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

Comparing libeio/eio.c (file contents):
Revision 1.8 by root, Sun May 11 20:08:37 2008 UTC vs.
Revision 1.64 by root, Thu May 26 04:05:18 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
1#include "eio.h" 40#include "eio.h"
41
42#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif
2#include "xthread.h" 45#include "xthread.h"
3 46
4#include <errno.h> 47#include <errno.h>
5#include <stddef.h> 48#include <stddef.h>
6#include <stdlib.h> 49#include <stdlib.h>
7#include <string.h> 50#include <string.h>
8#include <errno.h> 51#include <errno.h>
9#include <sys/types.h> 52#include <sys/types.h>
10#include <sys/stat.h> 53#include <sys/stat.h>
54#include <sys/statvfs.h>
11#include <limits.h> 55#include <limits.h>
12#include <fcntl.h> 56#include <fcntl.h>
13#include <assert.h> 57#include <assert.h>
14 58
15#ifndef EIO_FINISH 59#ifndef EIO_FINISH
… …
25#endif 69#endif
26 70
27#ifdef _WIN32 71#ifdef _WIN32
28 72
29 /*doh*/ 73 /*doh*/
30
31#else 74#else
32 75
33# include "config.h" 76# include "config.h"
34# include <sys/time.h> 77# include <sys/time.h>
35# include <sys/select.h> 78# include <sys/select.h>
36# include <unistd.h> 79# include <unistd.h>
37# include <utime.h> 80# include <utime.h>
38# include <signal.h> 81# include <signal.h>
39# include <dirent.h> 82# include <dirent.h>
40 83
84#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
85# include <sys/mman.h>
86#endif
87
88/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
89# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
90# define _DIRENT_HAVE_D_TYPE /* sigh */
91# define D_INO(de) (de)->d_fileno
92# define D_NAMLEN(de) (de)->d_namlen
93# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
94# define D_INO(de) (de)->d_ino
95# endif
96
97#ifdef _D_EXACT_NAMLEN
98# undef D_NAMLEN
99# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
100#endif
101
102# ifdef _DIRENT_HAVE_D_TYPE
103# define D_TYPE(de) (de)->d_type
104# endif
105
41# ifndef EIO_STRUCT_DIRENT 106# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 107# define EIO_STRUCT_DIRENT struct dirent
43# endif 108# endif
44 109
45#endif 110#endif
46 111
47# ifndef EIO_STRUCT_STAT
48# define EIO_STRUCT_STAT struct stat
49# endif
50
51#if HAVE_SENDFILE 112#if HAVE_SENDFILE
52# if __linux 113# if __linux
53# include <sys/sendfile.h> 114# include <sys/sendfile.h>
54# elif __freebsd 115# elif __FreeBSD__ || defined __APPLE__
55# include <sys/socket.h> 116# include <sys/socket.h>
56# include <sys/uio.h> 117# include <sys/uio.h>
57# elif __hpux 118# elif __hpux
58# include <sys/socket.h> 119# include <sys/socket.h>
59# elif __solaris /* not yet */ 120# elif __solaris
60# include <sys/sendfile.h> 121# include <sys/sendfile.h>
61# else 122# else
62# error sendfile support requested but not available 123# error sendfile support requested but not available
63# endif 124# endif
64#endif 125#endif
65 126
66/* number of seconds after which an idle threads exit */ 127#ifndef D_TYPE
67#define IDLE_TIMEOUT 10 128# define D_TYPE(de) 0
129#endif
130#ifndef D_INO
131# define D_INO(de) 0
132#endif
133#ifndef D_NAMLEN
134# define D_NAMLEN(de) strlen ((de)->d_name)
135#endif
68 136
69/* used for struct dirent, AIX doesn't provide it */ 137/* used for struct dirent, AIX doesn't provide it */
70#ifndef NAME_MAX 138#ifndef NAME_MAX
71# define NAME_MAX 4096 139# define NAME_MAX 4096
72#endif 140#endif
73 141
142/* used for readlink etc. */
143#ifndef PATH_MAX
144# define PATH_MAX 4096
145#endif
146
74/* buffer size for various temporary buffers */ 147/* buffer size for various temporary buffers */
75#define EIO_BUFSIZE 65536 148#define EIO_BUFSIZE 65536
76 149
77#define dBUF \ 150#define dBUF \
78 char *eio_buf; \ 151 char *eio_buf; \
79 X_LOCK (wrklock); \ 152 ETP_WORKER_LOCK (self); \
80 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 153 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
81 X_UNLOCK (wrklock); \ 154 ETP_WORKER_UNLOCK (self); \
155 errno = ENOMEM; \
82 if (!eio_buf) \ 156 if (!eio_buf) \
83 return -1; 157 return -1;
84 158
85#define EIO_TICKS ((1000000 + 1023) >> 10) 159#define EIO_TICKS ((1000000 + 1023) >> 10)
86 160
161/*****************************************************************************/
162
163#if __GNUC__ >= 3
164# define expect(expr,value) __builtin_expect ((expr),(value))
165#else
166# define expect(expr,value) (expr)
167#endif
168
169#define expect_false(expr) expect ((expr) != 0, 0)
170#define expect_true(expr) expect ((expr) != 0, 1)
171
172/*****************************************************************************/
173
174#define ETP_PRI_MIN EIO_PRI_MIN
175#define ETP_PRI_MAX EIO_PRI_MAX
176
177struct etp_worker;
178
179#define ETP_REQ eio_req
180#define ETP_DESTROY(req) eio_destroy (req)
181static int eio_finish (eio_req *req);
182#define ETP_FINISH(req) eio_finish (req)
183static void eio_execute (struct etp_worker *self, eio_req *req);
184#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
185
186#define ETP_WORKER_CLEAR(req) \
187 if (wrk->dbuf) \
188 { \
189 free (wrk->dbuf); \
190 wrk->dbuf = 0; \
191 } \
192 \
193 if (wrk->dirp) \
194 { \
195 closedir (wrk->dirp); \
196 wrk->dirp = 0; \
197 }
198
199#define ETP_WORKER_COMMON \
200 void *dbuf; \
201 DIR *dirp;
202
203/*****************************************************************************/
204
205#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
206
207/* calculate time difference in ~1/EIO_TICKS of a second */
208static int tvdiff (struct timeval *tv1, struct timeval *tv2)
209{
210 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
211 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
212}
213
214static unsigned int started, idle, wanted = 4;
215
87static void (*want_poll_cb) (void); 216static void (*want_poll_cb) (void);
88static void (*done_poll_cb) (void); 217static void (*done_poll_cb) (void);
89 218
90static unsigned int max_poll_time = 0; 219static unsigned int max_poll_time; /* reslock */
91static unsigned int max_poll_reqs = 0; 220static unsigned int max_poll_reqs; /* reslock */
92 221
93/* calculcate time difference in ~1/EIO_TICKS of a second */ 222static volatile unsigned int nreqs; /* reqlock */
94static int tvdiff (struct timeval *tv1, struct timeval *tv2) 223static volatile unsigned int nready; /* reqlock */
95{ 224static volatile unsigned int npending; /* reqlock */
96 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 225static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 226static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98}
99 227
100static unsigned int started, idle, wanted; 228static xmutex_t wrklock;
229static xmutex_t reslock;
230static xmutex_t reqlock;
231static xcond_t reqwait;
232
233#if !HAVE_PREADWRITE
234/*
235 * make our pread/pwrite emulation safe against themselves, but not against
236 * normal read/write by using a mutex. slows down execution a lot,
237 * but that's your problem, not mine.
238 */
239static xmutex_t preadwritelock = X_MUTEX_INIT;
240#endif
241
242typedef struct etp_worker
243{
244 /* locked by wrklock */
245 struct etp_worker *prev, *next;
246
247 xthread_t tid;
248
249 /* locked by reslock, reqlock or wrklock */
250 ETP_REQ *req; /* currently processed request */
251
252 ETP_WORKER_COMMON
253} etp_worker;
254
255static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
256
257#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
258#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
101 259
102/* worker threads management */ 260/* worker threads management */
103static mutex_t wrklock = X_MUTEX_INIT;
104 261
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) 262static void etp_worker_clear (etp_worker *wrk)
120{ 263{
121 if (wrk->dirp) 264 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} 265}
133 266
134static void worker_free (worker *wrk) 267static void etp_worker_free (etp_worker *wrk)
135{ 268{
136 wrk->next->prev = wrk->prev; 269 wrk->next->prev = wrk->prev;
137 wrk->prev->next = wrk->next; 270 wrk->prev->next = wrk->next;
138 271
139 free (wrk); 272 free (wrk);
140} 273}
141 274
142static volatile unsigned int nreqs, nready, npending;
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) 275static unsigned int etp_nreqs (void)
150{ 276{
277 int retval;
278 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
279 retval = nreqs;
280 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151 return nreqs; 281 return retval;
152} 282}
153 283
154unsigned int eio_nready (void) 284static unsigned int etp_nready (void)
155{ 285{
156 unsigned int retval; 286 unsigned int retval;
157 287
158 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 288 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159 retval = nready; 289 retval = nready;
160 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 290 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161 291
162 return retval; 292 return retval;
163} 293}
164 294
165unsigned int eio_npending (void) 295static unsigned int etp_npending (void)
166{ 296{
167 unsigned int retval; 297 unsigned int retval;
168 298
169 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 299 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170 retval = npending; 300 retval = npending;
171 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 301 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172 302
173 return retval; 303 return retval;
174} 304}
175 305
176unsigned int eio_nthreads (void) 306static unsigned int etp_nthreads (void)
177{ 307{
178 unsigned int retval; 308 unsigned int retval;
179 309
180 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 310 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181 retval = started; 311 retval = started;
… …
188 * a somewhat faster data structure might be nice, but 318 * a somewhat faster data structure might be nice, but
189 * with 8 priorities this actually needs <20 insns 319 * with 8 priorities this actually needs <20 insns
190 * per shift, the most expensive operation. 320 * per shift, the most expensive operation.
191 */ 321 */
192typedef struct { 322typedef struct {
193 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 323 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194 int size; 324 int size;
195} reqq; 325} etp_reqq;
196 326
197static reqq req_queue; 327static etp_reqq req_queue;
198static reqq res_queue; 328static etp_reqq res_queue;
199 329
200static int reqq_push (reqq *q, eio_req *req) 330static int reqq_push (etp_reqq *q, ETP_REQ *req)
201{ 331{
202 int pri = req->pri; 332 int pri = req->pri;
203 req->next = 0; 333 req->next = 0;
204 334
205 if (q->qe[pri]) 335 if (q->qe[pri])
… …
211 q->qe[pri] = q->qs[pri] = req; 341 q->qe[pri] = q->qs[pri] = req;
212 342
213 return q->size++; 343 return q->size++;
214} 344}
215 345
216static eio_req *reqq_shift (reqq *q) 346static ETP_REQ *reqq_shift (etp_reqq *q)
217{ 347{
218 int pri; 348 int pri;
219 349
220 if (!q->size) 350 if (!q->size)
221 return 0; 351 return 0;
222 352
223 --q->size; 353 --q->size;
224 354
225 for (pri = EIO_NUM_PRI; pri--; ) 355 for (pri = ETP_NUM_PRI; pri--; )
226 { 356 {
227 eio_req *req = q->qs[pri]; 357 eio_req *req = q->qs[pri];
228 358
229 if (req) 359 if (req)
230 { 360 {
… …
236 } 366 }
237 367
238 abort (); 368 abort ();
239} 369}
240 370
241static void grp_try_feed (eio_req *grp) 371static void etp_thread_init (void)
242{ 372{
243 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 373 X_MUTEX_CREATE (wrklock);
374 X_MUTEX_CREATE (reslock);
375 X_MUTEX_CREATE (reqlock);
376 X_COND_CREATE (reqwait);
377}
378
379static void etp_atfork_prepare (void)
380{
381 X_LOCK (wrklock);
382 X_LOCK (reqlock);
383 X_LOCK (reslock);
384#if !HAVE_PREADWRITE
385 X_LOCK (preadwritelock);
386#endif
387}
388
389static void etp_atfork_parent (void)
390{
391#if !HAVE_PREADWRITE
392 X_UNLOCK (preadwritelock);
393#endif
394 X_UNLOCK (reslock);
395 X_UNLOCK (reqlock);
396 X_UNLOCK (wrklock);
397}
398
399static void etp_atfork_child (void)
400{
401 ETP_REQ *prv;
402
403 while ((prv = reqq_shift (&req_queue)))
404 ETP_DESTROY (prv);
405
406 while ((prv = reqq_shift (&res_queue)))
407 ETP_DESTROY (prv);
408
409 while (wrk_first.next != &wrk_first)
244 { 410 {
245 int old_len = grp->size; 411 etp_worker *wrk = wrk_first.next;
246 412
247 EIO_FEED (grp); 413 if (wrk->req)
414 ETP_DESTROY (wrk->req);
248 415
249 /* stop if no progress has been made */ 416 etp_worker_clear (wrk);
250 if (old_len == grp->size) 417 etp_worker_free (wrk);
251 {
252 grp->feed = 0;
253 break;
254 }
255 } 418 }
256}
257 419
258static int eio_finish (eio_req *req); 420 started = 0;
421 idle = 0;
422 nreqs = 0;
423 nready = 0;
424 npending = 0;
259 425
260static int grp_dec (eio_req *grp) 426 etp_thread_init ();
261{ 427}
262 --grp->size;
263 428
264 /* call feeder, if applicable */ 429static void
265 grp_try_feed (grp); 430etp_once_init (void)
431{
432 etp_thread_init ();
433 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
434}
266 435
267 /* finish, if done */ 436static int
268 if (!grp->size && grp->int1) 437etp_init (void (*want_poll)(void), void (*done_poll)(void))
269 return eio_finish (grp); 438{
270 else 439 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
440
441 pthread_once (&doinit, etp_once_init);
442
443 want_poll_cb = want_poll;
444 done_poll_cb = done_poll;
445
271 return 0; 446 return 0;
272} 447}
273 448
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); 449X_THREAD_PROC (etp_proc);
325 450
326static void start_thread (void) 451static void etp_start_thread (void)
327{ 452{
328 worker *wrk = calloc (1, sizeof (worker)); 453 etp_worker *wrk = calloc (1, sizeof (etp_worker));
329 454
330 /*TODO*/ 455 /*TODO*/
331 assert (("unable to allocate worker thread data", !wrk)); 456 assert (("unable to allocate worker thread data", wrk));
332 457
333 X_LOCK (wrklock); 458 X_LOCK (wrklock);
334 459
335 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 460 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
336 { 461 {
337 wrk->prev = &wrk_first; 462 wrk->prev = &wrk_first;
338 wrk->next = wrk_first.next; 463 wrk->next = wrk_first.next;
339 wrk_first.next->prev = wrk; 464 wrk_first.next->prev = wrk;
340 wrk_first.next = wrk; 465 wrk_first.next = wrk;
… …
344 free (wrk); 469 free (wrk);
345 470
346 X_UNLOCK (wrklock); 471 X_UNLOCK (wrklock);
347} 472}
348 473
349static void maybe_start_thread (void) 474static void etp_maybe_start_thread (void)
350{ 475{
351 if (eio_nthreads () >= wanted) 476 if (expect_true (etp_nthreads () >= wanted))
352 return; 477 return;
353 478
354 /* todo: maybe use idle here, but might be less exact */ 479 /* todo: maybe use idle here, but might be less exact */
355 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 480 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
356 return; 481 return;
357 482
358 start_thread (); 483 etp_start_thread ();
359} 484}
360 485
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) 486static void etp_end_thread (void)
375{ 487{
376 eio_req *req = calloc (1, sizeof (eio_req)); 488 eio_req *req = calloc (1, sizeof (eio_req));
377 489
378 req->type = EIO_QUIT; 490 req->type = -1;
379 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 491 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
380 492
381 X_LOCK (reqlock); 493 X_LOCK (reqlock);
382 reqq_push (&req_queue, req); 494 reqq_push (&req_queue, req);
383 X_COND_SIGNAL (reqwait); 495 X_COND_SIGNAL (reqwait);
384 X_UNLOCK (reqlock); 496 X_UNLOCK (reqlock);
… …
386 X_LOCK (wrklock); 498 X_LOCK (wrklock);
387 --started; 499 --started;
388 X_UNLOCK (wrklock); 500 X_UNLOCK (wrklock);
389} 501}
390 502
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) 503static int etp_poll (void)
428{ 504{
429 int maxreqs = max_poll_reqs; 505 unsigned int maxreqs;
506 unsigned int maxtime;
430 struct timeval tv_start, tv_now; 507 struct timeval tv_start, tv_now;
431 eio_req *req;
432 508
509 X_LOCK (reslock);
510 maxreqs = max_poll_reqs;
511 maxtime = max_poll_time;
512 X_UNLOCK (reslock);
513
433 if (max_poll_time) 514 if (maxtime)
434 gettimeofday (&tv_start, 0); 515 gettimeofday (&tv_start, 0);
435 516
436 for (;;) 517 for (;;)
437 { 518 {
519 ETP_REQ *req;
520
438 maybe_start_thread (); 521 etp_maybe_start_thread ();
439 522
440 X_LOCK (reslock); 523 X_LOCK (reslock);
441 req = reqq_shift (&res_queue); 524 req = reqq_shift (&res_queue);
442 525
443 if (req) 526 if (req)
… …
451 X_UNLOCK (reslock); 534 X_UNLOCK (reslock);
452 535
453 if (!req) 536 if (!req)
454 return 0; 537 return 0;
455 538
539 X_LOCK (reqlock);
456 --nreqs; 540 --nreqs;
541 X_UNLOCK (reqlock);
457 542
458 if (req->type == EIO_GROUP && req->size) 543 if (expect_false (req->type == EIO_GROUP && req->size))
459 { 544 {
460 req->int1 = 1; /* mark request as delayed */ 545 req->int1 = 1; /* mark request as delayed */
461 continue; 546 continue;
462 } 547 }
463 else 548 else
464 { 549 {
465 int res = eio_finish (req); 550 int res = ETP_FINISH (req);
466 if (res) 551 if (expect_false (res))
467 return res; 552 return res;
468 } 553 }
469 554
470 if (maxreqs && !--maxreqs) 555 if (expect_false (maxreqs && !--maxreqs))
471 break; 556 break;
472 557
473 if (max_poll_time) 558 if (maxtime)
474 { 559 {
475 gettimeofday (&tv_now, 0); 560 gettimeofday (&tv_now, 0);
476 561
477 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 562 if (tvdiff (&tv_start, &tv_now) >= maxtime)
478 break; 563 break;
479 } 564 }
480 } 565 }
481 566
482 errno = EAGAIN; 567 errno = EAGAIN;
483 return -1; 568 return -1;
484} 569}
485 570
571static void etp_cancel (ETP_REQ *req)
572{
573 X_LOCK (wrklock);
574 req->flags |= EIO_FLAG_CANCELLED;
575 X_UNLOCK (wrklock);
576
577 eio_grp_cancel (req);
578}
579
580static void etp_submit (ETP_REQ *req)
581{
582 req->pri -= ETP_PRI_MIN;
583
584 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
585 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
586
587 if (expect_false (req->type == EIO_GROUP))
588 {
589 /* I hope this is worth it :/ */
590 X_LOCK (reqlock);
591 ++nreqs;
592 X_UNLOCK (reqlock);
593
594 X_LOCK (reslock);
595
596 ++npending;
597
598 if (!reqq_push (&res_queue, req) && want_poll_cb)
599 want_poll_cb ();
600
601 X_UNLOCK (reslock);
602 }
603 else
604 {
605 X_LOCK (reqlock);
606 ++nreqs;
607 ++nready;
608 reqq_push (&req_queue, req);
609 X_COND_SIGNAL (reqwait);
610 X_UNLOCK (reqlock);
611
612 etp_maybe_start_thread ();
613 }
614}
615
616static void etp_set_max_poll_time (double nseconds)
617{
618 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
619 max_poll_time = nseconds * EIO_TICKS;
620 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
621}
622
623static void etp_set_max_poll_reqs (unsigned int maxreqs)
624{
625 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
626 max_poll_reqs = maxreqs;
627 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
628}
629
630static void etp_set_max_idle (unsigned int nthreads)
631{
632 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
633 max_idle = nthreads;
634 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
635}
636
637static void etp_set_idle_timeout (unsigned int seconds)
638{
639 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
640 idle_timeout = seconds;
641 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
642}
643
644static void etp_set_min_parallel (unsigned int nthreads)
645{
646 if (wanted < nthreads)
647 wanted = nthreads;
648}
649
650static void etp_set_max_parallel (unsigned int nthreads)
651{
652 if (wanted > nthreads)
653 wanted = nthreads;
654
655 while (started > wanted)
656 etp_end_thread ();
657}
658
659/*****************************************************************************/
660
661static void grp_try_feed (eio_req *grp)
662{
663 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
664 {
665 grp->flags &= ~EIO_FLAG_GROUPADD;
666
667 EIO_FEED (grp);
668
669 /* stop if no progress has been made */
670 if (!(grp->flags & EIO_FLAG_GROUPADD))
671 {
672 grp->feed = 0;
673 break;
674 }
675 }
676}
677
678static int grp_dec (eio_req *grp)
679{
680 --grp->size;
681
682 /* call feeder, if applicable */
683 grp_try_feed (grp);
684
685 /* finish, if done */
686 if (!grp->size && grp->int1)
687 return eio_finish (grp);
688 else
689 return 0;
690}
691
692void eio_destroy (eio_req *req)
693{
694 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
695 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
696
697 EIO_DESTROY (req);
698}
699
700static int eio_finish (eio_req *req)
701{
702 int res = EIO_FINISH (req);
703
704 if (req->grp)
705 {
706 int res2;
707 eio_req *grp = req->grp;
708
709 /* unlink request */
710 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
711 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
712
713 if (grp->grp_first == req)
714 grp->grp_first = req->grp_next;
715
716 res2 = grp_dec (grp);
717
718 if (!res && res2)
719 res = res2;
720 }
721
722 eio_destroy (req);
723
724 return res;
725}
726
727void eio_grp_cancel (eio_req *grp)
728{
729 for (grp = grp->grp_first; grp; grp = grp->grp_next)
730 eio_cancel (grp);
731}
732
733void eio_cancel (eio_req *req)
734{
735 etp_cancel (req);
736}
737
738void eio_submit (eio_req *req)
739{
740 etp_submit (req);
741}
742
743unsigned int eio_nreqs (void)
744{
745 return etp_nreqs ();
746}
747
748unsigned int eio_nready (void)
749{
750 return etp_nready ();
751}
752
753unsigned int eio_npending (void)
754{
755 return etp_npending ();
756}
757
758unsigned int eio_nthreads (void)
759{
760 return etp_nthreads ();
761}
762
763void eio_set_max_poll_time (double nseconds)
764{
765 etp_set_max_poll_time (nseconds);
766}
767
768void eio_set_max_poll_reqs (unsigned int maxreqs)
769{
770 etp_set_max_poll_reqs (maxreqs);
771}
772
773void eio_set_max_idle (unsigned int nthreads)
774{
775 etp_set_max_idle (nthreads);
776}
777
778void eio_set_idle_timeout (unsigned int seconds)
779{
780 etp_set_idle_timeout (seconds);
781}
782
783void eio_set_min_parallel (unsigned int nthreads)
784{
785 etp_set_min_parallel (nthreads);
786}
787
788void eio_set_max_parallel (unsigned int nthreads)
789{
790 etp_set_max_parallel (nthreads);
791}
792
793int eio_poll (void)
794{
795 return etp_poll ();
796}
797
486/*****************************************************************************/ 798/*****************************************************************************/
487/* work around various missing functions */ 799/* work around various missing functions */
488 800
489#if !HAVE_PREADWRITE 801#if !HAVE_PREADWRITE
802# undef pread
803# undef pwrite
490# define pread aio_pread 804# define pread eio__pread
491# define pwrite aio_pwrite 805# define pwrite eio__pwrite
492 806
493/* 807static ssize_t
494 * make our pread/pwrite safe against themselves, but not against 808eio__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{ 809{
502 ssize_t res; 810 ssize_t res;
503 off_t ooffset; 811 off_t ooffset;
504 812
505 X_LOCK (preadwritelock); 813 X_LOCK (preadwritelock);
… …
510 X_UNLOCK (preadwritelock); 818 X_UNLOCK (preadwritelock);
511 819
512 return res; 820 return res;
513} 821}
514 822
515static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 823static ssize_t
824eio__pwrite (int fd, void *buf, size_t count, off_t offset)
516{ 825{
517 ssize_t res; 826 ssize_t res;
518 off_t ooffset; 827 off_t ooffset;
519 828
520 X_LOCK (preadwritelock); 829 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR); 830 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET); 831 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count); 832 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET); 833 lseek (fd, ooffset, SEEK_SET);
525 X_UNLOCK (preadwritelock); 834 X_UNLOCK (preadwritelock);
526 835
527 return res; 836 return res;
528} 837}
529#endif 838#endif
530 839
531#ifndef HAVE_FUTIMES 840#ifndef HAVE_UTIMES
532 841
842# undef utimes
533# define utimes(path,times) aio_utimes (path, times) 843# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) aio_futimes (fd, times)
535 844
845static int
536static int aio_utimes (const char *filename, const struct timeval times[2]) 846eio__utimes (const char *filename, const struct timeval times[2])
537{ 847{
538 if (times) 848 if (times)
539 { 849 {
540 struct utimbuf buf; 850 struct utimbuf buf;
541 851
… …
546 } 856 }
547 else 857 else
548 return utime (filename, 0); 858 return utime (filename, 0);
549} 859}
550 860
861#endif
862
863#ifndef HAVE_FUTIMES
864
865# undef futimes
866# define futimes(fd,times) eio__futimes (fd, times)
867
551static int aio_futimes (int fd, const struct timeval tv[2]) 868static int eio__futimes (int fd, const struct timeval tv[2])
552{ 869{
553 errno = ENOSYS; 870 errno = ENOSYS;
554 return -1; 871 return -1;
555} 872}
556 873
557#endif 874#endif
558 875
559#if !HAVE_FDATASYNC 876#if !HAVE_FDATASYNC
877# undef fdatasync
560# define fdatasync fsync 878# define fdatasync(fd) fsync (fd)
561#endif 879#endif
880
881/* sync_file_range always needs emulation */
882int
883eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
884{
885#if HAVE_SYNC_FILE_RANGE
886 int res;
887
888 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
889 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
890 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
891 {
892 flags = 0
893 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
894 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
895 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
896 }
897
898 res = sync_file_range (fd, offset, nbytes, flags);
899
900 if (!res || errno != ENOSYS)
901 return res;
902#endif
903
904 /* even though we could play tricks with the flags, it's better to always
905 * call fdatasync, as that matches the expectation of its users best */
906 return fdatasync (fd);
907}
562 908
563#if !HAVE_READAHEAD 909#if !HAVE_READAHEAD
910# undef readahead
564# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 911# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565 912
913static ssize_t
566static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 914eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
567{ 915{
568 size_t todo = count; 916 size_t todo = count;
569 dBUF; 917 dBUF;
570 918
571 while (todo > 0) 919 while (todo > 0)
… …
581 return count; 929 return count;
582} 930}
583 931
584#endif 932#endif
585 933
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 */ 934/* sendfile always needs emulation */
935static ssize_t
617static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 936eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
618{ 937{
619 ssize_t res; 938 ssize_t res;
620 939
621 if (!count) 940 if (!count)
622 return 0; 941 return 0;
623 942
624#if HAVE_SENDFILE 943#if HAVE_SENDFILE
625# if __linux 944# if __linux
626 res = sendfile (ofd, ifd, &offset, count); 945 res = sendfile (ofd, ifd, &offset, count);
627 946
628# elif __freebsd 947# elif __FreeBSD__
629 /* 948 /*
630 * Of course, the freebsd sendfile is a dire hack with no thoughts 949 * Of course, the freebsd sendfile is a dire hack with no thoughts
631 * wasted on making it similar to other I/O functions. 950 * wasted on making it similar to other I/O functions.
632 */ 951 */
633 { 952 {
634 off_t sbytes; 953 off_t sbytes;
635 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 954 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636 955
637 if (res < 0 && sbytes) 956 #if 0 /* according to the manpage, this is correct, but broken behaviour */
638 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 957 /* freebsd' sendfile will return 0 on success */
958 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
959 /* not on e.g. EIO or EPIPE - sounds broken */
960 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
961 res = sbytes;
962 #endif
963
964 /* according to source inspection, this is correct, and useful behaviour */
965 if (sbytes)
966 res = sbytes;
967 }
968
969# elif defined (__APPLE__)
970
971 {
972 off_t sbytes = count;
973 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
974
975 /* according to the manpage, sbytes is always valid */
976 if (sbytes)
639 res = sbytes; 977 res = sbytes;
640 } 978 }
641 979
642# elif __hpux 980# elif __hpux
643 res = sendfile (ofd, ifd, offset, count, 0, 0); 981 res = sendfile (ofd, ifd, offset, count, 0, 0);
… …
657 if (res < 0 && sbytes) 995 if (res < 0 && sbytes)
658 res = sbytes; 996 res = sbytes;
659 } 997 }
660 998
661# endif 999# endif
1000
1001#elif defined (_WIN32)
1002
1003 /* does not work, just for documentation of what would need to be done */
1004 {
1005 HANDLE h = TO_SOCKET (ifd);
1006 SetFilePointer (h, offset, 0, FILE_BEGIN);
1007 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1008 }
1009
662#else 1010#else
663 res = -1; 1011 res = -1;
664 errno = ENOSYS; 1012 errno = ENOSYS;
665#endif 1013#endif
666 1014
667 if (res < 0 1015 if (res < 0
668 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1016 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1017 /* BSDs */
1018#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1019 || errno == ENOTSUP
1020#endif
1021 || errno == EOPNOTSUPP /* BSDs */
669#if __solaris 1022#if __solaris
670 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1023 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
671#endif 1024#endif
672 ) 1025 )
673 ) 1026 )
… …
704 } 1057 }
705 1058
706 return res; 1059 return res;
707} 1060}
708 1061
1062static signed char
1063eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1064{
1065 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1066 : a->inode < b->inode ? -1
1067 : a->inode > b->inode ? 1
1068 : 0;
1069}
1070
1071#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1072
1073#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1074#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1075
1076static void
1077eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1078{
1079 unsigned char bits [9 + sizeof (ino_t) * 8];
1080 unsigned char *bit = bits;
1081
1082 assert (CHAR_BIT == 8);
1083 assert (sizeof (eio_dirent) * 8 < 256);
1084 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1085 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1086
1087 if (size <= EIO_SORT_FAST)
1088 return;
1089
1090 /* first prepare an array of bits to test in our radix sort */
1091 /* try to take endianness into account, as well as differences in ino_t sizes */
1092 /* inode_bits must contain all inodes ORed together */
1093 /* which is used to skip bits that are 0 everywhere, which is very common */
1094 {
1095 ino_t endianness;
1096 int i, j;
1097
1098 /* we store the byte offset of byte n into byte n of "endianness" */
1099 for (i = 0; i < sizeof (ino_t); ++i)
1100 ((unsigned char *)&endianness)[i] = i;
1101
1102 *bit++ = 0;
1103
1104 for (i = 0; i < sizeof (ino_t); ++i)
1105 {
1106 /* shifting off the byte offsets out of "endianness" */
1107 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1108 endianness >>= 8;
1109
1110 for (j = 0; j < 8; ++j)
1111 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1112 *bit++ = offs + j;
1113 }
1114
1115 for (j = 0; j < 8; ++j)
1116 if (score_bits & (1 << j))
1117 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1118 }
1119
1120 /* now actually do the sorting (a variant of MSD radix sort) */
1121 {
1122 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1123 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1124 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1125 int stk_idx = 0;
1126
1127 base_stk [stk_idx] = dents;
1128 end_stk [stk_idx] = dents + size;
1129 bit_stk [stk_idx] = bit - 1;
1130
1131 do
1132 {
1133 base = base_stk [stk_idx];
1134 end = end_stk [stk_idx];
1135 bit = bit_stk [stk_idx];
1136
1137 for (;;)
1138 {
1139 unsigned char O = *bit >> 3;
1140 unsigned char M = 1 << (*bit & 7);
1141
1142 eio_dirent *a = base;
1143 eio_dirent *b = end;
1144
1145 if (b - a < EIO_SORT_CUTOFF)
1146 break;
1147
1148 /* now bit-partition the array on the bit */
1149 /* this ugly asymmetric loop seems to perform much better than typical */
1150 /* partition algos found in the literature */
1151 do
1152 if (!(((unsigned char *)a)[O] & M))
1153 ++a;
1154 else if (!(((unsigned char *)--b)[O] & M))
1155 {
1156 eio_dirent tmp = *a; *a = *b; *b = tmp;
1157 ++a;
1158 }
1159 while (b > a);
1160
1161 /* next bit, or stop, if no bits left in this path */
1162 if (!*--bit)
1163 break;
1164
1165 base_stk [stk_idx] = a;
1166 end_stk [stk_idx] = end;
1167 bit_stk [stk_idx] = bit;
1168 ++stk_idx;
1169
1170 end = a;
1171 }
1172 }
1173 while (stk_idx--);
1174 }
1175}
1176
1177static void
1178eio_dent_insertion_sort (eio_dirent *dents, int size)
1179{
1180 /* first move the smallest element to the front, to act as a sentinel */
1181 {
1182 int i;
1183 eio_dirent *min = dents;
1184
1185 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1186 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1187 if (EIO_DENT_CMP (dents [i], <, *min))
1188 min = &dents [i];
1189
1190 /* swap elements 0 and j (minimum) */
1191 {
1192 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1193 }
1194 }
1195
1196 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1197 {
1198 eio_dirent *i, *j;
1199
1200 for (i = dents + 1; i < dents + size; ++i)
1201 {
1202 eio_dirent value = *i;
1203
1204 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1205 j [1] = j [0];
1206
1207 j [1] = value;
1208 }
1209 }
1210}
1211
1212static void
1213eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1214{
1215 if (size <= 1)
1216 return; /* our insertion sort relies on size > 0 */
1217
1218 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1219 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1220 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1221
1222 /* use an insertion sort at the end, or for small arrays, */
1223 /* as insertion sort is more efficient for small partitions */
1224 eio_dent_insertion_sort (dents, size);
1225}
1226
709/* read a full directory */ 1227/* read a full directory */
1228static void
710static void scandir_ (eio_req *req, worker *self) 1229eio__scandir (eio_req *req, etp_worker *self)
711{ 1230{
712 DIR *dirp; 1231 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; 1232 EIO_STRUCT_DIRENT *entp;
719 char *name, *names; 1233 char *name, *names;
720 int memlen = 4096; 1234 int namesalloc = 4096;
721 int memofs = 0; 1235 int namesoffs = 0;
1236 int flags = req->int1;
1237 eio_dirent *dents = 0;
1238 int dentalloc = 128;
722 int res = 0; 1239 int dentoffs = 0;
1240 ino_t inode_bits = 0;
1241
1242 req->result = -1;
1243
1244 if (!(flags & EIO_READDIR_DENTS))
1245 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
723 1246
724 X_LOCK (wrklock); 1247 X_LOCK (wrklock);
1248 /* the corresponding closedir is in ETP_WORKER_CLEAR */
725 self->dirp = dirp = opendir (req->ptr1); 1249 self->dirp = dirp = opendir (req->ptr1);
726 self->dbuf = u = malloc (sizeof (*u)); 1250
727 req->flags |= EIO_FLAG_PTR2_FREE; 1251 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1252 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
728 req->ptr2 = names = malloc (memlen); 1253 req->ptr2 = names = malloc (namesalloc);
729 X_UNLOCK (wrklock); 1254 X_UNLOCK (wrklock);
730 1255
731 if (dirp && u && names) 1256 if (dirp && names && (!flags || dents))
732 for (;;) 1257 for (;;)
733 { 1258 {
734 errno = 0; 1259 errno = 0;
735 readdir_r (dirp, &u->d, &entp); 1260 entp = readdir (dirp);
736 1261
737 if (!entp) 1262 if (!entp)
1263 {
1264 if (errno)
1265 break;
1266
1267 /* sort etc. */
1268 req->int1 = flags;
1269 req->result = dentoffs;
1270
1271 if (flags & EIO_READDIR_STAT_ORDER)
1272 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1273 else if (flags & EIO_READDIR_DIRS_FIRST)
1274 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1275 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1276 else
1277 {
1278 /* in this case, all is known, and we just put dirs first and sort them */
1279 eio_dirent *oth = dents + dentoffs;
1280 eio_dirent *dir = dents;
1281
1282 /* now partition dirs to the front, and non-dirs to the back */
1283 /* by walking from both sides and swapping if necessary */
1284 while (oth > dir)
1285 {
1286 if (dir->type == EIO_DT_DIR)
1287 ++dir;
1288 else if ((--oth)->type == EIO_DT_DIR)
1289 {
1290 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1291
1292 ++dir;
1293 }
1294 }
1295
1296 /* now sort the dirs only (dirs all have the same score) */
1297 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1298 }
1299
738 break; 1300 break;
1301 }
739 1302
1303 /* now add the entry to our list(s) */
740 name = entp->d_name; 1304 name = entp->d_name;
741 1305
1306 /* skip . and .. entries */
742 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1307 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
743 { 1308 {
744 int len = strlen (name) + 1; 1309 int len = D_NAMLEN (entp) + 1;
745 1310
746 res++; 1311 while (expect_false (namesoffs + len > namesalloc))
747
748 while (memofs + len > memlen)
749 { 1312 {
750 memlen *= 2; 1313 namesalloc *= 2;
751 X_LOCK (wrklock); 1314 X_LOCK (wrklock);
752 req->ptr2 = names = realloc (names, memlen); 1315 req->ptr2 = names = realloc (names, namesalloc);
753 X_UNLOCK (wrklock); 1316 X_UNLOCK (wrklock);
754 1317
755 if (!names) 1318 if (!names)
756 break; 1319 break;
757 } 1320 }
758 1321
759 memcpy (names + memofs, name, len); 1322 memcpy (names + namesoffs, name, len);
1323
1324 if (dents)
1325 {
1326 struct eio_dirent *ent;
1327
1328 if (expect_false (dentoffs == dentalloc))
1329 {
1330 dentalloc *= 2;
1331 X_LOCK (wrklock);
1332 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1333 X_UNLOCK (wrklock);
1334
1335 if (!dents)
1336 break;
1337 }
1338
1339 ent = dents + dentoffs;
1340
1341 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1342 ent->namelen = len - 1;
1343 ent->inode = D_INO (entp);
1344
1345 inode_bits |= ent->inode;
1346
1347 switch (D_TYPE (entp))
1348 {
1349 default:
1350 ent->type = EIO_DT_UNKNOWN;
1351 flags |= EIO_READDIR_FOUND_UNKNOWN;
1352 break;
1353
1354 #ifdef DT_FIFO
1355 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1356 #endif
1357 #ifdef DT_CHR
1358 case DT_CHR: ent->type = EIO_DT_CHR; break;
1359 #endif
1360 #ifdef DT_MPC
1361 case DT_MPC: ent->type = EIO_DT_MPC; break;
1362 #endif
1363 #ifdef DT_DIR
1364 case DT_DIR: ent->type = EIO_DT_DIR; break;
1365 #endif
1366 #ifdef DT_NAM
1367 case DT_NAM: ent->type = EIO_DT_NAM; break;
1368 #endif
1369 #ifdef DT_BLK
1370 case DT_BLK: ent->type = EIO_DT_BLK; break;
1371 #endif
1372 #ifdef DT_MPB
1373 case DT_MPB: ent->type = EIO_DT_MPB; break;
1374 #endif
1375 #ifdef DT_REG
1376 case DT_REG: ent->type = EIO_DT_REG; break;
1377 #endif
1378 #ifdef DT_NWK
1379 case DT_NWK: ent->type = EIO_DT_NWK; break;
1380 #endif
1381 #ifdef DT_CMP
1382 case DT_CMP: ent->type = EIO_DT_CMP; break;
1383 #endif
1384 #ifdef DT_LNK
1385 case DT_LNK: ent->type = EIO_DT_LNK; break;
1386 #endif
1387 #ifdef DT_SOCK
1388 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1389 #endif
1390 #ifdef DT_DOOR
1391 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1392 #endif
1393 #ifdef DT_WHT
1394 case DT_WHT: ent->type = EIO_DT_WHT; break;
1395 #endif
1396 }
1397
1398 ent->score = 7;
1399
1400 if (flags & EIO_READDIR_DIRS_FIRST)
1401 {
1402 if (ent->type == EIO_DT_UNKNOWN)
1403 {
1404 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1405 ent->score = 1;
1406 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1407 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1408 }
1409 else if (ent->type == EIO_DT_DIR)
1410 ent->score = 0;
1411 }
1412 }
1413
760 memofs += len; 1414 namesoffs += len;
1415 ++dentoffs;
1416 }
1417
1418 if (EIO_CANCELLED (req))
1419 {
1420 errno = ECANCELED;
1421 break;
761 } 1422 }
762 } 1423 }
1424}
763 1425
764 if (errno) 1426#ifdef PAGESIZE
765 res = -1; 1427# define eio_pagesize() PAGESIZE
1428#else
1429static intptr_t
1430eio_pagesize (void)
1431{
1432 static intptr_t page;
1433
1434 if (!page)
1435 page = sysconf (_SC_PAGESIZE);
1436
1437 return page;
1438}
1439#endif
1440
1441static void
1442eio_page_align (void **addr, size_t *length)
1443{
1444 intptr_t mask = eio_pagesize () - 1;
1445
1446 /* round down addr */
1447 intptr_t adj = mask & (intptr_t)*addr;
1448
1449 *addr = (void *)((intptr_t)*addr - adj);
1450 *length += adj;
1451
1452 /* round up length */
1453 *length = (*length + mask) & ~mask;
1454}
1455
1456#if !_POSIX_MEMLOCK
1457# define eio__mlockall(a) ((errno = ENOSYS), -1)
1458#else
1459
1460static int
1461eio__mlockall (int flags)
1462{
1463 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1464 extern int mallopt (int, int);
1465 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1466 #endif
1467
1468 if (EIO_MCL_CURRENT != MCL_CURRENT
1469 || EIO_MCL_FUTURE != MCL_FUTURE)
1470 {
1471 flags = 0
1472 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1473 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1474 }
1475
1476 return mlockall (flags);
1477}
1478#endif
1479
1480#if !_POSIX_MEMLOCK_RANGE
1481# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1482#else
1483
1484static int
1485eio__mlock (void *addr, size_t length)
1486{
1487 eio_page_align (&addr, &length);
1488
1489 return mlock (addr, length);
1490}
1491
1492#endif
1493
1494#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1495# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1496#else
1497
1498int
1499eio__msync (void *mem, size_t len, int flags)
1500{
1501 eio_page_align (&mem, &len);
1502
1503 if (EIO_MS_ASYNC != MS_SYNC
1504 || EIO_MS_INVALIDATE != MS_INVALIDATE
1505 || EIO_MS_SYNC != MS_SYNC)
1506 {
1507 flags = 0
1508 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1509 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1510 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1511 }
1512
1513 return msync (mem, len, flags);
1514}
1515
1516#endif
1517
1518int
1519eio__mtouch (void *mem, size_t len, int flags)
1520{
1521 eio_page_align (&mem, &len);
1522
766 1523 {
767 req->result = res; 1524 intptr_t addr = (intptr_t)mem;
1525 intptr_t end = addr + len;
1526 intptr_t page = eio_pagesize ();
1527
1528 if (addr < end)
1529 if (flags & EIO_MT_MODIFY) /* modify */
1530 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1531 else
1532 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1533 }
1534
1535 return 0;
768} 1536}
769 1537
770/*****************************************************************************/ 1538/*****************************************************************************/
771 1539
772#define ALLOC(len) \ 1540#define ALLOC(len) \
… …
782 req->result = -1; \ 1550 req->result = -1; \
783 break; \ 1551 break; \
784 } \ 1552 } \
785 } 1553 }
786 1554
787X_THREAD_PROC (eio_proc) 1555X_THREAD_PROC (etp_proc)
788{ 1556{
789 eio_req *req; 1557 ETP_REQ *req;
790 struct timespec ts; 1558 struct timespec ts;
791 worker *self = (worker *)thr_arg; 1559 etp_worker *self = (etp_worker *)thr_arg;
792 1560
793 /* try to distribute timeouts somewhat randomly */ 1561 /* try to distribute timeouts somewhat randomly */
794 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1562 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
795 1563
796 for (;;) 1564 for (;;)
797 { 1565 {
798 ts.tv_sec = time (0) + IDLE_TIMEOUT;
799
800 X_LOCK (reqlock); 1566 X_LOCK (reqlock);
801 1567
802 for (;;) 1568 for (;;)
803 { 1569 {
804 self->req = req = reqq_shift (&req_queue); 1570 self->req = req = reqq_shift (&req_queue);
… …
806 if (req) 1572 if (req)
807 break; 1573 break;
808 1574
809 ++idle; 1575 ++idle;
810 1576
1577 ts.tv_sec = time (0) + idle_timeout;
811 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1578 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
812 { 1579 {
813 if (idle > max_idle) 1580 if (idle > max_idle)
814 { 1581 {
815 --idle; 1582 --idle;
… …
820 goto quit; 1587 goto quit;
821 } 1588 }
822 1589
823 /* we are allowed to idle, so do so without any timeout */ 1590 /* we are allowed to idle, so do so without any timeout */
824 X_COND_WAIT (reqwait, reqlock); 1591 X_COND_WAIT (reqwait, reqlock);
825 ts.tv_sec = time (0) + IDLE_TIMEOUT;
826 } 1592 }
827 1593
828 --idle; 1594 --idle;
829 } 1595 }
830 1596
831 --nready; 1597 --nready;
832 1598
833 X_UNLOCK (reqlock); 1599 X_UNLOCK (reqlock);
834 1600
835 errno = 0; /* strictly unnecessary */ 1601 if (req->type < 0)
1602 goto quit;
836 1603
837 if (!EIO_CANCELLED (req)) 1604 if (!EIO_CANCELLED (req))
838 switch (req->type) 1605 ETP_EXECUTE (self, req);
839 {
840 case EIO_READ: ALLOC (req->size);
841 req->result = req->offs >= 0
842 ? pread (req->int1, req->ptr2, req->size, req->offs)
843 : read (req->int1, req->ptr2, req->size); break;
844 case EIO_WRITE: req->result = req->offs >= 0
845 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
846 : write (req->int1, req->ptr2, req->size); break;
847
848 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
849 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
850
851 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
852 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
853 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
854 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
855 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
856 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
857
858 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
859 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
860 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
861 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
862 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
863 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
864
865 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
866 case EIO_CLOSE: req->result = close (req->int1); break;
867 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
868 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
869 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
870 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
871 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
872 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
873 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
874 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
875
876 case EIO_READLINK: ALLOC (NAME_MAX);
877 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
878
879 case EIO_SYNC: req->result = 0; sync (); break;
880 case EIO_FSYNC: req->result = fsync (req->int1); break;
881 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
882
883 case EIO_READDIR: scandir_ (req, self); break;
884
885 case EIO_BUSY:
886#ifdef _WIN32
887 Sleep (req->nv1 * 1000.);
888#else
889 {
890 struct timeval tv;
891
892 tv.tv_sec = req->nv1;
893 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
894
895 req->result = select (0, 0, 0, 0, &tv);
896 }
897#endif
898 break;
899
900 case EIO_UTIME:
901 case EIO_FUTIME:
902 {
903 struct timeval tv[2];
904 struct timeval *times;
905
906 if (req->nv1 != -1. || req->nv2 != -1.)
907 {
908 tv[0].tv_sec = req->nv1;
909 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
910 tv[1].tv_sec = req->nv2;
911 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
912
913 times = tv;
914 }
915 else
916 times = 0;
917
918
919 req->result = req->type == EIO_FUTIME
920 ? futimes (req->int1, times)
921 : utimes (req->ptr1, times);
922 }
923
924 case EIO_GROUP:
925 case EIO_NOP:
926 break;
927
928 case EIO_QUIT:
929 goto quit;
930
931 default:
932 req->result = -1;
933 break;
934 }
935
936 req->errorno = errno;
937 1606
938 X_LOCK (reslock); 1607 X_LOCK (reslock);
939 1608
940 ++npending; 1609 ++npending;
941 1610
942 if (!reqq_push (&res_queue, req) && want_poll_cb) 1611 if (!reqq_push (&res_queue, req) && want_poll_cb)
943 want_poll_cb (); 1612 want_poll_cb ();
944 1613
945 self->req = 0; 1614 self->req = 0;
946 worker_clear (self); 1615 etp_worker_clear (self);
947 1616
948 X_UNLOCK (reslock); 1617 X_UNLOCK (reslock);
949 } 1618 }
950 1619
951quit: 1620quit:
952 X_LOCK (wrklock); 1621 X_LOCK (wrklock);
953 worker_free (self); 1622 etp_worker_free (self);
954 X_UNLOCK (wrklock); 1623 X_UNLOCK (wrklock);
955 1624
956 return 0; 1625 return 0;
957} 1626}
958 1627
959/*****************************************************************************/ 1628/*****************************************************************************/
960 1629
961static void eio_atfork_prepare (void)
962{
963 X_LOCK (wrklock);
964 X_LOCK (reqlock);
965 X_LOCK (reslock);
966#if !HAVE_PREADWRITE
967 X_LOCK (preadwritelock);
968#endif
969#if !HAVE_READDIR_R
970 X_LOCK (readdirlock);
971#endif
972}
973
974static void eio_atfork_parent (void)
975{
976#if !HAVE_READDIR_R
977 X_UNLOCK (readdirlock);
978#endif
979#if !HAVE_PREADWRITE
980 X_UNLOCK (preadwritelock);
981#endif
982 X_UNLOCK (reslock);
983 X_UNLOCK (reqlock);
984 X_UNLOCK (wrklock);
985}
986
987static void eio_atfork_child (void)
988{
989 eio_req *prv;
990
991 while (prv = reqq_shift (&req_queue))
992 eio_destroy (prv);
993
994 while (prv = reqq_shift (&res_queue))
995 eio_destroy (prv);
996
997 while (wrk_first.next != &wrk_first)
998 {
999 worker *wrk = wrk_first.next;
1000
1001 if (wrk->req)
1002 eio_destroy (wrk->req);
1003
1004 worker_clear (wrk);
1005 worker_free (wrk);
1006 }
1007
1008 started = 0;
1009 idle = 0;
1010 nreqs = 0;
1011 nready = 0;
1012 npending = 0;
1013
1014 eio_atfork_parent ();
1015}
1016
1017int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1630int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1018{ 1631{
1019 want_poll_cb = want_poll; 1632 return etp_init (want_poll, done_poll);
1020 done_poll_cb = done_poll;
1021
1022#ifdef _WIN32
1023 X_MUTEX_CHECK (wrklock);
1024 X_MUTEX_CHECK (reslock);
1025 X_MUTEX_CHECK (reqlock);
1026 X_MUTEX_CHECK (reqwait);
1027 X_MUTEX_CHECK (preadwritelock);
1028 X_MUTEX_CHECK (readdirlock);
1029
1030 X_COND_CHECK (reqwait);
1031#endif
1032
1033 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1034} 1633}
1035 1634
1036static void eio_api_destroy (eio_req *req) 1635static void eio_api_destroy (eio_req *req)
1037{ 1636{
1038 free (req); 1637 free (req);
… …
1043 \ 1642 \
1044 req = (eio_req *)calloc (1, sizeof *req); \ 1643 req = (eio_req *)calloc (1, sizeof *req); \
1045 if (!req) \ 1644 if (!req) \
1046 return 0; \ 1645 return 0; \
1047 \ 1646 \
1048 req->type = EIO_ ## rtype; \ 1647 req->type = rtype; \
1049 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 1648 req->pri = pri; \
1050 req->finish = cb; \ 1649 req->finish = cb; \
1650 req->data = data; \
1051 req->destroy = eio_api_destroy; 1651 req->destroy = eio_api_destroy;
1052 1652
1053#define SEND eio_submit (req); return req 1653#define SEND eio_submit (req); return req
1054 1654
1055#define PATH (void)0 1655#define PATH \
1656 req->flags |= EIO_FLAG_PTR1_FREE; \
1657 req->ptr1 = strdup (path); \
1658 if (!req->ptr1) \
1659 { \
1660 eio_api_destroy (req); \
1661 return 0; \
1662 }
1056 1663
1664static void eio_execute (etp_worker *self, eio_req *req)
1665{
1666 switch (req->type)
1667 {
1668 case EIO_READ: ALLOC (req->size);
1669 req->result = req->offs >= 0
1670 ? pread (req->int1, req->ptr2, req->size, req->offs)
1671 : read (req->int1, req->ptr2, req->size); break;
1672 case EIO_WRITE: req->result = req->offs >= 0
1673 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1674 : write (req->int1, req->ptr2, req->size); break;
1675
1676 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1677 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1678
1679 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1680 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1681 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1682 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1683 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1684 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1685
1686 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1687 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1688 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1689 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1690
1691 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1692 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1693 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1694 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1695 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1696 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1697
1698 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1699 case EIO_CLOSE: req->result = close (req->int1); break;
1700 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1701 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1702 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1703 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1704 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1705 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1706 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1707 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1708
1709 case EIO_READLINK: ALLOC (PATH_MAX);
1710 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1711
1712 case EIO_SYNC: req->result = 0; sync (); break;
1713 case EIO_FSYNC: req->result = fsync (req->int1); break;
1714 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1715 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1716 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1717 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1718 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1719 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1720
1721 case EIO_READDIR: eio__scandir (req, self); break;
1722
1723 case EIO_BUSY:
1724#ifdef _WIN32
1725 Sleep (req->nv1 * 1e3);
1726#else
1727 {
1728 struct timeval tv;
1729
1730 tv.tv_sec = req->nv1;
1731 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1732
1733 req->result = select (0, 0, 0, 0, &tv);
1734 }
1735#endif
1736 break;
1737
1738 case EIO_UTIME:
1739 case EIO_FUTIME:
1740 {
1741 struct timeval tv[2];
1742 struct timeval *times;
1743
1744 if (req->nv1 != -1. || req->nv2 != -1.)
1745 {
1746 tv[0].tv_sec = req->nv1;
1747 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1748 tv[1].tv_sec = req->nv2;
1749 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1750
1751 times = tv;
1752 }
1753 else
1754 times = 0;
1755
1756 req->result = req->type == EIO_FUTIME
1757 ? futimes (req->int1, times)
1758 : utimes (req->ptr1, times);
1759 }
1760 break;
1761
1762 case EIO_GROUP:
1763 abort (); /* handled in eio_request */
1764
1765 case EIO_NOP:
1766 req->result = 0;
1767 break;
1768
1769 case EIO_CUSTOM:
1770 ((void (*)(eio_req *))req->feed) (req);
1771 break;
1772
1773 default:
1774 errno = ENOSYS;
1775 req->result = -1;
1776 break;
1777 }
1778
1779 req->errorno = errno;
1780}
1781
1782#ifndef EIO_NO_WRAPPERS
1783
1784eio_req *eio_nop (int pri, eio_cb cb, void *data)
1785{
1786 REQ (EIO_NOP); SEND;
1787}
1788
1789eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1790{
1791 REQ (EIO_BUSY); req->nv1 = delay; SEND;
1792}
1793
1057eio_req *eio_fsync (int fd, eio_cb cb) 1794eio_req *eio_sync (int pri, eio_cb cb, void *data)
1058{ 1795{
1796 REQ (EIO_SYNC); SEND;
1797}
1798
1799eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1800{
1059 REQ (FSYNC); req->int1 = fd; SEND; 1801 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1060} 1802}
1061 1803
1804eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1805{
1806 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1807}
1808
1809eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1810{
1811 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1812}
1813
1814eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1815{
1816 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1817}
1818
1819eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1820{
1821 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1822}
1823
1824eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1825{
1826 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1827}
1828
1062eio_req *eio_fdatasync (int fd, eio_cb cb) 1829eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1063{ 1830{
1064 REQ (FDATASYNC); req->int1 = fd; SEND; 1831 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1065} 1832}
1066 1833
1834eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1835{
1836 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1837}
1838
1067eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb) 1839eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1068{ 1840{
1069 REQ (READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 1841 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1070} 1842}
1071 1843
1072eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb) 1844eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1073{ 1845{
1074 REQ (READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 1846 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1075} 1847}
1076 1848
1077eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb) 1849eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1078{ 1850{
1079 REQ (WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 1851 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1080} 1852}
1081 1853
1082eio_req *eio_fstat (int fd, eio_cb cb) 1854eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1083{ 1855{
1084 REQ (FSTAT); req->int1 = fd; SEND; 1856 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1085} 1857}
1086 1858
1859eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1860{
1861 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1862}
1863
1087eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb) 1864eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1088{ 1865{
1089 REQ (FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1866 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1090} 1867}
1091 1868
1092eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb) 1869eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1093{ 1870{
1094 REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 1871 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1095} 1872}
1096 1873
1097eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb) 1874eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1098{ 1875{
1099 REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 1876 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1100} 1877}
1101 1878
1102eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb) 1879eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1103{ 1880{
1104 REQ (FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 1881 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1105} 1882}
1106 1883
1107eio_req *eio_dup2 (int fd, int fd2, eio_cb cb) 1884eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1108{ 1885{
1109 REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND; 1886 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1110} 1887}
1111 1888
1112eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb) 1889eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1113{ 1890{
1114 REQ (SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 1891 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1115} 1892}
1116 1893
1117eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb) 1894eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1118{ 1895{
1119 REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 1896 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1120} 1897}
1121 1898
1122eio_req *eio_readlink (const char *path, eio_cb cb)
1123{
1124 REQ (READLINK); PATH; SEND;
1125}
1126
1127eio_req *eio_stat (const char *path, eio_cb cb)
1128{
1129 REQ (STAT); PATH; SEND;
1130}
1131
1132eio_req *eio_lstat (const char *path, eio_cb cb)
1133{
1134 REQ (LSTAT); PATH; SEND;
1135}
1136
1137eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb) 1899eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1138{ 1900{
1139 REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 1901 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1140} 1902}
1141 1903
1142eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb) 1904eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1143{ 1905{
1144 REQ (TRUNCATE); PATH; req->offs = offset; SEND; 1906 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1145} 1907}
1146 1908
1147eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 1909eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1148{ 1910{
1149 REQ (CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 1911 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1150} 1912}
1151 1913
1152eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb) 1914eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1153{ 1915{
1154 REQ (CHMOD); PATH; req->int2 = (long)mode; SEND; 1916 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1155} 1917}
1156 1918
1157eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb) 1919eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1158{ 1920{
1159 REQ (MKDIR); PATH; req->int2 = (long)mode; SEND; 1921 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1160} 1922}
1161 1923
1162eio_req *eio_unlink (const char *path, eio_cb cb) 1924static eio_req *
1925eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1163{ 1926{
1164 REQ (UNLINK); PATH; SEND;
1165}
1166
1167eio_req *eio_rmdir (const char *path, eio_cb cb)
1168{
1169 REQ (RMDIR); PATH; SEND; 1927 REQ (type); PATH; SEND;
1170} 1928}
1171 1929
1930eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1931{
1932 return eio__1path (EIO_READLINK, path, pri, cb, data);
1933}
1934
1935eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1936{
1937 return eio__1path (EIO_STAT, path, pri, cb, data);
1938}
1939
1940eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1941{
1942 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1943}
1944
1945eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1946{
1947 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1948}
1949
1950eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1951{
1952 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1953}
1954
1172eio_req *eio_readdir (const char *path, eio_cb cb) 1955eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1173{ 1956{
1174 REQ (READDIR); PATH; SEND; 1957 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1175} 1958}
1176 1959
1960eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1961{
1962 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1963}
1964
1177eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb) 1965eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1178{ 1966{
1179 REQ (MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1967 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1180} 1968}
1181 1969
1182eio_req *eio_busy (double delay, eio_cb cb) 1970static eio_req *
1971eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1183{ 1972{
1184 REQ (BUSY); req->nv1 = delay; SEND; 1973 REQ (type); PATH;
1185}
1186 1974
1975 req->flags |= EIO_FLAG_PTR2_FREE;
1976 req->ptr2 = strdup (new_path);
1977 if (!req->ptr2)
1978 {
1979 eio_api_destroy (req);
1980 return 0;
1981 }
1982
1983 SEND;
1984}
1985
1986eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1987{
1988 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
1989}
1990
1991eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1992{
1993 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
1994}
1995
1996eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1997{
1998 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1999}
2000
2001eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
2002{
2003 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
2004}
2005
2006#endif
2007
1187eio_req *eio_nop (eio_cb cb) 2008eio_req *eio_grp (eio_cb cb, void *data)
1188{ 2009{
2010 const int pri = EIO_PRI_MAX;
2011
1189 REQ (NOP); SEND; 2012 REQ (EIO_GROUP); SEND;
1190} 2013}
1191 2014
1192#undef REQ 2015#undef REQ
1193#undef PATH 2016#undef PATH
1194#undef SEND 2017#undef SEND
1195 2018
1196#if 0 2019/*****************************************************************************/
1197 2020/* grp functions */
1198void
1199aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1200 PROTOTYPE: $$$;$
1201 PPCODE:
1202{
1203 dREQ;
1204
1205 req->type = EIO_OPEN;
1206 req->sv1 = newSVsv (pathname);
1207 req->ptr1 = SvPVbyte_nolen (req->sv1);
1208 req->int1 = flags;
1209 req->int2 = mode;
1210
1211 EIO_SEND;
1212}
1213
1214void
1215aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1216 PROTOTYPE: $;$
1217 ALIAS:
1218 aio_fsync = EIO_FSYNC
1219 aio_fdatasync = EIO_FDATASYNC
1220 PPCODE:
1221{
1222 dREQ;
1223
1224 req->type = ix;
1225 req->sv1 = newSVsv (fh);
1226 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1227
1228 EIO_SEND (req);
1229}
1230
1231void
1232aio_close (SV *fh, SV *callback=&PL_sv_undef)
1233 PROTOTYPE: $;$
1234 PPCODE:
1235{
1236 dREQ;
1237
1238 req->type = EIO_CLOSE;
1239 req->sv1 = newSVsv (fh);
1240 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1241
1242 EIO_SEND (req);
1243}
1244
1245void
1246aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1247 ALIAS:
1248 aio_read = EIO_READ
1249 aio_write = EIO_WRITE
1250 PROTOTYPE: $$$$$;$
1251 PPCODE:
1252{
1253 STRLEN svlen;
1254 char *svptr = SvPVbyte (data, svlen);
1255 UV len = SvUV (length);
1256
1257 SvUPGRADE (data, SVt_PV);
1258 SvPOK_on (data);
1259
1260 if (dataoffset < 0)
1261 dataoffset += svlen;
1262
1263 if (dataoffset < 0 || dataoffset > svlen)
1264 croak ("dataoffset outside of data scalar");
1265
1266 if (ix == EIO_WRITE)
1267 {
1268 /* write: check length and adjust. */
1269 if (!SvOK (length) || len + dataoffset > svlen)
1270 len = svlen - dataoffset;
1271 }
1272 else
1273 {
1274 /* read: grow scalar as necessary */
1275 svptr = SvGROW (data, len + dataoffset + 1);
1276 }
1277
1278 if (len < 0)
1279 croak ("length must not be negative");
1280
1281 {
1282 dREQ;
1283
1284 req->type = ix;
1285 req->sv1 = newSVsv (fh);
1286 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1287 : IoOFP (sv_2io (fh)));
1288 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1289 req->size = len;
1290 req->sv2 = SvREFCNT_inc (data);
1291 req->ptr2 = (char *)svptr + dataoffset;
1292 req->stroffset = dataoffset;
1293
1294 if (!SvREADONLY (data))
1295 {
1296 SvREADONLY_on (data);
1297 req->flags |= FLAG_SV2_RO_OFF;
1298 }
1299
1300 EIO_SEND;
1301 }
1302}
1303
1304void
1305aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1306 PROTOTYPE: $$;$
1307 PPCODE:
1308{
1309 SV *data;
1310 dREQ;
1311
1312 data = newSV (NAME_MAX);
1313 SvPOK_on (data);
1314
1315 req->type = EIO_READLINK;
1316 req->sv1 = newSVsv (path);
1317 req->ptr1 = SvPVbyte_nolen (req->sv1);
1318 req->sv2 = data;
1319 req->ptr2 = SvPVbyte_nolen (data);
1320
1321 EIO_SEND;
1322}
1323
1324void
1325aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1326 PROTOTYPE: $$$$;$
1327 PPCODE:
1328{
1329 dREQ;
1330
1331 req->type = EIO_SENDFILE;
1332 req->sv1 = newSVsv (out_fh);
1333 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1334 req->sv2 = newSVsv (in_fh);
1335 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1336 req->offs = SvVAL64 (in_offset);
1337 req->size = length;
1338
1339 EIO_SEND;
1340}
1341
1342void
1343aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1344 PROTOTYPE: $$$;$
1345 PPCODE:
1346{
1347 dREQ;
1348
1349 req->type = EIO_READAHEAD;
1350 req->sv1 = newSVsv (fh);
1351 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1352 req->offs = SvVAL64 (offset);
1353 req->size = length;
1354
1355 EIO_SEND;
1356}
1357
1358void
1359aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1360 ALIAS:
1361 aio_stat = EIO_STAT
1362 aio_lstat = EIO_LSTAT
1363 PPCODE:
1364{
1365 dREQ;
1366
1367 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1368 if (!req->ptr2)
1369 {
1370 req_destroy (req);
1371 croak ("out of memory during aio_stat statdata allocation");
1372 }
1373
1374 req->flags |= FLAG_PTR2_FREE;
1375 req->sv1 = newSVsv (fh_or_path);
1376
1377 if (SvPOK (fh_or_path))
1378 {
1379 req->type = ix;
1380 req->ptr1 = SvPVbyte_nolen (req->sv1);
1381 }
1382 else
1383 {
1384 req->type = EIO_FSTAT;
1385 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1386 }
1387
1388 EIO_SEND;
1389}
1390
1391void
1392aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1393 PPCODE:
1394{
1395 dREQ;
1396
1397 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1398 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1399 req->sv1 = newSVsv (fh_or_path);
1400
1401 if (SvPOK (fh_or_path))
1402 {
1403 req->type = EIO_UTIME;
1404 req->ptr1 = SvPVbyte_nolen (req->sv1);
1405 }
1406 else
1407 {
1408 req->type = EIO_FUTIME;
1409 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1410 }
1411
1412 EIO_SEND;
1413}
1414
1415void
1416aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1417 PPCODE:
1418{
1419 dREQ;
1420
1421 req->sv1 = newSVsv (fh_or_path);
1422 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1423
1424 if (SvPOK (fh_or_path))
1425 {
1426 req->type = EIO_TRUNCATE;
1427 req->ptr1 = SvPVbyte_nolen (req->sv1);
1428 }
1429 else
1430 {
1431 req->type = EIO_FTRUNCATE;
1432 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1433 }
1434
1435 EIO_SEND;
1436}
1437
1438void
1439aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1440 ALIAS:
1441 aio_chmod = EIO_CHMOD
1442 aio_fchmod = EIO_FCHMOD
1443 aio_mkdir = EIO_MKDIR
1444 PPCODE:
1445{
1446 dREQ;
1447
1448 req->type = type;
1449 req->int2 = mode;
1450 req->sv1 = newSVsv (fh_or_path);
1451
1452 if (ix == EIO_FCHMOD)
1453 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1454 else
1455 req->ptr1 = SvPVbyte_nolen (req->sv1);
1456
1457 EIO_SEND;
1458}
1459
1460void
1461aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1462 PPCODE:
1463{
1464 dREQ;
1465
1466 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1467 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1468 req->sv1 = newSVsv (fh_or_path);
1469
1470 if (SvPOK (fh_or_path))
1471 {
1472 req->type = EIO_CHOWN;
1473 req->ptr1 = SvPVbyte_nolen (req->sv1);
1474 }
1475 else
1476 {
1477 req->type = EIO_FCHOWN;
1478 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1479 }
1480
1481 EIO_SEND;
1482}
1483
1484void
1485aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1486 ALIAS:
1487 aio_unlink = EIO_UNLINK
1488 aio_rmdir = EIO_RMDIR
1489 aio_readdir = EIO_READDIR
1490 PPCODE:
1491{
1492 dREQ;
1493
1494 req->type = ix;
1495 req->sv1 = newSVsv (pathname);
1496 req->ptr1 = SvPVbyte_nolen (req->sv1);
1497
1498 EIO_SEND;
1499}
1500
1501void
1502aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1503 ALIAS:
1504 aio_link = EIO_LINK
1505 aio_symlink = EIO_SYMLINK
1506 aio_rename = EIO_RENAME
1507 PPCODE:
1508{
1509 dREQ;
1510
1511 req->type = ix;
1512 req->sv1 = newSVsv (oldpath);
1513 req->ptr1 = SvPVbyte_nolen (req->sv1);
1514 req->sv2 = newSVsv (newpath);
1515 req->ptr2 = SvPVbyte_nolen (req->sv2);
1516
1517 EIO_SEND;
1518}
1519
1520void
1521aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1522 PPCODE:
1523{
1524 dREQ;
1525
1526 req->type = EIO_MKNOD;
1527 req->sv1 = newSVsv (pathname);
1528 req->ptr1 = SvPVbyte_nolen (req->sv1);
1529 req->int2 = (mode_t)mode;
1530 req->offs = dev;
1531
1532 EIO_SEND;
1533}
1534
1535void
1536aio_busy (double delay, SV *callback=&PL_sv_undef)
1537 PPCODE:
1538{
1539 dREQ;
1540
1541 req->type = EIO_BUSY;
1542 req->nv1 = delay < 0. ? 0. : delay;
1543
1544 EIO_SEND;
1545}
1546
1547void
1548aio_group (SV *callback=&PL_sv_undef)
1549 PROTOTYPE: ;$
1550 PPCODE:
1551{
1552 dREQ;
1553
1554 req->type = EIO_GROUP;
1555
1556 req_send (req);
1557 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1558}
1559
1560void
1561aio_nop (SV *callback=&PL_sv_undef)
1562 ALIAS:
1563 aio_nop = EIO_NOP
1564 aio_sync = EIO_SYNC
1565 PPCODE:
1566{
1567 dREQ;
1568
1569#endif
1570 2021
1571void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2022void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1572{ 2023{
1573 grp->int2 = limit; 2024 grp->int2 = limit;
1574 grp->feed = feed; 2025 grp->feed = feed;
… …
1585 2036
1586void eio_grp_add (eio_req *grp, eio_req *req) 2037void eio_grp_add (eio_req *grp, eio_req *req)
1587{ 2038{
1588 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2039 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1589 2040
2041 grp->flags |= EIO_FLAG_GROUPADD;
2042
1590 ++grp->size; 2043 ++grp->size;
1591 req->grp = grp; 2044 req->grp = grp;
1592 2045
1593 req->grp_prev = 0; 2046 req->grp_prev = 0;
1594 req->grp_next = grp->grp_first; 2047 req->grp_next = grp->grp_first;
… …
1597 grp->grp_first->grp_prev = req; 2050 grp->grp_first->grp_prev = req;
1598 2051
1599 grp->grp_first = req; 2052 grp->grp_first = req;
1600} 2053}
1601 2054
2055/*****************************************************************************/
2056/* misc garbage */
1602 2057
2058ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2059{
2060 etp_worker wrk;
2061 ssize_t ret;
2062
2063 wrk.dbuf = 0;
2064
2065 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2066
2067 if (wrk.dbuf)
2068 free (wrk.dbuf);
2069
2070 return ret;
2071}
2072

Diff Legend

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