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 1.13 by root, Tue May 13 19:34:11 2008 UTC vs.
Revision 1.32 by root, Sat Jun 6 18:06:55 2009 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009 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>
25#endif 68#endif
26 69
27#ifdef _WIN32 70#ifdef _WIN32
28 71
29 /*doh*/ 72 /*doh*/
30
31#else 73#else
32 74
33# include "config.h" 75# include "config.h"
34# include <sys/time.h> 76# include <sys/time.h>
35# include <sys/select.h> 77# include <sys/select.h>
36# include <unistd.h> 78# include <unistd.h>
37# include <utime.h> 79# include <utime.h>
38# include <signal.h> 80# include <signal.h>
39# include <dirent.h> 81# include <dirent.h>
82
83/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
85# define D_INO(de) (de)->d_fileno
86# define _DIRENT_HAVE_D_TYPE /* sigh */
87# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
88# define D_INO(de) (de)->d_ino
89# endif
90
91# ifdef _DIRENT_HAVE_D_TYPE
92# define D_TYPE(de) (de)->d_type
93# endif
40 94
41# ifndef EIO_STRUCT_DIRENT 95# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 96# define EIO_STRUCT_DIRENT struct dirent
43# endif 97# endif
44 98
57# else 111# else
58# error sendfile support requested but not available 112# error sendfile support requested but not available
59# endif 113# endif
60#endif 114#endif
61 115
116#ifndef D_TYPE
117# define D_TYPE(de) 0
118#endif
119
120#ifndef D_INO
121# define D_INO(de) 0
122#endif
123
62/* number of seconds after which an idle threads exit */ 124/* number of seconds after which an idle threads exit */
63#define IDLE_TIMEOUT 10 125#define IDLE_TIMEOUT 10
64 126
65/* used for struct dirent, AIX doesn't provide it */ 127/* used for struct dirent, AIX doesn't provide it */
66#ifndef NAME_MAX 128#ifndef NAME_MAX
70/* buffer size for various temporary buffers */ 132/* buffer size for various temporary buffers */
71#define EIO_BUFSIZE 65536 133#define EIO_BUFSIZE 65536
72 134
73#define dBUF \ 135#define dBUF \
74 char *eio_buf; \ 136 char *eio_buf; \
75 X_LOCK (etplock); \ 137 ETP_WORKER_LOCK (self); \
76 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 138 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77 X_UNLOCK (etplock); \ 139 ETP_WORKER_UNLOCK (self); \
78 errno = ENOMEM; \ 140 errno = ENOMEM; \
79 if (!eio_buf) \ 141 if (!eio_buf) \
80 return -1; 142 return -1;
81 143
82#define EIO_TICKS ((1000000 + 1023) >> 10) 144#define EIO_TICKS ((1000000 + 1023) >> 10)
83 145
84/*****************************************************************************/ 146/*****************************************************************************/
85 147
148#if __GNUC__ >= 3
149# define expect(expr,value) __builtin_expect ((expr),(value))
150#else
151# define expect(expr,value) (expr)
152#endif
153
154#define expect_false(expr) expect ((expr) != 0, 0)
155#define expect_true(expr) expect ((expr) != 0, 1)
156
157/*****************************************************************************/
158
159#define ETP_PRI_MIN EIO_PRI_MIN
160#define ETP_PRI_MAX EIO_PRI_MAX
161
162struct etp_worker;
163
164#define ETP_REQ eio_req
165#define ETP_DESTROY(req) eio_destroy (req)
166static int eio_finish (eio_req *req);
167#define ETP_FINISH(req) eio_finish (req)
168static void eio_execute (struct etp_worker *self, eio_req *req);
169#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
170
171#define ETP_WORKER_CLEAR(req) \
172 if (wrk->dbuf) \
173 { \
174 free (wrk->dbuf); \
175 wrk->dbuf = 0; \
176 } \
177 \
178 if (wrk->dirp) \
179 { \
180 closedir (wrk->dirp); \
181 wrk->dirp = 0; \
182 }
183
184#define ETP_WORKER_COMMON \
185 void *dbuf; \
186 DIR *dirp;
187
188/*****************************************************************************/
189
190#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
191
86/* calculcate time difference in ~1/EIO_TICKS of a second */ 192/* calculcate time difference in ~1/EIO_TICKS of a second */
87static int tvdiff (struct timeval *tv1, struct timeval *tv2) 193static int tvdiff (struct timeval *tv1, struct timeval *tv2)
88{ 194{
89 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 195 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
90 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 196 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
91} 197}
92 198
93static unsigned int started, idle, wanted = 4; 199static unsigned int started, idle, wanted = 4;
94 200
95typedef struct etp_pool
96{
97 void (*want_poll_cb) (void); 201static void (*want_poll_cb) (void);
98 void (*done_poll_cb) (void); 202static void (*done_poll_cb) (void);
99 203
100 unsigned int max_poll_time; 204static unsigned int max_poll_time; /* reslock */
101 unsigned int max_poll_reqs; 205static unsigned int max_poll_reqs; /* reslock */
102} etp_pool;
103 206
104static volatile unsigned int nreqs, nready, npending; 207static volatile unsigned int nreqs; /* reqlock */
208static volatile unsigned int nready; /* reqlock */
209static volatile unsigned int npending; /* reqlock */
105static volatile unsigned int max_idle = 4; 210static volatile unsigned int max_idle = 4;
106 211
107static mutex_t etplock = X_MUTEX_INIT; 212static mutex_t wrklock = X_MUTEX_INIT;
108static mutex_t reslock = X_MUTEX_INIT; 213static mutex_t reslock = X_MUTEX_INIT;
109static mutex_t reqlock = X_MUTEX_INIT; 214static mutex_t reqlock = X_MUTEX_INIT;
110static cond_t reqwait = X_COND_INIT; 215static cond_t reqwait = X_COND_INIT;
111 216
217#if !HAVE_PREADWRITE
218/*
219 * make our pread/pwrite emulation safe against themselves, but not against
220 * normal read/write by using a mutex. slows down execution a lot,
221 * but that's your problem, not mine.
222 */
223static mutex_t preadwritelock = X_MUTEX_INIT;
224#endif
225
112typedef struct worker 226typedef struct etp_worker
113{ 227{
114 /* locked by etplock */ 228 /* locked by wrklock */
115 struct worker *prev, *next; 229 struct etp_worker *prev, *next;
116 230
117 thread_t tid; 231 thread_t tid;
118 232
119 /* locked by reslock, reqlock or etplock */ 233 /* locked by reslock, reqlock or wrklock */
120 eio_req *req; /* currently processed request */ 234 ETP_REQ *req; /* currently processed request */
121 void *dbuf; 235
122 DIR *dirp; 236 ETP_WORKER_COMMON
123} worker; 237} etp_worker;
124 238
125static worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 239static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
240
241#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
242#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
126 243
127/* worker threads management */ 244/* worker threads management */
128 245
129static void worker_clear (worker *wrk) 246static void etp_worker_clear (etp_worker *wrk)
130{ 247{
131 if (wrk->dirp) 248 ETP_WORKER_CLEAR (wrk);
132 {
133 closedir (wrk->dirp);
134 wrk->dirp = 0;
135 }
136
137 if (wrk->dbuf)
138 {
139 free (wrk->dbuf);
140 wrk->dbuf = 0;
141 }
142} 249}
143 250
144static void worker_free (worker *wrk) 251static void etp_worker_free (etp_worker *wrk)
145{ 252{
146 wrk->next->prev = wrk->prev; 253 wrk->next->prev = wrk->prev;
147 wrk->prev->next = wrk->next; 254 wrk->prev->next = wrk->next;
148 255
149 free (wrk); 256 free (wrk);
150} 257}
151 258
152unsigned int eio_nreqs (void) 259static unsigned int etp_nreqs (void)
153{ 260{
261 int retval;
262 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
263 retval = nreqs;
264 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 265 return retval;
155} 266}
156 267
157unsigned int eio_nready (void) 268static unsigned int etp_nready (void)
158{ 269{
159 unsigned int retval; 270 unsigned int retval;
160 271
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 272 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 273 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 274 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 275
165 return retval; 276 return retval;
166} 277}
167 278
168unsigned int eio_npending (void) 279static unsigned int etp_npending (void)
169{ 280{
170 unsigned int retval; 281 unsigned int retval;
171 282
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 283 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 284 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 285 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 286
176 return retval; 287 return retval;
177} 288}
178 289
179unsigned int eio_nthreads (void) 290static unsigned int etp_nthreads (void)
180{ 291{
181 unsigned int retval; 292 unsigned int retval;
182 293
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 294 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 295 retval = started;
191 * a somewhat faster data structure might be nice, but 302 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 303 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 304 * per shift, the most expensive operation.
194 */ 305 */
195typedef struct { 306typedef struct {
196 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 307 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197 int size; 308 int size;
198} reqq; 309} etp_reqq;
199 310
200static reqq req_queue; 311static etp_reqq req_queue;
201static reqq res_queue; 312static etp_reqq res_queue;
202 313
203static int reqq_push (reqq *q, eio_req *req) 314static int reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 315{
205 int pri = req->pri; 316 int pri = req->pri;
206 req->next = 0; 317 req->next = 0;
207 318
208 if (q->qe[pri]) 319 if (q->qe[pri])
214 q->qe[pri] = q->qs[pri] = req; 325 q->qe[pri] = q->qs[pri] = req;
215 326
216 return q->size++; 327 return q->size++;
217} 328}
218 329
219static eio_req *reqq_shift (reqq *q) 330static ETP_REQ *reqq_shift (etp_reqq *q)
220{ 331{
221 int pri; 332 int pri;
222 333
223 if (!q->size) 334 if (!q->size)
224 return 0; 335 return 0;
225 336
226 --q->size; 337 --q->size;
227 338
228 for (pri = EIO_NUM_PRI; pri--; ) 339 for (pri = ETP_NUM_PRI; pri--; )
229 { 340 {
230 eio_req *req = q->qs[pri]; 341 eio_req *req = q->qs[pri];
231 342
232 if (req) 343 if (req)
233 { 344 {
241 abort (); 352 abort ();
242} 353}
243 354
244static void etp_atfork_prepare (void) 355static void etp_atfork_prepare (void)
245{ 356{
246 X_LOCK (etplock); 357 X_LOCK (wrklock);
247 X_LOCK (reqlock); 358 X_LOCK (reqlock);
248 X_LOCK (reslock); 359 X_LOCK (reslock);
249#if !HAVE_PREADWRITE 360#if !HAVE_PREADWRITE
250 X_LOCK (preadwritelock); 361 X_LOCK (preadwritelock);
251#endif 362#endif
252#if !HAVE_READDIR_R
253 X_LOCK (readdirlock);
254#endif
255} 363}
256 364
257static void etp_atfork_parent (void) 365static void etp_atfork_parent (void)
258{ 366{
259#if !HAVE_READDIR_R
260 X_UNLOCK (readdirlock);
261#endif
262#if !HAVE_PREADWRITE 367#if !HAVE_PREADWRITE
263 X_UNLOCK (preadwritelock); 368 X_UNLOCK (preadwritelock);
264#endif 369#endif
265 X_UNLOCK (reslock); 370 X_UNLOCK (reslock);
266 X_UNLOCK (reqlock); 371 X_UNLOCK (reqlock);
267 X_UNLOCK (etplock); 372 X_UNLOCK (wrklock);
268} 373}
269 374
270static void etp_atfork_child (void) 375static void etp_atfork_child (void)
271{ 376{
272 eio_req *prv; 377 ETP_REQ *prv;
273 378
274 while (prv = reqq_shift (&req_queue)) 379 while ((prv = reqq_shift (&req_queue)))
275 eio_destroy (prv); 380 ETP_DESTROY (prv);
276 381
277 while (prv = reqq_shift (&res_queue)) 382 while ((prv = reqq_shift (&res_queue)))
278 eio_destroy (prv); 383 ETP_DESTROY (prv);
279 384
280 while (wrk_first.next != &wrk_first) 385 while (wrk_first.next != &wrk_first)
281 { 386 {
282 worker *wrk = wrk_first.next; 387 etp_worker *wrk = wrk_first.next;
283 388
284 if (wrk->req) 389 if (wrk->req)
285 eio_destroy (wrk->req); 390 ETP_DESTROY (wrk->req);
286 391
287 worker_clear (wrk); 392 etp_worker_clear (wrk);
288 worker_free (wrk); 393 etp_worker_free (wrk);
289 } 394 }
290 395
291 started = 0; 396 started = 0;
292 idle = 0; 397 idle = 0;
293 nreqs = 0; 398 nreqs = 0;
302{ 407{
303 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 408 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
304} 409}
305 410
306static int 411static int
307etp_init (etp_pool *etp, void (*want_poll)(void), void (*done_poll)(void)) 412etp_init (void (*want_poll)(void), void (*done_poll)(void))
308{ 413{
309 static pthread_once_t doinit = PTHREAD_ONCE_INIT; 414 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
310 415
311 pthread_once (&doinit, etp_once_init); 416 pthread_once (&doinit, etp_once_init);
312 417
313 memset (etp, 0, sizeof *etp);
314
315 etp->want_poll_cb = want_poll; 418 want_poll_cb = want_poll;
316 etp->done_poll_cb = done_poll; 419 done_poll_cb = done_poll;
317}
318 420
319static etp_pool etp;
320
321/*****************************************************************************/
322
323static void grp_try_feed (eio_req *grp)
324{
325 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
326 {
327 int old_len = grp->size;
328
329 EIO_FEED (grp);
330
331 /* stop if no progress has been made */
332 if (old_len == grp->size)
333 {
334 grp->feed = 0;
335 break;
336 }
337 }
338}
339
340static int eio_finish (eio_req *req);
341
342static int grp_dec (eio_req *grp)
343{
344 --grp->size;
345
346 /* call feeder, if applicable */
347 grp_try_feed (grp);
348
349 /* finish, if done */
350 if (!grp->size && grp->int1)
351 return eio_finish (grp);
352 else
353 return 0; 421 return 0;
354} 422}
355 423
356void eio_destroy (eio_req *req)
357{
358 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
359 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
360
361 EIO_DESTROY (req);
362}
363
364static int eio_finish (eio_req *req)
365{
366 int res = EIO_FINISH (req);
367
368 if (req->grp)
369 {
370 int res2;
371 eio_req *grp = req->grp;
372
373 /* unlink request */
374 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
375 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
376
377 if (grp->grp_first == req)
378 grp->grp_first = req->grp_next;
379
380 res2 = grp_dec (grp);
381
382 if (!res && res2)
383 res = res2;
384 }
385
386 eio_destroy (req);
387
388 return res;
389}
390
391void eio_grp_cancel (eio_req *grp)
392{
393 for (grp = grp->grp_first; grp; grp = grp->grp_next)
394 eio_cancel (grp);
395}
396
397void eio_cancel (eio_req *req)
398{
399 X_LOCK (etplock);
400 req->flags |= EIO_FLAG_CANCELLED;
401 X_UNLOCK (etplock);
402
403 eio_grp_cancel (req);
404}
405
406X_THREAD_PROC (eio_proc); 424X_THREAD_PROC (etp_proc);
407 425
408static void start_thread (void) 426static void etp_start_thread (void)
409{ 427{
410 worker *wrk = calloc (1, sizeof (worker)); 428 etp_worker *wrk = calloc (1, sizeof (etp_worker));
411 429
412 /*TODO*/ 430 /*TODO*/
413 assert (("unable to allocate worker thread data", wrk)); 431 assert (("unable to allocate worker thread data", wrk));
414 432
415 X_LOCK (etplock); 433 X_LOCK (wrklock);
416 434
417 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 435 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
418 { 436 {
419 wrk->prev = &wrk_first; 437 wrk->prev = &wrk_first;
420 wrk->next = wrk_first.next; 438 wrk->next = wrk_first.next;
421 wrk_first.next->prev = wrk; 439 wrk_first.next->prev = wrk;
422 wrk_first.next = wrk; 440 wrk_first.next = wrk;
423 ++started; 441 ++started;
424 } 442 }
425 else 443 else
426 free (wrk); 444 free (wrk);
427 445
428 X_UNLOCK (etplock); 446 X_UNLOCK (wrklock);
429} 447}
430 448
431static void maybe_start_thread (void) 449static void etp_maybe_start_thread (void)
432{ 450{
433 if (eio_nthreads () >= wanted) 451 if (expect_true (etp_nthreads () >= wanted))
434 return; 452 return;
435 453
436 /* todo: maybe use idle here, but might be less exact */ 454 /* todo: maybe use idle here, but might be less exact */
437 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 455 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
438 return; 456 return;
439 457
440 start_thread (); 458 etp_start_thread ();
441} 459}
442 460
443void eio_submit (eio_req *req)
444{
445 req->pri += EIO_PRI_BIAS;
446
447 if (req->pri < EIO_PRI_MIN + EIO_PRI_BIAS) req->pri = EIO_PRI_MIN + EIO_PRI_BIAS;
448 if (req->pri > EIO_PRI_MAX + EIO_PRI_BIAS) req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
449
450 ++nreqs;
451
452 X_LOCK (reqlock);
453 ++nready;
454 reqq_push (&req_queue, req);
455 X_COND_SIGNAL (reqwait);
456 X_UNLOCK (reqlock);
457
458 maybe_start_thread ();
459}
460
461static void end_thread (void) 461static void etp_end_thread (void)
462{ 462{
463 eio_req *req = calloc (1, sizeof (eio_req)); 463 eio_req *req = calloc (1, sizeof (eio_req));
464 464
465 req->type = EIO_QUIT; 465 req->type = -1;
466 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 466 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
467 467
468 X_LOCK (reqlock); 468 X_LOCK (reqlock);
469 reqq_push (&req_queue, req); 469 reqq_push (&req_queue, req);
470 X_COND_SIGNAL (reqwait); 470 X_COND_SIGNAL (reqwait);
471 X_UNLOCK (reqlock); 471 X_UNLOCK (reqlock);
472 472
473 X_LOCK (etplock); 473 X_LOCK (wrklock);
474 --started; 474 --started;
475 X_UNLOCK (etplock); 475 X_UNLOCK (wrklock);
476} 476}
477 477
478void eio_set_max_poll_time (double nseconds)
479{
480 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
481 etp.max_poll_time = nseconds;
482 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
483}
484
485void eio_set_max_poll_reqs (unsigned int maxreqs)
486{
487 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
488 etp.max_poll_reqs = maxreqs;
489 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
490}
491
492void eio_set_max_idle (unsigned int nthreads)
493{
494 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
495 max_idle = nthreads <= 0 ? 1 : nthreads;
496 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
497}
498
499void eio_set_min_parallel (unsigned int nthreads)
500{
501 if (wanted < nthreads)
502 wanted = nthreads;
503}
504
505void eio_set_max_parallel (unsigned int nthreads)
506{
507 if (wanted > nthreads)
508 wanted = nthreads;
509
510 while (started > wanted)
511 end_thread ();
512}
513
514int eio_poll (void) 478static int etp_poll (void)
515{ 479{
516 int maxreqs = etp.max_poll_reqs; 480 unsigned int maxreqs;
481 unsigned int maxtime;
517 struct timeval tv_start, tv_now; 482 struct timeval tv_start, tv_now;
518 eio_req *req;
519 483
520 if (etp.max_poll_time) 484 X_LOCK (reslock);
485 maxreqs = max_poll_reqs;
486 maxtime = max_poll_time;
487 X_UNLOCK (reslock);
488
489 if (maxtime)
521 gettimeofday (&tv_start, 0); 490 gettimeofday (&tv_start, 0);
522 491
523 for (;;) 492 for (;;)
524 { 493 {
494 ETP_REQ *req;
495
525 maybe_start_thread (); 496 etp_maybe_start_thread ();
526 497
527 X_LOCK (reslock); 498 X_LOCK (reslock);
528 req = reqq_shift (&res_queue); 499 req = reqq_shift (&res_queue);
529 500
530 if (req) 501 if (req)
531 { 502 {
532 --npending; 503 --npending;
533 504
534 if (!res_queue.size && etp.done_poll_cb) 505 if (!res_queue.size && done_poll_cb)
535 etp.done_poll_cb (); 506 done_poll_cb ();
536 } 507 }
537 508
538 X_UNLOCK (reslock); 509 X_UNLOCK (reslock);
539 510
540 if (!req) 511 if (!req)
541 return 0; 512 return 0;
542 513
514 X_LOCK (reqlock);
543 --nreqs; 515 --nreqs;
516 X_UNLOCK (reqlock);
544 517
545 if (req->type == EIO_GROUP && req->size) 518 if (expect_false (req->type == EIO_GROUP && req->size))
546 { 519 {
547 req->int1 = 1; /* mark request as delayed */ 520 req->int1 = 1; /* mark request as delayed */
548 continue; 521 continue;
549 } 522 }
550 else 523 else
551 { 524 {
552 int res = eio_finish (req); 525 int res = ETP_FINISH (req);
553 if (res) 526 if (expect_false (res))
554 return res; 527 return res;
555 } 528 }
556 529
557 if (maxreqs && !--maxreqs) 530 if (expect_false (maxreqs && !--maxreqs))
558 break; 531 break;
559 532
560 if (etp.max_poll_time) 533 if (maxtime)
561 { 534 {
562 gettimeofday (&tv_now, 0); 535 gettimeofday (&tv_now, 0);
563 536
564 if (tvdiff (&tv_start, &tv_now) >= etp.max_poll_time) 537 if (tvdiff (&tv_start, &tv_now) >= maxtime)
565 break; 538 break;
566 } 539 }
567 } 540 }
568 541
569 errno = EAGAIN; 542 errno = EAGAIN;
570 return -1; 543 return -1;
571} 544}
572 545
546static void etp_cancel (ETP_REQ *req)
547{
548 X_LOCK (wrklock);
549 req->flags |= EIO_FLAG_CANCELLED;
550 X_UNLOCK (wrklock);
551
552 eio_grp_cancel (req);
553}
554
555static void etp_submit (ETP_REQ *req)
556{
557 req->pri -= ETP_PRI_MIN;
558
559 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
560 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
561
562 if (expect_false (req->type == EIO_GROUP))
563 {
564 /* I hope this is worth it :/ */
565 X_LOCK (reqlock);
566 ++nreqs;
567 X_UNLOCK (reqlock);
568
569 X_LOCK (reslock);
570
571 ++npending;
572
573 if (!reqq_push (&res_queue, req) && want_poll_cb)
574 want_poll_cb ();
575
576 X_UNLOCK (reslock);
577 }
578 else
579 {
580 X_LOCK (reqlock);
581 ++nreqs;
582 ++nready;
583 reqq_push (&req_queue, req);
584 X_COND_SIGNAL (reqwait);
585 X_UNLOCK (reqlock);
586
587 etp_maybe_start_thread ();
588 }
589}
590
591static void etp_set_max_poll_time (double nseconds)
592{
593 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
594 max_poll_time = nseconds;
595 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
596}
597
598static void etp_set_max_poll_reqs (unsigned int maxreqs)
599{
600 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
601 max_poll_reqs = maxreqs;
602 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
603}
604
605static void etp_set_max_idle (unsigned int nthreads)
606{
607 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
608 max_idle = nthreads <= 0 ? 1 : nthreads;
609 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
610}
611
612static void etp_set_min_parallel (unsigned int nthreads)
613{
614 if (wanted < nthreads)
615 wanted = nthreads;
616}
617
618static void etp_set_max_parallel (unsigned int nthreads)
619{
620 if (wanted > nthreads)
621 wanted = nthreads;
622
623 while (started > wanted)
624 etp_end_thread ();
625}
626
627/*****************************************************************************/
628
629static void grp_try_feed (eio_req *grp)
630{
631 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
632 {
633 grp->flags &= ~EIO_FLAG_GROUPADD;
634
635 EIO_FEED (grp);
636
637 /* stop if no progress has been made */
638 if (!(grp->flags & EIO_FLAG_GROUPADD))
639 {
640 grp->feed = 0;
641 break;
642 }
643 }
644}
645
646static int grp_dec (eio_req *grp)
647{
648 --grp->size;
649
650 /* call feeder, if applicable */
651 grp_try_feed (grp);
652
653 /* finish, if done */
654 if (!grp->size && grp->int1)
655 return eio_finish (grp);
656 else
657 return 0;
658}
659
660void eio_destroy (eio_req *req)
661{
662 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
663 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
664
665 EIO_DESTROY (req);
666}
667
668static int eio_finish (eio_req *req)
669{
670 int res = EIO_FINISH (req);
671
672 if (req->grp)
673 {
674 int res2;
675 eio_req *grp = req->grp;
676
677 /* unlink request */
678 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
679 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
680
681 if (grp->grp_first == req)
682 grp->grp_first = req->grp_next;
683
684 res2 = grp_dec (grp);
685
686 if (!res && res2)
687 res = res2;
688 }
689
690 eio_destroy (req);
691
692 return res;
693}
694
695void eio_grp_cancel (eio_req *grp)
696{
697 for (grp = grp->grp_first; grp; grp = grp->grp_next)
698 eio_cancel (grp);
699}
700
701void eio_cancel (eio_req *req)
702{
703 etp_cancel (req);
704}
705
706void eio_submit (eio_req *req)
707{
708 etp_submit (req);
709}
710
711unsigned int eio_nreqs (void)
712{
713 return etp_nreqs ();
714}
715
716unsigned int eio_nready (void)
717{
718 return etp_nready ();
719}
720
721unsigned int eio_npending (void)
722{
723 return etp_npending ();
724}
725
726unsigned int eio_nthreads (void)
727{
728 return etp_nthreads ();
729}
730
731void eio_set_max_poll_time (double nseconds)
732{
733 etp_set_max_poll_time (nseconds);
734}
735
736void eio_set_max_poll_reqs (unsigned int maxreqs)
737{
738 etp_set_max_poll_reqs (maxreqs);
739}
740
741void eio_set_max_idle (unsigned int nthreads)
742{
743 etp_set_max_idle (nthreads);
744}
745
746void eio_set_min_parallel (unsigned int nthreads)
747{
748 etp_set_min_parallel (nthreads);
749}
750
751void eio_set_max_parallel (unsigned int nthreads)
752{
753 etp_set_max_parallel (nthreads);
754}
755
756int eio_poll (void)
757{
758 return etp_poll ();
759}
760
573/*****************************************************************************/ 761/*****************************************************************************/
574/* work around various missing functions */ 762/* work around various missing functions */
575 763
576#if !HAVE_PREADWRITE 764#if !HAVE_PREADWRITE
765# undef pread
766# undef pwrite
577# define pread eio__pread 767# define pread eio__pread
578# define pwrite eio__pwrite 768# define pwrite eio__pwrite
579
580/*
581 * make our pread/pwrite safe against themselves, but not against
582 * normal read/write by using a mutex. slows down execution a lot,
583 * but that's your problem, not mine.
584 */
585static mutex_t preadwritelock = X_MUTEX_INIT;
586 769
587static ssize_t 770static ssize_t
588eio__pread (int fd, void *buf, size_t count, off_t offset) 771eio__pread (int fd, void *buf, size_t count, off_t offset)
589{ 772{
590 ssize_t res; 773 ssize_t res;
608 791
609 X_LOCK (preadwritelock); 792 X_LOCK (preadwritelock);
610 ooffset = lseek (fd, 0, SEEK_CUR); 793 ooffset = lseek (fd, 0, SEEK_CUR);
611 lseek (fd, offset, SEEK_SET); 794 lseek (fd, offset, SEEK_SET);
612 res = write (fd, buf, count); 795 res = write (fd, buf, count);
613 lseek (fd, offset, SEEK_SET); 796 lseek (fd, ooffset, SEEK_SET);
614 X_UNLOCK (preadwritelock); 797 X_UNLOCK (preadwritelock);
615 798
616 return res; 799 return res;
617} 800}
618#endif 801#endif
619 802
620#ifndef HAVE_FUTIMES 803#ifndef HAVE_FUTIMES
621 804
805# undef utimes
806# undef futimes
622# define utimes(path,times) eio__utimes (path, times) 807# define utimes(path,times) eio__utimes (path, times)
623# define futimes(fd,times) eio__futimes (fd, times) 808# define futimes(fd,times) eio__futimes (fd, times)
624 809
625static int 810static int
626eio__utimes (const char *filename, const struct timeval times[2]) 811eio__utimes (const char *filename, const struct timeval times[2])
645} 830}
646 831
647#endif 832#endif
648 833
649#if !HAVE_FDATASYNC 834#if !HAVE_FDATASYNC
835# undef fdatasync
650# define fdatasync fsync 836# define fdatasync(fd) fsync (fd)
651#endif 837#endif
838
839/* sync_file_range always needs emulation */
840int
841eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
842{
843#if HAVE_SYNC_FILE_RANGE
844 int res;
845
846 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
847 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
848 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
849 {
850 flags = 0
851 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
852 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
853 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
854 }
855
856 res = sync_file_range (fd, offset, nbytes, flags);
857
858 if (!res || errno != ENOSYS)
859 return res;
860#endif
861
862 /* even though we could play tricks with the flags, it's better to always
863 * call fdatasync, as thta matches the expectation of it's users best */
864 return fdatasync (fd);
865}
652 866
653#if !HAVE_READAHEAD 867#if !HAVE_READAHEAD
868# undef readahead
654# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 869# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
655 870
656static ssize_t 871static ssize_t
657eio__readahead (int fd, off_t offset, size_t count, worker *self) 872eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
658{ 873{
659 size_t todo = count; 874 size_t todo = count;
660 dBUF; 875 dBUF;
661 876
662 while (todo > 0) 877 while (todo > 0)
672 return count; 887 return count;
673} 888}
674 889
675#endif 890#endif
676 891
677#if !HAVE_READDIR_R
678# define readdir_r eio__readdir_r
679
680static mutex_t readdirlock = X_MUTEX_INIT;
681
682static int
683eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
684{
685 EIO_STRUCT_DIRENT *e;
686 int errorno;
687
688 X_LOCK (readdirlock);
689
690 e = readdir (dirp);
691 errorno = errno;
692
693 if (e)
694 {
695 *res = ent;
696 strcpy (ent->d_name, e->d_name);
697 }
698 else
699 *res = 0;
700
701 X_UNLOCK (readdirlock);
702
703 errno = errorno;
704 return e ? 0 : -1;
705}
706#endif
707
708/* sendfile always needs emulation */ 892/* sendfile always needs emulation */
709static ssize_t 893static ssize_t
710eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self) 894eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
711{ 895{
712 ssize_t res; 896 ssize_t res;
713 897
714 if (!count) 898 if (!count)
715 return 0; 899 return 0;
797 } 981 }
798 982
799 return res; 983 return res;
800} 984}
801 985
986static int
987eio_dent_cmp (const void *a_, const void *b_)
988{
989 const eio_dirent *a = (const eio_dirent *)a_;
990 const eio_dirent *b = (const eio_dirent *)b_;
991
992 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score
993 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */
994}
995
802/* read a full directory */ 996/* read a full directory */
803static void 997static void
804eio__scandir (eio_req *req, worker *self) 998eio__scandir (eio_req *req, etp_worker *self)
805{ 999{
806 DIR *dirp; 1000 DIR *dirp;
807 union
808 {
809 EIO_STRUCT_DIRENT d;
810 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
811 } *u;
812 EIO_STRUCT_DIRENT *entp; 1001 EIO_STRUCT_DIRENT *entp;
813 char *name, *names; 1002 unsigned char *name, *names;
814 int memlen = 4096; 1003 int namesalloc = 4096;
815 int memofs = 0; 1004 int namesoffs = 0;
1005 int flags = req->int1;
1006 eio_dirent *dents = 0;
1007 int dentalloc = 128;
816 int res = 0; 1008 int dentoffs = 0;
817 1009
1010 req->result = -1;
1011
1012 if (!(flags & EIO_READDIR_DENTS))
1013 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1014
818 X_LOCK (etplock); 1015 X_LOCK (wrklock);
1016 /* the corresponding closedir is in ETP_WORKER_CLEAR */
819 self->dirp = dirp = opendir (req->ptr1); 1017 self->dirp = dirp = opendir (req->ptr1);
820 self->dbuf = u = malloc (sizeof (*u));
821 req->flags |= EIO_FLAG_PTR2_FREE; 1018 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
822 req->ptr2 = names = malloc (memlen); 1019 req->ptr1 = names = malloc (namesalloc);
1020 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
823 X_UNLOCK (etplock); 1021 X_UNLOCK (wrklock);
824 1022
825 if (dirp && u && names) 1023 if (dirp && names && (!flags || dents))
826 for (;;) 1024 for (;;)
827 { 1025 {
828 errno = 0; 1026 errno = 0;
829 readdir_r (dirp, &u->d, &entp); 1027 entp = readdir (dirp);
830 1028
831 if (!entp) 1029 if (!entp)
1030 {
1031 if (errno)
1032 break;
1033
1034 /* sort etc. */
1035 req->int1 = flags;
1036 req->result = dentoffs;
1037
1038 if (dents)
1039 {
1040 eio_dirent *ent = dents + dentoffs;
1041
1042 while (ent > dents)
1043 (--ent)->name = names + (size_t)ent->name;
1044 }
1045
1046 if (flags & EIO_READDIR_STAT_ORDER
1047 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))
1048 {
1049 /* pray your qsort doesn't use quicksort */
1050 qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1051 }
1052 else if (flags & EIO_READDIR_DIRS_FIRST)
1053 {
1054 /* in this case, all is known, and we just put dirs first and sort them */
1055 eio_dirent *ent = dents + dentoffs;
1056 eio_dirent *dir = dents;
1057
1058 while (ent > dir)
1059 {
1060 if (dir->type == DT_DIR)
1061 ++dir;
1062 else
1063 {
1064 --ent;
1065
1066 if (ent->type == DT_DIR)
1067 {
1068 eio_dirent tmp = *dir;
1069 *dir = *ent;
1070 *ent = tmp;
1071
1072 ++dir;
1073 }
1074 }
1075 }
1076
1077 /* now sort the dirs only */
1078 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1079 }
1080
1081 /* only provide the names array unless DENTS is specified */
1082 if (!(flags & EIO_READDIR_DENTS))
1083 {
1084 X_LOCK (wrklock);
1085 assert (!dents);
1086 req->ptr1 = 0;
1087 req->ptr2 = names;
1088 X_UNLOCK (wrklock);
1089 }
1090
832 break; 1091 break;
1092 }
833 1093
1094 /* now add the entry to our list(s) */
834 name = entp->d_name; 1095 name = entp->d_name;
835 1096
1097 /* skip . and .. entries */
836 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1098 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
837 { 1099 {
838 int len = strlen (name) + 1; 1100 int len = strlen (name) + 1;
839 1101
840 res++; 1102 while (expect_false (namesoffs + len > namesalloc))
841
842 while (memofs + len > memlen)
843 { 1103 {
844 memlen *= 2; 1104 namesalloc *= 2;
845 X_LOCK (etplock); 1105 X_LOCK (wrklock);
846 req->ptr2 = names = realloc (names, memlen); 1106 req->ptr1 = names = realloc (names, namesalloc);
847 X_UNLOCK (etplock); 1107 X_UNLOCK (wrklock);
848 1108
849 if (!names) 1109 if (!names)
850 break; 1110 break;
851 } 1111 }
852 1112
853 memcpy (names + memofs, name, len); 1113 memcpy (names + namesoffs, name, len);
1114
1115 if (dents)
1116 {
1117 struct eio_dirent *ent;
1118
1119 if (expect_false (dentoffs == dentalloc))
1120 {
1121 dentalloc *= 2;
1122 X_LOCK (wrklock);
1123 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1124 X_UNLOCK (wrklock);
1125
1126 if (!dents)
1127 break;
1128 }
1129
1130 ent = dents + dentoffs;
1131
1132 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */
1133 ent->namelen = len - 1;
1134 ent->inode = D_INO (entp);
1135
1136 switch (D_TYPE (entp))
1137 {
1138 default:
1139 ent->type = EIO_DT_UNKNOWN;
1140 flags |= EIO_READDIR_FOUND_UNKNOWN;
1141 break;
1142
1143 #ifdef DT_FIFO
1144 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1145 #endif
1146 #ifdef DT_CHR
1147 case DT_CHR: ent->type = EIO_DT_CHR; break;
1148 #endif
1149 #ifdef DT_DIR
1150 case DT_DIR: ent->type = EIO_DT_DIR; break;
1151 #endif
1152 #ifdef DT_BLK
1153 case DT_BLK: ent->type = EIO_DT_BLK; break;
1154 #endif
1155 #ifdef DT_REG
1156 case DT_REG: ent->type = EIO_DT_REG; break;
1157 #endif
1158 #ifdef DT_LNK
1159 case DT_LNK: ent->type = EIO_DT_LNK; break;
1160 #endif
1161 #ifdef DT_SOCK
1162 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1163 #endif
1164 #ifdef DT_WHT
1165 case DT_WHT: ent->type = EIO_DT_WHT; break;
1166 #endif
1167 }
1168
1169 ent->score = 0;
1170
1171 if (flags & EIO_READDIR_DIRS_FIRST)
1172 {
1173 if (ent->type == EIO_DT_UNKNOWN)
1174 {
1175 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1176 ent->score = 98;
1177 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1178 ent->score = len <= 4 ? 5 : len <= 7 ? 4 : 1; /* shorter == more likely dir, but avoid too many classes */
1179 }
1180 else if (ent->type == DT_DIR)
1181 ent->score = 100;
1182 }
1183 }
1184
854 memofs += len; 1185 namesoffs += len;
1186 ++dentoffs;
855 } 1187 }
856 } 1188 }
857 1189 else
858 if (errno)
859 res = -1;
860
861 req->result = res; 1190 req->result = -1;
1191}
1192
1193#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1194# undef msync
1195# define msync(a,b,c) ((errno = ENOSYS), -1)
1196#endif
1197
1198int
1199eio__mtouch (void *mem, size_t len, int flags)
1200{
1201 intptr_t addr = (intptr_t)mem;
1202 intptr_t end = addr + len;
1203#ifdef PAGESIZE
1204 const intptr_t page = PAGESIZE;
1205#else
1206 static intptr_t page;
1207
1208 if (!page)
1209 page = sysconf (_SC_PAGESIZE);
1210#endif
1211
1212 addr &= ~(page - 1); /* assume page size is always a power of two */
1213
1214 if (addr < end)
1215 if (flags) /* modify */
1216 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1217 else
1218 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1219
1220 return 0;
862} 1221}
863 1222
864/*****************************************************************************/ 1223/*****************************************************************************/
865 1224
866#define ALLOC(len) \ 1225#define ALLOC(len) \
867 if (!req->ptr2) \ 1226 if (!req->ptr2) \
868 { \ 1227 { \
869 X_LOCK (etplock); \ 1228 X_LOCK (wrklock); \
870 req->flags |= EIO_FLAG_PTR2_FREE; \ 1229 req->flags |= EIO_FLAG_PTR2_FREE; \
871 X_UNLOCK (etplock); \ 1230 X_UNLOCK (wrklock); \
872 req->ptr2 = malloc (len); \ 1231 req->ptr2 = malloc (len); \
873 if (!req->ptr2) \ 1232 if (!req->ptr2) \
874 { \ 1233 { \
875 errno = ENOMEM; \ 1234 errno = ENOMEM; \
876 req->result = -1; \ 1235 req->result = -1; \
877 break; \ 1236 break; \
878 } \ 1237 } \
879 } 1238 }
880 1239
881X_THREAD_PROC (eio_proc) 1240X_THREAD_PROC (etp_proc)
882{ 1241{
883 eio_req *req; 1242 ETP_REQ *req;
884 struct timespec ts; 1243 struct timespec ts;
885 worker *self = (worker *)thr_arg; 1244 etp_worker *self = (etp_worker *)thr_arg;
886 1245
887 /* try to distribute timeouts somewhat randomly */ 1246 /* try to distribute timeouts somewhat randomly */
888 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1247 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
889 1248
890 for (;;) 1249 for (;;)
891 { 1250 {
892 ts.tv_sec = time (0) + IDLE_TIMEOUT;
893
894 X_LOCK (reqlock); 1251 X_LOCK (reqlock);
895 1252
896 for (;;) 1253 for (;;)
897 { 1254 {
898 self->req = req = reqq_shift (&req_queue); 1255 self->req = req = reqq_shift (&req_queue);
900 if (req) 1257 if (req)
901 break; 1258 break;
902 1259
903 ++idle; 1260 ++idle;
904 1261
1262 ts.tv_sec = time (0) + IDLE_TIMEOUT;
905 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1263 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
906 { 1264 {
907 if (idle > max_idle) 1265 if (idle > max_idle)
908 { 1266 {
909 --idle; 1267 --idle;
910 X_UNLOCK (reqlock); 1268 X_UNLOCK (reqlock);
911 X_LOCK (etplock); 1269 X_LOCK (wrklock);
912 --started; 1270 --started;
913 X_UNLOCK (etplock); 1271 X_UNLOCK (wrklock);
914 goto quit; 1272 goto quit;
915 } 1273 }
916 1274
917 /* we are allowed to idle, so do so without any timeout */ 1275 /* we are allowed to idle, so do so without any timeout */
918 X_COND_WAIT (reqwait, reqlock); 1276 X_COND_WAIT (reqwait, reqlock);
919 ts.tv_sec = time (0) + IDLE_TIMEOUT;
920 } 1277 }
921 1278
922 --idle; 1279 --idle;
923 } 1280 }
924 1281
925 --nready; 1282 --nready;
926 1283
927 X_UNLOCK (reqlock); 1284 X_UNLOCK (reqlock);
928 1285
929 errno = 0; /* strictly unnecessary */ 1286 if (req->type < 0)
1287 goto quit;
930 1288
931 if (!EIO_CANCELLED (req)) 1289 if (!EIO_CANCELLED (req))
932 switch (req->type) 1290 ETP_EXECUTE (self, req);
933 {
934 case EIO_READ: ALLOC (req->size);
935 req->result = req->offs >= 0
936 ? pread (req->int1, req->ptr2, req->size, req->offs)
937 : read (req->int1, req->ptr2, req->size); break;
938 case EIO_WRITE: req->result = req->offs >= 0
939 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
940 : write (req->int1, req->ptr2, req->size); break;
941
942 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
943 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
944
945 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
946 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
947 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
948 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
949 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
950 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
951
952 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
953 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
954 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
955 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
956 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
957 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
958
959 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
960 case EIO_CLOSE: req->result = close (req->int1); break;
961 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
962 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
963 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
964 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
965 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
966 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
967 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
968 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
969
970 case EIO_READLINK: ALLOC (NAME_MAX);
971 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
972
973 case EIO_SYNC: req->result = 0; sync (); break;
974 case EIO_FSYNC: req->result = fsync (req->int1); break;
975 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
976
977 case EIO_READDIR: eio__scandir (req, self); break;
978
979 case EIO_BUSY:
980#ifdef _WIN32
981 Sleep (req->nv1 * 1000.);
982#else
983 {
984 struct timeval tv;
985
986 tv.tv_sec = req->nv1;
987 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
988
989 req->result = select (0, 0, 0, 0, &tv);
990 }
991#endif
992 break;
993
994 case EIO_UTIME:
995 case EIO_FUTIME:
996 {
997 struct timeval tv[2];
998 struct timeval *times;
999
1000 if (req->nv1 != -1. || req->nv2 != -1.)
1001 {
1002 tv[0].tv_sec = req->nv1;
1003 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1004 tv[1].tv_sec = req->nv2;
1005 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1006
1007 times = tv;
1008 }
1009 else
1010 times = 0;
1011
1012
1013 req->result = req->type == EIO_FUTIME
1014 ? futimes (req->int1, times)
1015 : utimes (req->ptr1, times);
1016 }
1017
1018 case EIO_GROUP:
1019 case EIO_NOP:
1020 req->result = 0;
1021 break;
1022
1023 case EIO_QUIT:
1024 goto quit;
1025
1026 default:
1027 req->result = -1;
1028 break;
1029 }
1030
1031 req->errorno = errno;
1032 1291
1033 X_LOCK (reslock); 1292 X_LOCK (reslock);
1034 1293
1035 ++npending; 1294 ++npending;
1036 1295
1037 if (!reqq_push (&res_queue, req) && etp.want_poll_cb) 1296 if (!reqq_push (&res_queue, req) && want_poll_cb)
1038 etp.want_poll_cb (); 1297 want_poll_cb ();
1039 1298
1040 self->req = 0; 1299 self->req = 0;
1041 worker_clear (self); 1300 etp_worker_clear (self);
1042 1301
1043 X_UNLOCK (reslock); 1302 X_UNLOCK (reslock);
1044 } 1303 }
1045 1304
1046quit: 1305quit:
1047 X_LOCK (etplock); 1306 X_LOCK (wrklock);
1048 worker_free (self); 1307 etp_worker_free (self);
1049 X_UNLOCK (etplock); 1308 X_UNLOCK (wrklock);
1050 1309
1051 return 0; 1310 return 0;
1052} 1311}
1053 1312
1054/*****************************************************************************/ 1313/*****************************************************************************/
1055 1314
1056int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1315int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1057{ 1316{
1058 etp_init (&etp, want_poll, done_poll); 1317 return etp_init (want_poll, done_poll);
1059} 1318}
1060 1319
1061static void eio_api_destroy (eio_req *req) 1320static void eio_api_destroy (eio_req *req)
1062{ 1321{
1063 free (req); 1322 free (req);
1085 { \ 1344 { \
1086 eio_api_destroy (req); \ 1345 eio_api_destroy (req); \
1087 return 0; \ 1346 return 0; \
1088 } 1347 }
1089 1348
1349static void eio_execute (etp_worker *self, eio_req *req)
1350{
1351 errno = 0;
1352
1353 switch (req->type)
1354 {
1355 case EIO_READ: ALLOC (req->size);
1356 req->result = req->offs >= 0
1357 ? pread (req->int1, req->ptr2, req->size, req->offs)
1358 : read (req->int1, req->ptr2, req->size); break;
1359 case EIO_WRITE: req->result = req->offs >= 0
1360 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1361 : write (req->int1, req->ptr2, req->size); break;
1362
1363 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1364 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1365
1366 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1367 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1368 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1369 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1370 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1371 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1372
1373 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1374 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1375 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1376 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1377 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1378 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1379
1380 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1381 case EIO_CLOSE: req->result = close (req->int1); break;
1382 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1383 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1384 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1385 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1386 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1387 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1388 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1389 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1390
1391 case EIO_READLINK: ALLOC (NAME_MAX);
1392 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1393
1394 case EIO_SYNC: req->result = 0; sync (); break;
1395 case EIO_FSYNC: req->result = fsync (req->int1); break;
1396 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1397 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1398 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1399 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1400
1401 case EIO_READDIR: eio__scandir (req, self); break;
1402
1403 case EIO_BUSY:
1404#ifdef _WIN32
1405 Sleep (req->nv1 * 1000.);
1406#else
1407 {
1408 struct timeval tv;
1409
1410 tv.tv_sec = req->nv1;
1411 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
1412
1413 req->result = select (0, 0, 0, 0, &tv);
1414 }
1415#endif
1416 break;
1417
1418 case EIO_UTIME:
1419 case EIO_FUTIME:
1420 {
1421 struct timeval tv[2];
1422 struct timeval *times;
1423
1424 if (req->nv1 != -1. || req->nv2 != -1.)
1425 {
1426 tv[0].tv_sec = req->nv1;
1427 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1428 tv[1].tv_sec = req->nv2;
1429 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1430
1431 times = tv;
1432 }
1433 else
1434 times = 0;
1435
1436
1437 req->result = req->type == EIO_FUTIME
1438 ? futimes (req->int1, times)
1439 : utimes (req->ptr1, times);
1440 }
1441 break;
1442
1443 case EIO_GROUP:
1444 abort (); /* handled in eio_request */
1445
1446 case EIO_NOP:
1447 req->result = 0;
1448 break;
1449
1450 case EIO_CUSTOM:
1451 ((void (*)(eio_req *))req->feed) (req);
1452 break;
1453
1454 default:
1455 req->result = -1;
1456 break;
1457 }
1458
1459 req->errorno = errno;
1460}
1461
1090#ifndef EIO_NO_WRAPPERS 1462#ifndef EIO_NO_WRAPPERS
1091 1463
1092eio_req *eio_nop (int pri, eio_cb cb, void *data) 1464eio_req *eio_nop (int pri, eio_cb cb, void *data)
1093{ 1465{
1094 REQ (EIO_NOP); SEND; 1466 REQ (EIO_NOP); SEND;
1105} 1477}
1106 1478
1107eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 1479eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1108{ 1480{
1109 REQ (EIO_FSYNC); req->int1 = fd; SEND; 1481 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1482}
1483
1484eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1485{
1486 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1487}
1488
1489eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1490{
1491 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1492}
1493
1494eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1495{
1496 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1110} 1497}
1111 1498
1112eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 1499eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1113{ 1500{
1114 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1501 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1228eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1615eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1229{ 1616{
1230 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1617 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1231} 1618}
1232 1619
1233eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1620eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1234{ 1621{
1235 return eio__1path (EIO_READDIR, path, pri, cb, data); 1622 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1236} 1623}
1237 1624
1238eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1625eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1239{ 1626{
1240 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1627 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1241} 1628}
1242 1629
1243static eio_req * 1630static eio_req *
1244eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1631eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1245{ 1632{
1269eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1656eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1270{ 1657{
1271 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 1658 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1272} 1659}
1273 1660
1661eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1662{
1663 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1664}
1665
1274#endif 1666#endif
1275 1667
1276eio_req *eio_grp (eio_cb cb, void *data) 1668eio_req *eio_grp (eio_cb cb, void *data)
1277{ 1669{
1278 const int pri = EIO_PRI_MAX; 1670 const int pri = EIO_PRI_MAX;
1304 1696
1305void eio_grp_add (eio_req *grp, eio_req *req) 1697void eio_grp_add (eio_req *grp, eio_req *req)
1306{ 1698{
1307 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1699 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1308 1700
1701 grp->flags |= EIO_FLAG_GROUPADD;
1702
1309 ++grp->size; 1703 ++grp->size;
1310 req->grp = grp; 1704 req->grp = grp;
1311 1705
1312 req->grp_prev = 0; 1706 req->grp_prev = 0;
1313 req->grp_next = grp->grp_first; 1707 req->grp_next = grp->grp_first;
1321/*****************************************************************************/ 1715/*****************************************************************************/
1322/* misc garbage */ 1716/* misc garbage */
1323 1717
1324ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 1718ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1325{ 1719{
1326 worker wrk; 1720 etp_worker wrk;
1327 1721
1328 wrk.dbuf = 0; 1722 wrk.dbuf = 0;
1329 1723
1330 eio__sendfile (ofd, ifd, offset, count, &wrk); 1724 eio__sendfile (ofd, ifd, offset, count, &wrk);
1331 1725

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines