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

Comparing libeio/eio.c (file contents):
Revision 1.25 by root, Tue Sep 30 17:23:24 2008 UTC vs.
Revision 1.93 by root, Mon Jul 18 01:27:03 2011 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008 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;
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
213} etp_worker; 321} etp_worker;
214 322
215static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 323static etp_worker wrk_first; /* NOT etp */
216 324
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 int ecb_cold
459etp_init (void (*want_poll)(void), void (*done_poll)(void))
332{ 460{
333 X_LOCK (wrklock); 461 X_MUTEX_CREATE (wrklock);
334 X_LOCK (reqlock); 462 X_MUTEX_CREATE (reslock);
335 X_LOCK (reslock); 463 X_MUTEX_CREATE (reqlock);
336#if !HAVE_PREADWRITE 464 X_COND_CREATE (reqwait);
337 X_LOCK (preadwritelock);
338#endif
339}
340 465
341static void etp_atfork_parent (void) 466 reqq_init (&req_queue);
342{ 467 reqq_init (&res_queue);
343#if !HAVE_PREADWRITE
344 X_UNLOCK (preadwritelock);
345#endif
346 X_UNLOCK (reslock);
347 X_UNLOCK (reqlock);
348 X_UNLOCK (wrklock);
349}
350 468
351static void etp_atfork_child (void) 469 wrk_first.next =
352{ 470 wrk_first.prev = &wrk_first;
353 ETP_REQ *prv;
354
355 while ((prv = reqq_shift (&req_queue)))
356 ETP_DESTROY (prv);
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 }
371 471
372 started = 0; 472 started = 0;
373 idle = 0; 473 idle = 0;
374 nreqs = 0; 474 nreqs = 0;
375 nready = 0; 475 nready = 0;
376 npending = 0; 476 npending = 0;
377 477
378 etp_atfork_parent ();
379}
380
381static void
382etp_once_init (void)
383{
384 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
385}
386
387static int
388etp_init (void (*want_poll)(void), void (*done_poll)(void))
389{
390 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
391
392 pthread_once (&doinit, etp_once_init);
393
394 want_poll_cb = want_poll; 478 want_poll_cb = want_poll;
395 done_poll_cb = done_poll; 479 done_poll_cb = done_poll;
396 480
397 return 0; 481 return 0;
398} 482}
399 483
400X_THREAD_PROC (etp_proc); 484X_THREAD_PROC (etp_proc);
401 485
486static void ecb_cold
402static void etp_start_thread (void) 487etp_start_thread (void)
403{ 488{
404 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 489 etp_worker *wrk = calloc (1, sizeof (etp_worker));
405 490
406 /*TODO*/ 491 /*TODO*/
407 assert (("unable to allocate worker thread data", wrk)); 492 assert (("unable to allocate worker thread data", wrk));
420 free (wrk); 505 free (wrk);
421 506
422 X_UNLOCK (wrklock); 507 X_UNLOCK (wrklock);
423} 508}
424 509
510static void
425static void etp_maybe_start_thread (void) 511etp_maybe_start_thread (void)
426{ 512{
427 if (expect_true (etp_nthreads () >= wanted)) 513 if (ecb_expect_true (etp_nthreads () >= wanted))
428 return; 514 return;
429 515
430 /* todo: maybe use idle here, but might be less exact */ 516 /* todo: maybe use idle here, but might be less exact */
431 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 517 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
432 return; 518 return;
433 519
434 etp_start_thread (); 520 etp_start_thread ();
435} 521}
436 522
523static void ecb_cold
437static void etp_end_thread (void) 524etp_end_thread (void)
438{ 525{
439 eio_req *req = calloc (1, sizeof (eio_req)); 526 eio_req *req = calloc (1, sizeof (eio_req));
440 527
441 req->type = -1; 528 req->type = -1;
442 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 529 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
449 X_LOCK (wrklock); 536 X_LOCK (wrklock);
450 --started; 537 --started;
451 X_UNLOCK (wrklock); 538 X_UNLOCK (wrklock);
452} 539}
453 540
454static int etp_poll (void) 541static int
542etp_poll (void)
455{ 543{
456 unsigned int maxreqs; 544 unsigned int maxreqs;
457 unsigned int maxtime; 545 unsigned int maxtime;
458 struct timeval tv_start, tv_now; 546 struct timeval tv_start, tv_now;
459 547
489 577
490 X_LOCK (reqlock); 578 X_LOCK (reqlock);
491 --nreqs; 579 --nreqs;
492 X_UNLOCK (reqlock); 580 X_UNLOCK (reqlock);
493 581
494 if (expect_false (req->type == EIO_GROUP && req->size)) 582 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
495 { 583 {
496 req->int1 = 1; /* mark request as delayed */ 584 req->int1 = 1; /* mark request as delayed */
497 continue; 585 continue;
498 } 586 }
499 else 587 else
500 { 588 {
501 int res = ETP_FINISH (req); 589 int res = ETP_FINISH (req);
502 if (expect_false (res)) 590 if (ecb_expect_false (res))
503 return res; 591 return res;
504 } 592 }
505 593
506 if (expect_false (maxreqs && !--maxreqs)) 594 if (ecb_expect_false (maxreqs && !--maxreqs))
507 break; 595 break;
508 596
509 if (maxtime) 597 if (maxtime)
510 { 598 {
511 gettimeofday (&tv_now, 0); 599 gettimeofday (&tv_now, 0);
517 605
518 errno = EAGAIN; 606 errno = EAGAIN;
519 return -1; 607 return -1;
520} 608}
521 609
610static void
522static void etp_cancel (ETP_REQ *req) 611etp_cancel (ETP_REQ *req)
523{ 612{
524 X_LOCK (wrklock); 613 req->cancelled = 1;
525 req->flags |= EIO_FLAG_CANCELLED;
526 X_UNLOCK (wrklock);
527 614
528 eio_grp_cancel (req); 615 eio_grp_cancel (req);
529} 616}
530 617
618static void
531static void etp_submit (ETP_REQ *req) 619etp_submit (ETP_REQ *req)
532{ 620{
533 req->pri -= ETP_PRI_MIN; 621 req->pri -= ETP_PRI_MIN;
534 622
535 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 623 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
536 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 624 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
537 625
538 if (expect_false (req->type == EIO_GROUP)) 626 if (ecb_expect_false (req->type == EIO_GROUP))
539 { 627 {
540 /* I hope this is worth it :/ */ 628 /* I hope this is worth it :/ */
541 X_LOCK (reqlock); 629 X_LOCK (reqlock);
542 ++nreqs; 630 ++nreqs;
543 X_UNLOCK (reqlock); 631 X_UNLOCK (reqlock);
562 650
563 etp_maybe_start_thread (); 651 etp_maybe_start_thread ();
564 } 652 }
565} 653}
566 654
655static void ecb_cold
567static void etp_set_max_poll_time (double nseconds) 656etp_set_max_poll_time (double nseconds)
568{ 657{
569 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 658 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
570 max_poll_time = nseconds; 659 max_poll_time = nseconds * EIO_TICKS;
571 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 660 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
572} 661}
573 662
663static void ecb_cold
574static void etp_set_max_poll_reqs (unsigned int maxreqs) 664etp_set_max_poll_reqs (unsigned int maxreqs)
575{ 665{
576 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 666 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
577 max_poll_reqs = maxreqs; 667 max_poll_reqs = maxreqs;
578 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 668 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
579} 669}
580 670
671static void ecb_cold
581static void etp_set_max_idle (unsigned int nthreads) 672etp_set_max_idle (unsigned int nthreads)
582{ 673{
583 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 674 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
584 max_idle = nthreads <= 0 ? 1 : nthreads; 675 max_idle = nthreads;
585 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 676 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
586} 677}
587 678
679static void ecb_cold
680etp_set_idle_timeout (unsigned int seconds)
681{
682 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
683 idle_timeout = seconds;
684 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
685}
686
687static void ecb_cold
588static void etp_set_min_parallel (unsigned int nthreads) 688etp_set_min_parallel (unsigned int nthreads)
589{ 689{
590 if (wanted < nthreads) 690 if (wanted < nthreads)
591 wanted = nthreads; 691 wanted = nthreads;
592} 692}
593 693
694static void ecb_cold
594static void etp_set_max_parallel (unsigned int nthreads) 695etp_set_max_parallel (unsigned int nthreads)
595{ 696{
596 if (wanted > nthreads) 697 if (wanted > nthreads)
597 wanted = nthreads; 698 wanted = nthreads;
598 699
599 while (started > wanted) 700 while (started > wanted)
600 etp_end_thread (); 701 etp_end_thread ();
601} 702}
602 703
603/*****************************************************************************/ 704/*****************************************************************************/
604 705
706static void
605static void grp_try_feed (eio_req *grp) 707grp_try_feed (eio_req *grp)
606{ 708{
607 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 709 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
608 { 710 {
609 int old_len = grp->size; 711 grp->flags &= ~EIO_FLAG_GROUPADD;
610 712
611 EIO_FEED (grp); 713 EIO_FEED (grp);
612 714
613 /* stop if no progress has been made */ 715 /* stop if no progress has been made */
614 if (old_len == grp->size) 716 if (!(grp->flags & EIO_FLAG_GROUPADD))
615 { 717 {
616 grp->feed = 0; 718 grp->feed = 0;
617 break; 719 break;
618 } 720 }
619 } 721 }
620} 722}
621 723
724static int
622static int grp_dec (eio_req *grp) 725grp_dec (eio_req *grp)
623{ 726{
624 --grp->size; 727 --grp->size;
625 728
626 /* call feeder, if applicable */ 729 /* call feeder, if applicable */
627 grp_try_feed (grp); 730 grp_try_feed (grp);
631 return eio_finish (grp); 734 return eio_finish (grp);
632 else 735 else
633 return 0; 736 return 0;
634} 737}
635 738
739static void
636void eio_destroy (eio_req *req) 740eio_destroy (eio_req *req)
637{ 741{
638 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 742 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
639 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 743 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
640 744
641 EIO_DESTROY (req); 745 EIO_DESTROY (req);
642} 746}
643 747
748static int
644static int eio_finish (eio_req *req) 749eio_finish (eio_req *req)
645{ 750{
646 int res = EIO_FINISH (req); 751 int res = EIO_FINISH (req);
647 752
648 if (req->grp) 753 if (req->grp)
649 { 754 {
657 if (grp->grp_first == req) 762 if (grp->grp_first == req)
658 grp->grp_first = req->grp_next; 763 grp->grp_first = req->grp_next;
659 764
660 res2 = grp_dec (grp); 765 res2 = grp_dec (grp);
661 766
662 if (!res && res2) 767 if (!res)
663 res = res2; 768 res = res2;
664 } 769 }
665 770
666 eio_destroy (req); 771 eio_destroy (req);
667 772
668 return res; 773 return res;
669} 774}
670 775
776void
671void eio_grp_cancel (eio_req *grp) 777eio_grp_cancel (eio_req *grp)
672{ 778{
673 for (grp = grp->grp_first; grp; grp = grp->grp_next) 779 for (grp = grp->grp_first; grp; grp = grp->grp_next)
674 eio_cancel (grp); 780 eio_cancel (grp);
675} 781}
676 782
783void
677void eio_cancel (eio_req *req) 784eio_cancel (eio_req *req)
678{ 785{
679 etp_cancel (req); 786 etp_cancel (req);
680} 787}
681 788
789void
682void eio_submit (eio_req *req) 790eio_submit (eio_req *req)
683{ 791{
684 etp_submit (req); 792 etp_submit (req);
685} 793}
686 794
687unsigned int eio_nreqs (void) 795unsigned int
796eio_nreqs (void)
688{ 797{
689 return etp_nreqs (); 798 return etp_nreqs ();
690} 799}
691 800
692unsigned int eio_nready (void) 801unsigned int
802eio_nready (void)
693{ 803{
694 return etp_nready (); 804 return etp_nready ();
695} 805}
696 806
697unsigned int eio_npending (void) 807unsigned int
808eio_npending (void)
698{ 809{
699 return etp_npending (); 810 return etp_npending ();
700} 811}
701 812
702unsigned int eio_nthreads (void) 813unsigned int ecb_cold
814eio_nthreads (void)
703{ 815{
704 return etp_nthreads (); 816 return etp_nthreads ();
705} 817}
706 818
819void ecb_cold
707void eio_set_max_poll_time (double nseconds) 820eio_set_max_poll_time (double nseconds)
708{ 821{
709 etp_set_max_poll_time (nseconds); 822 etp_set_max_poll_time (nseconds);
710} 823}
711 824
825void ecb_cold
712void eio_set_max_poll_reqs (unsigned int maxreqs) 826eio_set_max_poll_reqs (unsigned int maxreqs)
713{ 827{
714 etp_set_max_poll_reqs (maxreqs); 828 etp_set_max_poll_reqs (maxreqs);
715} 829}
716 830
831void ecb_cold
717void eio_set_max_idle (unsigned int nthreads) 832eio_set_max_idle (unsigned int nthreads)
718{ 833{
719 etp_set_max_idle (nthreads); 834 etp_set_max_idle (nthreads);
720} 835}
721 836
837void ecb_cold
838eio_set_idle_timeout (unsigned int seconds)
839{
840 etp_set_idle_timeout (seconds);
841}
842
843void ecb_cold
722void eio_set_min_parallel (unsigned int nthreads) 844eio_set_min_parallel (unsigned int nthreads)
723{ 845{
724 etp_set_min_parallel (nthreads); 846 etp_set_min_parallel (nthreads);
725} 847}
726 848
849void ecb_cold
727void eio_set_max_parallel (unsigned int nthreads) 850eio_set_max_parallel (unsigned int nthreads)
728{ 851{
729 etp_set_max_parallel (nthreads); 852 etp_set_max_parallel (nthreads);
730} 853}
731 854
732int eio_poll (void) 855int eio_poll (void)
736 859
737/*****************************************************************************/ 860/*****************************************************************************/
738/* work around various missing functions */ 861/* work around various missing functions */
739 862
740#if !HAVE_PREADWRITE 863#if !HAVE_PREADWRITE
864# undef pread
865# undef pwrite
741# define pread eio__pread 866# define pread eio__pread
742# define pwrite eio__pwrite 867# define pwrite eio__pwrite
743 868
744static ssize_t 869static eio_ssize_t
745eio__pread (int fd, void *buf, size_t count, off_t offset) 870eio__pread (int fd, void *buf, size_t count, off_t offset)
746{ 871{
747 ssize_t res; 872 eio_ssize_t res;
748 off_t ooffset; 873 off_t ooffset;
749 874
750 X_LOCK (preadwritelock); 875 X_LOCK (preadwritelock);
751 ooffset = lseek (fd, 0, SEEK_CUR); 876 ooffset = lseek (fd, 0, SEEK_CUR);
752 lseek (fd, offset, SEEK_SET); 877 lseek (fd, offset, SEEK_SET);
755 X_UNLOCK (preadwritelock); 880 X_UNLOCK (preadwritelock);
756 881
757 return res; 882 return res;
758} 883}
759 884
760static ssize_t 885static eio_ssize_t
761eio__pwrite (int fd, void *buf, size_t count, off_t offset) 886eio__pwrite (int fd, void *buf, size_t count, off_t offset)
762{ 887{
763 ssize_t res; 888 eio_ssize_t res;
764 off_t ooffset; 889 off_t ooffset;
765 890
766 X_LOCK (preadwritelock); 891 X_LOCK (preadwritelock);
767 ooffset = lseek (fd, 0, SEEK_CUR); 892 ooffset = lseek (fd, 0, SEEK_CUR);
768 lseek (fd, offset, SEEK_SET); 893 lseek (fd, offset, SEEK_SET);
769 res = write (fd, buf, count); 894 res = write (fd, buf, count);
770 lseek (fd, offset, SEEK_SET); 895 lseek (fd, ooffset, SEEK_SET);
771 X_UNLOCK (preadwritelock); 896 X_UNLOCK (preadwritelock);
772 897
773 return res; 898 return res;
774} 899}
775#endif 900#endif
776 901
777#ifndef HAVE_FUTIMES 902#ifndef HAVE_UTIMES
778 903
904# undef utimes
779# define utimes(path,times) eio__utimes (path, times) 905# define utimes(path,times) eio__utimes (path, times)
780# define futimes(fd,times) eio__futimes (fd, times)
781 906
782static int 907static int
783eio__utimes (const char *filename, const struct timeval times[2]) 908eio__utimes (const char *filename, const struct timeval times[2])
784{ 909{
785 if (times) 910 if (times)
793 } 918 }
794 else 919 else
795 return utime (filename, 0); 920 return utime (filename, 0);
796} 921}
797 922
923#endif
924
925#ifndef HAVE_FUTIMES
926
927# undef futimes
928# define futimes(fd,times) eio__futimes (fd, times)
929
930static int
798static int eio__futimes (int fd, const struct timeval tv[2]) 931eio__futimes (int fd, const struct timeval tv[2])
799{ 932{
800 errno = ENOSYS; 933 errno = ENOSYS;
801 return -1; 934 return -1;
802} 935}
803 936
804#endif 937#endif
805 938
806#if !HAVE_FDATASYNC 939#if !HAVE_FDATASYNC
940# undef fdatasync
807# define fdatasync fsync 941# define fdatasync(fd) fsync (fd)
808#endif 942#endif
943
944/* sync_file_range always needs emulation */
945static int
946eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
947{
948#if HAVE_SYNC_FILE_RANGE
949 int res;
950
951 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
952 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
953 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
954 {
955 flags = 0
956 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
957 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
958 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
959 }
960
961 res = sync_file_range (fd, offset, nbytes, flags);
962
963 if (!res || errno != ENOSYS)
964 return res;
965#endif
966
967 /* even though we could play tricks with the flags, it's better to always
968 * call fdatasync, as that matches the expectation of its users best */
969 return fdatasync (fd);
970}
971
972static int
973eio__fallocate (int fd, int mode, off_t offset, size_t len)
974{
975#if HAVE_FALLOCATE
976 return fallocate (fd, mode, offset, len);
977#else
978 errno = ENOSYS;
979 return -1;
980#endif
981}
809 982
810#if !HAVE_READAHEAD 983#if !HAVE_READAHEAD
984# undef readahead
811# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 985# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
812 986
813static ssize_t 987static eio_ssize_t
814eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 988eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
815{ 989{
816 size_t todo = count; 990 size_t todo = count;
817 dBUF; 991 dBUF;
818 992
830} 1004}
831 1005
832#endif 1006#endif
833 1007
834/* sendfile always needs emulation */ 1008/* sendfile always needs emulation */
835static ssize_t 1009static eio_ssize_t
836eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1010eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
837{ 1011{
1012 eio_ssize_t written = 0;
838 ssize_t res; 1013 eio_ssize_t res;
839 1014
840 if (!count) 1015 if (!count)
841 return 0; 1016 return 0;
842 1017
1018 for (;;)
1019 {
1020#ifdef __APPLE__
1021# undef HAVE_SENDFILE /* broken, as everything on os x */
1022#endif
843#if HAVE_SENDFILE 1023#if HAVE_SENDFILE
844# if __linux 1024# if __linux
1025 off_t soffset = offset;
845 res = sendfile (ofd, ifd, &offset, count); 1026 res = sendfile (ofd, ifd, &soffset, count);
846 1027
847# elif __freebsd 1028# elif __FreeBSD__
848 /* 1029 /*
849 * Of course, the freebsd sendfile is a dire hack with no thoughts 1030 * Of course, the freebsd sendfile is a dire hack with no thoughts
850 * wasted on making it similar to other I/O functions. 1031 * wasted on making it similar to other I/O functions.
851 */ 1032 */
852 {
853 off_t sbytes; 1033 off_t sbytes;
854 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1034 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
855 1035
856 if (res < 0 && sbytes) 1036 #if 0 /* according to the manpage, this is correct, but broken behaviour */
857 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 1037 /* freebsd' sendfile will return 0 on success */
1038 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
1039 /* not on e.g. EIO or EPIPE - sounds broken */
1040 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
858 res = sbytes; 1041 res = sbytes;
859 } 1042 #endif
1043
1044 /* according to source inspection, this is correct, and useful behaviour */
1045 if (sbytes)
1046 res = sbytes;
1047
1048# elif defined (__APPLE__)
1049 off_t sbytes = count;
1050 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1051
1052 /* according to the manpage, sbytes is always valid */
1053 if (sbytes)
1054 res = sbytes;
860 1055
861# elif __hpux 1056# elif __hpux
862 res = sendfile (ofd, ifd, offset, count, 0, 0); 1057 res = sendfile (ofd, ifd, offset, count, 0, 0);
863 1058
864# elif __solaris 1059# elif __solaris
865 {
866 struct sendfilevec vec; 1060 struct sendfilevec vec;
867 size_t sbytes; 1061 size_t sbytes;
868 1062
869 vec.sfv_fd = ifd; 1063 vec.sfv_fd = ifd;
870 vec.sfv_flag = 0; 1064 vec.sfv_flag = 0;
871 vec.sfv_off = offset; 1065 vec.sfv_off = offset;
872 vec.sfv_len = count; 1066 vec.sfv_len = count;
873 1067
874 res = sendfilev (ofd, &vec, 1, &sbytes); 1068 res = sendfilev (ofd, &vec, 1, &sbytes);
875 1069
876 if (res < 0 && sbytes) 1070 if (res < 0 && sbytes)
877 res = sbytes; 1071 res = sbytes;
878 }
879 1072
880# endif 1073# endif
1074
1075#elif defined (_WIN32) && 0
1076 /* does not work, just for documentation of what would need to be done */
1077 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1078 /* libeio guarantees that the file offset does not change, and windows */
1079 /* has no way to get an independent handle to the same file description */
1080 HANDLE h = TO_SOCKET (ifd);
1081 SetFilePointer (h, offset, 0, FILE_BEGIN);
1082 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1083
881#else 1084#else
882 res = -1; 1085 res = -1;
883 errno = ENOSYS; 1086 errno = ENOSYS;
884#endif 1087#endif
885 1088
1089 /* we assume sendfile can copy at least 128mb in one go */
1090 if (res <= 128 * 1024 * 1024)
1091 {
1092 if (res > 0)
1093 written += res;
1094
1095 if (written)
1096 return written;
1097
1098 break;
1099 }
1100 else
1101 {
1102 /* if we requested more, then probably the kernel was lazy */
1103 written += res;
1104 offset += res;
1105 count -= res;
1106
1107 if (!count)
1108 return written;
1109 }
1110 }
1111
886 if (res < 0 1112 if (res < 0
887 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1113 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1114 /* BSDs */
1115#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1116 || errno == ENOTSUP
1117#endif
1118#ifdef EOPNOTSUPP /* windows */
1119 || errno == EOPNOTSUPP /* BSDs */
1120#endif
888#if __solaris 1121#if __solaris
889 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1122 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
890#endif 1123#endif
891 ) 1124 )
892 ) 1125 )
896 1129
897 res = 0; 1130 res = 0;
898 1131
899 while (count) 1132 while (count)
900 { 1133 {
901 ssize_t cnt; 1134 eio_ssize_t cnt;
902 1135
903 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1136 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
904 1137
905 if (cnt <= 0) 1138 if (cnt <= 0)
906 { 1139 {
923 } 1156 }
924 1157
925 return res; 1158 return res;
926} 1159}
927 1160
1161#ifdef PAGESIZE
1162# define eio_pagesize() PAGESIZE
1163#else
1164static intptr_t
1165eio_pagesize (void)
1166{
1167 static intptr_t page;
1168
1169 if (!page)
1170 page = sysconf (_SC_PAGESIZE);
1171
1172 return page;
1173}
1174#endif
1175
1176static void
1177eio_page_align (void **addr, size_t *length)
1178{
1179 intptr_t mask = eio_pagesize () - 1;
1180
1181 /* round down addr */
1182 intptr_t adj = mask & (intptr_t)*addr;
1183
1184 *addr = (void *)((intptr_t)*addr - adj);
1185 *length += adj;
1186
1187 /* round up length */
1188 *length = (*length + mask) & ~mask;
1189}
1190
1191#if !_POSIX_MEMLOCK
1192# define eio__mlockall(a) EIO_ENOSYS ()
1193#else
1194
1195static int
1196eio__mlockall (int flags)
1197{
1198 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1199 extern int mallopt (int, int);
1200 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1201 #endif
1202
1203 if (EIO_MCL_CURRENT != MCL_CURRENT
1204 || EIO_MCL_FUTURE != MCL_FUTURE)
1205 {
1206 flags = 0
1207 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1208 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1209 }
1210
1211 return mlockall (flags);
1212}
1213#endif
1214
1215#if !_POSIX_MEMLOCK_RANGE
1216# define eio__mlock(a,b) EIO_ENOSYS ()
1217#else
1218
1219static int
1220eio__mlock (void *addr, size_t length)
1221{
1222 eio_page_align (&addr, &length);
1223
1224 return mlock (addr, length);
1225}
1226
1227#endif
1228
1229#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1230# define eio__msync(a,b,c) EIO_ENOSYS ()
1231#else
1232
1233static int
1234eio__msync (void *mem, size_t len, int flags)
1235{
1236 eio_page_align (&mem, &len);
1237
1238 if (EIO_MS_ASYNC != MS_SYNC
1239 || EIO_MS_INVALIDATE != MS_INVALIDATE
1240 || EIO_MS_SYNC != MS_SYNC)
1241 {
1242 flags = 0
1243 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1244 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1245 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1246 }
1247
1248 return msync (mem, len, flags);
1249}
1250
1251#endif
1252
1253static int
1254eio__mtouch (eio_req *req)
1255{
1256 void *mem = req->ptr2;
1257 size_t len = req->size;
1258 int flags = req->int1;
1259
1260 eio_page_align (&mem, &len);
1261
1262 {
1263 intptr_t addr = (intptr_t)mem;
1264 intptr_t end = addr + len;
1265 intptr_t page = eio_pagesize ();
1266
1267 if (addr < end)
1268 if (flags & EIO_MT_MODIFY) /* modify */
1269 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1270 else
1271 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1272 }
1273
1274 return 0;
1275}
1276
1277/*****************************************************************************/
1278/* requests implemented outside eio_execute, because they are so large */
1279
1280static void
1281eio__realpath (eio_req *req, etp_worker *self)
1282{
1283 char *rel = req->ptr1;
1284 char *res;
1285 char *tmp1, *tmp2;
1286#if SYMLOOP_MAX > 32
1287 int symlinks = SYMLOOP_MAX;
1288#else
1289 int symlinks = 32;
1290#endif
1291
1292 req->result = -1;
1293
1294 errno = EINVAL;
1295 if (!rel)
1296 return;
1297
1298 errno = ENOENT;
1299 if (!*rel)
1300 return;
1301
1302 if (!req->ptr2)
1303 {
1304 X_LOCK (wrklock);
1305 req->flags |= EIO_FLAG_PTR2_FREE;
1306 X_UNLOCK (wrklock);
1307 req->ptr2 = malloc (PATH_MAX * 3);
1308
1309 errno = ENOMEM;
1310 if (!req->ptr2)
1311 return;
1312 }
1313
1314 res = req->ptr2;
1315 tmp1 = res + PATH_MAX;
1316 tmp2 = tmp1 + PATH_MAX;
1317
1318#if 0 /* disabled, the musl way to do things is just too racy */
1319#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1320 /* on linux we may be able to ask the kernel */
1321 {
1322 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1323
1324 if (fd >= 0)
1325 {
1326 sprintf (tmp1, "/proc/self/fd/%d", fd);
1327 req->result = readlink (tmp1, res, PATH_MAX);
1328 close (fd);
1329
1330 /* here we should probably stat the open file and the disk file, to make sure they still match */
1331
1332 if (req->result > 0)
1333 goto done;
1334 }
1335 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1336 return;
1337 }
1338#endif
1339#endif
1340
1341 if (*rel != '/')
1342 {
1343 if (!getcwd (res, PATH_MAX))
1344 return;
1345
1346 if (res [1]) /* only use if not / */
1347 res += strlen (res);
1348 }
1349
1350 while (*rel)
1351 {
1352 eio_ssize_t len, linklen;
1353 char *beg = rel;
1354
1355 while (*rel && *rel != '/')
1356 ++rel;
1357
1358 len = rel - beg;
1359
1360 if (!len) /* skip slashes */
1361 {
1362 ++rel;
1363 continue;
1364 }
1365
1366 if (beg [0] == '.')
1367 {
1368 if (len == 1)
1369 continue; /* . - nop */
1370
1371 if (beg [1] == '.' && len == 2)
1372 {
1373 /* .. - back up one component, if possible */
1374
1375 while (res != req->ptr2)
1376 if (*--res == '/')
1377 break;
1378
1379 continue;
1380 }
1381 }
1382
1383 errno = ENAMETOOLONG;
1384 if (res + 1 + len + 1 >= tmp1)
1385 return;
1386
1387 /* copy one component */
1388 *res = '/';
1389 memcpy (res + 1, beg, len);
1390
1391 /* zero-terminate, for readlink */
1392 res [len + 1] = 0;
1393
1394 /* now check if it's a symlink */
1395 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1396
1397 if (linklen < 0)
1398 {
1399 if (errno != EINVAL)
1400 return;
1401
1402 /* it's a normal directory. hopefully */
1403 res += len + 1;
1404 }
1405 else
1406 {
1407 /* yay, it was a symlink - build new path in tmp2 */
1408 int rellen = strlen (rel);
1409
1410 errno = ENAMETOOLONG;
1411 if (linklen + 1 + rellen >= PATH_MAX)
1412 return;
1413
1414 errno = ELOOP;
1415 if (!--symlinks)
1416 return;
1417
1418 if (*tmp1 == '/')
1419 res = req->ptr2; /* symlink resolves to an absolute path */
1420
1421 /* we need to be careful, as rel might point into tmp2 already */
1422 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1423 tmp2 [linklen] = '/';
1424 memcpy (tmp2, tmp1, linklen);
1425
1426 rel = tmp2;
1427 }
1428 }
1429
1430 /* special case for the lone root path */
1431 if (res == req->ptr2)
1432 *res++ = '/';
1433
1434 req->result = res - (char *)req->ptr2;
1435
1436done:
1437 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1438}
1439
1440static signed char
1441eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1442{
1443 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1444 : a->inode < b->inode ? -1
1445 : a->inode > b->inode ? 1
1446 : 0;
1447}
1448
1449#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1450
1451#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1452#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1453
1454static void
1455eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1456{
1457 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1458 unsigned char *bit = bits;
1459
1460 assert (CHAR_BIT == 8);
1461 assert (sizeof (eio_dirent) * 8 < 256);
1462 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1463 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1464
1465 if (size <= EIO_SORT_FAST)
1466 return;
1467
1468 /* first prepare an array of bits to test in our radix sort */
1469 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1470 /* inode_bits must contain all inodes ORed together */
1471 /* which is used to skip bits that are 0 everywhere, which is very common */
1472 {
1473 eio_ino_t endianness;
1474 int i, j;
1475
1476 /* we store the byte offset of byte n into byte n of "endianness" */
1477 for (i = 0; i < sizeof (eio_ino_t); ++i)
1478 ((unsigned char *)&endianness)[i] = i;
1479
1480 *bit++ = 0;
1481
1482 for (i = 0; i < sizeof (eio_ino_t); ++i)
1483 {
1484 /* shifting off the byte offsets out of "endianness" */
1485 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1486 endianness >>= 8;
1487
1488 for (j = 0; j < 8; ++j)
1489 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1490 *bit++ = offs + j;
1491 }
1492
1493 for (j = 0; j < 8; ++j)
1494 if (score_bits & (1 << j))
1495 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1496 }
1497
1498 /* now actually do the sorting (a variant of MSD radix sort) */
1499 {
1500 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1501 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1502 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1503 int stk_idx = 0;
1504
1505 base_stk [stk_idx] = dents;
1506 end_stk [stk_idx] = dents + size;
1507 bit_stk [stk_idx] = bit - 1;
1508
1509 do
1510 {
1511 base = base_stk [stk_idx];
1512 end = end_stk [stk_idx];
1513 bit = bit_stk [stk_idx];
1514
1515 for (;;)
1516 {
1517 unsigned char O = *bit >> 3;
1518 unsigned char M = 1 << (*bit & 7);
1519
1520 eio_dirent *a = base;
1521 eio_dirent *b = end;
1522
1523 if (b - a < EIO_SORT_CUTOFF)
1524 break;
1525
1526 /* now bit-partition the array on the bit */
1527 /* this ugly asymmetric loop seems to perform much better than typical */
1528 /* partition algos found in the literature */
1529 do
1530 if (!(((unsigned char *)a)[O] & M))
1531 ++a;
1532 else if (!(((unsigned char *)--b)[O] & M))
1533 {
1534 eio_dirent tmp = *a; *a = *b; *b = tmp;
1535 ++a;
1536 }
1537 while (b > a);
1538
1539 /* next bit, or stop, if no bits left in this path */
1540 if (!*--bit)
1541 break;
1542
1543 base_stk [stk_idx] = a;
1544 end_stk [stk_idx] = end;
1545 bit_stk [stk_idx] = bit;
1546 ++stk_idx;
1547
1548 end = a;
1549 }
1550 }
1551 while (stk_idx--);
1552 }
1553}
1554
1555static void
1556eio_dent_insertion_sort (eio_dirent *dents, int size)
1557{
1558 /* first move the smallest element to the front, to act as a sentinel */
1559 {
1560 int i;
1561 eio_dirent *min = dents;
1562
1563 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1564 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1565 if (EIO_DENT_CMP (dents [i], <, *min))
1566 min = &dents [i];
1567
1568 /* swap elements 0 and j (minimum) */
1569 {
1570 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1571 }
1572 }
1573
1574 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1575 {
1576 eio_dirent *i, *j;
1577
1578 for (i = dents + 1; i < dents + size; ++i)
1579 {
1580 eio_dirent value = *i;
1581
1582 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1583 j [1] = j [0];
1584
1585 j [1] = value;
1586 }
1587 }
1588}
1589
1590static void
1591eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1592{
1593 if (size <= 1)
1594 return; /* our insertion sort relies on size > 0 */
1595
1596 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1597 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1598 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1599
1600 /* use an insertion sort at the end, or for small arrays, */
1601 /* as insertion sort is more efficient for small partitions */
1602 eio_dent_insertion_sort (dents, size);
1603}
1604
928/* read a full directory */ 1605/* read a full directory */
929static void 1606static void
930eio__scandir (eio_req *req, etp_worker *self) 1607eio__scandir (eio_req *req, etp_worker *self)
931{ 1608{
1609 char *name, *names;
1610 int namesalloc = 4096 - sizeof (void *) * 4;
1611 int namesoffs = 0;
1612 int flags = req->int1;
1613 eio_dirent *dents = 0;
1614 int dentalloc = 128;
1615 int dentoffs = 0;
1616 eio_ino_t inode_bits = 0;
1617#ifdef _WIN32
1618 HANDLE dirp;
1619 WIN32_FIND_DATA entp;
1620#else
932 DIR *dirp; 1621 DIR *dirp;
933 EIO_STRUCT_DIRENT *entp; 1622 EIO_STRUCT_DIRENT *entp;
934 char *name, *names; 1623#endif
935 int memlen = 4096;
936 int memofs = 0;
937 int res = 0;
938 1624
939 X_LOCK (wrklock); 1625 req->result = -1;
940 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1626
1627 if (!(flags & EIO_READDIR_DENTS))
1628 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1629
1630#ifdef _WIN32
1631 {
1632 int len = strlen ((const char *)req->ptr1);
1633 char *path = malloc (MAX_PATH);
1634 const char *fmt;
1635
1636 if (!len)
1637 fmt = "./*";
1638 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1639 fmt = "%s*";
1640 else
1641 fmt = "%s/*";
1642
1643 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1644 dirp = FindFirstFile (path, &entp);
1645 free (path);
1646
1647 if (dirp == INVALID_HANDLE_VALUE)
1648 {
1649 dirp = 0;
1650
1651 switch (GetLastError ())
1652 {
1653 case ERROR_FILE_NOT_FOUND:
1654 req->result = 0;
1655 break;
1656
1657 case ERROR_INVALID_NAME:
1658 case ERROR_PATH_NOT_FOUND:
1659 case ERROR_NO_MORE_FILES:
1660 errno = ENOENT;
1661 break;
1662
1663 case ERROR_NOT_ENOUGH_MEMORY:
1664 errno = ENOMEM;
1665 break;
1666
1667 default:
1668 errno = EINVAL;
1669 break;
1670 }
1671 }
1672 }
1673#else
941 self->dirp = dirp = opendir (req->ptr1); 1674 dirp = opendir (req->ptr1);
1675#endif
1676
942 req->flags |= EIO_FLAG_PTR2_FREE; 1677 if (req->flags & EIO_FLAG_PTR1_FREE)
1678 free (req->ptr1);
1679
1680 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1681 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
943 req->ptr2 = names = malloc (memlen); 1682 req->ptr2 = names = malloc (namesalloc);
944 X_UNLOCK (wrklock);
945 1683
946 if (dirp && names) 1684 if (dirp && names && (!flags || dents))
947 for (;;) 1685 for (;;)
948 { 1686 {
1687 int done;
1688
1689#ifdef _WIN32
1690 done = !dirp;
1691#else
949 errno = 0; 1692 errno = 0;
950 entp = readdir (dirp); 1693 entp = readdir (dirp);
1694 done = !entp;
1695#endif
951 1696
952 if (!entp) 1697 if (done)
1698 {
1699#ifndef _WIN32
1700 int old_errno = errno;
1701 closedir (dirp);
1702 errno = old_errno;
1703
1704 if (errno)
1705 break;
1706#endif
1707
1708 /* sort etc. */
1709 req->int1 = flags;
1710 req->result = dentoffs;
1711
1712 if (flags & EIO_READDIR_STAT_ORDER)
1713 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1714 else if (flags & EIO_READDIR_DIRS_FIRST)
1715 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1716 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1717 else
1718 {
1719 /* in this case, all is known, and we just put dirs first and sort them */
1720 eio_dirent *oth = dents + dentoffs;
1721 eio_dirent *dir = dents;
1722
1723 /* now partition dirs to the front, and non-dirs to the back */
1724 /* by walking from both sides and swapping if necessary */
1725 while (oth > dir)
1726 {
1727 if (dir->type == EIO_DT_DIR)
1728 ++dir;
1729 else if ((--oth)->type == EIO_DT_DIR)
1730 {
1731 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1732
1733 ++dir;
1734 }
1735 }
1736
1737 /* now sort the dirs only (dirs all have the same score) */
1738 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1739 }
1740
953 break; 1741 break;
1742 }
954 1743
955 name = entp->d_name; 1744 /* now add the entry to our list(s) */
1745 name = D_NAME (entp);
956 1746
1747 /* skip . and .. entries */
957 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1748 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
958 { 1749 {
959 int len = strlen (name) + 1; 1750 int len = D_NAMLEN (entp) + 1;
960 1751
961 res++; 1752 while (ecb_expect_false (namesoffs + len > namesalloc))
962
963 while (memofs + len > memlen)
964 { 1753 {
965 memlen *= 2; 1754 namesalloc *= 2;
966 X_LOCK (wrklock);
967 req->ptr2 = names = realloc (names, memlen); 1755 req->ptr2 = names = realloc (names, namesalloc);
968 X_UNLOCK (wrklock);
969 1756
970 if (!names) 1757 if (!names)
971 break; 1758 break;
972 } 1759 }
973 1760
974 memcpy (names + memofs, name, len); 1761 memcpy (names + namesoffs, name, len);
1762
1763 if (dents)
1764 {
1765 struct eio_dirent *ent;
1766
1767 if (ecb_expect_false (dentoffs == dentalloc))
1768 {
1769 dentalloc *= 2;
1770 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1771
1772 if (!dents)
1773 break;
1774 }
1775
1776 ent = dents + dentoffs;
1777
1778 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1779 ent->namelen = len - 1;
1780 ent->inode = D_INO (entp);
1781
1782 inode_bits |= ent->inode;
1783
1784 switch (D_TYPE (entp))
1785 {
1786 default:
1787 ent->type = EIO_DT_UNKNOWN;
1788 flags |= EIO_READDIR_FOUND_UNKNOWN;
1789 break;
1790
1791 #ifdef DT_FIFO
1792 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1793 #endif
1794 #ifdef DT_CHR
1795 case DT_CHR: ent->type = EIO_DT_CHR; break;
1796 #endif
1797 #ifdef DT_MPC
1798 case DT_MPC: ent->type = EIO_DT_MPC; break;
1799 #endif
1800 #ifdef DT_DIR
1801 case DT_DIR: ent->type = EIO_DT_DIR; break;
1802 #endif
1803 #ifdef DT_NAM
1804 case DT_NAM: ent->type = EIO_DT_NAM; break;
1805 #endif
1806 #ifdef DT_BLK
1807 case DT_BLK: ent->type = EIO_DT_BLK; break;
1808 #endif
1809 #ifdef DT_MPB
1810 case DT_MPB: ent->type = EIO_DT_MPB; break;
1811 #endif
1812 #ifdef DT_REG
1813 case DT_REG: ent->type = EIO_DT_REG; break;
1814 #endif
1815 #ifdef DT_NWK
1816 case DT_NWK: ent->type = EIO_DT_NWK; break;
1817 #endif
1818 #ifdef DT_CMP
1819 case DT_CMP: ent->type = EIO_DT_CMP; break;
1820 #endif
1821 #ifdef DT_LNK
1822 case DT_LNK: ent->type = EIO_DT_LNK; break;
1823 #endif
1824 #ifdef DT_SOCK
1825 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1826 #endif
1827 #ifdef DT_DOOR
1828 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1829 #endif
1830 #ifdef DT_WHT
1831 case DT_WHT: ent->type = EIO_DT_WHT; break;
1832 #endif
1833 }
1834
1835 ent->score = 7;
1836
1837 if (flags & EIO_READDIR_DIRS_FIRST)
1838 {
1839 if (ent->type == EIO_DT_UNKNOWN)
1840 {
1841 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1842 ent->score = 1;
1843 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1844 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1845 }
1846 else if (ent->type == EIO_DT_DIR)
1847 ent->score = 0;
1848 }
1849 }
1850
975 memofs += len; 1851 namesoffs += len;
1852 ++dentoffs;
976 } 1853 }
1854
1855 if (EIO_CANCELLED (req))
1856 {
1857 errno = ECANCELED;
1858 break;
1859 }
1860
1861#ifdef _WIN32
1862 if (!FindNextFile (dirp, &entp))
1863 {
1864 FindClose (dirp);
1865 dirp = 0;
1866 }
1867#endif
977 } 1868 }
978
979 if (errno)
980 res = -1;
981
982 req->result = res;
983}
984
985#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
986# define msync(a,b,c) ENOSYS
987#endif
988
989int
990eio__mtouch (void *mem, size_t len, int flags)
991{
992 intptr_t addr = (intptr_t)mem;
993 intptr_t end = addr + len;
994#ifdef PAGESIZE
995 const intptr_t page = PAGESIZE;
996#else
997 static intptr_t page;
998
999 if (!page)
1000 page = sysconf (_SC_PAGESIZE);
1001#endif
1002
1003 addr &= ~(page - 1); /* assume page size is always a power of two */
1004
1005 if (addr < end)
1006 if (flags) /* modify */
1007 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1008 else
1009 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1010
1011 return 0;
1012} 1869}
1013 1870
1014/*****************************************************************************/ 1871/*****************************************************************************/
1015 1872
1016#define ALLOC(len) \ 1873#define ALLOC(len) \
1032{ 1889{
1033 ETP_REQ *req; 1890 ETP_REQ *req;
1034 struct timespec ts; 1891 struct timespec ts;
1035 etp_worker *self = (etp_worker *)thr_arg; 1892 etp_worker *self = (etp_worker *)thr_arg;
1036 1893
1037 /* try to distribute timeouts somewhat randomly */ 1894 /* try to distribute timeouts somewhat evenly */
1038 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1895 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1039 1896
1040 for (;;) 1897 for (;;)
1041 { 1898 {
1899 ts.tv_sec = 0;
1900
1042 X_LOCK (reqlock); 1901 X_LOCK (reqlock);
1043 1902
1044 for (;;) 1903 for (;;)
1045 { 1904 {
1046 self->req = req = reqq_shift (&req_queue); 1905 self->req = req = reqq_shift (&req_queue);
1047 1906
1048 if (req) 1907 if (req)
1049 break; 1908 break;
1050 1909
1910 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1911 {
1912 X_UNLOCK (reqlock);
1913 X_LOCK (wrklock);
1914 --started;
1915 X_UNLOCK (wrklock);
1916 goto quit;
1917 }
1918
1051 ++idle; 1919 ++idle;
1052 1920
1053 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1921 if (idle <= max_idle)
1054 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1922 /* we are allowed to idle, so do so without any timeout */
1923 X_COND_WAIT (reqwait, reqlock);
1924 else
1055 { 1925 {
1056 if (idle > max_idle) 1926 /* initialise timeout once */
1057 { 1927 if (!ts.tv_sec)
1058 --idle; 1928 ts.tv_sec = time (0) + idle_timeout;
1059 X_UNLOCK (reqlock);
1060 X_LOCK (wrklock);
1061 --started;
1062 X_UNLOCK (wrklock);
1063 goto quit;
1064 }
1065 1929
1066 /* we are allowed to idle, so do so without any timeout */
1067 X_COND_WAIT (reqwait, reqlock); 1930 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1931 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1068 } 1932 }
1069 1933
1070 --idle; 1934 --idle;
1071 } 1935 }
1072 1936
1075 X_UNLOCK (reqlock); 1939 X_UNLOCK (reqlock);
1076 1940
1077 if (req->type < 0) 1941 if (req->type < 0)
1078 goto quit; 1942 goto quit;
1079 1943
1080 if (!EIO_CANCELLED (req))
1081 ETP_EXECUTE (self, req); 1944 ETP_EXECUTE (self, req);
1082 1945
1083 X_LOCK (reslock); 1946 X_LOCK (reslock);
1084 1947
1085 ++npending; 1948 ++npending;
1086 1949
1101 return 0; 1964 return 0;
1102} 1965}
1103 1966
1104/*****************************************************************************/ 1967/*****************************************************************************/
1105 1968
1969int ecb_cold
1106int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1970eio_init (void (*want_poll)(void), void (*done_poll)(void))
1107{ 1971{
1972#if !HAVE_PREADWRITE
1973 X_MUTEX_CREATE (preadwritelock);
1974#endif
1975
1108 return etp_init (want_poll, done_poll); 1976 return etp_init (want_poll, done_poll);
1109} 1977}
1110 1978
1979ecb_inline void
1111static void eio_api_destroy (eio_req *req) 1980eio_api_destroy (eio_req *req)
1112{ 1981{
1113 free (req); 1982 free (req);
1114} 1983}
1115 1984
1116#define REQ(rtype) \ 1985#define REQ(rtype) \
1135 { \ 2004 { \
1136 eio_api_destroy (req); \ 2005 eio_api_destroy (req); \
1137 return 0; \ 2006 return 0; \
1138 } 2007 }
1139 2008
2009static void
1140static void eio_execute (etp_worker *self, eio_req *req) 2010eio_execute (etp_worker *self, eio_req *req)
1141{ 2011{
1142 errno = 0; 2012 if (ecb_expect_false (EIO_CANCELLED (req)))
2013 {
2014 req->result = -1;
2015 req->errorno = ECANCELED;
2016 return;
2017 }
1143 2018
1144 switch (req->type) 2019 switch (req->type)
1145 { 2020 {
1146 case EIO_READ: ALLOC (req->size); 2021 case EIO_READ: ALLOC (req->size);
1147 req->result = req->offs >= 0 2022 req->result = req->offs >= 0
1159 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2034 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1160 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2035 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1161 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2036 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1162 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2037 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1163 2038
2039 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2040 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2041 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2042 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2043
1164 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; 2044 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1165 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2045 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1166 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; 2046 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1167 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2047 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1168 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; 2048 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1175 case EIO_RMDIR: req->result = rmdir (req->ptr1); break; 2055 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1176 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; 2056 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1177 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2057 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1178 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2058 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1179 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2059 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1180 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; 2060 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1181 2061
2062 case EIO_REALPATH: eio__realpath (req, self); break;
2063
1182 case EIO_READLINK: ALLOC (NAME_MAX); 2064 case EIO_READLINK: ALLOC (PATH_MAX);
1183 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; 2065 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1184 2066
1185 case EIO_SYNC: req->result = 0; sync (); break; 2067 case EIO_SYNC: req->result = 0; sync (); break;
1186 case EIO_FSYNC: req->result = fsync (req->int1); break; 2068 case EIO_FSYNC: req->result = fsync (req->int1); break;
1187 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2069 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1188 case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; 2070 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2071 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1189 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2072 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2073 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2074 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2075 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1190 2076
1191 case EIO_READDIR: eio__scandir (req, self); break; 2077 case EIO_READDIR: eio__scandir (req, self); break;
1192 2078
1193 case EIO_BUSY: 2079 case EIO_BUSY:
1194#ifdef _WIN32 2080#ifdef _WIN32
1195 Sleep (req->nv1 * 1000.); 2081 Sleep (req->nv1 * 1e3);
1196#else 2082#else
1197 { 2083 {
1198 struct timeval tv; 2084 struct timeval tv;
1199 2085
1200 tv.tv_sec = req->nv1; 2086 tv.tv_sec = req->nv1;
1201 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; 2087 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1202 2088
1203 req->result = select (0, 0, 0, 0, &tv); 2089 req->result = select (0, 0, 0, 0, &tv);
1204 } 2090 }
1205#endif 2091#endif
1206 break; 2092 break;
1221 times = tv; 2107 times = tv;
1222 } 2108 }
1223 else 2109 else
1224 times = 0; 2110 times = 0;
1225 2111
1226
1227 req->result = req->type == EIO_FUTIME 2112 req->result = req->type == EIO_FUTIME
1228 ? futimes (req->int1, times) 2113 ? futimes (req->int1, times)
1229 : utimes (req->ptr1, times); 2114 : utimes (req->ptr1, times);
1230 } 2115 }
1231 break; 2116 break;
1236 case EIO_NOP: 2121 case EIO_NOP:
1237 req->result = 0; 2122 req->result = 0;
1238 break; 2123 break;
1239 2124
1240 case EIO_CUSTOM: 2125 case EIO_CUSTOM:
1241 ((void (*)(eio_req *))req->feed) (req); 2126 req->feed (req);
1242 break; 2127 break;
1243 2128
1244 default: 2129 default:
2130 errno = ENOSYS;
1245 req->result = -1; 2131 req->result = -1;
1246 break; 2132 break;
1247 } 2133 }
1248 2134
1249 req->errorno = errno; 2135 req->errorno = errno;
1279eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2165eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1280{ 2166{
1281 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2167 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1282} 2168}
1283 2169
2170eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2171{
2172 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2173}
2174
2175eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2176{
2177 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2178}
2179
2180eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2181{
2182 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2183}
2184
2185eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2186{
2187 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2188}
2189
1284eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2190eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1285{ 2191{
1286 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2192 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1287} 2193}
1288 2194
1309eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2215eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1310{ 2216{
1311 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2217 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1312} 2218}
1313 2219
2220eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2221{
2222 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2223}
2224
1314eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2225eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1315{ 2226{
1316 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2227 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1317} 2228}
1318 2229
1324eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2235eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1325{ 2236{
1326 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2237 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1327} 2238}
1328 2239
1329eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2240eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1330{ 2241{
1331 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2242 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1332} 2243}
1333 2244
1334eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2245eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1354eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2265eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1355{ 2266{
1356 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2267 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1357} 2268}
1358 2269
1359eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2270eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1360{ 2271{
1361 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2272 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1362} 2273}
1363 2274
1364eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2275eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1380eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2291eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1381{ 2292{
1382 return eio__1path (EIO_READLINK, path, pri, cb, data); 2293 return eio__1path (EIO_READLINK, path, pri, cb, data);
1383} 2294}
1384 2295
2296eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2297{
2298 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2299}
2300
1385eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2301eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1386{ 2302{
1387 return eio__1path (EIO_STAT, path, pri, cb, data); 2303 return eio__1path (EIO_STAT, path, pri, cb, data);
1388} 2304}
1389 2305
1390eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2306eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1391{ 2307{
1392 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2308 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1393} 2309}
1394 2310
2311eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2312{
2313 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2314}
2315
1395eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2316eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1396{ 2317{
1397 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2318 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1398} 2319}
1399 2320
1400eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2321eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1401{ 2322{
1402 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2323 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1403} 2324}
1404 2325
1405eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2326eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1406{ 2327{
1407 return eio__1path (EIO_READDIR, path, pri, cb, data); 2328 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1408} 2329}
1409 2330
1410eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2331eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1411{ 2332{
1412 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; 2333 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1413} 2334}
1414 2335
1415static eio_req * 2336static eio_req *
1416eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2337eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1417{ 2338{
1441eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2362eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1442{ 2363{
1443 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2364 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1444} 2365}
1445 2366
1446eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2367eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1447{ 2368{
1448 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2369 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1449} 2370}
1450 2371
1451#endif 2372#endif
1452 2373
1453eio_req *eio_grp (eio_cb cb, void *data) 2374eio_req *eio_grp (eio_cb cb, void *data)
1462#undef SEND 2383#undef SEND
1463 2384
1464/*****************************************************************************/ 2385/*****************************************************************************/
1465/* grp functions */ 2386/* grp functions */
1466 2387
2388void
1467void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2389eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1468{ 2390{
1469 grp->int2 = limit; 2391 grp->int2 = limit;
1470 grp->feed = feed; 2392 grp->feed = feed;
1471 2393
1472 grp_try_feed (grp); 2394 grp_try_feed (grp);
1473} 2395}
1474 2396
2397void
1475void eio_grp_limit (eio_req *grp, int limit) 2398eio_grp_limit (eio_req *grp, int limit)
1476{ 2399{
1477 grp->int2 = limit; 2400 grp->int2 = limit;
1478 2401
1479 grp_try_feed (grp); 2402 grp_try_feed (grp);
1480} 2403}
1481 2404
2405void
1482void eio_grp_add (eio_req *grp, eio_req *req) 2406eio_grp_add (eio_req *grp, eio_req *req)
1483{ 2407{
1484 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2408 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2409
2410 grp->flags |= EIO_FLAG_GROUPADD;
1485 2411
1486 ++grp->size; 2412 ++grp->size;
1487 req->grp = grp; 2413 req->grp = grp;
1488 2414
1489 req->grp_prev = 0; 2415 req->grp_prev = 0;
1496} 2422}
1497 2423
1498/*****************************************************************************/ 2424/*****************************************************************************/
1499/* misc garbage */ 2425/* misc garbage */
1500 2426
2427eio_ssize_t
1501ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2428eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1502{ 2429{
1503 etp_worker wrk; 2430 etp_worker wrk;
2431 eio_ssize_t ret;
1504 2432
1505 wrk.dbuf = 0; 2433 wrk.dbuf = 0;
1506 2434
1507 eio__sendfile (ofd, ifd, offset, count, &wrk); 2435 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1508 2436
1509 if (wrk.dbuf) 2437 if (wrk.dbuf)
1510 free (wrk.dbuf); 2438 free (wrk.dbuf);
1511}
1512 2439
2440 return ret;
2441}
2442

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines