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 1.33 by root, Sat Jun 6 19:44:17 2009 UTC vs.
Revision 1.88 by root, Thu Jul 14 19:31:43 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009 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"
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <limits.h> 59#include <limits.h>
55#include <fcntl.h> 60#include <fcntl.h>
56#include <assert.h> 61#include <assert.h>
57 62
63#include <sys/statvfs.h>
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H
67# include <stdint.h>
68#endif
69
70#ifndef ECANCELED
71# define ECANCELED EDOM
72#endif
73
74static void eio_destroy (eio_req *req);
75
58#ifndef EIO_FINISH 76#ifndef EIO_FINISH
59# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 77# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60#endif 78#endif
61 79
62#ifndef EIO_DESTROY 80#ifndef EIO_DESTROY
65 83
66#ifndef EIO_FEED 84#ifndef EIO_FEED
67# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 85# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68#endif 86#endif
69 87
88#ifndef EIO_FD_TO_WIN32_HANDLE
89# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
90#endif
91#ifndef EIO_WIN32_HANDLE_TO_FD
92# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
93#endif
94
95#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
96
97#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
98
70#ifdef _WIN32 99#ifdef _WIN32
71 100
72 /*doh*/ 101 #define PAGESIZE 4096 /* GetSystemInfo? */
102
103 #ifdef EIO_STRUCT_STATI64
104 #define stat(path,buf) _stati64 (path,buf)
105 #define fstat(fd,buf) _fstati64 (path,buf)
106 #endif
107 #define lstat(path,buf) stat (path,buf)
108 #define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
109 #define mkdir(path,mode) _mkdir (path)
110 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
111
112 #define chown(path,uid,gid) EIO_ENOSYS ()
113 #define fchown(fd,uid,gid) EIO_ENOSYS ()
114 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
115 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
116 #define mknod(path,mode,dev) EIO_ENOSYS ()
117 #define sync() EIO_ENOSYS ()
118
119 /* we could even stat and see if it exists */
120 static int
121 symlink (const char *old, const char *neu)
122 {
123 if (CreateSymbolicLink (neu, old, 1))
124 return 0;
125
126 if (CreateSymbolicLink (neu, old, 0))
127 return 0;
128
129 return EIO_ERRNO (ENOENT, -1);
130 }
131
73#else 132#else
74 133
75# include "config.h"
76# include <sys/time.h> 134 #include <sys/time.h>
77# include <sys/select.h> 135 #include <sys/select.h>
136 #include <sys/statvfs.h>
78# include <unistd.h> 137 #include <unistd.h>
79# include <utime.h> 138 #include <utime.h>
80# include <signal.h> 139 #include <signal.h>
81# include <dirent.h> 140 #include <dirent.h>
82 141
142 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
143 #include <sys/mman.h>
144 #endif
145
83/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 146 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) 147 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
85# define _DIRENT_HAVE_D_TYPE /* sigh */ 148 #define _DIRENT_HAVE_D_TYPE /* sigh */
86# define D_INO(de) (de)->d_fileno 149 #define D_INO(de) (de)->d_fileno
87# define D_NAMLEN(de) (de)->d_namlen 150 #define D_NAMLEN(de) (de)->d_namlen
88# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 151 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
89# define D_INO(de) (de)->d_ino 152 #define D_INO(de) (de)->d_ino
90# endif 153 #endif
91 154
92#ifdef _D_EXACT_NAMLEN 155 #ifdef _D_EXACT_NAMLEN
93# undef D_NAMLEN 156 #undef D_NAMLEN
94# define D_NAMLEN(de) _D_EXACT_NAMLEN (de) 157 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
95#endif 158 #endif
96 159
97# ifdef _DIRENT_HAVE_D_TYPE 160 #ifdef _DIRENT_HAVE_D_TYPE
98# define D_TYPE(de) (de)->d_type 161 #define D_TYPE(de) (de)->d_type
99# endif 162 #endif
100 163
101# ifndef EIO_STRUCT_DIRENT 164 #ifndef EIO_STRUCT_DIRENT
102# define EIO_STRUCT_DIRENT struct dirent 165 #define EIO_STRUCT_DIRENT struct dirent
103# endif 166 #endif
104 167
105#endif 168#endif
106 169
107#if HAVE_SENDFILE 170#if HAVE_SENDFILE
108# if __linux 171# if __linux
109# include <sys/sendfile.h> 172# include <sys/sendfile.h>
110# elif __freebsd 173# elif __FreeBSD__ || defined __APPLE__
111# include <sys/socket.h> 174# include <sys/socket.h>
112# include <sys/uio.h> 175# include <sys/uio.h>
113# elif __hpux 176# elif __hpux
114# include <sys/socket.h> 177# include <sys/socket.h>
115# elif __solaris /* not yet */ 178# elif __solaris
116# include <sys/sendfile.h> 179# include <sys/sendfile.h>
117# else 180# else
118# error sendfile support requested but not available 181# error sendfile support requested but not available
119# endif 182# endif
120#endif 183#endif
127#endif 190#endif
128#ifndef D_NAMLEN 191#ifndef D_NAMLEN
129# define D_NAMLEN(de) strlen ((de)->d_name) 192# define D_NAMLEN(de) strlen ((de)->d_name)
130#endif 193#endif
131 194
132/* number of seconds after which an idle threads exit */
133#define IDLE_TIMEOUT 10
134
135/* used for struct dirent, AIX doesn't provide it */ 195/* used for struct dirent, AIX doesn't provide it */
136#ifndef NAME_MAX 196#ifndef NAME_MAX
137# define NAME_MAX 4096 197# define NAME_MAX 4096
198#endif
199
200/* used for readlink etc. */
201#ifndef PATH_MAX
202# define PATH_MAX 4096
138#endif 203#endif
139 204
140/* buffer size for various temporary buffers */ 205/* buffer size for various temporary buffers */
141#define EIO_BUFSIZE 65536 206#define EIO_BUFSIZE 65536
142 207
148 errno = ENOMEM; \ 213 errno = ENOMEM; \
149 if (!eio_buf) \ 214 if (!eio_buf) \
150 return -1; 215 return -1;
151 216
152#define EIO_TICKS ((1000000 + 1023) >> 10) 217#define EIO_TICKS ((1000000 + 1023) >> 10)
153
154/*****************************************************************************/
155
156#if __GNUC__ >= 3
157# define expect(expr,value) __builtin_expect ((expr),(value))
158#else
159# define expect(expr,value) (expr)
160#endif
161
162#define expect_false(expr) expect ((expr) != 0, 0)
163#define expect_true(expr) expect ((expr) != 0, 1)
164
165/*****************************************************************************/
166 218
167#define ETP_PRI_MIN EIO_PRI_MIN 219#define ETP_PRI_MIN EIO_PRI_MIN
168#define ETP_PRI_MAX EIO_PRI_MAX 220#define ETP_PRI_MAX EIO_PRI_MAX
169 221
170struct etp_worker; 222struct etp_worker;
195 247
196/*****************************************************************************/ 248/*****************************************************************************/
197 249
198#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 250#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
199 251
200/* calculcate time difference in ~1/EIO_TICKS of a second */ 252/* calculate time difference in ~1/EIO_TICKS of a second */
253ecb_inline int
201static int tvdiff (struct timeval *tv1, struct timeval *tv2) 254tvdiff (struct timeval *tv1, struct timeval *tv2)
202{ 255{
203 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 256 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
204 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 257 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
205} 258}
206 259
213static unsigned int max_poll_reqs; /* reslock */ 266static unsigned int max_poll_reqs; /* reslock */
214 267
215static volatile unsigned int nreqs; /* reqlock */ 268static volatile unsigned int nreqs; /* reqlock */
216static volatile unsigned int nready; /* reqlock */ 269static volatile unsigned int nready; /* reqlock */
217static volatile unsigned int npending; /* reqlock */ 270static volatile unsigned int npending; /* reqlock */
218static volatile unsigned int max_idle = 4; 271static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
272static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
219 273
220static mutex_t wrklock = X_MUTEX_INIT; 274static xmutex_t wrklock;
221static mutex_t reslock = X_MUTEX_INIT; 275static xmutex_t reslock;
222static mutex_t reqlock = X_MUTEX_INIT; 276static xmutex_t reqlock;
223static cond_t reqwait = X_COND_INIT; 277static xcond_t reqwait;
224 278
225#if !HAVE_PREADWRITE 279#if !HAVE_PREADWRITE
226/* 280/*
227 * make our pread/pwrite emulation safe against themselves, but not against 281 * make our pread/pwrite emulation safe against themselves, but not against
228 * normal read/write by using a mutex. slows down execution a lot, 282 * normal read/write by using a mutex. slows down execution a lot,
229 * but that's your problem, not mine. 283 * but that's your problem, not mine.
230 */ 284 */
231static mutex_t preadwritelock = X_MUTEX_INIT; 285static xmutex_t preadwritelock = X_MUTEX_INIT;
232#endif 286#endif
233 287
234typedef struct etp_worker 288typedef struct etp_worker
235{ 289{
236 /* locked by wrklock */ 290 /* locked by wrklock */
237 struct etp_worker *prev, *next; 291 struct etp_worker *prev, *next;
238 292
239 thread_t tid; 293 xthread_t tid;
240 294
241 /* locked by reslock, reqlock or wrklock */ 295 /* locked by reslock, reqlock or wrklock */
242 ETP_REQ *req; /* currently processed request */ 296 ETP_REQ *req; /* currently processed request */
243 297
244 ETP_WORKER_COMMON 298 ETP_WORKER_COMMON
249#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 303#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
250#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 304#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
251 305
252/* worker threads management */ 306/* worker threads management */
253 307
308static void ecb_cold
254static void etp_worker_clear (etp_worker *wrk) 309etp_worker_clear (etp_worker *wrk)
255{ 310{
256 ETP_WORKER_CLEAR (wrk); 311 ETP_WORKER_CLEAR (wrk);
257} 312}
258 313
314static void ecb_cold
259static void etp_worker_free (etp_worker *wrk) 315etp_worker_free (etp_worker *wrk)
260{ 316{
261 wrk->next->prev = wrk->prev; 317 wrk->next->prev = wrk->prev;
262 wrk->prev->next = wrk->next; 318 wrk->prev->next = wrk->next;
263 319
264 free (wrk); 320 free (wrk);
265} 321}
266 322
267static unsigned int etp_nreqs (void) 323static unsigned int
324etp_nreqs (void)
268{ 325{
269 int retval; 326 int retval;
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 327 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = nreqs; 328 retval = nreqs;
272 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 329 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273 return retval; 330 return retval;
274} 331}
275 332
276static unsigned int etp_nready (void) 333static unsigned int
334etp_nready (void)
277{ 335{
278 unsigned int retval; 336 unsigned int retval;
279 337
280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 338 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281 retval = nready; 339 retval = nready;
282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 340 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283 341
284 return retval; 342 return retval;
285} 343}
286 344
287static unsigned int etp_npending (void) 345static unsigned int
346etp_npending (void)
288{ 347{
289 unsigned int retval; 348 unsigned int retval;
290 349
291 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 350 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
292 retval = npending; 351 retval = npending;
293 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 352 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
294 353
295 return retval; 354 return retval;
296} 355}
297 356
298static unsigned int etp_nthreads (void) 357static unsigned int
358etp_nthreads (void)
299{ 359{
300 unsigned int retval; 360 unsigned int retval;
301 361
302 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 362 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303 retval = started; 363 retval = started;
317} etp_reqq; 377} etp_reqq;
318 378
319static etp_reqq req_queue; 379static etp_reqq req_queue;
320static etp_reqq res_queue; 380static etp_reqq res_queue;
321 381
382static int ecb_noinline
322static int reqq_push (etp_reqq *q, ETP_REQ *req) 383reqq_push (etp_reqq *q, ETP_REQ *req)
323{ 384{
324 int pri = req->pri; 385 int pri = req->pri;
325 req->next = 0; 386 req->next = 0;
326 387
327 if (q->qe[pri]) 388 if (q->qe[pri])
333 q->qe[pri] = q->qs[pri] = req; 394 q->qe[pri] = q->qs[pri] = req;
334 395
335 return q->size++; 396 return q->size++;
336} 397}
337 398
399static ETP_REQ * ecb_noinline
338static ETP_REQ *reqq_shift (etp_reqq *q) 400reqq_shift (etp_reqq *q)
339{ 401{
340 int pri; 402 int pri;
341 403
342 if (!q->size) 404 if (!q->size)
343 return 0; 405 return 0;
358 } 420 }
359 421
360 abort (); 422 abort ();
361} 423}
362 424
363static void etp_atfork_prepare (void) 425static void ecb_cold
426etp_thread_init (void)
364{ 427{
365 X_LOCK (wrklock);
366 X_LOCK (reqlock);
367 X_LOCK (reslock);
368#if !HAVE_PREADWRITE 428#if !HAVE_PREADWRITE
369 X_LOCK (preadwritelock); 429 X_MUTEX_CREATE (preadwritelock);
370#endif 430#endif
431 X_MUTEX_CREATE (wrklock);
432 X_MUTEX_CREATE (reslock);
433 X_MUTEX_CREATE (reqlock);
434 X_COND_CREATE (reqwait);
371} 435}
372 436
437static void ecb_cold
438etp_atfork_prepare (void)
439{
440}
441
442static void ecb_cold
373static void etp_atfork_parent (void) 443etp_atfork_parent (void)
374{ 444{
375#if !HAVE_PREADWRITE
376 X_UNLOCK (preadwritelock);
377#endif
378 X_UNLOCK (reslock);
379 X_UNLOCK (reqlock);
380 X_UNLOCK (wrklock);
381} 445}
382 446
447static void ecb_cold
383static void etp_atfork_child (void) 448etp_atfork_child (void)
384{ 449{
385 ETP_REQ *prv; 450 ETP_REQ *prv;
386 451
387 while ((prv = reqq_shift (&req_queue))) 452 while ((prv = reqq_shift (&req_queue)))
388 ETP_DESTROY (prv); 453 ETP_DESTROY (prv);
405 idle = 0; 470 idle = 0;
406 nreqs = 0; 471 nreqs = 0;
407 nready = 0; 472 nready = 0;
408 npending = 0; 473 npending = 0;
409 474
410 etp_atfork_parent (); 475 etp_thread_init ();
411} 476}
412 477
413static void 478static void ecb_cold
414etp_once_init (void) 479etp_once_init (void)
415{ 480{
481 etp_thread_init ();
416 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 482 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
417} 483}
418 484
419static int 485static int ecb_cold
420etp_init (void (*want_poll)(void), void (*done_poll)(void)) 486etp_init (void (*want_poll)(void), void (*done_poll)(void))
421{ 487{
422 static pthread_once_t doinit = PTHREAD_ONCE_INIT; 488 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
423 489
424 pthread_once (&doinit, etp_once_init); 490 pthread_once (&doinit, etp_once_init);
429 return 0; 495 return 0;
430} 496}
431 497
432X_THREAD_PROC (etp_proc); 498X_THREAD_PROC (etp_proc);
433 499
500static void ecb_cold
434static void etp_start_thread (void) 501etp_start_thread (void)
435{ 502{
436 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 503 etp_worker *wrk = calloc (1, sizeof (etp_worker));
437 504
438 /*TODO*/ 505 /*TODO*/
439 assert (("unable to allocate worker thread data", wrk)); 506 assert (("unable to allocate worker thread data", wrk));
452 free (wrk); 519 free (wrk);
453 520
454 X_UNLOCK (wrklock); 521 X_UNLOCK (wrklock);
455} 522}
456 523
524static void
457static void etp_maybe_start_thread (void) 525etp_maybe_start_thread (void)
458{ 526{
459 if (expect_true (etp_nthreads () >= wanted)) 527 if (ecb_expect_true (etp_nthreads () >= wanted))
460 return; 528 return;
461 529
462 /* todo: maybe use idle here, but might be less exact */ 530 /* todo: maybe use idle here, but might be less exact */
463 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 531 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
464 return; 532 return;
465 533
466 etp_start_thread (); 534 etp_start_thread ();
467} 535}
468 536
537static void ecb_cold
469static void etp_end_thread (void) 538etp_end_thread (void)
470{ 539{
471 eio_req *req = calloc (1, sizeof (eio_req)); 540 eio_req *req = calloc (1, sizeof (eio_req));
472 541
473 req->type = -1; 542 req->type = -1;
474 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 543 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
481 X_LOCK (wrklock); 550 X_LOCK (wrklock);
482 --started; 551 --started;
483 X_UNLOCK (wrklock); 552 X_UNLOCK (wrklock);
484} 553}
485 554
486static int etp_poll (void) 555static int
556etp_poll (void)
487{ 557{
488 unsigned int maxreqs; 558 unsigned int maxreqs;
489 unsigned int maxtime; 559 unsigned int maxtime;
490 struct timeval tv_start, tv_now; 560 struct timeval tv_start, tv_now;
491 561
521 591
522 X_LOCK (reqlock); 592 X_LOCK (reqlock);
523 --nreqs; 593 --nreqs;
524 X_UNLOCK (reqlock); 594 X_UNLOCK (reqlock);
525 595
526 if (expect_false (req->type == EIO_GROUP && req->size)) 596 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
527 { 597 {
528 req->int1 = 1; /* mark request as delayed */ 598 req->int1 = 1; /* mark request as delayed */
529 continue; 599 continue;
530 } 600 }
531 else 601 else
532 { 602 {
533 int res = ETP_FINISH (req); 603 int res = ETP_FINISH (req);
534 if (expect_false (res)) 604 if (ecb_expect_false (res))
535 return res; 605 return res;
536 } 606 }
537 607
538 if (expect_false (maxreqs && !--maxreqs)) 608 if (ecb_expect_false (maxreqs && !--maxreqs))
539 break; 609 break;
540 610
541 if (maxtime) 611 if (maxtime)
542 { 612 {
543 gettimeofday (&tv_now, 0); 613 gettimeofday (&tv_now, 0);
549 619
550 errno = EAGAIN; 620 errno = EAGAIN;
551 return -1; 621 return -1;
552} 622}
553 623
624static void
554static void etp_cancel (ETP_REQ *req) 625etp_cancel (ETP_REQ *req)
555{ 626{
556 X_LOCK (wrklock); 627 req->cancelled = 1;
557 req->flags |= EIO_FLAG_CANCELLED;
558 X_UNLOCK (wrklock);
559 628
560 eio_grp_cancel (req); 629 eio_grp_cancel (req);
561} 630}
562 631
632static void
563static void etp_submit (ETP_REQ *req) 633etp_submit (ETP_REQ *req)
564{ 634{
565 req->pri -= ETP_PRI_MIN; 635 req->pri -= ETP_PRI_MIN;
566 636
567 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 637 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
568 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 638 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
569 639
570 if (expect_false (req->type == EIO_GROUP)) 640 if (ecb_expect_false (req->type == EIO_GROUP))
571 { 641 {
572 /* I hope this is worth it :/ */ 642 /* I hope this is worth it :/ */
573 X_LOCK (reqlock); 643 X_LOCK (reqlock);
574 ++nreqs; 644 ++nreqs;
575 X_UNLOCK (reqlock); 645 X_UNLOCK (reqlock);
594 664
595 etp_maybe_start_thread (); 665 etp_maybe_start_thread ();
596 } 666 }
597} 667}
598 668
669static void ecb_cold
599static void etp_set_max_poll_time (double nseconds) 670etp_set_max_poll_time (double nseconds)
600{ 671{
601 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 672 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
602 max_poll_time = nseconds; 673 max_poll_time = nseconds * EIO_TICKS;
603 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 674 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
604} 675}
605 676
677static void ecb_cold
606static void etp_set_max_poll_reqs (unsigned int maxreqs) 678etp_set_max_poll_reqs (unsigned int maxreqs)
607{ 679{
608 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 680 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609 max_poll_reqs = maxreqs; 681 max_poll_reqs = maxreqs;
610 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 682 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611} 683}
612 684
685static void ecb_cold
613static void etp_set_max_idle (unsigned int nthreads) 686etp_set_max_idle (unsigned int nthreads)
614{ 687{
615 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 688 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
616 max_idle = nthreads <= 0 ? 1 : nthreads; 689 max_idle = nthreads;
617 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 690 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
618} 691}
619 692
693static void ecb_cold
694etp_set_idle_timeout (unsigned int seconds)
695{
696 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
697 idle_timeout = seconds;
698 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
699}
700
701static void ecb_cold
620static void etp_set_min_parallel (unsigned int nthreads) 702etp_set_min_parallel (unsigned int nthreads)
621{ 703{
622 if (wanted < nthreads) 704 if (wanted < nthreads)
623 wanted = nthreads; 705 wanted = nthreads;
624} 706}
625 707
708static void ecb_cold
626static void etp_set_max_parallel (unsigned int nthreads) 709etp_set_max_parallel (unsigned int nthreads)
627{ 710{
628 if (wanted > nthreads) 711 if (wanted > nthreads)
629 wanted = nthreads; 712 wanted = nthreads;
630 713
631 while (started > wanted) 714 while (started > wanted)
632 etp_end_thread (); 715 etp_end_thread ();
633} 716}
634 717
635/*****************************************************************************/ 718/*****************************************************************************/
636 719
720static void
637static void grp_try_feed (eio_req *grp) 721grp_try_feed (eio_req *grp)
638{ 722{
639 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 723 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
640 { 724 {
641 grp->flags &= ~EIO_FLAG_GROUPADD; 725 grp->flags &= ~EIO_FLAG_GROUPADD;
642 726
649 break; 733 break;
650 } 734 }
651 } 735 }
652} 736}
653 737
738static int
654static int grp_dec (eio_req *grp) 739grp_dec (eio_req *grp)
655{ 740{
656 --grp->size; 741 --grp->size;
657 742
658 /* call feeder, if applicable */ 743 /* call feeder, if applicable */
659 grp_try_feed (grp); 744 grp_try_feed (grp);
663 return eio_finish (grp); 748 return eio_finish (grp);
664 else 749 else
665 return 0; 750 return 0;
666} 751}
667 752
753static void
668void eio_destroy (eio_req *req) 754eio_destroy (eio_req *req)
669{ 755{
670 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 756 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
671 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 757 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
672 758
673 EIO_DESTROY (req); 759 EIO_DESTROY (req);
674} 760}
675 761
762static int
676static int eio_finish (eio_req *req) 763eio_finish (eio_req *req)
677{ 764{
678 int res = EIO_FINISH (req); 765 int res = EIO_FINISH (req);
679 766
680 if (req->grp) 767 if (req->grp)
681 { 768 {
689 if (grp->grp_first == req) 776 if (grp->grp_first == req)
690 grp->grp_first = req->grp_next; 777 grp->grp_first = req->grp_next;
691 778
692 res2 = grp_dec (grp); 779 res2 = grp_dec (grp);
693 780
694 if (!res && res2) 781 if (!res)
695 res = res2; 782 res = res2;
696 } 783 }
697 784
698 eio_destroy (req); 785 eio_destroy (req);
699 786
700 return res; 787 return res;
701} 788}
702 789
790void
703void eio_grp_cancel (eio_req *grp) 791eio_grp_cancel (eio_req *grp)
704{ 792{
705 for (grp = grp->grp_first; grp; grp = grp->grp_next) 793 for (grp = grp->grp_first; grp; grp = grp->grp_next)
706 eio_cancel (grp); 794 eio_cancel (grp);
707} 795}
708 796
797void
709void eio_cancel (eio_req *req) 798eio_cancel (eio_req *req)
710{ 799{
711 etp_cancel (req); 800 etp_cancel (req);
712} 801}
713 802
803void
714void eio_submit (eio_req *req) 804eio_submit (eio_req *req)
715{ 805{
716 etp_submit (req); 806 etp_submit (req);
717} 807}
718 808
719unsigned int eio_nreqs (void) 809unsigned int
810eio_nreqs (void)
720{ 811{
721 return etp_nreqs (); 812 return etp_nreqs ();
722} 813}
723 814
724unsigned int eio_nready (void) 815unsigned int
816eio_nready (void)
725{ 817{
726 return etp_nready (); 818 return etp_nready ();
727} 819}
728 820
729unsigned int eio_npending (void) 821unsigned int
822eio_npending (void)
730{ 823{
731 return etp_npending (); 824 return etp_npending ();
732} 825}
733 826
734unsigned int eio_nthreads (void) 827unsigned int ecb_cold
828eio_nthreads (void)
735{ 829{
736 return etp_nthreads (); 830 return etp_nthreads ();
737} 831}
738 832
833void ecb_cold
739void eio_set_max_poll_time (double nseconds) 834eio_set_max_poll_time (double nseconds)
740{ 835{
741 etp_set_max_poll_time (nseconds); 836 etp_set_max_poll_time (nseconds);
742} 837}
743 838
839void ecb_cold
744void eio_set_max_poll_reqs (unsigned int maxreqs) 840eio_set_max_poll_reqs (unsigned int maxreqs)
745{ 841{
746 etp_set_max_poll_reqs (maxreqs); 842 etp_set_max_poll_reqs (maxreqs);
747} 843}
748 844
845void ecb_cold
749void eio_set_max_idle (unsigned int nthreads) 846eio_set_max_idle (unsigned int nthreads)
750{ 847{
751 etp_set_max_idle (nthreads); 848 etp_set_max_idle (nthreads);
752} 849}
753 850
851void ecb_cold
852eio_set_idle_timeout (unsigned int seconds)
853{
854 etp_set_idle_timeout (seconds);
855}
856
857void ecb_cold
754void eio_set_min_parallel (unsigned int nthreads) 858eio_set_min_parallel (unsigned int nthreads)
755{ 859{
756 etp_set_min_parallel (nthreads); 860 etp_set_min_parallel (nthreads);
757} 861}
758 862
863void ecb_cold
759void eio_set_max_parallel (unsigned int nthreads) 864eio_set_max_parallel (unsigned int nthreads)
760{ 865{
761 etp_set_max_parallel (nthreads); 866 etp_set_max_parallel (nthreads);
762} 867}
763 868
764int eio_poll (void) 869int eio_poll (void)
773# undef pread 878# undef pread
774# undef pwrite 879# undef pwrite
775# define pread eio__pread 880# define pread eio__pread
776# define pwrite eio__pwrite 881# define pwrite eio__pwrite
777 882
778static ssize_t 883static eio_ssize_t
779eio__pread (int fd, void *buf, size_t count, off_t offset) 884eio__pread (int fd, void *buf, size_t count, off_t offset)
780{ 885{
781 ssize_t res; 886 eio_ssize_t res;
782 off_t ooffset; 887 off_t ooffset;
783 888
784 X_LOCK (preadwritelock); 889 X_LOCK (preadwritelock);
785 ooffset = lseek (fd, 0, SEEK_CUR); 890 ooffset = lseek (fd, 0, SEEK_CUR);
786 lseek (fd, offset, SEEK_SET); 891 lseek (fd, offset, SEEK_SET);
789 X_UNLOCK (preadwritelock); 894 X_UNLOCK (preadwritelock);
790 895
791 return res; 896 return res;
792} 897}
793 898
794static ssize_t 899static eio_ssize_t
795eio__pwrite (int fd, void *buf, size_t count, off_t offset) 900eio__pwrite (int fd, void *buf, size_t count, off_t offset)
796{ 901{
797 ssize_t res; 902 eio_ssize_t res;
798 off_t ooffset; 903 off_t ooffset;
799 904
800 X_LOCK (preadwritelock); 905 X_LOCK (preadwritelock);
801 ooffset = lseek (fd, 0, SEEK_CUR); 906 ooffset = lseek (fd, 0, SEEK_CUR);
802 lseek (fd, offset, SEEK_SET); 907 lseek (fd, offset, SEEK_SET);
806 911
807 return res; 912 return res;
808} 913}
809#endif 914#endif
810 915
811#ifndef HAVE_FUTIMES 916#ifndef HAVE_UTIMES
812 917
813# undef utimes 918# undef utimes
814# undef futimes
815# define utimes(path,times) eio__utimes (path, times) 919# define utimes(path,times) eio__utimes (path, times)
816# define futimes(fd,times) eio__futimes (fd, times)
817 920
818static int 921static int
819eio__utimes (const char *filename, const struct timeval times[2]) 922eio__utimes (const char *filename, const struct timeval times[2])
820{ 923{
821 if (times) 924 if (times)
829 } 932 }
830 else 933 else
831 return utime (filename, 0); 934 return utime (filename, 0);
832} 935}
833 936
937#endif
938
939#ifndef HAVE_FUTIMES
940
941# undef futimes
942# define futimes(fd,times) eio__futimes (fd, times)
943
944static int
834static int eio__futimes (int fd, const struct timeval tv[2]) 945eio__futimes (int fd, const struct timeval tv[2])
835{ 946{
836 errno = ENOSYS; 947 errno = ENOSYS;
837 return -1; 948 return -1;
838} 949}
839 950
843# undef fdatasync 954# undef fdatasync
844# define fdatasync(fd) fsync (fd) 955# define fdatasync(fd) fsync (fd)
845#endif 956#endif
846 957
847/* sync_file_range always needs emulation */ 958/* sync_file_range always needs emulation */
848int 959static int
849eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 960eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
850{ 961{
851#if HAVE_SYNC_FILE_RANGE 962#if HAVE_SYNC_FILE_RANGE
852 int res; 963 int res;
853 964
866 if (!res || errno != ENOSYS) 977 if (!res || errno != ENOSYS)
867 return res; 978 return res;
868#endif 979#endif
869 980
870 /* even though we could play tricks with the flags, it's better to always 981 /* even though we could play tricks with the flags, it's better to always
871 * call fdatasync, as thta matches the expectation of it's users best */ 982 * call fdatasync, as that matches the expectation of its users best */
872 return fdatasync (fd); 983 return fdatasync (fd);
984}
985
986static int
987eio__fallocate (int fd, int mode, off_t offset, size_t len)
988{
989#if HAVE_FALLOCATE
990 return fallocate (fd, mode, offset, len);
991#else
992 errno = ENOSYS;
993 return -1;
994#endif
873} 995}
874 996
875#if !HAVE_READAHEAD 997#if !HAVE_READAHEAD
876# undef readahead 998# undef readahead
877# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 999# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
878 1000
879static ssize_t 1001static eio_ssize_t
880eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1002eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
881{ 1003{
882 size_t todo = count; 1004 size_t todo = count;
883 dBUF; 1005 dBUF;
884 1006
896} 1018}
897 1019
898#endif 1020#endif
899 1021
900/* sendfile always needs emulation */ 1022/* sendfile always needs emulation */
901static ssize_t 1023static eio_ssize_t
902eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1024eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
903{ 1025{
1026 eio_ssize_t written = 0;
904 ssize_t res; 1027 eio_ssize_t res;
905 1028
906 if (!count) 1029 if (!count)
907 return 0; 1030 return 0;
908 1031
1032 for (;;)
1033 {
1034#ifdef __APPLE__
1035# undef HAVE_SENDFILE /* broken, as everything on os x */
1036#endif
909#if HAVE_SENDFILE 1037#if HAVE_SENDFILE
910# if __linux 1038# if __linux
1039 off_t soffset = offset;
911 res = sendfile (ofd, ifd, &offset, count); 1040 res = sendfile (ofd, ifd, &soffset, count);
912 1041
913# elif __freebsd 1042# elif __FreeBSD__
914 /* 1043 /*
915 * Of course, the freebsd sendfile is a dire hack with no thoughts 1044 * Of course, the freebsd sendfile is a dire hack with no thoughts
916 * wasted on making it similar to other I/O functions. 1045 * wasted on making it similar to other I/O functions.
917 */ 1046 */
918 {
919 off_t sbytes; 1047 off_t sbytes;
920 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1048 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
921 1049
922 if (res < 0 && sbytes) 1050 #if 0 /* according to the manpage, this is correct, but broken behaviour */
923 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1051 /* freebsd' sendfile will return 0 on success */
1052 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1053 /* not on e.g. EIO or EPIPE - sounds broken */
1054 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
924 res = sbytes; 1055 res = sbytes;
925 } 1056 #endif
1057
1058 /* according to source inspection, this is correct, and useful behaviour */
1059 if (sbytes)
1060 res = sbytes;
1061
1062# elif defined (__APPLE__)
1063 off_t sbytes = count;
1064 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1065
1066 /* according to the manpage, sbytes is always valid */
1067 if (sbytes)
1068 res = sbytes;
926 1069
927# elif __hpux 1070# elif __hpux
928 res = sendfile (ofd, ifd, offset, count, 0, 0); 1071 res = sendfile (ofd, ifd, offset, count, 0, 0);
929 1072
930# elif __solaris 1073# elif __solaris
931 {
932 struct sendfilevec vec; 1074 struct sendfilevec vec;
933 size_t sbytes; 1075 size_t sbytes;
934 1076
935 vec.sfv_fd = ifd; 1077 vec.sfv_fd = ifd;
936 vec.sfv_flag = 0; 1078 vec.sfv_flag = 0;
937 vec.sfv_off = offset; 1079 vec.sfv_off = offset;
938 vec.sfv_len = count; 1080 vec.sfv_len = count;
939 1081
940 res = sendfilev (ofd, &vec, 1, &sbytes); 1082 res = sendfilev (ofd, &vec, 1, &sbytes);
941 1083
942 if (res < 0 && sbytes) 1084 if (res < 0 && sbytes)
943 res = sbytes; 1085 res = sbytes;
944 }
945 1086
946# endif 1087# endif
1088
1089#elif defined (_WIN32) && 0
1090 /* does not work, just for documentation of what would need to be done */
1091 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1092 /* libeio guarantees that the file offset does not change, and windows */
1093 /* has no way to get an independent handle to the same file description */
1094 HANDLE h = TO_SOCKET (ifd);
1095 SetFilePointer (h, offset, 0, FILE_BEGIN);
1096 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1097
947#else 1098#else
948 res = -1; 1099 res = -1;
949 errno = ENOSYS; 1100 errno = ENOSYS;
950#endif 1101#endif
951 1102
1103 /* we assume sendfile can copy at least 128mb in one go */
1104 if (res <= 128 * 1024 * 1024)
1105 {
1106 if (res > 0)
1107 written += res;
1108
1109 if (written)
1110 return written;
1111
1112 break;
1113 }
1114 else
1115 {
1116 /* if we requested more, then probably the kernel was lazy */
1117 written += res;
1118 offset += res;
1119 count -= res;
1120
1121 if (!count)
1122 return written;
1123 }
1124 }
1125
952 if (res < 0 1126 if (res < 0
953 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1127 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1128 /* BSDs */
1129#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1130 || errno == ENOTSUP
1131#endif
1132 || errno == EOPNOTSUPP /* BSDs */
954#if __solaris 1133#if __solaris
955 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1134 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
956#endif 1135#endif
957 ) 1136 )
958 ) 1137 )
962 1141
963 res = 0; 1142 res = 0;
964 1143
965 while (count) 1144 while (count)
966 { 1145 {
967 ssize_t cnt; 1146 eio_ssize_t cnt;
968 1147
969 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1148 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
970 1149
971 if (cnt <= 0) 1150 if (cnt <= 0)
972 { 1151 {
989 } 1168 }
990 1169
991 return res; 1170 return res;
992} 1171}
993 1172
1173#ifdef PAGESIZE
1174# define eio_pagesize() PAGESIZE
1175#else
1176static intptr_t
1177eio_pagesize (void)
1178{
1179 static intptr_t page;
1180
1181 if (!page)
1182 page = sysconf (_SC_PAGESIZE);
1183
1184 return page;
1185}
1186#endif
1187
1188static void
1189eio_page_align (void **addr, size_t *length)
1190{
1191 intptr_t mask = eio_pagesize () - 1;
1192
1193 /* round down addr */
1194 intptr_t adj = mask & (intptr_t)*addr;
1195
1196 *addr = (void *)((intptr_t)*addr - adj);
1197 *length += adj;
1198
1199 /* round up length */
1200 *length = (*length + mask) & ~mask;
1201}
1202
1203#if !_POSIX_MEMLOCK
1204# define eio__mlockall(a) eio_nosyscall()
1205#else
1206
994static int 1207static int
1208eio__mlockall (int flags)
1209{
1210 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1211 extern int mallopt (int, int);
1212 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1213 #endif
1214
1215 if (EIO_MCL_CURRENT != MCL_CURRENT
1216 || EIO_MCL_FUTURE != MCL_FUTURE)
1217 {
1218 flags = 0
1219 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1220 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1221 }
1222
1223 return mlockall (flags);
1224}
1225#endif
1226
1227#if !_POSIX_MEMLOCK_RANGE
1228# define eio__mlock(a,b) EIO_ENOSYS ()
1229#else
1230
1231static int
1232eio__mlock (void *addr, size_t length)
1233{
1234 eio_page_align (&addr, &length);
1235
1236 return mlock (addr, length);
1237}
1238
1239#endif
1240
1241#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1242# define eio__msync(a,b,c) EIO_ENOSYS ()
1243#else
1244
1245static int
1246eio__msync (void *mem, size_t len, int flags)
1247{
1248 eio_page_align (&mem, &len);
1249
1250 if (EIO_MS_ASYNC != MS_SYNC
1251 || EIO_MS_INVALIDATE != MS_INVALIDATE
1252 || EIO_MS_SYNC != MS_SYNC)
1253 {
1254 flags = 0
1255 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1256 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1257 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1258 }
1259
1260 return msync (mem, len, flags);
1261}
1262
1263#endif
1264
1265static int
1266eio__mtouch (eio_req *req)
1267{
1268 void *mem = req->ptr2;
1269 size_t len = req->size;
1270 int flags = req->int1;
1271
1272 eio_page_align (&mem, &len);
1273
1274 {
1275 intptr_t addr = (intptr_t)mem;
1276 intptr_t end = addr + len;
1277 intptr_t page = eio_pagesize ();
1278
1279 if (addr < end)
1280 if (flags & EIO_MT_MODIFY) /* modify */
1281 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1282 else
1283 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1284 }
1285
1286 return 0;
1287}
1288
1289/*****************************************************************************/
1290/* requests implemented outside eio_execute, because they are so large */
1291
1292static void
1293eio__realpath (eio_req *req, etp_worker *self)
1294{
1295 char *rel = req->ptr1;
1296 char *res;
1297 char *tmp1, *tmp2;
1298#if SYMLOOP_MAX > 32
1299 int symlinks = SYMLOOP_MAX;
1300#else
1301 int symlinks = 32;
1302#endif
1303
1304 req->result = -1;
1305
1306 errno = EINVAL;
1307 if (!rel)
1308 return;
1309
1310 errno = ENOENT;
1311 if (!*rel)
1312 return;
1313
1314 if (!req->ptr2)
1315 {
1316 X_LOCK (wrklock);
1317 req->flags |= EIO_FLAG_PTR2_FREE;
1318 X_UNLOCK (wrklock);
1319 req->ptr2 = malloc (PATH_MAX * 3);
1320
1321 errno = ENOMEM;
1322 if (!req->ptr2)
1323 return;
1324 }
1325
1326 res = req->ptr2;
1327 tmp1 = res + PATH_MAX;
1328 tmp2 = tmp1 + PATH_MAX;
1329
1330#if 0 /* disabled, the musl way to do things is just too racy */
1331#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1332 /* on linux we may be able to ask the kernel */
1333 {
1334 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1335
1336 if (fd >= 0)
1337 {
1338 sprintf (tmp1, "/proc/self/fd/%d", fd);
1339 req->result = readlink (tmp1, res, PATH_MAX);
1340 close (fd);
1341
1342 /* here we should probably stat the open file and the disk file, to make sure they still match */
1343
1344 if (req->result > 0)
1345 goto done;
1346 }
1347 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1348 return;
1349 }
1350#endif
1351#endif
1352
1353 if (*rel != '/')
1354 {
1355 if (!getcwd (res, PATH_MAX))
1356 return;
1357
1358 if (res [1]) /* only use if not / */
1359 res += strlen (res);
1360 }
1361
1362 while (*rel)
1363 {
1364 eio_ssize_t len, linklen;
1365 char *beg = rel;
1366
1367 while (*rel && *rel != '/')
1368 ++rel;
1369
1370 len = rel - beg;
1371
1372 if (!len) /* skip slashes */
1373 {
1374 ++rel;
1375 continue;
1376 }
1377
1378 if (beg [0] == '.')
1379 {
1380 if (len == 1)
1381 continue; /* . - nop */
1382
1383 if (beg [1] == '.' && len == 2)
1384 {
1385 /* .. - back up one component, if possible */
1386
1387 while (res != req->ptr2)
1388 if (*--res == '/')
1389 break;
1390
1391 continue;
1392 }
1393 }
1394
1395 errno = ENAMETOOLONG;
1396 if (res + 1 + len + 1 >= tmp1)
1397 return;
1398
1399 /* copy one component */
1400 *res = '/';
1401 memcpy (res + 1, beg, len);
1402
1403 /* zero-terminate, for readlink */
1404 res [len + 1] = 0;
1405
1406 /* now check if it's a symlink */
1407 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1408
1409 if (linklen < 0)
1410 {
1411 if (errno != EINVAL)
1412 return;
1413
1414 /* it's a normal directory. hopefully */
1415 res += len + 1;
1416 }
1417 else
1418 {
1419 /* yay, it was a symlink - build new path in tmp2 */
1420 int rellen = strlen (rel);
1421
1422 errno = ENAMETOOLONG;
1423 if (linklen + 1 + rellen >= PATH_MAX)
1424 return;
1425
1426 errno = ELOOP;
1427 if (!--symlinks)
1428 return;
1429
1430 if (*tmp1 == '/')
1431 res = req->ptr2; /* symlink resolves to an absolute path */
1432
1433 /* we need to be careful, as rel might point into tmp2 already */
1434 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1435 tmp2 [linklen] = '/';
1436 memcpy (tmp2, tmp1, linklen);
1437
1438 rel = tmp2;
1439 }
1440 }
1441
1442 /* special case for the lone root path */
1443 if (res == req->ptr2)
1444 *res++ = '/';
1445
1446 req->result = res - (char *)req->ptr2;
1447
1448done:
1449 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1450}
1451
1452static signed char
995eio_dent_cmp (const void *a_, const void *b_) 1453eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
996{ 1454{
997 const eio_dirent *a = (const eio_dirent *)a_; 1455 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
998 const eio_dirent *b = (const eio_dirent *)b_; 1456 : a->inode < b->inode ? -1
1457 : a->inode > b->inode ? 1
1458 : 0;
1459}
999 1460
1000 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score 1461#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1001 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; /* int might be < ino_t */ 1462
1463#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1464#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1465
1466static void
1467eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1468{
1469 unsigned char bits [9 + sizeof (ino_t) * 8];
1470 unsigned char *bit = bits;
1471
1472 assert (CHAR_BIT == 8);
1473 assert (sizeof (eio_dirent) * 8 < 256);
1474 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1475 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1476
1477 if (size <= EIO_SORT_FAST)
1478 return;
1479
1480 /* first prepare an array of bits to test in our radix sort */
1481 /* try to take endianness into account, as well as differences in ino_t sizes */
1482 /* inode_bits must contain all inodes ORed together */
1483 /* which is used to skip bits that are 0 everywhere, which is very common */
1484 {
1485 ino_t endianness;
1486 int i, j;
1487
1488 /* we store the byte offset of byte n into byte n of "endianness" */
1489 for (i = 0; i < sizeof (ino_t); ++i)
1490 ((unsigned char *)&endianness)[i] = i;
1491
1492 *bit++ = 0;
1493
1494 for (i = 0; i < sizeof (ino_t); ++i)
1495 {
1496 /* shifting off the byte offsets out of "endianness" */
1497 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1498 endianness >>= 8;
1499
1500 for (j = 0; j < 8; ++j)
1501 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1502 *bit++ = offs + j;
1503 }
1504
1505 for (j = 0; j < 8; ++j)
1506 if (score_bits & (1 << j))
1507 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1508 }
1509
1510 /* now actually do the sorting (a variant of MSD radix sort) */
1511 {
1512 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1513 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1514 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1515 int stk_idx = 0;
1516
1517 base_stk [stk_idx] = dents;
1518 end_stk [stk_idx] = dents + size;
1519 bit_stk [stk_idx] = bit - 1;
1520
1521 do
1522 {
1523 base = base_stk [stk_idx];
1524 end = end_stk [stk_idx];
1525 bit = bit_stk [stk_idx];
1526
1527 for (;;)
1528 {
1529 unsigned char O = *bit >> 3;
1530 unsigned char M = 1 << (*bit & 7);
1531
1532 eio_dirent *a = base;
1533 eio_dirent *b = end;
1534
1535 if (b - a < EIO_SORT_CUTOFF)
1536 break;
1537
1538 /* now bit-partition the array on the bit */
1539 /* this ugly asymmetric loop seems to perform much better than typical */
1540 /* partition algos found in the literature */
1541 do
1542 if (!(((unsigned char *)a)[O] & M))
1543 ++a;
1544 else if (!(((unsigned char *)--b)[O] & M))
1545 {
1546 eio_dirent tmp = *a; *a = *b; *b = tmp;
1547 ++a;
1548 }
1549 while (b > a);
1550
1551 /* next bit, or stop, if no bits left in this path */
1552 if (!*--bit)
1553 break;
1554
1555 base_stk [stk_idx] = a;
1556 end_stk [stk_idx] = end;
1557 bit_stk [stk_idx] = bit;
1558 ++stk_idx;
1559
1560 end = a;
1561 }
1562 }
1563 while (stk_idx--);
1564 }
1565}
1566
1567static void
1568eio_dent_insertion_sort (eio_dirent *dents, int size)
1569{
1570 /* first move the smallest element to the front, to act as a sentinel */
1571 {
1572 int i;
1573 eio_dirent *min = dents;
1574
1575 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1576 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1577 if (EIO_DENT_CMP (dents [i], <, *min))
1578 min = &dents [i];
1579
1580 /* swap elements 0 and j (minimum) */
1581 {
1582 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1583 }
1584 }
1585
1586 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1587 {
1588 eio_dirent *i, *j;
1589
1590 for (i = dents + 1; i < dents + size; ++i)
1591 {
1592 eio_dirent value = *i;
1593
1594 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1595 j [1] = j [0];
1596
1597 j [1] = value;
1598 }
1599 }
1600}
1601
1602static void
1603eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1604{
1605 if (size <= 1)
1606 return; /* our insertion sort relies on size > 0 */
1607
1608 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1609 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1610 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1611
1612 /* use an insertion sort at the end, or for small arrays, */
1613 /* as insertion sort is more efficient for small partitions */
1614 eio_dent_insertion_sort (dents, size);
1002} 1615}
1003 1616
1004/* read a full directory */ 1617/* read a full directory */
1005static void 1618static void
1006eio__scandir (eio_req *req, etp_worker *self) 1619eio__scandir (eio_req *req, etp_worker *self)
1007{ 1620{
1008 DIR *dirp; 1621 DIR *dirp;
1009 EIO_STRUCT_DIRENT *entp; 1622 EIO_STRUCT_DIRENT *entp;
1010 unsigned char *name, *names; 1623 char *name, *names;
1011 int namesalloc = 4096; 1624 int namesalloc = 4096;
1012 int namesoffs = 0; 1625 int namesoffs = 0;
1013 int flags = req->int1; 1626 int flags = req->int1;
1014 eio_dirent *dents = 0; 1627 eio_dirent *dents = 0;
1015 int dentalloc = 128; 1628 int dentalloc = 128;
1016 int dentoffs = 0; 1629 int dentoffs = 0;
1630 ino_t inode_bits = 0;
1017 1631
1018 req->result = -1; 1632 req->result = -1;
1019 1633
1020 if (!(flags & EIO_READDIR_DENTS)) 1634 if (!(flags & EIO_READDIR_DENTS))
1021 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1635 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1022 1636
1023 X_LOCK (wrklock); 1637 X_LOCK (wrklock);
1024 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1638 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1025 self->dirp = dirp = opendir (req->ptr1); 1639 self->dirp = dirp = opendir (req->ptr1);
1640
1641 if (req->flags & EIO_FLAG_PTR1_FREE)
1642 free (req->ptr1);
1643
1026 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1644 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1027 req->ptr1 = names = malloc (namesalloc);
1028 req->ptr2 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1645 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1646 req->ptr2 = names = malloc (namesalloc);
1029 X_UNLOCK (wrklock); 1647 X_UNLOCK (wrklock);
1030 1648
1031 if (dirp && names && (!flags || dents)) 1649 if (dirp && names && (!flags || dents))
1032 for (;;) 1650 for (;;)
1033 { 1651 {
1041 1659
1042 /* sort etc. */ 1660 /* sort etc. */
1043 req->int1 = flags; 1661 req->int1 = flags;
1044 req->result = dentoffs; 1662 req->result = dentoffs;
1045 1663
1046 if (dents)
1047 {
1048 eio_dirent *ent = dents + dentoffs;
1049
1050 while (ent > dents)
1051 (--ent)->name = names + (size_t)ent->name;
1052 }
1053
1054 if (flags & EIO_READDIR_STAT_ORDER 1664 if (flags & EIO_READDIR_STAT_ORDER)
1055 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1665 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1056 {
1057 /* pray your qsort doesn't use quicksort */
1058 qsort (dents, dentoffs, sizeof (*dents), eio_dent_cmp); /* score depends of DIRS_FIRST */
1059 }
1060 else if (flags & EIO_READDIR_DIRS_FIRST) 1666 else if (flags & EIO_READDIR_DIRS_FIRST)
1667 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1668 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1669 else
1061 { 1670 {
1062 /* in this case, all is known, and we just put dirs first and sort them */ 1671 /* in this case, all is known, and we just put dirs first and sort them */
1063 eio_dirent *ent = dents + dentoffs; 1672 eio_dirent *oth = dents + dentoffs;
1064 eio_dirent *dir = dents; 1673 eio_dirent *dir = dents;
1065 1674
1675 /* now partition dirs to the front, and non-dirs to the back */
1676 /* by walking from both sides and swapping if necessary */
1066 while (ent > dir) 1677 while (oth > dir)
1067 { 1678 {
1068 if (dir->type == DT_DIR) 1679 if (dir->type == EIO_DT_DIR)
1069 ++dir; 1680 ++dir;
1070 else
1071 {
1072 --ent;
1073
1074 if (ent->type == DT_DIR) 1681 else if ((--oth)->type == EIO_DT_DIR)
1075 { 1682 {
1076 eio_dirent tmp = *dir; 1683 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1077 *dir = *ent;
1078 *ent = tmp;
1079 1684
1080 ++dir; 1685 ++dir;
1081 } 1686 }
1082 } 1687 }
1688
1689 /* now sort the dirs only (dirs all have the same score) */
1690 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1083 } 1691 }
1084
1085 /* now sort the dirs only */
1086 qsort (dents, dir - dents, sizeof (*dents), eio_dent_cmp);
1087 }
1088
1089 /* only provide the names array unless DENTS is specified */
1090 if (!(flags & EIO_READDIR_DENTS))
1091 {
1092 X_LOCK (wrklock);
1093 assert (!dents);
1094 req->ptr1 = 0;
1095 req->ptr2 = names;
1096 X_UNLOCK (wrklock);
1097 }
1098 1692
1099 break; 1693 break;
1100 } 1694 }
1101 1695
1102 /* now add the entry to our list(s) */ 1696 /* now add the entry to our list(s) */
1105 /* skip . and .. entries */ 1699 /* skip . and .. entries */
1106 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1700 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1107 { 1701 {
1108 int len = D_NAMLEN (entp) + 1; 1702 int len = D_NAMLEN (entp) + 1;
1109 1703
1110 while (expect_false (namesoffs + len > namesalloc)) 1704 while (ecb_expect_false (namesoffs + len > namesalloc))
1111 { 1705 {
1112 namesalloc *= 2; 1706 namesalloc *= 2;
1113 X_LOCK (wrklock); 1707 X_LOCK (wrklock);
1114 req->ptr1 = names = realloc (names, namesalloc); 1708 req->ptr2 = names = realloc (names, namesalloc);
1115 X_UNLOCK (wrklock); 1709 X_UNLOCK (wrklock);
1116 1710
1117 if (!names) 1711 if (!names)
1118 break; 1712 break;
1119 } 1713 }
1122 1716
1123 if (dents) 1717 if (dents)
1124 { 1718 {
1125 struct eio_dirent *ent; 1719 struct eio_dirent *ent;
1126 1720
1127 if (expect_false (dentoffs == dentalloc)) 1721 if (ecb_expect_false (dentoffs == dentalloc))
1128 { 1722 {
1129 dentalloc *= 2; 1723 dentalloc *= 2;
1130 X_LOCK (wrklock); 1724 X_LOCK (wrklock);
1131 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1725 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1132 X_UNLOCK (wrklock); 1726 X_UNLOCK (wrklock);
1133 1727
1134 if (!dents) 1728 if (!dents)
1135 break; 1729 break;
1136 } 1730 }
1137 1731
1138 ent = dents + dentoffs; 1732 ent = dents + dentoffs;
1139 1733
1140 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1734 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1141 ent->namelen = len - 1; 1735 ent->namelen = len - 1;
1142 ent->inode = D_INO (entp); 1736 ent->inode = D_INO (entp);
1737
1738 inode_bits |= ent->inode;
1143 1739
1144 switch (D_TYPE (entp)) 1740 switch (D_TYPE (entp))
1145 { 1741 {
1146 default: 1742 default:
1147 ent->type = EIO_DT_UNKNOWN; 1743 ent->type = EIO_DT_UNKNOWN;
1190 #ifdef DT_WHT 1786 #ifdef DT_WHT
1191 case DT_WHT: ent->type = EIO_DT_WHT; break; 1787 case DT_WHT: ent->type = EIO_DT_WHT; break;
1192 #endif 1788 #endif
1193 } 1789 }
1194 1790
1195 ent->score = 0; 1791 ent->score = 7;
1196 1792
1197 if (flags & EIO_READDIR_DIRS_FIRST) 1793 if (flags & EIO_READDIR_DIRS_FIRST)
1198 { 1794 {
1199 if (ent->type == EIO_DT_UNKNOWN) 1795 if (ent->type == EIO_DT_UNKNOWN)
1200 { 1796 {
1201 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1797 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1202 ent->score = 98; 1798 ent->score = 1;
1203 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1799 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1204 ent->score = len <= 2 ? len + 60 : len <= 4 ? 50 : len <= 7 ? 40 : 10; /* shorter == more likely dir, but avoid too many classes */ 1800 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1205 } 1801 }
1206 else if (ent->type == EIO_DT_DIR) 1802 else if (ent->type == EIO_DT_DIR)
1207 ent->score = 100; 1803 ent->score = 0;
1208 } 1804 }
1209 } 1805 }
1210 1806
1211 namesoffs += len; 1807 namesoffs += len;
1212 ++dentoffs; 1808 ++dentoffs;
1213 } 1809 }
1810
1811 if (EIO_CANCELLED (req))
1812 {
1813 errno = ECANCELED;
1814 break;
1815 }
1214 } 1816 }
1215 else
1216 req->result = -1;
1217}
1218
1219#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1220# undef msync
1221# define msync(a,b,c) ((errno = ENOSYS), -1)
1222#endif
1223
1224int
1225eio__mtouch (void *mem, size_t len, int flags)
1226{
1227 intptr_t addr = (intptr_t)mem;
1228 intptr_t end = addr + len;
1229#ifdef PAGESIZE
1230 const intptr_t page = PAGESIZE;
1231#else
1232 static intptr_t page;
1233
1234 if (!page)
1235 page = sysconf (_SC_PAGESIZE);
1236#endif
1237
1238 addr &= ~(page - 1); /* assume page size is always a power of two */
1239
1240 if (addr < end)
1241 if (flags) /* modify */
1242 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1243 else
1244 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1245
1246 return 0;
1247} 1817}
1248 1818
1249/*****************************************************************************/ 1819/*****************************************************************************/
1250 1820
1251#define ALLOC(len) \ 1821#define ALLOC(len) \
1266X_THREAD_PROC (etp_proc) 1836X_THREAD_PROC (etp_proc)
1267{ 1837{
1268 ETP_REQ *req; 1838 ETP_REQ *req;
1269 struct timespec ts; 1839 struct timespec ts;
1270 etp_worker *self = (etp_worker *)thr_arg; 1840 etp_worker *self = (etp_worker *)thr_arg;
1841 int timeout;
1271 1842
1272 /* try to distribute timeouts somewhat randomly */ 1843 /* try to distribute timeouts somewhat evenly */
1273 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1844 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1274 1845
1275 for (;;) 1846 for (;;)
1276 { 1847 {
1848 ts.tv_sec = 0;
1849
1277 X_LOCK (reqlock); 1850 X_LOCK (reqlock);
1278 1851
1279 for (;;) 1852 for (;;)
1280 { 1853 {
1281 self->req = req = reqq_shift (&req_queue); 1854 self->req = req = reqq_shift (&req_queue);
1282 1855
1283 if (req) 1856 if (req)
1284 break; 1857 break;
1285 1858
1859 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1860 {
1861 X_UNLOCK (reqlock);
1862 X_LOCK (wrklock);
1863 --started;
1864 X_UNLOCK (wrklock);
1865 goto quit;
1866 }
1867
1286 ++idle; 1868 ++idle;
1287 1869
1288 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1870 if (idle <= max_idle)
1289 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1871 /* we are allowed to idle, so do so without any timeout */
1872 X_COND_WAIT (reqwait, reqlock);
1873 else
1290 { 1874 {
1291 if (idle > max_idle) 1875 /* initialise timeout once */
1292 { 1876 if (!ts.tv_sec)
1293 --idle; 1877 ts.tv_sec = time (0) + idle_timeout;
1294 X_UNLOCK (reqlock);
1295 X_LOCK (wrklock);
1296 --started;
1297 X_UNLOCK (wrklock);
1298 goto quit;
1299 }
1300 1878
1301 /* we are allowed to idle, so do so without any timeout */
1302 X_COND_WAIT (reqwait, reqlock); 1879 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1880 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1303 } 1881 }
1304 1882
1305 --idle; 1883 --idle;
1306 } 1884 }
1307 1885
1310 X_UNLOCK (reqlock); 1888 X_UNLOCK (reqlock);
1311 1889
1312 if (req->type < 0) 1890 if (req->type < 0)
1313 goto quit; 1891 goto quit;
1314 1892
1315 if (!EIO_CANCELLED (req))
1316 ETP_EXECUTE (self, req); 1893 ETP_EXECUTE (self, req);
1317 1894
1318 X_LOCK (reslock); 1895 X_LOCK (reslock);
1319 1896
1320 ++npending; 1897 ++npending;
1321 1898
1336 return 0; 1913 return 0;
1337} 1914}
1338 1915
1339/*****************************************************************************/ 1916/*****************************************************************************/
1340 1917
1918int ecb_cold
1341int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1919eio_init (void (*want_poll)(void), void (*done_poll)(void))
1342{ 1920{
1343 return etp_init (want_poll, done_poll); 1921 return etp_init (want_poll, done_poll);
1344} 1922}
1345 1923
1924ecb_inline void
1346static void eio_api_destroy (eio_req *req) 1925eio_api_destroy (eio_req *req)
1347{ 1926{
1348 free (req); 1927 free (req);
1349} 1928}
1350 1929
1351#define REQ(rtype) \ 1930#define REQ(rtype) \
1370 { \ 1949 { \
1371 eio_api_destroy (req); \ 1950 eio_api_destroy (req); \
1372 return 0; \ 1951 return 0; \
1373 } 1952 }
1374 1953
1954static void
1375static void eio_execute (etp_worker *self, eio_req *req) 1955eio_execute (etp_worker *self, eio_req *req)
1376{ 1956{
1377 errno = 0; 1957 if (ecb_expect_false (EIO_CANCELLED (req)))
1958 {
1959 req->result = -1;
1960 req->errorno = ECANCELED;
1961 return;
1962 }
1378 1963
1379 switch (req->type) 1964 switch (req->type)
1380 { 1965 {
1381 case EIO_READ: ALLOC (req->size); 1966 case EIO_READ: ALLOC (req->size);
1382 req->result = req->offs >= 0 1967 req->result = req->offs >= 0
1394 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1979 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1395 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1980 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1396 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1981 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1397 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1982 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1398 1983
1984 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1985 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1986 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1987 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1988
1399 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1989 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1400 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 1990 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1401 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 1991 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1402 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 1992 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1403 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 1993 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1410 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2000 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1411 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2001 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1412 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2002 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1413 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2003 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1414 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2004 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1415 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2005 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1416 2006
2007 case EIO_REALPATH: eio__realpath (req, self); break;
2008
1417 case EIO_READLINK: ALLOC (NAME_MAX); 2009 case EIO_READLINK: ALLOC (PATH_MAX);
1418 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2010 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1419 2011
1420 case EIO_SYNC: req->result = 0; sync (); break; 2012 case EIO_SYNC: req->result = 0; sync (); break;
1421 case EIO_FSYNC: req->result = fsync (req->int1); break; 2013 case EIO_FSYNC: req->result = fsync (req->int1); break;
1422 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2014 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1423 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 2015 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2016 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1424 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2017 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2018 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1425 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2019 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2020 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1426 2021
1427 case EIO_READDIR: eio__scandir (req, self); break; 2022 case EIO_READDIR: eio__scandir (req, self); break;
1428 2023
1429 case EIO_BUSY: 2024 case EIO_BUSY:
1430#ifdef _WIN32 2025#ifdef _WIN32
1431 Sleep (req->nv1 * 1000.); 2026 Sleep (req->nv1 * 1e3);
1432#else 2027#else
1433 { 2028 {
1434 struct timeval tv; 2029 struct timeval tv;
1435 2030
1436 tv.tv_sec = req->nv1; 2031 tv.tv_sec = req->nv1;
1437 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2032 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1438 2033
1439 req->result = select (0, 0, 0, 0, &tv); 2034 req->result = select (0, 0, 0, 0, &tv);
1440 } 2035 }
1441#endif 2036#endif
1442 break; 2037 break;
1457 times = tv; 2052 times = tv;
1458 } 2053 }
1459 else 2054 else
1460 times = 0; 2055 times = 0;
1461 2056
1462
1463 req->result = req->type == EIO_FUTIME 2057 req->result = req->type == EIO_FUTIME
1464 ? futimes (req->int1, times) 2058 ? futimes (req->int1, times)
1465 : utimes (req->ptr1, times); 2059 : utimes (req->ptr1, times);
1466 } 2060 }
1467 break; 2061 break;
1472 case EIO_NOP: 2066 case EIO_NOP:
1473 req->result = 0; 2067 req->result = 0;
1474 break; 2068 break;
1475 2069
1476 case EIO_CUSTOM: 2070 case EIO_CUSTOM:
1477 ((void (*)(eio_req *))req->feed) (req); 2071 req->feed (req);
1478 break; 2072 break;
1479 2073
1480 default: 2074 default:
2075 errno = ENOSYS;
1481 req->result = -1; 2076 req->result = -1;
1482 break; 2077 break;
1483 } 2078 }
1484 2079
1485 req->errorno = errno; 2080 req->errorno = errno;
1515eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2110eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1516{ 2111{
1517 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2112 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1518} 2113}
1519 2114
2115eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2116{
2117 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2118}
2119
2120eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2121{
2122 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2123}
2124
1520eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2125eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1521{ 2126{
1522 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2127 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1523} 2128}
1524 2129
2130eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2131{
2132 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2133}
2134
1525eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2135eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1526{ 2136{
1527 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2137 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1528} 2138}
1529 2139
1550eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2160eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1551{ 2161{
1552 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2162 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1553} 2163}
1554 2164
2165eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2166{
2167 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2168}
2169
1555eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2170eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1556{ 2171{
1557 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2172 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1558} 2173}
1559 2174
1565eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2180eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1566{ 2181{
1567 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2182 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1568} 2183}
1569 2184
1570eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2185eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1571{ 2186{
1572 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2187 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1573} 2188}
1574 2189
1575eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2190eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1595eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2210eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1596{ 2211{
1597 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2212 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1598} 2213}
1599 2214
1600eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2215eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1601{ 2216{
1602 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2217 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1603} 2218}
1604 2219
1605eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2220eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1621eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2236eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1622{ 2237{
1623 return eio__1path (EIO_READLINK, path, pri, cb, data); 2238 return eio__1path (EIO_READLINK, path, pri, cb, data);
1624} 2239}
1625 2240
2241eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2242{
2243 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2244}
2245
1626eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2246eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1627{ 2247{
1628 return eio__1path (EIO_STAT, path, pri, cb, data); 2248 return eio__1path (EIO_STAT, path, pri, cb, data);
1629} 2249}
1630 2250
1631eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2251eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1632{ 2252{
1633 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2253 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1634} 2254}
1635 2255
2256eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2257{
2258 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2259}
2260
1636eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2261eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1637{ 2262{
1638 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2263 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1639} 2264}
1640 2265
1648 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2273 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1649} 2274}
1650 2275
1651eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2276eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1652{ 2277{
1653 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2278 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1654} 2279}
1655 2280
1656static eio_req * 2281static eio_req *
1657eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2282eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1658{ 2283{
1682eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2307eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1683{ 2308{
1684 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2309 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1685} 2310}
1686 2311
1687eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2312eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1688{ 2313{
1689 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2314 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1690} 2315}
1691 2316
1692#endif 2317#endif
1693 2318
1694eio_req *eio_grp (eio_cb cb, void *data) 2319eio_req *eio_grp (eio_cb cb, void *data)
1703#undef SEND 2328#undef SEND
1704 2329
1705/*****************************************************************************/ 2330/*****************************************************************************/
1706/* grp functions */ 2331/* grp functions */
1707 2332
2333void
1708void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2334eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1709{ 2335{
1710 grp->int2 = limit; 2336 grp->int2 = limit;
1711 grp->feed = feed; 2337 grp->feed = feed;
1712 2338
1713 grp_try_feed (grp); 2339 grp_try_feed (grp);
1714} 2340}
1715 2341
2342void
1716void eio_grp_limit (eio_req *grp, int limit) 2343eio_grp_limit (eio_req *grp, int limit)
1717{ 2344{
1718 grp->int2 = limit; 2345 grp->int2 = limit;
1719 2346
1720 grp_try_feed (grp); 2347 grp_try_feed (grp);
1721} 2348}
1722 2349
2350void
1723void eio_grp_add (eio_req *grp, eio_req *req) 2351eio_grp_add (eio_req *grp, eio_req *req)
1724{ 2352{
1725 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2353 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1726 2354
1727 grp->flags |= EIO_FLAG_GROUPADD; 2355 grp->flags |= EIO_FLAG_GROUPADD;
1728 2356
1739} 2367}
1740 2368
1741/*****************************************************************************/ 2369/*****************************************************************************/
1742/* misc garbage */ 2370/* misc garbage */
1743 2371
2372eio_ssize_t
1744ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2373eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1745{ 2374{
1746 etp_worker wrk; 2375 etp_worker wrk;
2376 eio_ssize_t ret;
1747 2377
1748 wrk.dbuf = 0; 2378 wrk.dbuf = 0;
1749 2379
1750 eio__sendfile (ofd, ifd, offset, count, &wrk); 2380 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1751 2381
1752 if (wrk.dbuf) 2382 if (wrk.dbuf)
1753 free (wrk.dbuf); 2383 free (wrk.dbuf);
1754}
1755 2384
2385 return ret;
2386}
2387

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines