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

Comparing libeio/eio.c (file contents):
Revision 1.27 by root, Wed Oct 22 18:15:36 2008 UTC vs.
Revision 1.92 by root, Sun Jul 17 18:27:17 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
41#include "xthread.h" 50#include "xthread.h"
42 51
43#include <errno.h> 52#include <errno.h>
44#include <stddef.h> 53#include <stddef.h>
45#include <stdlib.h> 54#include <stdlib.h>
49#include <sys/stat.h> 58#include <sys/stat.h>
50#include <limits.h> 59#include <limits.h>
51#include <fcntl.h> 60#include <fcntl.h>
52#include <assert.h> 61#include <assert.h>
53 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76static void eio_destroy (eio_req *req);
77
54#ifndef EIO_FINISH 78#ifndef EIO_FINISH
55# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 79# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
56#endif 80#endif
57 81
58#ifndef EIO_DESTROY 82#ifndef EIO_DESTROY
61 85
62#ifndef EIO_FEED 86#ifndef EIO_FEED
63# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 87# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
64#endif 88#endif
65 89
90#ifndef EIO_FD_TO_WIN32_HANDLE
91# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
92#endif
93#ifndef EIO_WIN32_HANDLE_TO_FD
94# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
95#endif
96
97#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
98
99#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
100
66#ifdef _WIN32 101#ifdef _WIN32
67 102
68 /*doh*/ 103 #undef PAGESIZE
104 #define PAGESIZE 4096 /* GetSystemInfo? */
105
106 #ifdef EIO_STRUCT_STATI64
107 #define stat(path,buf) _stati64 (path,buf)
108 #define fstat(fd,buf) _fstati64 (path,buf)
109 #endif
110 #define lstat(path,buf) stat (path,buf)
111 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
112 #define mkdir(path,mode) _mkdir (path)
113 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
114
115 #define chmod(path,mode) _chmod (path, mode)
116 #define dup(fd) _dup (fd)
117 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
118
119 #define fchmod(fd,mode) EIO_ENOSYS ()
120 #define chown(path,uid,gid) EIO_ENOSYS ()
121 #define fchown(fd,uid,gid) EIO_ENOSYS ()
122 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
123 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
124 #define mknod(path,mode,dev) EIO_ENOSYS ()
125 #define sync() EIO_ENOSYS ()
126 #define readlink(path,buf,s) EIO_ENOSYS ()
127 #define statvfs(path,buf) EIO_ENOSYS ()
128 #define fstatvfs(fd,buf) EIO_ENOSYS ()
129
130 /* we could even stat and see if it exists */
131 static int
132 symlink (const char *old, const char *neu)
133 {
134 #if WINVER >= 0x0600
135 if (CreateSymbolicLink (neu, old, 1))
136 return 0;
137
138 if (CreateSymbolicLink (neu, old, 0))
139 return 0;
140 #endif
141
142 return EIO_ERRNO (ENOENT, -1);
143 }
144
145 /* POSIX API only */
146 #define CreateHardLink(neu,old,flags) 0
147 #define CreateSymbolicLink(neu,old,flags) 0
148
149 struct statvfs
150 {
151 int dummy;
152 };
153
154 #define DT_DIR EIO_DT_DIR
155 #define DT_REG EIO_DT_REG
156 #define D_NAME(entp) entp.cFileName
157 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
69 158
70#else 159#else
71 160
72# include "config.h"
73# include <sys/time.h> 161 #include <sys/time.h>
74# include <sys/select.h> 162 #include <sys/select.h>
163 #include <sys/statvfs.h>
75# include <unistd.h> 164 #include <unistd.h>
76# include <utime.h> 165 #include <utime.h>
77# include <signal.h> 166 #include <signal.h>
78# include <dirent.h> 167 #include <dirent.h>
79 168
169 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
170 #include <sys/mman.h>
171 #endif
172
173 #define D_NAME(entp) entp->d_name
174
175 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
176 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
177 #define _DIRENT_HAVE_D_TYPE /* sigh */
178 #define D_INO(de) (de)->d_fileno
179 #define D_NAMLEN(de) (de)->d_namlen
180 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
181 #define D_INO(de) (de)->d_ino
182 #endif
183
184 #ifdef _D_EXACT_NAMLEN
185 #undef D_NAMLEN
186 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
187 #endif
188
189 #ifdef _DIRENT_HAVE_D_TYPE
190 #define D_TYPE(de) (de)->d_type
191 #endif
192
80# ifndef EIO_STRUCT_DIRENT 193 #ifndef EIO_STRUCT_DIRENT
81# define EIO_STRUCT_DIRENT struct dirent 194 #define EIO_STRUCT_DIRENT struct dirent
82# endif 195 #endif
83 196
84#endif 197#endif
85 198
86#if HAVE_SENDFILE 199#if HAVE_SENDFILE
87# if __linux 200# if __linux
88# include <sys/sendfile.h> 201# include <sys/sendfile.h>
89# elif __freebsd 202# elif __FreeBSD__ || defined __APPLE__
90# include <sys/socket.h> 203# include <sys/socket.h>
91# include <sys/uio.h> 204# include <sys/uio.h>
92# elif __hpux 205# elif __hpux
93# include <sys/socket.h> 206# include <sys/socket.h>
94# elif __solaris /* not yet */ 207# elif __solaris
95# include <sys/sendfile.h> 208# include <sys/sendfile.h>
96# else 209# else
97# error sendfile support requested but not available 210# error sendfile support requested but not available
98# endif 211# endif
99#endif 212#endif
100 213
101/* number of seconds after which an idle threads exit */ 214#ifndef D_TYPE
102#define IDLE_TIMEOUT 10 215# define D_TYPE(de) 0
216#endif
217#ifndef D_INO
218# define D_INO(de) 0
219#endif
220#ifndef D_NAMLEN
221# define D_NAMLEN(entp) strlen (D_NAME (entp))
222#endif
103 223
104/* used for struct dirent, AIX doesn't provide it */ 224/* used for struct dirent, AIX doesn't provide it */
105#ifndef NAME_MAX 225#ifndef NAME_MAX
106# define NAME_MAX 4096 226# define NAME_MAX 4096
227#endif
228
229/* used for readlink etc. */
230#ifndef PATH_MAX
231# define PATH_MAX 4096
107#endif 232#endif
108 233
109/* buffer size for various temporary buffers */ 234/* buffer size for various temporary buffers */
110#define EIO_BUFSIZE 65536 235#define EIO_BUFSIZE 65536
111 236
118 if (!eio_buf) \ 243 if (!eio_buf) \
119 return -1; 244 return -1;
120 245
121#define EIO_TICKS ((1000000 + 1023) >> 10) 246#define EIO_TICKS ((1000000 + 1023) >> 10)
122 247
123/*****************************************************************************/
124
125#if __GNUC__ >= 3
126# define expect(expr,value) __builtin_expect ((expr),(value))
127#else
128# define expect(expr,value) (expr)
129#endif
130
131#define expect_false(expr) expect ((expr) != 0, 0)
132#define expect_true(expr) expect ((expr) != 0, 1)
133
134/*****************************************************************************/
135
136#define ETP_PRI_MIN EIO_PRI_MIN 248#define ETP_PRI_MIN EIO_PRI_MIN
137#define ETP_PRI_MAX EIO_PRI_MAX 249#define ETP_PRI_MAX EIO_PRI_MAX
138 250
139struct etp_worker; 251struct etp_worker;
140 252
148#define ETP_WORKER_CLEAR(req) \ 260#define ETP_WORKER_CLEAR(req) \
149 if (wrk->dbuf) \ 261 if (wrk->dbuf) \
150 { \ 262 { \
151 free (wrk->dbuf); \ 263 free (wrk->dbuf); \
152 wrk->dbuf = 0; \ 264 wrk->dbuf = 0; \
153 } \
154 \
155 if (wrk->dirp) \
156 { \
157 closedir (wrk->dirp); \
158 wrk->dirp = 0; \
159 } 265 }
266
160#define ETP_WORKER_COMMON \ 267#define ETP_WORKER_COMMON \
161 void *dbuf; \ 268 void *dbuf;
162 DIR *dirp;
163 269
164/*****************************************************************************/ 270/*****************************************************************************/
165 271
166#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 272#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
167 273
168/* calculcate time difference in ~1/EIO_TICKS of a second */ 274/* calculate time difference in ~1/EIO_TICKS of a second */
275ecb_inline int
169static int tvdiff (struct timeval *tv1, struct timeval *tv2) 276tvdiff (struct timeval *tv1, struct timeval *tv2)
170{ 277{
171 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 278 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
172 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 279 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
173} 280}
174 281
181static unsigned int max_poll_reqs; /* reslock */ 288static unsigned int max_poll_reqs; /* reslock */
182 289
183static volatile unsigned int nreqs; /* reqlock */ 290static volatile unsigned int nreqs; /* reqlock */
184static volatile unsigned int nready; /* reqlock */ 291static volatile unsigned int nready; /* reqlock */
185static volatile unsigned int npending; /* reqlock */ 292static volatile unsigned int npending; /* reqlock */
186static volatile unsigned int max_idle = 4; 293static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
294static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
187 295
188static mutex_t wrklock = X_MUTEX_INIT; 296static xmutex_t wrklock;
189static mutex_t reslock = X_MUTEX_INIT; 297static xmutex_t reslock;
190static mutex_t reqlock = X_MUTEX_INIT; 298static xmutex_t reqlock;
191static cond_t reqwait = X_COND_INIT; 299static xcond_t reqwait;
192 300
193#if !HAVE_PREADWRITE 301#if !HAVE_PREADWRITE
194/* 302/*
195 * make our pread/pwrite emulation safe against themselves, but not against 303 * make our pread/pwrite emulation safe against themselves, but not against
196 * normal read/write by using a mutex. slows down execution a lot, 304 * normal read/write by using a mutex. slows down execution a lot,
197 * but that's your problem, not mine. 305 * but that's your problem, not mine.
198 */ 306 */
199static mutex_t preadwritelock = X_MUTEX_INIT; 307static xmutex_t preadwritelock = X_MUTEX_INIT;
200#endif 308#endif
201 309
202typedef struct etp_worker 310typedef struct etp_worker
203{ 311{
204 /* locked by wrklock */ 312 /* locked by wrklock */
205 struct etp_worker *prev, *next; 313 struct etp_worker *prev, *next;
206 314
207 thread_t tid; 315 xthread_t tid;
208 316
209 /* locked by reslock, reqlock or wrklock */ 317 /* locked by reslock, reqlock or wrklock */
210 ETP_REQ *req; /* currently processed request */ 318 ETP_REQ *req; /* currently processed request */
211 319
212 ETP_WORKER_COMMON 320 ETP_WORKER_COMMON
217#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 325#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
218#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 326#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
219 327
220/* worker threads management */ 328/* worker threads management */
221 329
330static void ecb_cold
222static void etp_worker_clear (etp_worker *wrk) 331etp_worker_clear (etp_worker *wrk)
223{ 332{
224 ETP_WORKER_CLEAR (wrk); 333 ETP_WORKER_CLEAR (wrk);
225} 334}
226 335
336static void ecb_cold
227static void etp_worker_free (etp_worker *wrk) 337etp_worker_free (etp_worker *wrk)
228{ 338{
229 wrk->next->prev = wrk->prev; 339 wrk->next->prev = wrk->prev;
230 wrk->prev->next = wrk->next; 340 wrk->prev->next = wrk->next;
231 341
232 free (wrk); 342 free (wrk);
233} 343}
234 344
235static unsigned int etp_nreqs (void) 345static unsigned int
346etp_nreqs (void)
236{ 347{
237 int retval; 348 int retval;
238 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 349 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
239 retval = nreqs; 350 retval = nreqs;
240 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 351 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
241 return retval; 352 return retval;
242} 353}
243 354
244static unsigned int etp_nready (void) 355static unsigned int
356etp_nready (void)
245{ 357{
246 unsigned int retval; 358 unsigned int retval;
247 359
248 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 360 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
249 retval = nready; 361 retval = nready;
250 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 362 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
251 363
252 return retval; 364 return retval;
253} 365}
254 366
255static unsigned int etp_npending (void) 367static unsigned int
368etp_npending (void)
256{ 369{
257 unsigned int retval; 370 unsigned int retval;
258 371
259 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 372 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
260 retval = npending; 373 retval = npending;
261 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 374 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
262 375
263 return retval; 376 return retval;
264} 377}
265 378
266static unsigned int etp_nthreads (void) 379static unsigned int
380etp_nthreads (void)
267{ 381{
268 unsigned int retval; 382 unsigned int retval;
269 383
270 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 384 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
271 retval = started; 385 retval = started;
285} etp_reqq; 399} etp_reqq;
286 400
287static etp_reqq req_queue; 401static etp_reqq req_queue;
288static etp_reqq res_queue; 402static etp_reqq res_queue;
289 403
404static void ecb_noinline ecb_cold
405reqq_init (etp_reqq *q)
406{
407 int pri;
408
409 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
410 q->qs[pri] = q->qe[pri] = 0;
411
412 q->size = 0;
413}
414
415static int ecb_noinline
290static int reqq_push (etp_reqq *q, ETP_REQ *req) 416reqq_push (etp_reqq *q, ETP_REQ *req)
291{ 417{
292 int pri = req->pri; 418 int pri = req->pri;
293 req->next = 0; 419 req->next = 0;
294 420
295 if (q->qe[pri]) 421 if (q->qe[pri])
301 q->qe[pri] = q->qs[pri] = req; 427 q->qe[pri] = q->qs[pri] = req;
302 428
303 return q->size++; 429 return q->size++;
304} 430}
305 431
432static ETP_REQ * ecb_noinline
306static ETP_REQ *reqq_shift (etp_reqq *q) 433reqq_shift (etp_reqq *q)
307{ 434{
308 int pri; 435 int pri;
309 436
310 if (!q->size) 437 if (!q->size)
311 return 0; 438 return 0;
326 } 453 }
327 454
328 abort (); 455 abort ();
329} 456}
330 457
331static void etp_atfork_prepare (void) 458static void ecb_cold
459etp_thread_init (void)
332{ 460{
333 X_LOCK (wrklock);
334 X_LOCK (reqlock);
335 X_LOCK (reslock);
336#if !HAVE_PREADWRITE 461#if !HAVE_PREADWRITE
337 X_LOCK (preadwritelock);
338#endif
339}
340
341static void etp_atfork_parent (void)
342{
343#if !HAVE_PREADWRITE
344 X_UNLOCK (preadwritelock); 462 X_MUTEX_CREATE (preadwritelock);
345#endif 463#endif
346 X_UNLOCK (reslock); 464 X_MUTEX_CREATE (wrklock);
347 X_UNLOCK (reqlock); 465 X_MUTEX_CREATE (reslock);
348 X_UNLOCK (wrklock); 466 X_MUTEX_CREATE (reqlock);
467 X_COND_CREATE (reqwait);
349} 468}
350 469
470static void ecb_cold
351static void etp_atfork_child (void) 471etp_atfork_child (void)
352{ 472{
353 ETP_REQ *prv; 473 reqq_init (&req_queue);
474 reqq_init (&res_queue);
354 475
355 while ((prv = reqq_shift (&req_queue))) 476 wrk_first.next =
356 ETP_DESTROY (prv); 477 wrk_first.prev = &wrk_first;
357
358 while ((prv = reqq_shift (&res_queue)))
359 ETP_DESTROY (prv);
360
361 while (wrk_first.next != &wrk_first)
362 {
363 etp_worker *wrk = wrk_first.next;
364
365 if (wrk->req)
366 ETP_DESTROY (wrk->req);
367
368 etp_worker_clear (wrk);
369 etp_worker_free (wrk);
370 }
371 478
372 started = 0; 479 started = 0;
373 idle = 0; 480 idle = 0;
374 nreqs = 0; 481 nreqs = 0;
375 nready = 0; 482 nready = 0;
376 npending = 0; 483 npending = 0;
377 484
378 etp_atfork_parent (); 485 etp_thread_init ();
379} 486}
380 487
381static void 488static void ecb_cold
382etp_once_init (void) 489etp_once_init (void)
383{ 490{
384 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); 491 etp_thread_init ();
492 X_THREAD_ATFORK (0, 0, etp_atfork_child);
385} 493}
386 494
387static int 495static int ecb_cold
388etp_init (void (*want_poll)(void), void (*done_poll)(void)) 496etp_init (void (*want_poll)(void), void (*done_poll)(void))
389{ 497{
390 static pthread_once_t doinit = PTHREAD_ONCE_INIT; 498 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
391 499
392 pthread_once (&doinit, etp_once_init); 500 pthread_once (&doinit, etp_once_init);
397 return 0; 505 return 0;
398} 506}
399 507
400X_THREAD_PROC (etp_proc); 508X_THREAD_PROC (etp_proc);
401 509
510static void ecb_cold
402static void etp_start_thread (void) 511etp_start_thread (void)
403{ 512{
404 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 513 etp_worker *wrk = calloc (1, sizeof (etp_worker));
405 514
406 /*TODO*/ 515 /*TODO*/
407 assert (("unable to allocate worker thread data", wrk)); 516 assert (("unable to allocate worker thread data", wrk));
420 free (wrk); 529 free (wrk);
421 530
422 X_UNLOCK (wrklock); 531 X_UNLOCK (wrklock);
423} 532}
424 533
534static void
425static void etp_maybe_start_thread (void) 535etp_maybe_start_thread (void)
426{ 536{
427 if (expect_true (etp_nthreads () >= wanted)) 537 if (ecb_expect_true (etp_nthreads () >= wanted))
428 return; 538 return;
429 539
430 /* todo: maybe use idle here, but might be less exact */ 540 /* todo: maybe use idle here, but might be less exact */
431 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 541 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
432 return; 542 return;
433 543
434 etp_start_thread (); 544 etp_start_thread ();
435} 545}
436 546
547static void ecb_cold
437static void etp_end_thread (void) 548etp_end_thread (void)
438{ 549{
439 eio_req *req = calloc (1, sizeof (eio_req)); 550 eio_req *req = calloc (1, sizeof (eio_req));
440 551
441 req->type = -1; 552 req->type = -1;
442 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 553 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
449 X_LOCK (wrklock); 560 X_LOCK (wrklock);
450 --started; 561 --started;
451 X_UNLOCK (wrklock); 562 X_UNLOCK (wrklock);
452} 563}
453 564
454static int etp_poll (void) 565static int
566etp_poll (void)
455{ 567{
456 unsigned int maxreqs; 568 unsigned int maxreqs;
457 unsigned int maxtime; 569 unsigned int maxtime;
458 struct timeval tv_start, tv_now; 570 struct timeval tv_start, tv_now;
459 571
489 601
490 X_LOCK (reqlock); 602 X_LOCK (reqlock);
491 --nreqs; 603 --nreqs;
492 X_UNLOCK (reqlock); 604 X_UNLOCK (reqlock);
493 605
494 if (expect_false (req->type == EIO_GROUP && req->size)) 606 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
495 { 607 {
496 req->int1 = 1; /* mark request as delayed */ 608 req->int1 = 1; /* mark request as delayed */
497 continue; 609 continue;
498 } 610 }
499 else 611 else
500 { 612 {
501 int res = ETP_FINISH (req); 613 int res = ETP_FINISH (req);
502 if (expect_false (res)) 614 if (ecb_expect_false (res))
503 return res; 615 return res;
504 } 616 }
505 617
506 if (expect_false (maxreqs && !--maxreqs)) 618 if (ecb_expect_false (maxreqs && !--maxreqs))
507 break; 619 break;
508 620
509 if (maxtime) 621 if (maxtime)
510 { 622 {
511 gettimeofday (&tv_now, 0); 623 gettimeofday (&tv_now, 0);
517 629
518 errno = EAGAIN; 630 errno = EAGAIN;
519 return -1; 631 return -1;
520} 632}
521 633
634static void
522static void etp_cancel (ETP_REQ *req) 635etp_cancel (ETP_REQ *req)
523{ 636{
524 X_LOCK (wrklock); 637 req->cancelled = 1;
525 req->flags |= EIO_FLAG_CANCELLED;
526 X_UNLOCK (wrklock);
527 638
528 eio_grp_cancel (req); 639 eio_grp_cancel (req);
529} 640}
530 641
642static void
531static void etp_submit (ETP_REQ *req) 643etp_submit (ETP_REQ *req)
532{ 644{
533 req->pri -= ETP_PRI_MIN; 645 req->pri -= ETP_PRI_MIN;
534 646
535 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 647 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
536 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 648 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
537 649
538 if (expect_false (req->type == EIO_GROUP)) 650 if (ecb_expect_false (req->type == EIO_GROUP))
539 { 651 {
540 /* I hope this is worth it :/ */ 652 /* I hope this is worth it :/ */
541 X_LOCK (reqlock); 653 X_LOCK (reqlock);
542 ++nreqs; 654 ++nreqs;
543 X_UNLOCK (reqlock); 655 X_UNLOCK (reqlock);
562 674
563 etp_maybe_start_thread (); 675 etp_maybe_start_thread ();
564 } 676 }
565} 677}
566 678
679static void ecb_cold
567static void etp_set_max_poll_time (double nseconds) 680etp_set_max_poll_time (double nseconds)
568{ 681{
569 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 682 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
570 max_poll_time = nseconds; 683 max_poll_time = nseconds * EIO_TICKS;
571 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 684 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
572} 685}
573 686
687static void ecb_cold
574static void etp_set_max_poll_reqs (unsigned int maxreqs) 688etp_set_max_poll_reqs (unsigned int maxreqs)
575{ 689{
576 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 690 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
577 max_poll_reqs = maxreqs; 691 max_poll_reqs = maxreqs;
578 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 692 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
579} 693}
580 694
695static void ecb_cold
581static void etp_set_max_idle (unsigned int nthreads) 696etp_set_max_idle (unsigned int nthreads)
582{ 697{
583 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 698 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
584 max_idle = nthreads <= 0 ? 1 : nthreads; 699 max_idle = nthreads;
585 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 700 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
586} 701}
587 702
703static void ecb_cold
704etp_set_idle_timeout (unsigned int seconds)
705{
706 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
707 idle_timeout = seconds;
708 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
709}
710
711static void ecb_cold
588static void etp_set_min_parallel (unsigned int nthreads) 712etp_set_min_parallel (unsigned int nthreads)
589{ 713{
590 if (wanted < nthreads) 714 if (wanted < nthreads)
591 wanted = nthreads; 715 wanted = nthreads;
592} 716}
593 717
718static void ecb_cold
594static void etp_set_max_parallel (unsigned int nthreads) 719etp_set_max_parallel (unsigned int nthreads)
595{ 720{
596 if (wanted > nthreads) 721 if (wanted > nthreads)
597 wanted = nthreads; 722 wanted = nthreads;
598 723
599 while (started > wanted) 724 while (started > wanted)
600 etp_end_thread (); 725 etp_end_thread ();
601} 726}
602 727
603/*****************************************************************************/ 728/*****************************************************************************/
604 729
730static void
605static void grp_try_feed (eio_req *grp) 731grp_try_feed (eio_req *grp)
606{ 732{
607 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 733 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
608 { 734 {
609 grp->flags &= ~EIO_FLAG_GROUPADD; 735 grp->flags &= ~EIO_FLAG_GROUPADD;
610 736
617 break; 743 break;
618 } 744 }
619 } 745 }
620} 746}
621 747
748static int
622static int grp_dec (eio_req *grp) 749grp_dec (eio_req *grp)
623{ 750{
624 --grp->size; 751 --grp->size;
625 752
626 /* call feeder, if applicable */ 753 /* call feeder, if applicable */
627 grp_try_feed (grp); 754 grp_try_feed (grp);
631 return eio_finish (grp); 758 return eio_finish (grp);
632 else 759 else
633 return 0; 760 return 0;
634} 761}
635 762
763static void
636void eio_destroy (eio_req *req) 764eio_destroy (eio_req *req)
637{ 765{
638 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 766 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
639 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 767 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
640 768
641 EIO_DESTROY (req); 769 EIO_DESTROY (req);
642} 770}
643 771
772static int
644static int eio_finish (eio_req *req) 773eio_finish (eio_req *req)
645{ 774{
646 int res = EIO_FINISH (req); 775 int res = EIO_FINISH (req);
647 776
648 if (req->grp) 777 if (req->grp)
649 { 778 {
657 if (grp->grp_first == req) 786 if (grp->grp_first == req)
658 grp->grp_first = req->grp_next; 787 grp->grp_first = req->grp_next;
659 788
660 res2 = grp_dec (grp); 789 res2 = grp_dec (grp);
661 790
662 if (!res && res2) 791 if (!res)
663 res = res2; 792 res = res2;
664 } 793 }
665 794
666 eio_destroy (req); 795 eio_destroy (req);
667 796
668 return res; 797 return res;
669} 798}
670 799
800void
671void eio_grp_cancel (eio_req *grp) 801eio_grp_cancel (eio_req *grp)
672{ 802{
673 for (grp = grp->grp_first; grp; grp = grp->grp_next) 803 for (grp = grp->grp_first; grp; grp = grp->grp_next)
674 eio_cancel (grp); 804 eio_cancel (grp);
675} 805}
676 806
807void
677void eio_cancel (eio_req *req) 808eio_cancel (eio_req *req)
678{ 809{
679 etp_cancel (req); 810 etp_cancel (req);
680} 811}
681 812
813void
682void eio_submit (eio_req *req) 814eio_submit (eio_req *req)
683{ 815{
684 etp_submit (req); 816 etp_submit (req);
685} 817}
686 818
687unsigned int eio_nreqs (void) 819unsigned int
820eio_nreqs (void)
688{ 821{
689 return etp_nreqs (); 822 return etp_nreqs ();
690} 823}
691 824
692unsigned int eio_nready (void) 825unsigned int
826eio_nready (void)
693{ 827{
694 return etp_nready (); 828 return etp_nready ();
695} 829}
696 830
697unsigned int eio_npending (void) 831unsigned int
832eio_npending (void)
698{ 833{
699 return etp_npending (); 834 return etp_npending ();
700} 835}
701 836
702unsigned int eio_nthreads (void) 837unsigned int ecb_cold
838eio_nthreads (void)
703{ 839{
704 return etp_nthreads (); 840 return etp_nthreads ();
705} 841}
706 842
843void ecb_cold
707void eio_set_max_poll_time (double nseconds) 844eio_set_max_poll_time (double nseconds)
708{ 845{
709 etp_set_max_poll_time (nseconds); 846 etp_set_max_poll_time (nseconds);
710} 847}
711 848
849void ecb_cold
712void eio_set_max_poll_reqs (unsigned int maxreqs) 850eio_set_max_poll_reqs (unsigned int maxreqs)
713{ 851{
714 etp_set_max_poll_reqs (maxreqs); 852 etp_set_max_poll_reqs (maxreqs);
715} 853}
716 854
855void ecb_cold
717void eio_set_max_idle (unsigned int nthreads) 856eio_set_max_idle (unsigned int nthreads)
718{ 857{
719 etp_set_max_idle (nthreads); 858 etp_set_max_idle (nthreads);
720} 859}
721 860
861void ecb_cold
862eio_set_idle_timeout (unsigned int seconds)
863{
864 etp_set_idle_timeout (seconds);
865}
866
867void ecb_cold
722void eio_set_min_parallel (unsigned int nthreads) 868eio_set_min_parallel (unsigned int nthreads)
723{ 869{
724 etp_set_min_parallel (nthreads); 870 etp_set_min_parallel (nthreads);
725} 871}
726 872
873void ecb_cold
727void eio_set_max_parallel (unsigned int nthreads) 874eio_set_max_parallel (unsigned int nthreads)
728{ 875{
729 etp_set_max_parallel (nthreads); 876 etp_set_max_parallel (nthreads);
730} 877}
731 878
732int eio_poll (void) 879int eio_poll (void)
741# undef pread 888# undef pread
742# undef pwrite 889# undef pwrite
743# define pread eio__pread 890# define pread eio__pread
744# define pwrite eio__pwrite 891# define pwrite eio__pwrite
745 892
746static ssize_t 893static eio_ssize_t
747eio__pread (int fd, void *buf, size_t count, off_t offset) 894eio__pread (int fd, void *buf, size_t count, off_t offset)
748{ 895{
749 ssize_t res; 896 eio_ssize_t res;
750 off_t ooffset; 897 off_t ooffset;
751 898
752 X_LOCK (preadwritelock); 899 X_LOCK (preadwritelock);
753 ooffset = lseek (fd, 0, SEEK_CUR); 900 ooffset = lseek (fd, 0, SEEK_CUR);
754 lseek (fd, offset, SEEK_SET); 901 lseek (fd, offset, SEEK_SET);
757 X_UNLOCK (preadwritelock); 904 X_UNLOCK (preadwritelock);
758 905
759 return res; 906 return res;
760} 907}
761 908
762static ssize_t 909static eio_ssize_t
763eio__pwrite (int fd, void *buf, size_t count, off_t offset) 910eio__pwrite (int fd, void *buf, size_t count, off_t offset)
764{ 911{
765 ssize_t res; 912 eio_ssize_t res;
766 off_t ooffset; 913 off_t ooffset;
767 914
768 X_LOCK (preadwritelock); 915 X_LOCK (preadwritelock);
769 ooffset = lseek (fd, 0, SEEK_CUR); 916 ooffset = lseek (fd, 0, SEEK_CUR);
770 lseek (fd, offset, SEEK_SET); 917 lseek (fd, offset, SEEK_SET);
771 res = write (fd, buf, count); 918 res = write (fd, buf, count);
772 lseek (fd, offset, SEEK_SET); 919 lseek (fd, ooffset, SEEK_SET);
773 X_UNLOCK (preadwritelock); 920 X_UNLOCK (preadwritelock);
774 921
775 return res; 922 return res;
776} 923}
777#endif 924#endif
778 925
779#ifndef HAVE_FUTIMES 926#ifndef HAVE_UTIMES
780 927
781# undef utimes 928# undef utimes
782# undef futimes
783# define utimes(path,times) eio__utimes (path, times) 929# define utimes(path,times) eio__utimes (path, times)
784# define futimes(fd,times) eio__futimes (fd, times)
785 930
786static int 931static int
787eio__utimes (const char *filename, const struct timeval times[2]) 932eio__utimes (const char *filename, const struct timeval times[2])
788{ 933{
789 if (times) 934 if (times)
797 } 942 }
798 else 943 else
799 return utime (filename, 0); 944 return utime (filename, 0);
800} 945}
801 946
947#endif
948
949#ifndef HAVE_FUTIMES
950
951# undef futimes
952# define futimes(fd,times) eio__futimes (fd, times)
953
954static int
802static int eio__futimes (int fd, const struct timeval tv[2]) 955eio__futimes (int fd, const struct timeval tv[2])
803{ 956{
804 errno = ENOSYS; 957 errno = ENOSYS;
805 return -1; 958 return -1;
806} 959}
807 960
811# undef fdatasync 964# undef fdatasync
812# define fdatasync(fd) fsync (fd) 965# define fdatasync(fd) fsync (fd)
813#endif 966#endif
814 967
815/* sync_file_range always needs emulation */ 968/* sync_file_range always needs emulation */
816int 969static int
817eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 970eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
818{ 971{
819#if HAVE_SYNC_FILE_RANGE 972#if HAVE_SYNC_FILE_RANGE
820 int res; 973 int res;
821 974
829 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); 982 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
830 } 983 }
831 984
832 res = sync_file_range (fd, offset, nbytes, flags); 985 res = sync_file_range (fd, offset, nbytes, flags);
833 986
834 if (res != ENOSYS) 987 if (!res || errno != ENOSYS)
835 return res; 988 return res;
836#endif 989#endif
837 990
838 /* even though we could play tricks with the flags, it's better to always 991 /* even though we could play tricks with the flags, it's better to always
839 * call fdatasync, as thta matches the expectation of it's users best */ 992 * call fdatasync, as that matches the expectation of its users best */
840 return fdatasync (fd); 993 return fdatasync (fd);
994}
995
996static int
997eio__fallocate (int fd, int mode, off_t offset, size_t len)
998{
999#if HAVE_FALLOCATE
1000 return fallocate (fd, mode, offset, len);
1001#else
1002 errno = ENOSYS;
1003 return -1;
1004#endif
841} 1005}
842 1006
843#if !HAVE_READAHEAD 1007#if !HAVE_READAHEAD
844# undef readahead 1008# undef readahead
845# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1009# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
846 1010
847static ssize_t 1011static eio_ssize_t
848eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1012eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
849{ 1013{
850 size_t todo = count; 1014 size_t todo = count;
851 dBUF; 1015 dBUF;
852 1016
864} 1028}
865 1029
866#endif 1030#endif
867 1031
868/* sendfile always needs emulation */ 1032/* sendfile always needs emulation */
869static ssize_t 1033static eio_ssize_t
870eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1034eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
871{ 1035{
1036 eio_ssize_t written = 0;
872 ssize_t res; 1037 eio_ssize_t res;
873 1038
874 if (!count) 1039 if (!count)
875 return 0; 1040 return 0;
876 1041
1042 for (;;)
1043 {
1044#ifdef __APPLE__
1045# undef HAVE_SENDFILE /* broken, as everything on os x */
1046#endif
877#if HAVE_SENDFILE 1047#if HAVE_SENDFILE
878# if __linux 1048# if __linux
1049 off_t soffset = offset;
879 res = sendfile (ofd, ifd, &offset, count); 1050 res = sendfile (ofd, ifd, &soffset, count);
880 1051
881# elif __freebsd 1052# elif __FreeBSD__
882 /* 1053 /*
883 * Of course, the freebsd sendfile is a dire hack with no thoughts 1054 * Of course, the freebsd sendfile is a dire hack with no thoughts
884 * wasted on making it similar to other I/O functions. 1055 * wasted on making it similar to other I/O functions.
885 */ 1056 */
886 {
887 off_t sbytes; 1057 off_t sbytes;
888 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1058 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
889 1059
890 if (res < 0 && sbytes) 1060 #if 0 /* according to the manpage, this is correct, but broken behaviour */
891 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1061 /* freebsd' sendfile will return 0 on success */
1062 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1063 /* not on e.g. EIO or EPIPE - sounds broken */
1064 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
892 res = sbytes; 1065 res = sbytes;
893 } 1066 #endif
1067
1068 /* according to source inspection, this is correct, and useful behaviour */
1069 if (sbytes)
1070 res = sbytes;
1071
1072# elif defined (__APPLE__)
1073 off_t sbytes = count;
1074 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1075
1076 /* according to the manpage, sbytes is always valid */
1077 if (sbytes)
1078 res = sbytes;
894 1079
895# elif __hpux 1080# elif __hpux
896 res = sendfile (ofd, ifd, offset, count, 0, 0); 1081 res = sendfile (ofd, ifd, offset, count, 0, 0);
897 1082
898# elif __solaris 1083# elif __solaris
899 {
900 struct sendfilevec vec; 1084 struct sendfilevec vec;
901 size_t sbytes; 1085 size_t sbytes;
902 1086
903 vec.sfv_fd = ifd; 1087 vec.sfv_fd = ifd;
904 vec.sfv_flag = 0; 1088 vec.sfv_flag = 0;
905 vec.sfv_off = offset; 1089 vec.sfv_off = offset;
906 vec.sfv_len = count; 1090 vec.sfv_len = count;
907 1091
908 res = sendfilev (ofd, &vec, 1, &sbytes); 1092 res = sendfilev (ofd, &vec, 1, &sbytes);
909 1093
910 if (res < 0 && sbytes) 1094 if (res < 0 && sbytes)
911 res = sbytes; 1095 res = sbytes;
912 }
913 1096
914# endif 1097# endif
1098
1099#elif defined (_WIN32) && 0
1100 /* does not work, just for documentation of what would need to be done */
1101 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1102 /* libeio guarantees that the file offset does not change, and windows */
1103 /* has no way to get an independent handle to the same file description */
1104 HANDLE h = TO_SOCKET (ifd);
1105 SetFilePointer (h, offset, 0, FILE_BEGIN);
1106 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1107
915#else 1108#else
916 res = -1; 1109 res = -1;
917 errno = ENOSYS; 1110 errno = ENOSYS;
918#endif 1111#endif
919 1112
1113 /* we assume sendfile can copy at least 128mb in one go */
1114 if (res <= 128 * 1024 * 1024)
1115 {
1116 if (res > 0)
1117 written += res;
1118
1119 if (written)
1120 return written;
1121
1122 break;
1123 }
1124 else
1125 {
1126 /* if we requested more, then probably the kernel was lazy */
1127 written += res;
1128 offset += res;
1129 count -= res;
1130
1131 if (!count)
1132 return written;
1133 }
1134 }
1135
920 if (res < 0 1136 if (res < 0
921 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1137 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1138 /* BSDs */
1139#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1140 || errno == ENOTSUP
1141#endif
1142#ifdef EOPNOTSUPP /* windows */
1143 || errno == EOPNOTSUPP /* BSDs */
1144#endif
922#if __solaris 1145#if __solaris
923 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1146 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
924#endif 1147#endif
925 ) 1148 )
926 ) 1149 )
930 1153
931 res = 0; 1154 res = 0;
932 1155
933 while (count) 1156 while (count)
934 { 1157 {
935 ssize_t cnt; 1158 eio_ssize_t cnt;
936 1159
937 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1160 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
938 1161
939 if (cnt <= 0) 1162 if (cnt <= 0)
940 { 1163 {
957 } 1180 }
958 1181
959 return res; 1182 return res;
960} 1183}
961 1184
1185#ifdef PAGESIZE
1186# define eio_pagesize() PAGESIZE
1187#else
1188static intptr_t
1189eio_pagesize (void)
1190{
1191 static intptr_t page;
1192
1193 if (!page)
1194 page = sysconf (_SC_PAGESIZE);
1195
1196 return page;
1197}
1198#endif
1199
1200static void
1201eio_page_align (void **addr, size_t *length)
1202{
1203 intptr_t mask = eio_pagesize () - 1;
1204
1205 /* round down addr */
1206 intptr_t adj = mask & (intptr_t)*addr;
1207
1208 *addr = (void *)((intptr_t)*addr - adj);
1209 *length += adj;
1210
1211 /* round up length */
1212 *length = (*length + mask) & ~mask;
1213}
1214
1215#if !_POSIX_MEMLOCK
1216# define eio__mlockall(a) EIO_ENOSYS ()
1217#else
1218
1219static int
1220eio__mlockall (int flags)
1221{
1222 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1223 extern int mallopt (int, int);
1224 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1225 #endif
1226
1227 if (EIO_MCL_CURRENT != MCL_CURRENT
1228 || EIO_MCL_FUTURE != MCL_FUTURE)
1229 {
1230 flags = 0
1231 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1232 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1233 }
1234
1235 return mlockall (flags);
1236}
1237#endif
1238
1239#if !_POSIX_MEMLOCK_RANGE
1240# define eio__mlock(a,b) EIO_ENOSYS ()
1241#else
1242
1243static int
1244eio__mlock (void *addr, size_t length)
1245{
1246 eio_page_align (&addr, &length);
1247
1248 return mlock (addr, length);
1249}
1250
1251#endif
1252
1253#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1254# define eio__msync(a,b,c) EIO_ENOSYS ()
1255#else
1256
1257static int
1258eio__msync (void *mem, size_t len, int flags)
1259{
1260 eio_page_align (&mem, &len);
1261
1262 if (EIO_MS_ASYNC != MS_SYNC
1263 || EIO_MS_INVALIDATE != MS_INVALIDATE
1264 || EIO_MS_SYNC != MS_SYNC)
1265 {
1266 flags = 0
1267 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1268 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1269 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1270 }
1271
1272 return msync (mem, len, flags);
1273}
1274
1275#endif
1276
1277static int
1278eio__mtouch (eio_req *req)
1279{
1280 void *mem = req->ptr2;
1281 size_t len = req->size;
1282 int flags = req->int1;
1283
1284 eio_page_align (&mem, &len);
1285
1286 {
1287 intptr_t addr = (intptr_t)mem;
1288 intptr_t end = addr + len;
1289 intptr_t page = eio_pagesize ();
1290
1291 if (addr < end)
1292 if (flags & EIO_MT_MODIFY) /* modify */
1293 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1294 else
1295 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1296 }
1297
1298 return 0;
1299}
1300
1301/*****************************************************************************/
1302/* requests implemented outside eio_execute, because they are so large */
1303
1304static void
1305eio__realpath (eio_req *req, etp_worker *self)
1306{
1307 char *rel = req->ptr1;
1308 char *res;
1309 char *tmp1, *tmp2;
1310#if SYMLOOP_MAX > 32
1311 int symlinks = SYMLOOP_MAX;
1312#else
1313 int symlinks = 32;
1314#endif
1315
1316 req->result = -1;
1317
1318 errno = EINVAL;
1319 if (!rel)
1320 return;
1321
1322 errno = ENOENT;
1323 if (!*rel)
1324 return;
1325
1326 if (!req->ptr2)
1327 {
1328 X_LOCK (wrklock);
1329 req->flags |= EIO_FLAG_PTR2_FREE;
1330 X_UNLOCK (wrklock);
1331 req->ptr2 = malloc (PATH_MAX * 3);
1332
1333 errno = ENOMEM;
1334 if (!req->ptr2)
1335 return;
1336 }
1337
1338 res = req->ptr2;
1339 tmp1 = res + PATH_MAX;
1340 tmp2 = tmp1 + PATH_MAX;
1341
1342#if 0 /* disabled, the musl way to do things is just too racy */
1343#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1344 /* on linux we may be able to ask the kernel */
1345 {
1346 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1347
1348 if (fd >= 0)
1349 {
1350 sprintf (tmp1, "/proc/self/fd/%d", fd);
1351 req->result = readlink (tmp1, res, PATH_MAX);
1352 close (fd);
1353
1354 /* here we should probably stat the open file and the disk file, to make sure they still match */
1355
1356 if (req->result > 0)
1357 goto done;
1358 }
1359 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1360 return;
1361 }
1362#endif
1363#endif
1364
1365 if (*rel != '/')
1366 {
1367 if (!getcwd (res, PATH_MAX))
1368 return;
1369
1370 if (res [1]) /* only use if not / */
1371 res += strlen (res);
1372 }
1373
1374 while (*rel)
1375 {
1376 eio_ssize_t len, linklen;
1377 char *beg = rel;
1378
1379 while (*rel && *rel != '/')
1380 ++rel;
1381
1382 len = rel - beg;
1383
1384 if (!len) /* skip slashes */
1385 {
1386 ++rel;
1387 continue;
1388 }
1389
1390 if (beg [0] == '.')
1391 {
1392 if (len == 1)
1393 continue; /* . - nop */
1394
1395 if (beg [1] == '.' && len == 2)
1396 {
1397 /* .. - back up one component, if possible */
1398
1399 while (res != req->ptr2)
1400 if (*--res == '/')
1401 break;
1402
1403 continue;
1404 }
1405 }
1406
1407 errno = ENAMETOOLONG;
1408 if (res + 1 + len + 1 >= tmp1)
1409 return;
1410
1411 /* copy one component */
1412 *res = '/';
1413 memcpy (res + 1, beg, len);
1414
1415 /* zero-terminate, for readlink */
1416 res [len + 1] = 0;
1417
1418 /* now check if it's a symlink */
1419 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1420
1421 if (linklen < 0)
1422 {
1423 if (errno != EINVAL)
1424 return;
1425
1426 /* it's a normal directory. hopefully */
1427 res += len + 1;
1428 }
1429 else
1430 {
1431 /* yay, it was a symlink - build new path in tmp2 */
1432 int rellen = strlen (rel);
1433
1434 errno = ENAMETOOLONG;
1435 if (linklen + 1 + rellen >= PATH_MAX)
1436 return;
1437
1438 errno = ELOOP;
1439 if (!--symlinks)
1440 return;
1441
1442 if (*tmp1 == '/')
1443 res = req->ptr2; /* symlink resolves to an absolute path */
1444
1445 /* we need to be careful, as rel might point into tmp2 already */
1446 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1447 tmp2 [linklen] = '/';
1448 memcpy (tmp2, tmp1, linklen);
1449
1450 rel = tmp2;
1451 }
1452 }
1453
1454 /* special case for the lone root path */
1455 if (res == req->ptr2)
1456 *res++ = '/';
1457
1458 req->result = res - (char *)req->ptr2;
1459
1460done:
1461 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1462}
1463
1464static signed char
1465eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1466{
1467 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1468 : a->inode < b->inode ? -1
1469 : a->inode > b->inode ? 1
1470 : 0;
1471}
1472
1473#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1474
1475#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1476#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1477
1478static void
1479eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1480{
1481 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1482 unsigned char *bit = bits;
1483
1484 assert (CHAR_BIT == 8);
1485 assert (sizeof (eio_dirent) * 8 < 256);
1486 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1487 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1488
1489 if (size <= EIO_SORT_FAST)
1490 return;
1491
1492 /* first prepare an array of bits to test in our radix sort */
1493 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1494 /* inode_bits must contain all inodes ORed together */
1495 /* which is used to skip bits that are 0 everywhere, which is very common */
1496 {
1497 eio_ino_t endianness;
1498 int i, j;
1499
1500 /* we store the byte offset of byte n into byte n of "endianness" */
1501 for (i = 0; i < sizeof (eio_ino_t); ++i)
1502 ((unsigned char *)&endianness)[i] = i;
1503
1504 *bit++ = 0;
1505
1506 for (i = 0; i < sizeof (eio_ino_t); ++i)
1507 {
1508 /* shifting off the byte offsets out of "endianness" */
1509 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1510 endianness >>= 8;
1511
1512 for (j = 0; j < 8; ++j)
1513 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1514 *bit++ = offs + j;
1515 }
1516
1517 for (j = 0; j < 8; ++j)
1518 if (score_bits & (1 << j))
1519 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1520 }
1521
1522 /* now actually do the sorting (a variant of MSD radix sort) */
1523 {
1524 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1525 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1526 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1527 int stk_idx = 0;
1528
1529 base_stk [stk_idx] = dents;
1530 end_stk [stk_idx] = dents + size;
1531 bit_stk [stk_idx] = bit - 1;
1532
1533 do
1534 {
1535 base = base_stk [stk_idx];
1536 end = end_stk [stk_idx];
1537 bit = bit_stk [stk_idx];
1538
1539 for (;;)
1540 {
1541 unsigned char O = *bit >> 3;
1542 unsigned char M = 1 << (*bit & 7);
1543
1544 eio_dirent *a = base;
1545 eio_dirent *b = end;
1546
1547 if (b - a < EIO_SORT_CUTOFF)
1548 break;
1549
1550 /* now bit-partition the array on the bit */
1551 /* this ugly asymmetric loop seems to perform much better than typical */
1552 /* partition algos found in the literature */
1553 do
1554 if (!(((unsigned char *)a)[O] & M))
1555 ++a;
1556 else if (!(((unsigned char *)--b)[O] & M))
1557 {
1558 eio_dirent tmp = *a; *a = *b; *b = tmp;
1559 ++a;
1560 }
1561 while (b > a);
1562
1563 /* next bit, or stop, if no bits left in this path */
1564 if (!*--bit)
1565 break;
1566
1567 base_stk [stk_idx] = a;
1568 end_stk [stk_idx] = end;
1569 bit_stk [stk_idx] = bit;
1570 ++stk_idx;
1571
1572 end = a;
1573 }
1574 }
1575 while (stk_idx--);
1576 }
1577}
1578
1579static void
1580eio_dent_insertion_sort (eio_dirent *dents, int size)
1581{
1582 /* first move the smallest element to the front, to act as a sentinel */
1583 {
1584 int i;
1585 eio_dirent *min = dents;
1586
1587 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1588 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1589 if (EIO_DENT_CMP (dents [i], <, *min))
1590 min = &dents [i];
1591
1592 /* swap elements 0 and j (minimum) */
1593 {
1594 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1595 }
1596 }
1597
1598 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1599 {
1600 eio_dirent *i, *j;
1601
1602 for (i = dents + 1; i < dents + size; ++i)
1603 {
1604 eio_dirent value = *i;
1605
1606 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1607 j [1] = j [0];
1608
1609 j [1] = value;
1610 }
1611 }
1612}
1613
1614static void
1615eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1616{
1617 if (size <= 1)
1618 return; /* our insertion sort relies on size > 0 */
1619
1620 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1621 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1622 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1623
1624 /* use an insertion sort at the end, or for small arrays, */
1625 /* as insertion sort is more efficient for small partitions */
1626 eio_dent_insertion_sort (dents, size);
1627}
1628
962/* read a full directory */ 1629/* read a full directory */
963static void 1630static void
964eio__scandir (eio_req *req, etp_worker *self) 1631eio__scandir (eio_req *req, etp_worker *self)
965{ 1632{
1633 char *name, *names;
1634 int namesalloc = 4096 - sizeof (void *) * 4;
1635 int namesoffs = 0;
1636 int flags = req->int1;
1637 eio_dirent *dents = 0;
1638 int dentalloc = 128;
1639 int dentoffs = 0;
1640 eio_ino_t inode_bits = 0;
1641#ifdef _WIN32
1642 HANDLE dirp;
1643 WIN32_FIND_DATA entp;
1644#else
966 DIR *dirp; 1645 DIR *dirp;
967 EIO_STRUCT_DIRENT *entp; 1646 EIO_STRUCT_DIRENT *entp;
968 char *name, *names; 1647#endif
969 int memlen = 4096;
970 int memofs = 0;
971 int res = 0;
972 1648
973 X_LOCK (wrklock); 1649 req->result = -1;
974 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1650
1651 if (!(flags & EIO_READDIR_DENTS))
1652 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1653
1654#ifdef _WIN32
1655 {
1656 int len = strlen ((const char *)req->ptr1);
1657 char *path = malloc (MAX_PATH);
1658 const char *fmt;
1659
1660 if (!len)
1661 fmt = "./*";
1662 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1663 fmt = "%s*";
1664 else
1665 fmt = "%s/*";
1666
1667 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1668 dirp = FindFirstFile (path, &entp);
1669 free (path);
1670
1671 if (dirp == INVALID_HANDLE_VALUE)
1672 {
1673 dirp = 0;
1674
1675 switch (GetLastError ())
1676 {
1677 case ERROR_FILE_NOT_FOUND:
1678 req->result = 0;
1679 break;
1680
1681 case ERROR_INVALID_NAME:
1682 case ERROR_PATH_NOT_FOUND:
1683 case ERROR_NO_MORE_FILES:
1684 errno = ENOENT;
1685 break;
1686
1687 case ERROR_NOT_ENOUGH_MEMORY:
1688 errno = ENOMEM;
1689 break;
1690
1691 default:
1692 errno = EINVAL;
1693 break;
1694 }
1695 }
1696 }
1697#else
975 self->dirp = dirp = opendir (req->ptr1); 1698 dirp = opendir (req->ptr1);
1699#endif
1700
976 req->flags |= EIO_FLAG_PTR2_FREE; 1701 if (req->flags & EIO_FLAG_PTR1_FREE)
1702 free (req->ptr1);
1703
1704 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1705 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
977 req->ptr2 = names = malloc (memlen); 1706 req->ptr2 = names = malloc (namesalloc);
978 X_UNLOCK (wrklock);
979 1707
980 if (dirp && names) 1708 if (dirp && names && (!flags || dents))
981 for (;;) 1709 for (;;)
982 { 1710 {
1711 int done;
1712
1713#ifdef _WIN32
1714 done = !dirp;
1715#else
983 errno = 0; 1716 errno = 0;
984 entp = readdir (dirp); 1717 entp = readdir (dirp);
1718 done = !entp;
1719#endif
985 1720
986 if (!entp) 1721 if (done)
1722 {
1723#ifndef _WIN32
1724 int old_errno = errno;
1725 closedir (dirp);
1726 errno = old_errno;
1727
1728 if (errno)
1729 break;
1730#endif
1731
1732 /* sort etc. */
1733 req->int1 = flags;
1734 req->result = dentoffs;
1735
1736 if (flags & EIO_READDIR_STAT_ORDER)
1737 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1738 else if (flags & EIO_READDIR_DIRS_FIRST)
1739 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1740 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1741 else
1742 {
1743 /* in this case, all is known, and we just put dirs first and sort them */
1744 eio_dirent *oth = dents + dentoffs;
1745 eio_dirent *dir = dents;
1746
1747 /* now partition dirs to the front, and non-dirs to the back */
1748 /* by walking from both sides and swapping if necessary */
1749 while (oth > dir)
1750 {
1751 if (dir->type == EIO_DT_DIR)
1752 ++dir;
1753 else if ((--oth)->type == EIO_DT_DIR)
1754 {
1755 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1756
1757 ++dir;
1758 }
1759 }
1760
1761 /* now sort the dirs only (dirs all have the same score) */
1762 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1763 }
1764
987 break; 1765 break;
1766 }
988 1767
989 name = entp->d_name; 1768 /* now add the entry to our list(s) */
1769 name = D_NAME (entp);
990 1770
1771 /* skip . and .. entries */
991 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1772 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
992 { 1773 {
993 int len = strlen (name) + 1; 1774 int len = D_NAMLEN (entp) + 1;
994 1775
995 res++; 1776 while (ecb_expect_false (namesoffs + len > namesalloc))
996
997 while (memofs + len > memlen)
998 { 1777 {
999 memlen *= 2; 1778 namesalloc *= 2;
1000 X_LOCK (wrklock);
1001 req->ptr2 = names = realloc (names, memlen); 1779 req->ptr2 = names = realloc (names, namesalloc);
1002 X_UNLOCK (wrklock);
1003 1780
1004 if (!names) 1781 if (!names)
1005 break; 1782 break;
1006 } 1783 }
1007 1784
1008 memcpy (names + memofs, name, len); 1785 memcpy (names + namesoffs, name, len);
1786
1787 if (dents)
1788 {
1789 struct eio_dirent *ent;
1790
1791 if (ecb_expect_false (dentoffs == dentalloc))
1792 {
1793 dentalloc *= 2;
1794 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1795
1796 if (!dents)
1797 break;
1798 }
1799
1800 ent = dents + dentoffs;
1801
1802 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1803 ent->namelen = len - 1;
1804 ent->inode = D_INO (entp);
1805
1806 inode_bits |= ent->inode;
1807
1808 switch (D_TYPE (entp))
1809 {
1810 default:
1811 ent->type = EIO_DT_UNKNOWN;
1812 flags |= EIO_READDIR_FOUND_UNKNOWN;
1813 break;
1814
1815 #ifdef DT_FIFO
1816 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1817 #endif
1818 #ifdef DT_CHR
1819 case DT_CHR: ent->type = EIO_DT_CHR; break;
1820 #endif
1821 #ifdef DT_MPC
1822 case DT_MPC: ent->type = EIO_DT_MPC; break;
1823 #endif
1824 #ifdef DT_DIR
1825 case DT_DIR: ent->type = EIO_DT_DIR; break;
1826 #endif
1827 #ifdef DT_NAM
1828 case DT_NAM: ent->type = EIO_DT_NAM; break;
1829 #endif
1830 #ifdef DT_BLK
1831 case DT_BLK: ent->type = EIO_DT_BLK; break;
1832 #endif
1833 #ifdef DT_MPB
1834 case DT_MPB: ent->type = EIO_DT_MPB; break;
1835 #endif
1836 #ifdef DT_REG
1837 case DT_REG: ent->type = EIO_DT_REG; break;
1838 #endif
1839 #ifdef DT_NWK
1840 case DT_NWK: ent->type = EIO_DT_NWK; break;
1841 #endif
1842 #ifdef DT_CMP
1843 case DT_CMP: ent->type = EIO_DT_CMP; break;
1844 #endif
1845 #ifdef DT_LNK
1846 case DT_LNK: ent->type = EIO_DT_LNK; break;
1847 #endif
1848 #ifdef DT_SOCK
1849 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1850 #endif
1851 #ifdef DT_DOOR
1852 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1853 #endif
1854 #ifdef DT_WHT
1855 case DT_WHT: ent->type = EIO_DT_WHT; break;
1856 #endif
1857 }
1858
1859 ent->score = 7;
1860
1861 if (flags & EIO_READDIR_DIRS_FIRST)
1862 {
1863 if (ent->type == EIO_DT_UNKNOWN)
1864 {
1865 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1866 ent->score = 1;
1867 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1868 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1869 }
1870 else if (ent->type == EIO_DT_DIR)
1871 ent->score = 0;
1872 }
1873 }
1874
1009 memofs += len; 1875 namesoffs += len;
1876 ++dentoffs;
1010 } 1877 }
1878
1879 if (EIO_CANCELLED (req))
1880 {
1881 errno = ECANCELED;
1882 break;
1883 }
1884
1885#ifdef _WIN32
1886 if (!FindNextFile (dirp, &entp))
1887 {
1888 FindClose (dirp);
1889 dirp = 0;
1890 }
1891#endif
1011 } 1892 }
1012
1013 if (errno)
1014 res = -1;
1015
1016 req->result = res;
1017}
1018
1019#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1020# undef msync
1021# define msync(a,b,c) ENOSYS
1022#endif
1023
1024int
1025eio__mtouch (void *mem, size_t len, int flags)
1026{
1027 intptr_t addr = (intptr_t)mem;
1028 intptr_t end = addr + len;
1029#ifdef PAGESIZE
1030 const intptr_t page = PAGESIZE;
1031#else
1032 static intptr_t page;
1033
1034 if (!page)
1035 page = sysconf (_SC_PAGESIZE);
1036#endif
1037
1038 addr &= ~(page - 1); /* assume page size is always a power of two */
1039
1040 if (addr < end)
1041 if (flags) /* modify */
1042 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1043 else
1044 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1045
1046 return 0;
1047} 1893}
1048 1894
1049/*****************************************************************************/ 1895/*****************************************************************************/
1050 1896
1051#define ALLOC(len) \ 1897#define ALLOC(len) \
1067{ 1913{
1068 ETP_REQ *req; 1914 ETP_REQ *req;
1069 struct timespec ts; 1915 struct timespec ts;
1070 etp_worker *self = (etp_worker *)thr_arg; 1916 etp_worker *self = (etp_worker *)thr_arg;
1071 1917
1072 /* try to distribute timeouts somewhat randomly */ 1918 /* try to distribute timeouts somewhat evenly */
1073 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1919 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1074 1920
1075 for (;;) 1921 for (;;)
1076 { 1922 {
1923 ts.tv_sec = 0;
1924
1077 X_LOCK (reqlock); 1925 X_LOCK (reqlock);
1078 1926
1079 for (;;) 1927 for (;;)
1080 { 1928 {
1081 self->req = req = reqq_shift (&req_queue); 1929 self->req = req = reqq_shift (&req_queue);
1082 1930
1083 if (req) 1931 if (req)
1084 break; 1932 break;
1085 1933
1934 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1935 {
1936 X_UNLOCK (reqlock);
1937 X_LOCK (wrklock);
1938 --started;
1939 X_UNLOCK (wrklock);
1940 goto quit;
1941 }
1942
1086 ++idle; 1943 ++idle;
1087 1944
1088 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1945 if (idle <= max_idle)
1089 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1946 /* we are allowed to idle, so do so without any timeout */
1947 X_COND_WAIT (reqwait, reqlock);
1948 else
1090 { 1949 {
1091 if (idle > max_idle) 1950 /* initialise timeout once */
1092 { 1951 if (!ts.tv_sec)
1093 --idle; 1952 ts.tv_sec = time (0) + idle_timeout;
1094 X_UNLOCK (reqlock);
1095 X_LOCK (wrklock);
1096 --started;
1097 X_UNLOCK (wrklock);
1098 goto quit;
1099 }
1100 1953
1101 /* we are allowed to idle, so do so without any timeout */
1102 X_COND_WAIT (reqwait, reqlock); 1954 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1955 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1103 } 1956 }
1104 1957
1105 --idle; 1958 --idle;
1106 } 1959 }
1107 1960
1110 X_UNLOCK (reqlock); 1963 X_UNLOCK (reqlock);
1111 1964
1112 if (req->type < 0) 1965 if (req->type < 0)
1113 goto quit; 1966 goto quit;
1114 1967
1115 if (!EIO_CANCELLED (req))
1116 ETP_EXECUTE (self, req); 1968 ETP_EXECUTE (self, req);
1117 1969
1118 X_LOCK (reslock); 1970 X_LOCK (reslock);
1119 1971
1120 ++npending; 1972 ++npending;
1121 1973
1136 return 0; 1988 return 0;
1137} 1989}
1138 1990
1139/*****************************************************************************/ 1991/*****************************************************************************/
1140 1992
1993int ecb_cold
1141int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1994eio_init (void (*want_poll)(void), void (*done_poll)(void))
1142{ 1995{
1143 return etp_init (want_poll, done_poll); 1996 return etp_init (want_poll, done_poll);
1144} 1997}
1145 1998
1999ecb_inline void
1146static void eio_api_destroy (eio_req *req) 2000eio_api_destroy (eio_req *req)
1147{ 2001{
1148 free (req); 2002 free (req);
1149} 2003}
1150 2004
1151#define REQ(rtype) \ 2005#define REQ(rtype) \
1170 { \ 2024 { \
1171 eio_api_destroy (req); \ 2025 eio_api_destroy (req); \
1172 return 0; \ 2026 return 0; \
1173 } 2027 }
1174 2028
2029static void
1175static void eio_execute (etp_worker *self, eio_req *req) 2030eio_execute (etp_worker *self, eio_req *req)
1176{ 2031{
1177 errno = 0; 2032 if (ecb_expect_false (EIO_CANCELLED (req)))
2033 {
2034 req->result = -1;
2035 req->errorno = ECANCELED;
2036 return;
2037 }
1178 2038
1179 switch (req->type) 2039 switch (req->type)
1180 { 2040 {
1181 case EIO_READ: ALLOC (req->size); 2041 case EIO_READ: ALLOC (req->size);
1182 req->result = req->offs >= 0 2042 req->result = req->offs >= 0
1194 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2054 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1195 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2055 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1196 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2056 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1197 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2057 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1198 2058
2059 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2060 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2061 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2062 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2063
1199 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2064 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1200 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2065 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1201 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2066 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1202 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2067 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1203 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2068 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1210 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2075 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1211 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2076 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1212 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2077 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1213 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2078 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1214 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2079 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1215 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2080 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1216 2081
2082 case EIO_REALPATH: eio__realpath (req, self); break;
2083
1217 case EIO_READLINK: ALLOC (NAME_MAX); 2084 case EIO_READLINK: ALLOC (PATH_MAX);
1218 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2085 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1219 2086
1220 case EIO_SYNC: req->result = 0; sync (); break; 2087 case EIO_SYNC: req->result = 0; sync (); break;
1221 case EIO_FSYNC: req->result = fsync (req->int1); break; 2088 case EIO_FSYNC: req->result = fsync (req->int1); break;
1222 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2089 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1223 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 2090 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2091 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1224 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2092 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2093 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1225 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2094 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2095 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1226 2096
1227 case EIO_READDIR: eio__scandir (req, self); break; 2097 case EIO_READDIR: eio__scandir (req, self); break;
1228 2098
1229 case EIO_BUSY: 2099 case EIO_BUSY:
1230#ifdef _WIN32 2100#ifdef _WIN32
1231 Sleep (req->nv1 * 1000.); 2101 Sleep (req->nv1 * 1e3);
1232#else 2102#else
1233 { 2103 {
1234 struct timeval tv; 2104 struct timeval tv;
1235 2105
1236 tv.tv_sec = req->nv1; 2106 tv.tv_sec = req->nv1;
1237 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2107 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1238 2108
1239 req->result = select (0, 0, 0, 0, &tv); 2109 req->result = select (0, 0, 0, 0, &tv);
1240 } 2110 }
1241#endif 2111#endif
1242 break; 2112 break;
1257 times = tv; 2127 times = tv;
1258 } 2128 }
1259 else 2129 else
1260 times = 0; 2130 times = 0;
1261 2131
1262
1263 req->result = req->type == EIO_FUTIME 2132 req->result = req->type == EIO_FUTIME
1264 ? futimes (req->int1, times) 2133 ? futimes (req->int1, times)
1265 : utimes (req->ptr1, times); 2134 : utimes (req->ptr1, times);
1266 } 2135 }
1267 break; 2136 break;
1272 case EIO_NOP: 2141 case EIO_NOP:
1273 req->result = 0; 2142 req->result = 0;
1274 break; 2143 break;
1275 2144
1276 case EIO_CUSTOM: 2145 case EIO_CUSTOM:
1277 ((void (*)(eio_req *))req->feed) (req); 2146 req->feed (req);
1278 break; 2147 break;
1279 2148
1280 default: 2149 default:
2150 errno = ENOSYS;
1281 req->result = -1; 2151 req->result = -1;
1282 break; 2152 break;
1283 } 2153 }
1284 2154
1285 req->errorno = errno; 2155 req->errorno = errno;
1315eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2185eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1316{ 2186{
1317 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2187 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1318} 2188}
1319 2189
2190eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2191{
2192 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2193}
2194
2195eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2196{
2197 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2198}
2199
1320eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2200eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1321{ 2201{
1322 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2202 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1323} 2203}
1324 2204
2205eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2206{
2207 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2208}
2209
1325eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2210eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1326{ 2211{
1327 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2212 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1328} 2213}
1329 2214
1350eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2235eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1351{ 2236{
1352 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2237 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1353} 2238}
1354 2239
2240eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2241{
2242 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2243}
2244
1355eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2245eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1356{ 2246{
1357 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2247 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1358} 2248}
1359 2249
1365eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2255eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1366{ 2256{
1367 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2257 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1368} 2258}
1369 2259
1370eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2260eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1371{ 2261{
1372 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2262 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1373} 2263}
1374 2264
1375eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2265eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1395eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2285eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1396{ 2286{
1397 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2287 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1398} 2288}
1399 2289
1400eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2290eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1401{ 2291{
1402 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2292 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1403} 2293}
1404 2294
1405eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2295eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1421eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2311eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1422{ 2312{
1423 return eio__1path (EIO_READLINK, path, pri, cb, data); 2313 return eio__1path (EIO_READLINK, path, pri, cb, data);
1424} 2314}
1425 2315
2316eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2317{
2318 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2319}
2320
1426eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2321eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1427{ 2322{
1428 return eio__1path (EIO_STAT, path, pri, cb, data); 2323 return eio__1path (EIO_STAT, path, pri, cb, data);
1429} 2324}
1430 2325
1431eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2326eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1432{ 2327{
1433 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2328 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1434} 2329}
1435 2330
2331eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2332{
2333 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2334}
2335
1436eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2336eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1437{ 2337{
1438 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2338 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1439} 2339}
1440 2340
1441eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2341eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1442{ 2342{
1443 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2343 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1444} 2344}
1445 2345
1446eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2346eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1447{ 2347{
1448 return eio__1path (EIO_READDIR, path, pri, cb, data); 2348 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1449} 2349}
1450 2350
1451eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2351eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1452{ 2352{
1453 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2353 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1454} 2354}
1455 2355
1456static eio_req * 2356static eio_req *
1457eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2357eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1458{ 2358{
1482eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2382eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1483{ 2383{
1484 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2384 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1485} 2385}
1486 2386
1487eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2387eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1488{ 2388{
1489 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2389 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1490} 2390}
1491 2391
1492#endif 2392#endif
1493 2393
1494eio_req *eio_grp (eio_cb cb, void *data) 2394eio_req *eio_grp (eio_cb cb, void *data)
1503#undef SEND 2403#undef SEND
1504 2404
1505/*****************************************************************************/ 2405/*****************************************************************************/
1506/* grp functions */ 2406/* grp functions */
1507 2407
2408void
1508void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2409eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1509{ 2410{
1510 grp->int2 = limit; 2411 grp->int2 = limit;
1511 grp->feed = feed; 2412 grp->feed = feed;
1512 2413
1513 grp_try_feed (grp); 2414 grp_try_feed (grp);
1514} 2415}
1515 2416
2417void
1516void eio_grp_limit (eio_req *grp, int limit) 2418eio_grp_limit (eio_req *grp, int limit)
1517{ 2419{
1518 grp->int2 = limit; 2420 grp->int2 = limit;
1519 2421
1520 grp_try_feed (grp); 2422 grp_try_feed (grp);
1521} 2423}
1522 2424
2425void
1523void eio_grp_add (eio_req *grp, eio_req *req) 2426eio_grp_add (eio_req *grp, eio_req *req)
1524{ 2427{
1525 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2428 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1526 2429
1527 grp->flags |= EIO_FLAG_GROUPADD; 2430 grp->flags |= EIO_FLAG_GROUPADD;
1528 2431
1539} 2442}
1540 2443
1541/*****************************************************************************/ 2444/*****************************************************************************/
1542/* misc garbage */ 2445/* misc garbage */
1543 2446
2447eio_ssize_t
1544ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2448eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1545{ 2449{
1546 etp_worker wrk; 2450 etp_worker wrk;
2451 eio_ssize_t ret;
1547 2452
1548 wrk.dbuf = 0; 2453 wrk.dbuf = 0;
1549 2454
1550 eio__sendfile (ofd, ifd, offset, count, &wrk); 2455 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1551 2456
1552 if (wrk.dbuf) 2457 if (wrk.dbuf)
1553 free (wrk.dbuf); 2458 free (wrk.dbuf);
1554}
1555 2459
2460 return ret;
2461}
2462

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines