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

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.9 by root, Mon May 12 00:31:43 2008 UTC vs.
Revision 1.52 by root, Fri Jan 8 09:29:56 2010 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010 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>
79# include <sys/mman.h>
36# include <unistd.h> 80# include <unistd.h>
37# include <utime.h> 81# include <utime.h>
38# include <signal.h> 82# include <signal.h>
39# include <dirent.h> 83# include <dirent.h>
40 84
85/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
86# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
87# define _DIRENT_HAVE_D_TYPE /* sigh */
88# define D_INO(de) (de)->d_fileno
89# define D_NAMLEN(de) (de)->d_namlen
90# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
91# define D_INO(de) (de)->d_ino
92# endif
93
94#ifdef _D_EXACT_NAMLEN
95# undef D_NAMLEN
96# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
97#endif
98
99# ifdef _DIRENT_HAVE_D_TYPE
100# define D_TYPE(de) (de)->d_type
101# endif
102
41# ifndef EIO_STRUCT_DIRENT 103# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 104# define EIO_STRUCT_DIRENT struct dirent
43# endif 105# endif
44 106
45#endif 107#endif
46 108
47#if HAVE_SENDFILE 109#if HAVE_SENDFILE
48# if __linux 110# if __linux
49# include <sys/sendfile.h> 111# include <sys/sendfile.h>
50# elif __freebsd 112# elif __FreeBSD__ || defined __APPLE__
51# include <sys/socket.h> 113# include <sys/socket.h>
52# include <sys/uio.h> 114# include <sys/uio.h>
53# elif __hpux 115# elif __hpux
54# include <sys/socket.h> 116# include <sys/socket.h>
55# elif __solaris /* not yet */ 117# elif __solaris
56# include <sys/sendfile.h> 118# include <sys/sendfile.h>
57# else 119# else
58# error sendfile support requested but not available 120# error sendfile support requested but not available
59# endif 121# endif
60#endif 122#endif
61 123
124#ifndef D_TYPE
125# define D_TYPE(de) 0
126#endif
127#ifndef D_INO
128# define D_INO(de) 0
129#endif
130#ifndef D_NAMLEN
131# define D_NAMLEN(de) strlen ((de)->d_name)
132#endif
133
62/* number of seconds after which an idle threads exit */ 134/* number of seconds after which an idle threads exit */
63#define IDLE_TIMEOUT 10 135#define IDLE_TIMEOUT 10
64 136
65/* used for struct dirent, AIX doesn't provide it */ 137/* used for struct dirent, AIX doesn't provide it */
66#ifndef NAME_MAX 138#ifndef NAME_MAX
70/* buffer size for various temporary buffers */ 142/* buffer size for various temporary buffers */
71#define EIO_BUFSIZE 65536 143#define EIO_BUFSIZE 65536
72 144
73#define dBUF \ 145#define dBUF \
74 char *eio_buf; \ 146 char *eio_buf; \
75 X_LOCK (wrklock); \ 147 ETP_WORKER_LOCK (self); \
76 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 148 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77 X_UNLOCK (wrklock); \ 149 ETP_WORKER_UNLOCK (self); \
78 errno = ENOMEM; \ 150 errno = ENOMEM; \
79 if (!eio_buf) \ 151 if (!eio_buf) \
80 return -1; 152 return -1;
81 153
82#define EIO_TICKS ((1000000 + 1023) >> 10) 154#define EIO_TICKS ((1000000 + 1023) >> 10)
83 155
156/*****************************************************************************/
157
158#if __GNUC__ >= 3
159# define expect(expr,value) __builtin_expect ((expr),(value))
160#else
161# define expect(expr,value) (expr)
162#endif
163
164#define expect_false(expr) expect ((expr) != 0, 0)
165#define expect_true(expr) expect ((expr) != 0, 1)
166
167/*****************************************************************************/
168
169#define ETP_PRI_MIN EIO_PRI_MIN
170#define ETP_PRI_MAX EIO_PRI_MAX
171
172struct etp_worker;
173
174#define ETP_REQ eio_req
175#define ETP_DESTROY(req) eio_destroy (req)
176static int eio_finish (eio_req *req);
177#define ETP_FINISH(req) eio_finish (req)
178static void eio_execute (struct etp_worker *self, eio_req *req);
179#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
180
181#define ETP_WORKER_CLEAR(req) \
182 if (wrk->dbuf) \
183 { \
184 free (wrk->dbuf); \
185 wrk->dbuf = 0; \
186 } \
187 \
188 if (wrk->dirp) \
189 { \
190 closedir (wrk->dirp); \
191 wrk->dirp = 0; \
192 }
193
194#define ETP_WORKER_COMMON \
195 void *dbuf; \
196 DIR *dirp;
197
198/*****************************************************************************/
199
200#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
201
202/* calculate time difference in ~1/EIO_TICKS of a second */
203static int tvdiff (struct timeval *tv1, struct timeval *tv2)
204{
205 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
206 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
207}
208
209static unsigned int started, idle, wanted = 4;
210
84static void (*want_poll_cb) (void); 211static void (*want_poll_cb) (void);
85static void (*done_poll_cb) (void); 212static void (*done_poll_cb) (void);
86 213
87static unsigned int max_poll_time = 0; 214static unsigned int max_poll_time; /* reslock */
88static unsigned int max_poll_reqs = 0; 215static unsigned int max_poll_reqs; /* reslock */
89 216
90/* calculcate time difference in ~1/EIO_TICKS of a second */ 217static volatile unsigned int nreqs; /* reqlock */
91static int tvdiff (struct timeval *tv1, struct timeval *tv2) 218static volatile unsigned int nready; /* reqlock */
92{ 219static volatile unsigned int npending; /* reqlock */
93 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 220static volatile unsigned int max_idle = 4;
94 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
95}
96 221
97static unsigned int started, idle, wanted = 4;
98
99/* worker threads management */
100static mutex_t wrklock = X_MUTEX_INIT; 222static mutex_t wrklock = X_MUTEX_INIT;
101
102typedef struct worker
103{
104 /* locked by wrklock */
105 struct worker *prev, *next;
106
107 thread_t tid;
108
109 /* locked by reslock, reqlock or wrklock */
110 eio_req *req; /* currently processed request */
111 void *dbuf;
112 DIR *dirp;
113} worker;
114
115static worker wrk_first = { &wrk_first, &wrk_first, 0 };
116
117static void worker_clear (worker *wrk)
118{
119 if (wrk->dirp)
120 {
121 closedir (wrk->dirp);
122 wrk->dirp = 0;
123 }
124
125 if (wrk->dbuf)
126 {
127 free (wrk->dbuf);
128 wrk->dbuf = 0;
129 }
130}
131
132static void worker_free (worker *wrk)
133{
134 wrk->next->prev = wrk->prev;
135 wrk->prev->next = wrk->next;
136
137 free (wrk);
138}
139
140static volatile unsigned int nreqs, nready, npending;
141static volatile unsigned int max_idle = 4;
142
143static mutex_t reslock = X_MUTEX_INIT; 223static mutex_t reslock = X_MUTEX_INIT;
144static mutex_t reqlock = X_MUTEX_INIT; 224static mutex_t reqlock = X_MUTEX_INIT;
145static cond_t reqwait = X_COND_INIT; 225static cond_t reqwait = X_COND_INIT;
146 226
227#if !HAVE_PREADWRITE
228/*
229 * make our pread/pwrite emulation safe against themselves, but not against
230 * normal read/write by using a mutex. slows down execution a lot,
231 * but that's your problem, not mine.
232 */
233static mutex_t preadwritelock = X_MUTEX_INIT;
234#endif
235
236typedef struct etp_worker
237{
238 /* locked by wrklock */
239 struct etp_worker *prev, *next;
240
241 thread_t tid;
242
243 /* locked by reslock, reqlock or wrklock */
244 ETP_REQ *req; /* currently processed request */
245
246 ETP_WORKER_COMMON
247} etp_worker;
248
249static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
250
251#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
252#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
253
254/* worker threads management */
255
256static void etp_worker_clear (etp_worker *wrk)
257{
258 ETP_WORKER_CLEAR (wrk);
259}
260
261static void etp_worker_free (etp_worker *wrk)
262{
263 wrk->next->prev = wrk->prev;
264 wrk->prev->next = wrk->next;
265
266 free (wrk);
267}
268
147unsigned int eio_nreqs (void) 269static unsigned int etp_nreqs (void)
148{ 270{
271 int retval;
272 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
273 retval = nreqs;
274 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
149 return nreqs; 275 return retval;
150} 276}
151 277
152unsigned int eio_nready (void) 278static unsigned int etp_nready (void)
153{ 279{
154 unsigned int retval; 280 unsigned int retval;
155 281
156 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 282 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157 retval = nready; 283 retval = nready;
158 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 284 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159 285
160 return retval; 286 return retval;
161} 287}
162 288
163unsigned int eio_npending (void) 289static unsigned int etp_npending (void)
164{ 290{
165 unsigned int retval; 291 unsigned int retval;
166 292
167 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 293 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168 retval = npending; 294 retval = npending;
169 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 295 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170 296
171 return retval; 297 return retval;
172} 298}
173 299
174unsigned int eio_nthreads (void) 300static unsigned int etp_nthreads (void)
175{ 301{
176 unsigned int retval; 302 unsigned int retval;
177 303
178 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 304 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179 retval = started; 305 retval = started;
186 * a somewhat faster data structure might be nice, but 312 * a somewhat faster data structure might be nice, but
187 * with 8 priorities this actually needs <20 insns 313 * with 8 priorities this actually needs <20 insns
188 * per shift, the most expensive operation. 314 * per shift, the most expensive operation.
189 */ 315 */
190typedef struct { 316typedef struct {
191 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 317 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
192 int size; 318 int size;
193} reqq; 319} etp_reqq;
194 320
195static reqq req_queue; 321static etp_reqq req_queue;
196static reqq res_queue; 322static etp_reqq res_queue;
197 323
198static int reqq_push (reqq *q, eio_req *req) 324static int reqq_push (etp_reqq *q, ETP_REQ *req)
199{ 325{
200 int pri = req->pri; 326 int pri = req->pri;
201 req->next = 0; 327 req->next = 0;
202 328
203 if (q->qe[pri]) 329 if (q->qe[pri])
209 q->qe[pri] = q->qs[pri] = req; 335 q->qe[pri] = q->qs[pri] = req;
210 336
211 return q->size++; 337 return q->size++;
212} 338}
213 339
214static eio_req *reqq_shift (reqq *q) 340static ETP_REQ *reqq_shift (etp_reqq *q)
215{ 341{
216 int pri; 342 int pri;
217 343
218 if (!q->size) 344 if (!q->size)
219 return 0; 345 return 0;
220 346
221 --q->size; 347 --q->size;
222 348
223 for (pri = EIO_NUM_PRI; pri--; ) 349 for (pri = ETP_NUM_PRI; pri--; )
224 { 350 {
225 eio_req *req = q->qs[pri]; 351 eio_req *req = q->qs[pri];
226 352
227 if (req) 353 if (req)
228 { 354 {
234 } 360 }
235 361
236 abort (); 362 abort ();
237} 363}
238 364
239static void grp_try_feed (eio_req *grp) 365static void etp_atfork_prepare (void)
240{ 366{
241 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 367 X_LOCK (wrklock);
368 X_LOCK (reqlock);
369 X_LOCK (reslock);
370#if !HAVE_PREADWRITE
371 X_LOCK (preadwritelock);
372#endif
373}
374
375static void etp_atfork_parent (void)
376{
377#if !HAVE_PREADWRITE
378 X_UNLOCK (preadwritelock);
379#endif
380 X_UNLOCK (reslock);
381 X_UNLOCK (reqlock);
382 X_UNLOCK (wrklock);
383}
384
385static void etp_atfork_child (void)
386{
387 ETP_REQ *prv;
388
389 while ((prv = reqq_shift (&req_queue)))
390 ETP_DESTROY (prv);
391
392 while ((prv = reqq_shift (&res_queue)))
393 ETP_DESTROY (prv);
394
395 while (wrk_first.next != &wrk_first)
242 { 396 {
243 int old_len = grp->size; 397 etp_worker *wrk = wrk_first.next;
244 398
245 EIO_FEED (grp); 399 if (wrk->req)
400 ETP_DESTROY (wrk->req);
246 401
247 /* stop if no progress has been made */ 402 etp_worker_clear (wrk);
248 if (old_len == grp->size) 403 etp_worker_free (wrk);
249 {
250 grp->feed = 0;
251 break;
252 }
253 } 404 }
254}
255 405
256static int eio_finish (eio_req *req); 406 started = 0;
407 idle = 0;
408 nreqs = 0;
409 nready = 0;
410 npending = 0;
257 411
258static int grp_dec (eio_req *grp) 412 etp_atfork_parent ();
259{ 413}
260 --grp->size;
261 414
262 /* call feeder, if applicable */ 415static void
263 grp_try_feed (grp); 416etp_once_init (void)
417{
418 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
419}
264 420
265 /* finish, if done */ 421static int
266 if (!grp->size && grp->int1) 422etp_init (void (*want_poll)(void), void (*done_poll)(void))
267 return eio_finish (grp); 423{
268 else 424 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
425
426 pthread_once (&doinit, etp_once_init);
427
428 want_poll_cb = want_poll;
429 done_poll_cb = done_poll;
430
269 return 0; 431 return 0;
270} 432}
271 433
272void eio_destroy (eio_req *req)
273{
274 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
275 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
276
277 EIO_DESTROY (req);
278}
279
280static int eio_finish (eio_req *req)
281{
282 int res = EIO_FINISH (req);
283
284 if (req->grp)
285 {
286 int res2;
287 eio_req *grp = req->grp;
288
289 /* unlink request */
290 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
291 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
292
293 if (grp->grp_first == req)
294 grp->grp_first = req->grp_next;
295
296 res2 = grp_dec (grp);
297
298 if (!res && res2)
299 res = res2;
300 }
301
302 eio_destroy (req);
303
304 return res;
305}
306
307void eio_grp_cancel (eio_req *grp)
308{
309 for (grp = grp->grp_first; grp; grp = grp->grp_next)
310 eio_cancel (grp);
311}
312
313void eio_cancel (eio_req *req)
314{
315 X_LOCK (wrklock);
316 req->flags |= EIO_FLAG_CANCELLED;
317 X_UNLOCK (wrklock);
318
319 eio_grp_cancel (req);
320}
321
322X_THREAD_PROC (eio_proc); 434X_THREAD_PROC (etp_proc);
323 435
324static void start_thread (void) 436static void etp_start_thread (void)
325{ 437{
326 worker *wrk = calloc (1, sizeof (worker)); 438 etp_worker *wrk = calloc (1, sizeof (etp_worker));
327 439
328 /*TODO*/ 440 /*TODO*/
329 assert (("unable to allocate worker thread data", wrk)); 441 assert (("unable to allocate worker thread data", wrk));
330 442
331 X_LOCK (wrklock); 443 X_LOCK (wrklock);
332 444
333 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 445 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
334 { 446 {
335 wrk->prev = &wrk_first; 447 wrk->prev = &wrk_first;
336 wrk->next = wrk_first.next; 448 wrk->next = wrk_first.next;
337 wrk_first.next->prev = wrk; 449 wrk_first.next->prev = wrk;
338 wrk_first.next = wrk; 450 wrk_first.next = wrk;
342 free (wrk); 454 free (wrk);
343 455
344 X_UNLOCK (wrklock); 456 X_UNLOCK (wrklock);
345} 457}
346 458
347static void maybe_start_thread (void) 459static void etp_maybe_start_thread (void)
348{ 460{
349 if (eio_nthreads () >= wanted) 461 if (expect_true (etp_nthreads () >= wanted))
350 return; 462 return;
351 463
352 /* todo: maybe use idle here, but might be less exact */ 464 /* todo: maybe use idle here, but might be less exact */
353 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 465 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
354 return; 466 return;
355 467
356 start_thread (); 468 etp_start_thread ();
357} 469}
358 470
359void eio_submit (eio_req *req)
360{
361 ++nreqs;
362
363 X_LOCK (reqlock);
364 ++nready;
365 reqq_push (&req_queue, req);
366 X_COND_SIGNAL (reqwait);
367 X_UNLOCK (reqlock);
368
369 maybe_start_thread ();
370}
371
372static void end_thread (void) 471static void etp_end_thread (void)
373{ 472{
374 eio_req *req = calloc (1, sizeof (eio_req)); 473 eio_req *req = calloc (1, sizeof (eio_req));
375 474
376 req->type = EIO_QUIT; 475 req->type = -1;
377 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 476 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
378 477
379 X_LOCK (reqlock); 478 X_LOCK (reqlock);
380 reqq_push (&req_queue, req); 479 reqq_push (&req_queue, req);
381 X_COND_SIGNAL (reqwait); 480 X_COND_SIGNAL (reqwait);
382 X_UNLOCK (reqlock); 481 X_UNLOCK (reqlock);
384 X_LOCK (wrklock); 483 X_LOCK (wrklock);
385 --started; 484 --started;
386 X_UNLOCK (wrklock); 485 X_UNLOCK (wrklock);
387} 486}
388 487
389void eio_set_max_poll_time (double nseconds)
390{
391 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
392 max_poll_time = nseconds;
393 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
394}
395
396void eio_set_max_poll_reqs (unsigned int maxreqs)
397{
398 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
399 max_poll_reqs = maxreqs;
400 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
401}
402
403void eio_set_max_idle (unsigned int nthreads)
404{
405 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
406 max_idle = nthreads <= 0 ? 1 : nthreads;
407 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
408}
409
410void eio_set_min_parallel (unsigned int nthreads)
411{
412 if (wanted < nthreads)
413 wanted = nthreads;
414}
415
416void eio_set_max_parallel (unsigned int nthreads)
417{
418 if (wanted > nthreads)
419 wanted = nthreads;
420
421 while (started > wanted)
422 end_thread ();
423}
424
425int eio_poll (void) 488static int etp_poll (void)
426{ 489{
427 int maxreqs = max_poll_reqs; 490 unsigned int maxreqs;
491 unsigned int maxtime;
428 struct timeval tv_start, tv_now; 492 struct timeval tv_start, tv_now;
429 eio_req *req;
430 493
494 X_LOCK (reslock);
495 maxreqs = max_poll_reqs;
496 maxtime = max_poll_time;
497 X_UNLOCK (reslock);
498
431 if (max_poll_time) 499 if (maxtime)
432 gettimeofday (&tv_start, 0); 500 gettimeofday (&tv_start, 0);
433 501
434 for (;;) 502 for (;;)
435 { 503 {
504 ETP_REQ *req;
505
436 maybe_start_thread (); 506 etp_maybe_start_thread ();
437 507
438 X_LOCK (reslock); 508 X_LOCK (reslock);
439 req = reqq_shift (&res_queue); 509 req = reqq_shift (&res_queue);
440 510
441 if (req) 511 if (req)
449 X_UNLOCK (reslock); 519 X_UNLOCK (reslock);
450 520
451 if (!req) 521 if (!req)
452 return 0; 522 return 0;
453 523
524 X_LOCK (reqlock);
454 --nreqs; 525 --nreqs;
526 X_UNLOCK (reqlock);
455 527
456 if (req->type == EIO_GROUP && req->size) 528 if (expect_false (req->type == EIO_GROUP && req->size))
457 { 529 {
458 req->int1 = 1; /* mark request as delayed */ 530 req->int1 = 1; /* mark request as delayed */
459 continue; 531 continue;
460 } 532 }
461 else 533 else
462 { 534 {
463 int res = eio_finish (req); 535 int res = ETP_FINISH (req);
464 if (res) 536 if (expect_false (res))
465 return res; 537 return res;
466 } 538 }
467 539
468 if (maxreqs && !--maxreqs) 540 if (expect_false (maxreqs && !--maxreqs))
469 break; 541 break;
470 542
471 if (max_poll_time) 543 if (maxtime)
472 { 544 {
473 gettimeofday (&tv_now, 0); 545 gettimeofday (&tv_now, 0);
474 546
475 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 547 if (tvdiff (&tv_start, &tv_now) >= maxtime)
476 break; 548 break;
477 } 549 }
478 } 550 }
479 551
480 errno = EAGAIN; 552 errno = EAGAIN;
481 return -1; 553 return -1;
482} 554}
483 555
556static void etp_cancel (ETP_REQ *req)
557{
558 X_LOCK (wrklock);
559 req->flags |= EIO_FLAG_CANCELLED;
560 X_UNLOCK (wrklock);
561
562 eio_grp_cancel (req);
563}
564
565static void etp_submit (ETP_REQ *req)
566{
567 req->pri -= ETP_PRI_MIN;
568
569 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
570 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
571
572 if (expect_false (req->type == EIO_GROUP))
573 {
574 /* I hope this is worth it :/ */
575 X_LOCK (reqlock);
576 ++nreqs;
577 X_UNLOCK (reqlock);
578
579 X_LOCK (reslock);
580
581 ++npending;
582
583 if (!reqq_push (&res_queue, req) && want_poll_cb)
584 want_poll_cb ();
585
586 X_UNLOCK (reslock);
587 }
588 else
589 {
590 X_LOCK (reqlock);
591 ++nreqs;
592 ++nready;
593 reqq_push (&req_queue, req);
594 X_COND_SIGNAL (reqwait);
595 X_UNLOCK (reqlock);
596
597 etp_maybe_start_thread ();
598 }
599}
600
601static void etp_set_max_poll_time (double nseconds)
602{
603 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
604 max_poll_time = nseconds * EIO_TICKS;
605 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
606}
607
608static void etp_set_max_poll_reqs (unsigned int maxreqs)
609{
610 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
611 max_poll_reqs = maxreqs;
612 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
613}
614
615static void etp_set_max_idle (unsigned int nthreads)
616{
617 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
618 max_idle = nthreads <= 0 ? 1 : nthreads;
619 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
620}
621
622static void etp_set_min_parallel (unsigned int nthreads)
623{
624 if (wanted < nthreads)
625 wanted = nthreads;
626}
627
628static void etp_set_max_parallel (unsigned int nthreads)
629{
630 if (wanted > nthreads)
631 wanted = nthreads;
632
633 while (started > wanted)
634 etp_end_thread ();
635}
636
637/*****************************************************************************/
638
639static void grp_try_feed (eio_req *grp)
640{
641 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
642 {
643 grp->flags &= ~EIO_FLAG_GROUPADD;
644
645 EIO_FEED (grp);
646
647 /* stop if no progress has been made */
648 if (!(grp->flags & EIO_FLAG_GROUPADD))
649 {
650 grp->feed = 0;
651 break;
652 }
653 }
654}
655
656static int grp_dec (eio_req *grp)
657{
658 --grp->size;
659
660 /* call feeder, if applicable */
661 grp_try_feed (grp);
662
663 /* finish, if done */
664 if (!grp->size && grp->int1)
665 return eio_finish (grp);
666 else
667 return 0;
668}
669
670void eio_destroy (eio_req *req)
671{
672 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
673 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
674
675 EIO_DESTROY (req);
676}
677
678static int eio_finish (eio_req *req)
679{
680 int res = EIO_FINISH (req);
681
682 if (req->grp)
683 {
684 int res2;
685 eio_req *grp = req->grp;
686
687 /* unlink request */
688 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
689 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
690
691 if (grp->grp_first == req)
692 grp->grp_first = req->grp_next;
693
694 res2 = grp_dec (grp);
695
696 if (!res && res2)
697 res = res2;
698 }
699
700 eio_destroy (req);
701
702 return res;
703}
704
705void eio_grp_cancel (eio_req *grp)
706{
707 for (grp = grp->grp_first; grp; grp = grp->grp_next)
708 eio_cancel (grp);
709}
710
711void eio_cancel (eio_req *req)
712{
713 etp_cancel (req);
714}
715
716void eio_submit (eio_req *req)
717{
718 etp_submit (req);
719}
720
721unsigned int eio_nreqs (void)
722{
723 return etp_nreqs ();
724}
725
726unsigned int eio_nready (void)
727{
728 return etp_nready ();
729}
730
731unsigned int eio_npending (void)
732{
733 return etp_npending ();
734}
735
736unsigned int eio_nthreads (void)
737{
738 return etp_nthreads ();
739}
740
741void eio_set_max_poll_time (double nseconds)
742{
743 etp_set_max_poll_time (nseconds);
744}
745
746void eio_set_max_poll_reqs (unsigned int maxreqs)
747{
748 etp_set_max_poll_reqs (maxreqs);
749}
750
751void eio_set_max_idle (unsigned int nthreads)
752{
753 etp_set_max_idle (nthreads);
754}
755
756void eio_set_min_parallel (unsigned int nthreads)
757{
758 etp_set_min_parallel (nthreads);
759}
760
761void eio_set_max_parallel (unsigned int nthreads)
762{
763 etp_set_max_parallel (nthreads);
764}
765
766int eio_poll (void)
767{
768 return etp_poll ();
769}
770
484/*****************************************************************************/ 771/*****************************************************************************/
485/* work around various missing functions */ 772/* work around various missing functions */
486 773
487#if !HAVE_PREADWRITE 774#if !HAVE_PREADWRITE
775# undef pread
776# undef pwrite
488# define pread eio__pread 777# define pread eio__pread
489# define pwrite eio__pwrite 778# define pwrite eio__pwrite
490
491/*
492 * make our pread/pwrite safe against themselves, but not against
493 * normal read/write by using a mutex. slows down execution a lot,
494 * but that's your problem, not mine.
495 */
496static mutex_t preadwritelock = X_MUTEX_INIT;
497 779
498static ssize_t 780static ssize_t
499eio__pread (int fd, void *buf, size_t count, off_t offset) 781eio__pread (int fd, void *buf, size_t count, off_t offset)
500{ 782{
501 ssize_t res; 783 ssize_t res;
519 801
520 X_LOCK (preadwritelock); 802 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR); 803 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET); 804 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count); 805 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET); 806 lseek (fd, ooffset, SEEK_SET);
525 X_UNLOCK (preadwritelock); 807 X_UNLOCK (preadwritelock);
526 808
527 return res; 809 return res;
528} 810}
529#endif 811#endif
530 812
531#ifndef HAVE_FUTIMES 813#ifndef HAVE_FUTIMES
532 814
815# undef utimes
816# undef futimes
533# define utimes(path,times) eio__utimes (path, times) 817# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) eio__futimes (fd, times) 818# define futimes(fd,times) eio__futimes (fd, times)
535 819
536static int 820static int
537eio__utimes (const char *filename, const struct timeval times[2]) 821eio__utimes (const char *filename, const struct timeval times[2])
556} 840}
557 841
558#endif 842#endif
559 843
560#if !HAVE_FDATASYNC 844#if !HAVE_FDATASYNC
845# undef fdatasync
561# define fdatasync fsync 846# define fdatasync(fd) fsync (fd)
562#endif 847#endif
848
849/* sync_file_range always needs emulation */
850int
851eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
852{
853#if HAVE_SYNC_FILE_RANGE
854 int res;
855
856 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
857 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
858 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
859 {
860 flags = 0
861 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
862 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
863 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
864 }
865
866 res = sync_file_range (fd, offset, nbytes, flags);
867
868 if (!res || errno != ENOSYS)
869 return res;
870#endif
871
872 /* even though we could play tricks with the flags, it's better to always
873 * call fdatasync, as that matches the expectation of its users best */
874 return fdatasync (fd);
875}
563 876
564#if !HAVE_READAHEAD 877#if !HAVE_READAHEAD
878# undef readahead
565# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 879# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 880
567static ssize_t 881static ssize_t
568eio__readahead (int fd, off_t offset, size_t count, worker *self) 882eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
569{ 883{
570 size_t todo = count; 884 size_t todo = count;
571 dBUF; 885 dBUF;
572 886
573 while (todo > 0) 887 while (todo > 0)
583 return count; 897 return count;
584} 898}
585 899
586#endif 900#endif
587 901
588#if !HAVE_READDIR_R
589# define readdir_r eio__readdir_r
590
591static mutex_t readdirlock = X_MUTEX_INIT;
592
593static int
594eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
595{
596 EIO_STRUCT_DIRENT *e;
597 int errorno;
598
599 X_LOCK (readdirlock);
600
601 e = readdir (dirp);
602 errorno = errno;
603
604 if (e)
605 {
606 *res = ent;
607 strcpy (ent->d_name, e->d_name);
608 }
609 else
610 *res = 0;
611
612 X_UNLOCK (readdirlock);
613
614 errno = errorno;
615 return e ? 0 : -1;
616}
617#endif
618
619/* sendfile always needs emulation */ 902/* sendfile always needs emulation */
620static ssize_t 903static ssize_t
621eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self) 904eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
622{ 905{
623 ssize_t res; 906 ssize_t res;
624 907
625 if (!count) 908 if (!count)
626 return 0; 909 return 0;
627 910
628#if HAVE_SENDFILE 911#if HAVE_SENDFILE
629# if __linux 912# if __linux
630 res = sendfile (ofd, ifd, &offset, count); 913 res = sendfile (ofd, ifd, &offset, count);
631 914
632# elif __freebsd 915# elif __FreeBSD__
633 /* 916 /*
634 * Of course, the freebsd sendfile is a dire hack with no thoughts 917 * Of course, the freebsd sendfile is a dire hack with no thoughts
635 * wasted on making it similar to other I/O functions. 918 * wasted on making it similar to other I/O functions.
636 */ 919 */
637 { 920 {
638 off_t sbytes; 921 off_t sbytes;
639 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 922 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
640 923
641 if (res < 0 && sbytes) 924 /* freebsd' sendfile will return 0 on success */
642 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 925 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
926 /* not on e.g. EIO or EPIPE - sounds broken */
927 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
928 res = sbytes;
929 }
930
931# elif defined (__APPLE__)
932
933 {
934 off_t sbytes = count;
935 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
936
937 /* according to the manpage, sbytes is always valid */
938 if (sbytes)
643 res = sbytes; 939 res = sbytes;
644 } 940 }
645 941
646# elif __hpux 942# elif __hpux
647 res = sendfile (ofd, ifd, offset, count, 0, 0); 943 res = sendfile (ofd, ifd, offset, count, 0, 0);
661 if (res < 0 && sbytes) 957 if (res < 0 && sbytes)
662 res = sbytes; 958 res = sbytes;
663 } 959 }
664 960
665# endif 961# endif
962
963#elif defined (_WIN32)
964
965 /* does not work, just for documentation of what would need to be done */
966 {
967 HANDLE h = TO_SOCKET (ifd);
968 SetFilePointer (h, offset, 0, FILE_BEGIN);
969 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
970 }
971
666#else 972#else
667 res = -1; 973 res = -1;
668 errno = ENOSYS; 974 errno = ENOSYS;
669#endif 975#endif
670 976
671 if (res < 0 977 if (res < 0
672 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 978 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
979 /* BSDs */
980#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
981 || errno == ENOTSUP
982#endif
983 || errno == EOPNOTSUPP /* BSDs */
673#if __solaris 984#if __solaris
674 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 985 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
675#endif 986#endif
676 ) 987 )
677 ) 988 )
708 } 1019 }
709 1020
710 return res; 1021 return res;
711} 1022}
712 1023
1024static signed char
1025eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1026{
1027 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1028 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
1029}
1030
1031#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1032
1033#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1034#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1035
1036static void
1037eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1038{
1039 unsigned char bits [9 + sizeof (ino_t) * 8];
1040 unsigned char *bit = bits;
1041
1042 assert (CHAR_BIT == 8);
1043 assert (sizeof (eio_dirent) * 8 < 256);
1044 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
1045 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
1046
1047 if (size <= EIO_SORT_FAST)
1048 return;
1049
1050 /* first prepare an array of bits to test in our radix sort */
1051 /* try to take endianness into account, as well as differences in ino_t sizes */
1052 /* inode_bits must contain all inodes ORed together */
1053 /* which is used to skip bits that are 0 everywhere, which is very common */
1054 {
1055 ino_t endianness;
1056 int i, j;
1057
1058 /* we store the byte offset of byte n into byte n of "endianness" */
1059 for (i = 0; i < sizeof (ino_t); ++i)
1060 ((unsigned char *)&endianness)[i] = i;
1061
1062 *bit++ = 0;
1063
1064 for (i = 0; i < sizeof (ino_t); ++i)
1065 {
1066 /* shifting off the byte offsets out of "endianness" */
1067 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1068 endianness >>= 8;
1069
1070 for (j = 0; j < 8; ++j)
1071 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1072 *bit++ = offs + j;
1073 }
1074
1075 for (j = 0; j < 8; ++j)
1076 if (score_bits & (1 << j))
1077 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1078 }
1079
1080 /* now actually do the sorting (a variant of MSD radix sort) */
1081 {
1082 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1083 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1084 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1085 int stk_idx = 0;
1086
1087 base_stk [stk_idx] = dents;
1088 end_stk [stk_idx] = dents + size;
1089 bit_stk [stk_idx] = bit - 1;
1090
1091 do
1092 {
1093 base = base_stk [stk_idx];
1094 end = end_stk [stk_idx];
1095 bit = bit_stk [stk_idx];
1096
1097 for (;;)
1098 {
1099 unsigned char O = *bit >> 3;
1100 unsigned char M = 1 << (*bit & 7);
1101
1102 eio_dirent *a = base;
1103 eio_dirent *b = end;
1104
1105 if (b - a < EIO_SORT_CUTOFF)
1106 break;
1107
1108 /* now bit-partition the array on the bit */
1109 /* this ugly asymmetric loop seems to perform much better than typical */
1110 /* partition algos found in the literature */
1111 do
1112 if (!(((unsigned char *)a)[O] & M))
1113 ++a;
1114 else if (!(((unsigned char *)--b)[O] & M))
1115 {
1116 eio_dirent tmp = *a; *a = *b; *b = tmp;
1117 ++a;
1118 }
1119 while (b > a);
1120
1121 /* next bit, or stop, if no bits left in this path */
1122 if (!*--bit)
1123 break;
1124
1125 base_stk [stk_idx] = a;
1126 end_stk [stk_idx] = end;
1127 bit_stk [stk_idx] = bit;
1128 ++stk_idx;
1129
1130 end = a;
1131 }
1132 }
1133 while (stk_idx--);
1134 }
1135}
1136
1137static void
1138eio_dent_insertion_sort (eio_dirent *dents, int size)
1139{
1140 /* first move the smallest element to the front, to act as a sentinel */
1141 {
1142 int i;
1143 eio_dirent *min = dents;
1144
1145 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1146 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1147 if (EIO_DENT_CMP (dents [i], <, *min))
1148 min = &dents [i];
1149
1150 /* swap elements 0 and j (minimum) */
1151 {
1152 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1153 }
1154 }
1155
1156 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1157 {
1158 eio_dirent *i, *j;
1159
1160 for (i = dents + 1; i < dents + size; ++i)
1161 {
1162 eio_dirent value = *i;
1163
1164 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1165 j [1] = j [0];
1166
1167 j [1] = value;
1168 }
1169 }
1170}
1171
1172static void
1173eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1174{
1175 if (size <= 1)
1176 return; /* our insertion sort relies on size > 0 */
1177
1178 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1179 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1180 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1181
1182 /* use an insertion sort at the end, or for small arrays, */
1183 /* as insertion sort is more efficient for small partitions */
1184 eio_dent_insertion_sort (dents, size);
1185}
1186
713/* read a full directory */ 1187/* read a full directory */
714static void 1188static void
715eio__scandir (eio_req *req, worker *self) 1189eio__scandir (eio_req *req, etp_worker *self)
716{ 1190{
717 DIR *dirp; 1191 DIR *dirp;
718 union
719 {
720 EIO_STRUCT_DIRENT d;
721 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
722 } *u;
723 EIO_STRUCT_DIRENT *entp; 1192 EIO_STRUCT_DIRENT *entp;
724 char *name, *names; 1193 char *name, *names;
725 int memlen = 4096; 1194 int namesalloc = 4096;
726 int memofs = 0; 1195 int namesoffs = 0;
1196 int flags = req->int1;
1197 eio_dirent *dents = 0;
1198 int dentalloc = 128;
727 int res = 0; 1199 int dentoffs = 0;
1200 ino_t inode_bits = 0;
1201
1202 req->result = -1;
1203
1204 if (!(flags & EIO_READDIR_DENTS))
1205 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
728 1206
729 X_LOCK (wrklock); 1207 X_LOCK (wrklock);
1208 /* the corresponding closedir is in ETP_WORKER_CLEAR */
730 self->dirp = dirp = opendir (req->ptr1); 1209 self->dirp = dirp = opendir (req->ptr1);
731 self->dbuf = u = malloc (sizeof (*u));
732 req->flags |= EIO_FLAG_PTR2_FREE; 1210 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1211 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
733 req->ptr2 = names = malloc (memlen); 1212 req->ptr2 = names = malloc (namesalloc);
734 X_UNLOCK (wrklock); 1213 X_UNLOCK (wrklock);
735 1214
736 if (dirp && u && names) 1215 if (dirp && names && (!flags || dents))
737 for (;;) 1216 for (;;)
738 { 1217 {
739 errno = 0; 1218 errno = 0;
740 readdir_r (dirp, &u->d, &entp); 1219 entp = readdir (dirp);
741 1220
742 if (!entp) 1221 if (!entp)
1222 {
1223 if (errno)
1224 break;
1225
1226 /* sort etc. */
1227 req->int1 = flags;
1228 req->result = dentoffs;
1229
1230 if (flags & EIO_READDIR_STAT_ORDER)
1231 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
1232 else if (flags & EIO_READDIR_DIRS_FIRST)
1233 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1234 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1235 else
1236 {
1237 /* in this case, all is known, and we just put dirs first and sort them */
1238 eio_dirent *oth = dents + dentoffs;
1239 eio_dirent *dir = dents;
1240
1241 /* now partition dirs to the front, and non-dirs to the back */
1242 /* by walking from both sides and swapping if necessary */
1243 /* also clear score, so it doesn't influence sorting */
1244 while (oth > dir)
1245 {
1246 if (dir->type == EIO_DT_DIR)
1247 ++dir;
1248 else if ((--oth)->type == EIO_DT_DIR)
1249 {
1250 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1251
1252 ++dir;
1253 }
1254 }
1255
1256 /* now sort the dirs only */
1257 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1258 }
1259
743 break; 1260 break;
1261 }
744 1262
1263 /* now add the entry to our list(s) */
745 name = entp->d_name; 1264 name = entp->d_name;
746 1265
1266 /* skip . and .. entries */
747 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1267 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
748 { 1268 {
749 int len = strlen (name) + 1; 1269 int len = D_NAMLEN (entp) + 1;
750 1270
751 res++; 1271 while (expect_false (namesoffs + len > namesalloc))
752
753 while (memofs + len > memlen)
754 { 1272 {
755 memlen *= 2; 1273 namesalloc *= 2;
756 X_LOCK (wrklock); 1274 X_LOCK (wrklock);
757 req->ptr2 = names = realloc (names, memlen); 1275 req->ptr2 = names = realloc (names, namesalloc);
758 X_UNLOCK (wrklock); 1276 X_UNLOCK (wrklock);
759 1277
760 if (!names) 1278 if (!names)
761 break; 1279 break;
762 } 1280 }
763 1281
764 memcpy (names + memofs, name, len); 1282 memcpy (names + namesoffs, name, len);
1283
1284 if (dents)
1285 {
1286 struct eio_dirent *ent;
1287
1288 if (expect_false (dentoffs == dentalloc))
1289 {
1290 dentalloc *= 2;
1291 X_LOCK (wrklock);
1292 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1293 X_UNLOCK (wrklock);
1294
1295 if (!dents)
1296 break;
1297 }
1298
1299 ent = dents + dentoffs;
1300
1301 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1302 ent->namelen = len - 1;
1303 ent->inode = D_INO (entp);
1304
1305 inode_bits |= ent->inode;
1306
1307 switch (D_TYPE (entp))
1308 {
1309 default:
1310 ent->type = EIO_DT_UNKNOWN;
1311 flags |= EIO_READDIR_FOUND_UNKNOWN;
1312 break;
1313
1314 #ifdef DT_FIFO
1315 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1316 #endif
1317 #ifdef DT_CHR
1318 case DT_CHR: ent->type = EIO_DT_CHR; break;
1319 #endif
1320 #ifdef DT_MPC
1321 case DT_MPC: ent->type = EIO_DT_MPC; break;
1322 #endif
1323 #ifdef DT_DIR
1324 case DT_DIR: ent->type = EIO_DT_DIR; break;
1325 #endif
1326 #ifdef DT_NAM
1327 case DT_NAM: ent->type = EIO_DT_NAM; break;
1328 #endif
1329 #ifdef DT_BLK
1330 case DT_BLK: ent->type = EIO_DT_BLK; break;
1331 #endif
1332 #ifdef DT_MPB
1333 case DT_MPB: ent->type = EIO_DT_MPB; break;
1334 #endif
1335 #ifdef DT_REG
1336 case DT_REG: ent->type = EIO_DT_REG; break;
1337 #endif
1338 #ifdef DT_NWK
1339 case DT_NWK: ent->type = EIO_DT_NWK; break;
1340 #endif
1341 #ifdef DT_CMP
1342 case DT_CMP: ent->type = EIO_DT_CMP; break;
1343 #endif
1344 #ifdef DT_LNK
1345 case DT_LNK: ent->type = EIO_DT_LNK; break;
1346 #endif
1347 #ifdef DT_SOCK
1348 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1349 #endif
1350 #ifdef DT_DOOR
1351 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1352 #endif
1353 #ifdef DT_WHT
1354 case DT_WHT: ent->type = EIO_DT_WHT; break;
1355 #endif
1356 }
1357
1358 ent->score = 7;
1359
1360 if (flags & EIO_READDIR_DIRS_FIRST)
1361 {
1362 if (ent->type == EIO_DT_UNKNOWN)
1363 {
1364 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1365 ent->score = 1;
1366 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1367 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1368 }
1369 else if (ent->type == EIO_DT_DIR)
1370 ent->score = 0;
1371 }
1372 }
1373
765 memofs += len; 1374 namesoffs += len;
1375 ++dentoffs;
1376 }
1377
1378 if (EIO_CANCELLED (req))
1379 {
1380 errno = ECANCELED;
1381 break;
766 } 1382 }
767 } 1383 }
1384}
768 1385
769 if (errno) 1386#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
770 res = -1; 1387# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
771 1388#else
772 req->result = res; 1389
1390int
1391eio__msync (void *mem, size_t len, int flags)
1392{
1393 if (EIO_MS_ASYNC != MS_SYNC
1394 || EIO_MS_INVALIDATE != MS_INVALIDATE
1395 || EIO_MS_SYNC != MS_SYNC)
1396 {
1397 flags = 0
1398 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1399 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1400 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1401 }
1402
1403 return msync (mem, len, flags);
1404}
1405
1406#endif
1407
1408int
1409eio__mtouch (void *mem, size_t len, int flags)
1410{
1411 intptr_t addr = (intptr_t)mem;
1412 intptr_t end = addr + len;
1413#ifdef PAGESIZE
1414 const intptr_t page = PAGESIZE;
1415#else
1416 static intptr_t page;
1417
1418 if (!page)
1419 page = sysconf (_SC_PAGESIZE);
1420#endif
1421
1422 /* round down to start of page, although this is probably useless */
1423 addr &= ~(page - 1); /* assume page size is always a power of two */
1424
1425 if (addr < end)
1426 if (flags & EIO_MT_MODIFY) /* modify */
1427 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1428 else
1429 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1430
1431 return 0;
773} 1432}
774 1433
775/*****************************************************************************/ 1434/*****************************************************************************/
776 1435
777#define ALLOC(len) \ 1436#define ALLOC(len) \
787 req->result = -1; \ 1446 req->result = -1; \
788 break; \ 1447 break; \
789 } \ 1448 } \
790 } 1449 }
791 1450
792X_THREAD_PROC (eio_proc) 1451X_THREAD_PROC (etp_proc)
793{ 1452{
794 eio_req *req; 1453 ETP_REQ *req;
795 struct timespec ts; 1454 struct timespec ts;
796 worker *self = (worker *)thr_arg; 1455 etp_worker *self = (etp_worker *)thr_arg;
797 1456
798 /* try to distribute timeouts somewhat randomly */ 1457 /* try to distribute timeouts somewhat randomly */
799 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1458 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
800 1459
801 for (;;) 1460 for (;;)
802 { 1461 {
803 ts.tv_sec = time (0) + IDLE_TIMEOUT;
804
805 X_LOCK (reqlock); 1462 X_LOCK (reqlock);
806 1463
807 for (;;) 1464 for (;;)
808 { 1465 {
809 self->req = req = reqq_shift (&req_queue); 1466 self->req = req = reqq_shift (&req_queue);
811 if (req) 1468 if (req)
812 break; 1469 break;
813 1470
814 ++idle; 1471 ++idle;
815 1472
1473 ts.tv_sec = time (0) + IDLE_TIMEOUT;
816 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1474 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
817 { 1475 {
818 if (idle > max_idle) 1476 if (idle > max_idle)
819 { 1477 {
820 --idle; 1478 --idle;
825 goto quit; 1483 goto quit;
826 } 1484 }
827 1485
828 /* we are allowed to idle, so do so without any timeout */ 1486 /* we are allowed to idle, so do so without any timeout */
829 X_COND_WAIT (reqwait, reqlock); 1487 X_COND_WAIT (reqwait, reqlock);
830 ts.tv_sec = time (0) + IDLE_TIMEOUT;
831 } 1488 }
832 1489
833 --idle; 1490 --idle;
834 } 1491 }
835 1492
836 --nready; 1493 --nready;
837 1494
838 X_UNLOCK (reqlock); 1495 X_UNLOCK (reqlock);
839 1496
840 errno = 0; /* strictly unnecessary */ 1497 if (req->type < 0)
1498 goto quit;
841 1499
842 if (!EIO_CANCELLED (req)) 1500 if (!EIO_CANCELLED (req))
843 switch (req->type) 1501 ETP_EXECUTE (self, req);
844 {
845 case EIO_READ: ALLOC (req->size);
846 req->result = req->offs >= 0
847 ? pread (req->int1, req->ptr2, req->size, req->offs)
848 : read (req->int1, req->ptr2, req->size); break;
849 case EIO_WRITE: req->result = req->offs >= 0
850 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
851 : write (req->int1, req->ptr2, req->size); break;
852
853 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
854 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
855
856 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
857 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
858 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
859 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
860 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
861 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
862
863 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
864 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
865 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
866 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
867 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
868 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
869
870 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
871 case EIO_CLOSE: req->result = close (req->int1); break;
872 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
873 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
874 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
875 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
876 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
877 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
878 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
879 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
880
881 case EIO_READLINK: ALLOC (NAME_MAX);
882 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
883
884 case EIO_SYNC: req->result = 0; sync (); break;
885 case EIO_FSYNC: req->result = fsync (req->int1); break;
886 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
887
888 case EIO_READDIR: eio__scandir (req, self); break;
889
890 case EIO_BUSY:
891#ifdef _WIN32
892 Sleep (req->nv1 * 1000.);
893#else
894 {
895 struct timeval tv;
896
897 tv.tv_sec = req->nv1;
898 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
899
900 req->result = select (0, 0, 0, 0, &tv);
901 }
902#endif
903 break;
904
905 case EIO_UTIME:
906 case EIO_FUTIME:
907 {
908 struct timeval tv[2];
909 struct timeval *times;
910
911 if (req->nv1 != -1. || req->nv2 != -1.)
912 {
913 tv[0].tv_sec = req->nv1;
914 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
915 tv[1].tv_sec = req->nv2;
916 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
917
918 times = tv;
919 }
920 else
921 times = 0;
922
923
924 req->result = req->type == EIO_FUTIME
925 ? futimes (req->int1, times)
926 : utimes (req->ptr1, times);
927 }
928
929 case EIO_GROUP:
930 case EIO_NOP:
931 req->result = 0;
932 break;
933
934 case EIO_QUIT:
935 goto quit;
936
937 default:
938 req->result = -1;
939 break;
940 }
941
942 req->errorno = errno;
943 1502
944 X_LOCK (reslock); 1503 X_LOCK (reslock);
945 1504
946 ++npending; 1505 ++npending;
947 1506
948 if (!reqq_push (&res_queue, req) && want_poll_cb) 1507 if (!reqq_push (&res_queue, req) && want_poll_cb)
949 want_poll_cb (); 1508 want_poll_cb ();
950 1509
951 self->req = 0; 1510 self->req = 0;
952 worker_clear (self); 1511 etp_worker_clear (self);
953 1512
954 X_UNLOCK (reslock); 1513 X_UNLOCK (reslock);
955 } 1514 }
956 1515
957quit: 1516quit:
958 X_LOCK (wrklock); 1517 X_LOCK (wrklock);
959 worker_free (self); 1518 etp_worker_free (self);
960 X_UNLOCK (wrklock); 1519 X_UNLOCK (wrklock);
961 1520
962 return 0; 1521 return 0;
963} 1522}
964 1523
965/*****************************************************************************/ 1524/*****************************************************************************/
966 1525
967static void eio_atfork_prepare (void)
968{
969 X_LOCK (wrklock);
970 X_LOCK (reqlock);
971 X_LOCK (reslock);
972#if !HAVE_PREADWRITE
973 X_LOCK (preadwritelock);
974#endif
975#if !HAVE_READDIR_R
976 X_LOCK (readdirlock);
977#endif
978}
979
980static void eio_atfork_parent (void)
981{
982#if !HAVE_READDIR_R
983 X_UNLOCK (readdirlock);
984#endif
985#if !HAVE_PREADWRITE
986 X_UNLOCK (preadwritelock);
987#endif
988 X_UNLOCK (reslock);
989 X_UNLOCK (reqlock);
990 X_UNLOCK (wrklock);
991}
992
993static void eio_atfork_child (void)
994{
995 eio_req *prv;
996
997 while (prv = reqq_shift (&req_queue))
998 eio_destroy (prv);
999
1000 while (prv = reqq_shift (&res_queue))
1001 eio_destroy (prv);
1002
1003 while (wrk_first.next != &wrk_first)
1004 {
1005 worker *wrk = wrk_first.next;
1006
1007 if (wrk->req)
1008 eio_destroy (wrk->req);
1009
1010 worker_clear (wrk);
1011 worker_free (wrk);
1012 }
1013
1014 started = 0;
1015 idle = 0;
1016 nreqs = 0;
1017 nready = 0;
1018 npending = 0;
1019
1020 eio_atfork_parent ();
1021}
1022
1023int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1526int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1024{ 1527{
1025 want_poll_cb = want_poll; 1528 return etp_init (want_poll, done_poll);
1026 done_poll_cb = done_poll;
1027
1028#ifdef _WIN32
1029 X_MUTEX_CHECK (wrklock);
1030 X_MUTEX_CHECK (reslock);
1031 X_MUTEX_CHECK (reqlock);
1032 X_MUTEX_CHECK (reqwait);
1033 X_MUTEX_CHECK (preadwritelock);
1034 X_MUTEX_CHECK (readdirlock);
1035
1036 X_COND_CHECK (reqwait);
1037#endif
1038
1039 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1040} 1529}
1041 1530
1042static void eio_api_destroy (eio_req *req) 1531static void eio_api_destroy (eio_req *req)
1043{ 1532{
1044 free (req); 1533 free (req);
1049 \ 1538 \
1050 req = (eio_req *)calloc (1, sizeof *req); \ 1539 req = (eio_req *)calloc (1, sizeof *req); \
1051 if (!req) \ 1540 if (!req) \
1052 return 0; \ 1541 return 0; \
1053 \ 1542 \
1054 req->type = rtype; \ 1543 req->type = rtype; \
1055 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 1544 req->pri = pri; \
1056 req->finish = cb; \ 1545 req->finish = cb; \
1546 req->data = data; \
1057 req->destroy = eio_api_destroy; 1547 req->destroy = eio_api_destroy;
1058 1548
1059#define SEND eio_submit (req); return req 1549#define SEND eio_submit (req); return req
1060 1550
1061#define PATH \ 1551#define PATH \
1065 { \ 1555 { \
1066 eio_api_destroy (req); \ 1556 eio_api_destroy (req); \
1067 return 0; \ 1557 return 0; \
1068 } 1558 }
1069 1559
1070eio_req *eio_nop (eio_cb cb) 1560static void eio_execute (etp_worker *self, eio_req *req)
1561{
1562 errno = 0;
1563
1564 switch (req->type)
1565 {
1566 case EIO_READ: ALLOC (req->size);
1567 req->result = req->offs >= 0
1568 ? pread (req->int1, req->ptr2, req->size, req->offs)
1569 : read (req->int1, req->ptr2, req->size); break;
1570 case EIO_WRITE: req->result = req->offs >= 0
1571 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1572 : write (req->int1, req->ptr2, req->size); break;
1573
1574 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1575 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1576
1577 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1578 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1579 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1580 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1581 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1582 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1583
1584 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1585 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1586 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1587 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1588
1589 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1590 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1591 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1592 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1593 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1594 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1595
1596 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1597 case EIO_CLOSE: req->result = close (req->int1); break;
1598 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1599 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1600 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1601 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1602 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1603 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1604 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1605 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1606
1607 case EIO_READLINK: ALLOC (NAME_MAX);
1608 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1609
1610 case EIO_SYNC: req->result = 0; sync (); break;
1611 case EIO_FSYNC: req->result = fsync (req->int1); break;
1612 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1613 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1614 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1615 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1616
1617 case EIO_READDIR: eio__scandir (req, self); break;
1618
1619 case EIO_BUSY:
1620#ifdef _WIN32
1621 Sleep (req->nv1 * 1000.);
1622#else
1623 {
1624 struct timeval tv;
1625
1626 tv.tv_sec = req->nv1;
1627 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
1628
1629 req->result = select (0, 0, 0, 0, &tv);
1630 }
1631#endif
1632 break;
1633
1634 case EIO_UTIME:
1635 case EIO_FUTIME:
1636 {
1637 struct timeval tv[2];
1638 struct timeval *times;
1639
1640 if (req->nv1 != -1. || req->nv2 != -1.)
1641 {
1642 tv[0].tv_sec = req->nv1;
1643 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1644 tv[1].tv_sec = req->nv2;
1645 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1646
1647 times = tv;
1648 }
1649 else
1650 times = 0;
1651
1652
1653 req->result = req->type == EIO_FUTIME
1654 ? futimes (req->int1, times)
1655 : utimes (req->ptr1, times);
1656 }
1657 break;
1658
1659 case EIO_GROUP:
1660 abort (); /* handled in eio_request */
1661
1662 case EIO_NOP:
1663 req->result = 0;
1664 break;
1665
1666 case EIO_CUSTOM:
1667 ((void (*)(eio_req *))req->feed) (req);
1668 break;
1669
1670 default:
1671 req->result = -1;
1672 break;
1673 }
1674
1675 req->errorno = errno;
1676}
1677
1678#ifndef EIO_NO_WRAPPERS
1679
1680eio_req *eio_nop (int pri, eio_cb cb, void *data)
1071{ 1681{
1072 REQ (EIO_NOP); SEND; 1682 REQ (EIO_NOP); SEND;
1073} 1683}
1074 1684
1075eio_req *eio_busy (double delay, eio_cb cb) 1685eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1076{ 1686{
1077 REQ (EIO_BUSY); req->nv1 = delay; SEND; 1687 REQ (EIO_BUSY); req->nv1 = delay; SEND;
1078} 1688}
1079 1689
1080eio_req *eio_sync (eio_cb cb) 1690eio_req *eio_sync (int pri, eio_cb cb, void *data)
1081{ 1691{
1082 REQ (EIO_SYNC); SEND; 1692 REQ (EIO_SYNC); SEND;
1083} 1693}
1084 1694
1085eio_req *eio_fsync (int fd, eio_cb cb) 1695eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1086{ 1696{
1087 REQ (EIO_FSYNC); req->int1 = fd; SEND; 1697 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1088} 1698}
1089 1699
1700eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1701{
1702 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1703}
1704
1705eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1706{
1707 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1708}
1709
1710eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1711{
1712 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1713}
1714
1090eio_req *eio_fdatasync (int fd, eio_cb cb) 1715eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1091{ 1716{
1092 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1717 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1093} 1718}
1094 1719
1095eio_req *eio_close (int fd, eio_cb cb) 1720eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1096{ 1721{
1097 REQ (EIO_CLOSE); req->int1 = fd; SEND; 1722 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1098} 1723}
1099 1724
1100eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb) 1725eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1101{ 1726{
1102 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 1727 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1103} 1728}
1104 1729
1105eio_req *eio_read (int fd, void *data, size_t length, off_t offset, eio_cb cb) 1730eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1106{ 1731{
1107 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 1732 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1108} 1733}
1109 1734
1110eio_req *eio_write (int fd, void *data, size_t length, off_t offset, eio_cb cb) 1735eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1111{ 1736{
1112 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 1737 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1113} 1738}
1114 1739
1115eio_req *eio_fstat (int fd, eio_cb cb) 1740eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1116{ 1741{
1117 REQ (EIO_FSTAT); req->int1 = fd; SEND; 1742 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1118} 1743}
1119 1744
1745eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1746{
1747 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1748}
1749
1120eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb) 1750eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1121{ 1751{
1122 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 1752 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1123} 1753}
1124 1754
1125eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb) 1755eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1126{ 1756{
1127 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 1757 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1128} 1758}
1129 1759
1130eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb) 1760eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1131{ 1761{
1132 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 1762 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1133} 1763}
1134 1764
1135eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb) 1765eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1136{ 1766{
1137 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 1767 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1138} 1768}
1139 1769
1140eio_req *eio_dup2 (int fd, int fd2, eio_cb cb) 1770eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1141{ 1771{
1142 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND; 1772 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1143} 1773}
1144 1774
1145eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb) 1775eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1146{ 1776{
1147 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 1777 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1148} 1778}
1149 1779
1150eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb) 1780eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1151{ 1781{
1152 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 1782 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1153} 1783}
1154 1784
1155eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb) 1785eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1156{ 1786{
1157 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 1787 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1158} 1788}
1159 1789
1160eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb) 1790eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1161{ 1791{
1162 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 1792 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1163} 1793}
1164 1794
1165eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 1795eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1166{ 1796{
1167 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 1797 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1168} 1798}
1169 1799
1170eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb) 1800eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1171{ 1801{
1172 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND; 1802 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1173} 1803}
1174 1804
1175eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb) 1805eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1176{ 1806{
1177 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND; 1807 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1178} 1808}
1179 1809
1180static eio_req * 1810static eio_req *
1181eio__1path (int type, const char *path, eio_cb cb) 1811eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1182{ 1812{
1183 REQ (type); PATH; SEND; 1813 REQ (type); PATH; SEND;
1184} 1814}
1185 1815
1186eio_req *eio_readlink (const char *path, eio_cb cb) 1816eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1187{ 1817{
1188 return eio__1path (EIO_READLINK, path, cb); 1818 return eio__1path (EIO_READLINK, path, pri, cb, data);
1189} 1819}
1190 1820
1191eio_req *eio_stat (const char *path, eio_cb cb) 1821eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1192{ 1822{
1193 return eio__1path (EIO_STAT, path, cb); 1823 return eio__1path (EIO_STAT, path, pri, cb, data);
1194} 1824}
1195 1825
1196eio_req *eio_lstat (const char *path, eio_cb cb) 1826eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1197{ 1827{
1198 return eio__1path (EIO_LSTAT, path, cb); 1828 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1199} 1829}
1200 1830
1831eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1832{
1833 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1834}
1835
1201eio_req *eio_unlink (const char *path, eio_cb cb) 1836eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1202{ 1837{
1203 return eio__1path (EIO_UNLINK, path, cb); 1838 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1204} 1839}
1205 1840
1206eio_req *eio_rmdir (const char *path, eio_cb cb) 1841eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1207{ 1842{
1208 return eio__1path (EIO_RMDIR, path, cb); 1843 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1209} 1844}
1210 1845
1211eio_req *eio_readdir (const char *path, eio_cb cb) 1846eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1212{ 1847{
1213 return eio__1path (EIO_READDIR, path, cb); 1848 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1214} 1849}
1215 1850
1216eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb) 1851eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1217{ 1852{
1218 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1853 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1219} 1854}
1220 1855
1221static eio_req * 1856static eio_req *
1222eio__2path (int type, const char *path, const char *new_path, eio_cb cb) 1857eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1223{ 1858{
1224 REQ (type); PATH; 1859 REQ (type); PATH;
1225 1860
1226 req->flags |= EIO_FLAG_PTR2_FREE; 1861 req->flags |= EIO_FLAG_PTR2_FREE;
1227 req->ptr2 = strdup (new_path); 1862 req->ptr2 = strdup (new_path);
1232 } 1867 }
1233 1868
1234 SEND; 1869 SEND;
1235} 1870}
1236 1871
1237eio_req *eio_link (const char *path, const char *new_path, eio_cb cb) 1872eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1238{ 1873{
1239 return eio__2path (EIO_LINK, path, new_path, cb); 1874 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
1240} 1875}
1241 1876
1242eio_req *eio_symlink (const char *path, const char *new_path, eio_cb cb) 1877eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1243{ 1878{
1244 return eio__2path (EIO_SYMLINK, path, new_path, cb); 1879 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
1245} 1880}
1246 1881
1247eio_req *eio_rename (const char *path, const char *new_path, eio_cb cb) 1882eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1248{ 1883{
1249 return eio__2path (EIO_RENAME, path, new_path, cb); 1884 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1885}
1886
1887eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1888{
1889 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1890}
1891
1892#endif
1893
1894eio_req *eio_grp (eio_cb cb, void *data)
1895{
1896 const int pri = EIO_PRI_MAX;
1897
1898 REQ (EIO_GROUP); SEND;
1250} 1899}
1251 1900
1252#undef REQ 1901#undef REQ
1253#undef PATH 1902#undef PATH
1254#undef SEND 1903#undef SEND
1273 1922
1274void eio_grp_add (eio_req *grp, eio_req *req) 1923void eio_grp_add (eio_req *grp, eio_req *req)
1275{ 1924{
1276 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1925 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1277 1926
1927 grp->flags |= EIO_FLAG_GROUPADD;
1928
1278 ++grp->size; 1929 ++grp->size;
1279 req->grp = grp; 1930 req->grp = grp;
1280 1931
1281 req->grp_prev = 0; 1932 req->grp_prev = 0;
1282 req->grp_next = grp->grp_first; 1933 req->grp_next = grp->grp_first;
1290/*****************************************************************************/ 1941/*****************************************************************************/
1291/* misc garbage */ 1942/* misc garbage */
1292 1943
1293ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 1944ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1294{ 1945{
1295 worker wrk; 1946 etp_worker wrk;
1947 ssize_t ret;
1296 1948
1297 wrk.dbuf = 0; 1949 wrk.dbuf = 0;
1298 1950
1299 eio__sendfile (ofd, ifd, offset, count, &wrk); 1951 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1300 1952
1301 if (wrk.dbuf) 1953 if (wrk.dbuf)
1302 free (wrk.dbuf); 1954 free (wrk.dbuf);
1303}
1304 1955
1956 return ret;
1957}
1958

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines