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

Comparing libeio/eio.c (file contents):
Revision 1.33 by root, Sat Jun 6 19:44:17 2009 UTC vs.
Revision 1.89 by root, Thu Jul 14 22:36:17 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/* 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
73static void eio_destroy (eio_req *req);
74
58#ifndef EIO_FINISH 75#ifndef EIO_FINISH
59# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 76# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
60#endif 77#endif
61 78
62#ifndef EIO_DESTROY 79#ifndef EIO_DESTROY
65 82
66#ifndef EIO_FEED 83#ifndef EIO_FEED
67# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 84# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
68#endif 85#endif
69 86
87#ifndef EIO_FD_TO_WIN32_HANDLE
88# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
89#endif
90#ifndef EIO_WIN32_HANDLE_TO_FD
91# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
92#endif
93
94#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
95
96#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
97
70#ifdef _WIN32 98#ifdef _WIN32
71 99
72 /*doh*/ 100 #define PAGESIZE 4096 /* GetSystemInfo? */
101
102 #ifdef EIO_STRUCT_STATI64
103 #define stat(path,buf) _stati64 (path,buf)
104 #define fstat(fd,buf) _fstati64 (path,buf)
105 #endif
106 #define lstat(path,buf) stat (path,buf)
107 #define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
108 #define mkdir(path,mode) _mkdir (path)
109 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
110
111 #define chown(path,uid,gid) EIO_ENOSYS ()
112 #define fchown(fd,uid,gid) EIO_ENOSYS ()
113 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
114 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
115 #define mknod(path,mode,dev) EIO_ENOSYS ()
116 #define sync() EIO_ENOSYS ()
117
118 /* we could even stat and see if it exists */
119 static int
120 symlink (const char *old, const char *neu)
121 {
122 if (CreateSymbolicLink (neu, old, 1))
123 return 0;
124
125 if (CreateSymbolicLink (neu, old, 0))
126 return 0;
127
128 return EIO_ERRNO (ENOENT, -1);
129 }
130
73#else 131#else
74 132
75# include "config.h"
76# include <sys/time.h> 133 #include <sys/time.h>
77# include <sys/select.h> 134 #include <sys/select.h>
135 #include <sys/statvfs.h>
78# include <unistd.h> 136 #include <unistd.h>
79# include <utime.h> 137 #include <utime.h>
80# include <signal.h> 138 #include <signal.h>
81# include <dirent.h> 139 #include <dirent.h>
82 140
141 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
142 #include <sys/mman.h>
143 #endif
144
83/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 145 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
84# if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) 146 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
85# define _DIRENT_HAVE_D_TYPE /* sigh */ 147 #define _DIRENT_HAVE_D_TYPE /* sigh */
86# define D_INO(de) (de)->d_fileno 148 #define D_INO(de) (de)->d_fileno
87# define D_NAMLEN(de) (de)->d_namlen 149 #define D_NAMLEN(de) (de)->d_namlen
88# elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 150 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
89# define D_INO(de) (de)->d_ino 151 #define D_INO(de) (de)->d_ino
90# endif 152 #endif
91 153
92#ifdef _D_EXACT_NAMLEN 154 #ifdef _D_EXACT_NAMLEN
93# undef D_NAMLEN 155 #undef D_NAMLEN
94# define D_NAMLEN(de) _D_EXACT_NAMLEN (de) 156 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
95#endif 157 #endif
96 158
97# ifdef _DIRENT_HAVE_D_TYPE 159 #ifdef _DIRENT_HAVE_D_TYPE
98# define D_TYPE(de) (de)->d_type 160 #define D_TYPE(de) (de)->d_type
99# endif 161 #endif
100 162
101# ifndef EIO_STRUCT_DIRENT 163 #ifndef EIO_STRUCT_DIRENT
102# define EIO_STRUCT_DIRENT struct dirent 164 #define EIO_STRUCT_DIRENT struct dirent
103# endif 165 #endif
104 166
105#endif 167#endif
106 168
107#if HAVE_SENDFILE 169#if HAVE_SENDFILE
108# if __linux 170# if __linux
109# include <sys/sendfile.h> 171# include <sys/sendfile.h>
110# elif __freebsd 172# elif __FreeBSD__ || defined __APPLE__
111# include <sys/socket.h> 173# include <sys/socket.h>
112# include <sys/uio.h> 174# include <sys/uio.h>
113# elif __hpux 175# elif __hpux
114# include <sys/socket.h> 176# include <sys/socket.h>
115# elif __solaris /* not yet */ 177# elif __solaris
116# include <sys/sendfile.h> 178# include <sys/sendfile.h>
117# else 179# else
118# error sendfile support requested but not available 180# error sendfile support requested but not available
119# endif 181# endif
120#endif 182#endif
127#endif 189#endif
128#ifndef D_NAMLEN 190#ifndef D_NAMLEN
129# define D_NAMLEN(de) strlen ((de)->d_name) 191# define D_NAMLEN(de) strlen ((de)->d_name)
130#endif 192#endif
131 193
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 */ 194/* used for struct dirent, AIX doesn't provide it */
136#ifndef NAME_MAX 195#ifndef NAME_MAX
137# define NAME_MAX 4096 196# define NAME_MAX 4096
197#endif
198
199/* used for readlink etc. */
200#ifndef PATH_MAX
201# define PATH_MAX 4096
138#endif 202#endif
139 203
140/* buffer size for various temporary buffers */ 204/* buffer size for various temporary buffers */
141#define EIO_BUFSIZE 65536 205#define EIO_BUFSIZE 65536
142 206
148 errno = ENOMEM; \ 212 errno = ENOMEM; \
149 if (!eio_buf) \ 213 if (!eio_buf) \
150 return -1; 214 return -1;
151 215
152#define EIO_TICKS ((1000000 + 1023) >> 10) 216#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/*****************************************************************************/
166 217
167#define ETP_PRI_MIN EIO_PRI_MIN 218#define ETP_PRI_MIN EIO_PRI_MIN
168#define ETP_PRI_MAX EIO_PRI_MAX 219#define ETP_PRI_MAX EIO_PRI_MAX
169 220
170struct etp_worker; 221struct etp_worker;
195 246
196/*****************************************************************************/ 247/*****************************************************************************/
197 248
198#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 249#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
199 250
200/* calculcate time difference in ~1/EIO_TICKS of a second */ 251/* calculate time difference in ~1/EIO_TICKS of a second */
252ecb_inline int
201static int tvdiff (struct timeval *tv1, struct timeval *tv2) 253tvdiff (struct timeval *tv1, struct timeval *tv2)
202{ 254{
203 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 255 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
204 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 256 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
205} 257}
206 258
213static unsigned int max_poll_reqs; /* reslock */ 265static unsigned int max_poll_reqs; /* reslock */
214 266
215static volatile unsigned int nreqs; /* reqlock */ 267static volatile unsigned int nreqs; /* reqlock */
216static volatile unsigned int nready; /* reqlock */ 268static volatile unsigned int nready; /* reqlock */
217static volatile unsigned int npending; /* reqlock */ 269static volatile unsigned int npending; /* reqlock */
218static volatile unsigned int max_idle = 4; 270static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
271static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
219 272
220static mutex_t wrklock = X_MUTEX_INIT; 273static xmutex_t wrklock;
221static mutex_t reslock = X_MUTEX_INIT; 274static xmutex_t reslock;
222static mutex_t reqlock = X_MUTEX_INIT; 275static xmutex_t reqlock;
223static cond_t reqwait = X_COND_INIT; 276static xcond_t reqwait;
224 277
225#if !HAVE_PREADWRITE 278#if !HAVE_PREADWRITE
226/* 279/*
227 * make our pread/pwrite emulation safe against themselves, but not against 280 * make our pread/pwrite emulation safe against themselves, but not against
228 * normal read/write by using a mutex. slows down execution a lot, 281 * normal read/write by using a mutex. slows down execution a lot,
229 * but that's your problem, not mine. 282 * but that's your problem, not mine.
230 */ 283 */
231static mutex_t preadwritelock = X_MUTEX_INIT; 284static xmutex_t preadwritelock = X_MUTEX_INIT;
232#endif 285#endif
233 286
234typedef struct etp_worker 287typedef struct etp_worker
235{ 288{
236 /* locked by wrklock */ 289 /* locked by wrklock */
237 struct etp_worker *prev, *next; 290 struct etp_worker *prev, *next;
238 291
239 thread_t tid; 292 xthread_t tid;
240 293
241 /* locked by reslock, reqlock or wrklock */ 294 /* locked by reslock, reqlock or wrklock */
242 ETP_REQ *req; /* currently processed request */ 295 ETP_REQ *req; /* currently processed request */
243 296
244 ETP_WORKER_COMMON 297 ETP_WORKER_COMMON
249#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 302#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
250#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 303#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
251 304
252/* worker threads management */ 305/* worker threads management */
253 306
307static void ecb_cold
254static void etp_worker_clear (etp_worker *wrk) 308etp_worker_clear (etp_worker *wrk)
255{ 309{
256 ETP_WORKER_CLEAR (wrk); 310 ETP_WORKER_CLEAR (wrk);
257} 311}
258 312
313static void ecb_cold
259static void etp_worker_free (etp_worker *wrk) 314etp_worker_free (etp_worker *wrk)
260{ 315{
261 wrk->next->prev = wrk->prev; 316 wrk->next->prev = wrk->prev;
262 wrk->prev->next = wrk->next; 317 wrk->prev->next = wrk->next;
263 318
264 free (wrk); 319 free (wrk);
265} 320}
266 321
267static unsigned int etp_nreqs (void) 322static unsigned int
323etp_nreqs (void)
268{ 324{
269 int retval; 325 int retval;
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 326 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = nreqs; 327 retval = nreqs;
272 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 328 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
273 return retval; 329 return retval;
274} 330}
275 331
276static unsigned int etp_nready (void) 332static unsigned int
333etp_nready (void)
277{ 334{
278 unsigned int retval; 335 unsigned int retval;
279 336
280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 337 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281 retval = nready; 338 retval = nready;
282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 339 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283 340
284 return retval; 341 return retval;
285} 342}
286 343
287static unsigned int etp_npending (void) 344static unsigned int
345etp_npending (void)
288{ 346{
289 unsigned int retval; 347 unsigned int retval;
290 348
291 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 349 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
292 retval = npending; 350 retval = npending;
293 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 351 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
294 352
295 return retval; 353 return retval;
296} 354}
297 355
298static unsigned int etp_nthreads (void) 356static unsigned int
357etp_nthreads (void)
299{ 358{
300 unsigned int retval; 359 unsigned int retval;
301 360
302 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 361 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303 retval = started; 362 retval = started;
317} etp_reqq; 376} etp_reqq;
318 377
319static etp_reqq req_queue; 378static etp_reqq req_queue;
320static etp_reqq res_queue; 379static etp_reqq res_queue;
321 380
381static int ecb_noinline
322static int reqq_push (etp_reqq *q, ETP_REQ *req) 382reqq_push (etp_reqq *q, ETP_REQ *req)
323{ 383{
324 int pri = req->pri; 384 int pri = req->pri;
325 req->next = 0; 385 req->next = 0;
326 386
327 if (q->qe[pri]) 387 if (q->qe[pri])
333 q->qe[pri] = q->qs[pri] = req; 393 q->qe[pri] = q->qs[pri] = req;
334 394
335 return q->size++; 395 return q->size++;
336} 396}
337 397
398static ETP_REQ * ecb_noinline
338static ETP_REQ *reqq_shift (etp_reqq *q) 399reqq_shift (etp_reqq *q)
339{ 400{
340 int pri; 401 int pri;
341 402
342 if (!q->size) 403 if (!q->size)
343 return 0; 404 return 0;
358 } 419 }
359 420
360 abort (); 421 abort ();
361} 422}
362 423
363static void etp_atfork_prepare (void) 424static void ecb_cold
425etp_thread_init (void)
364{ 426{
365 X_LOCK (wrklock);
366 X_LOCK (reqlock);
367 X_LOCK (reslock);
368#if !HAVE_PREADWRITE 427#if !HAVE_PREADWRITE
369 X_LOCK (preadwritelock); 428 X_MUTEX_CREATE (preadwritelock);
370#endif 429#endif
430 X_MUTEX_CREATE (wrklock);
431 X_MUTEX_CREATE (reslock);
432 X_MUTEX_CREATE (reqlock);
433 X_COND_CREATE (reqwait);
371} 434}
372 435
436static void ecb_cold
437etp_atfork_prepare (void)
438{
439}
440
441static void ecb_cold
373static void etp_atfork_parent (void) 442etp_atfork_parent (void)
374{ 443{
375#if !HAVE_PREADWRITE
376 X_UNLOCK (preadwritelock);
377#endif
378 X_UNLOCK (reslock);
379 X_UNLOCK (reqlock);
380 X_UNLOCK (wrklock);
381} 444}
382 445
446static void ecb_cold
383static void etp_atfork_child (void) 447etp_atfork_child (void)
384{ 448{
385 ETP_REQ *prv; 449 ETP_REQ *prv;
386 450
387 while ((prv = reqq_shift (&req_queue))) 451 while ((prv = reqq_shift (&req_queue)))
388 ETP_DESTROY (prv); 452 ETP_DESTROY (prv);
405 idle = 0; 469 idle = 0;
406 nreqs = 0; 470 nreqs = 0;
407 nready = 0; 471 nready = 0;
408 npending = 0; 472 npending = 0;
409 473
410 etp_atfork_parent (); 474 etp_thread_init ();
411} 475}
412 476
413static void 477static void ecb_cold
414etp_once_init (void) 478etp_once_init (void)
415{ 479{
480 etp_thread_init ();
416 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 481 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
417} 482}
418 483
419static int 484static int ecb_cold
420etp_init (void (*want_poll)(void), void (*done_poll)(void)) 485etp_init (void (*want_poll)(void), void (*done_poll)(void))
421{ 486{
422 static pthread_once_t doinit = PTHREAD_ONCE_INIT; 487 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
423 488
424 pthread_once (&doinit, etp_once_init); 489 pthread_once (&doinit, etp_once_init);
429 return 0; 494 return 0;
430} 495}
431 496
432X_THREAD_PROC (etp_proc); 497X_THREAD_PROC (etp_proc);
433 498
499static void ecb_cold
434static void etp_start_thread (void) 500etp_start_thread (void)
435{ 501{
436 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 502 etp_worker *wrk = calloc (1, sizeof (etp_worker));
437 503
438 /*TODO*/ 504 /*TODO*/
439 assert (("unable to allocate worker thread data", wrk)); 505 assert (("unable to allocate worker thread data", wrk));
452 free (wrk); 518 free (wrk);
453 519
454 X_UNLOCK (wrklock); 520 X_UNLOCK (wrklock);
455} 521}
456 522
523static void
457static void etp_maybe_start_thread (void) 524etp_maybe_start_thread (void)
458{ 525{
459 if (expect_true (etp_nthreads () >= wanted)) 526 if (ecb_expect_true (etp_nthreads () >= wanted))
460 return; 527 return;
461 528
462 /* todo: maybe use idle here, but might be less exact */ 529 /* todo: maybe use idle here, but might be less exact */
463 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 530 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
464 return; 531 return;
465 532
466 etp_start_thread (); 533 etp_start_thread ();
467} 534}
468 535
536static void ecb_cold
469static void etp_end_thread (void) 537etp_end_thread (void)
470{ 538{
471 eio_req *req = calloc (1, sizeof (eio_req)); 539 eio_req *req = calloc (1, sizeof (eio_req));
472 540
473 req->type = -1; 541 req->type = -1;
474 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 542 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
481 X_LOCK (wrklock); 549 X_LOCK (wrklock);
482 --started; 550 --started;
483 X_UNLOCK (wrklock); 551 X_UNLOCK (wrklock);
484} 552}
485 553
486static int etp_poll (void) 554static int
555etp_poll (void)
487{ 556{
488 unsigned int maxreqs; 557 unsigned int maxreqs;
489 unsigned int maxtime; 558 unsigned int maxtime;
490 struct timeval tv_start, tv_now; 559 struct timeval tv_start, tv_now;
491 560
521 590
522 X_LOCK (reqlock); 591 X_LOCK (reqlock);
523 --nreqs; 592 --nreqs;
524 X_UNLOCK (reqlock); 593 X_UNLOCK (reqlock);
525 594
526 if (expect_false (req->type == EIO_GROUP && req->size)) 595 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
527 { 596 {
528 req->int1 = 1; /* mark request as delayed */ 597 req->int1 = 1; /* mark request as delayed */
529 continue; 598 continue;
530 } 599 }
531 else 600 else
532 { 601 {
533 int res = ETP_FINISH (req); 602 int res = ETP_FINISH (req);
534 if (expect_false (res)) 603 if (ecb_expect_false (res))
535 return res; 604 return res;
536 } 605 }
537 606
538 if (expect_false (maxreqs && !--maxreqs)) 607 if (ecb_expect_false (maxreqs && !--maxreqs))
539 break; 608 break;
540 609
541 if (maxtime) 610 if (maxtime)
542 { 611 {
543 gettimeofday (&tv_now, 0); 612 gettimeofday (&tv_now, 0);
549 618
550 errno = EAGAIN; 619 errno = EAGAIN;
551 return -1; 620 return -1;
552} 621}
553 622
623static void
554static void etp_cancel (ETP_REQ *req) 624etp_cancel (ETP_REQ *req)
555{ 625{
556 X_LOCK (wrklock); 626 req->cancelled = 1;
557 req->flags |= EIO_FLAG_CANCELLED;
558 X_UNLOCK (wrklock);
559 627
560 eio_grp_cancel (req); 628 eio_grp_cancel (req);
561} 629}
562 630
631static void
563static void etp_submit (ETP_REQ *req) 632etp_submit (ETP_REQ *req)
564{ 633{
565 req->pri -= ETP_PRI_MIN; 634 req->pri -= ETP_PRI_MIN;
566 635
567 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 636 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; 637 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
569 638
570 if (expect_false (req->type == EIO_GROUP)) 639 if (ecb_expect_false (req->type == EIO_GROUP))
571 { 640 {
572 /* I hope this is worth it :/ */ 641 /* I hope this is worth it :/ */
573 X_LOCK (reqlock); 642 X_LOCK (reqlock);
574 ++nreqs; 643 ++nreqs;
575 X_UNLOCK (reqlock); 644 X_UNLOCK (reqlock);
594 663
595 etp_maybe_start_thread (); 664 etp_maybe_start_thread ();
596 } 665 }
597} 666}
598 667
668static void ecb_cold
599static void etp_set_max_poll_time (double nseconds) 669etp_set_max_poll_time (double nseconds)
600{ 670{
601 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 671 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
602 max_poll_time = nseconds; 672 max_poll_time = nseconds * EIO_TICKS;
603 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 673 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
604} 674}
605 675
676static void ecb_cold
606static void etp_set_max_poll_reqs (unsigned int maxreqs) 677etp_set_max_poll_reqs (unsigned int maxreqs)
607{ 678{
608 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 679 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
609 max_poll_reqs = maxreqs; 680 max_poll_reqs = maxreqs;
610 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 681 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
611} 682}
612 683
684static void ecb_cold
613static void etp_set_max_idle (unsigned int nthreads) 685etp_set_max_idle (unsigned int nthreads)
614{ 686{
615 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 687 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
616 max_idle = nthreads <= 0 ? 1 : nthreads; 688 max_idle = nthreads;
617 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 689 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
618} 690}
619 691
692static void ecb_cold
693etp_set_idle_timeout (unsigned int seconds)
694{
695 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
696 idle_timeout = seconds;
697 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
698}
699
700static void ecb_cold
620static void etp_set_min_parallel (unsigned int nthreads) 701etp_set_min_parallel (unsigned int nthreads)
621{ 702{
622 if (wanted < nthreads) 703 if (wanted < nthreads)
623 wanted = nthreads; 704 wanted = nthreads;
624} 705}
625 706
707static void ecb_cold
626static void etp_set_max_parallel (unsigned int nthreads) 708etp_set_max_parallel (unsigned int nthreads)
627{ 709{
628 if (wanted > nthreads) 710 if (wanted > nthreads)
629 wanted = nthreads; 711 wanted = nthreads;
630 712
631 while (started > wanted) 713 while (started > wanted)
632 etp_end_thread (); 714 etp_end_thread ();
633} 715}
634 716
635/*****************************************************************************/ 717/*****************************************************************************/
636 718
719static void
637static void grp_try_feed (eio_req *grp) 720grp_try_feed (eio_req *grp)
638{ 721{
639 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 722 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
640 { 723 {
641 grp->flags &= ~EIO_FLAG_GROUPADD; 724 grp->flags &= ~EIO_FLAG_GROUPADD;
642 725
649 break; 732 break;
650 } 733 }
651 } 734 }
652} 735}
653 736
737static int
654static int grp_dec (eio_req *grp) 738grp_dec (eio_req *grp)
655{ 739{
656 --grp->size; 740 --grp->size;
657 741
658 /* call feeder, if applicable */ 742 /* call feeder, if applicable */
659 grp_try_feed (grp); 743 grp_try_feed (grp);
663 return eio_finish (grp); 747 return eio_finish (grp);
664 else 748 else
665 return 0; 749 return 0;
666} 750}
667 751
752static void
668void eio_destroy (eio_req *req) 753eio_destroy (eio_req *req)
669{ 754{
670 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 755 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
671 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 756 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
672 757
673 EIO_DESTROY (req); 758 EIO_DESTROY (req);
674} 759}
675 760
761static int
676static int eio_finish (eio_req *req) 762eio_finish (eio_req *req)
677{ 763{
678 int res = EIO_FINISH (req); 764 int res = EIO_FINISH (req);
679 765
680 if (req->grp) 766 if (req->grp)
681 { 767 {
689 if (grp->grp_first == req) 775 if (grp->grp_first == req)
690 grp->grp_first = req->grp_next; 776 grp->grp_first = req->grp_next;
691 777
692 res2 = grp_dec (grp); 778 res2 = grp_dec (grp);
693 779
694 if (!res && res2) 780 if (!res)
695 res = res2; 781 res = res2;
696 } 782 }
697 783
698 eio_destroy (req); 784 eio_destroy (req);
699 785
700 return res; 786 return res;
701} 787}
702 788
789void
703void eio_grp_cancel (eio_req *grp) 790eio_grp_cancel (eio_req *grp)
704{ 791{
705 for (grp = grp->grp_first; grp; grp = grp->grp_next) 792 for (grp = grp->grp_first; grp; grp = grp->grp_next)
706 eio_cancel (grp); 793 eio_cancel (grp);
707} 794}
708 795
796void
709void eio_cancel (eio_req *req) 797eio_cancel (eio_req *req)
710{ 798{
711 etp_cancel (req); 799 etp_cancel (req);
712} 800}
713 801
802void
714void eio_submit (eio_req *req) 803eio_submit (eio_req *req)
715{ 804{
716 etp_submit (req); 805 etp_submit (req);
717} 806}
718 807
719unsigned int eio_nreqs (void) 808unsigned int
809eio_nreqs (void)
720{ 810{
721 return etp_nreqs (); 811 return etp_nreqs ();
722} 812}
723 813
724unsigned int eio_nready (void) 814unsigned int
815eio_nready (void)
725{ 816{
726 return etp_nready (); 817 return etp_nready ();
727} 818}
728 819
729unsigned int eio_npending (void) 820unsigned int
821eio_npending (void)
730{ 822{
731 return etp_npending (); 823 return etp_npending ();
732} 824}
733 825
734unsigned int eio_nthreads (void) 826unsigned int ecb_cold
827eio_nthreads (void)
735{ 828{
736 return etp_nthreads (); 829 return etp_nthreads ();
737} 830}
738 831
832void ecb_cold
739void eio_set_max_poll_time (double nseconds) 833eio_set_max_poll_time (double nseconds)
740{ 834{
741 etp_set_max_poll_time (nseconds); 835 etp_set_max_poll_time (nseconds);
742} 836}
743 837
838void ecb_cold
744void eio_set_max_poll_reqs (unsigned int maxreqs) 839eio_set_max_poll_reqs (unsigned int maxreqs)
745{ 840{
746 etp_set_max_poll_reqs (maxreqs); 841 etp_set_max_poll_reqs (maxreqs);
747} 842}
748 843
844void ecb_cold
749void eio_set_max_idle (unsigned int nthreads) 845eio_set_max_idle (unsigned int nthreads)
750{ 846{
751 etp_set_max_idle (nthreads); 847 etp_set_max_idle (nthreads);
752} 848}
753 849
850void ecb_cold
851eio_set_idle_timeout (unsigned int seconds)
852{
853 etp_set_idle_timeout (seconds);
854}
855
856void ecb_cold
754void eio_set_min_parallel (unsigned int nthreads) 857eio_set_min_parallel (unsigned int nthreads)
755{ 858{
756 etp_set_min_parallel (nthreads); 859 etp_set_min_parallel (nthreads);
757} 860}
758 861
862void ecb_cold
759void eio_set_max_parallel (unsigned int nthreads) 863eio_set_max_parallel (unsigned int nthreads)
760{ 864{
761 etp_set_max_parallel (nthreads); 865 etp_set_max_parallel (nthreads);
762} 866}
763 867
764int eio_poll (void) 868int eio_poll (void)
773# undef pread 877# undef pread
774# undef pwrite 878# undef pwrite
775# define pread eio__pread 879# define pread eio__pread
776# define pwrite eio__pwrite 880# define pwrite eio__pwrite
777 881
778static ssize_t 882static eio_ssize_t
779eio__pread (int fd, void *buf, size_t count, off_t offset) 883eio__pread (int fd, void *buf, size_t count, off_t offset)
780{ 884{
781 ssize_t res; 885 eio_ssize_t res;
782 off_t ooffset; 886 off_t ooffset;
783 887
784 X_LOCK (preadwritelock); 888 X_LOCK (preadwritelock);
785 ooffset = lseek (fd, 0, SEEK_CUR); 889 ooffset = lseek (fd, 0, SEEK_CUR);
786 lseek (fd, offset, SEEK_SET); 890 lseek (fd, offset, SEEK_SET);
789 X_UNLOCK (preadwritelock); 893 X_UNLOCK (preadwritelock);
790 894
791 return res; 895 return res;
792} 896}
793 897
794static ssize_t 898static eio_ssize_t
795eio__pwrite (int fd, void *buf, size_t count, off_t offset) 899eio__pwrite (int fd, void *buf, size_t count, off_t offset)
796{ 900{
797 ssize_t res; 901 eio_ssize_t res;
798 off_t ooffset; 902 off_t ooffset;
799 903
800 X_LOCK (preadwritelock); 904 X_LOCK (preadwritelock);
801 ooffset = lseek (fd, 0, SEEK_CUR); 905 ooffset = lseek (fd, 0, SEEK_CUR);
802 lseek (fd, offset, SEEK_SET); 906 lseek (fd, offset, SEEK_SET);
806 910
807 return res; 911 return res;
808} 912}
809#endif 913#endif
810 914
811#ifndef HAVE_FUTIMES 915#ifndef HAVE_UTIMES
812 916
813# undef utimes 917# undef utimes
814# undef futimes
815# define utimes(path,times) eio__utimes (path, times) 918# define utimes(path,times) eio__utimes (path, times)
816# define futimes(fd,times) eio__futimes (fd, times)
817 919
818static int 920static int
819eio__utimes (const char *filename, const struct timeval times[2]) 921eio__utimes (const char *filename, const struct timeval times[2])
820{ 922{
821 if (times) 923 if (times)
829 } 931 }
830 else 932 else
831 return utime (filename, 0); 933 return utime (filename, 0);
832} 934}
833 935
936#endif
937
938#ifndef HAVE_FUTIMES
939
940# undef futimes
941# define futimes(fd,times) eio__futimes (fd, times)
942
943static int
834static int eio__futimes (int fd, const struct timeval tv[2]) 944eio__futimes (int fd, const struct timeval tv[2])
835{ 945{
836 errno = ENOSYS; 946 errno = ENOSYS;
837 return -1; 947 return -1;
838} 948}
839 949
843# undef fdatasync 953# undef fdatasync
844# define fdatasync(fd) fsync (fd) 954# define fdatasync(fd) fsync (fd)
845#endif 955#endif
846 956
847/* sync_file_range always needs emulation */ 957/* sync_file_range always needs emulation */
848int 958static int
849eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 959eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
850{ 960{
851#if HAVE_SYNC_FILE_RANGE 961#if HAVE_SYNC_FILE_RANGE
852 int res; 962 int res;
853 963
866 if (!res || errno != ENOSYS) 976 if (!res || errno != ENOSYS)
867 return res; 977 return res;
868#endif 978#endif
869 979
870 /* even though we could play tricks with the flags, it's better to always 980 /* 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 */ 981 * call fdatasync, as that matches the expectation of its users best */
872 return fdatasync (fd); 982 return fdatasync (fd);
983}
984
985static int
986eio__fallocate (int fd, int mode, off_t offset, size_t len)
987{
988#if HAVE_FALLOCATE
989 return fallocate (fd, mode, offset, len);
990#else
991 errno = ENOSYS;
992 return -1;
993#endif
873} 994}
874 995
875#if !HAVE_READAHEAD 996#if !HAVE_READAHEAD
876# undef readahead 997# undef readahead
877# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 998# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
878 999
879static ssize_t 1000static eio_ssize_t
880eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1001eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
881{ 1002{
882 size_t todo = count; 1003 size_t todo = count;
883 dBUF; 1004 dBUF;
884 1005
896} 1017}
897 1018
898#endif 1019#endif
899 1020
900/* sendfile always needs emulation */ 1021/* sendfile always needs emulation */
901static ssize_t 1022static eio_ssize_t
902eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1023eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
903{ 1024{
1025 eio_ssize_t written = 0;
904 ssize_t res; 1026 eio_ssize_t res;
905 1027
906 if (!count) 1028 if (!count)
907 return 0; 1029 return 0;
908 1030
1031 for (;;)
1032 {
1033#ifdef __APPLE__
1034# undef HAVE_SENDFILE /* broken, as everything on os x */
1035#endif
909#if HAVE_SENDFILE 1036#if HAVE_SENDFILE
910# if __linux 1037# if __linux
1038 off_t soffset = offset;
911 res = sendfile (ofd, ifd, &offset, count); 1039 res = sendfile (ofd, ifd, &soffset, count);
912 1040
913# elif __freebsd 1041# elif __FreeBSD__
914 /* 1042 /*
915 * Of course, the freebsd sendfile is a dire hack with no thoughts 1043 * Of course, the freebsd sendfile is a dire hack with no thoughts
916 * wasted on making it similar to other I/O functions. 1044 * wasted on making it similar to other I/O functions.
917 */ 1045 */
918 {
919 off_t sbytes; 1046 off_t sbytes;
920 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1047 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
921 1048
922 if (res < 0 && sbytes) 1049 #if 0 /* according to the manpage, this is correct, but broken behaviour */
923 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1050 /* freebsd' sendfile will return 0 on success */
1051 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1052 /* not on e.g. EIO or EPIPE - sounds broken */
1053 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
924 res = sbytes; 1054 res = sbytes;
925 } 1055 #endif
1056
1057 /* according to source inspection, this is correct, and useful behaviour */
1058 if (sbytes)
1059 res = sbytes;
1060
1061# elif defined (__APPLE__)
1062 off_t sbytes = count;
1063 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1064
1065 /* according to the manpage, sbytes is always valid */
1066 if (sbytes)
1067 res = sbytes;
926 1068
927# elif __hpux 1069# elif __hpux
928 res = sendfile (ofd, ifd, offset, count, 0, 0); 1070 res = sendfile (ofd, ifd, offset, count, 0, 0);
929 1071
930# elif __solaris 1072# elif __solaris
931 {
932 struct sendfilevec vec; 1073 struct sendfilevec vec;
933 size_t sbytes; 1074 size_t sbytes;
934 1075
935 vec.sfv_fd = ifd; 1076 vec.sfv_fd = ifd;
936 vec.sfv_flag = 0; 1077 vec.sfv_flag = 0;
937 vec.sfv_off = offset; 1078 vec.sfv_off = offset;
938 vec.sfv_len = count; 1079 vec.sfv_len = count;
939 1080
940 res = sendfilev (ofd, &vec, 1, &sbytes); 1081 res = sendfilev (ofd, &vec, 1, &sbytes);
941 1082
942 if (res < 0 && sbytes) 1083 if (res < 0 && sbytes)
943 res = sbytes; 1084 res = sbytes;
944 }
945 1085
946# endif 1086# endif
1087
1088#elif defined (_WIN32) && 0
1089 /* does not work, just for documentation of what would need to be done */
1090 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1091 /* libeio guarantees that the file offset does not change, and windows */
1092 /* has no way to get an independent handle to the same file description */
1093 HANDLE h = TO_SOCKET (ifd);
1094 SetFilePointer (h, offset, 0, FILE_BEGIN);
1095 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1096
947#else 1097#else
948 res = -1; 1098 res = -1;
949 errno = ENOSYS; 1099 errno = ENOSYS;
950#endif 1100#endif
951 1101
1102 /* we assume sendfile can copy at least 128mb in one go */
1103 if (res <= 128 * 1024 * 1024)
1104 {
1105 if (res > 0)
1106 written += res;
1107
1108 if (written)
1109 return written;
1110
1111 break;
1112 }
1113 else
1114 {
1115 /* if we requested more, then probably the kernel was lazy */
1116 written += res;
1117 offset += res;
1118 count -= res;
1119
1120 if (!count)
1121 return written;
1122 }
1123 }
1124
952 if (res < 0 1125 if (res < 0
953 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1126 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1127 /* BSDs */
1128#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1129 || errno == ENOTSUP
1130#endif
1131 || errno == EOPNOTSUPP /* BSDs */
954#if __solaris 1132#if __solaris
955 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1133 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
956#endif 1134#endif
957 ) 1135 )
958 ) 1136 )
962 1140
963 res = 0; 1141 res = 0;
964 1142
965 while (count) 1143 while (count)
966 { 1144 {
967 ssize_t cnt; 1145 eio_ssize_t cnt;
968 1146
969 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1147 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
970 1148
971 if (cnt <= 0) 1149 if (cnt <= 0)
972 { 1150 {
989 } 1167 }
990 1168
991 return res; 1169 return res;
992} 1170}
993 1171
1172#ifdef PAGESIZE
1173# define eio_pagesize() PAGESIZE
1174#else
1175static intptr_t
1176eio_pagesize (void)
1177{
1178 static intptr_t page;
1179
1180 if (!page)
1181 page = sysconf (_SC_PAGESIZE);
1182
1183 return page;
1184}
1185#endif
1186
1187static void
1188eio_page_align (void **addr, size_t *length)
1189{
1190 intptr_t mask = eio_pagesize () - 1;
1191
1192 /* round down addr */
1193 intptr_t adj = mask & (intptr_t)*addr;
1194
1195 *addr = (void *)((intptr_t)*addr - adj);
1196 *length += adj;
1197
1198 /* round up length */
1199 *length = (*length + mask) & ~mask;
1200}
1201
1202#if !_POSIX_MEMLOCK
1203# define eio__mlockall(a) EIO_ENOSYS ()
1204#else
1205
994static int 1206static int
1207eio__mlockall (int flags)
1208{
1209 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1210 extern int mallopt (int, int);
1211 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1212 #endif
1213
1214 if (EIO_MCL_CURRENT != MCL_CURRENT
1215 || EIO_MCL_FUTURE != MCL_FUTURE)
1216 {
1217 flags = 0
1218 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1219 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1220 }
1221
1222 return mlockall (flags);
1223}
1224#endif
1225
1226#if !_POSIX_MEMLOCK_RANGE
1227# define eio__mlock(a,b) EIO_ENOSYS ()
1228#else
1229
1230static int
1231eio__mlock (void *addr, size_t length)
1232{
1233 eio_page_align (&addr, &length);
1234
1235 return mlock (addr, length);
1236}
1237
1238#endif
1239
1240#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1241# define eio__msync(a,b,c) EIO_ENOSYS ()
1242#else
1243
1244static int
1245eio__msync (void *mem, size_t len, int flags)
1246{
1247 eio_page_align (&mem, &len);
1248
1249 if (EIO_MS_ASYNC != MS_SYNC
1250 || EIO_MS_INVALIDATE != MS_INVALIDATE
1251 || EIO_MS_SYNC != MS_SYNC)
1252 {
1253 flags = 0
1254 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1255 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1256 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1257 }
1258
1259 return msync (mem, len, flags);
1260}
1261
1262#endif
1263
1264static int
1265eio__mtouch (eio_req *req)
1266{
1267 void *mem = req->ptr2;
1268 size_t len = req->size;
1269 int flags = req->int1;
1270
1271 eio_page_align (&mem, &len);
1272
1273 {
1274 intptr_t addr = (intptr_t)mem;
1275 intptr_t end = addr + len;
1276 intptr_t page = eio_pagesize ();
1277
1278 if (addr < end)
1279 if (flags & EIO_MT_MODIFY) /* modify */
1280 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1281 else
1282 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1283 }
1284
1285 return 0;
1286}
1287
1288/*****************************************************************************/
1289/* requests implemented outside eio_execute, because they are so large */
1290
1291static void
1292eio__realpath (eio_req *req, etp_worker *self)
1293{
1294 char *rel = req->ptr1;
1295 char *res;
1296 char *tmp1, *tmp2;
1297#if SYMLOOP_MAX > 32
1298 int symlinks = SYMLOOP_MAX;
1299#else
1300 int symlinks = 32;
1301#endif
1302
1303 req->result = -1;
1304
1305 errno = EINVAL;
1306 if (!rel)
1307 return;
1308
1309 errno = ENOENT;
1310 if (!*rel)
1311 return;
1312
1313 if (!req->ptr2)
1314 {
1315 X_LOCK (wrklock);
1316 req->flags |= EIO_FLAG_PTR2_FREE;
1317 X_UNLOCK (wrklock);
1318 req->ptr2 = malloc (PATH_MAX * 3);
1319
1320 errno = ENOMEM;
1321 if (!req->ptr2)
1322 return;
1323 }
1324
1325 res = req->ptr2;
1326 tmp1 = res + PATH_MAX;
1327 tmp2 = tmp1 + PATH_MAX;
1328
1329#if 0 /* disabled, the musl way to do things is just too racy */
1330#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1331 /* on linux we may be able to ask the kernel */
1332 {
1333 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1334
1335 if (fd >= 0)
1336 {
1337 sprintf (tmp1, "/proc/self/fd/%d", fd);
1338 req->result = readlink (tmp1, res, PATH_MAX);
1339 close (fd);
1340
1341 /* here we should probably stat the open file and the disk file, to make sure they still match */
1342
1343 if (req->result > 0)
1344 goto done;
1345 }
1346 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1347 return;
1348 }
1349#endif
1350#endif
1351
1352 if (*rel != '/')
1353 {
1354 if (!getcwd (res, PATH_MAX))
1355 return;
1356
1357 if (res [1]) /* only use if not / */
1358 res += strlen (res);
1359 }
1360
1361 while (*rel)
1362 {
1363 eio_ssize_t len, linklen;
1364 char *beg = rel;
1365
1366 while (*rel && *rel != '/')
1367 ++rel;
1368
1369 len = rel - beg;
1370
1371 if (!len) /* skip slashes */
1372 {
1373 ++rel;
1374 continue;
1375 }
1376
1377 if (beg [0] == '.')
1378 {
1379 if (len == 1)
1380 continue; /* . - nop */
1381
1382 if (beg [1] == '.' && len == 2)
1383 {
1384 /* .. - back up one component, if possible */
1385
1386 while (res != req->ptr2)
1387 if (*--res == '/')
1388 break;
1389
1390 continue;
1391 }
1392 }
1393
1394 errno = ENAMETOOLONG;
1395 if (res + 1 + len + 1 >= tmp1)
1396 return;
1397
1398 /* copy one component */
1399 *res = '/';
1400 memcpy (res + 1, beg, len);
1401
1402 /* zero-terminate, for readlink */
1403 res [len + 1] = 0;
1404
1405 /* now check if it's a symlink */
1406 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1407
1408 if (linklen < 0)
1409 {
1410 if (errno != EINVAL)
1411 return;
1412
1413 /* it's a normal directory. hopefully */
1414 res += len + 1;
1415 }
1416 else
1417 {
1418 /* yay, it was a symlink - build new path in tmp2 */
1419 int rellen = strlen (rel);
1420
1421 errno = ENAMETOOLONG;
1422 if (linklen + 1 + rellen >= PATH_MAX)
1423 return;
1424
1425 errno = ELOOP;
1426 if (!--symlinks)
1427 return;
1428
1429 if (*tmp1 == '/')
1430 res = req->ptr2; /* symlink resolves to an absolute path */
1431
1432 /* we need to be careful, as rel might point into tmp2 already */
1433 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1434 tmp2 [linklen] = '/';
1435 memcpy (tmp2, tmp1, linklen);
1436
1437 rel = tmp2;
1438 }
1439 }
1440
1441 /* special case for the lone root path */
1442 if (res == req->ptr2)
1443 *res++ = '/';
1444
1445 req->result = res - (char *)req->ptr2;
1446
1447done:
1448 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1449}
1450
1451static signed char
995eio_dent_cmp (const void *a_, const void *b_) 1452eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
996{ 1453{
997 const eio_dirent *a = (const eio_dirent *)a_; 1454 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_; 1455 : a->inode < b->inode ? -1
1456 : a->inode > b->inode ? 1
1457 : 0;
1458}
999 1459
1000 return (int)b->score - (int)a->score ? (int)b->score - (int)a->score 1460#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 */ 1461
1462#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1463#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1464
1465static void
1466eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1467{
1468 unsigned char bits [9 + sizeof (ino_t) * 8];
1469 unsigned char *bit = bits;
1470
1471 assert (CHAR_BIT == 8);
1472 assert (sizeof (eio_dirent) * 8 < 256);
1473 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1474 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1475
1476 if (size <= EIO_SORT_FAST)
1477 return;
1478
1479 /* first prepare an array of bits to test in our radix sort */
1480 /* try to take endianness into account, as well as differences in ino_t sizes */
1481 /* inode_bits must contain all inodes ORed together */
1482 /* which is used to skip bits that are 0 everywhere, which is very common */
1483 {
1484 ino_t endianness;
1485 int i, j;
1486
1487 /* we store the byte offset of byte n into byte n of "endianness" */
1488 for (i = 0; i < sizeof (ino_t); ++i)
1489 ((unsigned char *)&endianness)[i] = i;
1490
1491 *bit++ = 0;
1492
1493 for (i = 0; i < sizeof (ino_t); ++i)
1494 {
1495 /* shifting off the byte offsets out of "endianness" */
1496 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1497 endianness >>= 8;
1498
1499 for (j = 0; j < 8; ++j)
1500 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1501 *bit++ = offs + j;
1502 }
1503
1504 for (j = 0; j < 8; ++j)
1505 if (score_bits & (1 << j))
1506 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1507 }
1508
1509 /* now actually do the sorting (a variant of MSD radix sort) */
1510 {
1511 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1512 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1513 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1514 int stk_idx = 0;
1515
1516 base_stk [stk_idx] = dents;
1517 end_stk [stk_idx] = dents + size;
1518 bit_stk [stk_idx] = bit - 1;
1519
1520 do
1521 {
1522 base = base_stk [stk_idx];
1523 end = end_stk [stk_idx];
1524 bit = bit_stk [stk_idx];
1525
1526 for (;;)
1527 {
1528 unsigned char O = *bit >> 3;
1529 unsigned char M = 1 << (*bit & 7);
1530
1531 eio_dirent *a = base;
1532 eio_dirent *b = end;
1533
1534 if (b - a < EIO_SORT_CUTOFF)
1535 break;
1536
1537 /* now bit-partition the array on the bit */
1538 /* this ugly asymmetric loop seems to perform much better than typical */
1539 /* partition algos found in the literature */
1540 do
1541 if (!(((unsigned char *)a)[O] & M))
1542 ++a;
1543 else if (!(((unsigned char *)--b)[O] & M))
1544 {
1545 eio_dirent tmp = *a; *a = *b; *b = tmp;
1546 ++a;
1547 }
1548 while (b > a);
1549
1550 /* next bit, or stop, if no bits left in this path */
1551 if (!*--bit)
1552 break;
1553
1554 base_stk [stk_idx] = a;
1555 end_stk [stk_idx] = end;
1556 bit_stk [stk_idx] = bit;
1557 ++stk_idx;
1558
1559 end = a;
1560 }
1561 }
1562 while (stk_idx--);
1563 }
1564}
1565
1566static void
1567eio_dent_insertion_sort (eio_dirent *dents, int size)
1568{
1569 /* first move the smallest element to the front, to act as a sentinel */
1570 {
1571 int i;
1572 eio_dirent *min = dents;
1573
1574 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1575 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1576 if (EIO_DENT_CMP (dents [i], <, *min))
1577 min = &dents [i];
1578
1579 /* swap elements 0 and j (minimum) */
1580 {
1581 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1582 }
1583 }
1584
1585 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1586 {
1587 eio_dirent *i, *j;
1588
1589 for (i = dents + 1; i < dents + size; ++i)
1590 {
1591 eio_dirent value = *i;
1592
1593 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1594 j [1] = j [0];
1595
1596 j [1] = value;
1597 }
1598 }
1599}
1600
1601static void
1602eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1603{
1604 if (size <= 1)
1605 return; /* our insertion sort relies on size > 0 */
1606
1607 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1608 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1609 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1610
1611 /* use an insertion sort at the end, or for small arrays, */
1612 /* as insertion sort is more efficient for small partitions */
1613 eio_dent_insertion_sort (dents, size);
1002} 1614}
1003 1615
1004/* read a full directory */ 1616/* read a full directory */
1005static void 1617static void
1006eio__scandir (eio_req *req, etp_worker *self) 1618eio__scandir (eio_req *req, etp_worker *self)
1007{ 1619{
1008 DIR *dirp; 1620 DIR *dirp;
1009 EIO_STRUCT_DIRENT *entp; 1621 EIO_STRUCT_DIRENT *entp;
1010 unsigned char *name, *names; 1622 char *name, *names;
1011 int namesalloc = 4096; 1623 int namesalloc = 4096;
1012 int namesoffs = 0; 1624 int namesoffs = 0;
1013 int flags = req->int1; 1625 int flags = req->int1;
1014 eio_dirent *dents = 0; 1626 eio_dirent *dents = 0;
1015 int dentalloc = 128; 1627 int dentalloc = 128;
1016 int dentoffs = 0; 1628 int dentoffs = 0;
1629 ino_t inode_bits = 0;
1017 1630
1018 req->result = -1; 1631 req->result = -1;
1019 1632
1020 if (!(flags & EIO_READDIR_DENTS)) 1633 if (!(flags & EIO_READDIR_DENTS))
1021 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1634 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1022 1635
1023 X_LOCK (wrklock); 1636 X_LOCK (wrklock);
1024 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1637 /* the corresponding closedir is in ETP_WORKER_CLEAR */
1025 self->dirp = dirp = opendir (req->ptr1); 1638 self->dirp = dirp = opendir (req->ptr1);
1639
1640 if (req->flags & EIO_FLAG_PTR1_FREE)
1641 free (req->ptr1);
1642
1026 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1643 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; 1644 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1645 req->ptr2 = names = malloc (namesalloc);
1029 X_UNLOCK (wrklock); 1646 X_UNLOCK (wrklock);
1030 1647
1031 if (dirp && names && (!flags || dents)) 1648 if (dirp && names && (!flags || dents))
1032 for (;;) 1649 for (;;)
1033 { 1650 {
1041 1658
1042 /* sort etc. */ 1659 /* sort etc. */
1043 req->int1 = flags; 1660 req->int1 = flags;
1044 req->result = dentoffs; 1661 req->result = dentoffs;
1045 1662
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 1663 if (flags & EIO_READDIR_STAT_ORDER)
1055 || !(~flags & (EIO_READDIR_DIRS_FIRST | EIO_READDIR_FOUND_UNKNOWN))) 1664 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) 1665 else if (flags & EIO_READDIR_DIRS_FIRST)
1666 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1667 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1668 else
1061 { 1669 {
1062 /* in this case, all is known, and we just put dirs first and sort them */ 1670 /* in this case, all is known, and we just put dirs first and sort them */
1063 eio_dirent *ent = dents + dentoffs; 1671 eio_dirent *oth = dents + dentoffs;
1064 eio_dirent *dir = dents; 1672 eio_dirent *dir = dents;
1065 1673
1674 /* now partition dirs to the front, and non-dirs to the back */
1675 /* by walking from both sides and swapping if necessary */
1066 while (ent > dir) 1676 while (oth > dir)
1067 { 1677 {
1068 if (dir->type == DT_DIR) 1678 if (dir->type == EIO_DT_DIR)
1069 ++dir; 1679 ++dir;
1070 else
1071 {
1072 --ent;
1073
1074 if (ent->type == DT_DIR) 1680 else if ((--oth)->type == EIO_DT_DIR)
1075 { 1681 {
1076 eio_dirent tmp = *dir; 1682 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1077 *dir = *ent;
1078 *ent = tmp;
1079 1683
1080 ++dir; 1684 ++dir;
1081 } 1685 }
1082 } 1686 }
1687
1688 /* now sort the dirs only (dirs all have the same score) */
1689 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1083 } 1690 }
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 1691
1099 break; 1692 break;
1100 } 1693 }
1101 1694
1102 /* now add the entry to our list(s) */ 1695 /* now add the entry to our list(s) */
1105 /* skip . and .. entries */ 1698 /* skip . and .. entries */
1106 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1699 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1107 { 1700 {
1108 int len = D_NAMLEN (entp) + 1; 1701 int len = D_NAMLEN (entp) + 1;
1109 1702
1110 while (expect_false (namesoffs + len > namesalloc)) 1703 while (ecb_expect_false (namesoffs + len > namesalloc))
1111 { 1704 {
1112 namesalloc *= 2; 1705 namesalloc *= 2;
1113 X_LOCK (wrklock); 1706 X_LOCK (wrklock);
1114 req->ptr1 = names = realloc (names, namesalloc); 1707 req->ptr2 = names = realloc (names, namesalloc);
1115 X_UNLOCK (wrklock); 1708 X_UNLOCK (wrklock);
1116 1709
1117 if (!names) 1710 if (!names)
1118 break; 1711 break;
1119 } 1712 }
1122 1715
1123 if (dents) 1716 if (dents)
1124 { 1717 {
1125 struct eio_dirent *ent; 1718 struct eio_dirent *ent;
1126 1719
1127 if (expect_false (dentoffs == dentalloc)) 1720 if (ecb_expect_false (dentoffs == dentalloc))
1128 { 1721 {
1129 dentalloc *= 2; 1722 dentalloc *= 2;
1130 X_LOCK (wrklock); 1723 X_LOCK (wrklock);
1131 req->ptr2 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1724 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1132 X_UNLOCK (wrklock); 1725 X_UNLOCK (wrklock);
1133 1726
1134 if (!dents) 1727 if (!dents)
1135 break; 1728 break;
1136 } 1729 }
1137 1730
1138 ent = dents + dentoffs; 1731 ent = dents + dentoffs;
1139 1732
1140 ent->name = (char *)(size_t)namesoffs; /* rather dirtily we store the offset in the pointer */ 1733 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1141 ent->namelen = len - 1; 1734 ent->namelen = len - 1;
1142 ent->inode = D_INO (entp); 1735 ent->inode = D_INO (entp);
1736
1737 inode_bits |= ent->inode;
1143 1738
1144 switch (D_TYPE (entp)) 1739 switch (D_TYPE (entp))
1145 { 1740 {
1146 default: 1741 default:
1147 ent->type = EIO_DT_UNKNOWN; 1742 ent->type = EIO_DT_UNKNOWN;
1190 #ifdef DT_WHT 1785 #ifdef DT_WHT
1191 case DT_WHT: ent->type = EIO_DT_WHT; break; 1786 case DT_WHT: ent->type = EIO_DT_WHT; break;
1192 #endif 1787 #endif
1193 } 1788 }
1194 1789
1195 ent->score = 0; 1790 ent->score = 7;
1196 1791
1197 if (flags & EIO_READDIR_DIRS_FIRST) 1792 if (flags & EIO_READDIR_DIRS_FIRST)
1198 { 1793 {
1199 if (ent->type == EIO_DT_UNKNOWN) 1794 if (ent->type == EIO_DT_UNKNOWN)
1200 { 1795 {
1201 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ 1796 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1202 ent->score = 98; 1797 ent->score = 1;
1203 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ 1798 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 */ 1799 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1205 } 1800 }
1206 else if (ent->type == EIO_DT_DIR) 1801 else if (ent->type == EIO_DT_DIR)
1207 ent->score = 100; 1802 ent->score = 0;
1208 } 1803 }
1209 } 1804 }
1210 1805
1211 namesoffs += len; 1806 namesoffs += len;
1212 ++dentoffs; 1807 ++dentoffs;
1213 } 1808 }
1809
1810 if (EIO_CANCELLED (req))
1811 {
1812 errno = ECANCELED;
1813 break;
1814 }
1214 } 1815 }
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} 1816}
1248 1817
1249/*****************************************************************************/ 1818/*****************************************************************************/
1250 1819
1251#define ALLOC(len) \ 1820#define ALLOC(len) \
1266X_THREAD_PROC (etp_proc) 1835X_THREAD_PROC (etp_proc)
1267{ 1836{
1268 ETP_REQ *req; 1837 ETP_REQ *req;
1269 struct timespec ts; 1838 struct timespec ts;
1270 etp_worker *self = (etp_worker *)thr_arg; 1839 etp_worker *self = (etp_worker *)thr_arg;
1840 int timeout;
1271 1841
1272 /* try to distribute timeouts somewhat randomly */ 1842 /* try to distribute timeouts somewhat evenly */
1273 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1843 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1274 1844
1275 for (;;) 1845 for (;;)
1276 { 1846 {
1847 ts.tv_sec = 0;
1848
1277 X_LOCK (reqlock); 1849 X_LOCK (reqlock);
1278 1850
1279 for (;;) 1851 for (;;)
1280 { 1852 {
1281 self->req = req = reqq_shift (&req_queue); 1853 self->req = req = reqq_shift (&req_queue);
1282 1854
1283 if (req) 1855 if (req)
1284 break; 1856 break;
1285 1857
1858 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1859 {
1860 X_UNLOCK (reqlock);
1861 X_LOCK (wrklock);
1862 --started;
1863 X_UNLOCK (wrklock);
1864 goto quit;
1865 }
1866
1286 ++idle; 1867 ++idle;
1287 1868
1288 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1869 if (idle <= max_idle)
1289 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1870 /* we are allowed to idle, so do so without any timeout */
1871 X_COND_WAIT (reqwait, reqlock);
1872 else
1290 { 1873 {
1291 if (idle > max_idle) 1874 /* initialise timeout once */
1292 { 1875 if (!ts.tv_sec)
1293 --idle; 1876 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 1877
1301 /* we are allowed to idle, so do so without any timeout */
1302 X_COND_WAIT (reqwait, reqlock); 1878 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1879 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1303 } 1880 }
1304 1881
1305 --idle; 1882 --idle;
1306 } 1883 }
1307 1884
1310 X_UNLOCK (reqlock); 1887 X_UNLOCK (reqlock);
1311 1888
1312 if (req->type < 0) 1889 if (req->type < 0)
1313 goto quit; 1890 goto quit;
1314 1891
1315 if (!EIO_CANCELLED (req))
1316 ETP_EXECUTE (self, req); 1892 ETP_EXECUTE (self, req);
1317 1893
1318 X_LOCK (reslock); 1894 X_LOCK (reslock);
1319 1895
1320 ++npending; 1896 ++npending;
1321 1897
1336 return 0; 1912 return 0;
1337} 1913}
1338 1914
1339/*****************************************************************************/ 1915/*****************************************************************************/
1340 1916
1917int ecb_cold
1341int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1918eio_init (void (*want_poll)(void), void (*done_poll)(void))
1342{ 1919{
1343 return etp_init (want_poll, done_poll); 1920 return etp_init (want_poll, done_poll);
1344} 1921}
1345 1922
1923ecb_inline void
1346static void eio_api_destroy (eio_req *req) 1924eio_api_destroy (eio_req *req)
1347{ 1925{
1348 free (req); 1926 free (req);
1349} 1927}
1350 1928
1351#define REQ(rtype) \ 1929#define REQ(rtype) \
1370 { \ 1948 { \
1371 eio_api_destroy (req); \ 1949 eio_api_destroy (req); \
1372 return 0; \ 1950 return 0; \
1373 } 1951 }
1374 1952
1953static void
1375static void eio_execute (etp_worker *self, eio_req *req) 1954eio_execute (etp_worker *self, eio_req *req)
1376{ 1955{
1377 errno = 0; 1956 if (ecb_expect_false (EIO_CANCELLED (req)))
1957 {
1958 req->result = -1;
1959 req->errorno = ECANCELED;
1960 return;
1961 }
1378 1962
1379 switch (req->type) 1963 switch (req->type)
1380 { 1964 {
1381 case EIO_READ: ALLOC (req->size); 1965 case EIO_READ: ALLOC (req->size);
1382 req->result = req->offs >= 0 1966 req->result = req->offs >= 0
1394 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1978 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1395 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1979 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1396 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1980 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1397 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 1981 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1398 1982
1983 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1984 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1985 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1986 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1987
1399 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 1988 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; 1989 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; 1990 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; 1991 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; 1992 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1410 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 1999 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1411 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2000 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; 2001 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1413 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2002 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1414 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2003 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; 2004 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1416 2005
2006 case EIO_REALPATH: eio__realpath (req, self); break;
2007
1417 case EIO_READLINK: ALLOC (NAME_MAX); 2008 case EIO_READLINK: ALLOC (PATH_MAX);
1418 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2009 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1419 2010
1420 case EIO_SYNC: req->result = 0; sync (); break; 2011 case EIO_SYNC: req->result = 0; sync (); break;
1421 case EIO_FSYNC: req->result = fsync (req->int1); break; 2012 case EIO_FSYNC: req->result = fsync (req->int1); break;
1422 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2013 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1423 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 2014 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2015 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1424 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2016 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2017 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; 2018 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2019 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1426 2020
1427 case EIO_READDIR: eio__scandir (req, self); break; 2021 case EIO_READDIR: eio__scandir (req, self); break;
1428 2022
1429 case EIO_BUSY: 2023 case EIO_BUSY:
1430#ifdef _WIN32 2024#ifdef _WIN32
1431 Sleep (req->nv1 * 1000.); 2025 Sleep (req->nv1 * 1e3);
1432#else 2026#else
1433 { 2027 {
1434 struct timeval tv; 2028 struct timeval tv;
1435 2029
1436 tv.tv_sec = req->nv1; 2030 tv.tv_sec = req->nv1;
1437 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2031 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1438 2032
1439 req->result = select (0, 0, 0, 0, &tv); 2033 req->result = select (0, 0, 0, 0, &tv);
1440 } 2034 }
1441#endif 2035#endif
1442 break; 2036 break;
1457 times = tv; 2051 times = tv;
1458 } 2052 }
1459 else 2053 else
1460 times = 0; 2054 times = 0;
1461 2055
1462
1463 req->result = req->type == EIO_FUTIME 2056 req->result = req->type == EIO_FUTIME
1464 ? futimes (req->int1, times) 2057 ? futimes (req->int1, times)
1465 : utimes (req->ptr1, times); 2058 : utimes (req->ptr1, times);
1466 } 2059 }
1467 break; 2060 break;
1472 case EIO_NOP: 2065 case EIO_NOP:
1473 req->result = 0; 2066 req->result = 0;
1474 break; 2067 break;
1475 2068
1476 case EIO_CUSTOM: 2069 case EIO_CUSTOM:
1477 ((void (*)(eio_req *))req->feed) (req); 2070 req->feed (req);
1478 break; 2071 break;
1479 2072
1480 default: 2073 default:
2074 errno = ENOSYS;
1481 req->result = -1; 2075 req->result = -1;
1482 break; 2076 break;
1483 } 2077 }
1484 2078
1485 req->errorno = errno; 2079 req->errorno = errno;
1515eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2109eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1516{ 2110{
1517 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2111 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1518} 2112}
1519 2113
2114eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2115{
2116 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2117}
2118
2119eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2120{
2121 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2122}
2123
1520eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2124eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1521{ 2125{
1522 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2126 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1523} 2127}
1524 2128
2129eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2130{
2131 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2132}
2133
1525eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2134eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1526{ 2135{
1527 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2136 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1528} 2137}
1529 2138
1550eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2159eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1551{ 2160{
1552 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2161 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1553} 2162}
1554 2163
2164eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2165{
2166 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2167}
2168
1555eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2169eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1556{ 2170{
1557 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2171 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1558} 2172}
1559 2173
1565eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2179eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1566{ 2180{
1567 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2181 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1568} 2182}
1569 2183
1570eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2184eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1571{ 2185{
1572 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2186 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1573} 2187}
1574 2188
1575eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2189eio_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) 2209eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1596{ 2210{
1597 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2211 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1598} 2212}
1599 2213
1600eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2214eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1601{ 2215{
1602 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2216 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1603} 2217}
1604 2218
1605eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2219eio_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) 2235eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1622{ 2236{
1623 return eio__1path (EIO_READLINK, path, pri, cb, data); 2237 return eio__1path (EIO_READLINK, path, pri, cb, data);
1624} 2238}
1625 2239
2240eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2241{
2242 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2243}
2244
1626eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2245eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1627{ 2246{
1628 return eio__1path (EIO_STAT, path, pri, cb, data); 2247 return eio__1path (EIO_STAT, path, pri, cb, data);
1629} 2248}
1630 2249
1631eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2250eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1632{ 2251{
1633 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2252 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1634} 2253}
1635 2254
2255eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2256{
2257 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2258}
2259
1636eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2260eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1637{ 2261{
1638 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2262 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1639} 2263}
1640 2264
1648 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; 2272 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1649} 2273}
1650 2274
1651eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2275eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1652{ 2276{
1653 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2277 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1654} 2278}
1655 2279
1656static eio_req * 2280static eio_req *
1657eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2281eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1658{ 2282{
1682eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2306eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1683{ 2307{
1684 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2308 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1685} 2309}
1686 2310
1687eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2311eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1688{ 2312{
1689 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2313 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1690} 2314}
1691 2315
1692#endif 2316#endif
1693 2317
1694eio_req *eio_grp (eio_cb cb, void *data) 2318eio_req *eio_grp (eio_cb cb, void *data)
1703#undef SEND 2327#undef SEND
1704 2328
1705/*****************************************************************************/ 2329/*****************************************************************************/
1706/* grp functions */ 2330/* grp functions */
1707 2331
2332void
1708void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2333eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1709{ 2334{
1710 grp->int2 = limit; 2335 grp->int2 = limit;
1711 grp->feed = feed; 2336 grp->feed = feed;
1712 2337
1713 grp_try_feed (grp); 2338 grp_try_feed (grp);
1714} 2339}
1715 2340
2341void
1716void eio_grp_limit (eio_req *grp, int limit) 2342eio_grp_limit (eio_req *grp, int limit)
1717{ 2343{
1718 grp->int2 = limit; 2344 grp->int2 = limit;
1719 2345
1720 grp_try_feed (grp); 2346 grp_try_feed (grp);
1721} 2347}
1722 2348
2349void
1723void eio_grp_add (eio_req *grp, eio_req *req) 2350eio_grp_add (eio_req *grp, eio_req *req)
1724{ 2351{
1725 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2352 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1726 2353
1727 grp->flags |= EIO_FLAG_GROUPADD; 2354 grp->flags |= EIO_FLAG_GROUPADD;
1728 2355
1739} 2366}
1740 2367
1741/*****************************************************************************/ 2368/*****************************************************************************/
1742/* misc garbage */ 2369/* misc garbage */
1743 2370
2371eio_ssize_t
1744ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2372eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1745{ 2373{
1746 etp_worker wrk; 2374 etp_worker wrk;
2375 eio_ssize_t ret;
1747 2376
1748 wrk.dbuf = 0; 2377 wrk.dbuf = 0;
1749 2378
1750 eio__sendfile (ofd, ifd, offset, count, &wrk); 2379 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1751 2380
1752 if (wrk.dbuf) 2381 if (wrk.dbuf)
1753 free (wrk.dbuf); 2382 free (wrk.dbuf);
1754}
1755 2383
2384 return ret;
2385}
2386

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines