
1391 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1392 #endif
1393 #ifdef DT_DOOR
1394 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1395 #endif
1396 #ifdef DT_WHT
1397 case DT_WHT: ent->type = EIO_DT_WHT; break;
1398 #endif
1399 }
1400
1401 ent->score = 7;
1402
1403 if (flags & EIO_READDIR_DIRS_FIRST)
1404 {
1405 if (ent->type == EIO_DT_UNKNOWN)
1406 {
1407 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1408 ent->score = 1;
1409 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1410 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1411 }
1412 else if (ent->type == EIO_DT_DIR)
1413 ent->score = 0;
1414 }
1415 }
1416
760 memofs += len; 1417 namesoffs += len;
1418 ++dentoffs;
1419 }
1420
1421 if (EIO_CANCELLED (req))
1422 {
1423 errno = ECANCELED;
1424 break;
761 } 1425 }
762 } 1426 }
1427}
763 1428
764 if (errno) 1429#ifdef PAGESIZE
765 res = -1; 1430# define eio_pagesize() PAGESIZE
1431#else
1432static intptr_t
1433eio_pagesize (void)
1434{
1435 static intptr_t page;
1436
1437 if (!page)
1438 page = sysconf (_SC_PAGESIZE);
1439
1440 return page;
1441}
1442#endif
1443
1444static void
1445eio_page_align (void **addr, size_t *length)
1446{
1447 intptr_t mask = eio_pagesize () - 1;
1448
1449 /* round down addr */
1450 intptr_t adj = mask & (intptr_t)*addr;
1451
1452 *addr = (void *)((intptr_t)*addr - adj);
1453 *length += adj;
1454
1455 /* round up length */
1456 *length = (*length + mask) & ~mask;
1457}
1458
1459#if !_POSIX_MEMLOCK
1460# define eio__mlockall(a) ((errno = ENOSYS), -1)
1461#else
1462
1463static int
1464eio__mlockall (int flags)
1465{
1466 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1467 extern int mallopt (int, int);
1468 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1469 #endif
1470
1471 if (EIO_MCL_CURRENT != MCL_CURRENT
1472 || EIO_MCL_FUTURE != MCL_FUTURE)
1473 {
1474 flags = 0
1475 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1476 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1477 }
1478
1479 return mlockall (flags);
1480}
1481#endif
1482
1483#if !_POSIX_MEMLOCK_RANGE
1484# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1485#else
1486
1487static int
1488eio__mlock (void *addr, size_t length)
1489{
1490 eio_page_align (&addr, &length);
1491
1492 return mlock (addr, length);
1493}
1494
1495#endif
1496
1497#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1498# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1499#else
1500
1501int
1502eio__msync (void *mem, size_t len, int flags)
1503{
1504 eio_page_align (&mem, &len);
1505
1506 if (EIO_MS_ASYNC != MS_SYNC
1507 || EIO_MS_INVALIDATE != MS_INVALIDATE
1508 || EIO_MS_SYNC != MS_SYNC)
1509 {
1510 flags = 0
1511 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1512 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1513 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1514 }
1515
1516 return msync (mem, len, flags);
1517}
1518
1519#endif
1520
1521int
1522eio__mtouch (void *mem, size_t len, int flags)
1523{
1524 eio_page_align (&mem, &len);
1525
766 1526 {
767 req->result = res; 1527 intptr_t addr = (intptr_t)mem;
1528 intptr_t end = addr + len;
1529 intptr_t page = eio_pagesize ();
1530
1531 if (addr < end)
1532 if (flags & EIO_MT_MODIFY) /* modify */
1533 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1534 else
1535 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1536 }
1537
1538 return 0;
768} 1539}
769 1540
770/*****************************************************************************/ 1541/*****************************************************************************/
771 1542
772#define ALLOC(len) \ 1543#define ALLOC(len) \
773 X_LOCK (wrklock); \
774 req->flags |= EIO_FLAG_PTR2_FREE; \
775 X_UNLOCK (wrklock); \
776 req->ptr2 = malloc (len); \
777 if (!req->ptr2) \ 1544 if (!req->ptr2) \
778 { \ 1545 { \
1546 X_LOCK (wrklock); \
1547 req->flags |= EIO_FLAG_PTR2_FREE; \
1548 X_UNLOCK (wrklock); \
1549 req->ptr2 = malloc (len); \
1550 if (!req->ptr2) \
1551 { \
779 errno = ENOMEM; \ 1552 errno = ENOMEM; \
780 req->result = -1; \ 1553 req->result = -1; \
781 break; \ 1554 break; \
1555 } \
782 } 1556 }
783 1557
784X_THREAD_PROC (eio_proc) 1558X_THREAD_PROC (etp_proc)
785{ 1559{
786 eio_req *req; 1560 ETP_REQ *req;
787 struct timespec ts; 1561 struct timespec ts;
788 worker *self = (worker *)thr_arg; 1562 etp_worker *self = (etp_worker *)thr_arg;
789 1563
790 /* try to distribute timeouts somewhat randomly */ 1564 /* try to distribute timeouts somewhat randomly */
791 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1565 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
792 1566
793 for (;;) 1567 for (;;)
794 { 1568 {
795 ts.tv_sec = time (0) + IDLE_TIMEOUT;
796
797 X_LOCK (reqlock); 1569 X_LOCK (reqlock);
798 1570
799 for (;;) 1571 for (;;)
800 { 1572 {
801 self->req = req = reqq_shift (&req_queue); 1573 self->req = req = reqq_shift (&req_queue);
… …
803 if (req) 1575 if (req)
804 break; 1576 break;
805 1577
806 ++idle; 1578 ++idle;
807 1579
1580 ts.tv_sec = time (0) + idle_timeout;
808 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1581 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
809 { 1582 {
810 if (idle > max_idle) 1583 if (idle > max_idle)
811 { 1584 {
812 --idle; 1585 --idle;
… …
817 goto quit; 1590 goto quit;
818 } 1591 }
819 1592
820 /* we are allowed to idle, so do so without any timeout */ 1593 /* we are allowed to idle, so do so without any timeout */
821 X_COND_WAIT (reqwait, reqlock); 1594 X_COND_WAIT (reqwait, reqlock);
822 ts.tv_sec = time (0) + IDLE_TIMEOUT;
823 } 1595 }
824 1596
825 --idle; 1597 --idle;
826 } 1598 }
827 1599
828 --nready; 1600 --nready;
829 1601
830 X_UNLOCK (reqlock); 1602 X_UNLOCK (reqlock);
831 1603
832 errno = 0; /* strictly unnecessary */ 1604 if (req->type < 0)
1605 goto quit;
833 1606
834 if (!EIO_CANCELLED (req)) 1607 if (!EIO_CANCELLED (req))
835 switch (req->type) 1608 ETP_EXECUTE (self, req);
836 {
837 case EIO_READ: req->result = req->offs >= 0
838 ? pread (req->int1, req->ptr2, req->size, req->offs)
839 : read (req->int1, req->ptr2, req->size); break;
840 case EIO_WRITE: req->result = req->offs >= 0
841 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
842 : write (req->int1, req->ptr2, req->size); break;
843
844 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
845 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
846
847 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
848 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
849 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
850 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
851 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
852 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
853
854 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
855 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
856 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
857 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
858 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
859 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
860
861 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
862 case EIO_CLOSE: req->result = close (req->int1); break;
863 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
864 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
865 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
866 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
867 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
868 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
869 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
870 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
871 case EIO_READLINK: ALLOC (NAME_MAX);
872 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
873
874 case EIO_SYNC: req->result = 0; sync (); break;
875 case EIO_FSYNC: req->result = fsync (req->int1); break;
876 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
877
878 case EIO_READDIR: scandir_ (req, self); break;
879
880 case EIO_BUSY:
881#ifdef _WIN32
882 Sleep (req->nv1 * 1000.);
883#else
884 {
885 struct timeval tv;
886
887 tv.tv_sec = req->nv1;
888 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
889
890 req->result = select (0, 0, 0, 0, &tv);
891 }
892#endif
893 break;
894
895 case EIO_UTIME:
896 case EIO_FUTIME:
897 {
898 struct timeval tv[2];
899 struct timeval *times;
900
901 if (req->nv1 != -1. || req->nv2 != -1.)
902 {
903 tv[0].tv_sec = req->nv1;
904 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
905 tv[1].tv_sec = req->nv2;
906 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
907
908 times = tv;
909 }
910 else
911 times = 0;
912
913
914 req->result = req->type == EIO_FUTIME
915 ? futimes (req->int1, times)
916 : utimes (req->ptr1, times);
917 }
918
919 case EIO_GROUP:
920 case EIO_NOP:
921 break;
922
923 case EIO_QUIT:
924 goto quit;
925
926 default:
927 req->result = -1;
928 break;
929 }
930
931 req->errorno = errno;
932 1609
933 X_LOCK (reslock); 1610 X_LOCK (reslock);
934 1611
935 ++npending; 1612 ++npending;
936 1613
937 if (!reqq_push (&res_queue, req) && want_poll_cb) 1614 if (!reqq_push (&res_queue, req) && want_poll_cb)
938 want_poll_cb (); 1615 want_poll_cb ();
939 1616
940 self->req = 0; 1617 self->req = 0;
941 worker_clear (self); 1618 etp_worker_clear (self);
942 1619
943 X_UNLOCK (reslock); 1620 X_UNLOCK (reslock);
944 } 1621 }
945 1622
946quit: 1623quit:
947 X_LOCK (wrklock); 1624 X_LOCK (wrklock);
948 worker_free (self); 1625 etp_worker_free (self);
949 X_UNLOCK (wrklock); 1626 X_UNLOCK (wrklock);
950 1627
951 return 0; 1628 return 0;
952} 1629}
953 1630
954/*****************************************************************************/ 1631/*****************************************************************************/
955 1632
956static void eio_atfork_prepare (void)
957{
958 X_LOCK (wrklock);
959 X_LOCK (reqlock);
960 X_LOCK (reslock);
961#if !HAVE_PREADWRITE
962 X_LOCK (preadwritelock);
963#endif
964#if !HAVE_READDIR_R
965 X_LOCK (readdirlock);
966#endif
967}
968
969static void eio_atfork_parent (void)
970{
971#if !HAVE_READDIR_R
972 X_UNLOCK (readdirlock);
973#endif
974#if !HAVE_PREADWRITE
975 X_UNLOCK (preadwritelock);
976#endif
977 X_UNLOCK (reslock);
978 X_UNLOCK (reqlock);
979 X_UNLOCK (wrklock);
980}
981
982static void eio_atfork_child (void)
983{
984 eio_req *prv;
985
986 while (prv = reqq_shift (&req_queue))
987 eio_destroy (prv);
988
989 while (prv = reqq_shift (&res_queue))
990 eio_destroy (prv);
991
992 while (wrk_first.next != &wrk_first)
993 {
994 worker *wrk = wrk_first.next;
995
996 if (wrk->req)
997 eio_destroy (wrk->req);
998
999 worker_clear (wrk);
1000 worker_free (wrk);
1001 }
1002
1003 started = 0;
1004 idle = 0;
1005 nreqs = 0;
1006 nready = 0;
1007 npending = 0;
1008
1009 eio_atfork_parent ();
1010}
1011
1012int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1633int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1013{ 1634{
1014 want_poll_cb = want_poll; 1635 return etp_init (want_poll, done_poll);
1015 done_poll_cb = done_poll;
1016
1017#ifdef _WIN32
1018 X_MUTEX_CHECK (wrklock);
1019 X_MUTEX_CHECK (reslock);
1020 X_MUTEX_CHECK (reqlock);
1021 X_MUTEX_CHECK (reqwait);
1022 X_MUTEX_CHECK (preadwritelock);
1023 X_MUTEX_CHECK (readdirlock);
1024
1025 X_COND_CHECK (reqwait);
1026#endif
1027
1028 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1029} 1636}
1030 1637
1031static void eio_api_destroy (eio_req *req) 1638static void eio_api_destroy (eio_req *req)
1032{ 1639{
1033 free (req); 1640 free (req);
… …
1038 \ 1645 \
1039 req = (eio_req *)calloc (1, sizeof *req); \ 1646 req = (eio_req *)calloc (1, sizeof *req); \
1040 if (!req) \ 1647 if (!req) \
1041 return 0; \ 1648 return 0; \
1042 \ 1649 \
1043 req->type = EIO_ ## rtype; \ 1650 req->type = rtype; \
1044 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 1651 req->pri = pri; \
1045 req->finish = cb; \ 1652 req->finish = cb; \
1653 req->data = data; \
1046 req->destroy = eio_api_destroy; 1654 req->destroy = eio_api_destroy;
1047 1655
1048#define SEND eio_submit (req); return req 1656#define SEND eio_submit (req); return req
1049 1657
1050#define PATH (void)0 1658#define PATH \
1659 req->flags |= EIO_FLAG_PTR1_FREE; \
1660 req->ptr1 = strdup (path); \
1661 if (!req->ptr1) \
1662 { \
1663 eio_api_destroy (req); \
1664 return 0; \
1665 }
1051 1666
1667static void eio_execute (etp_worker *self, eio_req *req)
1668{
1669 switch (req->type)
1670 {
1671 case EIO_READ: ALLOC (req->size);
1672 req->result = req->offs >= 0
1673 ? pread (req->int1, req->ptr2, req->size, req->offs)
1674 : read (req->int1, req->ptr2, req->size); break;
1675 case EIO_WRITE: req->result = req->offs >= 0
1676 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1677 : write (req->int1, req->ptr2, req->size); break;
1678
1679 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1680 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1681
1682 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1683 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1684 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1685 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1686 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1687 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1688
1689 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1690 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1691 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1692 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1693
1694 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1695 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1696 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1697 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1698 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1699 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1700
1701 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1702 case EIO_CLOSE: req->result = close (req->int1); break;
1703 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1704 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1705 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1706 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1707 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1708 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1709 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1710 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1711
1712 case EIO_READLINK: ALLOC (PATH_MAX);
1713 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1714
1715 case EIO_SYNC: req->result = 0; sync (); break;
1716 case EIO_FSYNC: req->result = fsync (req->int1); break;
1717 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1718 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1719 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1720 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1721 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1722 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1723
1724 case EIO_READDIR: eio__scandir (req, self); break;
1725
1726 case EIO_BUSY:
1727#ifdef _WIN32
1728 Sleep (req->nv1 * 1e3);
1729#else
1730 {
1731 struct timeval tv;
1732
1733 tv.tv_sec = req->nv1;
1734 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1735
1736 req->result = select (0, 0, 0, 0, &tv);
1737 }
1738#endif
1739 break;
1740
1741 case EIO_UTIME:
1742 case EIO_FUTIME:
1743 {
1744 struct timeval tv[2];
1745 struct timeval *times;
1746
1747 if (req->nv1 != -1. || req->nv2 != -1.)
1748 {
1749 tv[0].tv_sec = req->nv1;
1750 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1751 tv[1].tv_sec = req->nv2;
1752 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1753
1754 times = tv;
1755 }
1756 else
1757 times = 0;
1758
1759 req->result = req->type == EIO_FUTIME
1760 ? futimes (req->int1, times)
1761 : utimes (req->ptr1, times);
1762 }
1763 break;
1764
1765 case EIO_GROUP:
1766 abort (); /* handled in eio_request */
1767
1768 case EIO_NOP:
1769 req->result = 0;
1770 break;
1771
1772 case EIO_CUSTOM:
1773 ((void (*)(eio_req *))req->feed) (req);
1774 break;
1775
1776 default:
1777 errno = ENOSYS;
1778 req->result = -1;
1779 break;
1780 }
1781
1782 req->errorno = errno;
1783}
1784
1785#ifndef EIO_NO_WRAPPERS
1786
1787eio_req *eio_nop (int pri, eio_cb cb, void *data)
1788{
1789 REQ (EIO_NOP); SEND;
1790}
1791
1792eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1793{
1794 REQ (EIO_BUSY); req->nv1 = delay; SEND;
1795}
1796
1052eio_req *eio_fsync (int fd, eio_cb cb) 1797eio_req *eio_sync (int pri, eio_cb cb, void *data)
1053{ 1798{
1799 REQ (EIO_SYNC); SEND;
1800}
1801
1802eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1803{
1054 REQ (FSYNC); req->int1 = fd; SEND; 1804 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1055} 1805}
1056 1806
1807eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1808{
1809 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1810}
1811
1812eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1813{
1814 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1815}
1816
1817eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1818{
1819 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1820}
1821
1822eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1823{
1824 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1825}
1826
1827eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1828{
1829 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1830}
1831
1057eio_req *eio_fdatasync (int fd, eio_cb cb) 1832eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1058{ 1833{
1059 REQ (FDATASYNC); req->int1 = fd; SEND; 1834 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1060} 1835}
1061 1836
1837eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1838{
1839 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1840}
1841
1062eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb) 1842eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1063{ 1843{
1064 REQ (READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 1844 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1065} 1845}
1066 1846
1067eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb) 1847eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1068{ 1848{
1069 REQ (READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 1849 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1070} 1850}
1071 1851
1072eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb) 1852eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1073{ 1853{
1074 REQ (WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 1854 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1075} 1855}
1076 1856
1077eio_req *eio_fstat (int fd, eio_cb cb) 1857eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1078{ 1858{
1079 REQ (FSTAT); req->int1 = fd; SEND; 1859 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1080} 1860}
1081 1861
1862eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1863{
1864 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1865}
1866
1082eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb) 1867eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1083{ 1868{
1084 REQ (FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1869 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1085} 1870}
1086 1871
1087eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb) 1872eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1088{ 1873{
1089 REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 1874 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1090} 1875}
1091 1876
1092eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb) 1877eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1093{ 1878{
1094 REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 1879 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1095} 1880}
1096 1881
1097eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb) 1882eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1098{ 1883{
1099 REQ (FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 1884 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1100} 1885}
1101 1886
1102eio_req *eio_dup2 (int fd, int fd2, eio_cb cb) 1887eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1103{ 1888{
1104 REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND; 1889 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1105} 1890}
1106 1891
1107eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb) 1892eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1108{ 1893{
1109 REQ (SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 1894 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1110} 1895}
1111 1896
1112eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb) 1897eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1113{ 1898{
1114 REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 1899 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1115} 1900}
1116 1901
1117eio_req *eio_readlink (const char *path, eio_cb cb)
1118{
1119 REQ (READLINK); PATH; SEND;
1120}
1121
1122eio_req *eio_stat (const char *path, eio_cb cb)
1123{
1124 REQ (STAT); PATH; SEND;
1125}
1126
1127eio_req *eio_lstat (const char *path, eio_cb cb)
1128{
1129 REQ (LSTAT); PATH; SEND;
1130}
1131
1132eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb) 1902eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1133{ 1903{
1134 REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 1904 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1135} 1905}
1136 1906
1137eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb) 1907eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1138{ 1908{
1139 REQ (TRUNCATE); PATH; req->offs = offset; SEND; 1909 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1140} 1910}
1141 1911
1142eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 1912eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1143{ 1913{
1144 REQ (CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 1914 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1145} 1915}
1146 1916
1147eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb) 1917eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1148{ 1918{
1149 REQ (CHMOD); PATH; req->int2 = (long)mode; SEND; 1919 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1150} 1920}
1151 1921
1152eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb) 1922eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1153{ 1923{
1154 REQ (MKDIR); PATH; req->int2 = (long)mode; SEND; 1924 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1155} 1925}
1156 1926
1157eio_req *eio_unlink (const char *path, eio_cb cb) 1927static eio_req *
1928eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1158{ 1929{
1159 REQ (UNLINK); PATH; SEND;
1160}
1161
1162eio_req *eio_rmdir (const char *path, eio_cb cb)
1163{
1164 REQ (RMDIR); PATH; SEND; 1930 REQ (type); PATH; SEND;
1165} 1931}
1166 1932
1933eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1934{
1935 return eio__1path (EIO_READLINK, path, pri, cb, data);
1936}
1937
1938eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1939{
1940 return eio__1path (EIO_STAT, path, pri, cb, data);
1941}
1942
1943eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1944{
1945 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1946}
1947
1948eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1949{
1950 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1951}
1952
1953eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1954{
1955 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1956}
1957
1167eio_req *eio_readdir (const char *path, eio_cb cb) 1958eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1168{ 1959{
1169 REQ (READDIR); PATH; SEND; 1960 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1170} 1961}
1171 1962
1963eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1964{
1965 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1966}
1967
1172eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb) 1968eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1173{ 1969{
1174 REQ (MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1970 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1175} 1971}
1176 1972
1177eio_req *eio_busy (double delay, eio_cb cb) 1973static eio_req *
1974eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1178{ 1975{
1179 REQ (BUSY); req->nv1 = delay; SEND; 1976 REQ (type); PATH;
1180}
1181 1977
1978 req->flags |= EIO_FLAG_PTR2_FREE;
1979 req->ptr2 = strdup (new_path);
1980 if (!req->ptr2)
1981 {
1982 eio_api_destroy (req);
1983 return 0;
1984 }
1985
1986 SEND;
1987}
1988
1989eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1990{
1991 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
1992}
1993
1994eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1995{
1996 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
1997}
1998
1999eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2000{
2001 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2002}
2003
2004eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
2005{
2006 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
2007}
2008
2009#endif
2010
1182eio_req *eio_nop (eio_cb cb) 2011eio_req *eio_grp (eio_cb cb, void *data)
1183{ 2012{
2013 const int pri = EIO_PRI_MAX;
2014
1184 REQ (NOP); SEND; 2015 REQ (EIO_GROUP); SEND;
1185} 2016}
1186 2017
1187#undef REQ 2018#undef REQ
1188#undef PATH 2019#undef PATH
1189#undef SEND 2020#undef SEND
1190 2021
1191#if 0 2022/*****************************************************************************/
1192 2023/* grp functions */
1193void
1194aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1195 PROTOTYPE: $$$;$
1196 PPCODE:
1197{
1198 dREQ;
1199
1200 req->type = EIO_OPEN;
1201 req->sv1 = newSVsv (pathname);
1202 req->ptr1 = SvPVbyte_nolen (req->sv1);
1203 req->int1 = flags;
1204 req->int2 = mode;
1205
1206 EIO_SEND;
1207}
1208
1209void
1210aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1211 PROTOTYPE: $;$
1212 ALIAS:
1213 aio_fsync = EIO_FSYNC
1214 aio_fdatasync = EIO_FDATASYNC
1215 PPCODE:
1216{
1217 dREQ;
1218
1219 req->type = ix;
1220 req->sv1 = newSVsv (fh);
1221 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1222
1223 EIO_SEND (req);
1224}
1225
1226void
1227aio_close (SV *fh, SV *callback=&PL_sv_undef)
1228 PROTOTYPE: $;$
1229 PPCODE:
1230{
1231 dREQ;
1232
1233 req->type = EIO_CLOSE;
1234 req->sv1 = newSVsv (fh);
1235 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1236
1237 EIO_SEND (req);
1238}
1239
1240void
1241aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1242 ALIAS:
1243 aio_read = EIO_READ
1244 aio_write = EIO_WRITE
1245 PROTOTYPE: $$$$$;$
1246 PPCODE:
1247{
1248 STRLEN svlen;
1249 char *svptr = SvPVbyte (data, svlen);
1250 UV len = SvUV (length);
1251
1252 SvUPGRADE (data, SVt_PV);
1253 SvPOK_on (data);
1254
1255 if (dataoffset < 0)
1256 dataoffset += svlen;
1257
1258 if (dataoffset < 0 || dataoffset > svlen)
1259 croak ("dataoffset outside of data scalar");
1260
1261 if (ix == EIO_WRITE)
1262 {
1263 /* write: check length and adjust. */
1264 if (!SvOK (length) || len + dataoffset > svlen)
1265 len = svlen - dataoffset;
1266 }
1267 else
1268 {
1269 /* read: grow scalar as necessary */
1270 svptr = SvGROW (data, len + dataoffset + 1);
1271 }
1272
1273 if (len < 0)
1274 croak ("length must not be negative");
1275
1276 {
1277 dREQ;
1278
1279 req->type = ix;
1280 req->sv1 = newSVsv (fh);
1281 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1282 : IoOFP (sv_2io (fh)));
1283 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1284 req->size = len;
1285 req->sv2 = SvREFCNT_inc (data);
1286 req->ptr2 = (char *)svptr + dataoffset;
1287 req->stroffset = dataoffset;
1288
1289 if (!SvREADONLY (data))
1290 {
1291 SvREADONLY_on (data);
1292 req->flags |= FLAG_SV2_RO_OFF;
1293 }
1294
1295 EIO_SEND;
1296 }
1297}
1298
1299void
1300aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1301 PROTOTYPE: $$;$
1302 PPCODE:
1303{
1304 SV *data;
1305 dREQ;
1306
1307 data = newSV (NAME_MAX);
1308 SvPOK_on (data);
1309
1310 req->type = EIO_READLINK;
1311 req->sv1 = newSVsv (path);
1312 req->ptr1 = SvPVbyte_nolen (req->sv1);
1313 req->sv2 = data;
1314 req->ptr2 = SvPVbyte_nolen (data);
1315
1316 EIO_SEND;
1317}
1318
1319void
1320aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1321 PROTOTYPE: $$$$;$
1322 PPCODE:
1323{
1324 dREQ;
1325
1326 req->type = EIO_SENDFILE;
1327 req->sv1 = newSVsv (out_fh);
1328 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1329 req->sv2 = newSVsv (in_fh);
1330 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1331 req->offs = SvVAL64 (in_offset);
1332 req->size = length;
1333
1334 EIO_SEND;
1335}
1336
1337void
1338aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1339 PROTOTYPE: $$$;$
1340 PPCODE:
1341{
1342 dREQ;
1343
1344 req->type = EIO_READAHEAD;
1345 req->sv1 = newSVsv (fh);
1346 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1347 req->offs = SvVAL64 (offset);
1348 req->size = length;
1349
1350 EIO_SEND;
1351}
1352
1353void
1354aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1355 ALIAS:
1356 aio_stat = EIO_STAT
1357 aio_lstat = EIO_LSTAT
1358 PPCODE:
1359{
1360 dREQ;
1361
1362 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1363 if (!req->ptr2)
1364 {
1365 req_destroy (req);
1366 croak ("out of memory during aio_stat statdata allocation");
1367 }
1368
1369 req->flags |= FLAG_PTR2_FREE;
1370 req->sv1 = newSVsv (fh_or_path);
1371
1372 if (SvPOK (fh_or_path))
1373 {
1374 req->type = ix;
1375 req->ptr1 = SvPVbyte_nolen (req->sv1);
1376 }
1377 else
1378 {
1379 req->type = EIO_FSTAT;
1380 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1381 }
1382
1383 EIO_SEND;
1384}
1385
1386void
1387aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1388 PPCODE:
1389{
1390 dREQ;
1391
1392 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1393 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1394 req->sv1 = newSVsv (fh_or_path);
1395
1396 if (SvPOK (fh_or_path))
1397 {
1398 req->type = EIO_UTIME;
1399 req->ptr1 = SvPVbyte_nolen (req->sv1);
1400 }
1401 else
1402 {
1403 req->type = EIO_FUTIME;
1404 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1405 }
1406
1407 EIO_SEND;
1408}
1409
1410void
1411aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1412 PPCODE:
1413{
1414 dREQ;
1415
1416 req->sv1 = newSVsv (fh_or_path);
1417 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1418
1419 if (SvPOK (fh_or_path))
1420 {
1421 req->type = EIO_TRUNCATE;
1422 req->ptr1 = SvPVbyte_nolen (req->sv1);
1423 }
1424 else
1425 {
1426 req->type = EIO_FTRUNCATE;
1427 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1428 }
1429
1430 EIO_SEND;
1431}
1432
1433void
1434aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1435 ALIAS:
1436 aio_chmod = EIO_CHMOD
1437 aio_fchmod = EIO_FCHMOD
1438 aio_mkdir = EIO_MKDIR
1439 PPCODE:
1440{
1441 dREQ;
1442
1443 req->type = type;
1444 req->int2 = mode;
1445 req->sv1 = newSVsv (fh_or_path);
1446
1447 if (ix == EIO_FCHMOD)
1448 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1449 else
1450 req->ptr1 = SvPVbyte_nolen (req->sv1);
1451
1452 EIO_SEND;
1453}
1454
1455void
1456aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1457 PPCODE:
1458{
1459 dREQ;
1460
1461 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1462 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1463 req->sv1 = newSVsv (fh_or_path);
1464
1465 if (SvPOK (fh_or_path))
1466 {
1467 req->type = EIO_CHOWN;
1468 req->ptr1 = SvPVbyte_nolen (req->sv1);
1469 }
1470 else
1471 {
1472 req->type = EIO_FCHOWN;
1473 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1474 }
1475
1476 EIO_SEND;
1477}
1478
1479void
1480aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1481 ALIAS:
1482 aio_unlink = EIO_UNLINK
1483 aio_rmdir = EIO_RMDIR
1484 aio_readdir = EIO_READDIR
1485 PPCODE:
1486{
1487 dREQ;
1488
1489 req->type = ix;
1490 req->sv1 = newSVsv (pathname);
1491 req->ptr1 = SvPVbyte_nolen (req->sv1);
1492
1493 EIO_SEND;
1494}
1495
1496void
1497aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1498 ALIAS:
1499 aio_link = EIO_LINK
1500 aio_symlink = EIO_SYMLINK
1501 aio_rename = EIO_RENAME
1502 PPCODE:
1503{
1504 dREQ;
1505
1506 req->type = ix;
1507 req->sv1 = newSVsv (oldpath);
1508 req->ptr1 = SvPVbyte_nolen (req->sv1);
1509 req->sv2 = newSVsv (newpath);
1510 req->ptr2 = SvPVbyte_nolen (req->sv2);
1511
1512 EIO_SEND;
1513}
1514
1515void
1516aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1517 PPCODE:
1518{
1519 dREQ;
1520
1521 req->type = EIO_MKNOD;
1522 req->sv1 = newSVsv (pathname);
1523 req->ptr1 = SvPVbyte_nolen (req->sv1);
1524 req->int2 = (mode_t)mode;
1525 req->offs = dev;
1526
1527 EIO_SEND;
1528}
1529
1530void
1531aio_busy (double delay, SV *callback=&PL_sv_undef)
1532 PPCODE:
1533{
1534 dREQ;
1535
1536 req->type = EIO_BUSY;
1537 req->nv1 = delay < 0. ? 0. : delay;
1538
1539 EIO_SEND;
1540}
1541
1542void
1543aio_group (SV *callback=&PL_sv_undef)
1544 PROTOTYPE: ;$
1545 PPCODE:
1546{
1547 dREQ;
1548
1549 req->type = EIO_GROUP;
1550
1551 req_send (req);
1552 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1553}
1554
1555void
1556aio_nop (SV *callback=&PL_sv_undef)
1557 ALIAS:
1558 aio_nop = EIO_NOP
1559 aio_sync = EIO_SYNC
1560 PPCODE:
1561{
1562 dREQ;
1563
1564#endif
1565 2024
1566void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2025void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1567{ 2026{
1568 grp->int2 = limit; 2027 grp->int2 = limit;
1569 grp->feed = feed; 2028 grp->feed = feed;
… …
1580 2039
1581void eio_grp_add (eio_req *grp, eio_req *req) 2040void eio_grp_add (eio_req *grp, eio_req *req)
1582{ 2041{
1583 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2042 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1584 2043
2044 grp->flags |= EIO_FLAG_GROUPADD;
2045
1585 ++grp->size; 2046 ++grp->size;
1586 req->grp = grp; 2047 req->grp = grp;
1587 2048
1588 req->grp_prev = 0; 2049 req->grp_prev = 0;
1589 req->grp_next = grp->grp_first; 2050 req->grp_next = grp->grp_first;
… …
1592 grp->grp_first->grp_prev = req; 2053 grp->grp_first->grp_prev = req;
1593 2054
1594 grp->grp_first = req; 2055 grp->grp_first = req;
1595} 2056}
1596 2057
2058/*****************************************************************************/
2059/* misc garbage */
1597 2060
2061ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2062{
2063 etp_worker wrk;
2064 ssize_t ret;
2065
2066 wrk.dbuf = 0;
2067
2068 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2069
2070 if (wrk.dbuf)
2071 free (wrk.dbuf);
2072
2073 return ret;
2074}
2075

Diff Legend

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