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

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.19 by root, Tue Jun 17 23:34:20 2008 UTC vs.
Revision 1.90 by root, Sat Jul 16 16:46:10 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 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"
46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
41#include "xthread.h" 50#include "xthread.h"
42 51
43#include <errno.h> 52#include <errno.h>
44#include <stddef.h> 53#include <stddef.h>
45#include <stdlib.h> 54#include <stdlib.h>
49#include <sys/stat.h> 58#include <sys/stat.h>
50#include <limits.h> 59#include <limits.h>
51#include <fcntl.h> 60#include <fcntl.h>
52#include <assert.h> 61#include <assert.h>
53 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
54#ifndef EIO_FINISH 78#ifndef EIO_FINISH
55# 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
56#endif 80#endif
57 81
58#ifndef EIO_DESTROY 82#ifndef EIO_DESTROY
61 85
62#ifndef EIO_FEED 86#ifndef EIO_FEED
63# 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)
64#endif 88#endif
65 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
66#ifdef _WIN32 101#ifdef _WIN32
67 102
68 /*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 fchmod(fd,mode) EIO_ENOSYS ()
117 #define chown(path,uid,gid) EIO_ENOSYS ()
118 #define fchown(fd,uid,gid) EIO_ENOSYS ()
119 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
120 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
121 #define mknod(path,mode,dev) EIO_ENOSYS ()
122 #define sync() EIO_ENOSYS ()
123 #define readlink(path,buf,s) EIO_ENOSYS ()
124 #define statvfs(path,buf) EIO_ENOSYS ()
125 #define fstatvfs(fd,buf) EIO_ENOSYS ()
126
127 /* we could even stat and see if it exists */
128 static int
129 symlink (const char *old, const char *neu)
130 {
131 #if WINVER >= 0x0600
132 if (CreateSymbolicLink (neu, old, 1))
133 return 0;
134
135 if (CreateSymbolicLink (neu, old, 0))
136 return 0;
137 #endif
138
139 return EIO_ERRNO (ENOENT, -1);
140 }
141
142 /* POSIX API only */
143 #define CreateHardLink(neu,old) 0
144 #define CreateSymbolicLink(neu,old,flags) 0
145
146 struct statvfs
147 {
148 int dummy;
149 };
69 150
70#else 151#else
71 152
72# include "config.h"
73# include <sys/time.h> 153 #include <sys/time.h>
74# include <sys/select.h> 154 #include <sys/select.h>
155 #include <sys/statvfs.h>
75# include <unistd.h> 156 #include <unistd.h>
76# include <utime.h> 157 #include <utime.h>
77# include <signal.h> 158 #include <signal.h>
78# include <dirent.h> 159 #include <dirent.h>
79 160
161 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
162 #include <sys/mman.h>
163 #endif
164
165 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
166 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
167 #define _DIRENT_HAVE_D_TYPE /* sigh */
168 #define D_INO(de) (de)->d_fileno
169 #define D_NAMLEN(de) (de)->d_namlen
170 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
171 #define D_INO(de) (de)->d_ino
172 #endif
173
174 #ifdef _D_EXACT_NAMLEN
175 #undef D_NAMLEN
176 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
177 #endif
178
179 #ifdef _DIRENT_HAVE_D_TYPE
180 #define D_TYPE(de) (de)->d_type
181 #endif
182
80# ifndef EIO_STRUCT_DIRENT 183 #ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 184 #define EIO_STRUCT_DIRENT struct dirent
82# endif 185 #endif
83 186
84#endif 187#endif
85 188
86#if HAVE_SENDFILE 189#if HAVE_SENDFILE
87# if __linux 190# if __linux
88# include <sys/sendfile.h> 191# include <sys/sendfile.h>
89# elif __freebsd 192# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 193# include <sys/socket.h>
91# include <sys/uio.h> 194# include <sys/uio.h>
92# elif __hpux 195# elif __hpux
93# include <sys/socket.h> 196# include <sys/socket.h>
94# elif __solaris /* not yet */ 197# elif __solaris
95# include <sys/sendfile.h> 198# include <sys/sendfile.h>
96# else 199# else
97# error sendfile support requested but not available 200# error sendfile support requested but not available
98# endif 201# endif
99#endif 202#endif
100 203
101/* number of seconds after which an idle threads exit */ 204#ifndef D_TYPE
102#define IDLE_TIMEOUT 10 205# define D_TYPE(de) 0
206#endif
207#ifndef D_INO
208# define D_INO(de) 0
209#endif
210#ifndef D_NAMLEN
211# define D_NAMLEN(de) strlen ((de)->d_name)
212#endif
103 213
104/* used for struct dirent, AIX doesn't provide it */ 214/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 215#ifndef NAME_MAX
106# define NAME_MAX 4096 216# define NAME_MAX 4096
217#endif
218
219/* used for readlink etc. */
220#ifndef PATH_MAX
221# define PATH_MAX 4096
107#endif 222#endif
108 223
109/* buffer size for various temporary buffers */ 224/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 225#define EIO_BUFSIZE 65536
111 226
117 errno = ENOMEM; \ 232 errno = ENOMEM; \
118 if (!eio_buf) \ 233 if (!eio_buf) \
119 return -1; 234 return -1;
120 235
121#define EIO_TICKS ((1000000 + 1023) >> 10) 236#define EIO_TICKS ((1000000 + 1023) >> 10)
122
123/*****************************************************************************/
124 237
125#define ETP_PRI_MIN EIO_PRI_MIN 238#define ETP_PRI_MIN EIO_PRI_MIN
126#define ETP_PRI_MAX EIO_PRI_MAX 239#define ETP_PRI_MAX EIO_PRI_MAX
127 240
128struct etp_worker; 241struct etp_worker;
144 if (wrk->dirp) \ 257 if (wrk->dirp) \
145 { \ 258 { \
146 closedir (wrk->dirp); \ 259 closedir (wrk->dirp); \
147 wrk->dirp = 0; \ 260 wrk->dirp = 0; \
148 } 261 }
262
149#define ETP_WORKER_COMMON \ 263#define ETP_WORKER_COMMON \
150 void *dbuf; \ 264 void *dbuf; \
151 DIR *dirp; 265 DIR *dirp;
152 266
153/*****************************************************************************/ 267/*****************************************************************************/
154 268
155#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 269#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
156 270
157/* calculcate time difference in ~1/EIO_TICKS of a second */ 271/* calculate time difference in ~1/EIO_TICKS of a second */
272ecb_inline int
158static int tvdiff (struct timeval *tv1, struct timeval *tv2) 273tvdiff (struct timeval *tv1, struct timeval *tv2)
159{ 274{
160 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 275 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
161 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 276 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
162} 277}
163 278
170static unsigned int max_poll_reqs; /* reslock */ 285static unsigned int max_poll_reqs; /* reslock */
171 286
172static volatile unsigned int nreqs; /* reqlock */ 287static volatile unsigned int nreqs; /* reqlock */
173static volatile unsigned int nready; /* reqlock */ 288static volatile unsigned int nready; /* reqlock */
174static volatile unsigned int npending; /* reqlock */ 289static volatile unsigned int npending; /* reqlock */
175static volatile unsigned int max_idle = 4; 290static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
291static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
176 292
293static xmutex_t wrklock;
294static xmutex_t reslock;
295static xmutex_t reqlock;
296static xcond_t reqwait;
297
298#if !HAVE_PREADWRITE
299/*
300 * make our pread/pwrite emulation safe against themselves, but not against
301 * normal read/write by using a mutex. slows down execution a lot,
302 * but that's your problem, not mine.
303 */
177static mutex_t wrklock = X_MUTEX_INIT; 304static xmutex_t preadwritelock = X_MUTEX_INIT;
178static mutex_t reslock = X_MUTEX_INIT; 305#endif
179static mutex_t reqlock = X_MUTEX_INIT;
180static cond_t reqwait = X_COND_INIT;
181 306
182typedef struct etp_worker 307typedef struct etp_worker
183{ 308{
184 /* locked by wrklock */ 309 /* locked by wrklock */
185 struct etp_worker *prev, *next; 310 struct etp_worker *prev, *next;
186 311
187 thread_t tid; 312 xthread_t tid;
188 313
189 /* locked by reslock, reqlock or wrklock */ 314 /* locked by reslock, reqlock or wrklock */
190 ETP_REQ *req; /* currently processed request */ 315 ETP_REQ *req; /* currently processed request */
191 316
192 ETP_WORKER_COMMON 317 ETP_WORKER_COMMON
197#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 322#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
198#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 323#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
199 324
200/* worker threads management */ 325/* worker threads management */
201 326
327static void ecb_cold
202static void etp_worker_clear (etp_worker *wrk) 328etp_worker_clear (etp_worker *wrk)
203{ 329{
204 ETP_WORKER_CLEAR (wrk); 330 ETP_WORKER_CLEAR (wrk);
205} 331}
206 332
333static void ecb_cold
207static void etp_worker_free (etp_worker *wrk) 334etp_worker_free (etp_worker *wrk)
208{ 335{
209 wrk->next->prev = wrk->prev; 336 wrk->next->prev = wrk->prev;
210 wrk->prev->next = wrk->next; 337 wrk->prev->next = wrk->next;
211 338
212 free (wrk); 339 free (wrk);
213} 340}
214 341
215static unsigned int etp_nreqs (void) 342static unsigned int
343etp_nreqs (void)
216{ 344{
217 int retval; 345 int retval;
218 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 346 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
219 retval = nreqs; 347 retval = nreqs;
220 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 348 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
221 return retval; 349 return retval;
222} 350}
223 351
224static unsigned int etp_nready (void) 352static unsigned int
353etp_nready (void)
225{ 354{
226 unsigned int retval; 355 unsigned int retval;
227 356
228 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 357 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
229 retval = nready; 358 retval = nready;
230 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 359 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
231 360
232 return retval; 361 return retval;
233} 362}
234 363
235static unsigned int etp_npending (void) 364static unsigned int
365etp_npending (void)
236{ 366{
237 unsigned int retval; 367 unsigned int retval;
238 368
239 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 369 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
240 retval = npending; 370 retval = npending;
241 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 371 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
242 372
243 return retval; 373 return retval;
244} 374}
245 375
246static unsigned int etp_nthreads (void) 376static unsigned int
377etp_nthreads (void)
247{ 378{
248 unsigned int retval; 379 unsigned int retval;
249 380
250 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 381 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
251 retval = started; 382 retval = started;
265} etp_reqq; 396} etp_reqq;
266 397
267static etp_reqq req_queue; 398static etp_reqq req_queue;
268static etp_reqq res_queue; 399static etp_reqq res_queue;
269 400
401static int ecb_noinline
270static int reqq_push (etp_reqq *q, ETP_REQ *req) 402reqq_push (etp_reqq *q, ETP_REQ *req)
271{ 403{
272 int pri = req->pri; 404 int pri = req->pri;
273 req->next = 0; 405 req->next = 0;
274 406
275 if (q->qe[pri]) 407 if (q->qe[pri])
281 q->qe[pri] = q->qs[pri] = req; 413 q->qe[pri] = q->qs[pri] = req;
282 414
283 return q->size++; 415 return q->size++;
284} 416}
285 417
418static ETP_REQ * ecb_noinline
286static ETP_REQ *reqq_shift (etp_reqq *q) 419reqq_shift (etp_reqq *q)
287{ 420{
288 int pri; 421 int pri;
289 422
290 if (!q->size) 423 if (!q->size)
291 return 0; 424 return 0;
306 } 439 }
307 440
308 abort (); 441 abort ();
309} 442}
310 443
311static void etp_atfork_prepare (void) 444static void ecb_cold
445etp_thread_init (void)
312{ 446{
313 X_LOCK (wrklock);
314 X_LOCK (reqlock);
315 X_LOCK (reslock);
316#if !HAVE_PREADWRITE 447#if !HAVE_PREADWRITE
317 X_LOCK (preadwritelock); 448 X_MUTEX_CREATE (preadwritelock);
318#endif 449#endif
450 X_MUTEX_CREATE (wrklock);
451 X_MUTEX_CREATE (reslock);
452 X_MUTEX_CREATE (reqlock);
453 X_COND_CREATE (reqwait);
319} 454}
320 455
456static void ecb_cold
457etp_atfork_prepare (void)
458{
459}
460
461static void ecb_cold
321static void etp_atfork_parent (void) 462etp_atfork_parent (void)
322{ 463{
323#if !HAVE_PREADWRITE
324 X_UNLOCK (preadwritelock);
325#endif
326 X_UNLOCK (reslock);
327 X_UNLOCK (reqlock);
328 X_UNLOCK (wrklock);
329} 464}
330 465
466static void ecb_cold
331static void etp_atfork_child (void) 467etp_atfork_child (void)
332{ 468{
333 ETP_REQ *prv; 469 ETP_REQ *prv;
334 470
335 while (prv = reqq_shift (&req_queue)) 471 while ((prv = reqq_shift (&req_queue)))
336 ETP_DESTROY (prv); 472 ETP_DESTROY (prv);
337 473
338 while (prv = reqq_shift (&res_queue)) 474 while ((prv = reqq_shift (&res_queue)))
339 ETP_DESTROY (prv); 475 ETP_DESTROY (prv);
340 476
341 while (wrk_first.next != &wrk_first) 477 while (wrk_first.next != &wrk_first)
342 { 478 {
343 etp_worker *wrk = wrk_first.next; 479 etp_worker *wrk = wrk_first.next;
353 idle = 0; 489 idle = 0;
354 nreqs = 0; 490 nreqs = 0;
355 nready = 0; 491 nready = 0;
356 npending = 0; 492 npending = 0;
357 493
358 etp_atfork_parent (); 494 etp_thread_init ();
359} 495}
360 496
361static void 497static void ecb_cold
362etp_once_init (void) 498etp_once_init (void)
363{ 499{
500 etp_thread_init ();
364 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 501 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
365} 502}
366 503
367static int 504static int ecb_cold
368etp_init (void (*want_poll)(void), void (*done_poll)(void)) 505etp_init (void (*want_poll)(void), void (*done_poll)(void))
369{ 506{
370 static pthread_once_t doinit = PTHREAD_ONCE_INIT; 507 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
371 508
372 pthread_once (&doinit, etp_once_init); 509 pthread_once (&doinit, etp_once_init);
377 return 0; 514 return 0;
378} 515}
379 516
380X_THREAD_PROC (etp_proc); 517X_THREAD_PROC (etp_proc);
381 518
519static void ecb_cold
382static void etp_start_thread (void) 520etp_start_thread (void)
383{ 521{
384 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 522 etp_worker *wrk = calloc (1, sizeof (etp_worker));
385 523
386 /*TODO*/ 524 /*TODO*/
387 assert (("unable to allocate worker thread data", wrk)); 525 assert (("unable to allocate worker thread data", wrk));
400 free (wrk); 538 free (wrk);
401 539
402 X_UNLOCK (wrklock); 540 X_UNLOCK (wrklock);
403} 541}
404 542
543static void
405static void etp_maybe_start_thread (void) 544etp_maybe_start_thread (void)
406{ 545{
407 if (etp_nthreads () >= wanted) 546 if (ecb_expect_true (etp_nthreads () >= wanted))
408 return; 547 return;
409 548
410 /* todo: maybe use idle here, but might be less exact */ 549 /* todo: maybe use idle here, but might be less exact */
411 if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) 550 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
412 return; 551 return;
413 552
414 etp_start_thread (); 553 etp_start_thread ();
415} 554}
416 555
556static void ecb_cold
417static void etp_end_thread (void) 557etp_end_thread (void)
418{ 558{
419 eio_req *req = calloc (1, sizeof (eio_req)); 559 eio_req *req = calloc (1, sizeof (eio_req));
420 560
421 req->type = -1; 561 req->type = -1;
422 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 562 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
429 X_LOCK (wrklock); 569 X_LOCK (wrklock);
430 --started; 570 --started;
431 X_UNLOCK (wrklock); 571 X_UNLOCK (wrklock);
432} 572}
433 573
434static int etp_poll (void) 574static int
575etp_poll (void)
435{ 576{
436 unsigned int maxreqs; 577 unsigned int maxreqs;
437 unsigned int maxtime; 578 unsigned int maxtime;
438 struct timeval tv_start, tv_now; 579 struct timeval tv_start, tv_now;
439 580
469 610
470 X_LOCK (reqlock); 611 X_LOCK (reqlock);
471 --nreqs; 612 --nreqs;
472 X_UNLOCK (reqlock); 613 X_UNLOCK (reqlock);
473 614
474 if (req->type == EIO_GROUP && req->size) 615 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
475 { 616 {
476 req->int1 = 1; /* mark request as delayed */ 617 req->int1 = 1; /* mark request as delayed */
477 continue; 618 continue;
478 } 619 }
479 else 620 else
480 { 621 {
481 int res = ETP_FINISH (req); 622 int res = ETP_FINISH (req);
482 if (res) 623 if (ecb_expect_false (res))
483 return res; 624 return res;
484 } 625 }
485 626
486 if (maxreqs && !--maxreqs) 627 if (ecb_expect_false (maxreqs && !--maxreqs))
487 break; 628 break;
488 629
489 if (maxtime) 630 if (maxtime)
490 { 631 {
491 gettimeofday (&tv_now, 0); 632 gettimeofday (&tv_now, 0);
497 638
498 errno = EAGAIN; 639 errno = EAGAIN;
499 return -1; 640 return -1;
500} 641}
501 642
643static void
502static void etp_cancel (ETP_REQ *req) 644etp_cancel (ETP_REQ *req)
503{ 645{
504 X_LOCK (wrklock); 646 req->cancelled = 1;
505 req->flags |= EIO_FLAG_CANCELLED;
506 X_UNLOCK (wrklock);
507 647
508 eio_grp_cancel (req); 648 eio_grp_cancel (req);
509} 649}
510 650
651static void
511static void etp_submit (ETP_REQ *req) 652etp_submit (ETP_REQ *req)
512{ 653{
513 req->pri -= ETP_PRI_MIN; 654 req->pri -= ETP_PRI_MIN;
514 655
515 if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 656 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
516 if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 657 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
517 658
659 if (ecb_expect_false (req->type == EIO_GROUP))
660 {
661 /* I hope this is worth it :/ */
518 X_LOCK (reqlock); 662 X_LOCK (reqlock);
519 ++nreqs; 663 ++nreqs;
664 X_UNLOCK (reqlock);
665
666 X_LOCK (reslock);
667
668 ++npending;
669
670 if (!reqq_push (&res_queue, req) && want_poll_cb)
671 want_poll_cb ();
672
673 X_UNLOCK (reslock);
674 }
675 else
676 {
677 X_LOCK (reqlock);
678 ++nreqs;
520 ++nready; 679 ++nready;
521 reqq_push (&req_queue, req); 680 reqq_push (&req_queue, req);
522 X_COND_SIGNAL (reqwait); 681 X_COND_SIGNAL (reqwait);
523 X_UNLOCK (reqlock); 682 X_UNLOCK (reqlock);
524 683
525 etp_maybe_start_thread (); 684 etp_maybe_start_thread ();
685 }
526} 686}
527 687
688static void ecb_cold
528static void etp_set_max_poll_time (double nseconds) 689etp_set_max_poll_time (double nseconds)
529{ 690{
530 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 691 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
531 max_poll_time = nseconds; 692 max_poll_time = nseconds * EIO_TICKS;
532 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 693 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
533} 694}
534 695
696static void ecb_cold
535static void etp_set_max_poll_reqs (unsigned int maxreqs) 697etp_set_max_poll_reqs (unsigned int maxreqs)
536{ 698{
537 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 699 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
538 max_poll_reqs = maxreqs; 700 max_poll_reqs = maxreqs;
539 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 701 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
540} 702}
541 703
704static void ecb_cold
542static void etp_set_max_idle (unsigned int nthreads) 705etp_set_max_idle (unsigned int nthreads)
543{ 706{
544 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 707 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
545 max_idle = nthreads <= 0 ? 1 : nthreads; 708 max_idle = nthreads;
546 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 709 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
547} 710}
548 711
712static void ecb_cold
713etp_set_idle_timeout (unsigned int seconds)
714{
715 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
716 idle_timeout = seconds;
717 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
718}
719
720static void ecb_cold
549static void etp_set_min_parallel (unsigned int nthreads) 721etp_set_min_parallel (unsigned int nthreads)
550{ 722{
551 if (wanted < nthreads) 723 if (wanted < nthreads)
552 wanted = nthreads; 724 wanted = nthreads;
553} 725}
554 726
727static void ecb_cold
555static void etp_set_max_parallel (unsigned int nthreads) 728etp_set_max_parallel (unsigned int nthreads)
556{ 729{
557 if (wanted > nthreads) 730 if (wanted > nthreads)
558 wanted = nthreads; 731 wanted = nthreads;
559 732
560 while (started > wanted) 733 while (started > wanted)
561 etp_end_thread (); 734 etp_end_thread ();
562} 735}
563 736
564/*****************************************************************************/ 737/*****************************************************************************/
565 738
739static void
566static void grp_try_feed (eio_req *grp) 740grp_try_feed (eio_req *grp)
567{ 741{
568 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 742 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
569 { 743 {
570 int old_len = grp->size; 744 grp->flags &= ~EIO_FLAG_GROUPADD;
571 745
572 EIO_FEED (grp); 746 EIO_FEED (grp);
573 747
574 /* stop if no progress has been made */ 748 /* stop if no progress has been made */
575 if (old_len == grp->size) 749 if (!(grp->flags & EIO_FLAG_GROUPADD))
576 { 750 {
577 grp->feed = 0; 751 grp->feed = 0;
578 break; 752 break;
579 } 753 }
580 } 754 }
581} 755}
582 756
757static int
583static int grp_dec (eio_req *grp) 758grp_dec (eio_req *grp)
584{ 759{
585 --grp->size; 760 --grp->size;
586 761
587 /* call feeder, if applicable */ 762 /* call feeder, if applicable */
588 grp_try_feed (grp); 763 grp_try_feed (grp);
592 return eio_finish (grp); 767 return eio_finish (grp);
593 else 768 else
594 return 0; 769 return 0;
595} 770}
596 771
772static void
597void eio_destroy (eio_req *req) 773eio_destroy (eio_req *req)
598{ 774{
599 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 775 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
600 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 776 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
601 777
602 EIO_DESTROY (req); 778 EIO_DESTROY (req);
603} 779}
604 780
781static int
605static int eio_finish (eio_req *req) 782eio_finish (eio_req *req)
606{ 783{
607 int res = EIO_FINISH (req); 784 int res = EIO_FINISH (req);
608 785
609 if (req->grp) 786 if (req->grp)
610 { 787 {
618 if (grp->grp_first == req) 795 if (grp->grp_first == req)
619 grp->grp_first = req->grp_next; 796 grp->grp_first = req->grp_next;
620 797
621 res2 = grp_dec (grp); 798 res2 = grp_dec (grp);
622 799
623 if (!res && res2) 800 if (!res)
624 res = res2; 801 res = res2;
625 } 802 }
626 803
627 eio_destroy (req); 804 eio_destroy (req);
628 805
629 return res; 806 return res;
630} 807}
631 808
809void
632void eio_grp_cancel (eio_req *grp) 810eio_grp_cancel (eio_req *grp)
633{ 811{
634 for (grp = grp->grp_first; grp; grp = grp->grp_next) 812 for (grp = grp->grp_first; grp; grp = grp->grp_next)
635 eio_cancel (grp); 813 eio_cancel (grp);
636} 814}
637 815
816void
638void eio_cancel (eio_req *req) 817eio_cancel (eio_req *req)
639{ 818{
640 etp_cancel (req); 819 etp_cancel (req);
641} 820}
642 821
822void
643void eio_submit (eio_req *req) 823eio_submit (eio_req *req)
644{ 824{
645 etp_submit (req); 825 etp_submit (req);
646} 826}
647 827
648unsigned int eio_nreqs (void) 828unsigned int
829eio_nreqs (void)
649{ 830{
650 return etp_nreqs (); 831 return etp_nreqs ();
651} 832}
652 833
653unsigned int eio_nready (void) 834unsigned int
835eio_nready (void)
654{ 836{
655 return etp_nready (); 837 return etp_nready ();
656} 838}
657 839
658unsigned int eio_npending (void) 840unsigned int
841eio_npending (void)
659{ 842{
660 return etp_npending (); 843 return etp_npending ();
661} 844}
662 845
663unsigned int eio_nthreads (void) 846unsigned int ecb_cold
847eio_nthreads (void)
664{ 848{
665 return etp_nthreads (); 849 return etp_nthreads ();
666} 850}
667 851
852void ecb_cold
668void eio_set_max_poll_time (double nseconds) 853eio_set_max_poll_time (double nseconds)
669{ 854{
670 etp_set_max_poll_time (nseconds); 855 etp_set_max_poll_time (nseconds);
671} 856}
672 857
858void ecb_cold
673void eio_set_max_poll_reqs (unsigned int maxreqs) 859eio_set_max_poll_reqs (unsigned int maxreqs)
674{ 860{
675 etp_set_max_poll_reqs (maxreqs); 861 etp_set_max_poll_reqs (maxreqs);
676} 862}
677 863
864void ecb_cold
678void eio_set_max_idle (unsigned int nthreads) 865eio_set_max_idle (unsigned int nthreads)
679{ 866{
680 etp_set_max_idle (nthreads); 867 etp_set_max_idle (nthreads);
681} 868}
682 869
870void ecb_cold
871eio_set_idle_timeout (unsigned int seconds)
872{
873 etp_set_idle_timeout (seconds);
874}
875
876void ecb_cold
683void eio_set_min_parallel (unsigned int nthreads) 877eio_set_min_parallel (unsigned int nthreads)
684{ 878{
685 etp_set_min_parallel (nthreads); 879 etp_set_min_parallel (nthreads);
686} 880}
687 881
882void ecb_cold
688void eio_set_max_parallel (unsigned int nthreads) 883eio_set_max_parallel (unsigned int nthreads)
689{ 884{
690 etp_set_max_parallel (nthreads); 885 etp_set_max_parallel (nthreads);
691} 886}
692 887
693int eio_poll (void) 888int eio_poll (void)
697 892
698/*****************************************************************************/ 893/*****************************************************************************/
699/* work around various missing functions */ 894/* work around various missing functions */
700 895
701#if !HAVE_PREADWRITE 896#if !HAVE_PREADWRITE
897# undef pread
898# undef pwrite
702# define pread eio__pread 899# define pread eio__pread
703# define pwrite eio__pwrite 900# define pwrite eio__pwrite
704 901
705/*
706 * make our pread/pwrite safe against themselves, but not against
707 * normal read/write by using a mutex. slows down execution a lot,
708 * but that's your problem, not mine.
709 */
710static mutex_t preadwritelock = X_MUTEX_INIT;
711
712static ssize_t 902static eio_ssize_t
713eio__pread (int fd, void *buf, size_t count, off_t offset) 903eio__pread (int fd, void *buf, size_t count, off_t offset)
714{ 904{
715 ssize_t res; 905 eio_ssize_t res;
716 off_t ooffset; 906 off_t ooffset;
717 907
718 X_LOCK (preadwritelock); 908 X_LOCK (preadwritelock);
719 ooffset = lseek (fd, 0, SEEK_CUR); 909 ooffset = lseek (fd, 0, SEEK_CUR);
720 lseek (fd, offset, SEEK_SET); 910 lseek (fd, offset, SEEK_SET);
723 X_UNLOCK (preadwritelock); 913 X_UNLOCK (preadwritelock);
724 914
725 return res; 915 return res;
726} 916}
727 917
728static ssize_t 918static eio_ssize_t
729eio__pwrite (int fd, void *buf, size_t count, off_t offset) 919eio__pwrite (int fd, void *buf, size_t count, off_t offset)
730{ 920{
731 ssize_t res; 921 eio_ssize_t res;
732 off_t ooffset; 922 off_t ooffset;
733 923
734 X_LOCK (preadwritelock); 924 X_LOCK (preadwritelock);
735 ooffset = lseek (fd, 0, SEEK_CUR); 925 ooffset = lseek (fd, 0, SEEK_CUR);
736 lseek (fd, offset, SEEK_SET); 926 lseek (fd, offset, SEEK_SET);
737 res = write (fd, buf, count); 927 res = write (fd, buf, count);
738 lseek (fd, offset, SEEK_SET); 928 lseek (fd, ooffset, SEEK_SET);
739 X_UNLOCK (preadwritelock); 929 X_UNLOCK (preadwritelock);
740 930
741 return res; 931 return res;
742} 932}
743#endif 933#endif
744 934
745#ifndef HAVE_FUTIMES 935#ifndef HAVE_UTIMES
746 936
937# undef utimes
747# define utimes(path,times) eio__utimes (path, times) 938# define utimes(path,times) eio__utimes (path, times)
748# define futimes(fd,times) eio__futimes (fd, times)
749 939
750static int 940static int
751eio__utimes (const char *filename, const struct timeval times[2]) 941eio__utimes (const char *filename, const struct timeval times[2])
752{ 942{
753 if (times) 943 if (times)
761 } 951 }
762 else 952 else
763 return utime (filename, 0); 953 return utime (filename, 0);
764} 954}
765 955
956#endif
957
958#ifndef HAVE_FUTIMES
959
960# undef futimes
961# define futimes(fd,times) eio__futimes (fd, times)
962
963static int
766static int eio__futimes (int fd, const struct timeval tv[2]) 964eio__futimes (int fd, const struct timeval tv[2])
767{ 965{
768 errno = ENOSYS; 966 errno = ENOSYS;
769 return -1; 967 return -1;
770} 968}
771 969
772#endif 970#endif
773 971
774#if !HAVE_FDATASYNC 972#if !HAVE_FDATASYNC
973# undef fdatasync
775# define fdatasync fsync 974# define fdatasync(fd) fsync (fd)
776#endif 975#endif
976
977/* sync_file_range always needs emulation */
978static int
979eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
980{
981#if HAVE_SYNC_FILE_RANGE
982 int res;
983
984 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
985 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
986 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
987 {
988 flags = 0
989 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
990 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
991 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
992 }
993
994 res = sync_file_range (fd, offset, nbytes, flags);
995
996 if (!res || errno != ENOSYS)
997 return res;
998#endif
999
1000 /* even though we could play tricks with the flags, it's better to always
1001 * call fdatasync, as that matches the expectation of its users best */
1002 return fdatasync (fd);
1003}
1004
1005static int
1006eio__fallocate (int fd, int mode, off_t offset, size_t len)
1007{
1008#if HAVE_FALLOCATE
1009 return fallocate (fd, mode, offset, len);
1010#else
1011 errno = ENOSYS;
1012 return -1;
1013#endif
1014}
777 1015
778#if !HAVE_READAHEAD 1016#if !HAVE_READAHEAD
1017# undef readahead
779# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1018# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
780 1019
781static ssize_t 1020static eio_ssize_t
782eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1021eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
783{ 1022{
784 size_t todo = count; 1023 size_t todo = count;
785 dBUF; 1024 dBUF;
786 1025
798} 1037}
799 1038
800#endif 1039#endif
801 1040
802/* sendfile always needs emulation */ 1041/* sendfile always needs emulation */
803static ssize_t 1042static eio_ssize_t
804eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1043eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
805{ 1044{
1045 eio_ssize_t written = 0;
806 ssize_t res; 1046 eio_ssize_t res;
807 1047
808 if (!count) 1048 if (!count)
809 return 0; 1049 return 0;
810 1050
1051 for (;;)
1052 {
1053#ifdef __APPLE__
1054# undef HAVE_SENDFILE /* broken, as everything on os x */
1055#endif
811#if HAVE_SENDFILE 1056#if HAVE_SENDFILE
812# if __linux 1057# if __linux
1058 off_t soffset = offset;
813 res = sendfile (ofd, ifd, &offset, count); 1059 res = sendfile (ofd, ifd, &soffset, count);
814 1060
815# elif __freebsd 1061# elif __FreeBSD__
816 /* 1062 /*
817 * Of course, the freebsd sendfile is a dire hack with no thoughts 1063 * Of course, the freebsd sendfile is a dire hack with no thoughts
818 * wasted on making it similar to other I/O functions. 1064 * wasted on making it similar to other I/O functions.
819 */ 1065 */
820 {
821 off_t sbytes; 1066 off_t sbytes;
822 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1067 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
823 1068
824 if (res < 0 && sbytes) 1069 #if 0 /* according to the manpage, this is correct, but broken behaviour */
825 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1070 /* freebsd' sendfile will return 0 on success */
1071 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1072 /* not on e.g. EIO or EPIPE - sounds broken */
1073 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
826 res = sbytes; 1074 res = sbytes;
827 } 1075 #endif
1076
1077 /* according to source inspection, this is correct, and useful behaviour */
1078 if (sbytes)
1079 res = sbytes;
1080
1081# elif defined (__APPLE__)
1082 off_t sbytes = count;
1083 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1084
1085 /* according to the manpage, sbytes is always valid */
1086 if (sbytes)
1087 res = sbytes;
828 1088
829# elif __hpux 1089# elif __hpux
830 res = sendfile (ofd, ifd, offset, count, 0, 0); 1090 res = sendfile (ofd, ifd, offset, count, 0, 0);
831 1091
832# elif __solaris 1092# elif __solaris
833 {
834 struct sendfilevec vec; 1093 struct sendfilevec vec;
835 size_t sbytes; 1094 size_t sbytes;
836 1095
837 vec.sfv_fd = ifd; 1096 vec.sfv_fd = ifd;
838 vec.sfv_flag = 0; 1097 vec.sfv_flag = 0;
839 vec.sfv_off = offset; 1098 vec.sfv_off = offset;
840 vec.sfv_len = count; 1099 vec.sfv_len = count;
841 1100
842 res = sendfilev (ofd, &vec, 1, &sbytes); 1101 res = sendfilev (ofd, &vec, 1, &sbytes);
843 1102
844 if (res < 0 && sbytes) 1103 if (res < 0 && sbytes)
845 res = sbytes; 1104 res = sbytes;
846 }
847 1105
848# endif 1106# endif
1107
1108#elif defined (_WIN32) && 0
1109 /* does not work, just for documentation of what would need to be done */
1110 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1111 /* libeio guarantees that the file offset does not change, and windows */
1112 /* has no way to get an independent handle to the same file description */
1113 HANDLE h = TO_SOCKET (ifd);
1114 SetFilePointer (h, offset, 0, FILE_BEGIN);
1115 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1116
849#else 1117#else
850 res = -1; 1118 res = -1;
851 errno = ENOSYS; 1119 errno = ENOSYS;
852#endif 1120#endif
853 1121
1122 /* we assume sendfile can copy at least 128mb in one go */
1123 if (res <= 128 * 1024 * 1024)
1124 {
1125 if (res > 0)
1126 written += res;
1127
1128 if (written)
1129 return written;
1130
1131 break;
1132 }
1133 else
1134 {
1135 /* if we requested more, then probably the kernel was lazy */
1136 written += res;
1137 offset += res;
1138 count -= res;
1139
1140 if (!count)
1141 return written;
1142 }
1143 }
1144
854 if (res < 0 1145 if (res < 0
855 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1146 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1147 /* BSDs */
1148#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1149 || errno == ENOTSUP
1150#endif
1151#ifdef EOPNOTSUPP /* windows */
1152 || errno == EOPNOTSUPP /* BSDs */
1153#endif
856#if __solaris 1154#if __solaris
857 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1155 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
858#endif 1156#endif
859 ) 1157 )
860 ) 1158 )
864 1162
865 res = 0; 1163 res = 0;
866 1164
867 while (count) 1165 while (count)
868 { 1166 {
869 ssize_t cnt; 1167 eio_ssize_t cnt;
870 1168
871 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1169 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
872 1170
873 if (cnt <= 0) 1171 if (cnt <= 0)
874 { 1172 {
891 } 1189 }
892 1190
893 return res; 1191 return res;
894} 1192}
895 1193
1194#ifdef PAGESIZE
1195# define eio_pagesize() PAGESIZE
1196#else
1197static intptr_t
1198eio_pagesize (void)
1199{
1200 static intptr_t page;
1201
1202 if (!page)
1203 page = sysconf (_SC_PAGESIZE);
1204
1205 return page;
1206}
1207#endif
1208
1209static void
1210eio_page_align (void **addr, size_t *length)
1211{
1212 intptr_t mask = eio_pagesize () - 1;
1213
1214 /* round down addr */
1215 intptr_t adj = mask & (intptr_t)*addr;
1216
1217 *addr = (void *)((intptr_t)*addr - adj);
1218 *length += adj;
1219
1220 /* round up length */
1221 *length = (*length + mask) & ~mask;
1222}
1223
1224#if !_POSIX_MEMLOCK
1225# define eio__mlockall(a) EIO_ENOSYS ()
1226#else
1227
1228static int
1229eio__mlockall (int flags)
1230{
1231 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1232 extern int mallopt (int, int);
1233 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1234 #endif
1235
1236 if (EIO_MCL_CURRENT != MCL_CURRENT
1237 || EIO_MCL_FUTURE != MCL_FUTURE)
1238 {
1239 flags = 0
1240 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1241 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1242 }
1243
1244 return mlockall (flags);
1245}
1246#endif
1247
1248#if !_POSIX_MEMLOCK_RANGE
1249# define eio__mlock(a,b) EIO_ENOSYS ()
1250#else
1251
1252static int
1253eio__mlock (void *addr, size_t length)
1254{
1255 eio_page_align (&addr, &length);
1256
1257 return mlock (addr, length);
1258}
1259
1260#endif
1261
1262#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1263# define eio__msync(a,b,c) EIO_ENOSYS ()
1264#else
1265
1266static int
1267eio__msync (void *mem, size_t len, int flags)
1268{
1269 eio_page_align (&mem, &len);
1270
1271 if (EIO_MS_ASYNC != MS_SYNC
1272 || EIO_MS_INVALIDATE != MS_INVALIDATE
1273 || EIO_MS_SYNC != MS_SYNC)
1274 {
1275 flags = 0
1276 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1277 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1278 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1279 }
1280
1281 return msync (mem, len, flags);
1282}
1283
1284#endif
1285
1286static int
1287eio__mtouch (eio_req *req)
1288{
1289 void *mem = req->ptr2;
1290 size_t len = req->size;
1291 int flags = req->int1;
1292
1293 eio_page_align (&mem, &len);
1294
1295 {
1296 intptr_t addr = (intptr_t)mem;
1297 intptr_t end = addr + len;
1298 intptr_t page = eio_pagesize ();
1299
1300 if (addr < end)
1301 if (flags & EIO_MT_MODIFY) /* modify */
1302 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1303 else
1304 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1305 }
1306
1307 return 0;
1308}
1309
1310/*****************************************************************************/
1311/* requests implemented outside eio_execute, because they are so large */
1312
1313static void
1314eio__realpath (eio_req *req, etp_worker *self)
1315{
1316 char *rel = req->ptr1;
1317 char *res;
1318 char *tmp1, *tmp2;
1319#if SYMLOOP_MAX > 32
1320 int symlinks = SYMLOOP_MAX;
1321#else
1322 int symlinks = 32;
1323#endif
1324
1325 req->result = -1;
1326
1327 errno = EINVAL;
1328 if (!rel)
1329 return;
1330
1331 errno = ENOENT;
1332 if (!*rel)
1333 return;
1334
1335 if (!req->ptr2)
1336 {
1337 X_LOCK (wrklock);
1338 req->flags |= EIO_FLAG_PTR2_FREE;
1339 X_UNLOCK (wrklock);
1340 req->ptr2 = malloc (PATH_MAX * 3);
1341
1342 errno = ENOMEM;
1343 if (!req->ptr2)
1344 return;
1345 }
1346
1347 res = req->ptr2;
1348 tmp1 = res + PATH_MAX;
1349 tmp2 = tmp1 + PATH_MAX;
1350
1351#if 0 /* disabled, the musl way to do things is just too racy */
1352#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1353 /* on linux we may be able to ask the kernel */
1354 {
1355 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1356
1357 if (fd >= 0)
1358 {
1359 sprintf (tmp1, "/proc/self/fd/%d", fd);
1360 req->result = readlink (tmp1, res, PATH_MAX);
1361 close (fd);
1362
1363 /* here we should probably stat the open file and the disk file, to make sure they still match */
1364
1365 if (req->result > 0)
1366 goto done;
1367 }
1368 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1369 return;
1370 }
1371#endif
1372#endif
1373
1374 if (*rel != '/')
1375 {
1376 if (!getcwd (res, PATH_MAX))
1377 return;
1378
1379 if (res [1]) /* only use if not / */
1380 res += strlen (res);
1381 }
1382
1383 while (*rel)
1384 {
1385 eio_ssize_t len, linklen;
1386 char *beg = rel;
1387
1388 while (*rel && *rel != '/')
1389 ++rel;
1390
1391 len = rel - beg;
1392
1393 if (!len) /* skip slashes */
1394 {
1395 ++rel;
1396 continue;
1397 }
1398
1399 if (beg [0] == '.')
1400 {
1401 if (len == 1)
1402 continue; /* . - nop */
1403
1404 if (beg [1] == '.' && len == 2)
1405 {
1406 /* .. - back up one component, if possible */
1407
1408 while (res != req->ptr2)
1409 if (*--res == '/')
1410 break;
1411
1412 continue;
1413 }
1414 }
1415
1416 errno = ENAMETOOLONG;
1417 if (res + 1 + len + 1 >= tmp1)
1418 return;
1419
1420 /* copy one component */
1421 *res = '/';
1422 memcpy (res + 1, beg, len);
1423
1424 /* zero-terminate, for readlink */
1425 res [len + 1] = 0;
1426
1427 /* now check if it's a symlink */
1428 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1429
1430 if (linklen < 0)
1431 {
1432 if (errno != EINVAL)
1433 return;
1434
1435 /* it's a normal directory. hopefully */
1436 res += len + 1;
1437 }
1438 else
1439 {
1440 /* yay, it was a symlink - build new path in tmp2 */
1441 int rellen = strlen (rel);
1442
1443 errno = ENAMETOOLONG;
1444 if (linklen + 1 + rellen >= PATH_MAX)
1445 return;
1446
1447 errno = ELOOP;
1448 if (!--symlinks)
1449 return;
1450
1451 if (*tmp1 == '/')
1452 res = req->ptr2; /* symlink resolves to an absolute path */
1453
1454 /* we need to be careful, as rel might point into tmp2 already */
1455 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1456 tmp2 [linklen] = '/';
1457 memcpy (tmp2, tmp1, linklen);
1458
1459 rel = tmp2;
1460 }
1461 }
1462
1463 /* special case for the lone root path */
1464 if (res == req->ptr2)
1465 *res++ = '/';
1466
1467 req->result = res - (char *)req->ptr2;
1468
1469done:
1470 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1471}
1472
1473static signed char
1474eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1475{
1476 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1477 : a->inode < b->inode ? -1
1478 : a->inode > b->inode ? 1
1479 : 0;
1480}
1481
1482#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1483
1484#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1485#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1486
1487static void
1488eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1489{
1490 unsigned char bits [9 + sizeof (ino_t) * 8];
1491 unsigned char *bit = bits;
1492
1493 assert (CHAR_BIT == 8);
1494 assert (sizeof (eio_dirent) * 8 < 256);
1495 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1496 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1497
1498 if (size <= EIO_SORT_FAST)
1499 return;
1500
1501 /* first prepare an array of bits to test in our radix sort */
1502 /* try to take endianness into account, as well as differences in ino_t sizes */
1503 /* inode_bits must contain all inodes ORed together */
1504 /* which is used to skip bits that are 0 everywhere, which is very common */
1505 {
1506 ino_t endianness;
1507 int i, j;
1508
1509 /* we store the byte offset of byte n into byte n of "endianness" */
1510 for (i = 0; i < sizeof (ino_t); ++i)
1511 ((unsigned char *)&endianness)[i] = i;
1512
1513 *bit++ = 0;
1514
1515 for (i = 0; i < sizeof (ino_t); ++i)
1516 {
1517 /* shifting off the byte offsets out of "endianness" */
1518 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1519 endianness >>= 8;
1520
1521 for (j = 0; j < 8; ++j)
1522 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1523 *bit++ = offs + j;
1524 }
1525
1526 for (j = 0; j < 8; ++j)
1527 if (score_bits & (1 << j))
1528 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1529 }
1530
1531 /* now actually do the sorting (a variant of MSD radix sort) */
1532 {
1533 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1534 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1535 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1536 int stk_idx = 0;
1537
1538 base_stk [stk_idx] = dents;
1539 end_stk [stk_idx] = dents + size;
1540 bit_stk [stk_idx] = bit - 1;
1541
1542 do
1543 {
1544 base = base_stk [stk_idx];
1545 end = end_stk [stk_idx];
1546 bit = bit_stk [stk_idx];
1547
1548 for (;;)
1549 {
1550 unsigned char O = *bit >> 3;
1551 unsigned char M = 1 << (*bit & 7);
1552
1553 eio_dirent *a = base;
1554 eio_dirent *b = end;
1555
1556 if (b - a < EIO_SORT_CUTOFF)
1557 break;
1558
1559 /* now bit-partition the array on the bit */
1560 /* this ugly asymmetric loop seems to perform much better than typical */
1561 /* partition algos found in the literature */
1562 do
1563 if (!(((unsigned char *)a)[O] & M))
1564 ++a;
1565 else if (!(((unsigned char *)--b)[O] & M))
1566 {
1567 eio_dirent tmp = *a; *a = *b; *b = tmp;
1568 ++a;
1569 }
1570 while (b > a);
1571
1572 /* next bit, or stop, if no bits left in this path */
1573 if (!*--bit)
1574 break;
1575
1576 base_stk [stk_idx] = a;
1577 end_stk [stk_idx] = end;
1578 bit_stk [stk_idx] = bit;
1579 ++stk_idx;
1580
1581 end = a;
1582 }
1583 }
1584 while (stk_idx--);
1585 }
1586}
1587
1588static void
1589eio_dent_insertion_sort (eio_dirent *dents, int size)
1590{
1591 /* first move the smallest element to the front, to act as a sentinel */
1592 {
1593 int i;
1594 eio_dirent *min = dents;
1595
1596 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1597 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1598 if (EIO_DENT_CMP (dents [i], <, *min))
1599 min = &dents [i];
1600
1601 /* swap elements 0 and j (minimum) */
1602 {
1603 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1604 }
1605 }
1606
1607 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1608 {
1609 eio_dirent *i, *j;
1610
1611 for (i = dents + 1; i < dents + size; ++i)
1612 {
1613 eio_dirent value = *i;
1614
1615 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1616 j [1] = j [0];
1617
1618 j [1] = value;
1619 }
1620 }
1621}
1622
1623static void
1624eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1625{
1626 if (size <= 1)
1627 return; /* our insertion sort relies on size > 0 */
1628
1629 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1630 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1631 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1632
1633 /* use an insertion sort at the end, or for small arrays, */
1634 /* as insertion sort is more efficient for small partitions */
1635 eio_dent_insertion_sort (dents, size);
1636}
1637
896/* read a full directory */ 1638/* read a full directory */
897static void 1639static void
898eio__scandir (eio_req *req, etp_worker *self) 1640eio__scandir (eio_req *req, etp_worker *self)
899{ 1641{
900 DIR *dirp; 1642 DIR *dirp;
901 EIO_STRUCT_DIRENT *entp; 1643 EIO_STRUCT_DIRENT *entp;
902 char *name, *names; 1644 char *name, *names;
903 int memlen = 4096; 1645 int namesalloc = 4096;
904 int memofs = 0; 1646 int namesoffs = 0;
1647 int flags = req->int1;
1648 eio_dirent *dents = 0;
1649 int dentalloc = 128;
905 int res = 0; 1650 int dentoffs = 0;
1651 ino_t inode_bits = 0;
1652
1653 req->result = -1;
1654
1655 if (!(flags & EIO_READDIR_DENTS))
1656 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
906 1657
907 X_LOCK (wrklock); 1658 X_LOCK (wrklock);
1659 /* the corresponding closedir is in ETP_WORKER_CLEAR */
908 self->dirp = dirp = opendir (req->ptr1); 1660 self->dirp = dirp = opendir (req->ptr1);
1661
909 req->flags |= EIO_FLAG_PTR2_FREE; 1662 if (req->flags & EIO_FLAG_PTR1_FREE)
1663 free (req->ptr1);
1664
1665 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1666 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
910 req->ptr2 = names = malloc (memlen); 1667 req->ptr2 = names = malloc (namesalloc);
911 X_UNLOCK (wrklock); 1668 X_UNLOCK (wrklock);
912 1669
913 if (dirp && names) 1670 if (dirp && names && (!flags || dents))
914 for (;;) 1671 for (;;)
915 { 1672 {
916 errno = 0; 1673 errno = 0;
917 entp = readdir (dirp); 1674 entp = readdir (dirp);
918 1675
919 if (!entp) 1676 if (!entp)
1677 {
1678 if (errno)
1679 break;
1680
1681 /* sort etc. */
1682 req->int1 = flags;
1683 req->result = dentoffs;
1684
1685 if (flags & EIO_READDIR_STAT_ORDER)
1686 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1687 else if (flags & EIO_READDIR_DIRS_FIRST)
1688 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1689 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1690 else
1691 {
1692 /* in this case, all is known, and we just put dirs first and sort them */
1693 eio_dirent *oth = dents + dentoffs;
1694 eio_dirent *dir = dents;
1695
1696 /* now partition dirs to the front, and non-dirs to the back */
1697 /* by walking from both sides and swapping if necessary */
1698 while (oth > dir)
1699 {
1700 if (dir->type == EIO_DT_DIR)
1701 ++dir;
1702 else if ((--oth)->type == EIO_DT_DIR)
1703 {
1704 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1705
1706 ++dir;
1707 }
1708 }
1709
1710 /* now sort the dirs only (dirs all have the same score) */
1711 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1712 }
1713
920 break; 1714 break;
1715 }
921 1716
1717 /* now add the entry to our list(s) */
922 name = entp->d_name; 1718 name = entp->d_name;
923 1719
1720 /* skip . and .. entries */
924 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1721 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
925 { 1722 {
926 int len = strlen (name) + 1; 1723 int len = D_NAMLEN (entp) + 1;
927 1724
928 res++; 1725 while (ecb_expect_false (namesoffs + len > namesalloc))
929
930 while (memofs + len > memlen)
931 { 1726 {
932 memlen *= 2; 1727 namesalloc *= 2;
933 X_LOCK (wrklock); 1728 X_LOCK (wrklock);
934 req->ptr2 = names = realloc (names, memlen); 1729 req->ptr2 = names = realloc (names, namesalloc);
935 X_UNLOCK (wrklock); 1730 X_UNLOCK (wrklock);
936 1731
937 if (!names) 1732 if (!names)
938 break; 1733 break;
939 } 1734 }
940 1735
941 memcpy (names + memofs, name, len); 1736 memcpy (names + namesoffs, name, len);
1737
1738 if (dents)
1739 {
1740 struct eio_dirent *ent;
1741
1742 if (ecb_expect_false (dentoffs == dentalloc))
1743 {
1744 dentalloc *= 2;
1745 X_LOCK (wrklock);
1746 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1747 X_UNLOCK (wrklock);
1748
1749 if (!dents)
1750 break;
1751 }
1752
1753 ent = dents + dentoffs;
1754
1755 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1756 ent->namelen = len - 1;
1757 ent->inode = D_INO (entp);
1758
1759 inode_bits |= ent->inode;
1760
1761 switch (D_TYPE (entp))
1762 {
1763 default:
1764 ent->type = EIO_DT_UNKNOWN;
1765 flags |= EIO_READDIR_FOUND_UNKNOWN;
1766 break;
1767
1768 #ifdef DT_FIFO
1769 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1770 #endif
1771 #ifdef DT_CHR
1772 case DT_CHR: ent->type = EIO_DT_CHR; break;
1773 #endif
1774 #ifdef DT_MPC
1775 case DT_MPC: ent->type = EIO_DT_MPC; break;
1776 #endif
1777 #ifdef DT_DIR
1778 case DT_DIR: ent->type = EIO_DT_DIR; break;
1779 #endif
1780 #ifdef DT_NAM
1781 case DT_NAM: ent->type = EIO_DT_NAM; break;
1782 #endif
1783 #ifdef DT_BLK
1784 case DT_BLK: ent->type = EIO_DT_BLK; break;
1785 #endif
1786 #ifdef DT_MPB
1787 case DT_MPB: ent->type = EIO_DT_MPB; break;
1788 #endif
1789 #ifdef DT_REG
1790 case DT_REG: ent->type = EIO_DT_REG; break;
1791 #endif
1792 #ifdef DT_NWK
1793 case DT_NWK: ent->type = EIO_DT_NWK; break;
1794 #endif
1795 #ifdef DT_CMP
1796 case DT_CMP: ent->type = EIO_DT_CMP; break;
1797 #endif
1798 #ifdef DT_LNK
1799 case DT_LNK: ent->type = EIO_DT_LNK; break;
1800 #endif
1801 #ifdef DT_SOCK
1802 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1803 #endif
1804 #ifdef DT_DOOR
1805 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1806 #endif
1807 #ifdef DT_WHT
1808 case DT_WHT: ent->type = EIO_DT_WHT; break;
1809 #endif
1810 }
1811
1812 ent->score = 7;
1813
1814 if (flags & EIO_READDIR_DIRS_FIRST)
1815 {
1816 if (ent->type == EIO_DT_UNKNOWN)
1817 {
1818 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1819 ent->score = 1;
1820 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1821 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1822 }
1823 else if (ent->type == EIO_DT_DIR)
1824 ent->score = 0;
1825 }
1826 }
1827
942 memofs += len; 1828 namesoffs += len;
1829 ++dentoffs;
1830 }
1831
1832 if (EIO_CANCELLED (req))
1833 {
1834 errno = ECANCELED;
1835 break;
943 } 1836 }
944 } 1837 }
945
946 if (errno)
947 res = -1;
948
949 req->result = res;
950} 1838}
951 1839
952/*****************************************************************************/ 1840/*****************************************************************************/
953 1841
954#define ALLOC(len) \ 1842#define ALLOC(len) \
969X_THREAD_PROC (etp_proc) 1857X_THREAD_PROC (etp_proc)
970{ 1858{
971 ETP_REQ *req; 1859 ETP_REQ *req;
972 struct timespec ts; 1860 struct timespec ts;
973 etp_worker *self = (etp_worker *)thr_arg; 1861 etp_worker *self = (etp_worker *)thr_arg;
1862 int timeout;
974 1863
975 /* try to distribute timeouts somewhat randomly */ 1864 /* try to distribute timeouts somewhat evenly */
976 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1865 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
977 1866
978 for (;;) 1867 for (;;)
979 { 1868 {
1869 ts.tv_sec = 0;
1870
980 X_LOCK (reqlock); 1871 X_LOCK (reqlock);
981 1872
982 for (;;) 1873 for (;;)
983 { 1874 {
984 self->req = req = reqq_shift (&req_queue); 1875 self->req = req = reqq_shift (&req_queue);
985 1876
986 if (req) 1877 if (req)
987 break; 1878 break;
988 1879
1880 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1881 {
1882 X_UNLOCK (reqlock);
1883 X_LOCK (wrklock);
1884 --started;
1885 X_UNLOCK (wrklock);
1886 goto quit;
1887 }
1888
989 ++idle; 1889 ++idle;
990 1890
991 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1891 if (idle <= max_idle)
992 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1892 /* we are allowed to idle, so do so without any timeout */
1893 X_COND_WAIT (reqwait, reqlock);
1894 else
993 { 1895 {
994 if (idle > max_idle) 1896 /* initialise timeout once */
995 { 1897 if (!ts.tv_sec)
996 --idle; 1898 ts.tv_sec = time (0) + idle_timeout;
997 X_UNLOCK (reqlock);
998 X_LOCK (wrklock);
999 --started;
1000 X_UNLOCK (wrklock);
1001 goto quit;
1002 }
1003 1899
1004 /* we are allowed to idle, so do so without any timeout */
1005 X_COND_WAIT (reqwait, reqlock); 1900 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1901 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1006 } 1902 }
1007 1903
1008 --idle; 1904 --idle;
1009 } 1905 }
1010 1906
1013 X_UNLOCK (reqlock); 1909 X_UNLOCK (reqlock);
1014 1910
1015 if (req->type < 0) 1911 if (req->type < 0)
1016 goto quit; 1912 goto quit;
1017 1913
1018 if (!EIO_CANCELLED (req))
1019 ETP_EXECUTE (self, req); 1914 ETP_EXECUTE (self, req);
1020 1915
1021 X_LOCK (reslock); 1916 X_LOCK (reslock);
1022 1917
1023 ++npending; 1918 ++npending;
1024 1919
1039 return 0; 1934 return 0;
1040} 1935}
1041 1936
1042/*****************************************************************************/ 1937/*****************************************************************************/
1043 1938
1939int ecb_cold
1044int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1940eio_init (void (*want_poll)(void), void (*done_poll)(void))
1045{ 1941{
1046 return etp_init (want_poll, done_poll); 1942 return etp_init (want_poll, done_poll);
1047} 1943}
1048 1944
1945ecb_inline void
1049static void eio_api_destroy (eio_req *req) 1946eio_api_destroy (eio_req *req)
1050{ 1947{
1051 free (req); 1948 free (req);
1052} 1949}
1053 1950
1054#define REQ(rtype) \ 1951#define REQ(rtype) \
1073 { \ 1970 { \
1074 eio_api_destroy (req); \ 1971 eio_api_destroy (req); \
1075 return 0; \ 1972 return 0; \
1076 } 1973 }
1077 1974
1975static void
1078static void eio_execute (etp_worker *self, eio_req *req) 1976eio_execute (etp_worker *self, eio_req *req)
1079{ 1977{
1080 errno = 0; 1978 if (ecb_expect_false (EIO_CANCELLED (req)))
1979 {
1980 req->result = -1;
1981 req->errorno = ECANCELED;
1982 return;
1983 }
1081 1984
1082 switch (req->type) 1985 switch (req->type)
1083 { 1986 {
1084 case EIO_READ: ALLOC (req->size); 1987 case EIO_READ: ALLOC (req->size);
1085 req->result = req->offs >= 0 1988 req->result = req->offs >= 0
1097 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2000 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1098 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2001 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1099 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2002 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1100 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2003 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1101 2004
2005 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2006 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2007 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2008 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2009
1102 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2010 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1103 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2011 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1104 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2012 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1105 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2013 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1106 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2014 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1113 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2021 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1114 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2022 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1115 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2023 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1116 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2024 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1117 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2025 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1118 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2026 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1119 2027
2028 case EIO_REALPATH: eio__realpath (req, self); break;
2029
1120 case EIO_READLINK: ALLOC (NAME_MAX); 2030 case EIO_READLINK: ALLOC (PATH_MAX);
1121 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2031 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1122 2032
1123 case EIO_SYNC: req->result = 0; sync (); break; 2033 case EIO_SYNC: req->result = 0; sync (); break;
1124 case EIO_FSYNC: req->result = fsync (req->int1); break; 2034 case EIO_FSYNC: req->result = fsync (req->int1); break;
1125 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2035 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2036 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2037 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2038 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2039 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2040 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2041 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1126 2042
1127 case EIO_READDIR: eio__scandir (req, self); break; 2043 case EIO_READDIR: eio__scandir (req, self); break;
1128 2044
1129 case EIO_BUSY: 2045 case EIO_BUSY:
1130#ifdef _WIN32 2046#ifdef _WIN32
1131 Sleep (req->nv1 * 1000.); 2047 Sleep (req->nv1 * 1e3);
1132#else 2048#else
1133 { 2049 {
1134 struct timeval tv; 2050 struct timeval tv;
1135 2051
1136 tv.tv_sec = req->nv1; 2052 tv.tv_sec = req->nv1;
1137 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2053 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1138 2054
1139 req->result = select (0, 0, 0, 0, &tv); 2055 req->result = select (0, 0, 0, 0, &tv);
1140 } 2056 }
1141#endif 2057#endif
1142 break; 2058 break;
1157 times = tv; 2073 times = tv;
1158 } 2074 }
1159 else 2075 else
1160 times = 0; 2076 times = 0;
1161 2077
1162
1163 req->result = req->type == EIO_FUTIME 2078 req->result = req->type == EIO_FUTIME
1164 ? futimes (req->int1, times) 2079 ? futimes (req->int1, times)
1165 : utimes (req->ptr1, times); 2080 : utimes (req->ptr1, times);
1166 } 2081 }
2082 break;
1167 2083
1168 case EIO_GROUP: 2084 case EIO_GROUP:
2085 abort (); /* handled in eio_request */
2086
1169 case EIO_NOP: 2087 case EIO_NOP:
1170 req->result = 0; 2088 req->result = 0;
1171 break; 2089 break;
1172 2090
1173 case EIO_CUSTOM: 2091 case EIO_CUSTOM:
1174 req->feed (req); 2092 req->feed (req);
1175 break; 2093 break;
1176 2094
1177 default: 2095 default:
2096 errno = ENOSYS;
1178 req->result = -1; 2097 req->result = -1;
1179 break; 2098 break;
1180 } 2099 }
1181 2100
1182 req->errorno = errno; 2101 req->errorno = errno;
1202eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 2121eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1203{ 2122{
1204 REQ (EIO_FSYNC); req->int1 = fd; SEND; 2123 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1205} 2124}
1206 2125
2126eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2127{
2128 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2129}
2130
2131eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2132{
2133 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2134}
2135
2136eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2137{
2138 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2139}
2140
2141eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2142{
2143 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2144}
2145
2146eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2147{
2148 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2149}
2150
2151eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2152{
2153 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2154}
2155
1207eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2156eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1208{ 2157{
1209 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2158 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1210} 2159}
1211 2160
1232eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2181eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1233{ 2182{
1234 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2183 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1235} 2184}
1236 2185
2186eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2187{
2188 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2189}
2190
1237eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2191eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1238{ 2192{
1239 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2193 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1240} 2194}
1241 2195
1247eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2201eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1248{ 2202{
1249 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2203 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1250} 2204}
1251 2205
1252eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2206eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1253{ 2207{
1254 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2208 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1255} 2209}
1256 2210
1257eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2211eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1277eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2231eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1278{ 2232{
1279 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2233 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1280} 2234}
1281 2235
1282eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2236eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1283{ 2237{
1284 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2238 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1285} 2239}
1286 2240
1287eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2241eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1303eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2257eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1304{ 2258{
1305 return eio__1path (EIO_READLINK, path, pri, cb, data); 2259 return eio__1path (EIO_READLINK, path, pri, cb, data);
1306} 2260}
1307 2261
2262eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2263{
2264 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2265}
2266
1308eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2267eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1309{ 2268{
1310 return eio__1path (EIO_STAT, path, pri, cb, data); 2269 return eio__1path (EIO_STAT, path, pri, cb, data);
1311} 2270}
1312 2271
1313eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2272eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1314{ 2273{
1315 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2274 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1316} 2275}
1317 2276
2277eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2278{
2279 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2280}
2281
1318eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2282eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1319{ 2283{
1320 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2284 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1321} 2285}
1322 2286
1323eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2287eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1324{ 2288{
1325 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2289 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1326} 2290}
1327 2291
1328eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2292eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1329{ 2293{
1330 return eio__1path (EIO_READDIR, path, pri, cb, data); 2294 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1331} 2295}
1332 2296
1333eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2297eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1334{ 2298{
1335 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2299 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1336} 2300}
1337 2301
1338static eio_req * 2302static eio_req *
1339eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2303eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1340{ 2304{
1364eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2328eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1365{ 2329{
1366 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2330 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1367} 2331}
1368 2332
1369eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2333eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1370{ 2334{
1371 REQ (EIO_CUSTOM); req->feed = execute; SEND; 2335 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1372} 2336}
1373 2337
1374#endif 2338#endif
1385#undef SEND 2349#undef SEND
1386 2350
1387/*****************************************************************************/ 2351/*****************************************************************************/
1388/* grp functions */ 2352/* grp functions */
1389 2353
2354void
1390void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2355eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1391{ 2356{
1392 grp->int2 = limit; 2357 grp->int2 = limit;
1393 grp->feed = feed; 2358 grp->feed = feed;
1394 2359
1395 grp_try_feed (grp); 2360 grp_try_feed (grp);
1396} 2361}
1397 2362
2363void
1398void eio_grp_limit (eio_req *grp, int limit) 2364eio_grp_limit (eio_req *grp, int limit)
1399{ 2365{
1400 grp->int2 = limit; 2366 grp->int2 = limit;
1401 2367
1402 grp_try_feed (grp); 2368 grp_try_feed (grp);
1403} 2369}
1404 2370
2371void
1405void eio_grp_add (eio_req *grp, eio_req *req) 2372eio_grp_add (eio_req *grp, eio_req *req)
1406{ 2373{
1407 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2374 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2375
2376 grp->flags |= EIO_FLAG_GROUPADD;
1408 2377
1409 ++grp->size; 2378 ++grp->size;
1410 req->grp = grp; 2379 req->grp = grp;
1411 2380
1412 req->grp_prev = 0; 2381 req->grp_prev = 0;
1419} 2388}
1420 2389
1421/*****************************************************************************/ 2390/*****************************************************************************/
1422/* misc garbage */ 2391/* misc garbage */
1423 2392
2393eio_ssize_t
1424ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2394eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1425{ 2395{
1426 etp_worker wrk; 2396 etp_worker wrk;
2397 eio_ssize_t ret;
1427 2398
1428 wrk.dbuf = 0; 2399 wrk.dbuf = 0;
1429 2400
1430 eio__sendfile (ofd, ifd, offset, count, &wrk); 2401 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1431 2402
1432 if (wrk.dbuf) 2403 if (wrk.dbuf)
1433 free (wrk.dbuf); 2404 free (wrk.dbuf);
1434}
1435 2405
2406 return ret;
2407}
2408

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines