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

Comparing libeio/eio.c (file contents):
Revision 1.56 by root, Sun Sep 12 03:36:28 2010 UTC vs.
Revision 1.93 by root, Mon Jul 18 01:27:03 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
35 * and other provisions required by the GPL. If you do not delete the 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 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
41 46
42#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define XTHREAD_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.h"
49#include <stdlib.h> 54#include <stdlib.h>
50#include <string.h> 55#include <string.h>
51#include <errno.h> 56#include <errno.h>
52#include <sys/types.h> 57#include <sys/types.h>
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <sys/statvfs.h>
55#include <limits.h> 59#include <limits.h>
56#include <fcntl.h> 60#include <fcntl.h>
57#include <assert.h> 61#include <assert.h>
58 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76static void eio_destroy (eio_req *req);
77
59#ifndef EIO_FINISH 78#ifndef EIO_FINISH
60# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 79# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
61#endif 80#endif
62 81
63#ifndef EIO_DESTROY 82#ifndef EIO_DESTROY
66 85
67#ifndef EIO_FEED 86#ifndef EIO_FEED
68# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 87# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
69#endif 88#endif
70 89
90#ifndef EIO_FD_TO_WIN32_HANDLE
91# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
92#endif
93#ifndef EIO_WIN32_HANDLE_TO_FD
94# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
95#endif
96
97#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
98
99#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
100
71#ifdef _WIN32 101#ifdef _WIN32
72 102
73 /*doh*/ 103 #undef PAGESIZE
104 #define PAGESIZE 4096 /* GetSystemInfo? */
105
106 #ifdef EIO_STRUCT_STATI64
107 #define stat(path,buf) _stati64 (path,buf)
108 #define fstat(fd,buf) _fstati64 (path,buf)
109 #endif
110 #define lstat(path,buf) stat (path,buf)
111 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
112 #define mkdir(path,mode) _mkdir (path)
113 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
114
115 #define chmod(path,mode) _chmod (path, mode)
116 #define dup(fd) _dup (fd)
117 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
118
119 #define fchmod(fd,mode) EIO_ENOSYS ()
120 #define chown(path,uid,gid) EIO_ENOSYS ()
121 #define fchown(fd,uid,gid) EIO_ENOSYS ()
122 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
123 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
124 #define mknod(path,mode,dev) EIO_ENOSYS ()
125 #define sync() EIO_ENOSYS ()
126 #define readlink(path,buf,s) EIO_ENOSYS ()
127 #define statvfs(path,buf) EIO_ENOSYS ()
128 #define fstatvfs(fd,buf) EIO_ENOSYS ()
129
130 /* we could even stat and see if it exists */
131 static int
132 symlink (const char *old, const char *neu)
133 {
134 #if WINVER >= 0x0600
135 if (CreateSymbolicLink (neu, old, 1))
136 return 0;
137
138 if (CreateSymbolicLink (neu, old, 0))
139 return 0;
140 #endif
141
142 return EIO_ERRNO (ENOENT, -1);
143 }
144
145 /* POSIX API only */
146 #define CreateHardLink(neu,old,flags) 0
147 #define CreateSymbolicLink(neu,old,flags) 0
148
149 struct statvfs
150 {
151 int dummy;
152 };
153
154 #define DT_DIR EIO_DT_DIR
155 #define DT_REG EIO_DT_REG
156 #define D_NAME(entp) entp.cFileName
157 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
158
74#else 159#else
75 160
76# include "config.h"
77# include <sys/time.h> 161 #include <sys/time.h>
78# include <sys/select.h> 162 #include <sys/select.h>
163 #include <sys/statvfs.h>
79# include <unistd.h> 164 #include <unistd.h>
80# include <utime.h> 165 #include <utime.h>
81# include <signal.h> 166 #include <signal.h>
82# include <dirent.h> 167 #include <dirent.h>
83 168
84#if _POSIX_MEMLOCK || _POSIX_MAPPED_FILES 169 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
85# include <sys/mman.h> 170 #include <sys/mman.h>
86#endif 171 #endif
87 172
173 #define D_NAME(entp) entp->d_name
174
88/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 175 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
89# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 176 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
90# define _DIRENT_HAVE_D_TYPE /* sigh */ 177 #define _DIRENT_HAVE_D_TYPE /* sigh */
91# define D_INO(de) (de)->d_fileno 178 #define D_INO(de) (de)->d_fileno
92# define D_NAMLEN(de) (de)->d_namlen 179 #define D_NAMLEN(de) (de)->d_namlen
93# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 180 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
94# define D_INO(de) (de)->d_ino 181 #define D_INO(de) (de)->d_ino
95# endif 182 #endif
96 183
97#ifdef _D_EXACT_NAMLEN 184 #ifdef _D_EXACT_NAMLEN
98# undef D_NAMLEN 185 #undef D_NAMLEN
99# define D_NAMLEN(de) _D_EXACT_NAMLEN (de) 186 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
100#endif 187 #endif
101 188
102# ifdef _DIRENT_HAVE_D_TYPE 189 #ifdef _DIRENT_HAVE_D_TYPE
103# define D_TYPE(de) (de)->d_type 190 #define D_TYPE(de) (de)->d_type
104# endif 191 #endif
105 192
106# ifndef EIO_STRUCT_DIRENT 193 #ifndef EIO_STRUCT_DIRENT
107# define EIO_STRUCT_DIRENT struct dirent 194 #define EIO_STRUCT_DIRENT struct dirent
108# endif 195 #endif
109 196
110#endif 197#endif
111 198
112#if HAVE_SENDFILE 199#if HAVE_SENDFILE
113# if __linux 200# if __linux
129#endif 216#endif
130#ifndef D_INO 217#ifndef D_INO
131# define D_INO(de) 0 218# define D_INO(de) 0
132#endif 219#endif
133#ifndef D_NAMLEN 220#ifndef D_NAMLEN
134# define D_NAMLEN(de) strlen ((de)->d_name) 221# define D_NAMLEN(entp) strlen (D_NAME (entp))
135#endif 222#endif
136
137/* number of seconds after which an idle threads exit */
138#define IDLE_TIMEOUT 10
139 223
140/* used for struct dirent, AIX doesn't provide it */ 224/* used for struct dirent, AIX doesn't provide it */
141#ifndef NAME_MAX 225#ifndef NAME_MAX
142# define NAME_MAX 4096 226# define NAME_MAX 4096
143#endif 227#endif
159 if (!eio_buf) \ 243 if (!eio_buf) \
160 return -1; 244 return -1;
161 245
162#define EIO_TICKS ((1000000 + 1023) >> 10) 246#define EIO_TICKS ((1000000 + 1023) >> 10)
163 247
164/*****************************************************************************/
165
166#if __GNUC__ >= 3
167# define expect(expr,value) __builtin_expect ((expr),(value))
168#else
169# define expect(expr,value) (expr)
170#endif
171
172#define expect_false(expr) expect ((expr) != 0, 0)
173#define expect_true(expr) expect ((expr) != 0, 1)
174
175/*****************************************************************************/
176
177#define ETP_PRI_MIN EIO_PRI_MIN 248#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 249#define ETP_PRI_MAX EIO_PRI_MAX
179 250
180struct etp_worker; 251struct etp_worker;
181 252
189#define ETP_WORKER_CLEAR(req) \ 260#define ETP_WORKER_CLEAR(req) \
190 if (wrk->dbuf) \ 261 if (wrk->dbuf) \
191 { \ 262 { \
192 free (wrk->dbuf); \ 263 free (wrk->dbuf); \
193 wrk->dbuf = 0; \ 264 wrk->dbuf = 0; \
194 } \
195 \
196 if (wrk->dirp) \
197 { \
198 closedir (wrk->dirp); \
199 wrk->dirp = 0; \
200 } 265 }
201 266
202#define ETP_WORKER_COMMON \ 267#define ETP_WORKER_COMMON \
203 void *dbuf; \ 268 void *dbuf;
204 DIR *dirp;
205 269
206/*****************************************************************************/ 270/*****************************************************************************/
207 271
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 272#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209 273
210/* calculate time difference in ~1/EIO_TICKS of a second */ 274/* calculate time difference in ~1/EIO_TICKS of a second */
275ecb_inline int
211static int tvdiff (struct timeval *tv1, struct timeval *tv2) 276tvdiff (struct timeval *tv1, struct timeval *tv2)
212{ 277{
213 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 278 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
214 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 279 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
215} 280}
216 281
223static unsigned int max_poll_reqs; /* reslock */ 288static unsigned int max_poll_reqs; /* reslock */
224 289
225static volatile unsigned int nreqs; /* reqlock */ 290static volatile unsigned int nreqs; /* reqlock */
226static volatile unsigned int nready; /* reqlock */ 291static volatile unsigned int nready; /* reqlock */
227static volatile unsigned int npending; /* reqlock */ 292static volatile unsigned int npending; /* reqlock */
228static volatile unsigned int max_idle = 4; 293static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
294static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
229 295
230static xmutex_t wrklock = X_MUTEX_INIT; 296static xmutex_t wrklock;
231static xmutex_t reslock = X_MUTEX_INIT; 297static xmutex_t reslock;
232static xmutex_t reqlock = X_MUTEX_INIT; 298static xmutex_t reqlock;
233static xcond_t reqwait = X_COND_INIT; 299static xcond_t reqwait;
234 300
235#if !HAVE_PREADWRITE 301#if !HAVE_PREADWRITE
236/* 302/*
237 * make our pread/pwrite emulation safe against themselves, but not against 303 * make our pread/pwrite emulation safe against themselves, but not against
238 * normal read/write by using a mutex. slows down execution a lot, 304 * normal read/write by using a mutex. slows down execution a lot,
239 * but that's your problem, not mine. 305 * but that's your problem, not mine.
240 */ 306 */
241static xmutex_t preadwritelock = X_MUTEX_INIT; 307static xmutex_t preadwritelock;
242#endif 308#endif
243 309
244typedef struct etp_worker 310typedef struct etp_worker
245{ 311{
246 /* locked by wrklock */ 312 /* locked by wrklock */
252 ETP_REQ *req; /* currently processed request */ 318 ETP_REQ *req; /* currently processed request */
253 319
254 ETP_WORKER_COMMON 320 ETP_WORKER_COMMON
255} etp_worker; 321} etp_worker;
256 322
257static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 323static etp_worker wrk_first; /* NOT etp */
258 324
259#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 325#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
260#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 326#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
261 327
262/* worker threads management */ 328/* worker threads management */
263 329
330static void ecb_cold
264static void etp_worker_clear (etp_worker *wrk) 331etp_worker_clear (etp_worker *wrk)
265{ 332{
266 ETP_WORKER_CLEAR (wrk); 333 ETP_WORKER_CLEAR (wrk);
267} 334}
268 335
336static void ecb_cold
269static void etp_worker_free (etp_worker *wrk) 337etp_worker_free (etp_worker *wrk)
270{ 338{
271 wrk->next->prev = wrk->prev; 339 wrk->next->prev = wrk->prev;
272 wrk->prev->next = wrk->next; 340 wrk->prev->next = wrk->next;
273 341
274 free (wrk); 342 free (wrk);
275} 343}
276 344
277static unsigned int etp_nreqs (void) 345static unsigned int
346etp_nreqs (void)
278{ 347{
279 int retval; 348 int retval;
280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 349 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281 retval = nreqs; 350 retval = nreqs;
282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 351 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283 return retval; 352 return retval;
284} 353}
285 354
286static unsigned int etp_nready (void) 355static unsigned int
356etp_nready (void)
287{ 357{
288 unsigned int retval; 358 unsigned int retval;
289 359
290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 360 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
291 retval = nready; 361 retval = nready;
292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 362 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
293 363
294 return retval; 364 return retval;
295} 365}
296 366
297static unsigned int etp_npending (void) 367static unsigned int
368etp_npending (void)
298{ 369{
299 unsigned int retval; 370 unsigned int retval;
300 371
301 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 372 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
302 retval = npending; 373 retval = npending;
303 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 374 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
304 375
305 return retval; 376 return retval;
306} 377}
307 378
308static unsigned int etp_nthreads (void) 379static unsigned int
380etp_nthreads (void)
309{ 381{
310 unsigned int retval; 382 unsigned int retval;
311 383
312 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 384 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
313 retval = started; 385 retval = started;
327} etp_reqq; 399} etp_reqq;
328 400
329static etp_reqq req_queue; 401static etp_reqq req_queue;
330static etp_reqq res_queue; 402static etp_reqq res_queue;
331 403
404static void ecb_noinline ecb_cold
405reqq_init (etp_reqq *q)
406{
407 int pri;
408
409 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
410 q->qs[pri] = q->qe[pri] = 0;
411
412 q->size = 0;
413}
414
415static int ecb_noinline
332static int reqq_push (etp_reqq *q, ETP_REQ *req) 416reqq_push (etp_reqq *q, ETP_REQ *req)
333{ 417{
334 int pri = req->pri; 418 int pri = req->pri;
335 req->next = 0; 419 req->next = 0;
336 420
337 if (q->qe[pri]) 421 if (q->qe[pri])
343 q->qe[pri] = q->qs[pri] = req; 427 q->qe[pri] = q->qs[pri] = req;
344 428
345 return q->size++; 429 return q->size++;
346} 430}
347 431
432static ETP_REQ * ecb_noinline
348static ETP_REQ *reqq_shift (etp_reqq *q) 433reqq_shift (etp_reqq *q)
349{ 434{
350 int pri; 435 int pri;
351 436
352 if (!q->size) 437 if (!q->size)
353 return 0; 438 return 0;
368 } 453 }
369 454
370 abort (); 455 abort ();
371} 456}
372 457
373static void etp_atfork_prepare (void) 458static int ecb_cold
459etp_init (void (*want_poll)(void), void (*done_poll)(void))
374{ 460{
375 X_LOCK (wrklock); 461 X_MUTEX_CREATE (wrklock);
376 X_LOCK (reqlock); 462 X_MUTEX_CREATE (reslock);
377 X_LOCK (reslock); 463 X_MUTEX_CREATE (reqlock);
378#if !HAVE_PREADWRITE 464 X_COND_CREATE (reqwait);
379 X_LOCK (preadwritelock);
380#endif
381}
382 465
383static void etp_atfork_parent (void) 466 reqq_init (&req_queue);
384{ 467 reqq_init (&res_queue);
385#if !HAVE_PREADWRITE
386 X_UNLOCK (preadwritelock);
387#endif
388 X_UNLOCK (reslock);
389 X_UNLOCK (reqlock);
390 X_UNLOCK (wrklock);
391}
392 468
393static void etp_atfork_child (void) 469 wrk_first.next =
394{ 470 wrk_first.prev = &wrk_first;
395 ETP_REQ *prv;
396
397 while ((prv = reqq_shift (&req_queue)))
398 ETP_DESTROY (prv);
399
400 while ((prv = reqq_shift (&res_queue)))
401 ETP_DESTROY (prv);
402
403 while (wrk_first.next != &wrk_first)
404 {
405 etp_worker *wrk = wrk_first.next;
406
407 if (wrk->req)
408 ETP_DESTROY (wrk->req);
409
410 etp_worker_clear (wrk);
411 etp_worker_free (wrk);
412 }
413 471
414 started = 0; 472 started = 0;
415 idle = 0; 473 idle = 0;
416 nreqs = 0; 474 nreqs = 0;
417 nready = 0; 475 nready = 0;
418 npending = 0; 476 npending = 0;
419 477
420 etp_atfork_parent ();
421}
422
423static void
424etp_once_init (void)
425{
426 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
427}
428
429static int
430etp_init (void (*want_poll)(void), void (*done_poll)(void))
431{
432 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
433
434 pthread_once (&doinit, etp_once_init);
435
436 want_poll_cb = want_poll; 478 want_poll_cb = want_poll;
437 done_poll_cb = done_poll; 479 done_poll_cb = done_poll;
438 480
439 return 0; 481 return 0;
440} 482}
441 483
442X_THREAD_PROC (etp_proc); 484X_THREAD_PROC (etp_proc);
443 485
486static void ecb_cold
444static void etp_start_thread (void) 487etp_start_thread (void)
445{ 488{
446 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 489 etp_worker *wrk = calloc (1, sizeof (etp_worker));
447 490
448 /*TODO*/ 491 /*TODO*/
449 assert (("unable to allocate worker thread data", wrk)); 492 assert (("unable to allocate worker thread data", wrk));
462 free (wrk); 505 free (wrk);
463 506
464 X_UNLOCK (wrklock); 507 X_UNLOCK (wrklock);
465} 508}
466 509
510static void
467static void etp_maybe_start_thread (void) 511etp_maybe_start_thread (void)
468{ 512{
469 if (expect_true (etp_nthreads () >= wanted)) 513 if (ecb_expect_true (etp_nthreads () >= wanted))
470 return; 514 return;
471 515
472 /* todo: maybe use idle here, but might be less exact */ 516 /* todo: maybe use idle here, but might be less exact */
473 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 517 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
474 return; 518 return;
475 519
476 etp_start_thread (); 520 etp_start_thread ();
477} 521}
478 522
523static void ecb_cold
479static void etp_end_thread (void) 524etp_end_thread (void)
480{ 525{
481 eio_req *req = calloc (1, sizeof (eio_req)); 526 eio_req *req = calloc (1, sizeof (eio_req));
482 527
483 req->type = -1; 528 req->type = -1;
484 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 529 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
491 X_LOCK (wrklock); 536 X_LOCK (wrklock);
492 --started; 537 --started;
493 X_UNLOCK (wrklock); 538 X_UNLOCK (wrklock);
494} 539}
495 540
496static int etp_poll (void) 541static int
542etp_poll (void)
497{ 543{
498 unsigned int maxreqs; 544 unsigned int maxreqs;
499 unsigned int maxtime; 545 unsigned int maxtime;
500 struct timeval tv_start, tv_now; 546 struct timeval tv_start, tv_now;
501 547
531 577
532 X_LOCK (reqlock); 578 X_LOCK (reqlock);
533 --nreqs; 579 --nreqs;
534 X_UNLOCK (reqlock); 580 X_UNLOCK (reqlock);
535 581
536 if (expect_false (req->type == EIO_GROUP && req->size)) 582 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
537 { 583 {
538 req->int1 = 1; /* mark request as delayed */ 584 req->int1 = 1; /* mark request as delayed */
539 continue; 585 continue;
540 } 586 }
541 else 587 else
542 { 588 {
543 int res = ETP_FINISH (req); 589 int res = ETP_FINISH (req);
544 if (expect_false (res)) 590 if (ecb_expect_false (res))
545 return res; 591 return res;
546 } 592 }
547 593
548 if (expect_false (maxreqs && !--maxreqs)) 594 if (ecb_expect_false (maxreqs && !--maxreqs))
549 break; 595 break;
550 596
551 if (maxtime) 597 if (maxtime)
552 { 598 {
553 gettimeofday (&tv_now, 0); 599 gettimeofday (&tv_now, 0);
559 605
560 errno = EAGAIN; 606 errno = EAGAIN;
561 return -1; 607 return -1;
562} 608}
563 609
610static void
564static void etp_cancel (ETP_REQ *req) 611etp_cancel (ETP_REQ *req)
565{ 612{
566 X_LOCK (wrklock); 613 req->cancelled = 1;
567 req->flags |= EIO_FLAG_CANCELLED;
568 X_UNLOCK (wrklock);
569 614
570 eio_grp_cancel (req); 615 eio_grp_cancel (req);
571} 616}
572 617
618static void
573static void etp_submit (ETP_REQ *req) 619etp_submit (ETP_REQ *req)
574{ 620{
575 req->pri -= ETP_PRI_MIN; 621 req->pri -= ETP_PRI_MIN;
576 622
577 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 623 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
578 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 624 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
579 625
580 if (expect_false (req->type == EIO_GROUP)) 626 if (ecb_expect_false (req->type == EIO_GROUP))
581 { 627 {
582 /* I hope this is worth it :/ */ 628 /* I hope this is worth it :/ */
583 X_LOCK (reqlock); 629 X_LOCK (reqlock);
584 ++nreqs; 630 ++nreqs;
585 X_UNLOCK (reqlock); 631 X_UNLOCK (reqlock);
604 650
605 etp_maybe_start_thread (); 651 etp_maybe_start_thread ();
606 } 652 }
607} 653}
608 654
655static void ecb_cold
609static void etp_set_max_poll_time (double nseconds) 656etp_set_max_poll_time (double nseconds)
610{ 657{
611 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 658 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
612 max_poll_time = nseconds * EIO_TICKS; 659 max_poll_time = nseconds * EIO_TICKS;
613 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 660 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
614} 661}
615 662
663static void ecb_cold
616static void etp_set_max_poll_reqs (unsigned int maxreqs) 664etp_set_max_poll_reqs (unsigned int maxreqs)
617{ 665{
618 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 666 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
619 max_poll_reqs = maxreqs; 667 max_poll_reqs = maxreqs;
620 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 668 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
621} 669}
622 670
671static void ecb_cold
623static void etp_set_max_idle (unsigned int nthreads) 672etp_set_max_idle (unsigned int nthreads)
624{ 673{
625 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 674 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
626 max_idle = nthreads <= 0 ? 1 : nthreads; 675 max_idle = nthreads;
627 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 676 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
628} 677}
629 678
679static void ecb_cold
680etp_set_idle_timeout (unsigned int seconds)
681{
682 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
683 idle_timeout = seconds;
684 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
685}
686
687static void ecb_cold
630static void etp_set_min_parallel (unsigned int nthreads) 688etp_set_min_parallel (unsigned int nthreads)
631{ 689{
632 if (wanted < nthreads) 690 if (wanted < nthreads)
633 wanted = nthreads; 691 wanted = nthreads;
634} 692}
635 693
694static void ecb_cold
636static void etp_set_max_parallel (unsigned int nthreads) 695etp_set_max_parallel (unsigned int nthreads)
637{ 696{
638 if (wanted > nthreads) 697 if (wanted > nthreads)
639 wanted = nthreads; 698 wanted = nthreads;
640 699
641 while (started > wanted) 700 while (started > wanted)
642 etp_end_thread (); 701 etp_end_thread ();
643} 702}
644 703
645/*****************************************************************************/ 704/*****************************************************************************/
646 705
706static void
647static void grp_try_feed (eio_req *grp) 707grp_try_feed (eio_req *grp)
648{ 708{
649 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 709 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
650 { 710 {
651 grp->flags &= ~EIO_FLAG_GROUPADD; 711 grp->flags &= ~EIO_FLAG_GROUPADD;
652 712
659 break; 719 break;
660 } 720 }
661 } 721 }
662} 722}
663 723
724static int
664static int grp_dec (eio_req *grp) 725grp_dec (eio_req *grp)
665{ 726{
666 --grp->size; 727 --grp->size;
667 728
668 /* call feeder, if applicable */ 729 /* call feeder, if applicable */
669 grp_try_feed (grp); 730 grp_try_feed (grp);
673 return eio_finish (grp); 734 return eio_finish (grp);
674 else 735 else
675 return 0; 736 return 0;
676} 737}
677 738
739static void
678void eio_destroy (eio_req *req) 740eio_destroy (eio_req *req)
679{ 741{
680 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 742 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
681 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 743 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
682 744
683 EIO_DESTROY (req); 745 EIO_DESTROY (req);
684} 746}
685 747
748static int
686static int eio_finish (eio_req *req) 749eio_finish (eio_req *req)
687{ 750{
688 int res = EIO_FINISH (req); 751 int res = EIO_FINISH (req);
689 752
690 if (req->grp) 753 if (req->grp)
691 { 754 {
699 if (grp->grp_first == req) 762 if (grp->grp_first == req)
700 grp->grp_first = req->grp_next; 763 grp->grp_first = req->grp_next;
701 764
702 res2 = grp_dec (grp); 765 res2 = grp_dec (grp);
703 766
704 if (!res && res2) 767 if (!res)
705 res = res2; 768 res = res2;
706 } 769 }
707 770
708 eio_destroy (req); 771 eio_destroy (req);
709 772
710 return res; 773 return res;
711} 774}
712 775
776void
713void eio_grp_cancel (eio_req *grp) 777eio_grp_cancel (eio_req *grp)
714{ 778{
715 for (grp = grp->grp_first; grp; grp = grp->grp_next) 779 for (grp = grp->grp_first; grp; grp = grp->grp_next)
716 eio_cancel (grp); 780 eio_cancel (grp);
717} 781}
718 782
783void
719void eio_cancel (eio_req *req) 784eio_cancel (eio_req *req)
720{ 785{
721 etp_cancel (req); 786 etp_cancel (req);
722} 787}
723 788
789void
724void eio_submit (eio_req *req) 790eio_submit (eio_req *req)
725{ 791{
726 etp_submit (req); 792 etp_submit (req);
727} 793}
728 794
729unsigned int eio_nreqs (void) 795unsigned int
796eio_nreqs (void)
730{ 797{
731 return etp_nreqs (); 798 return etp_nreqs ();
732} 799}
733 800
734unsigned int eio_nready (void) 801unsigned int
802eio_nready (void)
735{ 803{
736 return etp_nready (); 804 return etp_nready ();
737} 805}
738 806
739unsigned int eio_npending (void) 807unsigned int
808eio_npending (void)
740{ 809{
741 return etp_npending (); 810 return etp_npending ();
742} 811}
743 812
744unsigned int eio_nthreads (void) 813unsigned int ecb_cold
814eio_nthreads (void)
745{ 815{
746 return etp_nthreads (); 816 return etp_nthreads ();
747} 817}
748 818
819void ecb_cold
749void eio_set_max_poll_time (double nseconds) 820eio_set_max_poll_time (double nseconds)
750{ 821{
751 etp_set_max_poll_time (nseconds); 822 etp_set_max_poll_time (nseconds);
752} 823}
753 824
825void ecb_cold
754void eio_set_max_poll_reqs (unsigned int maxreqs) 826eio_set_max_poll_reqs (unsigned int maxreqs)
755{ 827{
756 etp_set_max_poll_reqs (maxreqs); 828 etp_set_max_poll_reqs (maxreqs);
757} 829}
758 830
831void ecb_cold
759void eio_set_max_idle (unsigned int nthreads) 832eio_set_max_idle (unsigned int nthreads)
760{ 833{
761 etp_set_max_idle (nthreads); 834 etp_set_max_idle (nthreads);
762} 835}
763 836
837void ecb_cold
838eio_set_idle_timeout (unsigned int seconds)
839{
840 etp_set_idle_timeout (seconds);
841}
842
843void ecb_cold
764void eio_set_min_parallel (unsigned int nthreads) 844eio_set_min_parallel (unsigned int nthreads)
765{ 845{
766 etp_set_min_parallel (nthreads); 846 etp_set_min_parallel (nthreads);
767} 847}
768 848
849void ecb_cold
769void eio_set_max_parallel (unsigned int nthreads) 850eio_set_max_parallel (unsigned int nthreads)
770{ 851{
771 etp_set_max_parallel (nthreads); 852 etp_set_max_parallel (nthreads);
772} 853}
773 854
774int eio_poll (void) 855int eio_poll (void)
783# undef pread 864# undef pread
784# undef pwrite 865# undef pwrite
785# define pread eio__pread 866# define pread eio__pread
786# define pwrite eio__pwrite 867# define pwrite eio__pwrite
787 868
788static ssize_t 869static eio_ssize_t
789eio__pread (int fd, void *buf, size_t count, off_t offset) 870eio__pread (int fd, void *buf, size_t count, off_t offset)
790{ 871{
791 ssize_t res; 872 eio_ssize_t res;
792 off_t ooffset; 873 off_t ooffset;
793 874
794 X_LOCK (preadwritelock); 875 X_LOCK (preadwritelock);
795 ooffset = lseek (fd, 0, SEEK_CUR); 876 ooffset = lseek (fd, 0, SEEK_CUR);
796 lseek (fd, offset, SEEK_SET); 877 lseek (fd, offset, SEEK_SET);
799 X_UNLOCK (preadwritelock); 880 X_UNLOCK (preadwritelock);
800 881
801 return res; 882 return res;
802} 883}
803 884
804static ssize_t 885static eio_ssize_t
805eio__pwrite (int fd, void *buf, size_t count, off_t offset) 886eio__pwrite (int fd, void *buf, size_t count, off_t offset)
806{ 887{
807 ssize_t res; 888 eio_ssize_t res;
808 off_t ooffset; 889 off_t ooffset;
809 890
810 X_LOCK (preadwritelock); 891 X_LOCK (preadwritelock);
811 ooffset = lseek (fd, 0, SEEK_CUR); 892 ooffset = lseek (fd, 0, SEEK_CUR);
812 lseek (fd, offset, SEEK_SET); 893 lseek (fd, offset, SEEK_SET);
816 897
817 return res; 898 return res;
818} 899}
819#endif 900#endif
820 901
821#ifndef HAVE_FUTIMES 902#ifndef HAVE_UTIMES
822 903
823# undef utimes 904# undef utimes
824# undef futimes
825# define utimes(path,times) eio__utimes (path, times) 905# define utimes(path,times) eio__utimes (path, times)
826# define futimes(fd,times) eio__futimes (fd, times)
827 906
828static int 907static int
829eio__utimes (const char *filename, const struct timeval times[2]) 908eio__utimes (const char *filename, const struct timeval times[2])
830{ 909{
831 if (times) 910 if (times)
839 } 918 }
840 else 919 else
841 return utime (filename, 0); 920 return utime (filename, 0);
842} 921}
843 922
923#endif
924
925#ifndef HAVE_FUTIMES
926
927# undef futimes
928# define futimes(fd,times) eio__futimes (fd, times)
929
930static int
844static int eio__futimes (int fd, const struct timeval tv[2]) 931eio__futimes (int fd, const struct timeval tv[2])
845{ 932{
846 errno = ENOSYS; 933 errno = ENOSYS;
847 return -1; 934 return -1;
848} 935}
849 936
853# undef fdatasync 940# undef fdatasync
854# define fdatasync(fd) fsync (fd) 941# define fdatasync(fd) fsync (fd)
855#endif 942#endif
856 943
857/* sync_file_range always needs emulation */ 944/* sync_file_range always needs emulation */
858int 945static int
859eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 946eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
860{ 947{
861#if HAVE_SYNC_FILE_RANGE 948#if HAVE_SYNC_FILE_RANGE
862 int res; 949 int res;
863 950
880 /* even though we could play tricks with the flags, it's better to always 967 /* even though we could play tricks with the flags, it's better to always
881 * call fdatasync, as that matches the expectation of its users best */ 968 * call fdatasync, as that matches the expectation of its users best */
882 return fdatasync (fd); 969 return fdatasync (fd);
883} 970}
884 971
972static int
973eio__fallocate (int fd, int mode, off_t offset, size_t len)
974{
975#if HAVE_FALLOCATE
976 return fallocate (fd, mode, offset, len);
977#else
978 errno = ENOSYS;
979 return -1;
980#endif
981}
982
885#if !HAVE_READAHEAD 983#if !HAVE_READAHEAD
886# undef readahead 984# undef readahead
887# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 985# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
888 986
889static ssize_t 987static eio_ssize_t
890eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 988eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
891{ 989{
892 size_t todo = count; 990 size_t todo = count;
893 dBUF; 991 dBUF;
894 992
906} 1004}
907 1005
908#endif 1006#endif
909 1007
910/* sendfile always needs emulation */ 1008/* sendfile always needs emulation */
911static ssize_t 1009static eio_ssize_t
912eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1010eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
913{ 1011{
1012 eio_ssize_t written = 0;
914 ssize_t res; 1013 eio_ssize_t res;
915 1014
916 if (!count) 1015 if (!count)
917 return 0; 1016 return 0;
918 1017
1018 for (;;)
1019 {
1020#ifdef __APPLE__
1021# undef HAVE_SENDFILE /* broken, as everything on os x */
1022#endif
919#if HAVE_SENDFILE 1023#if HAVE_SENDFILE
920# if __linux 1024# if __linux
1025 off_t soffset = offset;
921 res = sendfile (ofd, ifd, &offset, count); 1026 res = sendfile (ofd, ifd, &soffset, count);
922 1027
923# elif __FreeBSD__ 1028# elif __FreeBSD__
924 /* 1029 /*
925 * Of course, the freebsd sendfile is a dire hack with no thoughts 1030 * Of course, the freebsd sendfile is a dire hack with no thoughts
926 * wasted on making it similar to other I/O functions. 1031 * wasted on making it similar to other I/O functions.
927 */ 1032 */
928 {
929 off_t sbytes; 1033 off_t sbytes;
930 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1034 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
931 1035
932 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1036 #if 0 /* according to the manpage, this is correct, but broken behaviour */
933 /* freebsd' sendfile will return 0 on success */ 1037 /* freebsd' sendfile will return 0 on success */
934 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1038 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
935 /* not on e.g. EIO or EPIPE - sounds broken */ 1039 /* not on e.g. EIO or EPIPE - sounds broken */
936 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1040 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
937 res = sbytes; 1041 res = sbytes;
938 #endif 1042 #endif
939 1043
940 /* according to source inspection, this is correct, and useful behaviour */ 1044 /* according to source inspection, this is correct, and useful behaviour */
941 if (sbytes) 1045 if (sbytes)
942 res = sbytes; 1046 res = sbytes;
943 }
944 1047
945# elif defined (__APPLE__) 1048# elif defined (__APPLE__)
946
947 {
948 off_t sbytes = count; 1049 off_t sbytes = count;
949 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1050 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
950 1051
951 /* according to the manpage, sbytes is always valid */ 1052 /* according to the manpage, sbytes is always valid */
952 if (sbytes) 1053 if (sbytes)
953 res = sbytes; 1054 res = sbytes;
954 }
955 1055
956# elif __hpux 1056# elif __hpux
957 res = sendfile (ofd, ifd, offset, count, 0, 0); 1057 res = sendfile (ofd, ifd, offset, count, 0, 0);
958 1058
959# elif __solaris 1059# elif __solaris
960 {
961 struct sendfilevec vec; 1060 struct sendfilevec vec;
962 size_t sbytes; 1061 size_t sbytes;
963 1062
964 vec.sfv_fd = ifd; 1063 vec.sfv_fd = ifd;
965 vec.sfv_flag = 0; 1064 vec.sfv_flag = 0;
966 vec.sfv_off = offset; 1065 vec.sfv_off = offset;
967 vec.sfv_len = count; 1066 vec.sfv_len = count;
968 1067
969 res = sendfilev (ofd, &vec, 1, &sbytes); 1068 res = sendfilev (ofd, &vec, 1, &sbytes);
970 1069
971 if (res < 0 && sbytes) 1070 if (res < 0 && sbytes)
972 res = sbytes; 1071 res = sbytes;
973 }
974 1072
975# endif 1073# endif
976 1074
977#elif defined (_WIN32) 1075#elif defined (_WIN32) && 0
978
979 /* does not work, just for documentation of what would need to be done */ 1076 /* does not work, just for documentation of what would need to be done */
980 { 1077 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1078 /* libeio guarantees that the file offset does not change, and windows */
1079 /* has no way to get an independent handle to the same file description */
981 HANDLE h = TO_SOCKET (ifd); 1080 HANDLE h = TO_SOCKET (ifd);
982 SetFilePointer (h, offset, 0, FILE_BEGIN); 1081 SetFilePointer (h, offset, 0, FILE_BEGIN);
983 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1082 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
984 }
985 1083
986#else 1084#else
987 res = -1; 1085 res = -1;
988 errno = ENOSYS; 1086 errno = ENOSYS;
989#endif 1087#endif
990 1088
1089 /* we assume sendfile can copy at least 128mb in one go */
1090 if (res <= 128 * 1024 * 1024)
1091 {
1092 if (res > 0)
1093 written += res;
1094
1095 if (written)
1096 return written;
1097
1098 break;
1099 }
1100 else
1101 {
1102 /* if we requested more, then probably the kernel was lazy */
1103 written += res;
1104 offset += res;
1105 count -= res;
1106
1107 if (!count)
1108 return written;
1109 }
1110 }
1111
991 if (res < 0 1112 if (res < 0
992 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1113 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
993 /* BSDs */ 1114 /* BSDs */
994#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1115#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
995 || errno == ENOTSUP 1116 || errno == ENOTSUP
996#endif 1117#endif
1118#ifdef EOPNOTSUPP /* windows */
997 || errno == EOPNOTSUPP /* BSDs */ 1119 || errno == EOPNOTSUPP /* BSDs */
1120#endif
998#if __solaris 1121#if __solaris
999 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1122 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1000#endif 1123#endif
1001 ) 1124 )
1002 ) 1125 )
1006 1129
1007 res = 0; 1130 res = 0;
1008 1131
1009 while (count) 1132 while (count)
1010 { 1133 {
1011 ssize_t cnt; 1134 eio_ssize_t cnt;
1012 1135
1013 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1136 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1014 1137
1015 if (cnt <= 0) 1138 if (cnt <= 0)
1016 { 1139 {
1033 } 1156 }
1034 1157
1035 return res; 1158 return res;
1036} 1159}
1037 1160
1161#ifdef PAGESIZE
1162# define eio_pagesize() PAGESIZE
1163#else
1164static intptr_t
1165eio_pagesize (void)
1166{
1167 static intptr_t page;
1168
1169 if (!page)
1170 page = sysconf (_SC_PAGESIZE);
1171
1172 return page;
1173}
1174#endif
1175
1176static void
1177eio_page_align (void **addr, size_t *length)
1178{
1179 intptr_t mask = eio_pagesize () - 1;
1180
1181 /* round down addr */
1182 intptr_t adj = mask & (intptr_t)*addr;
1183
1184 *addr = (void *)((intptr_t)*addr - adj);
1185 *length += adj;
1186
1187 /* round up length */
1188 *length = (*length + mask) & ~mask;
1189}
1190
1191#if !_POSIX_MEMLOCK
1192# define eio__mlockall(a) EIO_ENOSYS ()
1193#else
1194
1195static int
1196eio__mlockall (int flags)
1197{
1198 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1199 extern int mallopt (int, int);
1200 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1201 #endif
1202
1203 if (EIO_MCL_CURRENT != MCL_CURRENT
1204 || EIO_MCL_FUTURE != MCL_FUTURE)
1205 {
1206 flags = 0
1207 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1208 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1209 }
1210
1211 return mlockall (flags);
1212}
1213#endif
1214
1215#if !_POSIX_MEMLOCK_RANGE
1216# define eio__mlock(a,b) EIO_ENOSYS ()
1217#else
1218
1219static int
1220eio__mlock (void *addr, size_t length)
1221{
1222 eio_page_align (&addr, &length);
1223
1224 return mlock (addr, length);
1225}
1226
1227#endif
1228
1229#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1230# define eio__msync(a,b,c) EIO_ENOSYS ()
1231#else
1232
1233static int
1234eio__msync (void *mem, size_t len, int flags)
1235{
1236 eio_page_align (&mem, &len);
1237
1238 if (EIO_MS_ASYNC != MS_SYNC
1239 || EIO_MS_INVALIDATE != MS_INVALIDATE
1240 || EIO_MS_SYNC != MS_SYNC)
1241 {
1242 flags = 0
1243 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1244 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1245 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1246 }
1247
1248 return msync (mem, len, flags);
1249}
1250
1251#endif
1252
1253static int
1254eio__mtouch (eio_req *req)
1255{
1256 void *mem = req->ptr2;
1257 size_t len = req->size;
1258 int flags = req->int1;
1259
1260 eio_page_align (&mem, &len);
1261
1262 {
1263 intptr_t addr = (intptr_t)mem;
1264 intptr_t end = addr + len;
1265 intptr_t page = eio_pagesize ();
1266
1267 if (addr < end)
1268 if (flags & EIO_MT_MODIFY) /* modify */
1269 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1270 else
1271 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1272 }
1273
1274 return 0;
1275}
1276
1277/*****************************************************************************/
1278/* requests implemented outside eio_execute, because they are so large */
1279
1280static void
1281eio__realpath (eio_req *req, etp_worker *self)
1282{
1283 char *rel = req->ptr1;
1284 char *res;
1285 char *tmp1, *tmp2;
1286#if SYMLOOP_MAX > 32
1287 int symlinks = SYMLOOP_MAX;
1288#else
1289 int symlinks = 32;
1290#endif
1291
1292 req->result = -1;
1293
1294 errno = EINVAL;
1295 if (!rel)
1296 return;
1297
1298 errno = ENOENT;
1299 if (!*rel)
1300 return;
1301
1302 if (!req->ptr2)
1303 {
1304 X_LOCK (wrklock);
1305 req->flags |= EIO_FLAG_PTR2_FREE;
1306 X_UNLOCK (wrklock);
1307 req->ptr2 = malloc (PATH_MAX * 3);
1308
1309 errno = ENOMEM;
1310 if (!req->ptr2)
1311 return;
1312 }
1313
1314 res = req->ptr2;
1315 tmp1 = res + PATH_MAX;
1316 tmp2 = tmp1 + PATH_MAX;
1317
1318#if 0 /* disabled, the musl way to do things is just too racy */
1319#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1320 /* on linux we may be able to ask the kernel */
1321 {
1322 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1323
1324 if (fd >= 0)
1325 {
1326 sprintf (tmp1, "/proc/self/fd/%d", fd);
1327 req->result = readlink (tmp1, res, PATH_MAX);
1328 close (fd);
1329
1330 /* here we should probably stat the open file and the disk file, to make sure they still match */
1331
1332 if (req->result > 0)
1333 goto done;
1334 }
1335 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1336 return;
1337 }
1338#endif
1339#endif
1340
1341 if (*rel != '/')
1342 {
1343 if (!getcwd (res, PATH_MAX))
1344 return;
1345
1346 if (res [1]) /* only use if not / */
1347 res += strlen (res);
1348 }
1349
1350 while (*rel)
1351 {
1352 eio_ssize_t len, linklen;
1353 char *beg = rel;
1354
1355 while (*rel && *rel != '/')
1356 ++rel;
1357
1358 len = rel - beg;
1359
1360 if (!len) /* skip slashes */
1361 {
1362 ++rel;
1363 continue;
1364 }
1365
1366 if (beg [0] == '.')
1367 {
1368 if (len == 1)
1369 continue; /* . - nop */
1370
1371 if (beg [1] == '.' && len == 2)
1372 {
1373 /* .. - back up one component, if possible */
1374
1375 while (res != req->ptr2)
1376 if (*--res == '/')
1377 break;
1378
1379 continue;
1380 }
1381 }
1382
1383 errno = ENAMETOOLONG;
1384 if (res + 1 + len + 1 >= tmp1)
1385 return;
1386
1387 /* copy one component */
1388 *res = '/';
1389 memcpy (res + 1, beg, len);
1390
1391 /* zero-terminate, for readlink */
1392 res [len + 1] = 0;
1393
1394 /* now check if it's a symlink */
1395 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1396
1397 if (linklen < 0)
1398 {
1399 if (errno != EINVAL)
1400 return;
1401
1402 /* it's a normal directory. hopefully */
1403 res += len + 1;
1404 }
1405 else
1406 {
1407 /* yay, it was a symlink - build new path in tmp2 */
1408 int rellen = strlen (rel);
1409
1410 errno = ENAMETOOLONG;
1411 if (linklen + 1 + rellen >= PATH_MAX)
1412 return;
1413
1414 errno = ELOOP;
1415 if (!--symlinks)
1416 return;
1417
1418 if (*tmp1 == '/')
1419 res = req->ptr2; /* symlink resolves to an absolute path */
1420
1421 /* we need to be careful, as rel might point into tmp2 already */
1422 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1423 tmp2 [linklen] = '/';
1424 memcpy (tmp2, tmp1, linklen);
1425
1426 rel = tmp2;
1427 }
1428 }
1429
1430 /* special case for the lone root path */
1431 if (res == req->ptr2)
1432 *res++ = '/';
1433
1434 req->result = res - (char *)req->ptr2;
1435
1436done:
1437 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1438}
1439
1038static signed char 1440static signed char
1039eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1441eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1040{ 1442{
1041 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1443 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1042 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1444 : a->inode < b->inode ? -1
1445 : a->inode > b->inode ? 1
1446 : 0;
1043} 1447}
1044 1448
1045#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1449#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1046 1450
1047#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1451#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1048#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1452#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1049 1453
1050static void 1454static void
1051eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1455eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1052{ 1456{
1053 unsigned char bits [9 + sizeof (ino_t) * 8]; 1457 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1054 unsigned char *bit = bits; 1458 unsigned char *bit = bits;
1055 1459
1056 assert (CHAR_BIT == 8); 1460 assert (CHAR_BIT == 8);
1057 assert (sizeof (eio_dirent) * 8 < 256); 1461 assert (sizeof (eio_dirent) * 8 < 256);
1058 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1462 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1059 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1463 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1060 1464
1061 if (size <= EIO_SORT_FAST) 1465 if (size <= EIO_SORT_FAST)
1062 return; 1466 return;
1063 1467
1064 /* first prepare an array of bits to test in our radix sort */ 1468 /* first prepare an array of bits to test in our radix sort */
1065 /* try to take endianness into account, as well as differences in ino_t sizes */ 1469 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1066 /* inode_bits must contain all inodes ORed together */ 1470 /* inode_bits must contain all inodes ORed together */
1067 /* which is used to skip bits that are 0 everywhere, which is very common */ 1471 /* which is used to skip bits that are 0 everywhere, which is very common */
1068 { 1472 {
1069 ino_t endianness; 1473 eio_ino_t endianness;
1070 int i, j; 1474 int i, j;
1071 1475
1072 /* we store the byte offset of byte n into byte n of "endianness" */ 1476 /* we store the byte offset of byte n into byte n of "endianness" */
1073 for (i = 0; i < sizeof (ino_t); ++i) 1477 for (i = 0; i < sizeof (eio_ino_t); ++i)
1074 ((unsigned char *)&endianness)[i] = i; 1478 ((unsigned char *)&endianness)[i] = i;
1075 1479
1076 *bit++ = 0; 1480 *bit++ = 0;
1077 1481
1078 for (i = 0; i < sizeof (ino_t); ++i) 1482 for (i = 0; i < sizeof (eio_ino_t); ++i)
1079 { 1483 {
1080 /* shifting off the byte offsets out of "endianness" */ 1484 /* shifting off the byte offsets out of "endianness" */
1081 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1485 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1082 endianness >>= 8; 1486 endianness >>= 8;
1083 1487
1084 for (j = 0; j < 8; ++j) 1488 for (j = 0; j < 8; ++j)
1085 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1489 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1086 *bit++ = offs + j; 1490 *bit++ = offs + j;
1087 } 1491 }
1088 1492
1089 for (j = 0; j < 8; ++j) 1493 for (j = 0; j < 8; ++j)
1090 if (score_bits & (1 << j)) 1494 if (score_bits & (1 << j))
1091 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1495 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1092 } 1496 }
1093 1497
1094 /* now actually do the sorting (a variant of MSD radix sort) */ 1498 /* now actually do the sorting (a variant of MSD radix sort) */
1095 { 1499 {
1096 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1500 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1097 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1501 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1098 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1502 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1099 int stk_idx = 0; 1503 int stk_idx = 0;
1100 1504
1101 base_stk [stk_idx] = dents; 1505 base_stk [stk_idx] = dents;
1102 end_stk [stk_idx] = dents + size; 1506 end_stk [stk_idx] = dents + size;
1103 bit_stk [stk_idx] = bit - 1; 1507 bit_stk [stk_idx] = bit - 1;
1182 } 1586 }
1183 } 1587 }
1184} 1588}
1185 1589
1186static void 1590static void
1187eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1591eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1188{ 1592{
1189 if (size <= 1) 1593 if (size <= 1)
1190 return; /* our insertion sort relies on size > 0 */ 1594 return; /* our insertion sort relies on size > 0 */
1191 1595
1192 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1596 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1200 1604
1201/* read a full directory */ 1605/* read a full directory */
1202static void 1606static void
1203eio__scandir (eio_req *req, etp_worker *self) 1607eio__scandir (eio_req *req, etp_worker *self)
1204{ 1608{
1205 DIR *dirp;
1206 EIO_STRUCT_DIRENT *entp;
1207 char *name, *names; 1609 char *name, *names;
1208 int namesalloc = 4096; 1610 int namesalloc = 4096 - sizeof (void *) * 4;
1209 int namesoffs = 0; 1611 int namesoffs = 0;
1210 int flags = req->int1; 1612 int flags = req->int1;
1211 eio_dirent *dents = 0; 1613 eio_dirent *dents = 0;
1212 int dentalloc = 128; 1614 int dentalloc = 128;
1213 int dentoffs = 0; 1615 int dentoffs = 0;
1214 ino_t inode_bits = 0; 1616 eio_ino_t inode_bits = 0;
1617#ifdef _WIN32
1618 HANDLE dirp;
1619 WIN32_FIND_DATA entp;
1620#else
1621 DIR *dirp;
1622 EIO_STRUCT_DIRENT *entp;
1623#endif
1215 1624
1216 req->result = -1; 1625 req->result = -1;
1217 1626
1218 if (!(flags & EIO_READDIR_DENTS)) 1627 if (!(flags & EIO_READDIR_DENTS))
1219 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1628 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1220 1629
1221 X_LOCK (wrklock); 1630#ifdef _WIN32
1222 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1631 {
1632 int len = strlen ((const char *)req->ptr1);
1633 char *path = malloc (MAX_PATH);
1634 const char *fmt;
1635
1636 if (!len)
1637 fmt = "./*";
1638 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1639 fmt = "%s*";
1640 else
1641 fmt = "%s/*";
1642
1643 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1644 dirp = FindFirstFile (path, &entp);
1645 free (path);
1646
1647 if (dirp == INVALID_HANDLE_VALUE)
1648 {
1649 dirp = 0;
1650
1651 switch (GetLastError ())
1652 {
1653 case ERROR_FILE_NOT_FOUND:
1654 req->result = 0;
1655 break;
1656
1657 case ERROR_INVALID_NAME:
1658 case ERROR_PATH_NOT_FOUND:
1659 case ERROR_NO_MORE_FILES:
1660 errno = ENOENT;
1661 break;
1662
1663 case ERROR_NOT_ENOUGH_MEMORY:
1664 errno = ENOMEM;
1665 break;
1666
1667 default:
1668 errno = EINVAL;
1669 break;
1670 }
1671 }
1672 }
1673#else
1223 self->dirp = dirp = opendir (req->ptr1); 1674 dirp = opendir (req->ptr1);
1675#endif
1676
1677 if (req->flags & EIO_FLAG_PTR1_FREE)
1678 free (req->ptr1);
1679
1224 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1680 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1225 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1681 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1226 req->ptr2 = names = malloc (namesalloc); 1682 req->ptr2 = names = malloc (namesalloc);
1227 X_UNLOCK (wrklock);
1228 1683
1229 if (dirp && names && (!flags || dents)) 1684 if (dirp && names && (!flags || dents))
1230 for (;;) 1685 for (;;)
1231 { 1686 {
1687 int done;
1688
1689#ifdef _WIN32
1690 done = !dirp;
1691#else
1232 errno = 0; 1692 errno = 0;
1233 entp = readdir (dirp); 1693 entp = readdir (dirp);
1694 done = !entp;
1695#endif
1234 1696
1235 if (!entp) 1697 if (done)
1236 { 1698 {
1699#ifndef _WIN32
1700 int old_errno = errno;
1701 closedir (dirp);
1702 errno = old_errno;
1703
1237 if (errno) 1704 if (errno)
1238 break; 1705 break;
1706#endif
1239 1707
1240 /* sort etc. */ 1708 /* sort etc. */
1241 req->int1 = flags; 1709 req->int1 = flags;
1242 req->result = dentoffs; 1710 req->result = dentoffs;
1243 1711
1244 if (flags & EIO_READDIR_STAT_ORDER) 1712 if (flags & EIO_READDIR_STAT_ORDER)
1245 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1713 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1246 else if (flags & EIO_READDIR_DIRS_FIRST) 1714 else if (flags & EIO_READDIR_DIRS_FIRST)
1247 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1715 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1248 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1716 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1249 else 1717 else
1250 { 1718 {
1252 eio_dirent *oth = dents + dentoffs; 1720 eio_dirent *oth = dents + dentoffs;
1253 eio_dirent *dir = dents; 1721 eio_dirent *dir = dents;
1254 1722
1255 /* now partition dirs to the front, and non-dirs to the back */ 1723 /* now partition dirs to the front, and non-dirs to the back */
1256 /* by walking from both sides and swapping if necessary */ 1724 /* by walking from both sides and swapping if necessary */
1257 /* also clear score, so it doesn't influence sorting */
1258 while (oth > dir) 1725 while (oth > dir)
1259 { 1726 {
1260 if (dir->type == EIO_DT_DIR) 1727 if (dir->type == EIO_DT_DIR)
1261 ++dir; 1728 ++dir;
1262 else if ((--oth)->type == EIO_DT_DIR) 1729 else if ((--oth)->type == EIO_DT_DIR)
1265 1732
1266 ++dir; 1733 ++dir;
1267 } 1734 }
1268 } 1735 }
1269 1736
1270 /* now sort the dirs only */ 1737 /* now sort the dirs only (dirs all have the same score) */
1271 eio_dent_sort (dents, dir - dents, 0, inode_bits); 1738 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1272 } 1739 }
1273 1740
1274 break; 1741 break;
1275 } 1742 }
1276 1743
1277 /* now add the entry to our list(s) */ 1744 /* now add the entry to our list(s) */
1278 name = entp->d_name; 1745 name = D_NAME (entp);
1279 1746
1280 /* skip . and .. entries */ 1747 /* skip . and .. entries */
1281 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1748 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1282 { 1749 {
1283 int len = D_NAMLEN (entp) + 1; 1750 int len = D_NAMLEN (entp) + 1;
1284 1751
1285 while (expect_false (namesoffs + len > namesalloc)) 1752 while (ecb_expect_false (namesoffs + len > namesalloc))
1286 { 1753 {
1287 namesalloc *= 2; 1754 namesalloc *= 2;
1288 X_LOCK (wrklock);
1289 req->ptr2 = names = realloc (names, namesalloc); 1755 req->ptr2 = names = realloc (names, namesalloc);
1290 X_UNLOCK (wrklock);
1291 1756
1292 if (!names) 1757 if (!names)
1293 break; 1758 break;
1294 } 1759 }
1295 1760
1297 1762
1298 if (dents) 1763 if (dents)
1299 { 1764 {
1300 struct eio_dirent *ent; 1765 struct eio_dirent *ent;
1301 1766
1302 if (expect_false (dentoffs == dentalloc)) 1767 if (ecb_expect_false (dentoffs == dentalloc))
1303 { 1768 {
1304 dentalloc *= 2; 1769 dentalloc *= 2;
1305 X_LOCK (wrklock);
1306 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1770 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1307 X_UNLOCK (wrklock);
1308 1771
1309 if (!dents) 1772 if (!dents)
1310 break; 1773 break;
1311 } 1774 }
1312 1775
1392 if (EIO_CANCELLED (req)) 1855 if (EIO_CANCELLED (req))
1393 { 1856 {
1394 errno = ECANCELED; 1857 errno = ECANCELED;
1395 break; 1858 break;
1396 } 1859 }
1860
1861#ifdef _WIN32
1862 if (!FindNextFile (dirp, &entp))
1863 {
1864 FindClose (dirp);
1865 dirp = 0;
1866 }
1867#endif
1397 } 1868 }
1398}
1399
1400#ifdef PAGESIZE
1401# define eio_pagesize() PAGESIZE
1402#else
1403static intptr_t
1404eio_pagesize (void)
1405{
1406 static intptr_t page;
1407
1408 if (!page)
1409 page = sysconf (_SC_PAGESIZE);
1410
1411 return page;
1412}
1413#endif
1414
1415static void
1416eio_page_align (void **addr, size_t *length)
1417{
1418 intptr_t mask = eio_pagesize () - 1;
1419
1420 /* round down addr */
1421 intptr_t adj = mask & (intptr_t)*addr;
1422
1423 *addr = (void *)((intptr_t)*addr - adj);
1424 *length += adj;
1425
1426 /* round up length */
1427 *length = (*length + mask) & ~mask;
1428}
1429
1430#if !_POSIX_MEMLOCK
1431# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1432# define eio__mlockall(a) ((errno = ENOSYS), -1)
1433#else
1434
1435static int
1436eio__mlock (void *addr, size_t length)
1437{
1438 eio_page_align (&addr, &length);
1439
1440 mlock (addr, length);
1441}
1442
1443static int
1444eio__mlockall (int flags)
1445{
1446 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1447 extern int mallopt (int, int);
1448 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1449 #endif
1450
1451 if (EIO_MCL_CURRENT != MCL_CURRENT
1452 || EIO_MCL_FUTURE != MCL_FUTURE)
1453 {
1454 flags = 0
1455 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1456 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1457 }
1458
1459 mlockall (flags);
1460}
1461#endif
1462
1463#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1464# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1465#else
1466
1467int
1468eio__msync (void *mem, size_t len, int flags)
1469{
1470 eio_page_align (&mem, &len);
1471
1472 if (EIO_MS_ASYNC != MS_SYNC
1473 || EIO_MS_INVALIDATE != MS_INVALIDATE
1474 || EIO_MS_SYNC != MS_SYNC)
1475 {
1476 flags = 0
1477 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1478 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1479 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1480 }
1481
1482 return msync (mem, len, flags);
1483}
1484
1485#endif
1486
1487int
1488eio__mtouch (void *mem, size_t len, int flags)
1489{
1490 eio_page_align (&mem, &len);
1491
1492 {
1493 intptr_t addr = (intptr_t)mem;
1494 intptr_t end = addr + len;
1495 intptr_t page = eio_pagesize ();
1496
1497 if (addr < end)
1498 if (flags & EIO_MT_MODIFY) /* modify */
1499 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1500 else
1501 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1502 }
1503
1504 return 0;
1505} 1869}
1506 1870
1507/*****************************************************************************/ 1871/*****************************************************************************/
1508 1872
1509#define ALLOC(len) \ 1873#define ALLOC(len) \
1525{ 1889{
1526 ETP_REQ *req; 1890 ETP_REQ *req;
1527 struct timespec ts; 1891 struct timespec ts;
1528 etp_worker *self = (etp_worker *)thr_arg; 1892 etp_worker *self = (etp_worker *)thr_arg;
1529 1893
1530 /* try to distribute timeouts somewhat randomly */ 1894 /* try to distribute timeouts somewhat evenly */
1531 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1895 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1532 1896
1533 for (;;) 1897 for (;;)
1534 { 1898 {
1899 ts.tv_sec = 0;
1900
1535 X_LOCK (reqlock); 1901 X_LOCK (reqlock);
1536 1902
1537 for (;;) 1903 for (;;)
1538 { 1904 {
1539 self->req = req = reqq_shift (&req_queue); 1905 self->req = req = reqq_shift (&req_queue);
1540 1906
1541 if (req) 1907 if (req)
1542 break; 1908 break;
1543 1909
1910 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1911 {
1912 X_UNLOCK (reqlock);
1913 X_LOCK (wrklock);
1914 --started;
1915 X_UNLOCK (wrklock);
1916 goto quit;
1917 }
1918
1544 ++idle; 1919 ++idle;
1545 1920
1546 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1921 if (idle <= max_idle)
1547 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1922 /* we are allowed to idle, so do so without any timeout */
1923 X_COND_WAIT (reqwait, reqlock);
1924 else
1548 { 1925 {
1549 if (idle > max_idle) 1926 /* initialise timeout once */
1550 { 1927 if (!ts.tv_sec)
1551 --idle; 1928 ts.tv_sec = time (0) + idle_timeout;
1552 X_UNLOCK (reqlock);
1553 X_LOCK (wrklock);
1554 --started;
1555 X_UNLOCK (wrklock);
1556 goto quit;
1557 }
1558 1929
1559 /* we are allowed to idle, so do so without any timeout */
1560 X_COND_WAIT (reqwait, reqlock); 1930 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1931 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1561 } 1932 }
1562 1933
1563 --idle; 1934 --idle;
1564 } 1935 }
1565 1936
1568 X_UNLOCK (reqlock); 1939 X_UNLOCK (reqlock);
1569 1940
1570 if (req->type < 0) 1941 if (req->type < 0)
1571 goto quit; 1942 goto quit;
1572 1943
1573 if (!EIO_CANCELLED (req))
1574 ETP_EXECUTE (self, req); 1944 ETP_EXECUTE (self, req);
1575 1945
1576 X_LOCK (reslock); 1946 X_LOCK (reslock);
1577 1947
1578 ++npending; 1948 ++npending;
1579 1949
1594 return 0; 1964 return 0;
1595} 1965}
1596 1966
1597/*****************************************************************************/ 1967/*****************************************************************************/
1598 1968
1969int ecb_cold
1599int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1970eio_init (void (*want_poll)(void), void (*done_poll)(void))
1600{ 1971{
1972#if !HAVE_PREADWRITE
1973 X_MUTEX_CREATE (preadwritelock);
1974#endif
1975
1601 return etp_init (want_poll, done_poll); 1976 return etp_init (want_poll, done_poll);
1602} 1977}
1603 1978
1979ecb_inline void
1604static void eio_api_destroy (eio_req *req) 1980eio_api_destroy (eio_req *req)
1605{ 1981{
1606 free (req); 1982 free (req);
1607} 1983}
1608 1984
1609#define REQ(rtype) \ 1985#define REQ(rtype) \
1628 { \ 2004 { \
1629 eio_api_destroy (req); \ 2005 eio_api_destroy (req); \
1630 return 0; \ 2006 return 0; \
1631 } 2007 }
1632 2008
2009static void
1633static void eio_execute (etp_worker *self, eio_req *req) 2010eio_execute (etp_worker *self, eio_req *req)
1634{ 2011{
1635 errno = 0; 2012 if (ecb_expect_false (EIO_CANCELLED (req)))
2013 {
2014 req->result = -1;
2015 req->errorno = ECANCELED;
2016 return;
2017 }
1636 2018
1637 switch (req->type) 2019 switch (req->type)
1638 { 2020 {
1639 case EIO_READ: ALLOC (req->size); 2021 case EIO_READ: ALLOC (req->size);
1640 req->result = req->offs >= 0 2022 req->result = req->offs >= 0
1673 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2055 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1674 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2056 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1675 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2057 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1676 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2058 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1677 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2059 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1678 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2060 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2061
2062 case EIO_REALPATH: eio__realpath (req, self); break;
1679 2063
1680 case EIO_READLINK: ALLOC (PATH_MAX); 2064 case EIO_READLINK: ALLOC (PATH_MAX);
1681 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; 2065 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1682 2066
1683 case EIO_SYNC: req->result = 0; sync (); break; 2067 case EIO_SYNC: req->result = 0; sync (); break;
1684 case EIO_FSYNC: req->result = fsync (req->int1); break; 2068 case EIO_FSYNC: req->result = fsync (req->int1); break;
1685 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2069 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1686 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2070 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1687 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2071 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1688 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2072 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1689 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2073 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1690 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2074 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2075 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1691 2076
1692 case EIO_READDIR: eio__scandir (req, self); break; 2077 case EIO_READDIR: eio__scandir (req, self); break;
1693 2078
1694 case EIO_BUSY: 2079 case EIO_BUSY:
1695#ifdef _WIN32 2080#ifdef _WIN32
1696 Sleep (req->nv1 * 1000.); 2081 Sleep (req->nv1 * 1e3);
1697#else 2082#else
1698 { 2083 {
1699 struct timeval tv; 2084 struct timeval tv;
1700 2085
1701 tv.tv_sec = req->nv1; 2086 tv.tv_sec = req->nv1;
1702 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2087 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1703 2088
1704 req->result = select (0, 0, 0, 0, &tv); 2089 req->result = select (0, 0, 0, 0, &tv);
1705 } 2090 }
1706#endif 2091#endif
1707 break; 2092 break;
1722 times = tv; 2107 times = tv;
1723 } 2108 }
1724 else 2109 else
1725 times = 0; 2110 times = 0;
1726 2111
1727
1728 req->result = req->type == EIO_FUTIME 2112 req->result = req->type == EIO_FUTIME
1729 ? futimes (req->int1, times) 2113 ? futimes (req->int1, times)
1730 : utimes (req->ptr1, times); 2114 : utimes (req->ptr1, times);
1731 } 2115 }
1732 break; 2116 break;
1737 case EIO_NOP: 2121 case EIO_NOP:
1738 req->result = 0; 2122 req->result = 0;
1739 break; 2123 break;
1740 2124
1741 case EIO_CUSTOM: 2125 case EIO_CUSTOM:
1742 ((void (*)(eio_req *))req->feed) (req); 2126 req->feed (req);
1743 break; 2127 break;
1744 2128
1745 default: 2129 default:
2130 errno = ENOSYS;
1746 req->result = -1; 2131 req->result = -1;
1747 break; 2132 break;
1748 } 2133 }
1749 2134
1750 req->errorno = errno; 2135 req->errorno = errno;
1795eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2180eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1796{ 2181{
1797 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2182 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1798} 2183}
1799 2184
2185eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2186{
2187 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2188}
2189
1800eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2190eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1801{ 2191{
1802 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2192 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1803} 2193}
1804 2194
1845eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2235eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1846{ 2236{
1847 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2237 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1848} 2238}
1849 2239
1850eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2240eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1851{ 2241{
1852 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2242 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1853} 2243}
1854 2244
1855eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2245eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1875eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2265eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1876{ 2266{
1877 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2267 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1878} 2268}
1879 2269
1880eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2270eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1881{ 2271{
1882 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2272 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1883} 2273}
1884 2274
1885eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2275eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1901eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2291eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1902{ 2292{
1903 return eio__1path (EIO_READLINK, path, pri, cb, data); 2293 return eio__1path (EIO_READLINK, path, pri, cb, data);
1904} 2294}
1905 2295
2296eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2297{
2298 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2299}
2300
1906eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2301eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1907{ 2302{
1908 return eio__1path (EIO_STAT, path, pri, cb, data); 2303 return eio__1path (EIO_STAT, path, pri, cb, data);
1909} 2304}
1910 2305
1933 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2328 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1934} 2329}
1935 2330
1936eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2331eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1937{ 2332{
1938 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2333 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1939} 2334}
1940 2335
1941static eio_req * 2336static eio_req *
1942eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2337eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1943{ 2338{
1967eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2362eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1968{ 2363{
1969 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2364 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1970} 2365}
1971 2366
1972eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2367eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1973{ 2368{
1974 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2369 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1975} 2370}
1976 2371
1977#endif 2372#endif
1978 2373
1979eio_req *eio_grp (eio_cb cb, void *data) 2374eio_req *eio_grp (eio_cb cb, void *data)
1988#undef SEND 2383#undef SEND
1989 2384
1990/*****************************************************************************/ 2385/*****************************************************************************/
1991/* grp functions */ 2386/* grp functions */
1992 2387
2388void
1993void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2389eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1994{ 2390{
1995 grp->int2 = limit; 2391 grp->int2 = limit;
1996 grp->feed = feed; 2392 grp->feed = feed;
1997 2393
1998 grp_try_feed (grp); 2394 grp_try_feed (grp);
1999} 2395}
2000 2396
2397void
2001void eio_grp_limit (eio_req *grp, int limit) 2398eio_grp_limit (eio_req *grp, int limit)
2002{ 2399{
2003 grp->int2 = limit; 2400 grp->int2 = limit;
2004 2401
2005 grp_try_feed (grp); 2402 grp_try_feed (grp);
2006} 2403}
2007 2404
2405void
2008void eio_grp_add (eio_req *grp, eio_req *req) 2406eio_grp_add (eio_req *grp, eio_req *req)
2009{ 2407{
2010 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2408 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2011 2409
2012 grp->flags |= EIO_FLAG_GROUPADD; 2410 grp->flags |= EIO_FLAG_GROUPADD;
2013 2411
2024} 2422}
2025 2423
2026/*****************************************************************************/ 2424/*****************************************************************************/
2027/* misc garbage */ 2425/* misc garbage */
2028 2426
2427eio_ssize_t
2029ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2428eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2030{ 2429{
2031 etp_worker wrk; 2430 etp_worker wrk;
2032 ssize_t ret; 2431 eio_ssize_t ret;
2033 2432
2034 wrk.dbuf = 0; 2433 wrk.dbuf = 0;
2035 2434
2036 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2435 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2037 2436

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines