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

Comparing libeio/eio.c (file contents):
Revision 1.13 by root, Tue May 13 19:34:11 2008 UTC vs.
Revision 1.33 by root, Sat Jun 6 19:44:17 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 _DIRENT_HAVE_D_TYPE /* sigh */
86# define D_INO(de) (de)->d_fileno
87# define D_NAMLEN(de) (de)->d_namlen
88# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600
89# define D_INO(de) (de)->d_ino
90# endif
91
92#ifdef _D_EXACT_NAMLEN
93# undef D_NAMLEN
94# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
95#endif
96
97# ifdef _DIRENT_HAVE_D_TYPE
98# define D_TYPE(de) (de)->d_type
99# endif
40 100
41# ifndef EIO_STRUCT_DIRENT 101# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent 102# define EIO_STRUCT_DIRENT struct dirent
43# endif 103# endif
44 104
57# else 117# else
58# error sendfile support requested but not available 118# error sendfile support requested but not available
59# endif 119# endif
60#endif 120#endif
61 121
122#ifndef D_TYPE
123# define D_TYPE(de) 0
124#endif
125#ifndef D_INO
126# define D_INO(de) 0
127#endif
128#ifndef D_NAMLEN
129# define D_NAMLEN(de) strlen ((de)->d_name)
130#endif
131
62/* number of seconds after which an idle threads exit */ 132/* number of seconds after which an idle threads exit */
63#define IDLE_TIMEOUT 10 133#define IDLE_TIMEOUT 10
64 134
65/* used for struct dirent, AIX doesn't provide it */ 135/* used for struct dirent, AIX doesn't provide it */
66#ifndef NAME_MAX 136#ifndef NAME_MAX
70/* buffer size for various temporary buffers */ 140/* buffer size for various temporary buffers */
71#define EIO_BUFSIZE 65536 141#define EIO_BUFSIZE 65536
72 142
73#define dBUF \ 143#define dBUF \
74 char *eio_buf; \ 144 char *eio_buf; \
75 X_LOCK (etplock); \ 145 ETP_WORKER_LOCK (self); \
76 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 146 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
77 X_UNLOCK (etplock); \ 147 ETP_WORKER_UNLOCK (self); \
78 errno = ENOMEM; \ 148 errno = ENOMEM; \
79 if (!eio_buf) \ 149 if (!eio_buf) \
80 return -1; 150 return -1;
81 151
82#define EIO_TICKS ((1000000 + 1023) >> 10) 152#define EIO_TICKS ((1000000 + 1023) >> 10)
83 153
84/*****************************************************************************/ 154/*****************************************************************************/
85 155
156#if __GNUC__ >= 3
157# define expect(expr,value) __builtin_expect ((expr),(value))
158#else
159# define expect(expr,value) (expr)
160#endif
161
162#define expect_false(expr) expect ((expr) != 0, 0)
163#define expect_true(expr) expect ((expr) != 0, 1)
164
165/*****************************************************************************/
166
167#define ETP_PRI_MIN EIO_PRI_MIN
168#define ETP_PRI_MAX EIO_PRI_MAX
169
170struct etp_worker;
171
172#define ETP_REQ eio_req
173#define ETP_DESTROY(req) eio_destroy (req)
174static int eio_finish (eio_req *req);
175#define ETP_FINISH(req) eio_finish (req)
176static void eio_execute (struct etp_worker *self, eio_req *req);
177#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
178
179#define ETP_WORKER_CLEAR(req) \
180 if (wrk->dbuf) \
181 { \
182 free (wrk->dbuf); \
183 wrk->dbuf = 0; \
184 } \
185 \
186 if (wrk->dirp) \
187 { \
188 closedir (wrk->dirp); \
189 wrk->dirp = 0; \
190 }
191
192#define ETP_WORKER_COMMON \
193 void *dbuf; \
194 DIR *dirp;
195
196/*****************************************************************************/
197
198#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
199
86/* calculcate time difference in ~1/EIO_TICKS of a second */ 200/* calculcate time difference in ~1/EIO_TICKS of a second */
87static int tvdiff (struct timeval *tv1, struct timeval *tv2) 201static int tvdiff (struct timeval *tv1, struct timeval *tv2)
88{ 202{
89 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 203 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
90 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 204 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
91} 205}
92 206
93static unsigned int started, idle, wanted = 4; 207static unsigned int started, idle, wanted = 4;
94 208
95typedef struct etp_pool
96{
97 void (*want_poll_cb) (void); 209static void (*want_poll_cb) (void);
98 void (*done_poll_cb) (void); 210static void (*done_poll_cb) (void);
99 211
100 unsigned int max_poll_time; 212static unsigned int max_poll_time; /* reslock */
101 unsigned int max_poll_reqs; 213static unsigned int max_poll_reqs; /* reslock */
102} etp_pool;
103 214
104static volatile unsigned int nreqs, nready, npending; 215static volatile unsigned int nreqs; /* reqlock */
216static volatile unsigned int nready; /* reqlock */
217static volatile unsigned int npending; /* reqlock */
105static volatile unsigned int max_idle = 4; 218static volatile unsigned int max_idle = 4;
106 219
107static mutex_t etplock = X_MUTEX_INIT; 220static mutex_t wrklock = X_MUTEX_INIT;
108static mutex_t reslock = X_MUTEX_INIT; 221static mutex_t reslock = X_MUTEX_INIT;
109static mutex_t reqlock = X_MUTEX_INIT; 222static mutex_t reqlock = X_MUTEX_INIT;
110static cond_t reqwait = X_COND_INIT; 223static cond_t reqwait = X_COND_INIT;
111 224
225#if !HAVE_PREADWRITE
226/*
227 * make our pread/pwrite emulation safe against themselves, but not against
228 * normal read/write by using a mutex. slows down execution a lot,
229 * but that's your problem, not mine.
230 */
231static mutex_t preadwritelock = X_MUTEX_INIT;
232#endif
233
112typedef struct worker 234typedef struct etp_worker
113{ 235{
114 /* locked by etplock */ 236 /* locked by wrklock */
115 struct worker *prev, *next; 237 struct etp_worker *prev, *next;
116 238
117 thread_t tid; 239 thread_t tid;
118 240
119 /* locked by reslock, reqlock or etplock */ 241 /* locked by reslock, reqlock or wrklock */
120 eio_req *req; /* currently processed request */ 242 ETP_REQ *req; /* currently processed request */
121 void *dbuf; 243
122 DIR *dirp; 244 ETP_WORKER_COMMON
123} worker; 245} etp_worker;
124 246
125static worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 247static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
248
249#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
250#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
126 251
127/* worker threads management */ 252/* worker threads management */
128 253
129static void worker_clear (worker *wrk) 254static void etp_worker_clear (etp_worker *wrk)
130{ 255{
131 if (wrk->dirp) 256 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} 257}
143 258
144static void worker_free (worker *wrk) 259static void etp_worker_free (etp_worker *wrk)
145{ 260{
146 wrk->next->prev = wrk->prev; 261 wrk->next->prev = wrk->prev;
147 wrk->prev->next = wrk->next; 262 wrk->prev->next = wrk->next;
148 263
149 free (wrk); 264 free (wrk);
150} 265}
151 266
152unsigned int eio_nreqs (void) 267static unsigned int etp_nreqs (void)
153{ 268{
269 int retval;
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = nreqs;
272 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 273 return retval;
155} 274}
156 275
157unsigned int eio_nready (void) 276static unsigned int etp_nready (void)
158{ 277{
159 unsigned int retval; 278 unsigned int retval;
160 279
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 281 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 283
165 return retval; 284 return retval;
166} 285}
167 286
168unsigned int eio_npending (void) 287static unsigned int etp_npending (void)
169{ 288{
170 unsigned int retval; 289 unsigned int retval;
171 290
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 291 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 292 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 293 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 294
176 return retval; 295 return retval;
177} 296}
178 297
179unsigned int eio_nthreads (void) 298static unsigned int etp_nthreads (void)
180{ 299{
181 unsigned int retval; 300 unsigned int retval;
182 301
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 302 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 303 retval = started;
191 * a somewhat faster data structure might be nice, but 310 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 311 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 312 * per shift, the most expensive operation.
194 */ 313 */
195typedef struct { 314typedef struct {
196 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 315 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197 int size; 316 int size;
198} reqq; 317} etp_reqq;
199 318
200static reqq req_queue; 319static etp_reqq req_queue;
201static reqq res_queue; 320static etp_reqq res_queue;
202 321
203static int reqq_push (reqq *q, eio_req *req) 322static int reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 323{
205 int pri = req->pri; 324 int pri = req->pri;
206 req->next = 0; 325 req->next = 0;
207 326
208 if (q->qe[pri]) 327 if (q->qe[pri])
214 q->qe[pri] = q->qs[pri] = req; 333 q->qe[pri] = q->qs[pri] = req;
215 334
216 return q->size++; 335 return q->size++;
217} 336}
218 337
219static eio_req *reqq_shift (reqq *q) 338static ETP_REQ *reqq_shift (etp_reqq *q)
220{ 339{
221 int pri; 340 int pri;
222 341
223 if (!q->size) 342 if (!q->size)
224 return 0; 343 return 0;
225 344
226 --q->size; 345 --q->size;
227 346
228 for (pri = EIO_NUM_PRI; pri--; ) 347 for (pri = ETP_NUM_PRI; pri--; )
229 { 348 {
230 eio_req *req = q->qs[pri]; 349 eio_req *req = q->qs[pri];
231 350
232 if (req) 351 if (req)
233 { 352 {
241 abort (); 360 abort ();
242} 361}
243 362
244static void etp_atfork_prepare (void) 363static void etp_atfork_prepare (void)
245{ 364{
246 X_LOCK (etplock); 365 X_LOCK (wrklock);
247 X_LOCK (reqlock); 366 X_LOCK (reqlock);
248 X_LOCK (reslock); 367 X_LOCK (reslock);
249#if !HAVE_PREADWRITE 368#if !HAVE_PREADWRITE
250 X_LOCK (preadwritelock); 369 X_LOCK (preadwritelock);
251#endif 370#endif
252#if !HAVE_READDIR_R
253 X_LOCK (readdirlock);
254#endif
255} 371}
256 372
257static void etp_atfork_parent (void) 373static void etp_atfork_parent (void)
258{ 374{
259#if !HAVE_READDIR_R
260 X_UNLOCK (readdirlock);
261#endif
262#if !HAVE_PREADWRITE 375#if !HAVE_PREADWRITE
263 X_UNLOCK (preadwritelock); 376 X_UNLOCK (preadwritelock);
264#endif 377#endif
265 X_UNLOCK (reslock); 378 X_UNLOCK (reslock);
266 X_UNLOCK (reqlock); 379 X_UNLOCK (reqlock);
267 X_UNLOCK (etplock); 380 X_UNLOCK (wrklock);
268} 381}
269 382
270static void etp_atfork_child (void) 383static void etp_atfork_child (void)
271{ 384{
272 eio_req *prv; 385 ETP_REQ *prv;
273 386
274 while (prv = reqq_shift (&req_queue)) 387 while ((prv = reqq_shift (&req_queue)))
275 eio_destroy (prv); 388 ETP_DESTROY (prv);
276 389
277 while (prv = reqq_shift (&res_queue)) 390 while ((prv = reqq_shift (&res_queue)))
278 eio_destroy (prv); 391 ETP_DESTROY (prv);
279 392
280 while (wrk_first.next != &wrk_first) 393 while (wrk_first.next != &wrk_first)
281 { 394 {
282 worker *wrk = wrk_first.next; 395 etp_worker *wrk = wrk_first.next;
283 396
284 if (wrk->req) 397 if (wrk->req)
285 eio_destroy (wrk->req); 398 ETP_DESTROY (wrk->req);
286 399
287 worker_clear (wrk); 400 etp_worker_clear (wrk);
288 worker_free (wrk); 401 etp_worker_free (wrk);
289 } 402 }
290 403
291 started = 0; 404 started = 0;
292 idle = 0; 405 idle = 0;
293 nreqs = 0; 406 nreqs = 0;
302{ 415{
303 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 416 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
304} 417}
305 418
306static int 419static int
307etp_init (etp_pool *etp, void (*want_poll)(void), void (*done_poll)(void)) 420etp_init (void (*want_poll)(void), void (*done_poll)(void))
308{ 421{
309 static pthread_once_t doinit = PTHREAD_ONCE_INIT; 422 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
310 423
311 pthread_once (&doinit, etp_once_init); 424 pthread_once (&doinit, etp_once_init);
312 425
313 memset (etp, 0, sizeof *etp);
314
315 etp->want_poll_cb = want_poll; 426 want_poll_cb = want_poll;
316 etp->done_poll_cb = done_poll; 427 done_poll_cb = done_poll;
317}
318 428
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; 429 return 0;
354} 430}
355 431
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); 432X_THREAD_PROC (etp_proc);
407 433
408static void start_thread (void) 434static void etp_start_thread (void)
409{ 435{
410 worker *wrk = calloc (1, sizeof (worker)); 436 etp_worker *wrk = calloc (1, sizeof (etp_worker));
411 437
412 /*TODO*/ 438 /*TODO*/
413 assert (("unable to allocate worker thread data", wrk)); 439 assert (("unable to allocate worker thread data", wrk));
414 440
415 X_LOCK (etplock); 441 X_LOCK (wrklock);
416 442
417 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 443 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
418 { 444 {
419 wrk->prev = &wrk_first; 445 wrk->prev = &wrk_first;
420 wrk->next = wrk_first.next; 446 wrk->next = wrk_first.next;
421 wrk_first.next->prev = wrk; 447 wrk_first.next->prev = wrk;
422 wrk_first.next = wrk; 448 wrk_first.next = wrk;
423 ++started; 449 ++started;
424 } 450 }
425 else 451 else
426 free (wrk); 452 free (wrk);
427 453
428 X_UNLOCK (etplock); 454 X_UNLOCK (wrklock);
429} 455}
430 456
431static void maybe_start_thread (void) 457static void etp_maybe_start_thread (void)
432{ 458{
433 if (eio_nthreads () >= wanted) 459 if (expect_true (etp_nthreads () >= wanted))
434 return; 460 return;
435 461
436 /* todo: maybe use idle here, but might be less exact */ 462 /* todo: maybe use idle here, but might be less exact */
437 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 463 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
438 return; 464 return;
439 465
440 start_thread (); 466 etp_start_thread ();
441} 467}
442 468
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) 469static void etp_end_thread (void)
462{ 470{
463 eio_req *req = calloc (1, sizeof (eio_req)); 471 eio_req *req = calloc (1, sizeof (eio_req));
464 472
465 req->type = EIO_QUIT; 473 req->type = -1;
466 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 474 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
467 475
468 X_LOCK (reqlock); 476 X_LOCK (reqlock);
469 reqq_push (&req_queue, req); 477 reqq_push (&req_queue, req);
470 X_COND_SIGNAL (reqwait); 478 X_COND_SIGNAL (reqwait);
471 X_UNLOCK (reqlock); 479 X_UNLOCK (reqlock);
472 480
473 X_LOCK (etplock); 481 X_LOCK (wrklock);
474 --started; 482 --started;
475 X_UNLOCK (etplock); 483 X_UNLOCK (wrklock);
476} 484}
477 485
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) 486static int etp_poll (void)
515{ 487{
516 int maxreqs = etp.max_poll_reqs; 488 unsigned int maxreqs;
489 unsigned int maxtime;
517 struct timeval tv_start, tv_now; 490 struct timeval tv_start, tv_now;
518 eio_req *req;
519 491
520 if (etp.max_poll_time) 492 X_LOCK (reslock);
493 maxreqs = max_poll_reqs;
494 maxtime = max_poll_time;
495 X_UNLOCK (reslock);
496
497 if (maxtime)
521 gettimeofday (&tv_start, 0); 498 gettimeofday (&tv_start, 0);
522 499
523 for (;;) 500 for (;;)
524 { 501 {
502 ETP_REQ *req;
503
525 maybe_start_thread (); 504 etp_maybe_start_thread ();
526 505
527 X_LOCK (reslock); 506 X_LOCK (reslock);
528 req = reqq_shift (&res_queue); 507 req = reqq_shift (&res_queue);
529 508
530 if (req) 509 if (req)
531 { 510 {
532 --npending; 511 --npending;
533 512
534 if (!res_queue.size && etp.done_poll_cb) 513 if (!res_queue.size && done_poll_cb)
535 etp.done_poll_cb (); 514 done_poll_cb ();
536 } 515 }
537 516
538 X_UNLOCK (reslock); 517 X_UNLOCK (reslock);
539 518
540 if (!req) 519 if (!req)
541 return 0; 520 return 0;
542 521
522 X_LOCK (reqlock);
543 --nreqs; 523 --nreqs;
524 X_UNLOCK (reqlock);
544 525
545 if (req->type == EIO_GROUP && req->size) 526 if (expect_false (req->type == EIO_GROUP && req->size))
546 { 527 {
547 req->int1 = 1; /* mark request as delayed */ 528 req->int1 = 1; /* mark request as delayed */
548 continue; 529 continue;
549 } 530 }
550 else 531 else
551 { 532 {
552 int res = eio_finish (req); 533 int res = ETP_FINISH (req);
553 if (res) 534 if (expect_false (res))
554 return res; 535 return res;
555 } 536 }
556 537
557 if (maxreqs && !--maxreqs) 538 if (expect_false (maxreqs && !--maxreqs))
558 break; 539 break;
559 540
560 if (etp.max_poll_time) 541 if (maxtime)
561 { 542 {
562 gettimeofday (&tv_now, 0); 543 gettimeofday (&tv_now, 0);
563 544
564 if (tvdiff (&tv_start, &tv_now) >= etp.max_poll_time) 545 if (tvdiff (&tv_start, &tv_now) >= maxtime)
565 break; 546 break;
566 } 547 }
567 } 548 }
568 549
569 errno = EAGAIN; 550 errno = EAGAIN;
570 return -1; 551 return -1;
571} 552}
572 553
554static void etp_cancel (ETP_REQ *req)
555{
556 X_LOCK (wrklock);
557 req->flags |= EIO_FLAG_CANCELLED;
558 X_UNLOCK (wrklock);
559
560 eio_grp_cancel (req);
561}
562
563static void etp_submit (ETP_REQ *req)
564{
565 req->pri -= ETP_PRI_MIN;
566
567 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
568 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
569
570 if (expect_false (req->type == EIO_GROUP))
571 {
572 /* I hope this is worth it :/ */
573 X_LOCK (reqlock);
574 ++nreqs;
575 X_UNLOCK (reqlock);
576
577 X_LOCK (reslock);
578
579 ++npending;
580
581 if (!reqq_push (&res_queue, req) && want_poll_cb)
582 want_poll_cb ();
583
584 X_UNLOCK (reslock);
585 }
586 else
587 {
588 X_LOCK (reqlock);
589 ++nreqs;
590 ++nready;
591 reqq_push (&req_queue, req);
592 X_COND_SIGNAL (reqwait);
593 X_UNLOCK (reqlock);
594
595 etp_maybe_start_thread ();
596 }
597}
598
599static void etp_set_max_poll_time (double nseconds)
600{
601 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
602 max_poll_time = nseconds;
603 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
604}
605
606static void etp_set_max_poll_reqs (unsigned int maxreqs)
607{
608 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609 max_poll_reqs = maxreqs;
610 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611}
612
613static void etp_set_max_idle (unsigned int nthreads)
614{
615 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
616 max_idle = nthreads <= 0 ? 1 : nthreads;
617 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
618}
619
620static void etp_set_min_parallel (unsigned int nthreads)
621{
622 if (wanted < nthreads)
623 wanted = nthreads;
624}
625
626static void etp_set_max_parallel (unsigned int nthreads)
627{
628 if (wanted > nthreads)
629 wanted = nthreads;
630
631 while (started > wanted)
632 etp_end_thread ();
633}
634
635/*****************************************************************************/
636
637static void grp_try_feed (eio_req *grp)
638{
639 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
640 {
641 grp->flags &= ~EIO_FLAG_GROUPADD;
642
643 EIO_FEED (grp);
644
645 /* stop if no progress has been made */
646 if (!(grp->flags & EIO_FLAG_GROUPADD))
647 {
648 grp->feed = 0;
649 break;
650 }
651 }
652}
653
654static int grp_dec (eio_req *grp)
655{
656 --grp->size;
657
658 /* call feeder, if applicable */
659 grp_try_feed (grp);
660
661 /* finish, if done */
662 if (!grp->size && grp->int1)
663 return eio_finish (grp);
664 else
665 return 0;
666}
667
668void eio_destroy (eio_req *req)
669{
670 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
671 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
672
673 EIO_DESTROY (req);
674}
675
676static int eio_finish (eio_req *req)
677{
678 int res = EIO_FINISH (req);
679
680 if (req->grp)
681 {
682 int res2;
683 eio_req *grp = req->grp;
684
685 /* unlink request */
686 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
687 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
688
689 if (grp->grp_first == req)
690 grp->grp_first = req->grp_next;
691
692 res2 = grp_dec (grp);
693
694 if (!res && res2)
695 res = res2;
696 }
697
698 eio_destroy (req);
699
700 return res;
701}
702
703void eio_grp_cancel (eio_req *grp)
704{
705 for (grp = grp->grp_first; grp; grp = grp->grp_next)
706 eio_cancel (grp);
707}
708
709void eio_cancel (eio_req *req)
710{
711 etp_cancel (req);
712}
713
714void eio_submit (eio_req *req)
715{
716 etp_submit (req);
717}
718
719unsigned int eio_nreqs (void)
720{
721 return etp_nreqs ();
722}
723
724unsigned int eio_nready (void)
725{
726 return etp_nready ();
727}
728
729unsigned int eio_npending (void)
730{
731 return etp_npending ();
732}
733
734unsigned int eio_nthreads (void)
735{
736 return etp_nthreads ();
737}
738
739void eio_set_max_poll_time (double nseconds)
740{
741 etp_set_max_poll_time (nseconds);
742}
743
744void eio_set_max_poll_reqs (unsigned int maxreqs)
745{
746 etp_set_max_poll_reqs (maxreqs);
747}
748
749void eio_set_max_idle (unsigned int nthreads)
750{
751 etp_set_max_idle (nthreads);
752}
753
754void eio_set_min_parallel (unsigned int nthreads)
755{
756 etp_set_min_parallel (nthreads);
757}
758
759void eio_set_max_parallel (unsigned int nthreads)
760{
761 etp_set_max_parallel (nthreads);
762}
763
764int eio_poll (void)
765{
766 return etp_poll ();
767}
768
573/*****************************************************************************/ 769/*****************************************************************************/
574/* work around various missing functions */ 770/* work around various missing functions */
575 771
576#if !HAVE_PREADWRITE 772#if !HAVE_PREADWRITE
773# undef pread
774# undef pwrite
577# define pread eio__pread 775# define pread eio__pread
578# define pwrite eio__pwrite 776# 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;
586 777
587static ssize_t 778static ssize_t
588eio__pread (int fd, void *buf, size_t count, off_t offset) 779eio__pread (int fd, void *buf, size_t count, off_t offset)
589{ 780{
590 ssize_t res; 781 ssize_t res;
608 799
609 X_LOCK (preadwritelock); 800 X_LOCK (preadwritelock);
610 ooffset = lseek (fd, 0, SEEK_CUR); 801 ooffset = lseek (fd, 0, SEEK_CUR);
611 lseek (fd, offset, SEEK_SET); 802 lseek (fd, offset, SEEK_SET);
612 res = write (fd, buf, count); 803 res = write (fd, buf, count);
613 lseek (fd, offset, SEEK_SET); 804 lseek (fd, ooffset, SEEK_SET);
614 X_UNLOCK (preadwritelock); 805 X_UNLOCK (preadwritelock);
615 806
616 return res; 807 return res;
617} 808}
618#endif 809#endif
619 810
620#ifndef HAVE_FUTIMES 811#ifndef HAVE_FUTIMES
621 812
813# undef utimes
814# undef futimes
622# define utimes(path,times) eio__utimes (path, times) 815# define utimes(path,times) eio__utimes (path, times)
623# define futimes(fd,times) eio__futimes (fd, times) 816# define futimes(fd,times) eio__futimes (fd, times)
624 817
625static int 818static int
626eio__utimes (const char *filename, const struct timeval times[2]) 819eio__utimes (const char *filename, const struct timeval times[2])
645} 838}
646 839
647#endif 840#endif
648 841
649#if !HAVE_FDATASYNC 842#if !HAVE_FDATASYNC
843# undef fdatasync
650# define fdatasync fsync 844# define fdatasync(fd) fsync (fd)
651#endif 845#endif
846
847/* sync_file_range always needs emulation */
848int
849eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
850{
851#if HAVE_SYNC_FILE_RANGE
852 int res;
853
854 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
855 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
856 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
857 {
858 flags = 0
859 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
860 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
861 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
862 }
863
864 res = sync_file_range (fd, offset, nbytes, flags);
865
866 if (!res || errno != ENOSYS)
867 return res;
868#endif
869
870 /* even though we could play tricks with the flags, it's better to always
871 * call fdatasync, as thta matches the expectation of it's users best */
872 return fdatasync (fd);
873}
652 874
653#if !HAVE_READAHEAD 875#if !HAVE_READAHEAD
876# undef readahead
654# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 877# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
655 878
656static ssize_t 879static ssize_t
657eio__readahead (int fd, off_t offset, size_t count, worker *self) 880eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
658{ 881{
659 size_t todo = count; 882 size_t todo = count;
660 dBUF; 883 dBUF;
661 884
662 while (todo > 0) 885 while (todo > 0)
672 return count; 895 return count;
673} 896}
674 897
675#endif 898#endif
676 899
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 */ 900/* sendfile always needs emulation */
709static ssize_t 901static ssize_t
710eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self) 902eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
711{ 903{
712 ssize_t res; 904 ssize_t res;
713 905
714 if (!count) 906 if (!count)
715 return 0; 907 return 0;
797 } 989 }
798 990
799 return res; 991 return res;
800} 992}
801 993
994static int
995eio_dent_cmp (const void *a_, const void *b_)
996{
997 const eio_dirent *a = (const eio_dirent *)a_;
998 const eio_dirent *b = (const eio_dirent *)b_;
999
1000 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score
1001 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */
1002}
1003
802/* read a full directory */ 1004/* read a full directory */
803static void 1005static void
804eio__scandir (eio_req *req, worker *self) 1006eio__scandir (eio_req *req, etp_worker *self)
805{ 1007{
806 DIR *dirp; 1008 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; 1009 EIO_STRUCT_DIRENT *entp;
813 char *name, *names; 1010 unsigned char *name, *names;
814 int memlen = 4096; 1011 int namesalloc = 4096;
815 int memofs = 0; 1012 int namesoffs = 0;
1013 int flags = req->int1;
1014 eio_dirent *dents = 0;
1015 int dentalloc = 128;
816 int res = 0; 1016 int dentoffs = 0;
817 1017
1018 req->result = -1;
1019
1020 if (!(flags & EIO_READDIR_DENTS))
1021 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1022
818 X_LOCK (etplock); 1023 X_LOCK (wrklock);
1024 /* the corresponding closedir is in ETP_WORKER_CLEAR */
819 self->dirp = dirp = opendir (req->ptr1); 1025 self->dirp = dirp = opendir (req->ptr1);
820 self->dbuf = u = malloc (sizeof (*u));
821 req->flags |= EIO_FLAG_PTR2_FREE; 1026 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
822 req->ptr2 = names = malloc (memlen); 1027 req->ptr1 = names = malloc (namesalloc);
1028 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
823 X_UNLOCK (etplock); 1029 X_UNLOCK (wrklock);
824 1030
825 if (dirp && u && names) 1031 if (dirp && names && (!flags || dents))
826 for (;;) 1032 for (;;)
827 { 1033 {
828 errno = 0; 1034 errno = 0;
829 readdir_r (dirp, &u->d, &entp); 1035 entp = readdir (dirp);
830 1036
831 if (!entp) 1037 if (!entp)
1038 {
1039 if (errno)
1040 break;
1041
1042 /* sort etc. */
1043 req->int1 = flags;
1044 req->result = dentoffs;
1045
1046 if (dents)
1047 {
1048 eio_dirent *ent = dents + dentoffs;
1049
1050 while (ent > dents)
1051 (--ent)->name = names + (size_t)ent->name;
1052 }
1053
1054 if (flags & EIO_READDIR_STAT_ORDER
1055 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN)))
1056 {
1057 /* pray your qsort doesn't use quicksort */
1058 qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1059 }
1060 else if (flags & EIO_READDIR_DIRS_FIRST)
1061 {
1062 /* in this case, all is known, and we just put dirs first and sort them */
1063 eio_dirent *ent = dents + dentoffs;
1064 eio_dirent *dir = dents;
1065
1066 while (ent > dir)
1067 {
1068 if (dir->type == DT_DIR)
1069 ++dir;
1070 else
1071 {
1072 --ent;
1073
1074 if (ent->type == DT_DIR)
1075 {
1076 eio_dirent tmp = *dir;
1077 *dir = *ent;
1078 *ent = tmp;
1079
1080 ++dir;
1081 }
1082 }
1083 }
1084
1085 /* now sort the dirs only */
1086 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1087 }
1088
1089 /* only provide the names array unless DENTS is specified */
1090 if (!(flags & EIO_READDIR_DENTS))
1091 {
1092 X_LOCK (wrklock);
1093 assert (!dents);
1094 req->ptr1 = 0;
1095 req->ptr2 = names;
1096 X_UNLOCK (wrklock);
1097 }
1098
832 break; 1099 break;
1100 }
833 1101
1102 /* now add the entry to our list(s) */
834 name = entp->d_name; 1103 name = entp->d_name;
835 1104
1105 /* skip . and .. entries */
836 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1106 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
837 { 1107 {
838 int len = strlen (name) + 1; 1108 int len = D_NAMLEN (entp) + 1;
839 1109
840 res++; 1110 while (expect_false (namesoffs + len > namesalloc))
841
842 while (memofs + len > memlen)
843 { 1111 {
844 memlen *= 2; 1112 namesalloc *= 2;
845 X_LOCK (etplock); 1113 X_LOCK (wrklock);
846 req->ptr2 = names = realloc (names, memlen); 1114 req->ptr1 = names = realloc (names, namesalloc);
847 X_UNLOCK (etplock); 1115 X_UNLOCK (wrklock);
848 1116
849 if (!names) 1117 if (!names)
850 break; 1118 break;
851 } 1119 }
852 1120
853 memcpy (names + memofs, name, len); 1121 memcpy (names + namesoffs, name, len);
1122
1123 if (dents)
1124 {
1125 struct eio_dirent *ent;
1126
1127 if (expect_false (dentoffs == dentalloc))
1128 {
1129 dentalloc *= 2;
1130 X_LOCK (wrklock);
1131 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1132 X_UNLOCK (wrklock);
1133
1134 if (!dents)
1135 break;
1136 }
1137
1138 ent = dents + dentoffs;
1139
1140 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */
1141 ent->namelen = len - 1;
1142 ent->inode = D_INO (entp);
1143
1144 switch (D_TYPE (entp))
1145 {
1146 default:
1147 ent->type = EIO_DT_UNKNOWN;
1148 flags |= EIO_READDIR_FOUND_UNKNOWN;
1149 break;
1150
1151 #ifdef DT_FIFO
1152 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1153 #endif
1154 #ifdef DT_CHR
1155 case DT_CHR: ent->type = EIO_DT_CHR; break;
1156 #endif
1157 #ifdef DT_MPC
1158 case DT_MPC: ent->type = EIO_DT_MPC; break;
1159 #endif
1160 #ifdef DT_DIR
1161 case DT_DIR: ent->type = EIO_DT_DIR; break;
1162 #endif
1163 #ifdef DT_NAM
1164 case DT_NAM: ent->type = EIO_DT_NAM; break;
1165 #endif
1166 #ifdef DT_BLK
1167 case DT_BLK: ent->type = EIO_DT_BLK; break;
1168 #endif
1169 #ifdef DT_MPB
1170 case DT_MPB: ent->type = EIO_DT_MPB; break;
1171 #endif
1172 #ifdef DT_REG
1173 case DT_REG: ent->type = EIO_DT_REG; break;
1174 #endif
1175 #ifdef DT_NWK
1176 case DT_NWK: ent->type = EIO_DT_NWK; break;
1177 #endif
1178 #ifdef DT_CMP
1179 case DT_CMP: ent->type = EIO_DT_CMP; break;
1180 #endif
1181 #ifdef DT_LNK
1182 case DT_LNK: ent->type = EIO_DT_LNK; break;
1183 #endif
1184 #ifdef DT_SOCK
1185 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1186 #endif
1187 #ifdef DT_DOOR
1188 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1189 #endif
1190 #ifdef DT_WHT
1191 case DT_WHT: ent->type = EIO_DT_WHT; break;
1192 #endif
1193 }
1194
1195 ent->score = 0;
1196
1197 if (flags & EIO_READDIR_DIRS_FIRST)
1198 {
1199 if (ent->type == EIO_DT_UNKNOWN)
1200 {
1201 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1202 ent->score = 98;
1203 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1204 ent->score = len <= 2 ? len + 60 : len <= 4 ? 50 : len <= 7 ? 40 : 10; /* shorter == more likely dir, but avoid too many classes */
1205 }
1206 else if (ent->type == EIO_DT_DIR)
1207 ent->score = 100;
1208 }
1209 }
1210
854 memofs += len; 1211 namesoffs += len;
1212 ++dentoffs;
855 } 1213 }
856 } 1214 }
857 1215 else
858 if (errno)
859 res = -1;
860
861 req->result = res; 1216 req->result = -1;
1217}
1218
1219#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1220# undef msync
1221# define msync(a,b,c) ((errno = ENOSYS), -1)
1222#endif
1223
1224int
1225eio__mtouch (void *mem, size_t len, int flags)
1226{
1227 intptr_t addr = (intptr_t)mem;
1228 intptr_t end = addr + len;
1229#ifdef PAGESIZE
1230 const intptr_t page = PAGESIZE;
1231#else
1232 static intptr_t page;
1233
1234 if (!page)
1235 page = sysconf (_SC_PAGESIZE);
1236#endif
1237
1238 addr &= ~(page - 1); /* assume page size is always a power of two */
1239
1240 if (addr < end)
1241 if (flags) /* modify */
1242 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1243 else
1244 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1245
1246 return 0;
862} 1247}
863 1248
864/*****************************************************************************/ 1249/*****************************************************************************/
865 1250
866#define ALLOC(len) \ 1251#define ALLOC(len) \
867 if (!req->ptr2) \ 1252 if (!req->ptr2) \
868 { \ 1253 { \
869 X_LOCK (etplock); \ 1254 X_LOCK (wrklock); \
870 req->flags |= EIO_FLAG_PTR2_FREE; \ 1255 req->flags |= EIO_FLAG_PTR2_FREE; \
871 X_UNLOCK (etplock); \ 1256 X_UNLOCK (wrklock); \
872 req->ptr2 = malloc (len); \ 1257 req->ptr2 = malloc (len); \
873 if (!req->ptr2) \ 1258 if (!req->ptr2) \
874 { \ 1259 { \
875 errno = ENOMEM; \ 1260 errno = ENOMEM; \
876 req->result = -1; \ 1261 req->result = -1; \
877 break; \ 1262 break; \
878 } \ 1263 } \
879 } 1264 }
880 1265
881X_THREAD_PROC (eio_proc) 1266X_THREAD_PROC (etp_proc)
882{ 1267{
883 eio_req *req; 1268 ETP_REQ *req;
884 struct timespec ts; 1269 struct timespec ts;
885 worker *self = (worker *)thr_arg; 1270 etp_worker *self = (etp_worker *)thr_arg;
886 1271
887 /* try to distribute timeouts somewhat randomly */ 1272 /* try to distribute timeouts somewhat randomly */
888 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1273 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
889 1274
890 for (;;) 1275 for (;;)
891 { 1276 {
892 ts.tv_sec = time (0) + IDLE_TIMEOUT;
893
894 X_LOCK (reqlock); 1277 X_LOCK (reqlock);
895 1278
896 for (;;) 1279 for (;;)
897 { 1280 {
898 self->req = req = reqq_shift (&req_queue); 1281 self->req = req = reqq_shift (&req_queue);
900 if (req) 1283 if (req)
901 break; 1284 break;
902 1285
903 ++idle; 1286 ++idle;
904 1287
1288 ts.tv_sec = time (0) + IDLE_TIMEOUT;
905 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1289 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
906 { 1290 {
907 if (idle > max_idle) 1291 if (idle > max_idle)
908 { 1292 {
909 --idle; 1293 --idle;
910 X_UNLOCK (reqlock); 1294 X_UNLOCK (reqlock);
911 X_LOCK (etplock); 1295 X_LOCK (wrklock);
912 --started; 1296 --started;
913 X_UNLOCK (etplock); 1297 X_UNLOCK (wrklock);
914 goto quit; 1298 goto quit;
915 } 1299 }
916 1300
917 /* we are allowed to idle, so do so without any timeout */ 1301 /* we are allowed to idle, so do so without any timeout */
918 X_COND_WAIT (reqwait, reqlock); 1302 X_COND_WAIT (reqwait, reqlock);
919 ts.tv_sec = time (0) + IDLE_TIMEOUT;
920 } 1303 }
921 1304
922 --idle; 1305 --idle;
923 } 1306 }
924 1307
925 --nready; 1308 --nready;
926 1309
927 X_UNLOCK (reqlock); 1310 X_UNLOCK (reqlock);
928 1311
929 errno = 0; /* strictly unnecessary */ 1312 if (req->type < 0)
1313 goto quit;
930 1314
931 if (!EIO_CANCELLED (req)) 1315 if (!EIO_CANCELLED (req))
932 switch (req->type) 1316 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 1317
1033 X_LOCK (reslock); 1318 X_LOCK (reslock);
1034 1319
1035 ++npending; 1320 ++npending;
1036 1321
1037 if (!reqq_push (&res_queue, req) && etp.want_poll_cb) 1322 if (!reqq_push (&res_queue, req) && want_poll_cb)
1038 etp.want_poll_cb (); 1323 want_poll_cb ();
1039 1324
1040 self->req = 0; 1325 self->req = 0;
1041 worker_clear (self); 1326 etp_worker_clear (self);
1042 1327
1043 X_UNLOCK (reslock); 1328 X_UNLOCK (reslock);
1044 } 1329 }
1045 1330
1046quit: 1331quit:
1047 X_LOCK (etplock); 1332 X_LOCK (wrklock);
1048 worker_free (self); 1333 etp_worker_free (self);
1049 X_UNLOCK (etplock); 1334 X_UNLOCK (wrklock);
1050 1335
1051 return 0; 1336 return 0;
1052} 1337}
1053 1338
1054/*****************************************************************************/ 1339/*****************************************************************************/
1055 1340
1056int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1341int eio_init (void (*want_poll)(void), void (*done_poll)(void))
1057{ 1342{
1058 etp_init (&etp, want_poll, done_poll); 1343 return etp_init (want_poll, done_poll);
1059} 1344}
1060 1345
1061static void eio_api_destroy (eio_req *req) 1346static void eio_api_destroy (eio_req *req)
1062{ 1347{
1063 free (req); 1348 free (req);
1085 { \ 1370 { \
1086 eio_api_destroy (req); \ 1371 eio_api_destroy (req); \
1087 return 0; \ 1372 return 0; \
1088 } 1373 }
1089 1374
1375static void eio_execute (etp_worker *self, eio_req *req)
1376{
1377 errno = 0;
1378
1379 switch (req->type)
1380 {
1381 case EIO_READ: ALLOC (req->size);
1382 req->result = req->offs >= 0
1383 ? pread (req->int1, req->ptr2, req->size, req->offs)
1384 : read (req->int1, req->ptr2, req->size); break;
1385 case EIO_WRITE: req->result = req->offs >= 0
1386 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1387 : write (req->int1, req->ptr2, req->size); break;
1388
1389 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1390 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
1391
1392 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1393 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1394 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1395 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1396 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1397 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1398
1399 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1400 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1401 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1402 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1403 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1404 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1405
1406 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1407 case EIO_CLOSE: req->result = close (req->int1); break;
1408 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1409 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1410 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1411 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1412 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1413 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1414 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1415 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break;
1416
1417 case EIO_READLINK: ALLOC (NAME_MAX);
1418 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1419
1420 case EIO_SYNC: req->result = 0; sync (); break;
1421 case EIO_FSYNC: req->result = fsync (req->int1); break;
1422 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1423 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;
1424 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
1425 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1426
1427 case EIO_READDIR: eio__scandir (req, self); break;
1428
1429 case EIO_BUSY:
1430#ifdef _WIN32
1431 Sleep (req->nv1 * 1000.);
1432#else
1433 {
1434 struct timeval tv;
1435
1436 tv.tv_sec = req->nv1;
1437 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
1438
1439 req->result = select (0, 0, 0, 0, &tv);
1440 }
1441#endif
1442 break;
1443
1444 case EIO_UTIME:
1445 case EIO_FUTIME:
1446 {
1447 struct timeval tv[2];
1448 struct timeval *times;
1449
1450 if (req->nv1 != -1. || req->nv2 != -1.)
1451 {
1452 tv[0].tv_sec = req->nv1;
1453 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1454 tv[1].tv_sec = req->nv2;
1455 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1456
1457 times = tv;
1458 }
1459 else
1460 times = 0;
1461
1462
1463 req->result = req->type == EIO_FUTIME
1464 ? futimes (req->int1, times)
1465 : utimes (req->ptr1, times);
1466 }
1467 break;
1468
1469 case EIO_GROUP:
1470 abort (); /* handled in eio_request */
1471
1472 case EIO_NOP:
1473 req->result = 0;
1474 break;
1475
1476 case EIO_CUSTOM:
1477 ((void (*)(eio_req *))req->feed) (req);
1478 break;
1479
1480 default:
1481 req->result = -1;
1482 break;
1483 }
1484
1485 req->errorno = errno;
1486}
1487
1090#ifndef EIO_NO_WRAPPERS 1488#ifndef EIO_NO_WRAPPERS
1091 1489
1092eio_req *eio_nop (int pri, eio_cb cb, void *data) 1490eio_req *eio_nop (int pri, eio_cb cb, void *data)
1093{ 1491{
1094 REQ (EIO_NOP); SEND; 1492 REQ (EIO_NOP); SEND;
1105} 1503}
1106 1504
1107eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 1505eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1108{ 1506{
1109 REQ (EIO_FSYNC); req->int1 = fd; SEND; 1507 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1508}
1509
1510eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1511{
1512 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1513}
1514
1515eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1516{
1517 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1518}
1519
1520eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1521{
1522 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1110} 1523}
1111 1524
1112eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 1525eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1113{ 1526{
1114 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 1527 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1228eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 1641eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1229{ 1642{
1230 return eio__1path (EIO_RMDIR, path, pri, cb, data); 1643 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1231} 1644}
1232 1645
1233eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 1646eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1234{ 1647{
1235 return eio__1path (EIO_READDIR, path, pri, cb, data); 1648 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1236} 1649}
1237 1650
1238eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 1651eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1239{ 1652{
1240 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 1653 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
1241} 1654}
1242 1655
1243static eio_req * 1656static eio_req *
1244eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1657eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1245{ 1658{
1269eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 1682eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1270{ 1683{
1271 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 1684 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1272} 1685}
1273 1686
1687eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data)
1688{
1689 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND;
1690}
1691
1274#endif 1692#endif
1275 1693
1276eio_req *eio_grp (eio_cb cb, void *data) 1694eio_req *eio_grp (eio_cb cb, void *data)
1277{ 1695{
1278 const int pri = EIO_PRI_MAX; 1696 const int pri = EIO_PRI_MAX;
1304 1722
1305void eio_grp_add (eio_req *grp, eio_req *req) 1723void eio_grp_add (eio_req *grp, eio_req *req)
1306{ 1724{
1307 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 1725 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1308 1726
1727 grp->flags |= EIO_FLAG_GROUPADD;
1728
1309 ++grp->size; 1729 ++grp->size;
1310 req->grp = grp; 1730 req->grp = grp;
1311 1731
1312 req->grp_prev = 0; 1732 req->grp_prev = 0;
1313 req->grp_next = grp->grp_first; 1733 req->grp_next = grp->grp_first;
1321/*****************************************************************************/ 1741/*****************************************************************************/
1322/* misc garbage */ 1742/* misc garbage */
1323 1743
1324ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 1744ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1325{ 1745{
1326 worker wrk; 1746 etp_worker wrk;
1327 1747
1328 wrk.dbuf = 0; 1748 wrk.dbuf = 0;
1329 1749
1330 eio__sendfile (ofd, ifd, offset, count, &wrk); 1750 eio__sendfile (ofd, ifd, offset, count, &wrk);
1331 1751

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines